- Email: [email protected]
Providing ancillary service with commercial buildings: the Swiss perspective∗
9th International Symposium on Advanced Control of Chemical Processes 9th International Symposium on Advanced Chemical Processes June 7-10, 2015. Whistler, British Columbia,Control Canadaof 9th International Symposium on Control of Chemical Processes 9th International Symposium on Advanced Advanced June 7-10, 2015. Whistler, British Columbia,Control Canadaof Chemical Processes Available online at www.sciencedirect.com June Canada June 7-10, 7-10, 2015. 2015. Whistler, Whistler, British British Columbia, Columbia, Canada
ScienceDirect IFAC-PapersOnLine 48-8 (2015) 006–013
Providing ancillary Providing ancillary Providing ancillary Providing ancillary buildings: the buildings: the buildings: buildings: the the
service with commercial service with commercial service with serviceperspective with commercial commercial Swiss Swiss perspective Swiss Swiss perspective perspective
Lymperopoulos ∗∗ Faran A. Qureshi ∗∗ Truong Nghiem ∗∗ Lymperopoulos Faran A. Qureshi Nghiem ∗ Truong ∗ ∗∗ ∗ ∗ Truong Lymperopoulos Faran A. Qureshi Nghiem Ali Ahmadi ∗∗Khatir N. Jones ∗ Lymperopoulos Faran ∗∗ A.Colin Qureshi Truong Nghiem ∗ Ali Ahmadi Khatir ∗∗ Colin N. Jones ∗ Ali Ali Ahmadi Ahmadi Khatir Khatir ∗∗ Colin Colin N. N. Jones Jones ∗ ∗ Automatic Control Lab, Ecole Polytechnique Federal de Lausanne, ∗ ∗ Automatic Control Lab, Ecole Polytechnique Federal de Lausanne, ∗Lausanne, Automatic Control Lab, Ecole Polytechnique Federal CH-1015 Switzerland (e-mail: ioannis.lymperopoulos@-, AutomaticCH-1015 Control Lab, Ecole Polytechnique Federal de de Lausanne, Lausanne, Lausanne, Switzerland (e-mail: ioannis.lymperopoulos@-, Lausanne, CH-1015 Switzerland (e-mail: ioannis.lymperopoulos@-, faran.qureshi@-, xuan.nghiem@-, [email protected]). Lausanne, CH-1015 Switzerland (e-mail: ioannis.lymperopoulos@-, faran.qureshi@-, xuan.nghiem@-, [email protected]). ∗∗ faran.qureshi@-, xuan.nghiem@-, [email protected]). Ltd, Laufenburg, CH-5080 Switzerland (e-mail: ∗∗ Swissgrid faran.qureshi@-, xuan.nghiem@-, [email protected]). ∗∗ Swissgrid Ltd, Laufenburg, CH-5080 Switzerland (e-mail: ∗∗ Swissgrid Ltd, Laufenburg, CH-5080 [email protected]) Swissgrid Ltd, Laufenburg, CH-5080 Switzerland Switzerland (e-mail: (e-mail: [email protected]) [email protected]) [email protected])
Ioannis Ioannis Ioannis Ioannis
Abstract: Abstract: Abstract:services constitute the cornerstone of the power grid. They allow for an efficient system Ancillary Abstract: Ancillary services constitute the cornerstone of the power grid. They allow for an efficient system Ancillary services of the grid. They for system operation, provideconstitute resiliencethe to cornerstone uncertainties establish unprecedented Ancillary services constitute the cornerstone of and the power power grid.safeguards They allow allowagainst for an an efficient efficient system operation, provide resilience to uncertainties and establish safeguards against unprecedented operation, provide resilience to uncertainties and establish safeguards against unprecedented events. Their importance is growing due to the rise of grid decentralisation and integration of operation, provide resilience to uncertainties and establish safeguards against unprecedented events. Their importance is growing due to the rise of grid decentralisation and integration of events. Their importance is growing due to the rise of grid decentralisation and integration of intermittent, renewable power sources, which lead to more variability and uncertainty in the events. Their renewable importancepower is growing duewhich to thelead rise of more grid decentralisation and integration of intermittent, sources, variability and uncertainty in intermittent, renewable power sources, which lead to to more variability and uncertainty in the the system. Today, the vast power share sources, of ancillary services is more provided by largeand generating units. An intermittent, renewable which lead to variability uncertainty in the system. Today, the vast share of ancillary services is provided by large generating units. An system. Today, the vast ancillary services is large units. ongoing effort by and of business entities focuses on usingby of loads connected system. Today, theresearch vast share share of ancillary services is provided provided byvariation large generating generating units. An An ongoing effort by research and business entities focuses on using variation of loads connected ongoing effort by research and business entities focuses on using variation of loads connected to the power grid in order to increase significantly the provision of such services, hopefully at a ongoing effort by research and business entities focuses on using variation of loads connected to the power grid in order to increase significantly the provision of such services, hopefully at aa to in to the such services, at reduced cost. grid We examine from ansignificantly economic perspective, theof of commercial buildings to the the power power grid in order orderhere, to increase increase significantly the provision provision ofuse such services, hopefully hopefully atas a reduced cost. We examine here, from an economic perspective, the use of commercial buildings as reduced cost. cost. We providers examine here, here, from an economic perspective, the use of of commercial commercial buildings as ancillary service based on an realeconomic prices from the Swissthe electricity market. We calculate reduced We examine from perspective, use buildings as ancillary service providers based on real prices from the Swiss electricity market. We calculate ancillary providers based on real prices from Swiss market. We calculate the effectservice of retail electrical prices economic performance of a building that ancillary service providers based on on realthe prices from the the Swiss electricity electricity market.and We find calculate the effect of retail electrical prices on the economic performance of aa building and find that the effect of retail electrical prices on the economic performance of building and find that for the rates charged in the least expensive cantons a single building can reduce its overall the effect of retail electrical prices on the economic performance of a building and find that for the rates charged in the least expensive cantons a single building can reduce its overall for the rates charged in the least expensive cantons a single building can reduce its overall energy costs, when participating in the ancillary services market. For the high end of prices this for the costs, rates when charged in the least expensive cantons a market. single building can reduce its overall energy participating in the ancillary services For the high end of prices this energy costs, when participating the ancillary services end of prices this gradually becomes prohibitive butin can be alleviated for a market. buildingFor thatthe hashigh a need for electricity energy costs, when participating in the ancillary services market. For the high end of prices this gradually becomes prohibitive but can be alleviated for a building that has a need for electricity gradually becomes prohibitive but can be alleviated for a building that has a need for electricity during nighttime hours, as well as daytime. Finally, we show, the counter-intuitive result that gradually becomes prohibitive but can be alleviated for a building that has a need for electricity during nighttime hours, as well as daytime. Finally, we show, the counter-intuitive result that during nighttime as as show, counter-intuitive result that providing ancillaryhours, services can increase the Finally, comfort we levels of athe building at a decreased cost. during nighttime hours, as well well as daytime. daytime. Finally, we show, the counter-intuitive result that providing ancillary services can increase the comfort levels of a building at a decreased cost. providing ancillary services can increase the comfort levels of a building at a decreased cost. providing ancillary services can increase the comfort levels of a building at a decreased cost. © 2015, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved. Keywords: Ancillary services, stochastic programming, buildings Keywords: Ancillary services, services, stochastic programming, programming, buildings Keywords: Keywords: Ancillary Ancillary services, stochastic stochastic programming, buildings buildings 1. INTRODUCTION and reduce energy costs. For this reason, they are also 1. INTRODUCTION INTRODUCTION and reduce energy costs. For this reason, they are also 1. and reduce energy this they also being envisioned as costs. storageFor elements that could increase 1. INTRODUCTION and reduce energy costs. For this reason, reason, they are are also being envisioned as storage elements that could increase being envisioned as storage elements that could increase the performance and the robustness of the power grid Accommodating for the fluctuating mismatch between being envisioned as storage elements that could increase performance and the robustness of the power grid Accommodating for for the the fluctuating fluctuating mismatch mismatch between between the the performance and robustness of the grid ancillary service providers without Accommodating electrical consumption andfluctuating generation,mismatch either because of by the becoming performance and the the robustness of (ASP), the power power grid Accommodating for the between by becoming ancillary service providers (ASP), without electrical consumption and generation, either because of by becoming ancillary service providers (ASP), without the need for investing in new generation capacity (or electrical consumption and generation, either because of operational uncertainties or large irregular events, is perby becoming ancillary service providers (ASP), without electrical consumption and generation, either because of the need for investing in new generation capacity (or operational uncertainties uncertainties or or large large irregular irregular events, events, is is perper- underutilising the need for investing in new generation capacity (or the existing one). operational formed at different hierarchical levels. A spot market, the need for investing in new generation capacity (or operational uncertainties or large levels. irregular events, is per- underutilising the existing one). formed at different hierarchical A spot market, underutilising the existing one). formed at different hierarchical levels. A spot market, usually operated on hierarchical an hourly basis (but also cleared a underutilising the existing one). formed at different levels. A spot market, building accumulates energy through its thermal causually operated operated on on an an hourly basis basis (but (but also also cleared cleared aa A building accumulates energy through its thermal causually day in advance), matches or alters the (but generation schedulea A usually operatedmatches on an hourly hourly basis also cleared A building energy through its thermal capacity. Thisaccumulates slowly dissipating employed day in advance), or alters the generation schedule A building accumulates energy energy throughcan its be thermal capacity. This slowly dissipating energy can be employed day in advance), matches or alters the generation schedule of power plants to meet the projected demand of load day in advance), matches or alters the generation schedule pacity. This slowly dissipating energy can be employed to shift the electrical load of the HVAC (Heating, Venof power plants to meet the projected demand of load pacity. This slowly dissipating energy can be employed the load HVAC (Heating, Venof power plants to the projected demand load serving The grid operator, due to to of powerentities plants (LSE). to meet meet thepower projected demand of of load to shift shift and the electrical electrical load of of the the HVAC (Heating, Ventilation Air-Conditioning) system to more beneficial serving entities (LSE). The power grid operator, operator, due to to shift the electrical load of the HVAC (Heating, Ventilation and Air-Conditioning) system to more beneficial serving entities (LSE). The power to the inherent unpredictabilities in thegrid actual deliverydue of this serving entities (LSE). The power grid operator, due to tilation and Air-Conditioning) system to more beneficial time windows (a demand response service) or to absorb the inherent unpredictabilities in the actual delivery of this tilation and Air-Conditioning) system to more beneficial windows (a demand response service) or to absorb the inherent in actual delivery of as this transaction, procures additional generation capacityof a time the inherent unpredictabilities unpredictabilities in the the actual delivery this time windows (a response service) to arising from tracking that transaction, procures additional additional generation capacity as as a fluctuations time windows (a demand demand responseaa regulation service) or orsignal to