IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 30, NO. 4 JULY 2015, Load Modeling and Identification Based on Ant Colony Algorithms for EV Charging Stations Shaobing Yang, Mingli Wu, Xiu Yao, Student Member, IEEE, and Jiuchun Jiang. Senior Member, IEEE Abstract—Charging load modeling for electric vehicles (EVs) is. challenge due tots complexity. However, Iserves as a foundation {or related studies such as the Impact assessment of EV charging behaviors on power system and power demand side management for EVs, The decisive factors affecting charging load profile include ‘the power eurve, the duration, and the start time of each charging process. This paper introduces the charging traffic low (CTF) as a aiserete sequence to describe charging start events, where CTF contains both spatial and temporal properties of a charging load. A set of equations are proposed to build a probabilistic load model, followed by simulation iteration steps using a flow chart. The pa- rameter identification method based on ant colony (AC) algorithms {s then studied in depth, and the pheromone update and the state ‘transition probability are used to implement route finding and eity selection, respectively. Finally, an actual ease of battery swapplng station is applied fo Verify the proposed model in both identiiea- ‘Hon and simulation, The results sHow that the modet has satistac- tory accuracy and applicability. Index Terms—Ant e010 vehicles, load model. 1 algorithms, charging station, electric 1. Betropuction OWADAYS, electric vehicle (EV) industries are rapidly N developing. The large-scale application of EVs will come soon with improvements of power battery technologies. Thus, related research topics have been receiving great attention, whose spectrum includes impact analysis of EVs on a power system, power demand side management (PDSM) for EV charging load, and economic operation optimization of a micro arid involving EVs. In particular, modeling of EV charging loads becomes very important since itis a fundamental work to support the above stuclies. There are three charging modes for EVs: nonmal/slow. quick, and swap. The charging load in the last two modes has the fol- lowing characteristics: 1) higher superposition power values; 2) larger peak-valley differentials; 3) much more difficult to shift peak loads becanse the waiting time greatly affects the quality of charging service. Therefore, the load of a charging station ‘Manuscript received March 1, 2014; revised uly 03,2014; scooped August 10,2014 Date of publication September C1, 2014; date of cure version June 46,2015. This work was supported bythe Fundamental Research Funds forthe Cail Universities @2013/BMOT?), Paper no, TPWRS.00353-2014 ‘Ying M. Wu and J. fang ae wath Bovine Saotone Unversity Bejing 100084, China (e-mail shbyangglyta aun: mliruabjtu eduen: Jana Uhm ada cn) 'X. Yaoie wath The Oi State Univer, Columbus, O84 “iodos ad). (Color versions of one or more ofthe figures i ths paper ae svalable online a np ieeenplor eee or Digital Object Identifier 10 1109/TPWRS 2014.35: OUSA (¢-ma: 63 will cause significant impacts on a power distribution network (1). [2]. Iris necessary to assess these impacts. For this purpose, a highly resolved load model will be very helpful, as it can di- rectly support power flow calculation involving EVs [3]. [4]. Generally. there are two traditional approaches in power load modeling: the component-based statistical synthesis and the measurement-based identification. For EV charging load, the former could have three steps: First, classify EVs as different groups according to the uses. Second, analyze the operational pattern and the proportion for each group. Last forecast the total charging load. This approach was widely used for charging load forecasting [5]-[16]. In most studies. the distribution of charging start time was considered a crucial factor for EV charging load profile. In [5], it was analyzed based on the four charging scenarios. In [6]-{8], the EV plug-in time was represented by the normal distribution. In [9], it was obtained from the actual traffic flow measurement. In [10], it ‘was controlled by the charging strategies. In [11] and [12], it was simulated based on the vehicle mobility pattems. In [13]-[15]. it was assumed with the actual conditions. In [16], it ‘was estimated withthe arrival time distribution of an EV. Thus, many different ways are