EV stochastic modelling and its impacts on the Dutch distribution network

Transcription

1 EV stochastic modelling and its impacts on the Dutch distribution network Rick Scharrenberg Department of Electrical Engineering Eindhoven University of Technology 56MB Eindhoven, The Netherlands Bram Vonk, IEEE Student member Department of Electrical Engineering Eindhoven University of Technology 56MB Eindhoven, The Netherlands Phuong H. Nguyen, IEEE member Department of Electrical Engineering Eindhoven University of Technology 56MB Eindhoven, The Netherlands Abstract This paper presents the impact of increasing penetration of electric vehicle (EV) on distribution networks using a stochastic modelling approach based on Monte Carlo simulations. The proposed method aims to derive the stochastic characteristics of EVs with detailed transportation data together with geographic information, and vehicle properties, thus formulating suitable charging patterns. Output of the model is then coupled with different network scenarios of the existing planning tool to evaluate impacts of uncontrolled and controlled EV charging by running load flow analyses for hundreds of times. A case study for a typical Dutch distribution network is conducted for scenarios of the years 5,, 5 and. Simulation results show that the approach reflects accurately effects of controlled/smart charging to reduce the number of overloaded MV/LV transformers as well as the average power losses in developed scenarios. I. INTRODUCTION Electric vehicles (EVs) are expected to have a great impact on the electrical power grid operation in the future []. The Dutch government is planning to have,, EVs in the Netherlands in the year 5, which will increase the electric power demand []. This raises questions about the effects of EV charging on the distribution networks, both at lower voltage (LV) and medium voltage (MV) level []. To prevent problems with the electrical distribution network in the future, the influence of EVs needs to be investigated. It is an important research topic within the emerging smart grid context [4]. Modelling EVs is crucial to analyze its impact to the network infrastructure. Several research works have focused on building up a suitable model of EV charging behaviour [5] [6] [7]. In [8] [9], impacts of EV model has been studied and coupled with the planning process of a network operator. As a trend, application of distributed intelligence was introduced for modelling and controlling EV behaviour in []. In this paper the effect of different stochastic charging patterns of Plug-in Electric Vehicles (EVs) on the Dutch MV electrical power grid is researched for the years 5,, 5 and. Furthermore, the difference between uncontrolled and controlled charging is researched. The paper mainly focuses on results that are relevant to distribution network operators, such as maxium power usage, overloads and power losses. In order to simulate a realistic future charging pattern, the EVs are modelled as stochastic loads which are dependent on different stochastic parameters. Data from a mobility research of the Dutch government, geographic data and vehicle properties are used for these parameters and are combined into a data file []. After this, the actual charging pattern is calculated and added to the base load profile by using. The modelled load patterns can then be analyzed by running load flow calculations with Vision to obtain results about the impact on the grid, such as power flows, power losses and overloads. This process, based on Monte Carlo simulation, will be performed iteratively for hundreds of times to get a result which shows the stochastic variation effects of EV charging []. Furthermore, the results can be used to develop an optimal dispatching strategy to charge the EVs. This optimal dispatching strategy can be added to the initial script and simulations can be run again to obtain new results. Finally, a comparison between the two charging methods is made. An overview of the required steps for this approach is shown in Fig.. Section II of this paper explains how the EVs are modelled based on the different kinds of information, datasets and assumptions. Next, section III describes the two charging methods. Section IV explains the generation of the charging pattern and section V shows the results of load flow analyses based on the EV models. Finally, a conclusion is given in section VI. A. Geographic information II. PROCESSING INPUT DATA For this research the MV distribution network of Wierden city is used. Wierden city has been chosen because its distribution