- Email: [email protected]
Impact of offshore wind power forecast error in a carbon constraint electricity market
Energy xxx (2014) 1e11
Contents lists available at ScienceDirect
Energy journal homepage: www.elsevier.com/locate/energy
Impact of offshore wind power forecast error in a carbon constraint electricity market P. Higgins a, *, A.M. Foley a, R. Douglas b, K. Li a a b
School of Mechanical and Aerospace Engineering, Queen's University Belfast, Belfast BT9 5AH, United Kingdom School of Electronics, Electrical Engineering and Computer Science, Queen's University Belfast, Belfast BT9 5AH, United Kingdom
a r t i c l e i n f o
a b s t r a c t
Article history: Received 11 November 2013 Received in revised form 6 June 2014 Accepted 9 June 2014 Available online xxx
This paper investigates the impacts of offshore wind power forecast error on the operation and management of a pool-based electricity market in 2050. The impact from offshore wind power forecast errors of up to 2000 MW on system generation costs, emission costs, dispatch-down of wind, number of startups and system marginal price are analysed. The main findings of this research are an increase in system marginal prices of approximately 1% for every percentage point rise in the offshore wind power forecast error regardless of the average forecast error sign. If offshore wind power generates less than forecasted (13%) generation costs and system marginal prices increases by 10%. However, if offshore wind power generates more than forecasted (4%) the generation costs decrease yet the system marginal prices increase by 3%. The dispatch down of large quantities of wind power highlights the need for flexible interconnector capacity. From a system operator's perspective it is more beneficial when scheduling wind ahead of the trading period to forecast less wind than will be generated. © 2014 Elsevier Ltd. All rights reserved.
Keywords: Offshore wind Electricity markets Forecast error Scheduling Dispatch
1. Introduction The European Union (EU) has developed a long-term framework for reducing greenhouse gas emissions by 80%e95% by 2050 compared to 1990 levels [1]. This framework aims to help the EU become a competitive low carbon economy by 2050 by setting policy plans in areas such as transport, energy and climate change. The power sector is the most significant for achieving the 2050 targets as the EU roadmap specifies that carbon dioxide (CO2) emissions from the power sector are to be almost completely eliminated by 2050. The Republic of Ireland and Northern Ireland have developed national roadmaps for achieving the EU 2050 targets [2,3]. They estimate renewable resources could provide 74% and 60% of the electricity generated in the Republic of Ireland and Northern Ireland respectively. In both countries the renewable electricity comes mostly from onshore and offshore wind power. The total installed wind power in the Republic of Ireland will be 12,000 MW consisting of 6000 MW offshore and 6000 MW onshore. In Northern Ireland the total installed wind power by 2050 will be 2800 MW with 1200 MW offshore and 1600 MW onshore [2,3]. The all-island generating capacity for 2050 will be * Corresponding author. E-mail addresses: [email protected] (P. Higgins), [email protected] (A.M. Foley), [email protected] (R. Douglas), [email protected] (K. Li).
24,942 MW with the offshore wind in the Republic of Ireland and Northern Ireland making up 24% and 5% of the generating capacity respectively. The disadvantage of wind power is that it is variable and stochastic by nature which makes the role of the system operator (SO) even more challenging. Studies have shown that the inclusion of significant levels of wind generation into an electricity system can have major impacts on the system security, scheduling and dispatching of units [4,5]. Connolly et al. [6] concluded the ramping capabilities of the generating units are crucial to the deployed of high wind generation. The operating reserve and flexibility of the power system were found to be crucial to the development of wind energy in electricity markets [7,8]. Hong et al. [9] concluded that to accommodate the increasing amount of wind energy onto the Jiangsu's power system more flexible gas generating units are required rather than coal and nuclear power plants. Inflexible power systems with large penetrations of wind energy have been shown to have major problems with wind curtailment [10]. The variability of the wind power and responses of conventional generation at short notice is crucial to system balancing and understanding of high levels of renewable generation that a grid in the future can incorporate [11]. The SOs of the Single Electricity Market (SEM) in the Republic of Ireland and Northern Ireland released statistics on the wind forecast accuracy for the month of December 2012 [12]. The report highlighted a maximum and minimum
http://dx.doi.org/10.1016/j.energy.2014.06.037 0360-5442/© 2014 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
2
P. Higgins et al. / Energy xxx (2014) 1e11
forecast error of þ798 MW and 634 MW for a total installed wind capacity of 2088 MW. A maximum normalised mean absolute error of 48.8% and average 12.2% were experienced. The large planned wind capacity for 2050 will consequently mean wind power forecasting will become a critical power system planning tool. Onshore wind power forecasting techniques have improved dramatically and continue to advance but Offshore Wind Power (OWP) forecasting is more difficult due to limited datasets and knowledge [13]. As offshore wind generation increases the variability will become more pronounced and better planning techniques and forecasting will be required [14]. The accuracy of wind forecast has been shown to be considerably lower over an entire transmission zone rather than over a single zone and that averaging over wider areas increases forecast accuracy [15,16]. The wider spatial spread of wind farms and increased offshore wind generation has been found to be the main factors in the reduction of wind curtailment [17]. In Ref. [18] the impact of Ireland's 2020 wind energy targets with varying limits for wind curtailment was examined and it was shown that approximately 7%e14% of wind production was not dispatched. It was also shown that increases in wind forecast accuracy also resulted in reductions in wind curtailment but provided no improvement in emissions production [19]. A study of the SEM with 6000 MW of installed wind capacity stated ‘improving wind forecasting will lead to relatively small savings in system costs on a percentage basis but may account for millions of euros’ [20]. For a system like the SEM in 2050 with over 14 GW of wind power, wind forecasting could have a substantial impact on the operation of the system. Studies have highlighted the interconnector capacity between Great Britain and Ireland as critical during moments of high wind and low load [17,21,22]. Tuohy et al. [23] included Monte Carlo simulations for wind and load uncertainties and found that the mid-merit, peaking units and the interconnections were the most affected components of the electricity system. The method implemented in Ref. [23] involved analysing the difference between a perfect foresighted model and a stochastic optimisation model. Other studies with similar techniques have determined that a stochastic optimisation for an electricity system produces an increase in production costs in comparison to a model with perfect foresight [24,25]. The difference with this research is the impact of the magnitude of the OWP forecast error is being analysed not the optimising technique. The OWP forecast error not only affects the operation and management of the electricity system but also the financial aspect. In Ref. [26] wind power predictability as an investment factor for selecting onshore wind farm sites was investigated and it was found that predictability can play an important role in the operation and maintenance of offshore wind farms due to improved availability from reduced downtime periods. In Ref. [27] the financial profits of individual generating units are also affected by increased wind generation. An increase of wind energy on the system pushes more thermal/conventional generation off the system. This reduces the System Marginal Price (SMP) and therefore the potential profits for conventional plants. This new wind generation also requires the existing plants to operate more flexible which can increase running costs and downtime for maintenance as well as shortening the plant life. The increased pressure from conventional generators running with variability and the impact offshore wind forecast error has on the SMP and merit order means some conventional generators could find it difficult to survive on unpredictable future SMPs. This research shows that a considerable amount of work has been performed on the impacts of increasing wind generation on electricity systems, however, only a small amount has analysed the impact of OWP forecast error. The novelty of this work is the
analysis of the impact of offshore wind forecast error on an electricity system with over 60% of generating capacity originating from renewable energy resources. This paper is structured in five sections. Section 1 is the introduction. Section 2 presents the test system and describes the methodology used to investigate the impact of OWP forecast error in the SEM in 2050. Section 3 presents the results of the analysis. Section 4 discusses the impacts of OWP forecast error in the SEM in 2050 and conclusions. 2. Test system and methodology 2.1. Single electricity market (SEM) The SEM is a mandatory all-island wholesale pool market through which generators and suppliers trade electricity. The market operates over three trading periods: before, intra-day and after. The before trading period consists of each generating unit bidding their commercial offer data and technical offer data for each half hour interval in the intra-day. The commercial offer data and technical offer data contain price/quantity bids and no load costs. The SOs produce the reserve constrained unit commitment (RCUC) schedule based on forecasted wind generation, price/ quantity pairs from all generating units and the SEM system stability requirements. The difference between an unconstrained unit commitment and RCUC is the inclusion of transmission constraints. From the RCUC schedule the SOs inform each generating unit of their required generation and operating hours before the start of intra-day trading. The intra-day period involves the real time application of the generation schedule. Both SOs implement the schedule and are responsible for delivering an efficient operation of the wholesale power market. The SO's continuously analyse the wind forecast, generating capacity and system demand requirements prior to the trading period based on updates to maintain system stability. The four days after the intra-day the Single Energy Market Operator (SEMO) calculates the SMP for each trading period during the intra-day. These four days are known as the after period. This SMP is the price applicable to both generators and suppliers receiving/making payments for electricity generated/ used [28]. 2.2. PLEXOS model The majority of research for modelling unit commitment and dispatch scheduling of electricity systems have used either deterministic or stochastic optimisation techniques [18e20,23,25,29]. The main difference between both techniques is that deterministic modelling with perfect foresight ‘may provide lowest cost solutions for system dispatch but also may provide unrealistic results’ [30], whereas stochastic optimisations mimic real power systems dispatches. For this research a mixed integer, stochastic optimisation will be applied using Energy Exemplar's PLEXOS for Power Systems version 6.208R03 with the Xpress optimiser [31]. Xpress was set with mixed integer programming at a relative gap of 0.05%. Xpress aims to minimise the objective function, shown in Equation (1) which is conditional of a number of constraints:
min
X
cjt :Ujt þ njt :Vjt þ mjt :Pjt þ vl:uset þ vl:usrt
(1)
t2T
where t indexes time periods in chronological order t ¼ 1, …, T, cjt is the start cost of unit j in period t, Ujt is a binary quantity representing if unit j has started the period before t, njt is the no load cost of unit j, Vjt is a binary quantity representing the generating status of unit j, mjt is the production cost of unit j, Pjt is the power output of
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
P. Higgins et al. / Energy xxx (2014) 1e11
unit j, vl is the penalty for loss load, uset is the unserved energy of unit j and usrt is the reserve energy not met by unit j. The objective function is subjected to constraints such as maximum generation, minimum stable generation, minimum up and down times and ramp rates. PLEXOS requires generating/ transmission parameters such as maximum and minimum ramp up/down time, minimum stable level, start-up cost, forced outage rate, wind input data, operational and maintenance costs and fuel costs to populate the objective function. The model optimises each of the 365 days in 2050 at 1 h intervals. A study found the time intervals can have a significant effect on the results [32]. A small time interval of five to 15 min is required when analysing the ramping of the generation system and can have an impact on the generation costs. The issue with smaller time intervals is the increase in computational costs, a 5 min interval had a run time of 70 h whereas the 1 h interval had a run time of 3 h. For this research the 1 h interval is used as there are a large number of scenarios to be implemented and the model should still provide realistic results that represent the SEM dispatch scheduling. 2.3. Replicating SEM activities in PLEXOS The test system must mimic the SEM operation to fully investigate the impact OWP forecast error has on the wholesale market, as seen in Fig 1. Therefore a ‘day ahead’ model is required to produce a dispatch schedule similar to the RCUC of the before period in the SEM. The day-ahead model has a 24 h look ahead and maintenance outages included. The 24 h look ahead provides the model with enough time to develop an optimum schedule including the maintenance outages. The forecasted wind with no errors is included in this model. A ‘real time’ model incorporates the OWP forecast error and the dispatch schedule from the day ahead model which represents the intra-day trading period. The look ahead of the real time model was shortened to 3 h to replicate the short notice of OWP forecasting errors. A 3 h look ahead was found to be the optimum frequency of rescheduling the system due to changes in load and wind forecasts for a stochastic optimisation model [23]. Maintenance and forced outages are applied to the real time model to replicate the realistic short notice plant outages. The wind forecast error scenarios are applied to the real time model to investigate the impact OWP forecast error has on the market operation and management of the SEM. The after trading period of the SEM is not modelled for this
3
research. Further research projects will involve the development of the after trading period. 2.4. Test system The test system was built using 2011 as the base case (SEM_2011) [33]. The SEM_2011 is the all-island market consisting of 88 different generating units. The test system includes two interconnectors to the British Electricity Trading and Transmission Arrangement (BETTA) market and a single gas generating unit to represent the BETTA market. Interconnection to the BETTA has a combined capacity of 900 MW which is increased to 2000 MW for 2050 [3]. 2.4.1. Demand and generation portfolios The yearly demands for electricity in the Republic of Ireland and Northern Ireland are expected to increase from their 2010 levels of 28,000 GWh and 9000 GWh respectively to 48,000 GWh and 12,000 GWh in 2050 [2,3]. The required generation portfolio to meet the 2050 electricity demand is shown in Table 1. The total installed 2050 generating capacities for the Republic of Ireland and Northern Ireland are approximately 20,800 MW and 4400 MW respectively. A number of existing generators are expected to be decommissioned over the next few years. The Great Island units with a combined generating capacity of 212 MW, all the Tarbert units (595 MW) and Ballylumford gas units (510 MW) are all to be decommissioned. It is assumed that Kilroot coal generation will be decommissioned by 2050 as it will have to comply with the Industrial Emissions Directive [34]. The peat, oil and distillate generating units in both jurisdictions and the coal generating units in Northern Ireland are expected to be decommissioned by 2050 [3]. The introduction of carbon capture storage (CCS) technology should maintain a small amount of coal generation in 2050 [35]. Therefore it is assumed that gas will be the primary fossil fuel generation in 2050. 2.4.2. Onshore and offshore wind generation The all-island wind was modelled by separating the onshore wind into 13 regions and the offshore wind into seven regions [36], as shown in Fig 2. Each onshore wind region is modelled with a max capacity and an excel file containing a percentage value for every hour period in a year. The wind generation for every hour
Fig. 1. Test system structure.
