Assessing the Energy Content of System Frequency and Electric Vehicle Charging Efficiency for Ancillary Service Provision

Transcription

1 Downloaded from orbit.dtu.dk on: Dec 9, 207 Assessing te Energy Content of System Frequency and Electric Veicle Carging Efficiency for Ancillary Service Provision Tingvad, Andreas; Ziras, Caralampos; Hu, Junjie; Marinelli, Mattia Publised in: Proceedings of te 52nd International Universities' Power Engineering Conference Publication date: 207 Document Version Peer reviewed version Link back to DTU Orbit Citation (APA): Tingvad, A., Ziras, C., Hu, J., & Marinelli, M. (207). Assessing te Energy Content of System Frequency and Electric Veicle Carging Efficiency for Ancillary Service Provision. In Proceedings of te 52nd International Universities' Power Engineering Conference IEEE. General rigts Copyrigt and moral rigts for te publications made accessible in te public portal are retained by te autors and/or oter copyrigt owners and it is a condition of accessing publications tat users recognise and abide by te legal requirements associated wit tese rigts. Users may download and print one copy of any publication from te public portal for te purpose of private study or researc. You may not furter distribute te material or use it for any profit-making activity or commercial gain You may freely distribute te URL identifying te publication in te public portal If you believe tat tis document breaces copyrigt please contact us providing details, and we will remove access to te work immediately and investigate your claim.

2 Assessing te Energy Content of System Frequency and Electric Veicle Carging Efficiency for Ancillary Service Provision Andreas Tingvad, Caralampos Ziras, Junjie Hu, Mattia Marinelli Center for Electric Power and Energy, Department of Electrical Engineering DTU - Tecnical University of Denmark, Roskilde, Denmark Contact: {ating, cazi, junu, matm}@elektro.dtu.dk Abstract Te purpose of tis paper is to quantify te effect of biased system frequency deviations and carger losses in order for an aggregation of electric veicles (EVs) to provide reliable primary frequency control (PFC). A data set consisting of one year of frequency measurements of te Nordic syncronous zone is used for te analysis. Te average system frequency can be biased over te our, wic can lead storage units, performing PFC, to become fully carged or depleted. Tis paper presents statistical bounds on ow variable te average system frequency can be on different time scales. Additionally, a metod for calculating te expected energy loss caused by continuous carging and discarging is presented togeter wit efficiency measurements of a commercial bidirectional EV carger. It is found tat during a year, te energy balance of te service provider, relative to te grid, is witin te calculated bounds. Te efficiency losses are calculated and validated to ave a linear relationsip wit te reserve capacity and te provision time. Index Terms Ancillary Services, Electric Veicles, Frequency Control, Veicle-to-Grid I. INTRODUCTION Te electrification of te transportation sector is expected to substantially increase te use of electric veicles (EVs) []. EVs can be aggregated and participate in te electricity markets in order to decrease teir carging costs by consuming energy wen prices are lower [2]. Te use of EVs for providing ancillary services to te power system can be an additional revenue for EV owners and can assist te integration of larger amounts of renewable sources [3, 4]. Te use of bidirectional power converters can substantially increase te capabilities and profitability of EVs, since tey are able to offer power to te system, a concept referred to as Veicle-to-Grid (V2G) [5, 6]. External V2G cargers will most likely be installed at commercial locations as tey can provide iger power and potentially iger efficiency tan te internal EV carger, wic for some models ave been found to ave an efficiency between 50 80% [7]. A suitable ancillary service for EVs is primary frequency regulation, because it is compensated per power capacity and te energy requirements are relatively small, wic is beneficial as EVs, wit te rigt carger can ave a large power capacity and very fast response [8]. In tis service, loads/generators are expected to modify teir consumption/production according to te frequency deviation signal in a linear way [9]. For verification and reliability