Analysis of Optimal Integration of EVs and DGs Into CIGRE’s MV Benchmark Model

In today’s transportation sector, the growing number of electric vehicles (EVs) is progressively replacing petroleum-fueled vehicles, which are also expected to minimize greenhouse gas emissions. The main problem with EVs is the requirement of charging energy, which is fed through distribution network, while simultaneously feeding already connected load. The EV’s integration with the distribution network will overload the network due to EV’s charging load, which will eventually trip the power system protection. In this context, it is necessary to minimize power losses, improve voltage profile with sustainable power supply network. Therefore, optimal placement of EVs is required in the distribution network. Conventionally, researchers have used DGs to minimize power losses and improve voltage profile. In this paper, authors analyzed the effect of EV’s integration with simultaneous placement of distributed generations (DGs). The integration of EVs with higher penetration of DGs is cumbersome due to higher power losses and voltage variances that are outside allowable boundaries. The optimal placement of EVs into the distribution system with higher penetration of DGs is proposed in this paper using a battle royale optimization (BRO) algorithm. Since it is a new problem on CIGRE network which is not discussed before. Hence, the authors compared results with three most famous algorithms namely genetic algorithm (GA), particle swarm optimization (PSO), and accelerated PSO (APSO). The optimization problem is developed as a multi-objective function while decreasing active and reactive power losses, and minimising maximum voltage deviation index. The studied distribution network is the CIGRE 14-bus medium voltage (MV) distribution network. Three case studies are taken in which EVs are integrated in two scenarios with optimally sized and located DGs systems in the CIGRE distribution network using MATLAB. Case-1 includes simple network with DGs integration. Case-2 includes EVs only in the simple network, and case-3 finally includes EVs and DGs for optimal placement with minimum losses. The placement of the EVs results in a decrease in power losses and voltage deviation indices. The bus voltages of case-2, on the other hand, stay unchanged when the EVs are integrated. Case 3 with BRO showed the large reduction in power losses owing to the addition of EVs to the distribution network with DGs (from 19.98 kW, and 19.89 kVar to 2.54 kW, and 3.35 kVar).


I. INTRODUCTION
Distributed generations (DGs) such as photovoltaic (PV) 36 systems and electric vehicle (EV) are rapidly transforming 37 modern electricity systems. PV module and accessory prices 38 are continuing to fall, resulting in a rise in DGs adoption 39 in many countries [1]. The word prosumers was coined to only proven to be environmentally friendly as a source 44 of renewable energy, but it has also been shown to be critical 45 to the electrical distribution system due to less power losses 46 and improved voltage profiles [3]. These factors result in 47 lower electricity bills paid to the utility provider for con-48 sumers (prosumers) [4]. 49 The speedy penetration of EVs in the transportation sec- 50 tor, on the other hand, is revolutionary, and combining this 51 technological advencement with DGs is the very anticipating 52 option to reduce reliance on fossil based fuels with reduction 53 in GHG emission [5]. Moreover, as crude oil depletes and 54 has a negative effect on environment, the future of oil-based 55 vehicles is getting bleaker around the globe as EVs gain 56 popularity [6]. Furthermore, EVs have Benefits such as being 57 noise-free, fuel-efficient, and emission-free [7]. 58 Therefore, the fast penetration of EVs in the transport sec- 59 tor, similar to solar PV systems, will have positive impact on 60 the environmental as well as the electrical distribution system, 61 as they could help supply frequency and voltage backup while 62 being employed as spinning reverses to cope with abrupt load 63 surges or loss in some generators [8]. However, EVs should 64 be integrated into the electrical distribution system with great 65 care because they can overload the network, resulting in 66 increased power loss [9]. There have also been reports of 67 power quality degradation and voltage variations that exceed 68 allowed limits [10]. The problem becomes much more dif-69 ficult when EVs are integrated into a electrical distribution 70 system with a penetration of DGs systems. This needs the 71 best feasible placement of EVs and DGs in the electrical dis-72 tribution system to minimise their negative impact on network 73 power losses and voltage stability. 75 Many studies on the best location for EVs have been pub-76 lished. In [11], the authors used TSO and compared with 77 HHO and TLBO strategies to minimise the multi-objective 78 problem, which intended to reduce active power loss and 79 average VDI while maximizing VSI. The sizes and place-80 ment of DGs were optimized to reduce the network's impact 81 from EVs. Based on an intelligent EMS, the authors in [12] 82 recommended the optimal location of EVs into the electri-83 cal distribution network. This technique included two lev-84 els: one for active PMS at the nodes, which intended to 85 lower the daily total cost of active power used by EVs, 86 and another for reactive PMS at the system level, which 87 aimed to reduce power loss in the system by utilising the 88 EVs' reactive power rating. In [13], the authors proposed a 89 strategy for optimally locating EVs. The authors in [14] sug-90 gested stackelberg game theory for the optimal allocation of 91 EVs, with the theory being applied to analyse the interacting 92 behaviour between distribution corporations and EVs own-93 ers. In [15], the authors discussed case study for designing 94