absorb absorb fluctuations arising from tracking regulation signal that transaction, procures generation capacity a backup to meet the mismatch at faster time scales, when transaction, procures additional generation capacity as a fulfils fluctuations arising from tracking a regulation signal an AS obligation towards the grid operator. The backup to meet the mismatch at faster time scales, when arising from towards tracking the a regulation signal that that fulfils an AS obligation grid operator. The backup to meet the mismatch at faster time scales, when required. These services are named ancillary services (AS) fluctuations backup to meet the mismatch at faster time scales, when fulfils an an AS obligation obligation towards the grid grid operator. operator. The critical constraint is to ensure a comfortable environment required. These services are named ancillary services (AS) fulfils AS towards the The critical constraint is to ensure aa comfortable environment required. These services are named ancillary services (AS) and are commonly procured from large generation units, required. These services are named ancillary services (AS) critical constraint is to ensure comfortable environment to the occupants while providing the promised service. and are commonly procured from large generation units, critical constraint is to ensure a comfortable environment the occupants while providing the promised service. and are procured from large units, operated at lower than nominal at all times to to and are commonly commonly procured fromcapacity large generation generation units, the while providing the promised operated at lower lower than than nominal capacity at all all times times to to to the occupants occupants whileexplored providingthe thebenefits promisedofservice. service. operated at nominal capacity at to Several studies have accommodate for the fluctuating regulation signal of the operated at lower than nominal capacity at signal all times to Several studies have explored the benefits of buildings buildings accommodate for the fluctuating regulation of the Several studies have explored the benefits of accommodate participating in have electricity markets, usually by adapting grid operator. for studies explored the benefits of buildings buildings accommodate for the the fluctuating fluctuating regulation regulation signal signal of of the the Several participating in electricity markets, usually by adapting grid operator. participating in electricity adapting grid operator. to the variations of electricalmarkets, price to usually achieve by a less costly participating in electricity markets, usually by adapting grid operator. the variations of electrical price to achieve a less costly Commercial buildings are large consumers of electrical to to the variations of electrical price to achieve a less operation by shifting load. Qureshi et al. (2014) investigate Commercial buildings are large consumers of electrical to the variations of electrical price et to al. achieve a investigate less costly costly by shifting load. Qureshi (2014) Commercial buildings of electrical power and are movingare to large more consumers sophisticated levels of operation Commercial buildings are large consumers of electrical operation by by shifting shifting load. Qureshi et al. al. (2014) (2014) investigate participation in the load. New Qureshi York demand response (DR) power and and are are moving moving to to more more sophisticated sophisticated levels levels of of operation et investigate participation in the New York demand response (DR) power automation in order to increase oversight, comfort levels power and are moving to moreoversight, sophisticated levels of participation the New York demand response (DR) program usingin model predictive control (MPC), Vrettos automation in order to increase comfort levels participation in the New York demand response (DR) program using model predictive control (MPC), Vrettos automation in order to increase oversight, comfort levels automation in order to increase oversight, comfort levels et program using model predictive control (MPC), Vrettos al. (2013) study DR for residential buildings coupled program using model predictive control (MPC), Vrettos This work has received support from the Swiss National Science et al. (2013) study DR for residential buildings coupled et (2013) study DR residential buildings coupled with storage. approaches optimise the This work has received support from the Swiss National Science et al. al.extra (2013) study These DR for for residentialmostly buildings coupled with extra storage. These approaches mostly optimise the Foundation under the GEMS project (grant number 200021-137985) This work has received support from the Swiss National Science with extra storage. These approaches mostly optimise This workunder has received support from the number Swiss National Science operation ofstorage. the building without a priori promises to the Foundation the GEMS project (grant 200021-137985) with extra These approaches mostly optimise the operation of the building without a priori promises to the and the European Research Council under the European Unions SevFoundation under the GEMS project (grant number 200021-137985) Foundation under Research the GEMS projectunder (grant number 200021-137985) operation of the building without aa priori promises to the grid operator. and the European Council the European Unions Sevoperation of the building without priori promises to the enth Framework Programme (FP/2007-2013) / ERC Grant Agreeand the European Research Council under the European Unions Sevgrid operator. and the EuropeanProgramme Research Council under the European Unions Sevgrid operator. enth Framework (FP/2007-2013) / ERC Grant Agreegrid operator. ment n. 307608 (BuildNet) enth Framework Programme (FP/2007-2013) / ERC Grant Agree-
enth Programme (FP/2007-2013) / ERC Grant Agreement Framework n. 307608 (BuildNet) ment ment n. n. 307608 307608 (BuildNet) (BuildNet) Copyright © 2015, 2015 IFAC 6 Hosting by Elsevier Ltd. All rights reserved. 2405-8963 © IFAC (International Federation of Automatic Control) Copyright © 2015 IFAC 6 Copyright © 2015 IFAC 6 Peer review under responsibility of International Federation of Automatic Copyright © 2015 IFAC 6 Control. 10.1016/j.ifacol.2015.08.149
IFAC ADCHEM 2015 June 7-10, 2015. Whistler, BC, Canada Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
7
quency is outside of a deadband, which is set ±10 mHz in continental Europe.
ASPs have to commit before being able to provide a service to the power grid (e.g. increasing or decreasing consumption at request) that is determined in advance (here weekly). Based on this commitment the grid operator sends a high frequency regulation signal, according to his needs, that the ASP has to follow while also fulfilling his primary operational objectives (for a building that would be delivering comfort to occupants at a low cost). Buildings providing AS have been explored in Maasoumy et al. (2014), Maasoumy et al. (2013), Vrettos et al. (2014), Pavlak et al. (2014), Hao et al. (2013) and Balandat et al. (2014). These studies examine the capacity for providing such services with artificial economic criteria and for a limited prediction horizon (at most daily).
Frequency Restoration Reserve (FRR) control is a centralized and continuous control (i.e., is used in both normal operation and in contingency cases) to restore the frequency and exchanges with other control areas to their target values. Such an indispensable frequency control can be implemented automatically (aFRR) with a central proportional-integral (PI) controller, or manually (mFRR) by directly calling providers. Replacement Reserve (RR) is a manually activated product that is used i) for further imbalances in the case FRR has already been activated and is unable to restore the frequency to its target value, ii) to release FRR in the case of large disturbances and iii) to anticipate expected imbalances.
We use here real economic data from the Swiss EPEX market (from 2013) and explore commitments that have a weekly horizon. We employ a two stage stochastic formulation, that allows us to exploit certain statistical properties of the regulation signal. We find this approach to be effective while also satisfying constraints when tested against real regulation signals. Moreover, we try to assess the economic feasibility, of a building providing AS. For this, we study how the retail price of electricity plays a significant role in the decision for participation. For a single unit this is due to variable constraints (night setback), between working and non-working hours, that create a requirement for the procurement of electricity even at times that the building might not usually need it. We find that a building, or an aggregation, that requires some energy consumption at all times during the week would always have an incentive to participate in the secondary control market for even the highest, realistic, prices of electricity.
In this paper, we focus on the Swiss Ancillary Services market and an aFRR product, which is known as the Secondary Control Reserve (SCR). SwissGrid, the operator of Swiss electrical grid, procures about ±400 MW SCR in a weekly auction for every hour of the following week from a set of pre-qualified Ancillary Service Providers (ASPs), which can be either loads or generators. During real-time operation, SwissGrid activates the procured SCR through an Automatic Generation Controller (AGC) by sending each second an activation signal in parallel to all ASPs that have been accepted in the secondary control market. Note that the activation signal, which is called the Area Control Signal (ACS) in this paper, is proportional to the awarded capacities of all accepted ASPs.
Finally, we evaluate the comfort levels attained, by using standard long-term satisfaction metrics and find that participation in AS improves the average comfort satisfaction levels compared to operating the building to minimise costs without AS participation.
We detail the operational and financial structures of the Swiss secondary ancillary services system in the following sections.
In Section 2 we introduce the Swiss AS market and present some of the statistical qualities of the frequency regulation signal. In Section 3 we present the methodology for acquiring a simulation model for the building, while in Section 4 we examine how a commercial bidding can bid in the AS market while operating inside its comfort constraints. Finally, in Section 5 we present our simulation set-up, describe the long term comfort metric we use and explain our results.
At the start of each week, the provider of ancillary services (the building) must submit a capacity bid, α MW, and a baseline energy schedule e¯ = {¯ ek }, where e¯k is measured in units of MWh and is provided in 15min time increments for the entire week.
2.1 Swiss Secondary Ancillary Services Market
Throughout the paper we take the perspective of the service provider, and adopt the convention that energy is positive when it flows into the service provider and that monetary values are positive when they flow out from the provider.
2. SECONDARY ANCILLARY SERVICES IN SWITZERLAND
If the baseline is negative (the service provider is a generator), then the baseline is a production schedule, and is sold on the spot market at the hourly spot rate. However, if the service provider is a consumer, as we assume throughout this paper, then the baseline energy is purchased from the retail market, at fixed retail rates. As a result, we can define the cost to the service provider for the baseline schedule e¯ as Cbaseline (¯ e) := cretail e¯k
According to The European Network of Transmission System Operators for Electricity (ENTSO-E) definition, three levels of frequency control are generally used to maintain the real-time balance between load and generation. Frequency Containment Reserve (FCR) control is a local and decentralized proportional controller that contains the system frequency within a maximum deviation bound following large generation or load outages in a synchronous area of continental Europe. Note that FCR is only activated in the case of contingency when the system fre-
where cretail is the retail price of electricity.
The financial reward paid to the provider for offering a capacity of α MW is based on the bid price of the provider 7
IFAC ADCHEM 2015 June 7-10, 2015. Whistler, BC, Canada 8 Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
100 spot price ACS rk
90
70
# of occurrences
Price of Electricity (CHF/MWh)
ACS
ck
80
60 50 40 30 20 10 0
0
1
2
3
4
5
6
-400
7
-300
-200
-100
0
100
200
300
400
Requested MW
time (days)
Fig. 1. Incentives for tracking the ACS.
Fig. 2. Distribution of the averaged 15 min ACS for 2013
(pay-as-bid), and is proportional to α. Rcapacity (α) := ccapacity · α (1) For the purposes of the study in this paper, we fix the bid price of the provider ccapacity to be the average of accepted bids in the market, measured in CHF / MW.