presented to describe the EV plug-in time. Although charging load profiles can be simulated with these ways. its accuracy still depends on certain assumptions or the given scenarios. For the latter approach, the charging load model is built with unknown parameters. These parameters need to be identified based on actual measurement. In [17], an aggregation model ‘was proposed, and the curve-fitting method was directly applied for parameter identification, However. its applicability is seri- ously limited because EV load profiles are significantly different fiom each other. In [1S], a novel way was proposed to identify the number of charging EVs for the nationwide power system. However, it did not consider the charging progress dependency and the spatial feature of traffic flow. ‘Therefore, it is necessary to establish a generic EV charging oad model with both spatial and time properties. This task has two key issues: how to describe charging start time with a uni- ‘versal parameter, and how to identify the parameter with an ef ficient algorithm based on measured data. ‘This paper aims to provide a more general and applicable way to describe the charging load profile and build a highly resolved probabilistic model. Fist, three decisive factors are analyzed individually. These are the start time, the charging power pro- file, and the duration of a charging process, Then, a simulation approach and its steps are presented based on the EV charging ttalfic flow (CTF). For CTF identification, suitable ant colony (AC) algorithms are selected and designed. An actual case is then studied for the validation. 0885-8050 © 2014 IEEE. Parsonal us is permite, but sapicston redatrbution equies IEEE permission ‘See hip fee ore publications standard pubiatons rights ndex hia fr more infortaton ‘Aumnonzed licensed use imted to: Lulea Lnversty of Tecnology. Downloaded on June 08,2020 at 08:31:29 UTG trom IEEE Xplore. Restrictions apply1998 IL. DECISIVE FACTORS OF CHARGING LOAD A. Charging Start Time The distribution of the charging start time greatly affects the load profile ofa charging station, It is decided by the CTF which indicates the number of EVs coming for charging services in a fixed interval. Generally, CTF has three major factors: location, charging price, and day type. The day type, including weekday, ‘weekend, and holiday is the erucial factor since the location is fixed and the price is rarely changing. ‘In transportation field, the prediction of urban traffic flow is 1 very important research subject, and the day type isa decisive condition for classification. Thus, CTF can be obtained from the results of traffic activity studies. Also, it can be classified with day types. Because CTF directly contains both spatial and ‘temporal dimensions, it brings many benefits for the charging load forecasting, especially in the case of large-scale application of EVs. More importantly, many advanced research methods in the transportation field can be used to study EV charging issues, Define At as the interval in minutes, then the CTF of a day can be divided into an integer array. It can be written in the following form: {am}s where m is the index of the array dy is the EV number entering ‘the charging station during the time segment [m — 1,1], and L is the length of the array. mele) a B. Charging Power Profle A typical charging process of power batteries is shown in Fig. 1, and it has two charging stages: constant-current and con- stant-voltage. The power curve can be divided into multiple seg- ments with the interval A¢, The average power values of all sea- ments are saved into areal array. In addition, during the stage of constant-current, the charging power maintains a relative con- stant value. Thus, the charging power can be described by the following equation: pe(0), Solt) {pie 4, where pe is the real array which stores the discrete average power values, tis an integer to indicate the remaining charging time as ¢At, and Tis an integer to indicate the length of mea sed power curve as 7, (t> Te) (ct) ° C. Charging Durations In most cases, charging durations are different from each other in length. The charging duration of a battery is determined by the initial state of charge (SOC). For most EV drivers, the de- cision for charging is mainly up to the instruction from the SOC indicator. According to the law of large numbers and the central limit theorem, charging