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
4
P. Higgins et al. / Energy xxx (2014) 1e11
Table 1 2050 SEM generation portfolio. Max capacity (MW) Fuel source
All-island (2011)
Republic of Ireland (ROI-2050)
Northern Ireland (NI-2050)
Total (2050)
Offshore wind Onshore wind Wave/tidal Hydro/PHESa Gas CCGTb Gas OCGTc Coal (CCS)d Total
e 1935 e 508 4072 720 1331 8566
6000 6000 1080 508 5201 1185 850 20,824
1200 1600 200 10 900 500 e 4410
7200 7600 1280 518 6101 1685 850 25,234
a b c d
Pumped Hydro Energy Storage. Combined Cycle Gas Turbine. Open Cycle Gas Turbine. Carbon Capture Storage.
period in the year is a combination of the percentage file and the max capacity. The 2050 max capacity for each region is determined with the Grid 25 strategy in consideration [37]. The total wind for 2050 in the Republic of Ireland is expected to be 12,000 MW consisting of 6000 MW of onshore and 6000 MW of offshore. Northern Ireland will have 2800 MW of wind with 1600 MW of onshore wind and 1200 MW of offshore wind. Wind profiles for offshore wind regions are more difficult to obtain as there is a lack of data. Therefore the offshore files were manipulated from the onshore regions. Research showed the offshore wind regions of the east coast of the Island have a 1 h time lag to the respective onshore wind regions and the spatial correlation of neighbouring wind regions was found to be in the range of 0.94e0.97 [18]. The offshore wind resource is more powerful than onshore and as a result the capacity factor of offshore wind turbines is approximately 40%. Taking these factors into consideration the offshore wind data files were time and correlation adjusted accordingly using the same procedures as applied in Ref. [18].
2.4.3. Fuel and carbon price Forecasted fuel prices were obtained from an EU world energy technology outlook to 2050 [38] and are shown in Table 2. The price of coal does not experience the same projected increases as gas as it is envisaged that coal could return as an important source of electricity and will be increasingly converted using new advanced technologies [38]. The CO2 price of V234/tCO2 is based on figures from the United Kingdom government for 2050 [39]. The CO2 price is added to the model through a tax on the price of fuel. The sensitivity of the CO2 price is investigated by applying two additional prices of V117/tCO2 and V351/tCO2 to the system. 2.4.4. Transmission constraints The SOs have implemented a number of constraints to ensure the efficient and secure operation of the transmission system [40]. The main transmission constraint influencing OWP generation is the System Non-Synchronous Penetration (SNSP) limit. The SNSP is a measure of the non-synchronous generation on the system at any time. It is a ratio of non-synchronous generation (onshore wind, offshore wind, wave and tidal) and high voltage direct current (HVDC) imports to demand and HVDC exports. Currently the SOs has set the System Non Synchronous Penetration (SNSP) at 50%, this value is expected to rise to between 70% and 80% by 2020 [18]. For this 2050 model the SNSP is set at 75%. The formula for SNSP is
SNSP ¼
Non synchronous generation þ HVDC Imports % ¼ 75% System Demand þ HVDC Exports (2)
2.4.5. OWP forecast error scenarios The following method was implemented to analyse the magnitude and variance of the OWP forecast error. EirGrid publicly provides the wind generation and respective wind forecast data for every 15 min interval from 2010 onwards [41]. This wind data is recorded from the SEM which consists of onshore wind generation only. OWP forecasted and generation data is difficult to source as it is not publicly available. Therefore it will be assumed that the onshore wind forecast error from the EirGrid data will be similar to the OWP forecast error. From this data the magnitude of the wind forecast error was determined using the following formula.
Wind Forecast Error ¼
Wg Wf % Wg
(3)
where Wg is the generated wind (MW) and Wf is the wind forecast (MW). Fig 3 shows the distribution of the wind forecast error of wind generated in the SEM in 2012. The figure illustrates the frequency of each forecast error and which are the most dominate. It can be noted from the figure that the most dominate wind forecast error occurs between ±10% of the wind generation. Since the 60% of forecast error occurs between ±20% of wind generation eight scenarios ranging from 20% to þ20% will be analysed, as seen in Table 4. For each scenario a constant OWP forecast error is applied to the offshore wind generation in the real time model. The purpose of this work is to analyse the OWP forecast error therefore onshore wind generation will have no forecast error applied to it.
Table 2 Fuel prices.
Fig. 2. SEM wind regions.
Fuel type
Price (V/GJ)
Coal Gas
3.29 12.45
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
P. Higgins et al. / Energy xxx (2014) 1e11
5
Fig. 3. Onshore wind forecast error in the SEM (2012).
A Dutch study [42] using weather data and multi-turbine power curves determined the actual and forecasted wind power for 12 GW of installed onshore and offshore wind capacity. The research produced the normalised standard deviation of the wind power forecast error and highlighted that the forecast error decreased to approximately 5% for 1 h prior to operation and 9% for 6 h prior to operation. For this research the look ahead or forecasted error lag was fixed at 3 h, as discussed in Section 2.3. The wind generated and forecasted data for 2010, 2011 and 2012 was used to provide three forecast error profiles that include variance. The forecast error was found for every period in each year and converted into a percentage. The percentage forecast error file was multiplied by the wind forecast data to provide three generation profile with forecast error and variance, as shown by the 24 h sample in Fig 4. The new generation profiles were implemented in the real time model and were used to examine the impact of OWP forecast error variance. The variances and average errors of three profiles are listed in Table 3. The scenarios for this analysis are summarised in Table 4. The 2%, 5%, 10% and 20% scenarios will be referred to as deficit forecast errors scenarios and the 2%, 5%, 10% and 20% scenarios will be referred to as surplus forecast errors scenarios. The deficit and surplus scenarios are used to analyse the impact from the OWP forecast error magnitude. The 13%, 4% and 7% are used to analyse the impact from the OWP forecast error variance. The ‘Base e V117’ and ‘Base e V351’ scenarios are used to analyse the price of carbon impact.
possible to look at the impact of the variance of OWP forecast error and the sensitivity of the price of carbon. 3.1. Generation Fig 5 shows the generation comparison between all the scenarios in terms of GWh. The figure illustrates the annual generation for each fuel type over the year, from the 1st of January 2050 to 31st of December 2050. The major generating units in the SEM in 2050 are onshore wind, OWP and gas generation, with all three combining to generate over 80% of the annual system generation. In the SEM the major generation changes due to OWP forecast error are gas and onshore wind. The more efficient Combined Cycle Gas Turbines (CCGTs) are the most effected due to the offshore wind forecast error. The usage of CCGTs instead of Open Cycle Gas Turbines (OCGTs) or Combined Heat Plant (CHP) to match the change Table 4 List of scenarios. Scenario
Details
20%
A constant 20% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 10% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 5% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 2% error applied to the forecasted data, no variance, carbon price of V234/tCO2 No error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 2% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 5% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 10% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 20% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A varying error with an average of 13% applied to the forecasted data, variance of 0.09, carbon price of V234/tCO2 A varying error with an average of 4% applied to the forecasted data, variance of 0.146, carbon price of V234/tCO2 A varying error with an average of 7% applied to the forecasted data, variance of 0.15, carbon price of V234/tCO2 No error applied to the forecasted data, no variance, carbon price of V117/tCO2 No error applied to the forecasted data, no variance, carbon price of V351/tCO2
10% 5% 2% Base 2%
3. Results and analysis 5%
A comparison of the results was performed and is presented in the following tables and figures. The impact of the magnitude of OWP forecast error has on the electricity market was compared over a range of different parameters such as: unit generation, generation costs, net interchange between the SEM and the BETTA, emission production and costs, SMP of electricity, unit operational time and dispatch down of wind. From this analysis it was then
20% 13%
4%
7%
Table 3 Average OWP forecast error and variance.
Average error Variance
10%
2010
2011
2012
13% 0.09
4% 0.146
7% 0.15
Base e V117 Base e V351
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
6
P. Higgins et al. / Energy xxx (2014) 1e11
Fig. 4. Forecast error with variance.