SwissGrid levies a penalty against providers for tracking imbalances. Using a similar argument to the previous section, different penalties are paid for positive and negative deviation: penalty Cpenalty () = ck max{k , 0} − rkpenalty max{−k , 0} (4)
2.2 Real-time Operation
and rkpenalty are the cost and rebate paid for where cpenalty k tracking deviations.
Each 15 minute period, SwissGrid measures the error in energy between the ACS ak and the energy consumed / produced by the provider ek , and a financial remuneration / penalty is paid / imposed with a price that is coupled to the Swiss spot market price (SwissIX). This financial adjustment is based on two components: An incentivisation price for energy consumed / produced as a result of tracking the ACS, and a penalty price for deviations against this signal.
The reader is referred to the SwissGrid website 1 for full details on the incentivisation and imbalance penalties involved. 2.3 Statistical Properties of the ACS Important elements of the ACS can be captured by analysing the statistics of the signal provided by Swissgrid. We restrict our analysis here to the average ACS over a 15 min period (the actual signal comes every 1 sec with a clearing of 15 min), since we assume that a lower level controller captures the high frequency deviations, while our optimization scheme provides the mean level of power at each time period. We also find that the time correlation of the ACS is strong for this time period (0.55 after 15 min), so that the signals do not differ excessively.
Incentivisation If the ACS ak is positive, then the provider is being asked to consume more energy, and so this energy is sold at a discount. If the signal ak is negative, then the request is for a reduction in consumption. However, the provider has already purchased this energy at retail rates in the form of its baseline schedule and as a result, SwissGrid returns a rebate to the provider for the energy not used. The reward for tracking an ACS {ak } is therefore −cACS max{ak , 0} + rkACS max{−ak , 0} RACS (a) := k (2)
The distribution of the ACS throughout 2013 can be observed in Figure 2. The maximum absolute magnitude of the ACS is 400 MW, however the mean of the absolute magnitude is roughly 50 MW. The ACS is almost zero mean, with a small negative bias of almost -11 MW, however this varies per week and especially per day (there exist days for which the ACS is almost entirely positive or entirely negative).
where cACS and rkACS are the cost and rebate paid for enk ergy respectively. Figure 1 shows the relationship between these various prices for the 30th week of 2013.
From the probability density distribution of the signal we can deduce that there exist only a few significant outliers, while the mass of the signal is concentrated close to zero. The standard deviation of the ACS is almost 70 MW, while the kurtosis of the distribution is higher (∼ 5) than that of the normal distribution.
Note that the reward for ACS tracking may be either positive or negative, depending on whether the service provider gains energy as a result of tracking the signal, or produces it. Imbalance Penalties The tracking imbalance k between the ACS ak , and the energy consumed / produced by the provider ek at time k is (3) k = ek − e¯k − ak
By saturating tracking for very extreme outliers (track an outlier up to a portion of its full magnitude) we can protect 1 http://www.swissgrid.ch/swissgrid/en/home/experts/topics/ ancillary_services.html
8
IFAC ADCHEM 2015 June 7-10, 2015. Whistler, BC, Canada Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
input (external air temperature, sky temperature, solar radiation on each building surface, long-wave radiation, internal gain, etc.), and yk ∈ Rq is the room air temperature at time step k. The total thermal energy input for the building ek is then given by (6) ek = η(1T uk ) where we assume a constant electrical to thermal conversion efficiently of η = 1.
100 Energy remaining in the ACS signal (%)
99 95
90
90
80 80
70
The building is deemed satisfactorily comfortable if certain temperature constraints are satisfied. This is used throughout our simulations and imposes linear constraints on the optimization problem. We define the comfort constraints at a level δ ◦ C as |yk − Tideal | ≤ δ, where Tideal ◦ C is the ideal temperature provided by the occupants (set to 24 ◦ C in the simulations).
60
Energy remaining in the ACS signal
50
Select Points - Remaining Energy (%)
40
0
10
20
30
40
50
60
70
9
80
% of magnitude saturated
Fig. 3. Power Remaining in the ACS after truncating outliers at different saturation levels the equipment, facilitate larger bids while also fulfilling the AS obligation at an acceptably high level of compliance. We track asat = min(ak , (1 − sat) · ak ), where ak is the k original ACS and sat ∈ [0, 1] the saturation level.
As has been reported several times in the literature, a minimum cost approach, with no ACS tracking, would tend to follow the upper or lower temperature constraints in order to minimize energy spend. However, an ACS tracking approach would attempt to place the temperature somewhere between the upper and lower constraints in order to provide a greater flexibility for ACS tracking, which will result in a greater comfort for the occupants. This counter-intuitive behaviour can be observed in Section 5.
In Figure 3, we observe that shaving 60% of the magnitude of outliers preserves more than 96% of the energy in the ACS. This means that tracking is still effective for a very large portion of the requests. Here, we restrict to a 40% saturation level that allows for tracking more than 99% of the energy contained in the ACS. This results on average to only one to three 15 min periods in a week where the tracking is suboptimal. Most operators (including SwissGrid) allow a small percentage of time where the tracking can be suboptimal. Thus, the tracking capacity (and the revenue) is increased by tracking truncated ACS and taking advantage of the allowed tracking suboptimality.
The building model, coupled with a forecast of weather and occupancy, allows the characterization of the set of all energy signals that the building is able to consume throughout the coming week, while still ensuring the comfort of its occupants. This comfort set is given as xk+1 = Axk + Bu uk + Bd dk |Cxk − Tideal | ≤ δ Bcomfort (δ) := {ek } uk ∈ U e = η(1T u ) k k (7)
3. BUILDING MODEL We use the MATLAB toolbox OpenBuild, see Gorecki et al. (2014), to extract a linear state-space model from a detailed EnergyPlus building model. EnergyPlus, Crawley et al. (2001), is a tool for comprehensive modeling of building thermodynamics and HVAC systems. However, this representation is highly detailed, and not suitable as a prediction model in an optimization-based control scheme. OpenBuild extracts all the required data (material properties, orientation of different surfaces, etc.) from an EnergyPlus model and uses it to automatically construct a first principles linear model of the building thermodynamics. Moreover, EnergyPlus is also used to compute the effect of the external weather conditions and the internal gains on the linear model. Once constructed, the linear model is again validated against EnergyPlus. For more details see Gorecki et al. (2014) .
4. THE BIDDING PROBLEM FOR COMMERCIAL BUILDINGS In this section, we formulate the weekly bidding problem that a building faces when participating in the secondary ancillary services market of Switzerland. The goal is to characterize the weekly set of capacity bids α and baseline energy consumptions e¯ such that the building can ensure comfort of its occupants with a high probability. Amongst this set, the building can then select the element that minimizes its expected cost of operation. From Section 2, the total financial cost of the service provider can be summarized as e) − Rcapacity (α) C(α, e¯, a, e) := Cbaseline (¯ + Cpenalty (e − e¯ − a) − RACS (a)
The linear model obtained from OpenBuild has 70 states and is reduced in order to 20 states using the Hankel-Norm based balanced truncation method. The reduced linear model is discretized with a sampling time of 15 min. to obtain a linear dynamic model (5) xk+1 = Axk + Bu uk + Bd dk yk = Cxk where xk ∈ Rn is the state, uk ∈ Rm is the thermal energy input to each of the m zones, dk ∈ Rp is the disturbance
We can see that given an ACS a to track and a baseline schedule e¯, the optimal action of the building is to minimize any imbalance penalty Cpenalty . We define the ◦ function Cpenalty as the minimum penalty that can be obtained subject to comfort of the building’s occupants ◦ (γ) := min Cpenalty (e − γ) Cpenalty e
s.t. e ∈ Bcomfort (δ) 9
IFAC ADCHEM 2015 10 June 7-10, 2015. Whistler, BC, Canada Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
We can now formally define the bidding problem as e) − Rcapacity (α) (8) min Cbaseline (¯ e¯,α ◦ + Eaˆ Cpenalty (¯ e + αˆ a) − RACS (αˆ a) where the expectation is taken over the distribution of the normalized ACS a ˆ.
32
The optimizer (¯ e , α ) of (8) is then the baseline schedule and capacity bid that will result in lowest expected operating costs of the building, subject to the occupants of the building remaining comfortable.
24 20 16 12
4 0
0
1
2
3
4
5
6
7
Time [Days]
The bidding problem (8) can be solved using a twostage stochastic programming approach. The first stage variables are those that must be provided to SwissGrid at the start of the week : the capacity bid α, and the baseline schedule e¯. The second stage variables are then the response of the building ej to the randomly drawn samples a ˆj of the normalized ACS.
Fig. 4. Baseline consumption declared for a week and the ACS superimposed when it is realised. 32 30
We can now pose the two-stage stochastic optimization problem approximating (8) e) − Rcapacity (α) (9) min Cbaseline (¯
Temperature ( oC)
28
e¯,α
s.t.
28
8
4.1 Computation : Two Stage Stochastic Programming
Ns 1 −RACS (αˆ aj ) + Ns j=1
xjk+1 = Axjk + Bu ujk |Cxjk − Tideal | ≤ δ ujk ∈ U ejk = η(1T ujk ) ajk ejk − e¯k = αˆ
Baseline ACS superimposed
36
Power Consumption [W/m2 ]
ACS Tracking
40
26 24 22 Upper Constraints Lower Constraints Zone 1 Zone 2 Zone 3 Zone 4 Zone 5
20
+ Bd dk
18 16
0
1
2
3
4
5
6
7
Time [Days]
Fig. 5. Zones temperature profile for time varying comfort constraint (night setback).
α≥0 where Ns is the number of samples drawn from the normalized distributed of the ACS. The horizon length of the optimization problem (9) is one week with a time step of 15 min. Note that a tracking constraint is added to ensure perfect tracking of the samples a ˆj of the normalized ACS. Thus, the imbalance penalty Cpenalty term is removed from the cost function.
a week long basis. Weather and occupancy disturbances are assumed to be perfectly forecast, while the ACS (provided by Swissgrid) is revealed causally and is unknown at the time of the bid. For the ACS scenarios we use the recorded ACS from the previous weeks. For the financial variables, we use spot prices from the 2013 Swiss market, bonus and penalties for tracking (or deviating from ACS) from Swissgrid for the same year and the average capacity bid per week as published.
We highlight that problem (9) can be re-cast as a linear programming problem, enabling the solution of very largescale problems. The constraints are clearly linear, and the functions Cbaseline and Rcapacity are also linear functions. From equations (2) and (4), we see that the functions Cpenalty and RACS are convex piecewise affine functions of their arguments, and can therefore be re-formulated as linear functions subject to linear constraints using a standard epigraph formulation.