purposes, te provided reserve is calculated based on a power reference, wic is te submitted scedule of te service provider to te operator. If te frequency deviation is unbiased on an ourly basis, i.e. its integral is zero, ten tere would be no overall excange of energy between te EVs and te grid. However, frequency deviations can be significantly biased in certain ours, wic could lead to relatively large energy excanges. If tese excanges are neglected, it is very likely tat te batteries will be fully carged or depleted witin te reserve provision orizon; in tat case, te aggregation will not be able to provide te committed reserves. An additional callenge is te efficiency of te cargers, wic causes energy losses to te EV batteries, and wic must be also accounted for wen bidding in te power markets. Te main contributions of tis paper are te treefold. First, an assessment of te Nordic power system s frequency bias on different time scales is conducted. Second, we quantify te carger losses in relation to te reserve capacity of frequency regulation, including its effect on te revenue. Te remainder of tis paper is organised as follows. In Section II, te revenue for frequency provision is calculated along wit a statistical analysis of te energy content and carger efficiency. Section II presents te calculated confidence intervals for te excanged energy and simulations based on te real frequency data are used for validation. Finally, Section IV is te concludes and discusses te direction of future work. II. METHODOLOGY A. Frequency Normal operation Reserve Te Transmission System Operators (TSOs) of te Regional Group Nordic (RG-N) syncronous zone ave a sared set of grid codes wit te purpose of maintaining system stability. Primary frequency control is divided into two services, Frequency Normal-operation Reserve (FNR), activated for all system frequency deviations up to ±00 mhz and Frequency Disturbance-operation Reserve (FDR), only activated wen

3 frequency goes below 49.9 Hz. FNR is te most fitting reserve for EVs, as te EV bot would carge and discarge, depending on te frequency, and tereby use te battery as a buffer rater tan a source of production. In te case of a frequency deviation, te purpose of FNR is to re-establis an equilibrium between production and consumption. Te TSOs in RG-N are jointly responsible for procuring 600 MW of FNR reserves, proportional to eac TSO s sare of te production. Te service is bougt on market terms or 2 days aead but can only be provided by a Balance Responsible Party (BRP) wit a minimum bid size of 0.3 MW. FNR is a symmetrical service, wic requires te provider to offer te same power capacity for upwards and downwards regulation. Frequency reserves must be provided linearly, wit full activation for deviations of ±00 mhz. As te service istorically as been provided by termal power plants, te maximum response time is relatively large, and equal to 50 s [9]. For a frequency value f t at time t, te normalised response y t is calculated as If P cap is expressed in kw, ten Ebias,k is expressed in kw. Te energy tat te BRP as delivered to te grid during an our is compensated wit te power price for upwards regulation, wic is always equal or iger tan te spot price. Te energy tat te BRP as received from te grid during an our is bougt wit te power price for downwards regulation, wic always equal or lower tan te spot price. Te different payments and compensations are sown in Fig. 2., if f t < 49.9 Hz y t = (f t 50)/0., if 49.9 Hz f t 50. Hz, if f t > 50. Hz () Fig. 2. Te different payments between BRPs and TSO in te case of FNR service Fig.. Droop control caracteristic of units providing FNR Te relationsip between y t and f t can also be described grapically, as sown in Fig.. FNR is paid in EURO per available power capacity, P cap, per our, independent of ow often and ow muc te reserve is activated. Te power required by te service provider at time t is calculated as P t = P cap y t (2) Since FNR is considered a power availability service, te BRPs are required to pay or receive money to/from te TSO if te average frequency during an our is different tan 50 Hz, wic results in an energy excange wit te grid. In tat case frequency is called biased, or equivalently tat frequency carries an energy content. Te energy content of te previous our is calculated by te TSO as in Eq. 3. Te energy content of every our is denoted by Ebias,k, were k is te our