Fig. 5. OWP forecast error on generation.
in OWP forecast error results in lower generation costs. For surplus scenarios when the offshore wind generation is greater than forecasted, onshore wind and gas are directed to curtail/reduce generation and less is imported from the BETTA. The majority of the dispatching down of non-synchronous generation is onshore and offshore wind as the renewable generation for Northern Ireland (NI Renewables) and the Republic of Ireland (ROI Renewables) remains relatively constant. For the deficit scenarios where offshore wind generation is less than forecasted, an increase in onshore wind generation and gas generation is experienced. Approximately 55% of the increase in gas generation is provided by CCGTs and 43% by OCGTs. The gas peaking plants of CHP and steam turbines (ST) are the most affected gas generators in terms of start-ups and shut downs. The increase in onshore wind generation is due to the curtailing effects of the SNSP limit. Table 5 shows the impact the OWP forecast error has on the net interchange between the SEM and the BETTA. The net interchange is the HVDC exports from the SEM less the HVDC imports to the SEM. A positive value represents exports from the SEM to the BETTA and a negative represents an import from the BETTA. As expected there is a close correlation between the increase in surplus offshore wind forecast error and an increase in exports to the BETTA. 3.2. Generation costs Fig 6 illustrates the impact on generation costs due to the OWP forecast error. The offshore wind, onshore wind and both NI and ROI renewables are modelled as a zero generation cost as they are
must run units and have to be dispatched first. When OWP forecast error is a surplus the generation costs are cheaper as more OWP is on the system. The least efficient thermal generators that were dispatched in the base scenarios are ramped down or switched off. The majority of generation costs are from gas generators as over 40% of the generated electricity is from gas and approximately 50% is from renewable energy sources. Similar to 2014 the SEM system costs and SMPs in 2050 are still driven by the price of gas. The CCGT plants provide the majority of generation changes to balance the OWP forecast error and as a result it is their generation costs that affect the system generation costs the most. When less offshore wind is generated than forecasted, it is more economical to keep the coal generators online for longer periods than shut them down as per the day ahead schedule. The number of coal generator startups decrease as a result longer operational periods.
Table 5 OWP forecast error on the net interchange. Scenario
Net interchange (GWh)
20% 10% 5% 2% Base 2% 5% 10% 20%
11,403 11,790 11,981 12,086 12,137 12,197 12,269 12,390 12,597
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
P. Higgins et al. / Energy xxx (2014) 1e11
7
Fig. 6. OWP forecast error on generation costs.
The difference in generation costs due to the offshore wind forecast error is highlighted in Table 6. The values in the table show an OWP forecast error could increase the generation costs of the entire system by V193 million or an increase of 4%. The deficit scenarios produce a greater change in generation costs than the surplus scenarios. This is due to less efficient and more expensive coal and gas generators coming online to balance the loss from offshore wind generation. The generation costs change by 0.2% or V9 million for every percentage change in the OWP forecast error. 3.3. Emission production and costs Fig 7 shows the changes in emission production for the different OWP forecast scenarios. The deficit OWP forecast error scenarios produce more emissions than the emissions saved from the similar surplus OWP forecast scenarios. This is due to less efficient plants running in the deficit scenarios and being shut down in the surplus scenarios as discussed in section 3.1. The V234/tCO2 price of carbon results in a large emissions costs of approximately V2.65 trillion for the base scenario. The 20% offshore wind forecast error results in an increase of V0.17 trillion. Similar to the generation costs this highlights the need to reduce the OWP forecast error as much as possible. A comparison of the 2011 and 2050 emissions production shows a 50% reduction from 0.466 t/MWh to 0.185 t/MWh but a 600% increase in emissions cost from V6.06/MWh to V37/MWh. The significant impact the price of carbon has on emission costs is discussed in Section 3.7. 3.4. System marginal price Fig 8 shows the change in the yearly average SMP due to the OWP forecast errors. It highlights the yearly average SMP with the
Table 6 Difference in generation costs from base scenario. Scenario 20% 10% 5% 2% Base 2% 5% 10% 20%
Change in generation costs (V million) 193.0 88.2 47.0 21.5 e 16.5 45.1 87.3 167.5
price cap and without the price cap. The price cap is a constraint on the system which limits the maximum SMP that can occur and is set at V1000/MWh. Usually the price cap is applied when generation is less than the system demand. Before such an instance the SO implements operating reserve to meet the difference demand and generation. A recent report showed the price cap was reached once during the year 2013 [43]. When the price cap is modelled in the 2050 system it is reached eleven times in the base scenario and over one hundred times in the 20% scenario. The significant increase in the price cap occurrences for the 20% forecast error scenario is due to the large difference in generation and forecast, and most importantly the short time window restricting the scheduling of extra generation. The test system has produced an SMP for 8766 h periods and only 128 of those periods hit the price cap. Fig. 8 illustrates the impact the 1% (128 periods) of price capped SMP values can have on the yearly average SMP for the 20% offshore wind forecast error. If the price capped periods were equal to the average SMP a significant drop occurs for all the deficit scenarios. The most affected is the 20% scenario with a reduction of V11.44/MWh or 6%. The problem is not the price cap but the transmission and reserve constraints. The transmission constraints are not providing enough reserve to ensure security of supply and as a result the SMP price hits the price cap. Similar to the finding of Ref. [18] the transmission constraints need to be addressed as the 2011 set will not be suitable for 2050 when large quantities of nonsynchronous generation (onshore wind, offshore wind and wave) are present. It was found that the exclusion of the price cap influences the þ20% OWP forecast error results in a decrease of V3/MWh or 2% from the base scenario in the annual average SMP. Fig. 8 shows an increase of 0.1% for every percentage point rise in the surplus forecast error scenarios. 3.5. Dispatch-down of wind The reduction of electricity generated from wind can be described as the dispatch-down of wind generation. Wind that is dispatch-down can be due to constraints in the network or due to curtailment issues. The EirGrid ‘policy for scheduling and dispatch decisions’ prioritises the order of dispatching generating units [44]. EirGrid currently dispatches generators in the following order after the closure of the market gate, interconnector, peat, hydro and then wind. Fig 9 shows the monthly figures for the dispatch-down wind and the corresponding wind generation. As expected a close
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
8
P. Higgins et al. / Energy xxx (2014) 1e11
Fig. 7. Offshore wind forecast error impact on emissions production and costs.
correlation between the winter months (OctobereMarch) with high wind generation and the highest dispatch-down values was experienced. In 2011 80% of dispatch-down on the SEM was from the curtailment effects of the SNSP limit [45]. In 2050, with 2000 MW of interconnector capacity, the majority of the dispatchdown wind is due the SNSP and the lack of interconnector capacity. The impact of the OWP forecast error on the dispatch-down wind generation is shown in Fig 10. The base model has 10,196 GWh of dispatch down wind, 7245 GWh of onshore wind and 2951 GWh of offshore wind. Under the current renewable energy feed in tariffs this unused wind energy would still have to be paid for therefore this is clearly an undesirable amount of dispatched down wind energy. The surplus þ20% scenario generates an extra 4195 GWh of offshore wind. Approximately 43% (1797 GWh) of this extra wind generation is allowed onto the system while 57% (2398 GWh) is added to the dispatched down of wind energy. In the deficit scenarios when offshore wind generation is less than expected, more onshore wind generation is allowed onto the system to replace the offshore wind generation. For the deficit 20% scenario 4985 GWh of offshore wind is missing from the system resulting in an increase of 2102 GWh of onshore wind generation and a decrease of 2883 GWh in dispatched down wind. The SNSP limit that is restricting the amount of non-synchronous generation on the system is causing more than 50% of the extra offshore wind generation in the surplus scenarios to be curtailed. In
the deficit scenarios the SNSP limit allows more onshore wind generation onto the system to compensate for the offshore wind forecasting error. 3.6. Impact of wind power forecast error variance When variance is included to a deficit 7% OWP forecast error scenario 13.97 million tonnes of CO2 are produced at a cost of V2.79 trillion, as shown in Fig 11. This emission production is 2% more than a deficit 10% offshore wind forecast error which produces 13.77 million tonnes of CO2 at a cost of V2.75 trillion. As discussed in Section 3.3 when the error of the OWP forecast is smaller the emission production should reduce however when variance is included the emission production increases. The offshore wind generation is approximately the same but the introduction of variance results in an increase in start-ups for gas generators and longer operational time for coal generators which increases the emissions production and costs. An increase in emissions production and costs are seen for both deficit and surplus scenarios that include variance as the number of startups for gas generators and operational hours for coal generators increases. The same trend is recorded for the SMP when variance is applied to the forecast error. The introduction of variance into the OWP forecast error causes a significant impact on the SMP as seen in
Fig. 8. Offshore wind forecast error impact of SMP.
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
P. Higgins et al. / Energy xxx (2014) 1e11
9
Fig. 9. Monthly dispatch-down wind and wind generation for 2050 base scenario.
Fig. 10. Offshore wind forecast error on the dispatch down of wind.
Fig. 12. The surplus forecast error (4%) has more offshore wind generation and therefore should result in a decrease in the SMP. However, the variance results in an increase in the SMP. The introduction of the variance increases the amount of price cap instances and this is the cause for the significantly larger SMP (with Price Cap) figures.
Fig. 12 shows an increase of approximately 1% for every percentage point rise in the forecast error scenarios no matter the sign of the average forecast error. The inclusion of variance results in extra shut downs, start-ups and generating of expensive gas peaking plants for the positive and negative OWP forecast error scenarios. The additional operating of these peaking plants results
Fig. 11. OWP forecast error impact on emission production and costs (variance).
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
10
P. Higgins et al. / Energy xxx (2014) 1e11
Fig. 12. OWP forecast error impact on SMP (variance).
in the SMP increases. The significant increases on the base scenario SMP shows the need for an accurate OWP forecast when such a large percentage of the generation portfolio consists of offshore wind energy. 3.7. Carbon price sensitivity A UK government report [39] provided a low, medium and high price of carbon for 2050 as V117/tCO2, V234/tCO2 and V351/tCO2 respectively. The impact from the price of carbon on generation costs, emissions costs and SMP, for the no forecast error scenario, are listed in Table 7. The large carbon price variations produce differences in the generation and emissions costs of approximately V1 billion and V1 trillion respectively. Table 8 shows the effect the OWP forecast has on the SMP with the different carbon prices. The lower the carbon price the greater the impact on SMP from OWP forecast error. 4. Discussion and conclusion The analysis shows that OWP forecast error has the potential to have significant operation and management impacts on the SEM in 2050 in terms of SMP, generation costs, emissions costs, ramping, operational hours and dispatching of wind. The OWP capacity modelled for 2050 accounted for 30% of the total installed capacity. This significant OWP penetration resulted in periods throughout the year where the OWP forecast error experienced a difference of 2000 MW between forecasted and generated offshore wind power. The research showed the OWP forecast error caused small percentage changes to the SMP but considerable V/MWh changes. A percentage point change in the OWP forecast error resulted in a 0.1% variation to the SMP for all of the scenarios where no variance is included. The introduction of variance into the OWP forecast error produces different changes to the SMP. The surplus forecast error (4%) has more offshore wind generation and therefore should
Table 7 Carbon price sensitivity for base scenario with no forecast error.
result in a decrease in the SMP. However the variance results in an increase in the SMP as additional generation, start-ups and shut downs of gas peaking plants are required to balance the OWP forecast error. An SMP increase of approximately 1% for every percentage point rise in the forecast error scenarios was noted regardless of the average forecast error sign. This increase on the base scenario SMP shows the need for an accurate OWP forecast when such a large percentage of the generation portfolio consists of offshore wind energy. Further analysis of the operating reserve is required to reduce the impact of the price cap on the SMP. A comparison of the 2011 and 2050 emissions production shows a 50% reduction from 0.466 t/MWh to 0.185 t/MWh but a 600% increase in emissions cost from V6.06/MWh to V37/MWh. In 2050 the price of carbon is one of the biggest contributors to the SMP. The sensitivity analysis of the price of carbon shows the larger the price of carbon the less impact OWP forecast error has on SMP. All of the results in this paper are based on a carbon price of V234/tCO2. The V234/tCO2 price of carbon results in a large emission costs of V2.67 trillion for the base scenario. The surplus OWP forecast error scenarios produce a decrease in emissions production and cost. However, when variance is added to the scenario the emissions production and costs increase. This is due to an increase in coal generation which has a large CO2 emission per gigajoule of energy. Increased OWP forecast accuracy could allow more efficient units to be dispatched rather than fast ramping and inefficient peaking generators. The OWP forecast error has a significant impact on the onshore wind generation as the SNSP limit restricts the amount of nonsynchronous generation on the system. The base scenario with no forecast error dispatches down nearly 25% of the available wind generation. The dispatch down of large quantities of wind power in the base scenario highlights the need for additional interconnector capacity or energy storage. The lack of sufficient interconnector capacity and SNSP limit is resulting in less than 50% of the offshore wind forecast error generation being utilised. As highlighted by
Table 8 OWP forecast error with different carbon prices.