durations approximately obey a normal distribution in a large sample space. Therefore, a normally dis- tributed random variable can be used to describe charging du- rations. It is written in the following form: Ra ~ Neiwuss(t,97) a) [EEE TRANSACTIONS ON POWER SYSTEMS, VOL. 30,NO.4, JULY 2015 Active power / WW Time min Fig 1. Cus of active power ofa charger . Fring density 3 & Time min Fig. 2. Scales of probability density of carping duration of batteries where Noa i a normal distribution function, and its mean value and standard deviation are j and oy, respectively. The probability density shown in Fig, 2 with scatters is based con 4000 changing processes of electric buses recorded by the charging management system of the battery swapping station for the Beijing Olympic Games. The nonlinear curve-fitting is applied with (3), and the fitting result is shown by a solid line The determination coefficient, also called R-Square, indicates the fiting degree, which is 0.985. The mean value and the stan- dard deviation equal 123 and 17 min, respectively. Thus, (3) is considered a satisfactory way to describe the statistical features of charging durations. IIL, SIMULATION APPROACH OF CHARGING LoaD According to (1)-(3), a probabilistic model can be built for the EV charging load sinnulation. Its simulation steps are shown in Fig. 3. In most cases, the charging power profile of the chargers is already a known condition, and the distribution of charging durations can be obtained based on historical data. Therefore, this model can be used in two ways. First, forecast or estimate the CTF of an EV charging station, then simulate ‘Aumnonzed licensed use imted to: Lulea Lnversty of Tecnology. Downloaded on June 08,2020 at 08:31:29 UTG trom IEEE Xplore. Restrictions apply‘YANG oral: LOAD MODELING AND IDENTIFICATION BASED ON ANT COLONY ALGORITEMS FOR EV CHARGING STATIONS Init parameters Simulation interval and length The ray of CT: en he ary of eharging power! pe “The random duration: Ae “Gbtan the umber of Evsentenngin this nterval tale SOC foreach EV with the random durations Update the numberof eve waltng forenarging ‘there any Ev wating fr charging? ves Tart charang process ofa Gare for aE Deersate the number of Es waiting for charging Decrease the numberof avaiable chargers No x Proce a charging progesies sop or neeareSOC| Caleulate the total charangoad Increase snation time withthe tava Tse simulation length reached? Simulation feshes Fig. 3, Flow chat of charging load simulation its charging load. Second, based on testing data, identify the CTF and then build a load model. The former can be used for charging station planning, and the latter can be used for load modeling of a built station The proposed probabilistic model has the following features: 1) The model has a good applicability since it combines the injor factors affecting charging load profiles. 2) The CTF is related to the location of a charging station and the type of day. Thus, the model can represent the influences of both space and time. 3) The model can be implemented based on the previous, studies in transportation field, Thus, a universal anxl con- vincing interface is expected! which will be very helpful for charging load simulation of large-scale EV applications, Fora probabilistic model, the random variable such as (3) will make the parameter identification extremely difficult, Thus, the ‘mean value of charging durations is used forthe identification instead of the random variable, and the CTF result will be con- sidered accurate enough since a normally distributed random 1999 TABLET ARRAY FoR THE CHARGNG POWER CURVE We = 125 Ms, At = 10M) Timerange Aveagepower —Timerange Average power ri Ww ‘ain kW 110) 39297 771,80], Tea7 (11,20) 39382 (81,90) Los7s. (21,30) 39.500 (91,100) 9.885 (1.40) 39641 6067 a1, $0) 39.808 S612 {s1,60) 40038 2208 11,70) 32.463 - state kt ealaton ne: ut stop time (at ing atin: 3 Time min Fig. 4. Definition diagram of 4) variable has a stable expectation. The mean length is also die vided into a certain number of segments with interval At. cording to (2), a real array is obtained to describe the charging power profile. A sample power array is shown in Table L ‘The charging power array can be expressed as. {um} where m is the index which indicates the charging progress, and Nis the array length for the mean value of charging durations. ‘The charging load at a given time