index. If t s is te sampling time, N is te number of samples in an our, Ebias,k is calculated as E bias,k = k N t=k (N )+ P t t s (3) B. System frequency beaviour over a year Our analysis as been based on a data set consisting of one year of system frequency measurements from RG-N wit a sample rate of 0 seconds, and a resolution of mhz, measured by te Norwegian TSO, Statnett. Fig. 3 sows a istogram of every sample of te system frequency in 206. Over te te grid frequency is Gaussian distributed wit an average of 50 Hz wic means tat te expected energy excange wit te grid is zero. C. Energy content of system frequency Assuming a constant reserve capacity P cap, te accumulated energy E t up to time step t relative to te grid since te beginning of service provision, can be calculated as E t = E t + P t t s (4) Fig. 3. Histogram of frequency measurements for 206, wic follow a Gaussian distribution f N ( Hz, σ 2 = Hz 2 )

4 Fig. 4 sows E bias,t, from te first of January and during te rest of te mont for P cap = kw. Overall, E t varies around 0, but it can cange substantially in te course of one day, wic is te time orizon of FNR auction and provision. energy tat guarantees robustness in e.g. 99% of te cases; in Table I te energy requirements for different confidence intervals are presented. In practice, tis must be done in a way tat is compatible wit te operator s requirements, to ensure system security. By allocating ±0.79 kw in te battery for every kw of FNR, te EV would be able to deliver FNR in an our, in 99% of te cases. TABLE I CONFIDENCE INTERVALS FOR THE HOURLY ENERGY BIAS IN kw Fig. 4. Accumulated energy for January 206 for P cap = kw Even toug E t can cange significantly trougout a long time period, we are interested in te ourly variations during eac day, because FNR is procured daily. It is terefore more relevant to investigate te energy bias on a sorter, i.e. ourly, time scale. Fig. 5 sows te energy bias of te individual ours, Ebias,k, of te first 3 days of 206, wen providing FNR wit P cap = kw. Te energy bias is calculated by (3) as te summation of te frequency deviations witin eac our. Confidence Minimum Expected value Maximum 99% % % Fig. 5 sows ow Ebias,k of consecutive ours often canges between positive and negative values, tereby cancelling out previous imbalances in many cases, resulting in a lower E t. Terefore, providing FNR for ours does not require times more storage capacity to guarantee tat te service can be provided. To quantify tis effect, te distribution of te energy bias is calculated for periods of ours. Te vector of year of frequency data is divided into slots of ours tat starts our apart. We denote te energy bias of consecutive ours by Ebias,k and we calculate it as E bias,k = N k t=n (k )+ P t t s (5) An example of ow te data is divided into 3-our periods wit te index k in a moving window fasion is sown in Fig. 7.Ebias,k is essentially a moving sum over te previous ours and is defined for k. Fig. 5. Example of te energy content per our Ebias,k for te first 3 days of 206, wit P cap = kw. Te black line is te zero reference. Fig. 7. Example of data divided into periods of 3 ours to calculate E 3 bias,k. Fig. 6. Histogram of Ebias,k tat is fit wit a Gaussian distribution wit E bias,k N ( kw, σ 2 = kw 2 ) Ebias,k is calculated for every our of 206 and te distribution is sown in Fig. 6. Ebias,k is symmetrical around zero and approximately Gaussian distributed. Te Gaussian distribution of Ebias,k makes it possible to calculate a confidence interval tat will contain te frequency bias wit a certain probability. Wen providing FNR wit a storage unit, te energy is limited so it is not preferable to reserve enoug energy for te worst case tat almost never occurs. Instead, te EV can reserve D. Caracterising te carger efficiency Since large amounts of energy are excanged wit te grid during FNR provision, te efficiency of te power converter as a ig impact on te energy consumption. ENEL as produced a bidirectional carger wit a capacity of 0 kw tat via te CHAdeMO DC connection can be used to perform FNR wit an EV [0]. Te carger is currently being used to perform FNR wit up to P cap = 0 kw per EV in te Danis researc project called te Parker Project. Te efficiency of tis bidirectional power converter as been found from measurements in a laboratory environment. Since an EV performing FNR usually reserves a fraction of its power capacity to use it as an energy offset,

