Techno-economic assessment of forecasting and communication on centralized voltage control with high PV penetration

Electric Power Systems Research 151 (2017) 338–347

Contents lists available at ScienceDirect

Electric Power Systems Research journal homepage: www.elsevier.com/locate/epsr

Techno-economic assessment of forecasting and communication on centralized voltage control with high PV penetration Luis González-Sotres ∗ , Pablo Frías, Carlos Mateo Institute for Research in Technology (IIT), Comillas Pontifical University, Madrid, Spain

a r t i c l e

i n f o

Article history: Received 18 November 2016 Received in revised form 22 May 2017 Accepted 26 May 2017 Keywords: Techno-economic analysis Centralized voltage control Solar PV Forecasting Communication

a b s t r a c t The current increase of solar generation in the power system, especially at low-voltage (LV) and mediumvoltage (MV) levels, poses a great challenge from the distribution system operation point of view. Centralized voltage control schemes have been presented in the literature as a promising solution to deal with the voltage violations that may be caused by these high solar PV penetration levels. However, this approach relies on the accuracy of forecasted data to optimize the set-points and on the effectiveness of communication infrastructures to send them from the central unit to the inverters at the PV panels. This article presents a novel methodology developed in the European project SUSTAINABLE to assess the impact of forecasting and communication on centralized voltage control. This methodology has been applied trough simulation in two Portuguese networks used for the demonstration activities of the project, one in LV and another in MV. The results show that the accuracy of the data used to calculate the set-points as well as the time interval used for their transmission have a significant impact in these applications and highlight the importance of dedicating resources to improve both forecasting tools and communication systems. © 2017 Elsevier B.V. All rights reserved.

1. Introduction In the last 10 years, the integration of solar PV in power systems all over the world has increased significantly. By 2019 the total installed capacity is expected to be around 450 GW according to market forecasts [1]. The vast majority of these PV facilities are installed in low-voltage (LV) and medium-voltage (MV) networks, which leads to new challenges for the integration of Renewable Energy Sources (RES) as Distributed Generation (DG), such as reverse power flows, voltage violations or congestions in distribution grids [2]. Additionally, high PV penetration levels may cause power and voltage fluctuations due to cloud shadows, and even an increase on energy losses when reverse power flows are large enough [3,4]. There are different alternatives to deal with these issues, which can be performed by Distribution System Operators (DSOs), PV generators, or as interaction of both agents. A description of a wide set of available technical solutions for LV and MV networks can be found in Ref. [5], where active and reactive power control of PV units were identified as high effectiveness solutions to mitigate these problems. The same conclusion was obtained in

∗ Corresponding author. E-mail address: [email protected] (L. González-Sotres). http://dx.doi.org/10.1016/j.epsr.2017.05.034 0378-7796/© 2017 Elsevier B.V. All rights reserved.

Ref. [6], where authors investigated the use of smart inverters with high PV penetration under real operation conditions. However, the deployment of these solutions highly depends on the regulatory framework, which should provide adequate incentives to establish a more efficient and coordinated power system [7]. With regard to voltage violations, several authors have analyzed the impact of different voltage control strategies. For instance, authors in Ref. [8] assessed from a qualitative point of view the technical and market implications of current trends for voltage control, which were classified into four categories: decentralized autonomous control, decentralized peer-to-peer coordination, hierarchical control, and centralized control. In the decentralized autonomous control, the inverters determine the control actions based on local algorithms from each individual DG unit, and no communication between devices is required. In the decentralized peer-to-peer coordination, the control devices can exchange information and coordinate with each other to achieve a more efficient system operation. The same idea is used in the hierarchical control, but in this case different levels of coordination are defined, for instance at each substation. Finally, in the centralized control all the control actions are obtained from the same control center, which allows optimizing the system operation from a regional perspective.

L. González-Sotres et al. / Electric Power Systems Research 151 (2017) 338–347

In Ref. [9], autonomous and coordinated voltage control strategies based on reactive power adjustment at the Point of Common Coupling (PCC) were analyzed in LV networks, where authors concluded that this mechanism alone is not enough to fully resolve the overvoltage problems caused by DG due to the high R/X ratio at this voltage level. In Ref. [10] different active power curtailment techniques were analyzed, where authors concluded that the amount of curtailed energy was highly affected by the curtailment method used. The combination of real and reactive power control of PV inverters was investigated in Ref. [11], which results showed that the inclusion of reactive power compensation helps to reduce curtailment losses that may be derived from the implementation of an active power control strategy alone. This conclusion was also envisaged in Ref. [10], although in this case authors also noticed that the reactive power control is less effective where the major overvoltage problems are expected to arise, for instance in rural networks where R/X ratios tend to be higher. A coordinated voltage control strategy including also real-time thermal rating was presented in Ref. [12]. In this case, authors proposed an optimal control setting defined by a centralized control that might be used with set-point intervals between few minutes and one day ahead. This type of active distribution network operation planning highly relies not only on the communication infrastructure, but also on the accuracy of the forecasted data required to determine the control actions. In this sense, probabilistic and forecasting tools may help to reduce this uncertainty [13,14]. Moreover, voltage control mechanisms may be also performed as ancillary services considering a market framework as the one presented in Ref. [15], where the market settlement is performed using an Optimal Power Flow (OPF) formulation. All these concepts raise the need for reliable data and effective communication infrastructures, so advanced architectures combining technical and market operations developed so far just in pilot projects may become a reality at a larger scale [16]. However, currently there is a high uncertainty related to the level of smartness that should be deployed in the power system, which can be measured based on different Key Performance Indicators (KPIs) [17]. Indeed, despite the number of publications analyzing the requirements of the so called Smart Grid [18–21], most of them provide general recommendations that have to be verified from a technical point of view in order to reveal more practical conclusions [22]. In this sense, simulation tools can provide very valuable information to assess the performance of Smart Grid applications such as advanced voltage control considering both power and communication sides [23]. In addition, technical benefits have to be quantified in monetary terms to adopt the most cost-effective solutions. With this regard, the development of Cost-Benefit Analysis (CBA) methodologies may help to understand the economic implications of these advanced functionalities, but compared to the large amount of previous publications related to technical analyses, a much lower number of authors have tried to provide conclusions combining technical and economic approaches. For instance Refs. [24] and [25] compared from a technical and economic points of view different voltage control strategies in MV and LV networks, respectively. ® The authors used MATLAB and MATPOWER to run power flow simulations for the different voltage control mechanisms and then compared the investment, operation and maintenance costs of each solution. The results for the case of MV networks showed that central control can be much more interesting than grid reinforcement from an economic point of view, and also than local control in scenarios with large amount of DG. In the LV case the authors concluded that the combination of voltage dependent active and reactive power control mechanisms may provide the highest benefits. Authors also showed that distribution transformers with autonomously controlled on-load tap changers (OLTC) may be