The building model is run with a 15 min time step and the problem is solved using a YALMIP interface, L¨ ofberg (2004), to the Gurobi numerical solver. The number of ACS scenarios is quite limited (14 to 15) due to the extreme memory requirements from the numerous constraints generated for a week long simulation. Sensitivity studies suggest that this small number of scenarios is still representative in this case. The average time required for a single week optimization is approximately 5 min.
5. SIMULATION RESULTS 5.1 Simulation Setup
5.2 Comfort Metrics
We employ an EnergyPlus model of a five zone commercial building from the reference building database of US DoE: Office of Energy Efficiency and Renewable Energy (2011), with a total surface of 500 m2 . The optimization is run during the cooling season (mainly the summer months) on
We evaluate comfort levels using the Predicted Percentage of Dissatisfied (PPD) index which is the predicted percentage of people among a large group who will be dissatisfied with the thermal environment, based on ASHRAE (2009). PPD is a function of deviations from an ideal temperature. 10
IFAC ADCHEM 2015 June 7-10, 2015. Whistler, BC, Canada Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
0.28
0.5
ACS tracking Revenue ACS Capacity Bid Revenue ACS Baseline cost Total Cost for no ACS tracking Total cost for ACS tracking
0.24
0.45
Baseline Cost Revenue from ACS tracking
0.4
Operational Cost (CHF/m 2 )
0.2
Cost (CHF/m2 )
11
0.16 0.12 0.08 0.04 0
0.35 0.3 0.25 0.2 0.15 0.1
-0.04
0.05
-0.08 25
26
27
28
29
30
31
32
33
34
0
35
0.05
0.065
0.08
0.095
0.11
0.125
0.14
0.155
0.17
0.185
0.2
Retail Price of Electricity (CHF/kWH)
Week of the year
Fig. 8. Tracking revenues and cost of energy for different electricity prices for variable constraints (night setback).
Fig. 6. Overall cost of AS vs no AS participation, per week. 0.24
0.2
0.5
0.18
0.45
0.16 0.14 0.12 0.1 0.08 0.06 0.04
ACS tracking Minimum cost (no ACS tracking)
0.02 0
6
Baseline Cost Revenue from ACS tracking
0.4
Operational Cost (CHF/m 2 )
Cost per week (CHF/m 2 )
0.22
6.5
7
7.5
8
8.5
9
9.5
10
10.5 11
11.5 12
0.35 0.3 0.25 0.2 0.15 0.1 0.05
12.5 13
Percentage of People Dissatisfied (%)
0
0.05
0.065
0.08
0.095
0.11
0.125
0.14
0.155
0.17
0.185
0.2
Retail Price of Electricity (CHF/kWH)
Fig. 7. Pareto curve of comfort vs cost for the same weeks. Fig. 9. Tracking revenues and cost of energy for different electricity prices for constant constraints.
The metric depends also on many other comfort quantities (such as humidity, irradiation, metabolic rate, etc) which we consider fixed. This provides the PPD for each zone of the building at every time period. To acquire a metric for the total building comfort throughout a week we resort to the long-termed percentage of dissatisfied (LPD) as recommended in Carlucci (2013)
LPD =
N
k=1
Z
N
z=1
k=1
pz,k PPDz,k
Z
z=1
pz,k
average and 50% of the maximum power consumption of the building). The resulting temperature profile for all zones, when the actual ACS is received, seen in Figure 5, rarely violate constraints and when it does the violation is usually not significant. We compare the overall energy cost of operating this building (0.168 CHF/m2 ) for a specific week to minimum cost without ACS tracking (0.192 CHF/m2 ), for a retail electricity price of 0.10 CHF/kWh (close to the minimum price of electricity in Switzerland for a commercial building). The overall savings are in the order of 13% with the capacity bid resulting in a revenue of approximately 0.054 CHF/m2 and the tracking bonus in 0.0055 CHF/m2 . To provide AS the building had to consume 13.2 W/m2 of power on average compared to 11.4 W/m2 when no tracking was involved. This is mainly due to procuring energy during non-working hours in order to provide a feasible baseline for the magnitude of the ACS bid.
(10)
where k is the time step, z is the zone index, and pz,k is the occupancy rate. We use this quantity only as an output metric to evaluate comfort levels and do not impose it as a constraint. 5.3 Results Our objective is to determine under which conditions it is preferable for a building to participate in ACS tracking and what is the expected benefit. We compare against the situation where the building attempts to minimise energy costs without ACS tracking, a scheme studied extensively in the literature. The result is assessed both in terms of comfort and cost.
Similar results are obtained for all the summer weeks of 2013. In Figure 6 we observe the variation in revenue and cost for nine different summer weeks (for a retail price of electricity of 0.10 CHF/kWh). On average ACS tracking has an overall cost of 0.126CHF/m2 compared to 0.154 CHF/m2 for minimum cost without ACS tracking, an average improvement of 18%. Most of the revenue comes from the capacity bid rather than the bonus for tracking the real-time signal. A Pareto curve of the comfort
We run our algorithm for a specific summer week (30th of 2013). The baseline declared by the algorithm, displayed in Figure 4, allows it to bid a capacity of 16 W/m2 for the following week (which is more than 140% of the 11
IFAC ADCHEM 2015 12 June 7-10, 2015. Whistler, BC, Canada Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
attained vs cost is depicted in Figure 7. We see that ACS tracking can provide better comfort levels for the same temperature constraints at a reduced cost.
0.24 minimum cost (no ACS tracking) ACS tracking
0.22
As we have mentioned, the retail price of electricity plays a crucial role in the participation of the building in the secondary control market. In Figure 8 we can see the sensitivity of the ACS revenue as the price increases. At approximately 0.155 CHF/kWh the algorithm considers tracking ACS not profitable enough compared to the costs of procuring an expensive baseline and decides to bid zero capacity in the market. Alternatively, if we consider a building with comfort constraints at all times, then we always have to buy a minimum of power. This allows ACS tracking without significant sensitivity on the price of electricity since the building is not using much more power on average (for most cases up to 5% more). In Figure 9 we observe that for a building with active constraints throughout the week the optimization algorithm will always choose participation in the AS market. The benefit of participation, as a percentage of overall costs is diminishing because the cost of electricity is becoming very large, but the absolute value of the revenue remains virtually the same.
Week cost (CHF/m 2 )
0.2
0.18
0.16
0.14
0.12
6
8
10
12
14
16
18
20
22
% of people dissatisfied (PPD)
Fig. 10. Pareto curve of comfort levels vs cost for ACS tracking and no ACS tracking. ACKNOWLEDGEMENTS We would like to thank Swissgrid Ltd. for providing us with all the relevant data for our simulation studies.
It should be noted, that the building is buying energy at a very high price compared to the actual spot market because Swiss regulations currently require the baseline schedule to be purchased at retail rates. Access to the price levels of the spot market could provide considerable benefits for AS participation. Moreover, the building is being incentivized for tracking the ACS on a bonus calculated on the spot market (while purchasing electricity at higher rates).
REFERENCES ASHRAE (2009). Fundamentals. ASHRAE Handbook. ASHRAE, Atlanta, GA. Balandat, M., Oldewurtel, F., Chen, M., and Tomlin, C. (2014). Contract design for frequency regulation by aggregations of commercial buildings. In 2014 52nd Annual Allerton Conference on Communication, Control, and Computing (Allerton), 38–45. Carlucci, S. (2013). Thermal comfort assessment of buildings. Springer. Crawley, D.B., Lawrie, L.K., Winkelmann, F.C., Buhl, W.F., Huang, Y.J., Pedersen, C.O., Strand, R.K., Liesen, R.J., Fisher, D.E., Witte, M.J., and Glazer, J. (2001). EnergyPlus: creating a new-generation building energy simulation program. Energy and Buildings, 33(4), 319–331. Gorecki, T.T., Qureshi, F.A., and Jones, C.N. (2014). OpenBuild : An Integrated Simulation Environment for Building Control. Technical report, EPFL. Hao, H., Kowli, A., Lin, Y., Barooah, P., and Meyn, S. (2013). Ancillary service for the grid via control of commercial building hvac systems. In American Control Conference (ACC), 2013, 467–472. IEEE. L¨ofberg, J. (2004). Yalmip : A toolbox for modeling and optimization in MATLAB. In Proceedings of the CACSD Conference. Taipei, Taiwan. Maasoumy, M., Rosenberg, C., Sangiovanni-Vincentelli, A., and Callaway, D. (2014). Model predictive control approach to online computation of demand-side flexibility of commercial buildings HVAC systems for supply following. In American Control Conference (ACC), 2014, 1082–1089. Maasoumy, M., Ortiz, J., Culler, D., and SangiovanniVincentelli, A. (2013). Flexibility of commercial building hvac fan as ancillary service for smart grid. arXiv preprint arXiv:1311.6094. Pavlak, G.S., Henze, G.P., and Cushing, V.J. (2014). Optimizing commercial building participation in energy
Finally, we compare the LPD comfort levels offered by ACS tracking versus no ACS tracking by performing the following simulation test: we gradually tighten the temperature constraints towards the ideal temperature. This way the minimum cost approach creates better comfort levels (but at an increased cost) and the ACS tracking is forced to declare its baseline close to where the ideal temperature is manifested (with less room for large ACS signals). The results are depicted in Figure 10. We can observe that ACS tracking achieves the same level of LPD with considerably lower costs, except when the temperature constraints are very tight around the ideal level (where ACS tracking is not operationally possible).
6. CONCLUSION Participation of buildings in the secondary AS market is highly sensitive on the price at which they procure electricity. For the lowest of the prices for commercial buildings, currently seen in Switzerland, ACS tracking improves the comfort levels considerably and the overall costs modestly. For price levels close to the spot market the economical benefits could be also significant. The two stage stochastic optimization algorithm can produce both the optimal bid (in terms of economic performance) and the suitable baseline consumption to be able to track the ACS as it arrives. 12
IFAC ADCHEM 2015 June 7-10, 2015. Whistler, BC, Canada Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
and ancillary service markets. Energy and Buildings, 81, 115–126. Qureshi, F.A., Gorecki, T.T., and Jones, C.N. (2014). Model predictive control for market-based demand response participation. In Proceedings of the 19th IFAC World Congress, volume 19, 11153–11158. Cape Town, South Africa. US DoE: Office of Energy Efficiency and Renewable Energy (2011). Commercial Reference Buildings. Vrettos, E., Lai, K., Oldewurtel, F., and Andersson, G. (2013). Predictive control of buildings for demand response with dynamic day-ahead and real-time prices. In Control Conference (ECC), 2013 European, 2527– 2534. Vrettos, E., Oldewurtel, F., Andersson, G., and Zhu, F. (2014). Robust Provision of Frequency Reserves by Office Building Aggregations. In Proceedings of the 19th IFAC World Congress. Cape Town, South Africa.