5 te efficiency measurements ave been made for a loading between 0 and up to 9 kw in bot directions. Te measurements are performed wit a Nissan LEAF EV, wit a State of Carge (SOC) of 30 60%. Te measurements were made every second during 24 ours wile te EV was providing FNR wit P cap = 9.25 kw, tereby operating on most possible power levels. Te active power, measured on te AC side, is sown in Fig. 8 togeter wit te SOC, informed by te EV. previous power level. Te caracteristic of te curve is typical for power converters tat are less efficient wen operated at a low power levels. As long as te power is at least 50% of te capacity, te efficiency is between 80% and 90% and quickly drops wen te power decreases. State of te art power converters wit similar capacity can acieve an efficiency tat ranges between 88% and 96% for loading between 0% and 00% in bot directions []. Even toug laboratory equipment faces oter callenges tan mass-produced power converters, in terms of cost, tere is room for improvement in te future. As te commitment to te grid is independent of te carger equipment, all efficiency losses will be covered by te EV. Te battery will terefore receive less power tan wat is drawn from te grid and will ave to deliver more tan is extracted from te battery. Te power drawn or injected from/to te battery can be calculated as { P bat,t = P t η c, if P t 0 P t /η d, if P t < 0 (6) III. RESULTS Fig. 8. Measurements of a Nissan LEAF performing FNR wit a V2G carger from ENEL Te yellow curve is te teoretical accumulated energy and represents te ideal SOC if te efficiency was 00%. Te orange curve sows te actual SOC, wic decreases due to te losses associated wit FNR provision. Te difference between te real and te ideal SOC in te end of te reserve provision, is te total energy loss due to te carger efficiency. After 24 ours of FNR provision te SOC of te EV is 4. kw but it would ave been 33.7 kw if tere were no loss during tis period, resulting in a loss of 9.7 kw. Te voltage and current on te DC side is also measured and used to calculate te efficiency wen carging and discarging. Te efficiency is calculated as te ratio between te active power on te AC side and power on te DC side. We distinguis between te carging η c = P DC /P AC and discarging η d = P AC /P DC efficiencies. Fig. 9 sows a scatter plot of te efficiency at different power levels. Fig. 9. Power efficiency for different carging powers measured during 24 ours of FNR provision Because of te fast canging power reference and a small delay in te carger, te ratio between te power on te AC and DC side at a certain power level also depends on te A. Effect of Energy Content on Reserve Provision In tis section we sow te effect of te frequency energy content on FNR provision. We first divided te data into periods of ours and calculated Ebias,k for = We ten calculated te confidence intervals, were 99% of Ebias,k will lie witin for eac different ; te results are sown in Table II. Te imbalances are symmetrical so te positive and negative sides are equal. TABLE II 99 % CONFIDENCE INTERVALS FOR Ebias,k IN ± KW Period [] ± kw Period [] ± kw Te results from Table II validate our previously stated argument, tat te rate of cange of te confidence interval of Ebias,k is negative wit respect to. Te derived confidence intervals can be validated by calculating E t for an EV providing FNR for 5 consecutive ours. Te simulations are conducted for te wole frequency data set of 206, wic resulted in 8770 samples. For eac of te 5-our samples, E t is calculated during te wole FNR provision time and te results of 30 samples are sown in Fig. 0. Notice tat due to symmetry, te upper and lower limit of te confidence intervals are equal in absolute values. Due to te relatively small number of samples, E t violates te confidence intervals of Ebias,k wit iger probabilities. To evaluate te results statistically, we must use a large number of samples. In Fig. we sow te distribution of E t at te end of eac 5-our sample period for all te 8770 samples. Te accumulated energy follows a Gaussian