339

economically effective in the long term despite their higher investment and maintenance costs. Finally, they analyzed the economic effectiveness of a centralized control for network loss reduction calculating the optimal reactive power of PV inverters and OLTC setting. They concluded that the sole reduction of energy losses does not justify the investment in this control strategy, but it can be interesting when the required infrastructure is already deployed to obtain additional benefits. This paper, based on all the aforementioned concepts and conclusions, details a novel methodology to assess the impact of two critical aspects of a centralized voltage control application: the accuracy of the data used to calculate the control actions, and the time interval used to send the active and reactive power setpoints from the central unit to the PV inverters. These aspects are directly related to the performance of the forecasting tools and communication systems, and their assessment provides a significant contribution to understand and improve the performance of voltage control applications. The presented methodology has been developed in the context of the SUSTAINABLE EU project [26] and applied in two different Portuguese networks used for the demonstration activities, one in LV and another in MV. The details of the technical and economic assessment are presented in Section 2, whereas the case studies are described in Section 3. The details for the forecasting and communication analysis are presented in Section 4, and the results for both networks are presented in Section 5. Finally, the conclusions are discussed in Section 6. 2. Centralized voltage control assessment 2.1. Technical assessment This first part of the assessment describes the formulation of the centralized voltage control as well as the methodology to analyze the impact of the forecast accuracy and the set-point timing in this application. In this study a centralized voltage control strategy ® based on an OPF formulation has been modelled with MATLAB and MATPOWER [27]. The objective of the OPF is to minimize the G ) of all generation units (M) at each time total generation costs (cj,h h, according to the standard formulation presented in Eqs. from (1) to (7). The slack bus produces energy at an hourly cost and PV is dispatched at zero cost, so the ultimate objective is to minimize PV curtailment (i.e. maximize PV generation).



min

G cj,h

∀h ∈ H

(1)

j∈M

Subject to: D pGi,h − Pi,h =



vi,h · vk,h · (Gi,k · cosi,k,h + Bi,k · sini,k,h ) ∀i ∈ N, h ∈ H

(2)

k∈N

D qGi,h − Qi,h =



vi,h · vk,h · (Gi,k · sini,k,h − Bi,k · cosi,k,h ) ∀i ∈ N, h ∈ H

(3)

∀i, k ∈ N, h ∈ H

(4)

k∈N

max si,k,h ≤ Si,k

Vrmin ≤ vi,h ≤ Vrmax 0≤

pG j,h

≤

Gmax Pj,h

∀i ∈ N, h ∈ H, r ∈ R

∀j ∈ M, h ∈ H

Gmin ≤ qG ≤ Q Gmax Qj,h j,h j,h

∀j ∈ M, h ∈ H

(5) (6) (7)

The power balance equations for each bus i and time h are represented in Eq. (2) and (3) for all the nodes of the network (N), where D are the active power generation and demand, respecpG and Pi,h i,h D are the analogue for the reactive power, G and tively, qG and Qi,h ik i,h Bik the real and imaginary parts of the element located in the row

340

L. González-Sotres et al. / Electric Power Systems Research 151 (2017) 338–347

i and column k of the admittance matrix, vi,h the voltage magnitude at bus i and time h, and  i,k,h the angle difference between buses i and k at time h. The technical constraints are defined in Eqs. (4)–(7), where si,k,h is the apparent power flow between buses i max the maximum apparent power flow for buses i and k at time h, Sik

and k, Vrmin and Vrmax the minimum and maximum admissible voltage limits at iteration r until the feasibility of the OPF is reached, Gmin and Q Gmax the minimum and maximum reactive power and Qj,h j,h generation of unit j at time h. The modelling of the centralized voltage control takes into account the computation of the set-points for the active and reactive power control as well as for the OLTC for every time h in the analyzed period (H). The OLTC is modelled by changing the voltage of the node connected to the slack bus (v1,h ). The active power production of the PV units (pG ) is determined according to the j,h profile of PV generation, and the reactive power provided by the PV inverter (qG ) is adjusted considering the limits of the reactive i,h