13
13
ScienceDirect IFAC-PapersOnLine 48-8 (2015) 006–013
Providing ancillary Providing ancillary Providing ancillary Providing ancillary buildings: the buildings: the buildings: buildings: the the
service with commercial service with commercial service with serviceperspective with commercial commercial Swiss Swiss perspective Swiss Swiss perspective perspective
Lymperopoulos ∗∗ Faran A. Qureshi ∗∗ Truong Nghiem ∗∗ Lymperopoulos Faran A. Qureshi Nghiem ∗ Truong ∗ ∗∗ ∗ ∗ Truong Lymperopoulos Faran A. Qureshi Nghiem Ali Ahmadi ∗∗Khatir N. Jones ∗ Lymperopoulos Faran ∗∗ A.Colin Qureshi Truong Nghiem ∗ Ali Ahmadi Khatir ∗∗ Colin N. Jones ∗ Ali Ali Ahmadi Ahmadi Khatir Khatir ∗∗ Colin Colin N. N. Jones Jones ∗ ∗ Automatic Control Lab, Ecole Polytechnique Federal de Lausanne, ∗ ∗ Automatic Control Lab, Ecole Polytechnique Federal de Lausanne, ∗Lausanne, Automatic Control Lab, Ecole Polytechnique Federal CH-1015 Switzerland (e-mail: ioannis.lymperopoulos@-, AutomaticCH-1015 Control Lab, Ecole Polytechnique Federal de de Lausanne, Lausanne, Lausanne, Switzerland (e-mail: ioannis.lymperopoulos@-, Lausanne, CH-1015 Switzerland (e-mail: ioannis.lymperopoulos@-, faran.qureshi@-, xuan.nghiem@-, [email protected]). Lausanne, CH-1015 Switzerland (e-mail: ioannis.lymperopoulos@-, faran.qureshi@-, xuan.nghiem@-, [email protected]). ∗∗ faran.qureshi@-, xuan.nghiem@-, [email protected]). Ltd, Laufenburg, CH-5080 Switzerland (e-mail: ∗∗ Swissgrid faran.qureshi@-, xuan.nghiem@-, [email protected]). ∗∗ Swissgrid Ltd, Laufenburg, CH-5080 Switzerland (e-mail: ∗∗ Swissgrid Ltd, Laufenburg, CH-5080 [email protected]) Swissgrid Ltd, Laufenburg, CH-5080 Switzerland Switzerland (e-mail: (e-mail: [email protected]) [email protected]) [email protected])
Ioannis Ioannis Ioannis Ioannis
Abstract: Abstract: Abstract:services constitute the cornerstone of the power grid. They allow for an efficient system Ancillary Abstract: Ancillary services constitute the cornerstone of the power grid. They allow for an efficient system Ancillary services of the grid. They for system operation, provideconstitute resiliencethe to cornerstone uncertainties establish unprecedented Ancillary services constitute the cornerstone of and the power power grid.safeguards They allow allowagainst for an an efficient efficient system operation, provide resilience to uncertainties and establish safeguards against unprecedented operation, provide resilience to uncertainties and establish safeguards against unprecedented events. Their importance is growing due to the rise of grid decentralisation and integration of operation, provide resilience to uncertainties and establish safeguards against unprecedented events. Their importance is growing due to the rise of grid decentralisation and integration of events. Their importance is growing due to the rise of grid decentralisation and integration of intermittent, renewable power sources, which lead to more variability and uncertainty in the events. Their renewable importancepower is growing duewhich to thelead rise of more grid decentralisation and integration of intermittent, sources, variability and uncertainty in intermittent, renewable power sources, which lead to to more variability and uncertainty in the the system. Today, the vast power share sources, of ancillary services is more provided by largeand generating units. An intermittent, renewable which lead to variability uncertainty in the system. Today, the vast share of ancillary services is provided by large generating units. An system. Today, the vast ancillary services is large units. ongoing effort by and of business entities focuses on usingby of loads connected system. Today, theresearch vast share share of ancillary services is provided provided byvariation large generating generating units. An An ongoing effort by research and business entities focuses on using variation of loads connected ongoing effort by research and business entities focuses on using variation of loads connected to the power grid in order to increase significantly the provision of such services, hopefully at a ongoing effort by research and business entities focuses on using variation of loads connected to the power grid in order to increase significantly the provision of such services, hopefully at aa to in to the such services, at reduced cost. grid We examine from ansignificantly economic perspective, theof of commercial buildings to the the power power grid in order orderhere, to increase increase significantly the provision provision ofuse such services, hopefully hopefully atas a reduced cost. We examine here, from an economic perspective, the use of commercial buildings as reduced cost. cost. We providers examine here, here, from an economic perspective, the use of of commercial commercial buildings as ancillary service based on an realeconomic prices from the Swissthe electricity market. We calculate reduced We examine from perspective, use buildings as ancillary service providers based on real prices from the Swiss electricity market. We calculate ancillary providers based on real prices from Swiss market. We calculate the effectservice of retail electrical prices economic performance of a building that ancillary service providers based on on realthe prices from the the Swiss electricity electricity market.and We find calculate the effect of retail electrical prices on the economic performance of aa building and find that the effect of retail electrical prices on the economic performance of building and find that for the rates charged in the least expensive cantons a single building can reduce its overall the effect of retail electrical prices on the economic performance of a building and find that for the rates charged in the least expensive cantons a single building can reduce its overall for the rates charged in the least expensive cantons a single building can reduce its overall energy costs, when participating in the ancillary services market. For the high end of prices this for the costs, rates when charged in the least expensive cantons a market. single building can reduce its overall energy participating in the ancillary services For the high end of prices this energy costs, when participating the ancillary services end of prices this gradually becomes prohibitive butin can be alleviated for a market. buildingFor thatthe hashigh a need for electricity energy costs, when participating in the ancillary services market. For the high end of prices this gradually becomes prohibitive but can be alleviated for a building that has a need for electricity gradually becomes prohibitive but can be alleviated for a building that has a need for electricity during nighttime hours, as well as daytime. Finally, we show, the counter-intuitive result that gradually becomes prohibitive but can be alleviated for a building that has a need for electricity during nighttime hours, as well as daytime. Finally, we show, the counter-intuitive result that during nighttime as as show, counter-intuitive result that providing ancillaryhours, services can increase the Finally, comfort we levels of athe building at a decreased cost. during nighttime hours, as well well as daytime. daytime. Finally, we show, the counter-intuitive result that providing ancillary services can increase the comfort levels of a building at a decreased cost. providing ancillary services can increase the comfort levels of a building at a decreased cost. providing ancillary services can increase the comfort levels of a building at a decreased cost. © 2015, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved. Keywords: Ancillary services, stochastic programming, buildings Keywords: Ancillary services, services, stochastic programming, programming, buildings Keywords: Keywords: Ancillary Ancillary services, stochastic stochastic programming, buildings buildings 1. INTRODUCTION and reduce energy costs. For this reason, they are also 1. INTRODUCTION INTRODUCTION and reduce energy costs. For this reason, they are also 1. and reduce energy this they also being envisioned as costs. storageFor elements that could increase 1. INTRODUCTION and reduce energy costs. For this reason, reason, they are are also being envisioned as storage elements that could increase being envisioned as storage elements that could increase the performance and the robustness of the power grid Accommodating for the fluctuating mismatch between being envisioned as storage elements that could increase performance and the robustness of the power grid Accommodating for for the the fluctuating fluctuating mismatch mismatch between between the the performance and robustness of the grid ancillary service providers without Accommodating electrical consumption andfluctuating generation,mismatch either because of by the becoming performance and the the robustness of (ASP), the power power grid Accommodating for the between by becoming ancillary service providers (ASP), without electrical consumption and generation, either because of by becoming ancillary service providers (ASP), without the need for investing in new generation capacity (or electrical consumption and generation, either because of operational uncertainties or large irregular events, is perby becoming ancillary service providers (ASP), without electrical consumption and generation, either because of the need for investing in new generation capacity (or operational uncertainties uncertainties or or large large irregular irregular events, events, is is perper- underutilising the need for investing in new generation capacity (or the existing one). operational formed at different hierarchical levels. A spot market, the need for investing in new generation capacity (or operational uncertainties or large levels. irregular events, is per- underutilising the existing one). formed at different hierarchical A spot market, underutilising the existing one). formed at different hierarchical levels. A spot market, usually operated on hierarchical an hourly basis (but also cleared a underutilising the existing one). formed at different levels. A spot market, building accumulates energy through its thermal causually operated operated on on an an hourly basis basis (but (but also also cleared cleared aa A building accumulates energy through its thermal causually day in advance), matches or alters the (but generation schedulea A usually operatedmatches on an hourly hourly basis also cleared A building energy through its thermal capacity. Thisaccumulates slowly dissipating employed day in advance), or alters the generation schedule A building accumulates energy energy throughcan its be thermal capacity. This slowly dissipating energy can be employed day in advance), matches or alters the generation schedule of power plants to meet the projected demand of load day in advance), matches or alters the generation schedule pacity. This slowly dissipating energy can be employed to shift the electrical load of the HVAC (Heating, Venof power plants to meet the projected demand of load pacity. This slowly dissipating energy can be employed the load HVAC (Heating, Venof power plants to the projected demand load serving The grid operator, due to to of powerentities plants (LSE). to meet meet thepower projected demand of of load to shift shift and the electrical electrical load of of the the HVAC (Heating, Ventilation Air-Conditioning) system to more beneficial serving entities (LSE). The power grid operator, operator, due to to shift the electrical load of the HVAC (Heating, Ventilation and Air-Conditioning) system to more beneficial serving entities (LSE). The power to the inherent unpredictabilities in thegrid actual deliverydue of this serving entities (LSE). The power grid operator, due to tilation and Air-Conditioning) system to more beneficial time windows (a demand response service) or to absorb the inherent unpredictabilities in the actual delivery of this tilation and Air-Conditioning) system to more beneficial windows (a demand response service) or to absorb the inherent in actual delivery of as this transaction, procures additional generation capacityof a time the inherent unpredictabilities unpredictabilities in the the actual delivery this time windows (a response service) to arising from tracking that transaction, procures additional additional generation capacity as as a fluctuations time windows (a demand demand responseaa regulation service) or orsignal to absorb absorb fluctuations arising from tracking regulation signal that transaction, procures generation capacity a backup to meet the mismatch at faster time scales, when transaction, procures additional generation capacity as a fulfils fluctuations arising from tracking a regulation signal an AS obligation towards the grid operator. The backup to meet the mismatch at faster time scales, when arising from towards tracking the a regulation signal that that fulfils an AS obligation grid