6 expresses te expected energy loss in one our. Statistically te EV battery is expected to continuously lose energy wen providing FNR, due to te carger efficiency, even if Ebat,k can take positive values in some ours. We repeated te simulations for 2-our and 3-our frequency samples and 2 3 calculated te expected values of Ebat,k and Ebat,k, te results are summarised in Table III. Fig % confidence intervals of Ebias,k and simulated Et for =... 5 for 30 different 5-our frequency samples. distribution wit a mean value very close to 0 and a variance of.5276 kw2. TABLE III NORMALISED HOURLY ENERGY LOSS Efficiency [-] our [kw/kw/] 2 ours ours As seen from te results, losses, denoted by l, can be well approximated as a linear function of and η. It is straigtforward to sow from (5) tat l is also proportional to Pcap. Consequently, we express l as l = ( η)pcap Fig.. Distribution of Et at te end of te eac sample period for 8770 samples. Et t=end N ( kw, σ 2 =.5276 kw2 ) By allocating ±4 kw in te battery, te EV could provide FNR for 5 consecutive ours wit Pcap = kw in 00% of te cases but te requirements could be met in 99% of te cases by allocating only ±2.7 kw. B. Effect of carging efficiency on reserve provision In tis section we examine te effect of bot carger efficiency and frequency biases in FNR provision. For simplicity, te carging and discarging efficiency are assumed equal and we denote bot by η. We introduce te term Ebat,k, wic is equivalent to Ebias,k, but includes carger losses; for efficiency equal to, te 2 terms give te same results. We calculated Ebat,k via simulations by considering ourly frequency samples and 4 different efficiency values. In all cases, Ebat,k is Gaussian distributed and te fitted probability density functions are sown in Fig. 2. (7) As seen in Fig. 8, efficiency is not constant but depends on te carger loading. Even toug efficiency is very small for low power values, its effect on te overall losses is not so significant, due to te relatively small amounts of excanged energy. By using (7) and assuming an average efficiency of 0.8, providing FNR for 24 ours wit a capacity of 9.25 kw, results in 20.3 kw of losses. Tis value is very close to te actual losses of 9.7 kw, calculated from te real data. Ebat,k is Gaussian distributed for different values of te efficiency and for different durations. Similarly to te confidence intervals calculated for Ebias,k, we can derive bounds for Ebat,k wit specified probabilities. Fig. 3 sows Et and te 99% confidence intervals for Ebat,k of an EV providing FNR wit Pcap = kw for 5 ours and a constant efficiency of 0.8, for 30 different frequency samples. Fig. 3. Evolution of Et during 5 ours of FNR provision wit a constant efficiency of 0.8 for 30 frequency samples Fig. 2. Probability density function of Ebat,k, for FNR provision wit Pcap = kw and different carger efficiency values Te expected energy excange decreases as efficiency takes smaller values. Tis value, wic is negative for η <, Similar to Section III-A, in Fig. 4 we present te distribution of Et at te end of eac 5-our frequency period for all 8770 samples. Te service provider would need to allocate [ 4.2, 2] kw in te battery to be able to deliver tis service wit 99% confidence. Te range is larger compared to te no-loss case, but is substantially moved to te negative side due to te losses. Te expected value of Et at te end of te service provision is.29 kw, wic is very close to te

7 approximation from (7), wic gives an expected loss of.36 kw. Fig. 4. Te cross section of te end of 8770 periods, wit distribution E t t=end N (.29 kw, σ 2 =.59 kw 2 ) In order to compensate for te losses incurred by FNR provision, te service provider can cange te set-point of an EV from zero to a positive (carging) value suc tat te EV on average carges enoug to cover tese losses. If te SOC at te end of te service provision is required to be iger tan in te beginning, tis set-point cange can be added on te scedule of te EV. FNR could be provided wit EVs of te type Nissan LEAF wen tey are connected to a V2G carger via a plug suc as CHAdeMO. If it is connected to te grid constantly in te period 7:00-08:00, it could provide FNR for 5 ours. From our analysis it was found tat E t t=end [ 4.2, 2] kw for kw of capacity, wic results in a necessary capacity range of 6.2 kw per kw of P cap. Te Nissan LEAF as a capacity of 30 kw but it is often recommended tat it is only operated between 20 90% of te SOC to minimise te wear of te battery, leaving 70% of te capacity for FNR. Te maximum capacity P cap tat can be sold in te 5 ours period can be calculated as P cap = 70% 30kW 6.2kW /kw = 3.4kW (8) 3.4 kw is te largest capacity of FNR te Nissan LEAF can be used to deliver in 5 ours, if no actions to