Gmin ) or generated (Q Gmax ) with the limitation power consumed (Qj,h j,h

of the apparent power (S2 = P2 + Q2 ). According to Eq. (1), PV curtailment is only used when no other control variable is available to mitigate voltage constraints. This represents how the DSO may decide the control actions. In case the voltage limits cannot be fulfilled with all the available control actions, the voltage limits are relaxed by ε to assess the amount of energy that would be served under undervoltage or overvoltage conditions. Then, the OPF is run again with the relaxed voltage limits repeating this process iteratively until the OPF converges (R). These voltage limits are obtained according to Eqs. (8) and (9). min = V min − ε Vr+1 r max Vr+1

=

Vrmax

+ε

∀r ∈ R

(8)

∀r ∈ R

(9)

With the aim of taking into account the effect of the forecasting tools and communication systems in the simulations, the following methodology is presented. First, a forecasting stage is defined to take into account the fact that the central unit has to determine the set-points for a future scenario without perfect information. So, a forecast error is included in the demand and generation profiles used for the calculation of the set-points. After this, a control stage is implemented to take into account the effect of the set-point timing that can be provided by the communication system. Hence, in this stage the time interval used to send the set-points from the central unit to the PV inverters is considered. The objective of the forecasting stage is to obtain the set-points that have to be applied to the actual profiles in the control stage, which are subject to forecast errors. Since forecast errors affect both load and PV estimation, two extreme scenarios can be considered in order to assess their impact: (a) load overestimation and generation underestimation; and (b) load underestimation and generation overestimation. The sensitivity to the forecast error is modelled by applying different errors to the actual profiles, adding or subtracting these errors depending on which case of overestimation or underestimation is analyzed. The set-points obtained in the previous stage can be sent periodically considering different time intervals taking into account the performance of the communication network. For instance, the central unit may perform the OPF with a time resolution of 1 min, but the communication system may be limited to transmit this information to the control devices every 15 min. So, when the periodic control actions are taken in time intervals longer than in the forecasting stage, a method to decide the set-points for each interval of the control stage has to be defined. In this case, the median of all the set-points from the forecasting stage within the same time interval is used. In the previous example, it would be equivalent to use the 8th largest set-point in all the simulations of the 15-min

interval. This process represents how the DSO may choose a representative set-point for every period considering the constraints of the communication network. With this process, every set-point is kept constant until the next set-point arrives. Then, another OPF is performed with the same time resolution used in the forecasting stage, but fixing the values of the set-points based on the previous criteria and being the OLTC located in the substation (slack bus) the only control variable. Finally, the results of the OPF at each time h for the following KPIs are obtained according to Eqs. from (10) to (12), respectively: voltage deviation over the admissible limit at  each 





node i vVD , energy curtailment at each generation unit j i,h



 EL

and network energy losses ph

vVD i,h

=

.

⎧ v − V0max , vi,h > V0max ⎪ ⎨ i,h ⎪ ⎩

0, V0min

− vi,h ,

Gmax − pG = Pj,h pEC j,h j,h

pEL = pG − h 1,h



≤ vi,h ≤

vi,h <

∀i ∈ N, h ∈ H V0max

V0min

∀j ∈ M, h ∈ H

D + Pi,h

i∈N

V0min



pG j,h

pEC , j,h

h ∈ H

∀i ∈ N, h ∈ H

(10)

∀i ∈ N, h ∈ H (11) (12)

j∈M

2.2. Economic assessment The economic impact of forecast accuracy and set-point timing has been analyzed considering the results of the technical assessment through different monetization formulas to convert the technical KPIs into economic KPIs, according to common CBA methodologies for smart grid projects [28,29]. Eqs. (13)–(15) show the computation of the cost of voltage deviation (cVD ), energy curtailment (cEC ) and energy losses (cEL ), where UC stands for the unitary cost of these parameters. c VD =



D VD Pi,h · UCi,h

(13)

pEC · UChEC j,h

(14)

h∈H i∈N

c

EC

=

 h∈Hj∈M

c EL =



pEL · UChEL h

(15)

h∈H

Note that these KPIs are computed ex-post, so they just quantify the costs from the technical assessment in monetary terms. The unitary costs used for this economic assessment have been chosen considering a social perspective, which means that they are not actual costs but they somehow represent the benefits obtained from a better performance of this application. For instance, the presence of voltage deviations over the technical limits jeopardizes the electricity supply. This means that in an extreme situation of voltage deviation, the social cost that may be borne is the cost of non-served energy (NSE), also referred as Value of Loss Load (VoLL), which represents the value that electricity users attribute to continuity of supply. This value can be used as an upper bound for the voltage deviation cost, so when the voltage exceeds a predetermined threshold, the associated cost can be approximated with the VoLL. For lower values, a linear or quadratic function can be used. In Fig. 1 an illustrative scheme of the cost function with a linear approach is shown, where the thresholds for the maximum and minimum voltage magnitudes associated to the VoLL are indicated as VVoLL max and VVoLL min , respectively. So, the social cost of a voltage deviation at a certain node can be obtained by multiplying the consumption of that node by the corresponding penalty. The value of VoLL can be very different depending on the region and the customer sector (e.g. residential, industrial or commercial).

L. González-Sotres et al. / Electric Power Systems Research 151 (2017) 338–347

341

Fig. 1. Scheme of the voltage deviation cost function.

Table 1 Characteristics of the case studies.