network covers urban, rural and industrial areas and also has a medium-size population. Many small cities in the Netherlands have similar properties and therefore similar results can be expected for these cities. The distribution network also contains data from nearby villages and industrial areas. An overlay of the network on a map is shown in Fig.. This area covers a population of,87 citizens []. To determine the charging locations of EVs, a geographic dataset from the CBS is used [4]. In this dataset, Wierden city is divided in several areas. For each area the number of citizens and the number of companies are available. These numbers are used to determine if an area is an industrial area /4/$. c 4 IEEE PMAPS 4

2 6 OViN Mobility research data Internet PDF Wierden city data EV Parameters x Deviations of the number of EVs Expected number of EVs.5 Number of EVs [ ] SPSS Data file.5.5 Generate charging pattern 5 Year [ ] 5 Fig. : Predicted number of EVs Vision Load flow analysis 5 Determine number of overloads Process data Generate plots Number of people [ ] Fig. : Overview load flow calculations Departure time [hour] Fig. 4: Departure times of vehicles For this normal distribution, the expected value µ is chosen to be equal to the average predicted number of EVs. The value for the standard deviation has been chosen to match the predictions of both Netbeheer Nederland and the Dutch Government as good as possible [5][6]. In this case a of.66µ is used for all the simulated years. The predicted number of EVs and its deviations are shown in Fig.. In this figure the dark green line and bars indicates the expected values µ together with the standard deviations. The light green shaded area indicates the predicted amount of EVs from the Netbeheer Nederland research [6]. Fig. : Distribution network of Wierden city C. Mobility research data or a residential area. Because industrial areas will have no EVs charging at houses and almost no public charging spots, EV charging is neglected in these areas. B. Penetration level of EVs The current number of EVs in the Netherlands is about 7,97 [5]. The predicted number of EVs in The Netherlands for the years 5,, 5 and is on average 7,5,,, 8, and,95, respectively with deviations of 5% [5][6]. It is assumed that these predicted number of EVs can be modelled as a normal distribution, which is shown in Eq. [7]. f (x) = p e (x µ) / () Information about the mobility behavior of car drivers in the Netherlands is used to determine the departure times, arrival times and travelled distances of vehicles []. This data is shown in Fig. 4, Fig. 5 and Fig. 6. In this data, commuting times are clearly visible. Most cars will departure between 7: and 9: am, and between 6: and 8: pm. Furthermore, most of the cars will arrive at their destination between 8: and 9: am, and between 7: and 8: pm. Finally, it is shown that cars will probably travel short distances, which will result in short charging durations. D. Load profile The base load profile of the grid is based upon data from the Dutch network operator Enexis. This load profile has

3 Number of people [ ] Number of people [ ] Arrival time [hour].5 x Fig. 5: Arrival times of vehicles Distance [km] Fig. 6: Travelled distances of vehicles been calculated by averaging load profiles of different Enexis transformers that are used in areas comparable to those of Wierden city. The resulting load profile is shown in Fig. 7. A. Assumptions III. CHARGING STRATEGIES In this paper the focus will be on the impact of EVs on the grid and therefore some assumptions are made. Firstly, the load profile shown in II-D will be used as a base level and will not be changed. Increase of power usage throughout the simulated years caused by effects other than EVs will therefore be neglected. Secondly, the amount of EVs in Wierden will Load [%] Load profile 5 5 Fig. 7: Load profile be proportional to its population. Finally, some assumptions about the charging process are made: EVs will charge with a power of.7 kw (single phase charging) [8], an EV can drive 5 kms with kwh of charge [6], and an EV will start charging when it arrives at a destination in the uncontrolled case [9]. B. Smart charging In section III-A it is stated that EVs will start charging when they arrive at a destination if uncontrolled charging is used. For controlled or smart charging a different charging strategy is used. With smart charging, the arrival time of the last trip on a day is used as a first boundary. The departure time of the first trip on the following day is used as another boundary. Furthermore, the charging