is the sum of all charging processes that ase started within the previous charging duration. According to (1) and (4), a rapid computation formula is pro- posed in the following form: = YF las ml o me [1 Ny ® where # represents the given calculation time, ois the charging progress, and their definitions are also shown in Fig. 4 With (5), both the simulation and the identification ean be greatly simplified and improved. IV. IDENTIFICATION DEston WITH AC ALGORITHMS A. Resolution Space Analysis ‘The size of resolution space is determined by the number and the range of CTF elements. According to (1), Z will reach 144 if At is set as 10 min, Ifthe value of the CTF elements isin the range of [0, 6] the size of the resolution space will reach 7!4 ‘Aumnonzed licensed use imted to: Lulea Lnversty of Tecnology. Downloaded on June 08,2020 at 08:31:29 UTG trom IEEE Xplore. Restrictions applyVie aK vik Miva MN a) wo Fig. 5. Grid eingram of rsotion space thus an optimization algorithm shall be selected in searching for ‘global optimal solution. More importantly, CTF elements must be simultaneously resolved de to their correlation. ‘Most optimization algorithms such as genetic algorithms (GAs) and simulated annealing algorithms (SAAS), rely on the lobal error to establish the objective function, However, the lobal error is not suitable forthe parameter identification case due to two major reasons: First, ata given time, the simulation error is only related to the charging start events during the previous duration instead of all events. Second, the variation of each CTF element only affects the simulated load in the next charging duration. As such, both global and local optimizations shall be achieved at the same time, The latter is especially important. Therefore, AC is the best choice for the parameter identification because its objective function can be designed with local errors. This leads to better performance in terms of convergence speed and precision, Here, if the range of CTF elements is set as (0, K], the reso- lution space can be defined to be a nodal grid of Lx (K+ 1) dimensions, as shown in Fig. 5. Each node represents a city in a route for AC algorithms and its location isthe intersection of the column and the row. Thus, each column corresponds to a time segment, and each row is considered as a CTE element value. For example, ty, indicates that the element value equals y in the segment +. It means that y EVs enter the charging station in the segment [7 — 1,2] According to AC algorithus, in one iteration, each ant must accomplish a search route consisting of L nodes (cities) from the left-most column to the right-most one, and the row selection at each column for ants depends on the state transition probabil The measured data for a changing load can be saved as areal array according to the interval At, The array is written in the following form: {wide where w, is the total load at the time ‘At. Once a charging process starts, the charger will reach its rating power in a Very short time, such as a few minutes. Thus, the CTF element value of the segment i can be restricted by the following equation: ie (Le) © Ky = min(int(wi/p,), K) o where é is the column index, min is the funetion to return the smallest one from the given values, and p, is the rated power of the changer According to (7), the tabu table of AC algorithm can be es- tablished, which means the size of solution space can be signif- icantly reduced. [EEE TRANSACTIONS ON POWER SYSTEMS, VOL. 30,NO.4, JULY 2015 B. AC Operator Designs AC algorithms have two Key operators: pheromone update and city selection, The former is designed using feedback fiom the simulation error, and the latter is achieved with the state tran- sition probability. Each ant needs to complete two steps when it arrives at a column: First, update its related pheromone ele- ments according tothe error caused by the row selections within the previous charging duration. Second, select a node (row) at this column according to the state transition probability In the procedure of parameter identification, (5) is used to simulate the charging power values for all columms. Define as the index of a column. The sinmulation error is written asthe following equation: ee = pu(t) — ws © where 2 is the column index, According to (5), pu(z) is determined by multiple charging events started during the range of [x N,,, 2 1). Thuse, can be