Voltage (kV) Network length (m) Number of LV customers (LV network) or MV/LV secondary substations (MV network) Contracted power (MVA) Number of PV units Installed capacity (MVA)

Estimations of VoLL for different countries and sectors can be found in [30], where authors concluded that in the long term VoLL values could lie in intervals of 5–25D /kWh for developed countries, and 2–5D /kWh for developing countries. In Ref. [31] an estimation of VoLL for the different sectors and regions of Spain can be found, where authors concluded that the cost of one kWh of NSE was above 4D /kWh but the economic signals sent to DSOs to avoid these interruptions were lower, which might result in underinvestment in short-term energy security. With regard to energy curtailment, the economic effect from the system point of view of reducing the production of renewable energy is that the same amount of energy has to be produced with conventional generation. This effect can be represented in a simplified way as an energy purchase at the wholesale market hourly price. This approach can be also applied to assess the economic impact of energy losses. As in the case of VoLL, wholesale energy prices present different values in each country. For instance, the wholesale energy prices in European regions are currently in the range of 25–50D /MWh [32]. 3. Case studies description In this section, the case studies used to apply the presented assessment methodology are described. These case studies represent a rural LV network and a semi-urban MV network located in the region of Évora in Portugal [33]. In Table 1 the main characteristics of these networks are summarized, whereas their geographical representation is shown in Figs. 2 and 3, respectively. The contracted power and installed capacity used in these case studies are significantly higher than the actual values to analyze the effect of the centralized voltage control in future scenarios. Although these scenarios are severe for the analyzed networks, the conclusions obtained from the proposed methodology provide a useful insight about the impact of the forecast accuracy and set-

LV network

MV network

0.4 650 12 0.65 5 0.25

15 7550 35 28.41 12 46

Fig. 2. Portuguese rural LV network case study (blue line) (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.).

point timing in a centralized voltage control application, which can be extrapolated to other networks facing voltage problems. In the simulations, demand and generation profiles for a 24-h interval with 1 min resolution are used. The load profiles of the LV customers are obtained with the Load Profile Generator tool [34], a modelling tool for residential energy consumption. The characteristics of the load profiles used for the analysis are the ones corresponding to the type of customer CHS01 (family, 2 children) in HT06 (normal detached house). Fig. 4 shows a set of 6 samples of residential consumer profiles modelled with the previous software used in the LV case study. For the MV network, an aggregated consumer profile obtained with 65 residential profiles is used to represent a MV/LV secondary substation, which is shown in Fig. 5. Fig. 6 shows the PV generation profiles that are used in the study, which are from a real PV unit in the area, one for a sunny day and

342

L. González-Sotres et al. / Electric Power Systems Research 151 (2017) 338–347

1 Sunny day Cloudy day

Power (p.u.)

0.8 0.6

0.4

0.2

0

0

5

10

15

20

25

Hour Fig. 6. PV generation profiles with 1 min resolution. Fig. 3. Portuguese semi-urban MV network case study (red line) (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.).

Table 2 Parameters of the voltage deviation cost function and used values. Parameter

Value

V0min V0max V VoLL V VoLL VoLL

0.95 p.u. 1.05 p.u. 0.8 p.u. 1.2 p.u. 3D /kWh

min max

Fig. 4. Residential consumer profiles with 1 min resolution.

Fig. 7. Energy production cost function.

Fig. 5. MV consumer profile with 1 min resolution.

other for a cloudy day. Note that all the profiles are defined in p.u., so the actual consumption and supply are obtained multiplying these load and generation profiles by the contracted power or installed capacity of each customer or PV unit, respectively. In Table 2, the parameters used to obtain the voltage deviation cost function illustrated in Fig. 1 are presented. The admissible voltage limits are set to ±5% according to Portuguese legislation, whereas the threshold for the maximum penalty is set to ±20% to represent the situation where electricity supply can be damaged. The value for the VoLL is set to 3D /kWh according to Ref. [35], and the intermediate values are obtained with a linear interpolation. Besides, the value of the parameter ␧ to relax the admissible voltage limits when the OPF does not converge is set to 0.01 p.u. Finally, the energy production cost function used to estimate the costs of energy curtailment and electricity losses related to these KPIs is represented in Fig. 7. This function is also used in the OPF to define the hourly energy production cost of the slack bus. The

L. González-Sotres et al. / Electric Power Systems Research 151 (2017) 338–347

343

Table 3 Implementation details of the analyzed scenarios in the case studies. Forecasting analysis

PV forecast error

Communication analysis

Scenario 1

Scenario 2

Scenario 3

Scenario 4

0%, 10%, 20%, 30%, 40% underestimation 0%, 10%, 20%, 30%, 40% overestimation

0%, 10%, 20%, 30%, 40% overestimation

0%

0%

0%

0%

1 min, 5 min, 15 min, 30 min, 1 h, 2 h, 4 h, 8 h, 24 h Sunny day

1 min, 5 min, 15 min, 30 min, 1 h, 2 h, 4 h, 8 h, 24 h Cloudy day

Set-point time interval

1 min

0%, 10%, 20%, 30%, 40% underestimation 1 min

PV profile

Cloudy day

Cloudy day

Load forecast error

values are obtained with the average hourly prices of the Iberian day-ahead electricity market in 2013 [36]. 4. Forecasting and communication analysis

40

40

35

35

30

30

25

25 Cost (€)

Cost (€)