Carbon price
Generation costs (Vbillion) Emissions costs (Vtrillion) Annual average SMP
Price of carbon
V117/tCO2
V234/tCO2
V351/tCO2
V3.76 V1.40 V168.57
V4.70 V2.67 V190.46
V5.66 V3.99 V230.29
Base SMP 4% SMP 13% SMP
V117/tCO2
V234/tCO2
V351/tCO2
V168.57 V174.87 V186.34
V190.46 V195.93 V208.72
V230.29 V236.87 V246.81
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
P. Higgins et al. / Energy xxx (2014) 1e11
Refs. [17,21e23] the flexibility of the interconnectors from Ireland is crucial to reducing the dispatched down wind. An increase in interconnector capacity should also reduce the impact of OWP forecast error on the operation and management of the electricity system. With variance included the OWP forecast error scenario of 4% resulted in a decrease in generation cost of V31.5 million or 0.67%, whereas the 13% scenario resulted in an increase of V160.3 million or 3.41%. Both of these scenarios show the different extremes that could occur depending on the sign of the average error of the OWP forecast. Therefore, similar to comments from Ref. [15], when scheduling ahead it is more beneficial to forecast less wind than will be generated (surplus scenario). From an SO's perspective it is more beneficial to develop a schedule with less wind generation than expected and during real time allow the extra wind from the forecast error onto the system while reducing thermal generation. Constraining thermal generators in such a way incurs constraint costs and future work is required to determine the full financial costs associated with constraining thermal generators. In conclusion this paper investigates the impacts of OWP forecast error on the operation and management of the SEM in 2050, which is a carbon constraint electricity market. Acknowledgements The authors acknowledge the support of Energy Exemplar through the academic license for PLEXOS, the Engineering and Physical Sciences Research Council for funding support and Edward McGarrigle from University College Cork for his assistance. References [1] European Commission. A roadmap for moving to a competitive low carbon economy in 2050. Communication from the commission to the European parliament, the council, the European economic and social committee and the committee of the regions 2011. COM; 2011. p. 112. final. [2] Sustainable Energy Authority Ireland. Smartgrid roadmap 2050; 2012. [3] Ricardo-AEA. Envisioning the future e considering energy in Northern Ireland to 2050; 2013. Ricardo-AEA/R/ED58435. [4] Gil E. Evaluating the impact of wind power uncertainty on power system adequacy. In: Proceedings of the PMAPS; 2012. Istanbul, Turkey. [5] Tuohy A, Denny E, Meibom P, Barth R, O'Malley M. Operating the Irish power system with increased levels of wind power. In: Proceedings of the IEEE Power and Energy Society General Meeting e Conversion and Delivery of Electrical Energy in the 21st Century; 2008. pp. 1e4. Pittsburgh, America. [6] Connolly D, Lund H, Mathiesen BV, Leahy M. Modelling the existing Irish energy-system to identify future energy costs and the maximum wind penetration feasible. Energy 2010;35:2164e73. [7] Lannoye E, Flynn D, O'Malley M. Evaluation of power system flexibility. Proc IEEE Trans Power Syst 2012;27(2):922e31. [8] Bjelic I, Rajakovic N, Cosic B, Duic N. Increasing wind power penetration into the existing Serbian energy system. Energy 2013;57:30e7. [9] Hong L, Lund H, Moller B. The importance of flexible power plant operation for Jiangsu's wind integration. Energy 2012;41:499e507. [10] Finn P, Fitzpatrick C, Connolly D, Leahy M, Relihan L. Facilitation of renewable electricity using price based appliance control in Ireland's electricity market. Energy 2011;36:2952e60. [11] Kubik ML, Coker PJ, Barlow JF, Hunt C. A study into the accuracy of using meteorological wind data to estimate turbine generation output. Renew Energy 2012;51:153e8. [12] EirGrid, SONI. Wind forecast accuracy statistics; 2012. Available from: http:// www.eirgrid.com/media/AllIslandWindForecastAccuracyReportDecember 2012.pdf. [13] Trombe P, Pinson P, Madsen H. A general probabilistic forecasting framework for offshore wind power fluctuations. Energies 2012;5(3):621e57. [14] Foley A, Leahy P, Marvuglia A, McKeogh E. Current methods and advances in forecasting of wind power generation. Renew Energy 2011;37:1e8. [15] Newbery D. Contracting for wind generation. Econ Energy Environ Policy 2011;1(2).
11
[16] Kiviluoma J, O'Malley M, Touhy A, Meibom P, Milligan M, Lange B, Holttinen H, Gibescu M. Impact of wind power on the unit commitment, operating reserves, and market design. In: Proceedings of the IEEE Power and Energy Society General Meeting; 2011. pp. 1e8. California, USA. [17] Foley A, Leahy P, McKeogh E. Wind energy integration and the Wales-Ireland interconnector. In: Proceedings of the IEEE PES/IAS Conference on Sustainable Alternative Energy; 2009. pp. 1e6. Valencia, Spain. [18] McGarrigle E, Deane P, Leahy P. How much wind energy will be curtailed on the 2020 Irish power system? Renew Energy 2013;55:544e53. [19] McGarrigle E, Leahy P. The value of accuracy in wind energy forecasts. In: Proceedings of the 12th International Conference on Environmental and Electrical Engineering; 2013. Wroclaw, Poland. [20] Meibom P, Barth R, Hasche B, Brand H, Weber C, O'Malley M. Stochastic optimization model to study the operational impacts of high wind penetrations in Ireland. Proc IEEE Trans Power Syst 2011;26(3):1367e79. [21] Denny E, Touhy A, Meibom P, Keane A, Flynn D, Mullane A, et al. The impact of increased interconnection on electricity systems with large penetrations of wind generation: a case study of Ireland and Great Britain. Energy Policy 2010;38:6946e54. [22] Foley AM, O'Gallachoir BP, McKeogh E, Milborrow D, Leahy P. Addressing the technical and market challenges to high wind power integration in Ireland. Renew Sustain Energy Rev 2013;19:692e703. [23] Tuohy A, Meibom P, Denny E, O'Malley M. Unit commitment for systems with significant wind penetration. Proc IEEE Trans Power Syst 2009;24(2): 592e601. [24] Meibom Barth R, Brand H, Hasche B, Swider D, Ravn H, Weber C. Wind variability management studies e all island grid study workstream B. Department of Enterprise, Trade and Investment; 2007. [25] Wang J, Botterud A, Miranda V, Monteiro C, Sheble G. Impact of wind power forecasting on unit commitment and dispatch. In: Proceedings of the 8th International Wind Integration Workshop; 2009. Bremen, Germany. [26] Girard R, Laquaine K, Kariniotakis G. Assessment of wind power predictability as a decision factor in the investment phase of wind farms. Appl Energy 2013;101:609e17. [27] Kubik M, Coker P, Hunt C. The role of conventional generation in managing variability. Energy Policy 2012;50:253e61. [28] Single Electricity Market Operator. Available from: http://www.sem-o.com/ Pages/default.aspx; 2012. [29] Denny E, O'Malley M. Quantifying the total net benefits of grid integrated wind. Proc IEEE Trans Power Syst 2007;22(2):605e15. [30] Deane P, McKeogh E, O'Gallachoir B. Deviation of robust storage targets for large scale pumped hydro energy storage using PLEXOS. In: Proceedings of the 2010 European PLEXOS User Meeting; 2010. London, United Kingdom. [31] FICO Xpress Optimiser. Available from: http://www.fico.com/en/Products/ DMTools/xpress-overview/Pages/Xpress-Optimizer.aspx; 2013. [32] Deane J, Drayton G, Gallachoir B. The impact of sub-hourly modelling in power systems with significant levels of renewable generation. Appl Energy 2014;113:152e8. [33] Commission for Energy Regulation (CER). Available from: http://www.cer.ie/; 2012. [34] EirGrid, SONI. All-island generation capacity statement 2013-2022. EirGrid and System Operator of Northern Ireland (SONI); 2012. [35] Poyry Energy Consulting. Options for low-carbon power sector flexibility to 2050; 2010. p. 655. [36] Redpoint Energy Ltd. Single electricity market operator (SEMO) forecast model of 2011-2012; 2011. SEM-11-072. [37] EirGrid. GRID 25 e a strategy for the development of Ireland's electricity grid for a sustainable and competitive future; 2012. [38] European Commission. World energy technology outlook to 2050; 2007. [39] Department of Energy and Climate Change. Guidance on estimating carbon values beyond 2050: an interim approach. Available from: https://www.gov. uk/government/uploads/system/uploads/attachment_data/file/48108/1_2010 0120165619_e____carbonvaluesbeyond2050.pdf. [accessed 10.05.13]. [40] EirGrid, SONI. Transmission constraint groups. EirGrid and System Operator of Northern Ireland (SONI); 2012. V1.4. [41] EirGrid, SONI. Wind generated and forecasted data. Available from: http:// www.eirgrid.com/. [42] Gibescu M, Kling W, Ummels B. Case study for the integration of 12 GW wind power in the Dutch power system by 2020. In: Proceedings of the CIGRE/IEEE PES joint symposium on Integration of wind-scale renewable resources into the power delivery system; 2009. USA. [43] Commission for Energy Regulation (CER). Trading and settlement code e policy parameters 2014; 2013. SEM-13-053. [44] EirGrid. Policy for implementing scheduling and dispatch decisions; 2011. SEM-11-062. [45] EirGrid, SONI. 2011 Curtailment report; 2013. Available from: http://www. eirgrid.com/media/2011_Curtailment_Report.pdf.