used to update the pheromone elements within this range. The updating equation is waitten as a oa, @ = (eit Be (6=2 Mayet Le p=a) © where + is the pheromone matrix of size L x (K +1), é is the cofumn index of each element within the range of [1 jv — 1), j is the selected row index of the column #,@ is the pheromone intensity, pis the evaporation coefficient, and e isa very small positive real constant to prevent zero division. Meanwhile, for the column z, the row selection depends on the state transition probability. It can be calculated by the fol- owing equation: G=0,....Ks) 10) ‘where j isthe selected row index of the column ¢, # isan integer, Kis an integer obtained with (7), a is the heuristic factor of information, is the heuristic factor of expiration, and 1 is the heuristic function, With (10). the state transition probabilitcs ofall the rows of cofuinn «can be calculated. It follows that a row can be sé- lected according to the uniform distribution. In AC algorithms, the heuristic function plays a very important role since it can increase the local searching capability. It needs to be designed ‘with the full effects caused by the row selection, First, for column 2, if row y is selected, the simulated load results during the index range of [x +1, r+, are all affected. Based on (5) the simulated load at this column can be obtained by the following equation: Ns pels) = So (te mt) bv te as (ielet+1e2+N,). aD ‘Aumnonzed licensed use imted to: Lulea Lnversty of Tecnology. Downloaded on June 08,2020 at 08:31:29 UTG trom IEEE Xplore. Restrictions apply‘YANG ral: LOAD MODELING AND IDENTIFICATION BASED ON ANT COLONY Thats man parameters The number of erations Usd measur daw) The number of ante ‘Te number of columns) The numberof rows 1} ‘The pheromone mati (2 ‘he pheromone ners (0) The evaporation coefcent (0) ‘Selec the nent cola forthe ant. Updos the invohing phoromane leone wth (8) ‘elect rw according 000) ee alterations competed? Second, the total error can be written in the following form: a4 Ny DY beliiesy) — w Sa 1 a2) Lastly, the heuristic function can be given as the following equation: 1 =e as) Tey ‘The detailed steps for the parameter identification with AC algorithms are presented in Fig. 6 V. Case VALIDATION ‘The charging station to serve electric buses for the Beijing (Olympic Games is taken as an example. This station has been serving electric buses since 2008. The measured data were ac- (quired in 2010, At that time, there were already two electric bus lines and over 30 buses supported by this charging station. These ‘buses min 16 b per day like the regular buses. The I-h average power values of the five Thursdays of September in 2010 are selected for CTF identification. They are shown in Fig. 7 by a histogram. The conditions are set as follows: ALGORITHMS FOREV CHARGING STATIONS 200 Measurement “+ Simulation Tineth Fig. 7. Comparison with simulation to measurement of chasing lod & 2 8 a 3 e a Time sh Fig. 8. CTF identification resut 1) At = 10min, N,, = 13, and Table Tis used to describe the charging power curve, 2) The array w is loaded with the 24 1-h power values. 3) Set IC = 6, then row indexes are in the range of [0.6] 4) Q = 5000, p = 0.5,a= 1,8 = 1,e= Le=10, and set all values of the + matrix to be 20. 5) Set the ant number to be 600, and set the iteration times to be 100. The computation of charging load has a satisfying speed since (5) is based on the fundamental operations of arithmetic. The identified CTF values are shown in Fig. 8, and each one is pre- sented as a pulse, The pulse height indicates the number of EVs entering for charging in the previous 10 min. According to the CTF, simulation can be implemented with (5), and the results are shown with the solid dots and lines in Fig. 7. In this case, the R-Square value reaches 1.000, In ad- dition, the errors are given in Table Il. The mean absotute per- centage ertor (MAPE) is less than 1.3%, Thus, the identification accuracy with AC is areaily satisfactory. ‘Aumnonzed licensed use imted to: Lulea Lnversty of Tecnology. Downloaded on June 08,2020 at 08:31:29 UTG trom IEEE Xplore. Restrictions applyTABLE IL Coupanisow WitH SIMELATION To MEASUREMENT Mawral Simulated Siac Hour anew aa /kW_—EOCI% ean raw HOW Tao 4097 00 2 392A HS 3 rs 1976 1803 480 T64 19 me 32 som 4868 sa 44s 16 6 83 HS 35 46 of 7 Ri si Ht 60 113 8 es 2a oA 9 rs 1023 4 10st 29 83 OH ss 3636823 b 588 S37 