operator. The backup to meet the mismatch at faster time scales, when required. These services are named ancillary services (AS) fluctuations backup to meet the mismatch at faster time scales, when fulfils an an AS obligation obligation towards the grid grid operator. operator. The critical constraint is to ensure a comfortable environment required. These services are named ancillary services (AS) fulfils AS towards the The critical constraint is to ensure aa comfortable environment required. These services are named ancillary services (AS) and are commonly procured from large generation units, required. These services are named ancillary services (AS) critical constraint is to ensure comfortable environment to the occupants while providing the promised service. and are commonly procured from large generation units, critical constraint is to ensure a comfortable environment the occupants while providing the promised service. and are procured from large units, operated at lower than nominal at all times to to and are commonly commonly procured fromcapacity large generation generation units, the while providing the promised operated at lower lower than than nominal capacity at all all times times to to to the occupants occupants whileexplored providingthe thebenefits promisedofservice. service. operated at nominal capacity at to Several studies have accommodate for the fluctuating regulation signal of the operated at lower than nominal capacity at signal all times to Several studies have explored the benefits of buildings buildings accommodate for the fluctuating regulation of the Several studies have explored the benefits of accommodate participating in have electricity markets, usually by adapting grid operator. for studies explored the benefits of buildings buildings accommodate for the the fluctuating fluctuating regulation regulation signal signal of of the the Several participating in electricity markets, usually by adapting grid operator. participating in electricity adapting grid operator. to the variations of electricalmarkets, price to usually achieve by a less costly participating in electricity markets, usually by adapting grid operator. the variations of electrical price to achieve a less costly Commercial buildings are large consumers of electrical to to the variations of electrical price to achieve a less operation by shifting load. Qureshi et al. (2014) investigate Commercial buildings are large consumers of electrical to the variations of electrical price et to al. achieve a investigate less costly costly by shifting load. Qureshi (2014) Commercial buildings of electrical power and are movingare to large more consumers sophisticated levels of operation Commercial buildings are large consumers of electrical operation by by shifting shifting load. Qureshi et al. al. (2014) (2014) investigate participation in the load. New Qureshi York demand response (DR) power and and are are moving moving to to more more sophisticated sophisticated levels levels of of operation et investigate participation in the New York demand response (DR) power automation in order to increase oversight, comfort levels power and are moving to moreoversight, sophisticated levels of participation the New York demand response (DR) program usingin model predictive control (MPC), Vrettos automation in order to increase comfort levels participation in the New York demand response (DR) program using model predictive control (MPC), Vrettos automation in order to increase oversight, comfort levels automation in order to increase oversight, comfort levels et program using model predictive control (MPC), Vrettos al. (2013) study DR for residential buildings coupled program using model predictive control (MPC), Vrettos This work has received support from the Swiss National Science et al. (2013) study DR for residential buildings coupled et (2013) study DR residential buildings coupled with storage. approaches optimise the This work has received support from the Swiss National Science et al. al.extra (2013) study These DR for for residentialmostly buildings coupled with extra storage. These approaches mostly optimise the Foundation under the GEMS project (grant number 200021-137985) This work has received support from the Swiss National Science with extra storage. These approaches mostly optimise This workunder has received support from the number Swiss National Science operation ofstorage. the building without a priori promises to the Foundation the GEMS project (grant 200021-137985) with extra These approaches mostly optimise the operation of the building without a priori promises to the and the European Research Council under the European Unions SevFoundation under the GEMS project (grant number 200021-137985) Foundation under Research the GEMS projectunder (grant number 200021-137985) operation of the building without aa priori promises to the grid operator. and the European Council the European Unions Sevoperation of the building without priori promises to the enth Framework Programme (FP/2007-2013) / ERC Grant Agreeand the European Research Council under the European Unions Sevgrid operator. and the EuropeanProgramme Research Council under the European Unions Sevgrid operator. enth Framework (FP/2007-2013) / ERC Grant Agreegrid operator. ment n. 307608 (BuildNet) enth Framework Programme (FP/2007-2013) / ERC Grant Agree-
enth Programme (FP/2007-2013) / ERC Grant Agreement Framework n. 307608 (BuildNet) ment ment n. n. 307608 307608 (BuildNet) (BuildNet) Copyright © 2015, 2015 IFAC 6 Hosting by Elsevier Ltd. All rights reserved. 2405-8963 © IFAC (International Federation of Automatic Control) Copyright © 2015 IFAC 6 Copyright © 2015 IFAC 6 Peer review under responsibility of International Federation of Automatic Copyright © 2015 IFAC 6 Control. 10.1016/j.ifacol.2015.08.149
IFAC ADCHEM 2015 June 7-10, 2015. Whistler, BC, Canada Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
7
quency is outside of a deadband, which is set ±10 mHz in continental Europe.
ASPs have to commit before being able to provide a service to the power grid (e.g. increasing or decreasing consumption at request) that is determined in advance (here weekly). Based on this commitment the grid operator sends a high frequency regulation signal, according to his needs, that the ASP has to follow while also fulfilling his primary operational objectives (for a building that would be delivering comfort to occupants at a low cost). Buildings providing AS have been explored in Maasoumy et al. (2014), Maasoumy et al. (2013), Vrettos et al. (2014), Pavlak et al. (2014), Hao et al. (2013) and Balandat et al. (2014). These studies examine the capacity for providing such services with artificial economic criteria and for a limited prediction horizon (at most daily).
Frequency Restoration Reserve (FRR) control is a centralized and continuous control (i.e., is used in both normal operation and in contingency cases) to restore the frequency and exchanges with other control areas to their target values. Such an indispensable frequency control can be implemented automatically (aFRR) with a central proportional-integral (PI) controller, or manually (mFRR) by directly calling providers. Replacement Reserve (RR) is a manually activated product that is used i) for further imbalances in the case FRR has already been activated and is unable to restore the frequency to its target value, ii) to release FRR in the case of large disturbances and iii) to anticipate expected imbalances.
We use here real economic data from the Swiss EPEX market (from 2013) and explore commitments that have a weekly horizon. We employ a two stage stochastic formulation, that allows us to exploit certain statistical properties of the regulation signal. We find this approach to be effective while also satisfying constraints when tested against real regulation signals. Moreover, we try to assess the economic feasibility, of a building providing AS. For this, we study how the retail price of electricity plays a significant role in the decision for participation. For a single unit this is due to variable constraints (night setback), between working and non-working hours, that create a requirement for the procurement of electricity even at times that the building might not usually need it. We find that a building, or an aggregation, that requires some energy consumption at all times during the week would always have an incentive to participate in the secondary control market for even the highest, realistic, prices of electricity.
In this paper, we focus on the Swiss Ancillary Services market and an aFRR product, which is known as the Secondary Control Reserve (SCR). SwissGrid, the operator of Swiss electrical grid, procures about ±400 MW SCR in a weekly auction for every hour of the following week from a set of pre-qualified Ancillary Service Providers (ASPs), which can be either loads or generators. During real-time operation, SwissGrid activates the procured SCR through an Automatic Generation Controller (AGC) by sending each second an activation signal in parallel to all ASPs that have been accepted in the secondary control market. Note that the activation signal, which is called the Area Control Signal (ACS) in this paper, is proportional to the awarded capacities of all accepted ASPs.
Finally, we evaluate the comfort levels attained, by using standard long-term satisfaction metrics and find that participation in AS improves the average comfort satisfaction levels compared to operating the building to minimise costs without AS participation.
We detail the operational and financial structures of the Swiss secondary ancillary services system in the following sections.
In Section 2 we introduce the Swiss AS market and present some of the statistical qualities of the frequency regulation signal. In Section 3 we present the methodology for acquiring a simulation model for the building, while in Section 4 we examine how a commercial bidding can bid in the AS market while operating inside its comfort constraints. Finally, in Section 5 we present our simulation set-up, describe the long term comfort metric we use and explain our results.
At the start of each week, the provider of ancillary services (the building) must submit a capacity bid, α MW, and a baseline energy schedule e¯ = {¯ ek }, where e¯k is measured in units of MWh and is provided in 15min time increments for the entire week.
2.1 Swiss Secondary Ancillary Services Market
Throughout the paper we take the perspective of the service provider, and adopt the convention that energy is positive when it flows into the service provider and that monetary values are positive when they flow out from the provider.
2. SECONDARY ANCILLARY SERVICES IN SWITZERLAND
If the baseline is negative (the service provider is a generator), then the baseline is a production schedule, and is sold on the spot market at the hourly spot rate. However, if the service provider is a consumer, as we assume throughout this paper, then the baseline energy is purchased from the retail market, at fixed retail rates. As a result, we can define the cost to the service provider for the baseline schedule e¯ as Cbaseline (¯ e) := cretail e¯k
According to The European Network of Transmission System Operators for Electricity (ENTSO-E) definition, three levels of frequency control are generally used to maintain the real-time balance between load and generation. Frequency Containment Reserve (FCR) control is a local and decentralized proportional controller that contains the system frequency within a maximum deviation bound following large generation or load outages in a synchronous area of continental Europe. Note that FCR is only activated in the case of contingency when the system fre-
where cretail is the retail price of electricity.
The financial reward paid to the provider for offering a capacity of α MW is based on the bid price of the provider 7
IFAC ADCHEM 2015 June 7-10, 2015. Whistler, BC, Canada 8 Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
100 spot price ACS rk
90
70
# of occurrences
Price of Electricity (CHF/MWh)
ACS
ck
80
60 50 40 30 20 10 0
0
1
2
3
4
5
6
-400
7
-300
-200
-100
0
100
200
300
400
Requested MW
time (days)
Fig. 1. Incentives for tracking the ACS.
Fig. 2. Distribution of the averaged 15 min ACS for 2013
(pay-as-bid), and is proportional to α. Rcapacity (α) := ccapacity · α (1) For the purposes of the study in this paper, we fix the bid price of the provider ccapacity to be the average of accepted bids in the market, measured in CHF / MW.
SwissGrid levies a penalty against providers for tracking imbalances. Using a similar argument to the previous section, different penalties are paid for positive and negative deviation: penalty Cpenalty () = ck max{k , 0} − rkpenalty max{−k , 0} (4)
2.2 Real-time Operation
and rkpenalty are the cost and rebate paid for where cpenalty k tracking deviations.