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
Contents lists available at ScienceDirect
Energy journal homepage: www.elsevier.com/locate/energy
Impact of offshore wind power forecast error in a carbon constraint electricity market P. Higgins a, *, A.M. Foley a, R. Douglas b, K. Li a a b
School of Mechanical and Aerospace Engineering, Queen's University Belfast, Belfast BT9 5AH, United Kingdom School of Electronics, Electrical Engineering and Computer Science, Queen's University Belfast, Belfast BT9 5AH, United Kingdom
a r t i c l e i n f o
a b s t r a c t
Article history: Received 11 November 2013 Received in revised form 6 June 2014 Accepted 9 June 2014 Available online xxx
This paper investigates the impacts of offshore wind power forecast error on the operation and management of a pool-based electricity market in 2050. The impact from offshore wind power forecast errors of up to 2000 MW on system generation costs, emission costs, dispatch-down of wind, number of startups and system marginal price are analysed. The main findings of this research are an increase in system marginal prices of approximately 1% for every percentage point rise in the offshore wind power forecast error regardless of the average forecast error sign. If offshore wind power generates less than forecasted (13%) generation costs and system marginal prices increases by 10%. However, if offshore wind power generates more than forecasted (4%) the generation costs decrease yet the system marginal prices increase by 3%. The dispatch down of large quantities of wind power highlights the need for flexible interconnector capacity. From a system operator's perspective it is more beneficial when scheduling wind ahead of the trading period to forecast less wind than will be generated. © 2014 Elsevier Ltd. All rights reserved.
Keywords: Offshore wind Electricity markets Forecast error Scheduling Dispatch
1. Introduction The European Union (EU) has developed a long-term framework for reducing greenhouse gas emissions by 80%e95% by 2050 compared to 1990 levels [1]. This framework aims to help the EU become a competitive low carbon economy by 2050 by setting policy plans in areas such as transport, energy and climate change. The power sector is the most significant for achieving the 2050 targets as the EU roadmap specifies that carbon dioxide (CO2) emissions from the power sector are to be almost completely eliminated by 2050. The Republic of Ireland and Northern Ireland have developed national roadmaps for achieving the EU 2050 targets [2,3]. They estimate renewable resources could provide 74% and 60% of the electricity generated in the Republic of Ireland and Northern Ireland respectively. In both countries the renewable electricity comes mostly from onshore and offshore wind power. The total installed wind power in the Republic of Ireland will be 12,000 MW consisting of 6000 MW offshore and 6000 MW onshore. In Northern Ireland the total installed wind power by 2050 will be 2800 MW with 1200 MW offshore and 1600 MW onshore [2,3]. The all-island generating capacity for 2050 will be * Corresponding author. E-mail addresses: [email protected] (P. Higgins), [email protected] (A.M. Foley), [email protected] (R. Douglas), [email protected] (K. Li).
24,942 MW with the offshore wind in the Republic of Ireland and Northern Ireland making up 24% and 5% of the generating capacity respectively. The disadvantage of wind power is that it is variable and stochastic by nature which makes the role of the system operator (SO) even more challenging. Studies have shown that the inclusion of significant levels of wind generation into an electricity system can have major impacts on the system security, scheduling and dispatching of units [4,5]. Connolly et al. [6] concluded the ramping capabilities of the generating units are crucial to the deployed of high wind generation. The operating reserve and flexibility of the power system were found to be crucial to the development of wind energy in electricity markets [7,8]. Hong et al. [9] concluded that to accommodate the increasing amount of wind energy onto the Jiangsu's power system more flexible gas generating units are required rather than coal and nuclear power plants. Inflexible power systems with large penetrations of wind energy have been shown to have major problems with wind curtailment [10]. The variability of the wind power and responses of conventional generation at short notice is crucial to system balancing and understanding of high levels of renewable generation that a grid in the future can incorporate [11]. The SOs of the Single Electricity Market (SEM) in the Republic of Ireland and Northern Ireland released statistics on the wind forecast accuracy for the month of December 2012 [12]. The report highlighted a maximum and minimum
http://dx.doi.org/10.1016/j.energy.2014.06.037 0360-5442/© 2014 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
2
P. Higgins et al. / Energy xxx (2014) 1e11
forecast error of þ798 MW and 634 MW for a total installed wind capacity of 2088 MW. A maximum normalised mean absolute error of 48.8% and average 12.2% were experienced. The large planned wind capacity for 2050 will consequently mean wind power forecasting will become a critical power system planning tool. Onshore wind power forecasting techniques have improved dramatically and continue to advance but Offshore Wind Power (OWP) forecasting is more difficult due to limited datasets and knowledge [13]. As offshore wind generation increases the variability will become more pronounced and better planning techniques and forecasting will be required [14]. The accuracy of wind forecast has been shown to be considerably lower over an entire transmission zone rather than over a single zone and that averaging over wider areas increases forecast accuracy [15,16]. The wider spatial spread of wind farms and increased offshore wind generation has been found to be the main factors in the reduction of wind curtailment [17]. In Ref. [18] the impact of Ireland's 2020 wind energy targets with varying limits for wind curtailment was examined and it was shown that approximately 7%e14% of wind production was not dispatched. It was also shown that increases in wind forecast accuracy also resulted in reductions in wind curtailment but provided no improvement in emissions production [19]. A study of the SEM with 6000 MW of installed wind capacity stated ‘improving wind forecasting will lead to relatively small savings in system costs on a percentage basis but may account for millions of euros’ [20]. For a system like the SEM in 2050 with over 14 GW of wind power, wind forecasting could have a substantial impact on the operation of the system. Studies have highlighted the interconnector capacity between Great Britain and Ireland as critical during moments of high wind and low load [17,21,22]. Tuohy et al. [23] included Monte Carlo simulations for wind and load uncertainties and found that the mid-merit, peaking units and the interconnections were the most affected components of the electricity system. The method implemented in Ref. [23] involved analysing the difference between a perfect foresighted model and a stochastic optimisation model. Other studies with similar techniques have determined that a stochastic optimisation for an electricity system produces an increase in production costs in comparison to a model with perfect foresight [24,25]. The difference with this research is the impact of the magnitude of the OWP forecast error is being analysed not the optimising technique. The OWP forecast error not only affects the operation and management of the electricity system but also the financial aspect. In Ref. [26] wind power predictability as an investment factor for selecting onshore wind farm sites was investigated and it was found that predictability can play an important role in the operation and maintenance of offshore wind farms due to improved availability from reduced downtime periods. In Ref. [27] the financial profits of individual generating units are also affected by increased wind generation. An increase of wind energy on the system pushes more thermal/conventional generation off the system. This reduces the System Marginal Price (SMP) and therefore the potential profits for conventional plants. This new wind generation also requires the existing plants to operate more flexible which can increase running costs and downtime for maintenance as well as shortening the plant life. The increased pressure from conventional generators running with variability and the impact offshore wind forecast error has on the SMP and merit order means some conventional generators could find it difficult to survive on unpredictable future SMPs. This research shows that a considerable amount of work has been performed on the impacts of increasing wind generation on electricity systems, however, only a small amount has analysed the impact of OWP forecast error. The novelty of this work is the
analysis of the impact of offshore wind forecast error on an electricity system with over 60% of generating capacity originating from renewable energy resources. This paper is structured in five sections. Section 1 is the introduction. Section 2 presents the test system and describes the methodology used to investigate the impact of OWP forecast error in the SEM in 2050. Section 3 presents the results of the analysis. Section 4 discusses the impacts of OWP forecast error in the SEM in 2050 and conclusions. 2. Test system and methodology 2.1. Single electricity market (SEM) The SEM is a mandatory all-island wholesale pool market through which generators and suppliers trade electricity. The market operates over three trading periods: before, intra-day and after. The before trading period consists of each generating unit bidding their commercial offer data and technical offer data for each half hour interval in the intra-day. The commercial offer data and technical offer data contain price/quantity bids and no load costs. The SOs produce the reserve constrained unit commitment (RCUC) schedule based on forecasted wind generation, price/ quantity pairs from all generating units and the SEM system stability requirements. The difference between an unconstrained unit commitment and RCUC is the inclusion of transmission constraints. From the RCUC schedule the SOs inform each generating unit of their required generation and operating hours before the start of intra-day trading. The intra-day period involves the real time application of the generation schedule. Both SOs implement the schedule and are responsible for delivering an efficient operation of the wholesale power market. The SO's continuously analyse the wind forecast, generating capacity and system demand requirements prior to the trading period based on updates to maintain system stability. The four days after the intra-day the Single Energy Market Operator (SEMO) calculates the SMP for each trading period during the intra-day. These four days are known as the after period. This SMP is the price applicable to both generators and suppliers receiving/making payments for electricity generated/ used [28]. 2.2. PLEXOS model The majority of research for modelling unit commitment and dispatch scheduling of electricity systems have used either deterministic or stochastic optimisation techniques [18e20,23,25,29]. The main difference between both techniques is that deterministic modelling with perfect foresight ‘may provide lowest cost solutions for system dispatch but also may provide unrealistic results’ [30], whereas stochastic optimisations mimic real power systems dispatches. For this research a mixed integer, stochastic optimisation will be applied using Energy Exemplar's PLEXOS for Power Systems version 6.208R03 with the Xpress optimiser [31]. Xpress was set with mixed integer programming at a relative gap of 0.05%. Xpress aims to minimise the objective function, shown in Equation (1) which is conditional of a number of constraints:
min
X
cjt :Ujt þ njt :Vjt þ mjt :Pjt þ vl:uset þ vl:usrt
(1)
t2T
where t indexes time periods in chronological order t ¼ 1, …, T, cjt is the start cost of unit j in period t, Ujt is a binary quantity representing if unit j has started the period before t, njt is the no load cost of unit j, Vjt is a binary quantity representing the generating status of unit j, mjt is the production cost of unit j, Pjt is the power output of
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
P. Higgins et al. / Energy xxx (2014) 1e11
unit j, vl is the penalty for loss load, uset is the unserved energy of unit j and usrt is the reserve energy not met by unit j. The objective function is subjected to constraints such as maximum generation, minimum stable generation, minimum up and down times and ramp rates. PLEXOS requires generating/ transmission parameters such as maximum and minimum ramp up/down time, minimum stable level, start-up cost, forced outage rate, wind input data, operational and maintenance costs and fuel costs to populate the objective function. The model optimises each of the 365 days in 2050 at 1 h intervals. A study found the time intervals can have a significant effect on the results [32]. A small time interval of five to 15 min is required when analysing the ramping of the generation system and can have an impact on the generation costs. The issue with smaller time intervals is the increase in computational costs, a 5 min interval had a run time of 70 h whereas the 1 h interval had a run time of 3 h. For this research the 1 h interval is used as there are a large number of scenarios to be implemented and the model should still provide realistic results that represent the SEM dispatch scheduling. 2.3. Replicating SEM activities in PLEXOS The test system must mimic the SEM operation to fully investigate the impact OWP forecast error has on the wholesale market, as seen in Fig 1. Therefore a ‘day ahead’ model is required to produce a dispatch schedule similar to the RCUC of the before period in the SEM. The day-ahead model has a 24 h look ahead and maintenance outages included. The 24 h look ahead provides the model with enough time to develop an optimum schedule including the maintenance outages. The forecasted wind with no errors is included in this model. A ‘real time’ model incorporates the OWP forecast error and the dispatch schedule from the day ahead model which represents the intra-day trading period. The look ahead of the real time model was shortened to 3 h to replicate the short notice of OWP forecasting errors. A 3 h look ahead was found to be the optimum frequency of rescheduling the system due to changes in load and wind forecasts for a stochastic optimisation model [23]. Maintenance and forced outages are applied to the real time model to replicate the realistic short notice plant outages. The wind forecast error scenarios are applied to the real time model to investigate the impact OWP forecast error has on the market operation and management of the SEM. The after trading period of the SEM is not modelled for this
3
research. Further research projects will involve the development of the after trading period. 2.4. Test system The test system was built using 2011 as the base case (SEM_2011) [33]. The SEM_2011 is the all-island market consisting of 88 different generating units. The test system includes two interconnectors to the British Electricity Trading and Transmission Arrangement (BETTA) market and a single gas generating unit to represent the BETTA market. Interconnection to the BETTA has a combined capacity of 900 MW which is increased to 2000 MW for 2050 [3]. 2.4.1. Demand and generation portfolios The yearly demands for electricity in the Republic of Ireland and Northern Ireland are expected to increase from their 2010 levels of 28,000 GWh and 9000 GWh respectively to 48,000 GWh and 12,000 GWh in 2050 [2,3]. The required generation portfolio to meet the 2050 electricity demand is shown in Table 1. The total installed 2050 generating capacities for the Republic of Ireland and Northern Ireland are approximately 20,800 MW and 4400 MW respectively. A number of existing generators are expected to be decommissioned over the next few years. The Great Island units with a combined generating capacity of 212 MW, all the Tarbert units (595 MW) and Ballylumford gas units (510 MW) are all to be decommissioned. It is assumed that Kilroot coal generation will be decommissioned by 2050 as it will have to comply with the Industrial Emissions Directive [34]. The peat, oil and distillate generating units in both jurisdictions and the coal generating units in Northern Ireland are expected to be decommissioned by 2050 [3]. The introduction of carbon capture storage (CCS) technology should maintain a small amount of coal generation in 2050 [35]. Therefore it is assumed that gas will be the primary fossil fuel generation in 2050. 2.4.2. Onshore and offshore wind generation The all-island wind was modelled by separating the onshore wind into 13 regions and the offshore wind into seven regions [36], as shown in Fig 2. Each onshore wind region is modelled with a max capacity and an excel file containing a percentage value for every hour period in a year. The wind generation for every hour
Fig. 1. Test system structure.