OSA SM wm 90D re ) Is 6300 8807 eso a8 16 aso 2g 8D re re es) e063 SOLE SIS 1) 4330 M673 a8 2 17 BSS HRA a1 Sod SR OSs Domo BT OTT 27363 AL 7661 De sa63 SLAs a —+Measurement Simulation =~ Statistical mean Active Power kW Time /h Fig. 9. Probebiistic simulation rvult of charging load ‘Moreover. the proposed model can be used to produce prob- abilistic results with Monte Carlo method, According to (3), set jeand as 123 and 17 min, respectively. Then, pseudo-random numbers can be produced as the charging durations. Based on Fig. 3, a computer program has been built for simulation. The 1-h power values for 100 Thursday samples are produced and shown in Fig. 9 with crosses. The statistical mean values are also presented with a dash line. Although the charging duration finctuates significantly, the charging load profite still presents a clear and stable feature, At each hour, the power values have an expectation which is very close to the real one. The errors of mean values are given in ‘Table II, and its MAPE is less than 4.4%, [EEE TRANSACTIONS ON POWER SYSTEMS, VOL. 30,NO.4, JULY 2015 —1G = GA. Fig 10. Accuracy comparison with AC to OA, TABLE I PearoRMAnce CoNPafisos Witt ACTO GA Algorithms Computation cost MAPE: Resquate xe Toa 0.985 Furthermore, a comparative analysis has been carried out to benchmark the identification approach of AC algorithms in per- formance. Based on the global error of charging load, a com puter program with GA has been developed for CTF identifica- tion. A computer is used to run AC and GA in single-treaded mode. The software SupesP! which calculates the mathematical constant x to a specific number of decimal digits is selected to test the single-tneaded computing capability of this compute: ‘The computational cost of x up to 1 million decimal digits is about 13 s using the version 1.9 of SupesPl. The comparison re- sult of AC and Ga is presented in Table II and Fig. 10. The AC algorithm has significant advantages in accuracy and efficiency since (11)-(13) have been utilized to enhance its local searching capability. ‘VI. CoNcLUSION The decisive factors affecting charging load characteristics are the distribution of durations, the power profile of chargers, and the charging start time. The proposed CTF can be intro- duced to describe the charging start time, Furthermore, it con- tains the properties of both space and time. ‘The proposed model provides flexible modes for simulation and parameter identification. In simulation, it can be used in both stochastic and deterministic situations. In parameter iden- tification, (5) provides a rapid calculation method to improve the efficiency. ‘All elements of CTF are linked one by one, and the charging power at a certain time is determined by a group of elements ‘within the previous charging duration. Thus, the parameter iden- tification does not entirely depend on a global error, With a par- ticular design on the objective fimetion, the proposed AC algo- rithm has shown outstanding performance in accuracy and com- putational efficiency. ‘Aumnonzed licensed use imted to: Lulea Lnversty of Tecnology. Downloaded on June 08,2020 at 08:31:29 UTG trom IEEE Xplore. Restrictions apply‘YANG oral: LOAD MODELING AND IDENTIFICATION BASED ON ANT COLONY ALGORITEMS FOR EV CHARGING STATIONS CTF brings a new approach to the study on EV charging load profiles, To benefit fiom the interdisciplinary cooperation, the research results on traffic activities can be used on EV load ‘modeling. In the future, a universal platform can be built for large-scale EV applications based on the concept of CTF. REFERENCES (1a P Lopes, FJ. Soeres, and P.M R_ Almeida, “ategration of levtrc veicles in the elecne power system,” Proc JEEE, val. 98, fo 1 pp L6e-182 Jan 2011 (0) M Etszedi-Amol, K- Choma, and J. Stefi, “Repidcharge ele ‘mevshiele stations” JEEE That Pover Del, val 25 00.3. pp 1883-1887, Jul. 2010. [51 1.6. Vlachopiauis, Probabilistic constrained load ow considering intesration of wind power generation and elec Vehicle,” IEEE Thane. Power Set, sol 24,0. 4. pp 1808-1817, Nov. 2008 [4] G Li and X-P Zhang, “Modeling of plug-in hybrid eletne vehicle charging demand un peobabistis pomer ovr ealsulatins” JEEE Trans. Smar: Grid, vl 3, 20.1, pp. 492198, Mar 2012 K- Qian C. Zips, M Allan and Y, Yuan, “Modeling of load demand due to EV battery charging in dstibition systems” IEEE Trans ‘Power St, vol 26,n0. 