Each 15 minute period, SwissGrid measures the error in energy between the ACS ak and the energy consumed / produced by the provider ek , and a financial remuneration / penalty is paid / imposed with a price that is coupled to the Swiss spot market price (SwissIX). This financial adjustment is based on two components: An incentivisation price for energy consumed / produced as a result of tracking the ACS, and a penalty price for deviations against this signal.
The reader is referred to the SwissGrid website 1 for full details on the incentivisation and imbalance penalties involved. 2.3 Statistical Properties of the ACS Important elements of the ACS can be captured by analysing the statistics of the signal provided by Swissgrid. We restrict our analysis here to the average ACS over a 15 min period (the actual signal comes every 1 sec with a clearing of 15 min), since we assume that a lower level controller captures the high frequency deviations, while our optimization scheme provides the mean level of power at each time period. We also find that the time correlation of the ACS is strong for this time period (0.55 after 15 min), so that the signals do not differ excessively.
Incentivisation If the ACS ak is positive, then the provider is being asked to consume more energy, and so this energy is sold at a discount. If the signal ak is negative, then the request is for a reduction in consumption. However, the provider has already purchased this energy at retail rates in the form of its baseline schedule and as a result, SwissGrid returns a rebate to the provider for the energy not used. The reward for tracking an ACS {ak } is therefore −cACS max{ak , 0} + rkACS max{−ak , 0} RACS (a) := k (2)
The distribution of the ACS throughout 2013 can be observed in Figure 2. The maximum absolute magnitude of the ACS is 400 MW, however the mean of the absolute magnitude is roughly 50 MW. The ACS is almost zero mean, with a small negative bias of almost -11 MW, however this varies per week and especially per day (there exist days for which the ACS is almost entirely positive or entirely negative).
where cACS and rkACS are the cost and rebate paid for enk ergy respectively. Figure 1 shows the relationship between these various prices for the 30th week of 2013.
From the probability density distribution of the signal we can deduce that there exist only a few significant outliers, while the mass of the signal is concentrated close to zero. The standard deviation of the ACS is almost 70 MW, while the kurtosis of the distribution is higher (∼ 5) than that of the normal distribution.
Note that the reward for ACS tracking may be either positive or negative, depending on whether the service provider gains energy as a result of tracking the signal, or produces it. Imbalance Penalties The tracking imbalance k between the ACS ak , and the energy consumed / produced by the provider ek at time k is (3) k = ek − e¯k − ak
By saturating tracking for very extreme outliers (track an outlier up to a portion of its full magnitude) we can protect 1 http://www.swissgrid.ch/swissgrid/en/home/experts/topics/ ancillary_services.html
8
IFAC ADCHEM 2015 June 7-10, 2015. Whistler, BC, Canada Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
input (external air temperature, sky temperature, solar radiation on each building surface, long-wave radiation, internal gain, etc.), and yk ∈ Rq is the room air temperature at time step k. The total thermal energy input for the building ek is then given by (6) ek = η(1T uk ) where we assume a constant electrical to thermal conversion efficiently of η = 1.
100 Energy remaining in the ACS signal (%)
99 95
90
90
80 80
70
The building is deemed satisfactorily comfortable if certain temperature constraints are satisfied. This is used throughout our simulations and imposes linear constraints on the optimization problem. We define the comfort constraints at a level δ ◦ C as |yk − Tideal | ≤ δ, where Tideal ◦ C is the ideal temperature provided by the occupants (set to 24 ◦ C in the simulations).
60
Energy remaining in the ACS signal
50
Select Points - Remaining Energy (%)
40
0
10
20
30
40
50
60
70
9
80
% of magnitude saturated
Fig. 3. Power Remaining in the ACS after truncating outliers at different saturation levels the equipment, facilitate larger bids while also fulfilling the AS obligation at an acceptably high level of compliance. We track asat = min(ak , (1 − sat) · ak ), where ak is the k original ACS and sat ∈ [0, 1] the saturation level.
As has been reported several times in the literature, a minimum cost approach, with no ACS tracking, would tend to follow the upper or lower temperature constraints in order to minimize energy spend. However, an ACS tracking approach would attempt to place the temperature somewhere between the upper and lower constraints in order to provide a greater flexibility for ACS tracking, which will result in a greater comfort for the occupants. This counter-intuitive behaviour can be observed in Section 5.
In Figure 3, we observe that shaving 60% of the magnitude of outliers preserves more than 96% of the energy in the ACS. This means that tracking is still effective for a very large portion of the requests. Here, we restrict to a 40% saturation level that allows for tracking more than 99% of the energy contained in the ACS. This results on average to only one to three 15 min periods in a week where the tracking is suboptimal. Most operators (including SwissGrid) allow a small percentage of time where the tracking can be suboptimal. Thus, the tracking capacity (and the revenue) is increased by tracking truncated ACS and taking advantage of the allowed tracking suboptimality.
The building model, coupled with a forecast of weather and occupancy, allows the characterization of the set of all energy signals that the building is able to consume throughout the coming week, while still ensuring the comfort of its occupants. This comfort set is given as xk+1 = Axk + Bu uk + Bd dk |Cxk − Tideal | ≤ δ Bcomfort (δ) := {ek } uk ∈ U e = η(1T u ) k k (7)
3. BUILDING MODEL We use the MATLAB toolbox OpenBuild, see Gorecki et al. (2014), to extract a linear state-space model from a detailed EnergyPlus building model. EnergyPlus, Crawley et al. (2001), is a tool for comprehensive modeling of building thermodynamics and HVAC systems. However, this representation is highly detailed, and not suitable as a prediction model in an optimization-based control scheme. OpenBuild extracts all the required data (material properties, orientation of different surfaces, etc.) from an EnergyPlus model and uses it to automatically construct a first principles linear model of the building thermodynamics. Moreover, EnergyPlus is also used to compute the effect of the external weather conditions and the internal gains on the linear model. Once constructed, the linear model is again validated against EnergyPlus. For more details see Gorecki et al. (2014) .
4. THE BIDDING PROBLEM FOR COMMERCIAL BUILDINGS In this section, we formulate the weekly bidding problem that a building faces when participating in the secondary ancillary services market of Switzerland. The goal is to characterize the weekly set of capacity bids α and baseline energy consumptions e¯ such that the building can ensure comfort of its occupants with a high probability. Amongst this set, the building can then select the element that minimizes its expected cost of operation. From Section 2, the total financial cost of the service provider can be summarized as e) − Rcapacity (α) C(α, e¯, a, e) := Cbaseline (¯ + Cpenalty (e − e¯ − a) − RACS (a)
The linear model obtained from OpenBuild has 70 states and is reduced in order to 20 states using the Hankel-Norm based balanced truncation method. The reduced linear model is discretized with a sampling time of 15 min. to obtain a linear dynamic model (5) xk+1 = Axk + Bu uk + Bd dk yk = Cxk where xk ∈ Rn is the state, uk ∈ Rm is the thermal energy input to each of the m zones, dk ∈ Rp is the disturbance
We can see that given an ACS a to track and a baseline schedule e¯, the optimal action of the building is to minimize any imbalance penalty Cpenalty . We define the ◦ function Cpenalty as the minimum penalty that can be obtained subject to comfort of the building’s occupants ◦ (γ) := min Cpenalty (e − γ) Cpenalty e
s.t. e ∈ Bcomfort (δ) 9
IFAC ADCHEM 2015 10 June 7-10, 2015. Whistler, BC, Canada Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
We can now formally define the bidding problem as e) − Rcapacity (α) (8) min Cbaseline (¯ e¯,α ◦ + Eaˆ Cpenalty (¯ e + αˆ a) − RACS (αˆ a) where the expectation is taken over the distribution of the normalized ACS a ˆ.
32
The optimizer (¯ e , α ) of (8) is then the baseline schedule and capacity bid that will result in lowest expected operating costs of the building, subject to the occupants of the building remaining comfortable.
24 20 16 12
4 0
0
1
2
3
4
5
6
7
Time [Days]
The bidding problem (8) can be solved using a twostage stochastic programming approach. The first stage variables are those that must be provided to SwissGrid at the start of the week : the capacity bid α, and the baseline schedule e¯. The second stage variables are then the response of the building ej to the randomly drawn samples a ˆj of the normalized ACS.
Fig. 4. Baseline consumption declared for a week and the ACS superimposed when it is realised. 32 30
We can now pose the two-stage stochastic optimization problem approximating (8) e) − Rcapacity (α) (9) min Cbaseline (¯
Temperature ( oC)
28
e¯,α
s.t.
28
8
4.1 Computation : Two Stage Stochastic Programming
Ns 1 −RACS (αˆ aj ) + Ns j=1
xjk+1 = Axjk + Bu ujk |Cxjk − Tideal | ≤ δ ujk ∈ U ejk = η(1T ujk ) ajk ejk − e¯k = αˆ
Baseline ACS superimposed
36
Power Consumption [W/m2 ]
ACS Tracking
40
26 24 22 Upper Constraints Lower Constraints Zone 1 Zone 2 Zone 3 Zone 4 Zone 5
20
+ Bd dk
18 16
0
1
2
3
4
5
6
7
Time [Days]
Fig. 5. Zones temperature profile for time varying comfort constraint (night setback).
α≥0 where Ns is the number of samples drawn from the normalized distributed of the ACS. The horizon length of the optimization problem (9) is one week with a time step of 15 min. Note that a tracking constraint is added to ensure perfect tracking of the samples a ˆj of the normalized ACS. Thus, the imbalance penalty Cpenalty term is removed from the cost function.
a week long basis. Weather and occupancy disturbances are assumed to be perfectly forecast, while the ACS (provided by Swissgrid) is revealed causally and is unknown at the time of the bid. For the ACS scenarios we use the recorded ACS from the previous weeks. For the financial variables, we use spot prices from the 2013 Swiss market, bonus and penalties for tracking (or deviating from ACS) from Swissgrid for the same year and the average capacity bid per week as published.
We highlight that problem (9) can be re-cast as a linear programming problem, enabling the solution of very largescale problems. The constraints are clearly linear, and the functions Cbaseline and Rcapacity are also linear functions. From equations (2) and (4), we see that the functions Cpenalty and RACS are convex piecewise affine functions of their arguments, and can therefore be re-formulated as linear functions subject to linear constraints using a standard epigraph formulation.
The building model is run with a 15 min time step and the problem is solved using a YALMIP interface, L¨ ofberg (2004), to the Gurobi numerical solver. The number of ACS scenarios is quite limited (14 to 15) due to the extreme memory requirements from the numerous constraints generated for a week long simulation. Sensitivity studies suggest that this small number of scenarios is still representative in this case. The average time required for a single week optimization is approximately 5 min.