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
4
P. Higgins et al. / Energy xxx (2014) 1e11
Table 1 2050 SEM generation portfolio. Max capacity (MW) Fuel source
All-island (2011)
Republic of Ireland (ROI-2050)
Northern Ireland (NI-2050)
Total (2050)
Offshore wind Onshore wind Wave/tidal Hydro/PHESa Gas CCGTb Gas OCGTc Coal (CCS)d Total
e 1935 e 508 4072 720 1331 8566
6000 6000 1080 508 5201 1185 850 20,824
1200 1600 200 10 900 500 e 4410
7200 7600 1280 518 6101 1685 850 25,234
a b c d
Pumped Hydro Energy Storage. Combined Cycle Gas Turbine. Open Cycle Gas Turbine. Carbon Capture Storage.
period in the year is a combination of the percentage file and the max capacity. The 2050 max capacity for each region is determined with the Grid 25 strategy in consideration [37]. The total wind for 2050 in the Republic of Ireland is expected to be 12,000 MW consisting of 6000 MW of onshore and 6000 MW of offshore. Northern Ireland will have 2800 MW of wind with 1600 MW of onshore wind and 1200 MW of offshore wind. Wind profiles for offshore wind regions are more difficult to obtain as there is a lack of data. Therefore the offshore files were manipulated from the onshore regions. Research showed the offshore wind regions of the east coast of the Island have a 1 h time lag to the respective onshore wind regions and the spatial correlation of neighbouring wind regions was found to be in the range of 0.94e0.97 [18]. The offshore wind resource is more powerful than onshore and as a result the capacity factor of offshore wind turbines is approximately 40%. Taking these factors into consideration the offshore wind data files were time and correlation adjusted accordingly using the same procedures as applied in Ref. [18].
2.4.3. Fuel and carbon price Forecasted fuel prices were obtained from an EU world energy technology outlook to 2050 [38] and are shown in Table 2. The price of coal does not experience the same projected increases as gas as it is envisaged that coal could return as an important source of electricity and will be increasingly converted using new advanced technologies [38]. The CO2 price of V234/tCO2 is based on figures from the United Kingdom government for 2050 [39]. The CO2 price is added to the model through a tax on the price of fuel. The sensitivity of the CO2 price is investigated by applying two additional prices of V117/tCO2 and V351/tCO2 to the system. 2.4.4. Transmission constraints The SOs have implemented a number of constraints to ensure the efficient and secure operation of the transmission system [40]. The main transmission constraint influencing OWP generation is the System Non-Synchronous Penetration (SNSP) limit. The SNSP is a measure of the non-synchronous generation on the system at any time. It is a ratio of non-synchronous generation (onshore wind, offshore wind, wave and tidal) and high voltage direct current (HVDC) imports to demand and HVDC exports. Currently the SOs has set the System Non Synchronous Penetration (SNSP) at 50%, this value is expected to rise to between 70% and 80% by 2020 [18]. For this 2050 model the SNSP is set at 75%. The formula for SNSP is
SNSP ¼
Non synchronous generation þ HVDC Imports % ¼ 75% System Demand þ HVDC Exports (2)
2.4.5. OWP forecast error scenarios The following method was implemented to analyse the magnitude and variance of the OWP forecast error. EirGrid publicly provides the wind generation and respective wind forecast data for every 15 min interval from 2010 onwards [41]. This wind data is recorded from the SEM which consists of onshore wind generation only. OWP forecasted and generation data is difficult to source as it is not publicly available. Therefore it will be assumed that the onshore wind forecast error from the EirGrid data will be similar to the OWP forecast error. From this data the magnitude of the wind forecast error was determined using the following formula.
Wind Forecast Error ¼
Wg Wf % Wg
(3)
where Wg is the generated wind (MW) and Wf is the wind forecast (MW). Fig 3 shows the distribution of the wind forecast error of wind generated in the SEM in 2012. The figure illustrates the frequency of each forecast error and which are the most dominate. It can be noted from the figure that the most dominate wind forecast error occurs between ±10% of the wind generation. Since the 60% of forecast error occurs between ±20% of wind generation eight scenarios ranging from 20% to þ20% will be analysed, as seen in Table 4. For each scenario a constant OWP forecast error is applied to the offshore wind generation in the real time model. The purpose of this work is to analyse the OWP forecast error therefore onshore wind generation will have no forecast error applied to it.
Table 2 Fuel prices.
Fig. 2. SEM wind regions.
Fuel type
Price (V/GJ)
Coal Gas
3.29 12.45
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
P. Higgins et al. / Energy xxx (2014) 1e11
5
Fig. 3. Onshore wind forecast error in the SEM (2012).
A Dutch study [42] using weather data and multi-turbine power curves determined the actual and forecasted wind power for 12 GW of installed onshore and offshore wind capacity. The research produced the normalised standard deviation of the wind power forecast error and highlighted that the forecast error decreased to approximately 5% for 1 h prior to operation and 9% for 6 h prior to operation. For this research the look ahead or forecasted error lag was fixed at 3 h, as discussed in Section 2.3. The wind generated and forecasted data for 2010, 2011 and 2012 was used to provide three forecast error profiles that include variance. The forecast error was found for every period in each year and converted into a percentage. The percentage forecast error file was multiplied by the wind forecast data to provide three generation profile with forecast error and variance, as shown by the 24 h sample in Fig 4. The new generation profiles were implemented in the real time model and were used to examine the impact of OWP forecast error variance. The variances and average errors of three profiles are listed in Table 3. The scenarios for this analysis are summarised in Table 4. The 2%, 5%, 10% and 20% scenarios will be referred to as deficit forecast errors scenarios and the 2%, 5%, 10% and 20% scenarios will be referred to as surplus forecast errors scenarios. The deficit and surplus scenarios are used to analyse the impact from the OWP forecast error magnitude. The 13%, 4% and 7% are used to analyse the impact from the OWP forecast error variance. The ‘Base e V117’ and ‘Base e V351’ scenarios are used to analyse the price of carbon impact.
possible to look at the impact of the variance of OWP forecast error and the sensitivity of the price of carbon. 3.1. Generation Fig 5 shows the generation comparison between all the scenarios in terms of GWh. The figure illustrates the annual generation for each fuel type over the year, from the 1st of January 2050 to 31st of December 2050. The major generating units in the SEM in 2050 are onshore wind, OWP and gas generation, with all three combining to generate over 80% of the annual system generation. In the SEM the major generation changes due to OWP forecast error are gas and onshore wind. The more efficient Combined Cycle Gas Turbines (CCGTs) are the most effected due to the offshore wind forecast error. The usage of CCGTs instead of Open Cycle Gas Turbines (OCGTs) or Combined Heat Plant (CHP) to match the change Table 4 List of scenarios. Scenario
Details
20%
A constant 20% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 10% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 5% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 2% error applied to the forecasted data, no variance, carbon price of V234/tCO2 No error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 2% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 5% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 10% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A constant 20% error applied to the forecasted data, no variance, carbon price of V234/tCO2 A varying error with an average of 13% applied to the forecasted data, variance of 0.09, carbon price of V234/tCO2 A varying error with an average of 4% applied to the forecasted data, variance of 0.146, carbon price of V234/tCO2 A varying error with an average of 7% applied to the forecasted data, variance of 0.15, carbon price of V234/tCO2 No error applied to the forecasted data, no variance, carbon price of V117/tCO2 No error applied to the forecasted data, no variance, carbon price of V351/tCO2
10% 5% 2% Base 2%
3. Results and analysis 5%
A comparison of the results was performed and is presented in the following tables and figures. The impact of the magnitude of OWP forecast error has on the electricity market was compared over a range of different parameters such as: unit generation, generation costs, net interchange between the SEM and the BETTA, emission production and costs, SMP of electricity, unit operational time and dispatch down of wind. From this analysis it was then
20% 13%
4%
7%
Table 3 Average OWP forecast error and variance.
Average error Variance
10%
2010
2011
2012
13% 0.09
4% 0.146
7% 0.15
Base e V117 Base e V351
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
6
P. Higgins et al. / Energy xxx (2014) 1e11
Fig. 4. Forecast error with variance.