2, pp £02-810, May 2011 (e] ¥ Cao, 5 Teng C Li, P. Zhang, ¥. Ten, Z Zhang, end Li, "An optimized BV cheng model consider TOU price and SOC eure IEEE Tras Sart Grid OL 3.20. 1. pp. 31-138, Mar. 2012 171 P. Papadopoulos, 8. Skarves-Kazakos,L Grau. LM Cipcigan, and 1 Janine “Elesti vehislet inapacton Bntiehdstbution netwe HET Elecer Sit Trap, vol 2,20. pp. 91-102, Apr 201 Is] § Shao. Mt Pipatansscmpora and § "Rahman, “Cri integration of slectri vehicles and demand response with customer choice” IEEE Trans. Smart Gri, vol. 3.0.3, pp. 543-550, Mar. 2012 [8] ¥ 2hens, 2 ¥ Done, ¥ 3, Mens, 1 18 Zhao, ands Qiu, “Eletric vehicle battery charging swap stations in distibution systems: Com- parson study and optimal planing,” IEEE Trans. Power $)st, vol 28, no. pp. 221-999, Jan. 2012 [io] 4 Lassa Y Tikka, 1 Hansa, and J Partanen, “Elects cae a8 past ‘fstectrcity distiution-—Who pays, wh benefits?” ZET Bec. St Trangp. vol 2 no. 4 pp. 186-194 Feb 2012 (ul) P. Seichce Marin, G: Sanchez, and G, Morals Ezpais, “Direct load contol decision model for agarezeed EV charzia points,” /EEE Grane, Power Set, vol. 27, m0. 3, pp. 1577-1584, Aug. 2012 [12] A. Lojouska, D. Korowicka, G.Papoefaou, and Evan der Shi Stoctartie modeling of poser demand due to EVs using copula, IEEE Trane Power Sy, $9) 27-134 pp 1960-1968, Ne 2012. 103] ¥ Ma, T Houghton. A. Cruden, endD. ined, “Modeling the benefits of ahila to ard technolo toa power sytem" LEEE shane. Power ‘jst, vol 27, no. 2, pp. 1012-1020, Mey (04) K Clement-Nyus,E-Haesen and J-Driesen, “The ampac of charuing plug-in yb elcric vehicles on aresidential distribution grid” IEEE Trans. Power Sst. vol 25, no. 1, pp. 371-380, Feb. 2010 [is] M-A. Onege-vaaquez,F. Boufford, and V. Silva, “Elect vehicle aggrepator system operator coowdiation for charging scheduling a fercices procurement” EEE Trans Power 5). fo be pubkisied [Hg] 8. Huang andD Intel, The impact of domestic plug-in hybrid elec. ‘wie vehicles on power distribution system lads in Proc Te. Conf Power Sit Tecino, Ox 2010, [17] 7-Zheng Wang Men C Zhu, and. Zin, “Agasepstion model ised optimization for electric vebicls charging araegy, EEE Trane 2, 13 Smart Grd, vl. 4, 0.2, pp. 131-138, Jun 18] P Zhang. K. Qian, © Zhou, BG Stewart, and DM. Hepbum, “A smathodology for optimization of poner eyes dem duet clatic Yehicl charging load" ZEEE Tanz, Power Sjst vol 27. m0. pp 1628-1636, Ava. 2012, Shaobing Yang received the BS. desree Som Sangha Jiaotong University, Shanzhat China 1995 and the MS. degre ffom Beijing Tiaotong University, eying China, im 2009, both electncl “le ss an Associate Professor of the School of tecrical Enpinsenng it Bajing listen Unive sy. His resarch ifersse include power system Smulation, load modeling, demand ‘side manaze- tant slecine power qusliy, and load forseasting fbr cles veces Mingli Wu recsivad the BS. and MS. degrees in sloctncalenainoonng from SouestSgotong Unt versity, Chengdu, China, in 1993 and 1996, respec twvely and the PhD. degree elecncal engsneerng ‘fom Befing Jiaotong University, Bejing, Chine 2006 Hs isa Professor ofthe School of Electrical En gineering in Bayjing Taotong University. is current fessor areas anclads pore supply for elec ‘says dita simulation of power =ystams, and else tec poner quality iu Yao (S'10) received the BS. and MS. depres fiom Xi'an Jiaotong Univers, China, in 2007 and 2010, respectively. She is curenily pursung the PAD. degree in The Ohio State Unversity Columbus, OF, USA, Her current reewch interests include do ate ft detection in high vottags aad power eletronte systems armonic elimination oF multilevel verter, tnd the ution of toda snore! converter tm igh power systems Jiachun ang (MEL2-SMEL4) wae Bom ia Ji ‘Province, Chita. He received the BS. depree in ‘lsctacal enainsering and the Ph.D deseo in power ‘stem automation from Northem Jiaotong Univer ‘iy, Being, China, 21993 and 1999, spectively. He is curently a Profesor with the Sehool of ‘Elecrical Engineering) Being Jaoag University main interests are related to ater application technology, elevine car charging stations, and tmicr-ardtshneloay ‘Prof, hang recesved the Notional Scisoe and ss 2nd Award for hi wore cn EV Bus sytem. and the Bajing Scrnce and Technology Progress 2nd Award for his work oa EV harping sytem, Teemnclogy Po ‘Aumnonzed licensed use imted to: Lulea Lnversty of Tecnology. Downloaded on June 08,2020 at 08:31:29 UTG trom IEEE Xplore. Restrictions apply