The analysis of the impact of forecasting and communication on a centralized voltage control application is presented in this section, where the methodology explained in Section 2 is applied to the case studies described in Section 3. Then, two type of analysis are carried out for each case study: a forecasting analysis to assess the impact of the forecast accuracy achieved to calculate the control signals, and a communication analysis to assess the impact of the set-point timing used to send these control signals. The forecasting analysis is aimed to quantify the impact of the forecast errors included in the load and generation profiles. These profiles are used by the centralized voltage control application to compute the set-points for the OLTC and the PV inverters. For instance, the forecasting tool developed in Ref. [37] can achieve a normalized root mean square error of approximately 10%, being the highest performance reached when the forecast is carried out for short term periods. In this analysis, the following forecast errors are used: 0% (perfect forecast), 10%, 20%, 30% and 40%. These errors are applied for the scenario of PV underestimation and demand overestimation, and for the PV overestimation and demand underestimation. Then, in this analysis the OPF is first performed in the case studies with the 24-h profiles considering the corresponding forecast error and 1 min resolution, and then the obtained

set-points are applied in another OPF with the actual profiles considering 1 min set-point time interval to analyze the forecast error alone. The communication analysis aims to evaluate the impact of using different time intervals to send the set-points from the central unit to the PV inverters, which can be determined by the communication technology. For instance, the communication technologies used in the SUSTAINABLE project are PLC and GPRS [14]. PLC based on PRIME standard provides data rates up to 130 kbps, whereas GPRS provides 110 kbps in the downlink and up to 26.8 kbps in the uplink, so both can theoretically achieve time intervals up to 5 or 15 min [22]. However, their real performance depends on the local network conditions. In the case of PLC, the performance of this technology is highly affected by noise and signal attenuation [38,39]. Conversely, GPRS is more affected by the quality of the local coverage. When better performance is required, optical fiber can be used to provide higher data rates and reduce transmission delays. In this analysis, the following time intervals are used to analyze the impact of set-point timing for any type of communication technology: 1 min, 5 min, 15 min, 30 min, 1 h, 2 h, 4 h, 8 h and 24 h. Notice that longer intervals are multiple of shorter intervals for a better comparison. The initial OPF is computed considering only a perfect forecast of the load and generation profiles, so the forecasting analysis does not interfere with this one. Then, this time the OPF is carried out in the forecasting stage with 1-min resolution and 0% forecast error, but in the control stage is performed for the 9 sensitivities of set-point interval. As explained in Section 2.1, when a

20 Voltage cost Curtailment cost Losses cost Total cost

15 10

20 15 10

5 0

Voltage cost Curtailment cost Losses cost Total cost

5

0

5

10

15 20 25 Forecast error (%)

a

30

35

40

0

0

5

10

15 20 25 Forecast error (%)

30

35

40

b

Fig. 8. Impact of the forecast error in a cloudy day in the LV case study. (a) PV underestimation and demand overestimation. (b) PV overestimation and demand underestimation.

L. González-Sotres et al. / Electric Power Systems Research 151 (2017) 338–347

1000

1000

900

900

800

800

700

700

600

600 Cost (€)

Cost (€)

344

500 400

500 400

300

300 Voltage cost Curtailment cost Losses cost Total cost

200 100 0

Voltage cost Curtailment cost Losses cost Total cost

0

5

10

15 20 25 Forecast error (%)

30

35

200 100 40

a

0

0

5

10

15 20 25 Forecast error (%)

30

35

40

b

Fig. 9. Impact of the forecast error in a cloudy day in the MV case study. (a) PV underestimation and demand overestimation. (b) PV overestimation and demand underestimation.

time interval longer than 1-min is analyzed, the set-points applied in each interval are chosen using the median of all the set-points from the previous simulation. Finally, this analysis is applied in two scenarios, one with the PV profile of the sunny day and the other with the cloudy day. In the forecasting analysis, only the cloudy day profile is used. Then, in both analysis the same number of simulations are performed. A summary of the described implementation details for each scenario is included in Table 3. 5. Results 5.1. Forecasting analysis Fig. 8(a) shows the economic results of the defined KPIs (cost of voltage deviation, energy curtailment and energy losses described in Section 2.2) for the sensitivity of the forecast error using the PV profile of a cloudy day, for the case of PV generation underestimation and load overestimation in the LV network. It can be seen that when underestimating PV generation, higher forecast errors cause a significant increase of voltage costs. This is mainly because the control signals dedicated to compensate the overvoltages are less intense than actually required. By contrast, it can be noted that the curtailment cost decreases with a forecast error of 10% compared to the situation of perfect forecast, but with higher forecast errors the curtailment costs also increase. Moreover, in the case of 40% forecast error, the higher increase in the curtailment cost causes a decrement in the voltage cost. This effect is due to the differences in the demand profiles between the different customers, which makes that for higher levels of demand, the OLTC is not able to keep all voltages within limits. Thus, more curtailment is required because reactive power compensation has a very limited effect in the LV network. Indeed, in this network the demand is the dominant part. Then, it can be concluded that even when the amount of PV is small, the effect in voltage control can be substantial. Finally, it can be observed that the effect of the losses cost is similar to the curtailment cost, and is not very affected by the control actions. Fig. 8(b) shows the results in the case of overestimating generation and underestimating demand, where it can be seen that curtailment costs increase with the increase of the forecast error. However, in this case this effect is more natural than in the previous scenario because when generation is overestimated, control signals for energy curtailment are higher than actually needed. Fur-