Fig. 5. OWP forecast error on generation.
in OWP forecast error results in lower generation costs. For surplus scenarios when the offshore wind generation is greater than forecasted, onshore wind and gas are directed to curtail/reduce generation and less is imported from the BETTA. The majority of the dispatching down of non-synchronous generation is onshore and offshore wind as the renewable generation for Northern Ireland (NI Renewables) and the Republic of Ireland (ROI Renewables) remains relatively constant. For the deficit scenarios where offshore wind generation is less than forecasted, an increase in onshore wind generation and gas generation is experienced. Approximately 55% of the increase in gas generation is provided by CCGTs and 43% by OCGTs. The gas peaking plants of CHP and steam turbines (ST) are the most affected gas generators in terms of start-ups and shut downs. The increase in onshore wind generation is due to the curtailing effects of the SNSP limit. Table 5 shows the impact the OWP forecast error has on the net interchange between the SEM and the BETTA. The net interchange is the HVDC exports from the SEM less the HVDC imports to the SEM. A positive value represents exports from the SEM to the BETTA and a negative represents an import from the BETTA. As expected there is a close correlation between the increase in surplus offshore wind forecast error and an increase in exports to the BETTA. 3.2. Generation costs Fig 6 illustrates the impact on generation costs due to the OWP forecast error. The offshore wind, onshore wind and both NI and ROI renewables are modelled as a zero generation cost as they are
must run units and have to be dispatched first. When OWP forecast error is a surplus the generation costs are cheaper as more OWP is on the system. The least efficient thermal generators that were dispatched in the base scenarios are ramped down or switched off. The majority of generation costs are from gas generators as over 40% of the generated electricity is from gas and approximately 50% is from renewable energy sources. Similar to 2014 the SEM system costs and SMPs in 2050 are still driven by the price of gas. The CCGT plants provide the majority of generation changes to balance the OWP forecast error and as a result it is their generation costs that affect the system generation costs the most. When less offshore wind is generated than forecasted, it is more economical to keep the coal generators online for longer periods than shut them down as per the day ahead schedule. The number of coal generator startups decrease as a result longer operational periods.
Table 5 OWP forecast error on the net interchange. Scenario
Net interchange (GWh)
20% 10% 5% 2% Base 2% 5% 10% 20%
11,403 11,790 11,981 12,086 12,137 12,197 12,269 12,390 12,597
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
P. Higgins et al. / Energy xxx (2014) 1e11
7
Fig. 6. OWP forecast error on generation costs.
The difference in generation costs due to the offshore wind forecast error is highlighted in Table 6. The values in the table show an OWP forecast error could increase the generation costs of the entire system by V193 million or an increase of 4%. The deficit scenarios produce a greater change in generation costs than the surplus scenarios. This is due to less efficient and more expensive coal and gas generators coming online to balance the loss from offshore wind generation. The generation costs change by 0.2% or V9 million for every percentage change in the OWP forecast error. 3.3. Emission production and costs Fig 7 shows the changes in emission production for the different OWP forecast scenarios. The deficit OWP forecast error scenarios produce more emissions than the emissions saved from the similar surplus OWP forecast scenarios. This is due to less efficient plants running in the deficit scenarios and being shut down in the surplus scenarios as discussed in section 3.1. The V234/tCO2 price of carbon results in a large emissions costs of approximately V2.65 trillion for the base scenario. The 20% offshore wind forecast error results in an increase of V0.17 trillion. Similar to the generation costs this highlights the need to reduce the OWP forecast error as much as possible. A comparison of the 2011 and 2050 emissions production shows a 50% reduction from 0.466 t/MWh to 0.185 t/MWh but a 600% increase in emissions cost from V6.06/MWh to V37/MWh. The significant impact the price of carbon has on emission costs is discussed in Section 3.7. 3.4. System marginal price Fig 8 shows the change in the yearly average SMP due to the OWP forecast errors. It highlights the yearly average SMP with the
Table 6 Difference in generation costs from base scenario. Scenario 20% 10% 5% 2% Base 2% 5% 10% 20%
Change in generation costs (V million) 193.0 88.2 47.0 21.5 e 16.5 45.1 87.3 167.5
price cap and without the price cap. The price cap is a constraint on the system which limits the maximum SMP that can occur and is set at V1000/MWh. Usually the price cap is applied when generation is less than the system demand. Before such an instance the SO implements operating reserve to meet the difference demand and generation. A recent report showed the price cap was reached once during the year 2013 [43]. When the price cap is modelled in the 2050 system it is reached eleven times in the base scenario and over one hundred times in the 20% scenario. The significant increase in the price cap occurrences for the 20% forecast error scenario is due to the large difference in generation and forecast, and most importantly the short time window restricting the scheduling of extra generation. The test system has produced an SMP for 8766 h periods and only 128 of those periods hit the price cap. Fig. 8 illustrates the impact the 1% (128 periods) of price capped SMP values can have on the yearly average SMP for the 20% offshore wind forecast error. If the price capped periods were equal to the average SMP a significant drop occurs for all the deficit scenarios. The most affected is the 20% scenario with a reduction of V11.44/MWh or 6%. The problem is not the price cap but the transmission and reserve constraints. The transmission constraints are not providing enough reserve to ensure security of supply and as a result the SMP price hits the price cap. Similar to the finding of Ref. [18] the transmission constraints need to be addressed as the 2011 set will not be suitable for 2050 when large quantities of nonsynchronous generation (onshore wind, offshore wind and wave) are present. It was found that the exclusion of the price cap influences the þ20% OWP forecast error results in a decrease of V3/MWh or 2% from the base scenario in the annual average SMP. Fig. 8 shows an increase of 0.1% for every percentage point rise in the surplus forecast error scenarios. 3.5. Dispatch-down of wind The reduction of electricity generated from wind can be described as the dispatch-down of wind generation. Wind that is dispatch-down can be due to constraints in the network or due to curtailment issues. The EirGrid ‘policy for scheduling and dispatch decisions’ prioritises the order of dispatching generating units [44]. EirGrid currently dispatches generators in the following order after the closure of the market gate, interconnector, peat, hydro and then wind. Fig 9 shows the monthly figures for the dispatch-down wind and the corresponding wind generation. As expected a close
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
8
P. Higgins et al. / Energy xxx (2014) 1e11
Fig. 7. Offshore wind forecast error impact on emissions production and costs.
correlation between the winter months (OctobereMarch) with high wind generation and the highest dispatch-down values was experienced. In 2011 80% of dispatch-down on the SEM was from the curtailment effects of the SNSP limit [45]. In 2050, with 2000 MW of interconnector capacity, the majority of the dispatchdown wind is due the SNSP and the lack of interconnector capacity. The impact of the OWP forecast error on the dispatch-down wind generation is shown in Fig 10. The base model has 10,196 GWh of dispatch down wind, 7245 GWh of onshore wind and 2951 GWh of offshore wind. Under the current renewable energy feed in tariffs this unused wind energy would still have to be paid for therefore this is clearly an undesirable amount of dispatched down wind energy. The surplus þ20% scenario generates an extra 4195 GWh of offshore wind. Approximately 43% (1797 GWh) of this extra wind generation is allowed onto the system while 57% (2398 GWh) is added to the dispatched down of wind energy. In the deficit scenarios when offshore wind generation is less than expected, more onshore wind generation is allowed onto the system to replace the offshore wind generation. For the deficit 20% scenario 4985 GWh of offshore wind is missing from the system resulting in an increase of 2102 GWh of onshore wind generation and a decrease of 2883 GWh in dispatched down wind. The SNSP limit that is restricting the amount of non-synchronous generation on the system is causing more than 50% of the extra offshore wind generation in the surplus scenarios to be curtailed. In
the deficit scenarios the SNSP limit allows more onshore wind generation onto the system to compensate for the offshore wind forecasting error. 3.6. Impact of wind power forecast error variance When variance is included to a deficit 7% OWP forecast error scenario 13.97 million tonnes of CO2 are produced at a cost of V2.79 trillion, as shown in Fig 11. This emission production is 2% more than a deficit 10% offshore wind forecast error which produces 13.77 million tonnes of CO2 at a cost of V2.75 trillion. As discussed in Section 3.3 when the error of the OWP forecast is smaller the emission production should reduce however when variance is included the emission production increases. The offshore wind generation is approximately the same but the introduction of variance results in an increase in start-ups for gas generators and longer operational time for coal generators which increases the emissions production and costs. An increase in emissions production and costs are seen for both deficit and surplus scenarios that include variance as the number of startups for gas generators and operational hours for coal generators increases. The same trend is recorded for the SMP when variance is applied to the forecast error. The introduction of variance into the OWP forecast error causes a significant impact on the SMP as seen in
Fig. 8. Offshore wind forecast error impact of SMP.
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
P. Higgins et al. / Energy xxx (2014) 1e11
9
Fig. 9. Monthly dispatch-down wind and wind generation for 2050 base scenario.
Fig. 10. Offshore wind forecast error on the dispatch down of wind.
Fig. 12. The surplus forecast error (4%) has more offshore wind generation and therefore should result in a decrease in the SMP. However, the variance results in an increase in the SMP. The introduction of the variance increases the amount of price cap instances and this is the cause for the significantly larger SMP (with Price Cap) figures.
Fig. 12 shows an increase of approximately 1% for every percentage point rise in the forecast error scenarios no matter the sign of the average forecast error. The inclusion of variance results in extra shut downs, start-ups and generating of expensive gas peaking plants for the positive and negative OWP forecast error scenarios. The additional operating of these peaking plants results
Fig. 11. OWP forecast error impact on emission production and costs (variance).
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
10
P. Higgins et al. / Energy xxx (2014) 1e11
Fig. 12. OWP forecast error impact on SMP (variance).
in the SMP increases. The significant increases on the base scenario SMP shows the need for an accurate OWP forecast when such a large percentage of the generation portfolio consists of offshore wind energy. 3.7. Carbon price sensitivity A UK government report [39] provided a low, medium and high price of carbon for 2050 as V117/tCO2, V234/tCO2 and V351/tCO2 respectively. The impact from the price of carbon on generation costs, emissions costs and SMP, for the no forecast error scenario, are listed in Table 7. The large carbon price variations produce differences in the generation and emissions costs of approximately V1 billion and V1 trillion respectively. Table 8 shows the effect the OWP forecast has on the SMP with the different carbon prices. The lower the carbon price the greater the impact on SMP from OWP forecast error. 4. Discussion and conclusion The analysis shows that OWP forecast error has the potential to have significant operation and management impacts on the SEM in 2050 in terms of SMP, generation costs, emissions costs, ramping, operational hours and dispatching of wind. The OWP capacity modelled for 2050 accounted for 30% of the total installed capacity. This significant OWP penetration resulted in periods throughout the year where the OWP forecast error experienced a difference of 2000 MW between forecasted and generated offshore wind power. The research showed the OWP forecast error caused small percentage changes to the SMP but considerable V/MWh changes. A percentage point change in the OWP forecast error resulted in a 0.1% variation to the SMP for all of the scenarios where no variance is included. The introduction of variance into the OWP forecast error produces different changes to the SMP. The surplus forecast error (4%) has more offshore wind generation and therefore should
Table 7 Carbon price sensitivity for base scenario with no forecast error.
result in a decrease in the SMP. However the variance results in an increase in the SMP as additional generation, start-ups and shut downs of gas peaking plants are required to balance the OWP forecast error. An SMP increase of approximately 1% for every percentage point rise in the forecast error scenarios was noted regardless of the average forecast error sign. This increase on the base scenario SMP shows the need for an accurate OWP forecast when such a large percentage of the generation portfolio consists of offshore wind energy. Further analysis of the operating reserve is required to reduce the impact of the price cap on the SMP. A comparison of the 2011 and 2050 emissions production shows a 50% reduction from 0.466 t/MWh to 0.185 t/MWh but a 600% increase in emissions cost from V6.06/MWh to V37/MWh. In 2050 the price of carbon is one of the biggest contributors to the SMP. The sensitivity analysis of the price of carbon shows the larger the price of carbon the less impact OWP forecast error has on SMP. All of the results in this paper are based on a carbon price of V234/tCO2. The V234/tCO2 price of carbon results in a large emission costs of V2.67 trillion for the base scenario. The surplus OWP forecast error scenarios produce a decrease in emissions production and cost. However, when variance is added to the scenario the emissions production and costs increase. This is due to an increase in coal generation which has a large CO2 emission per gigajoule of energy. Increased OWP forecast accuracy could allow more efficient units to be dispatched rather than fast ramping and inefficient peaking generators. The OWP forecast error has a significant impact on the onshore wind generation as the SNSP limit restricts the amount of nonsynchronous generation on the system. The base scenario with no forecast error dispatches down nearly 25% of the available wind generation. The dispatch down of large quantities of wind power in the base scenario highlights the need for additional interconnector capacity or energy storage. The lack of sufficient interconnector capacity and SNSP limit is resulting in less than 50% of the offshore wind forecast error generation being utilised. As highlighted by
Table 8 OWP forecast error with different carbon prices.