duration is calculated and an optimal charging point is chosen where the power consumption in the distribution network is low. The charging load will then be placed around this charging hour until one of the boundaries is reached. When a boundary is reached, placing of the loads in that direction will be stopped and will continue in the other direction. By using this valley filling method the EV load is spread as much as possible and overloads are less likely to occur []. A mathematical formulation of this method is shown in Eq.. minimize x X4 t= (P L (t) + NX x(i, t) P avg ) i= subject to t arrival (i) apple t start (i) apple t departure (i) T charge, t start (i)+t charge (i) X t start (i) x(i, t) = E charge (i), P charge if charging x(i, t) = otherwise 8i [, N], 8t [, 4] () In this equation, t is the time of the day, i the number of the EV movement and P L the base load of the grid. N is the amount of EV movements on a given day, x(i,t) the charging load of the EVs and P avg the average base load of the grid. t arrival indicates the arrival time of a given EV, t start the started charging time of a certain EV, t departure the departure time of a given EV and T charge the neccesary charging time for an EV. E charge indicates the energy that is required to fully charge an EV, while P charge indicates the EV charging power. IV. GENERATION OF CHARGING PATTERN By using the expected numbers of EVs, mobility research data and information about Wierden city, a charging pattern is generated using a script based on the schematic shown in Fig. 8. In the first step, user s input about the number of iterations, expected amount of vehicles, standard deviation of vehicles and output file name are obtained. Also a choice for smart or uncontrolled charging has to be made. After this step a,

4 Process EV and grid input data i= Generate base load profile Determine number of trips i++ j= Add EV load to load profile j++ Power usage [MW] If japplenumber of trips no If iapplemaximum iteration no Write output yes yes Fig. 8: Stochastic charging pattern generation data file is read which contains samples of EV arrival times, travelled distances and the average number of trips per day for a single EV. This data file also contains information about the population of the Netherlands, the population of Wierden, EV efficiency (kms per kwh) and the charging power. Furthermore this data file contains the base load profile parameters and all the loads of the distribution network. After the input processing has been done, an iteration counter is started. In the second step, an empty load matrix is created with dimensions [4 x n loads ], where the number 4 indicates the amount of hours per day and n loads the number of connected loads in the grid. This load matrix is filled with the base load profile for each individual load. After this, the number of trips is determined by picking a random value from the normal distribution of EVs and multiplying this by the average number of trips per EV. Next, the iteration counter is raised and a trip counter is started. In the third step, a random entry from the mobility research data is picked. This entry contains an arrival time, departure time and travelled distance. Based upon this information the EV load is calculated and added to a random load in the load matrix. This EV load can be added on a different timespan in the load matrix, depending on the used charging method (smart charging or uncontrolled charging). After this the trip counter is raised. In the fourth step, the trip counter is compared with the determined number of trips. If the trip counter exceeds the number of trips, then all the EV loads for that iteration has been added to the load matrix and the script will proceed with the next iteration. If not, the script will proceed with adding the next EV load. In the fifth step, the iteration counter is compared with Fig. 9: Power usage EVs with uncontrolled charging the maximum iteration value. If the iteration counter exceeds this value, then the script is finished with the generation of a charging pattern and the script will proceed with writing the output. If not, the script will proceed with the next iteration. The last step consists of writing an output file which contains the load matrices from every iteration. This output file is then imported into Vision for the load flow analyses. V. SIMULATION RESULTS In this section, impacts of uncontrolled and controlled EV charging on a typical Dutch distribution network will be modeled and simulated. Modelling of EV charging patterns is developed in and used for analysing load flows recursively in the Vision program of Phase-to-Phase for the years 5,, 5, and. A. Peak demand impacts ) Uncontrolled EV charging: Fig. 9 shows the average EV power usage for uncontrolled charging together with the standard deviations. In this plot a maximum EV power usage peak can be seen at 6: pm. The load flow analyses show, that when no EVs are charging, the peak load in the grid is 4. MW. For the years 5,, 5 and the increase in this peak load due to EV charging is.