thermore, since the demand underestimation reduces the amount of expected consumption, the estimated voltages are significantly higher than in the previous case, so the control actions sent to the PV inverters are even more restrictive and make voltage costs keep practically constant. Comparing both figures, it can be seen that the impact of the forecast error in total costs is higher in the PV underestimation than in the overestimation scenario, and the main driver in both cases is the voltage cost. Additionally, PV curtailment has been proved as an effective way to mitigate voltage deviations in this network, since small control actions can provide significant improvements in voltage profiles. Fig. 9(a) and (b) shows the sensitivity to the forecast error in the MV case study for the scenarios of underestimating PV generation and overestimating demand, and overestimating PV generation and underestimating demand, respectively. In the case of PV underestimation, it can be seen that voltage costs increase with the increase of the forecast error, whereas curtailment costs decrease. Then, when the forecast error for PV underestimation is higher and overvoltages are not identified, the obtained control actions do not include PV curtailment. In the case of PV overestimation, it can be noted that higher forecast errors result in higher curtailment costs, because the higher expected PV generation involves higher overvoltages that have to be mitigated. However, most of these actions are not necessary, because the scenario with 10% of forecast error already presents no voltage deviation. Finally, the effect of network losses can be better identified in this case study, although small variations are observed. Despite in Fig. 9(a) the main cost driver is voltage deviation and in Fig. 9 (b) the curtailment cost, for very high values of forecast error the total cost in both cases is very similar. However, for lower values of this parameter, generation underestimation scenarios show higher total costs than in the case of generation overestimation. From this analysis, it can be concluded that the worst scenario for the forecasting applications is the PV underestimation and demand overestimation, mainly in terms of voltage costs. Indeed, it can be seen that in both networks a 10% forecast error for this scenario produces a total cost similar to the one obtained with a 40% forecast error in the PV overestimation and load underestimation scenario. However, scenarios with PV overestimation usually involve higher curtailment requirements, which penalize the DG. Then, the reduction of the forecast error leads to better results of the voltage control application, because more accurate forecasts enable

40

40

35

35

30

30

25

25 Cost (€)

Cost (€)

L. González-Sotres et al. / Electric Power Systems Research 151 (2017) 338–347

20 Voltage cost Curtailment cost Losses cost Total cost

15 10

20 Voltage cost Curtailment cost Losses cost Total cost

15 10

5 0

345

5

0

240

480 720 960 Set-point interval (min)

1200

0

1440

0

240

a

480 720 960 Set-point interval (min)

1200

1440

b

Fig. 10. Impact of the set-point interval with perfect forecast in the LV case study. (a) Sunny day. (b) Cloudy day.

more effective control actions. Since forecasting tools achieve better performance for short term forecasting periods, control actions can be more effective when information is frequently updated. This idea is also highlighted in Ref. [40], and the impact of this effect is precisely analyzed in the next section. 5.2. Communication analysis

1000

1000

900

900

800

800

700

700

600

600 Cost (€)

Cost (€)

Fig. 10(a) and (b) shows the results for the sensitivity of the setpoint interval in the LV network considering perfect forecasts for the PV profiles of the sunny and cloudy days, respectively. In both cases it can be seen that the reduction of the set-point interval involves lower voltage costs thanks to a better adjustment to the actual voltage profiles, especially when the time period is lower than one hour. Note that within this time frame the reduction of the voltage cost is related to a slight increase of the curtailment cost, which allows maintaining the voltages closer to the admissible limits. This effect is more intensified in the cloudy day, since the PV

profile shows higher peak values. Therefore, even though in the cloudy day the curtailment cost is higher than in the sunny day for longer set-point intervals, the voltage costs are still higher. Then, in the sunny day PV voltage control is more effective. Finally, it can be seen that the effect of the set-point interval in energy losses in this network is not very high. Fig. 11 (a) and (b) shows the same sensitivity in the MV network. In both cases total costs show a dramatic reduction with a set-point interval of 1 min and saturate with intervals between 2 h and 4 h, and very similar values are obtained. However, in the sunny day voltage costs are considerably lower than in the cloudy day. This effect is caused by the lower voltage used at the slack bus in this scenario, which reduces the overvoltages but increases the energy losses. Comparing the results of Figs. 10 and 11, it can be noted that the total cost reduction obtained with lower set-point intervals is higher in the LV network, despite the fact that the ratio between PV generation and demand is higher in the MV case study. All these

500 400

400

300

300 Voltage cost Curtailment cost Losses cost Total cost

200 100 0

500

0

240

480 720 960 Set-point interval (min)

a

1200

1440

Voltage cost Curtailment cost Losses cost Total cost

200 100 0

0

240

480 720 960 Set-point interval (min)

b

Fig. 11. Impact of the set-point interval with perfect forecast in the MV case study. (a) Sunny day. (b) Cloudy day.

1200

1440

346

L. González-Sotres et al. / Electric Power Systems Research 151 (2017) 338–347

results show the complexity of the analysis, which is affected by many factors but yields important conclusions. The results from this communication assessment confirm that centralized voltage control applications with shorter set-point intervals are beneficial for the grid, regardless of the type of PV profile. Nevertheless, these improvements may be achieved only below a certain value of set-point interval. From the scenarios analyzed in this study, it can be concluded that this threshold is approximately 1 h. In this sense, the communication technology and strategy used to transmit the set-points is very important. Since the voltage control application obtains the best performance with a set-point timing of 1 min, especially in the MV, communication technologies with high data rates like optical fiber could be considered for this application.