5. SIMULATION RESULTS 5.1 Simulation Setup
5.2 Comfort Metrics
We employ an EnergyPlus model of a five zone commercial building from the reference building database of US DoE: Office of Energy Efficiency and Renewable Energy (2011), with a total surface of 500 m2 . The optimization is run during the cooling season (mainly the summer months) on
We evaluate comfort levels using the Predicted Percentage of Dissatisfied (PPD) index which is the predicted percentage of people among a large group who will be dissatisfied with the thermal environment, based on ASHRAE (2009). PPD is a function of deviations from an ideal temperature. 10
IFAC ADCHEM 2015 June 7-10, 2015. Whistler, BC, Canada Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
0.28
0.5
ACS tracking Revenue ACS Capacity Bid Revenue ACS Baseline cost Total Cost for no ACS tracking Total cost for ACS tracking
0.24
0.45
Baseline Cost Revenue from ACS tracking
0.4
Operational Cost (CHF/m 2 )
0.2
Cost (CHF/m2 )
11
0.16 0.12 0.08 0.04 0
0.35 0.3 0.25 0.2 0.15 0.1
-0.04
0.05
-0.08 25
26
27
28
29
30
31
32
33
34
0
35
0.05
0.065
0.08
0.095
0.11
0.125
0.14
0.155
0.17
0.185
0.2
Retail Price of Electricity (CHF/kWH)
Week of the year
Fig. 8. Tracking revenues and cost of energy for different electricity prices for variable constraints (night setback).
Fig. 6. Overall cost of AS vs no AS participation, per week. 0.24
0.2
0.5
0.18
0.45
0.16 0.14 0.12 0.1 0.08 0.06 0.04
ACS tracking Minimum cost (no ACS tracking)
0.02 0
6
Baseline Cost Revenue from ACS tracking
0.4
Operational Cost (CHF/m 2 )
Cost per week (CHF/m 2 )
0.22
6.5
7
7.5
8
8.5
9
9.5
10
10.5 11
11.5 12
0.35 0.3 0.25 0.2 0.15 0.1 0.05
12.5 13
Percentage of People Dissatisfied (%)
0
0.05
0.065
0.08
0.095
0.11
0.125
0.14
0.155
0.17
0.185
0.2
Retail Price of Electricity (CHF/kWH)
Fig. 7. Pareto curve of comfort vs cost for the same weeks. Fig. 9. Tracking revenues and cost of energy for different electricity prices for constant constraints.
The metric depends also on many other comfort quantities (such as humidity, irradiation, metabolic rate, etc) which we consider fixed. This provides the PPD for each zone of the building at every time period. To acquire a metric for the total building comfort throughout a week we resort to the long-termed percentage of dissatisfied (LPD) as recommended in Carlucci (2013)
LPD =
N
k=1
Z
N
z=1
k=1
pz,k PPDz,k
Z
z=1
pz,k
average and 50% of the maximum power consumption of the building). The resulting temperature profile for all zones, when the actual ACS is received, seen in Figure 5, rarely violate constraints and when it does the violation is usually not significant. We compare the overall energy cost of operating this building (0.168 CHF/m2 ) for a specific week to minimum cost without ACS tracking (0.192 CHF/m2 ), for a retail electricity price of 0.10 CHF/kWh (close to the minimum price of electricity in Switzerland for a commercial building). The overall savings are in the order of 13% with the capacity bid resulting in a revenue of approximately 0.054 CHF/m2 and the tracking bonus in 0.0055 CHF/m2 . To provide AS the building had to consume 13.2 W/m2 of power on average compared to 11.4 W/m2 when no tracking was involved. This is mainly due to procuring energy during non-working hours in order to provide a feasible baseline for the magnitude of the ACS bid.
(10)
where k is the time step, z is the zone index, and pz,k is the occupancy rate. We use this quantity only as an output metric to evaluate comfort levels and do not impose it as a constraint. 5.3 Results Our objective is to determine under which conditions it is preferable for a building to participate in ACS tracking and what is the expected benefit. We compare against the situation where the building attempts to minimise energy costs without ACS tracking, a scheme studied extensively in the literature. The result is assessed both in terms of comfort and cost.
Similar results are obtained for all the summer weeks of 2013. In Figure 6 we observe the variation in revenue and cost for nine different summer weeks (for a retail price of electricity of 0.10 CHF/kWh). On average ACS tracking has an overall cost of 0.126CHF/m2 compared to 0.154 CHF/m2 for minimum cost without ACS tracking, an average improvement of 18%. Most of the revenue comes from the capacity bid rather than the bonus for tracking the real-time signal. A Pareto curve of the comfort
We run our algorithm for a specific summer week (30th of 2013). The baseline declared by the algorithm, displayed in Figure 4, allows it to bid a capacity of 16 W/m2 for the following week (which is more than 140% of the 11
IFAC ADCHEM 2015 12 June 7-10, 2015. Whistler, BC, Canada Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
attained vs cost is depicted in Figure 7. We see that ACS tracking can provide better comfort levels for the same temperature constraints at a reduced cost.
0.24 minimum cost (no ACS tracking) ACS tracking
0.22
As we have mentioned, the retail price of electricity plays a crucial role in the participation of the building in the secondary control market. In Figure 8 we can see the sensitivity of the ACS revenue as the price increases. At approximately 0.155 CHF/kWh the algorithm considers tracking ACS not profitable enough compared to the costs of procuring an expensive baseline and decides to bid zero capacity in the market. Alternatively, if we consider a building with comfort constraints at all times, then we always have to buy a minimum of power. This allows ACS tracking without significant sensitivity on the price of electricity since the building is not using much more power on average (for most cases up to 5% more). In Figure 9 we observe that for a building with active constraints throughout the week the optimization algorithm will always choose participation in the AS market. The benefit of participation, as a percentage of overall costs is diminishing because the cost of electricity is becoming very large, but the absolute value of the revenue remains virtually the same.
Week cost (CHF/m 2 )
0.2
0.18
0.16
0.14
0.12
6
8
10
12
14
16
18
20
22
% of people dissatisfied (PPD)
Fig. 10. Pareto curve of comfort levels vs cost for ACS tracking and no ACS tracking. ACKNOWLEDGEMENTS We would like to thank Swissgrid Ltd. for providing us with all the relevant data for our simulation studies.
It should be noted, that the building is buying energy at a very high price compared to the actual spot market because Swiss regulations currently require the baseline schedule to be purchased at retail rates. Access to the price levels of the spot market could provide considerable benefits for AS participation. Moreover, the building is being incentivized for tracking the ACS on a bonus calculated on the spot market (while purchasing electricity at higher rates).
REFERENCES ASHRAE (2009). Fundamentals. ASHRAE Handbook. ASHRAE, Atlanta, GA. Balandat, M., Oldewurtel, F., Chen, M., and Tomlin, C. (2014). Contract design for frequency regulation by aggregations of commercial buildings. In 2014 52nd Annual Allerton Conference on Communication, Control, and Computing (Allerton), 38–45. Carlucci, S. (2013). Thermal comfort assessment of buildings. Springer. Crawley, D.B., Lawrie, L.K., Winkelmann, F.C., Buhl, W.F., Huang, Y.J., Pedersen, C.O., Strand, R.K., Liesen, R.J., Fisher, D.E., Witte, M.J., and Glazer, J. (2001). EnergyPlus: creating a new-generation building energy simulation program. Energy and Buildings, 33(4), 319–331. Gorecki, T.T., Qureshi, F.A., and Jones, C.N. (2014). OpenBuild : An Integrated Simulation Environment for Building Control. Technical report, EPFL. Hao, H., Kowli, A., Lin, Y., Barooah, P., and Meyn, S. (2013). Ancillary service for the grid via control of commercial building hvac systems. In American Control Conference (ACC), 2013, 467–472. IEEE. L¨ofberg, J. (2004). Yalmip : A toolbox for modeling and optimization in MATLAB. In Proceedings of the CACSD Conference. Taipei, Taiwan. Maasoumy, M., Rosenberg, C., Sangiovanni-Vincentelli, A., and Callaway, D. (2014). Model predictive control approach to online computation of demand-side flexibility of commercial buildings HVAC systems for supply following. In American Control Conference (ACC), 2014, 1082–1089. Maasoumy, M., Ortiz, J., Culler, D., and SangiovanniVincentelli, A. (2013). Flexibility of commercial building hvac fan as ancillary service for smart grid. arXiv preprint arXiv:1311.6094. Pavlak, G.S., Henze, G.P., and Cushing, V.J. (2014). Optimizing commercial building participation in energy
Finally, we compare the LPD comfort levels offered by ACS tracking versus no ACS tracking by performing the following simulation test: we gradually tighten the temperature constraints towards the ideal temperature. This way the minimum cost approach creates better comfort levels (but at an increased cost) and the ACS tracking is forced to declare its baseline close to where the ideal temperature is manifested (with less room for large ACS signals). The results are depicted in Figure 10. We can observe that ACS tracking achieves the same level of LPD with considerably lower costs, except when the temperature constraints are very tight around the ideal level (where ACS tracking is not operationally possible).
6. CONCLUSION Participation of buildings in the secondary AS market is highly sensitive on the price at which they procure electricity. For the lowest of the prices for commercial buildings, currently seen in Switzerland, ACS tracking improves the comfort levels considerably and the overall costs modestly. For price levels close to the spot market the economical benefits could be also significant. The two stage stochastic optimization algorithm can produce both the optimal bid (in terms of economic performance) and the suitable baseline consumption to be able to track the ACS as it arrives. 12
IFAC ADCHEM 2015 June 7-10, 2015. Whistler, BC, Canada Ioannis Lymperopoulos et al. / IFAC-PapersOnLine 48-8 (2015) 006–013
and ancillary service markets. Energy and Buildings, 81, 115–126. Qureshi, F.A., Gorecki, T.T., and Jones, C.N. (2014). Model predictive control for market-based demand response participation. In Proceedings of the 19th IFAC World Congress, volume 19, 11153–11158. Cape Town, South Africa. US DoE: Office of Energy Efficiency and Renewable Energy (2011). Commercial Reference Buildings. Vrettos, E., Lai, K., Oldewurtel, F., and Andersson, G. (2013). Predictive control of buildings for demand response with dynamic day-ahead and real-time prices. In Control Conference (ECC), 2013 European, 2527– 2534. Vrettos, E., Oldewurtel, F., Andersson, G., and Zhu, F. (2014). Robust Provision of Frequency Reserves by Office Building Aggregations. In Proceedings of the 19th IFAC World Congress. Cape Town, South Africa.
13
13