%,.46%, 6.9% and 4.7% respectively. The increase in peak power usage is resulting into overloads in the distribution network. The number of overloads for each year can be seen in Table I. These overloads are mainly caused by EVs that are trying to charge in the evening or large groups of EVs which are charging at the same transformer. In this table is shown that one MV/LV transformer is overloaded even when no EVs are charging (year ). This transformer is only overloaded for a small period of time which does not cause it to fail. For the years 5- it is visible that each year the number of overloads increases up to a maximum of 48 MV/LV transformers and HV/MV transformers. To decrease the amount of overloads, smart charging is used. ) Controlled EV charging: Fig. shows the average EV power usage for smart charging together with the standards deviations. In this plot a maximum EV power usage peak can be seen at 7: am. The loadflow analyses show, that when

5 TABLE I: Overloads with uncontrolled charging Year MV/LV transformers HV/MV transformers MV cables (.6%) (%) (%) 5 (.6%) (%) (%) (.9%) (%) (%) 5 6 (.%) (6.7%) (%) 48 (.%) (6.7%) (%) TABLE II: Overloads with smart charging Year MV/LV transformers HV/MV transformers MV cables (.6%) (%) (%) 5 (.6%) (%) (%) (.9%) (%) (%) 5 9 (5.7%) (8.%) (%) (.6%) (6.7%) (%) smart charging is used, the increase in peak load due to EV charging is.%,.77%,.% and 7.% for the years 5,, 5 and respectively. For the latter years, this is significantly lower than the case with uncontrolled charging. The number of overloads for this smart charging strategy are shown in Table II. In this table it is shown that the number of overloads for the year 5 and with smart charging are equal to the number of overloads with uncontrolled charging. For the year 5 and the number of overloads is significantly reduced when smart charging is used. With smart charging the maximum amount of overloaded MV/LV transformers in the year is reduced with 58.%. This reduction of overloads is mainly the result of shifting the charging load from the evening towards the night, as can be seen in Fig.. The remaining.6% of overloads cannot be decreased by load shifting because it is the result of large groups of EVs that are charging at the same transformer instead of spreading throughout the city. B. Power losses impacts Besides the peak power usage and the number of overloads, also the power losses are important. In Fig. and Fig. the power losses due to EV charging are shown for both smart charging and uncontrolled charging. From the loadflow analyses follows, that in the peak power losses due to EV Power usage [MW] Smart Uncontrolled 5 5 Time [hour] Fig. : Power usage EVs with and without smart charging Power losses [MW] Fig. : Power losses EVs with uncontrolled charging charging will be increased with 79.8% for the uncontrolled case. With smart charging these losses will only increase with 4.66%. The average power losses in the distribution network without EVs charging are.4 MW. In Fig. 4 the increase in power losses related to charging EVs are shown for both smart charging and uncontrolled charging. In this figure is shown that the power losses with smart charging are much more constant compared to the power losses of uncontrolled charging. Furthermore, the average EV power losses are shown Power usage [MW] Power losses [MW] Fig. : Power usage EVs with smart charging. 5 5 Fig. : Power losses EVs with smart charging