6. Conclusions In this article, a methodology to analyze the impact of forecast accuracy and set-point timing for a centralized voltage control application has been presented. The main novelty of the presented approach is that it is focused on the assessment of the value of information considering forecasting and communication issues from a technical and economic point of view, which is a gap that has been found in the literature review. Additionally, sensitivities related to the type of forecast error (underestimation or overestimation) and the type of day for the PV generation profile (sunny or cloudy) have been taken into account to provide more practical conclusions. This methodology has been applied through simulations in a LV and in a MV real network, where the following conclusions have been obtained: From the forecasting analysis, in the case of PV underestimation and load overestimation, higher forecast errors have resulted in higher voltage costs and lower curtailment costs, because overvoltages are not identified and then the obtained control actions do not include PV curtailment. In the case of PV overestimation and load underestimation, higher forecast errors have resulted in higher curtailment costs, because the higher expected PV generation involves higher overvoltages that have to be mitigated. The impact of the forecast error has been found higher in the PV underestimation than in the overestimation scenario, being the voltage deviation the main cost driver. As a general conclusion, the reduction of the forecast error leads to better results of the voltage control application, because more accurate forecasts enable more effective control actions. From the communication analysis, the reduction of the setpoint interval has involved lower voltage costs thanks to a better adjustment to the actual voltage profiles, especially when the time interval was lower than 1 h. This cost reduction obtained with shorter time intervals has been related to a slight increase of the curtailment cost, which allows maintaining the voltages closer to the admissible limits. Moreover, the total cost reduction obtained with shorter set-point intervals has been higher in the LV network, although the ratio between PV generation and demand was higher in the MV case study. Nevertheless, in the case of MV networks, there is a huge difference when using this application with 1 min set-point interval, so in this voltage level the communication requirements would be higher than in the LV. The previous conclusions highlight the importance of dedicating resources to improve both communication systems and forecasting tools. However, the identified cost reductions should be compared to the cost to further reduce the forecast error and set-point interval, so this methodology can be used as a first step for a cost-benefit analysis. As regards to the three defined KPIs, voltage deviation has been proved as the most important cost driver in all the simulated scenarios except for the case of PV overestimation in the MV

network, where energy curtailment is clearly most important. PV curtailment has been proved as an effective way to mitigate voltage deviations from a technical point of view, since small control actions can provide significant improvements in voltage profiles. Energy losses have not been identified as an important cost driver, since only small differences have been obtained between all the scenarios. However, they can significantly contribute to the total costs, so they should be taken into account for any economic analysis. Nevertheless, the validity of this assessment is subject to the availability of the DSO to control DG that is established by the regulation at each country, as well as to the voltage quality requirements and remuneration schemes. In this sense, the presented methodology is particularly interesting in those cases where curtailment is allowed and voltage quality requirements are more stringent. Notwithstanding, the economic trade-off between voltage quality and RES curtailment highlighted in this chapter is a topic that has to be further investigated.

Acknowledgement The authors would like to acknowledge the SUSTAINABLE project partners for their contributions and support in data collecting. This project has received funding from the European Union’s Seventh Framework Programme for Research and Technological Development under grant agreement No: 308755.

References [1] SolarPower Europe, Global Market Outlook for Solar Power 2015–2019, SolarPower Europe, 2014, p. 32. [2] J. von Appen, M. Braun, T. Stetz, K. Diwold, D. Geibel, Time in the sun: the challenge of high PV penetration in the German electric grid, IEEE Power Energy Mag. 3 (2) (2013) 55–64. [3] M. Ebad, W.M. Grady, An approach for assessing high-penetration PV impact on distribution feeders, Electr. Power Syst. Res. 133 (2016) 347–354. [4] L. Gonzalez-Sotres, C. Mateo, T. Gomez, J. Reneses, M. Rivier, Á. Sanchez-Miralles, Assessing the impact of distributed generation on energy losses using reference network models, Cigrè International Symposium on The Electric Power System of the Future: Integrating Supergrids and Microgrids (2011). [5] M. Vandenbergh, D. Craciun, V. Helmbrecht, R. Hermes, R. Lama, P. M. Sonvilla, M. Rekinger, G. Concas, Prioritisation of Technical Solutions Available for the Integration of PV into the Distribution Grid, 2013. [6] E. Ghiani, F. Pilo, Smart inverter operation in distribution networks with high penetration of photovoltaic systems, J. Mod. Power Syst. Clean Energy 3 (4) (2015) 504–511. [7] I.J. Pérez-Arriaga, S. Ruester, S. Schwenen, C. Battle, J. Glachant, From Distribution Networks to Smart Distribution Systems: Rethinking the Regulation of European Electricity DSOs, 2013. [8] X. Han, A.M. Kosek, D.E. Morales Bondy, H.W. Bindner, S. You, D.V. Tackie, J. Mehmedalic, F. Thordarson, Assessment of distribution grid voltage control strategies in view of deployment, 2014 IEEE Int. Work. Intell. Energy Syst. (2014) 46–51. [9] M. Delfanti, M. Merlo, G. Monfredini, Voltage control on LV distribution network: local regulation strategies for DG exploitation, Res. J. Appl. Sci. Eng. Technol. 7 (23) (2014) 4891–4905. [10] N. Etherden, M.H.J. Bollen, Overload and overvoltage in low-voltage and medium-voltage networks due to renewable energy—some illustrative case studies, Electr. Power Syst. Res. 114 (2014) 39–48. [11] L. Collins, J.K. Ward, Real and reactive power control of distributed PV inverters for overvoltage prevention and increased renewable generation hosting capacity, Renew. Energy 81 (2015) 464–471. [12] M.Z. Degefa, M. Lehtonen, R.J. Millar, A. Alahäivälä, E. Saarijärvi, Optimal voltage control strategies for day-ahead active distribution network operation, Electr. Power Syst. Res. 127 (2015) 41–52. [13] F.V.V. Klonari, J.F. Toubeau, Z. De Greve, O. Durieux, J. Lobry, Probabilistic analysis tool of the voltage profile in low voltage grids, in: 23rd CIRED, June, 2015, pp. 15–18. [14] R.J. Bessa, A. Trindade, A. Monteiro, V. Miranda, Solar power forecasting in smart grids using distributed information, Proc. 18th Power Syst. Comput. Conf. (2014). [15] A.G. Madureira, J.A. Pec¸as Lopes, Ancillary services market framework for voltage control in distribution networks with microgrids, Electr. Power Syst. Res. 86 (2012) 1–7.