reduce te variance of te energy bias are taken. Providing FNR wit 3.4 kw for tese 5 ours every day in a year would amount to a capacity payment of 42 EUR, based on an average of te market prices , as already sown in [5]. However, compared to te previous analysis were losses were not taken into account, te expected daily losses are found to be kW = 4.4 kw. Wit te average Danis retail electricity price of 0.3 EUR/kW, including taxes, tis would amount to a cost of 48 EUR per year, terefore nullifying te revenue. Additionally te EV would ave a SOC, normally distributed between % at te end of te period wic could be an inconvenience for te EV owner. IV. CONCLUSION AND FUTURE WORK Te business model can be improved by increasing te efficiency. If te efficiency of te converter is increased to 90 %, te losses would be reduced by 50 %. As te power system becomes more based on intermittent energy sources, te need and tereby probably te price for fast frequency control would increase. Since FNR is a service to te grid, it could also be argued tat it sould be exempted from taxes wic are about 64 % of te electricity cost. Te previous sections ave sown tat te efficiency losses can be predicted and tereby covered by adjusting te setpoint. It is terefore a cost problem rater tan an operational problem. Te variance of te energy content in eac our is owever very problematic as it reduces te power capacity used for FNR, wic is proportional to te payment. A way to reduce te energy imbalance ratio is to allow for canging te set-point during operation depending on te energy balance at te current time. If te V2G carger as a capacity of 0 kw, a sare of tis could be reserved for set-point canges, by buying or selling te additionally needed energy on te intra-day market. Future work would involve analysing ow systemic imbalances of te system frequency at specific times of te day can be used to make more accurate estimates te energy content and ow centralised control can be used to increase te efficiency by carging some EVs at full power and oters not at all, making a combined delivery. V. ACKNOWLEDGEMENTS Te work in tis paper as been supported by te researc projects ACES (EUDP grant nr: EUDP7-I-2499) and Parker (ForskEL grant nr ttp://parker-project.com/). REFERENCES [] Energinet.dk. Electricity security of supply report. ttp: //energinet.dk/sitecollectiondocuments/engelske%20dokumenter/ Om%20os/Elforsyningssikkered 206 UK.pdf, 206. [2] R. Halvgaard, N.K. Poulsen, H. Madsen, J. B. Jørgensen, F Marra, and D.E.M. Bondy. Electric veicle carge planning using economic model predictive control. Electric Veicle Conference (IEVC) IEEE, 202. [3] A. Sculler, C. M. Flat, and S. Gottwalt. Quantifying load flexibility of electric veicles for renewable energy integration. Applied Energy, 5: , 205. [4] M. Marinelli, S. Martinenas, K. Knezović, and P.B. Andersen. Validating a centralized approac to primary frequency control wit series-produced electric veicles. Journal of Energy Storage, 7:63 73, 206. [5] A. Tingvad, S. Martinenas, P.B Andersen, M. Marinelli, B.E. Cristensen, and O.J. Olesen. Economic comparison of electric veicles performing unidirectional and bidirectional frequency control in denmark wit practical validation. 206 Proceedings of te 5st International Universities Power Engineering Conference, 206. [6] J. Hu, H. Morais, T. Sousa, and M. Lind. Electric veicle fleet management in smart grids: A review of services, optimization and control aspects. Renewable & Sustainable Energy Reviews, 56: , 206. [7] A. Kieldsen, A. Tingvad, S. Martinenas, and T. Meier Sørensen. Efficiency test metod for electric veicle cargers. Proceedings of EVS29-International Battery, Hybrid and Fuel Cell Electric Veicle Symposium, 206. [8] W. Kempton and J Tomić. Veicle-to-grid power fundamentals: Calculating capacity and net revenue. Journal of Power Sources, 44(): , [9] Energinet.dk. Ancillary services to be delivered in denmark - tender conditions. ttp://energinet.dk/en/el/systemydelser-for-el/sider/ Systemydelserforel.aspx, 202. [0] ENEL. Nissan, enel and nuvve operate world s first fully commercial veicle-to-grid ub in denmark. ttps:// dam/enel-com/pressrelease/porting pressrelease/nuvve-nissan-enel% 20ENG.pdf, 206. [] H. S. Kim, M. H. Ryu, J. W. Baek, and J. H. Jung. Hig-efficiency isolated bidirectional ac;dc converter for a dc distribution system. IEEE Transactions on Power Electronics, 28(4): , 203.