6 Power losses [MW] Smart Uncontrolled 5 5 Time [hour] Fig. 4: Power losses EVs with and without smart charging TABLE III: Average power losses due to EV charging Year Smart charging Uncontrolled charging 5.7 MW.7 MW.6 MW.44 MW MW.578 MW.8 MW.654 MW in Table III. In this table is shown that the average power losses related to EV charging are in general much lower if smart smart charging is used instead of uncontrolled charging. VI. CONCLUSION The stochastic charging behavior of EVs has a big impact on the distribution network of Wierden. The total power consumption of EVs is not causing much problems, because it is only an increase of 4.7% of the peak power usage in the network. Problems arise when large groups of EVs are not spread evenly throughout the city and want to charge at the same time at a specific transformer. This is locally causing a big increase of power usage which causes the MV/LV transformers to be overloaded. With uncontrolled charging.% of the MV/LV transformers, 6.7% of the HV/MV transformers and none of the cables will be overloaded in. Also the average power losses will increase with 8.8%. These number can be reduced significantly by using smart charging. With smart charging.6% of the MV/LV transformers, 6.7% of the HV/MV transformers and none of the cables will be overloaded in. Furthermore, the average power losses will increase with.6% instead of the 8.8% of the uncontrolled case. From these results follow that problems in the MV distribution network due to EV charging can be reduced, but cannot be prevented by only using smart charging. Furthermore, it shows that components, mainly MV/LV transformers, in the distribution network have to be upgraded to support large scale EV charging. ACKNOWLEDGMENT This work is part of the DAME project sponsored by the Electromobility+ program []. We would like to thank Gertjan Mulder and Jos Klein Bleumink from Stichting e-laad for providing public charging data. Finally, the authors would like to thank Berto Jansen from Phase to Phase for providing a license for Vision. REFERENCES [] J. Pillai and B. Bak-Jensen, Impacts of electric vehicle loads on power distribution systems, in Vehicle Power and Propulsion Conference (VPPC), IEEE, Lille, France, Sep., pp. 6. [] (, Oct.) Elektrisch rijden in de versnelling. Agentschap NL. [Online]. Available: van%aanpak%-elektrisch%rijden%in%de%versnelling-.pdf/ [] (9) Nederland gaat elektrisch rijden. Enexis. [Online]. Available: [4] M. Simoes, R. Roche, E. Kyriakides, S. Suryanarayanan, B. Blunier, K. McBee, P. Nguyen, P. Ribeiro, and A. Miraoui, A comparison of smart grid technologies and progresses in europe and the u.s. Industry Applications, IEEE Transactions on, vol. 48, no. 4, pp. 54 6,. [5] A. Lojowska, D. Kurowicka, G. Papaefthymiou, and L. van der Sluis, From transportation patterns to power demand: Stochastic modeling of uncontrolled domestic charging of electric vehicles, in Power and Energy Society General Meeting, IEEE, Detroit, USA, Jul., pp. 7. [6] R. Verzijlbergh, Z. Lukszo, J. Slootweg, and M. Ilic, The impact of controlled electric vehicle charging on residential low voltage networks, in Networking, Sensing and Control (ICNSC), IEEE International Conference, Delft, The Netherlands, Apr., pp [7] J. Lopes, F. Soares, and P. Almeida, Integration of electric vehicles in the electric power system, Proceedings of the IEEE, vol. 99, no., pp. 68 8,. [8] E. Veldman, A. Postma, H. Slootweg, and M. Gibescu, Electric vehicles charge ahead, Transmission & Distribution World, vol. 6, no., pp ,48,5,. [9] M. Grond, Impact of future residential loads on medium voltage networks, Master s thesis, Delft University of Technology, Delft, The Netherlands, May. [] E. Karfopoulos and N. Hatziargyriou, A multi-agent system for controlled charging of a large population of electric vehicles, Power Systems, IEEE Transactions, vol. 8, no., pp. 96 4,. [] () Ovin research. CBS. [Online]. Available: [] () Monte carlo methods. University of Bristol. [Online]. Available: manpw/teaching/notes.pdf [] () Wierden in cijfers. Gemeente Wierden. [Online]. Available: 44 [4] Toelichting Wijk- en Buurtkaart, CBS,. [5] () Cijfers elektrisch vervoer. Agentschap NL. [Online]. Available: [6] (, Jan.) Laadstrategie elektrisch wegvervoer. Netbeheer Nederland. [Online]. Available: [7] F. Koyanagi, T. Inuzuka, Y. Uriu, and R. Yokoyama, Monte carlo simulation on the demand impact by quick chargers for electric vehicles, Power Engineering Society Summer Meeting, vol., pp. 6, 999. [8] R. Prateek, Electric vehicles, charging infrastructures and impacts, Master s thesis, Eindhoven University of Technology, Eindhoven, The Netherlands, Dec.. [9] A. Lojowska, D. Kurowicka, G. Papaefthymiou, and L. van der Sluis, Stochastic modeling of power demand due to evs using copula, Power Systems, IEEE Transactions on, vol. 7, no. 4, pp ,. [] M. Galus, F. Wietor, and G. Andersson, Incorporating valley filling and peak shaving in a utility function based management of an electric vehicle aggregator, in Innovative Smart Grid Technologies (ISGT Europe), rd IEEE PES International Conference and Exhibition, Berlin, Germany, Oct., pp. 8. [] (, Sep.) Dame project. Electromobility+. [Online]. Available: seminar/4 DAME Presentation DAME.pdf