L. González-Sotres et al. / Electric Power Systems Research 151 (2017) 338–347 [16] C. Gouveia, D. Rua, F.J. Soares, C. Moreira, P.G. Matos, J.A.P. Lopes, Development and implementation of Portuguese smart distribution system, Electr. Power Syst. Res. 120 (2015) 150–162. [17] B. Dupont, L. Meeus, R. Belmans, Measuring the ‘smartness’ of the electricity grid, 2010 7th International Conference on the European Energy Market, EEM (2010) 1–6, 2010. [18] S. Rohjans, C. Danekas, M. Uslar, Requirements for smart grid ICT-architectures, in: 2012 3rd IEEE PES Innov. Smart Grid Technol. Eur. (ISGT Eur., Oct., 2012, pp. 1–8. [19] A. Patel, J. Aparicio, N. Tas, M. Loiacono, J. Rosca, Assessing communications technology options for smart grid applications, in: 2011 IEEE Int. Conf. Smart Grid Commun., Oct., 2011, pp. 126–131. [20] R.H. Khan, J.Y. Khan, A comprehensive review of the application characteristics and traffic requirements of a smart grid communications network, Comput. Netw. 57 (February (3)) (2013) 825–845. [21] V.C. Gungor, D. Sahin, T. Kocak, S. Ergut, C. Buccella, C. Cecati, G.P. Hancke, A Survey on smart grid potential applications and communication requirements, IEEE Trans. Ind. Informatics 9 (1) (2013) 28–42. [22] G. López, J.I. Moreno, H. Amarís, F. Salazar, Paving the road toward smart grids through large-scale advanced metering infrastructures, Electr. Power Syst. Res. 120 (2015) 194–205. [23] K. Mets, J.A. Ojea, C. Develder, Combining power and communication network simulation for cost-effective smart grid analysis, IEEE Commun. Surv. Tutorials 16 (3) (2014) 1771–1796. [24] B. Idlbi, K. Diwold, T. Stetz, H. Wang, M. Braun, Cost-benefit analysis of central and local voltage control provided by distributed generators in MV networks, IEEE PowerTech (2013). [25] T. Stetz, K. Diwold, M. Kraiczy, D. Geibel, S. Schmidt, M. Braun, Techno-economic assessment of voltage control strategies in low voltage grids, IEEE Trans. Smart Grid 5 (4) (2014) 2125–2132. [26] SUSTAINABLE project, 2016. [Online]. Available: http://www. sustainableproject.eu/. [27] R.D. Zimmerman, C.E. Murillo-Sanchez, R.J. Thomas, Matpower: steady-state operations, planning, and analysis tools for power systems research and education, IEEE Trans. Power Syst. 26 (1) (2011) 12–19.

347

[28] EPRI, Methodological Approach for Estimating the Benefits and Costs of Smart Grid Demonstration Projects, 2010. [29] JRC, Guidelines for Conducting a Cost-benefit Analysis of Smart Grid Projects, 2012. [30] I. Losa, O. Bertoldi, Regulation of continuity of supply in the electricity sector and cost of energy not supplied, in: Int. Energy Work, 2009, Venice, Italy, June 17–19, 2009, pp. 1–10, 2009. [31] P. Linares, L. Rey, The costs of electricity interruptions in Spain: are we sending the right signals? Energy Policy 61 (2013) 751–760. [32] European Commission, Energy Prices and Costs in Europe, 2016. [33] SUSTAINABLE project, Deliverable 6.2: Description of tools integration on existing infrastructure, 2016. [34] LoadProfileGenerator, 2016. [Online]. Available: http://www. loadprofilegenerator.de/. [35] A. Rodriguez-Calvo, P. Frías, J. Reneses, R. Cossent, C. Mateo, Optimal investment in smart MV/LV substations to improve continuity of supply, Int. J. Electr. Power Energy Syst. 62 (2014) 410–418. [36] OMIE, 2016. [Online]. Available: http://www.omie.es/en/inicio. [37] A. Madureira, R. Bessa, J. Meirinhos, D. Fayzur, J. Filipe, A.A. Messias, D.A. Lopes, P.G. Matos, The impact of solar power forecast errors on voltage control in smart distribution grids, in: Proceedings of the 23rd International Conference on Electricity Distribution (CIRED 2015), Lyon, France, 2015. [38] L. González-Sotres, C. Mateo, P. Frías, C. Rodríguez-morcillo, J. Matanza, Replicability analysis of PLC PRIME networks for smart metering applications, IEEE Trans. Smart Grid (2016) 1–9. [39] S. Galli, A. Scaglione, Z. Wang, For the grid and through the grid: the role of power line communications in the smart grid, Proc. IEEE 99 (June (6)) (2011) 998–1027. [40] Z. Ziadi, M. Oshiro, T. Senjyu, A. Yona, N. Urasaki, T. Funabashi, C.H. Kim, Optimal voltage control using inverters interfaced with PV systems considering forecast error in a distribution system, IEEE Trans. Sustain. Energy 5 (2) (2014) 682–690.