Carbon price
Generation costs (Vbillion) Emissions costs (Vtrillion) Annual average SMP
Price of carbon
V117/tCO2
V234/tCO2
V351/tCO2
V3.76 V1.40 V168.57
V4.70 V2.67 V190.46
V5.66 V3.99 V230.29
Base SMP 4% SMP 13% SMP
V117/tCO2
V234/tCO2
V351/tCO2
V168.57 V174.87 V186.34
V190.46 V195.93 V208.72
V230.29 V236.87 V246.81
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037
P. Higgins et al. / Energy xxx (2014) 1e11
Refs. [17,21e23] the flexibility of the interconnectors from Ireland is crucial to reducing the dispatched down wind. An increase in interconnector capacity should also reduce the impact of OWP forecast error on the operation and management of the electricity system. With variance included the OWP forecast error scenario of 4% resulted in a decrease in generation cost of V31.5 million or 0.67%, whereas the 13% scenario resulted in an increase of V160.3 million or 3.41%. Both of these scenarios show the different extremes that could occur depending on the sign of the average error of the OWP forecast. Therefore, similar to comments from Ref. [15], when scheduling ahead it is more beneficial to forecast less wind than will be generated (surplus scenario). From an SO's perspective it is more beneficial to develop a schedule with less wind generation than expected and during real time allow the extra wind from the forecast error onto the system while reducing thermal generation. Constraining thermal generators in such a way incurs constraint costs and future work is required to determine the full financial costs associated with constraining thermal generators. In conclusion this paper investigates the impacts of OWP forecast error on the operation and management of the SEM in 2050, which is a carbon constraint electricity market. Acknowledgements The authors acknowledge the support of Energy Exemplar through the academic license for PLEXOS, the Engineering and Physical Sciences Research Council for funding support and Edward McGarrigle from University College Cork for his assistance. References [1] European Commission. A roadmap for moving to a competitive low carbon economy in 2050. Communication from the commission to the European parliament, the council, the European economic and social committee and the committee of the regions 2011. COM; 2011. p. 112. final. [2] Sustainable Energy Authority Ireland. Smartgrid roadmap 2050; 2012. [3] Ricardo-AEA. Envisioning the future e considering energy in Northern Ireland to 2050; 2013. Ricardo-AEA/R/ED58435. [4] Gil E. Evaluating the impact of wind power uncertainty on power system adequacy. In: Proceedings of the PMAPS; 2012. Istanbul, Turkey. [5] Tuohy A, Denny E, Meibom P, Barth R, O'Malley M. Operating the Irish power system with increased levels of wind power. In: Proceedings of the IEEE Power and Energy Society General Meeting e Conversion and Delivery of Electrical Energy in the 21st Century; 2008. pp. 1e4. Pittsburgh, America. [6] Connolly D, Lund H, Mathiesen BV, Leahy M. Modelling the existing Irish energy-system to identify future energy costs and the maximum wind penetration feasible. Energy 2010;35:2164e73. [7] Lannoye E, Flynn D, O'Malley M. Evaluation of power system flexibility. Proc IEEE Trans Power Syst 2012;27(2):922e31. [8] Bjelic I, Rajakovic N, Cosic B, Duic N. Increasing wind power penetration into the existing Serbian energy system. Energy 2013;57:30e7. [9] Hong L, Lund H, Moller B. The importance of flexible power plant operation for Jiangsu's wind integration. Energy 2012;41:499e507. [10] Finn P, Fitzpatrick C, Connolly D, Leahy M, Relihan L. Facilitation of renewable electricity using price based appliance control in Ireland's electricity market. Energy 2011;36:2952e60. [11] Kubik ML, Coker PJ, Barlow JF, Hunt C. A study into the accuracy of using meteorological wind data to estimate turbine generation output. Renew Energy 2012;51:153e8. [12] EirGrid, SONI. Wind forecast accuracy statistics; 2012. Available from: http:// www.eirgrid.com/media/AllIslandWindForecastAccuracyReportDecember 2012.pdf. [13] Trombe P, Pinson P, Madsen H. A general probabilistic forecasting framework for offshore wind power fluctuations. Energies 2012;5(3):621e57. [14] Foley A, Leahy P, Marvuglia A, McKeogh E. Current methods and advances in forecasting of wind power generation. Renew Energy 2011;37:1e8. [15] Newbery D. Contracting for wind generation. Econ Energy Environ Policy 2011;1(2).
11
[16] Kiviluoma J, O'Malley M, Touhy A, Meibom P, Milligan M, Lange B, Holttinen H, Gibescu M. Impact of wind power on the unit commitment, operating reserves, and market design. In: Proceedings of the IEEE Power and Energy Society General Meeting; 2011. pp. 1e8. California, USA. [17] Foley A, Leahy P, McKeogh E. Wind energy integration and the Wales-Ireland interconnector. In: Proceedings of the IEEE PES/IAS Conference on Sustainable Alternative Energy; 2009. pp. 1e6. Valencia, Spain. [18] McGarrigle E, Deane P, Leahy P. How much wind energy will be curtailed on the 2020 Irish power system? Renew Energy 2013;55:544e53. [19] McGarrigle E, Leahy P. The value of accuracy in wind energy forecasts. In: Proceedings of the 12th International Conference on Environmental and Electrical Engineering; 2013. Wroclaw, Poland. [20] Meibom P, Barth R, Hasche B, Brand H, Weber C, O'Malley M. Stochastic optimization model to study the operational impacts of high wind penetrations in Ireland. Proc IEEE Trans Power Syst 2011;26(3):1367e79. [21] Denny E, Touhy A, Meibom P, Keane A, Flynn D, Mullane A, et al. The impact of increased interconnection on electricity systems with large penetrations of wind generation: a case study of Ireland and Great Britain. Energy Policy 2010;38:6946e54. [22] Foley AM, O'Gallachoir BP, McKeogh E, Milborrow D, Leahy P. Addressing the technical and market challenges to high wind power integration in Ireland. Renew Sustain Energy Rev 2013;19:692e703. [23] Tuohy A, Meibom P, Denny E, O'Malley M. Unit commitment for systems with significant wind penetration. Proc IEEE Trans Power Syst 2009;24(2): 592e601. [24] Meibom Barth R, Brand H, Hasche B, Swider D, Ravn H, Weber C. Wind variability management studies e all island grid study workstream B. Department of Enterprise, Trade and Investment; 2007. [25] Wang J, Botterud A, Miranda V, Monteiro C, Sheble G. Impact of wind power forecasting on unit commitment and dispatch. In: Proceedings of the 8th International Wind Integration Workshop; 2009. Bremen, Germany. [26] Girard R, Laquaine K, Kariniotakis G. Assessment of wind power predictability as a decision factor in the investment phase of wind farms. Appl Energy 2013;101:609e17. [27] Kubik M, Coker P, Hunt C. The role of conventional generation in managing variability. Energy Policy 2012;50:253e61. [28] Single Electricity Market Operator. Available from: http://www.sem-o.com/ Pages/default.aspx; 2012. [29] Denny E, O'Malley M. Quantifying the total net benefits of grid integrated wind. Proc IEEE Trans Power Syst 2007;22(2):605e15. [30] Deane P, McKeogh E, O'Gallachoir B. Deviation of robust storage targets for large scale pumped hydro energy storage using PLEXOS. In: Proceedings of the 2010 European PLEXOS User Meeting; 2010. London, United Kingdom. [31] FICO Xpress Optimiser. Available from: http://www.fico.com/en/Products/ DMTools/xpress-overview/Pages/Xpress-Optimizer.aspx; 2013. [32] Deane J, Drayton G, Gallachoir B. The impact of sub-hourly modelling in power systems with significant levels of renewable generation. Appl Energy 2014;113:152e8. [33] Commission for Energy Regulation (CER). Available from: http://www.cer.ie/; 2012. [34] EirGrid, SONI. All-island generation capacity statement 2013-2022. EirGrid and System Operator of Northern Ireland (SONI); 2012. [35] Poyry Energy Consulting. Options for low-carbon power sector flexibility to 2050; 2010. p. 655. [36] Redpoint Energy Ltd. Single electricity market operator (SEMO) forecast model of 2011-2012; 2011. SEM-11-072. [37] EirGrid. GRID 25 e a strategy for the development of Ireland's electricity grid for a sustainable and competitive future; 2012. [38] European Commission. World energy technology outlook to 2050; 2007. [39] Department of Energy and Climate Change. Guidance on estimating carbon values beyond 2050: an interim approach. Available from: https://www.gov. uk/government/uploads/system/uploads/attachment_data/file/48108/1_2010 0120165619_e____carbonvaluesbeyond2050.pdf. [accessed 10.05.13]. [40] EirGrid, SONI. Transmission constraint groups. EirGrid and System Operator of Northern Ireland (SONI); 2012. V1.4. [41] EirGrid, SONI. Wind generated and forecasted data. Available from: http:// www.eirgrid.com/. [42] Gibescu M, Kling W, Ummels B. Case study for the integration of 12 GW wind power in the Dutch power system by 2020. In: Proceedings of the CIGRE/IEEE PES joint symposium on Integration of wind-scale renewable resources into the power delivery system; 2009. USA. [43] Commission for Energy Regulation (CER). Trading and settlement code e policy parameters 2014; 2013. SEM-13-053. [44] EirGrid. Policy for implementing scheduling and dispatch decisions; 2011. SEM-11-062. [45] EirGrid, SONI. 2011 Curtailment report; 2013. Available from: http://www. eirgrid.com/media/2011_Curtailment_Report.pdf.
Please cite this article in press as: Higgins P, et al., Impact of offshore wind power forecast error in a carbon constraint electricity market, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.06.037