152
• RERs integration is not considered in other distribution 153 networks (IEEE-33/69/118) which are purely radial net-154 works. Therefore, the authors considered CIGRE 14-bus 155 MV distribution network for the analysis, which already 156 used placement of RERs and energy storage devices. 157 The conventional multi-objective function which were 158 most commonly used in distribution networks, is consid-159 ered in this research for CIGRE 14-bus MV distribution 160 network to optimal place the EVs along with DGs. The remainder of the paper is laid out as follows. The role 200 of the BRO, GA, PSO, and APSO algorithms is outlined 201 in Section II of the methodology of the examined CIGRE 202 14-bus MV distribution network. The steps for implement-203 ing four algorithms are also described. This section also 204 VOLUME 10, 2022 includes distribution system modelling. This section also 205 covers problem formulation, including objective functions 206 and constraints. The results are analysed and discussed in 207 Section III. The comparison of BRO with three algorithms is 208 also detailed. Critical analysis and discussion are presented 209 in Section IV. Section V contains the conclusion.

211
As the EVs are assigned in the distribution system, the goal 212 of the optimization problem is to reduce the system's power 213 loss, as well as the maximum VDI. i.e., f obj2 as defined below [26], [27]: where ω 1 &ω 2 are weights assigned to each objective func-221 tion. ω 1 &ω 2 are taken as 0.5 and 0.5, respectively.

222
The normalization for multi-objective function is carried 223 out using mathematical functions which are discussed in [28], 224 [29], and [30]. In first step of normalization, each objective 225 is converted into unitless quantity by using power equation
Similarly, power capacity limit supplied by DGs is [27] is considered at any instant in the simulation.

307
In this study, the CIGRE 14-bus MV distribution network 308 is taken as the study system. As indicated in figure 1 [ Furthermore, BRO, like other swarm-based algorithms, 326 uses random populations that is uniform in distribution across 327 the search space. Then, using a weapon, each individual 328 attempts to harm the closest soldier. As a result, soldier in best 329 position harms its immediate surroundings. The procedure of 330 applying BRO method is described as follows:

331
Step1: Initialize the solutions population, i.e., the maxi-332 mum threshold value, total number of iterations, and ran-333 domly generated initial population;

334
Step2: Each player in the game aims to injure the closest 335 enemy soldier. When one soldier is injured by another sol-336 dier, the damage scale of that soldier is raised by one. It is 337 mathematically expressed as [46]: where X iDSol represents damage scale of the i th soldier in the 340 whole population.

341
Step3: Soldiers prefer to change positions as soon as they 342 are damaged, attacking opponents from a different direction. 343 As a result, in order to concentrate on exploitation, the injured 344 solder travels to a spot between the prior position and the best 345 position determined thus far (elite player where rand represents random numbers which are uniformly 349 distributed between 0 and 1. X DSol,d describes location of 350 the injured individual in dimension d. X EliteSol,d is the best 351 location of the injured (elite) individual in dimension d.

352
Step4: If a soldier's damage scale exceeds a predeter-353 mined threshold value (a value of 3 is taken to avoid pre-354 mature convergence), the soldier dies and the soldier will be 355 revived/reallocated at random from the viable problem space. 356 The value of X iDSol will be reset to zero (i.e., X iDSol =0). 357 Mathematically, it can be related as follows [47], [48]:  non-linear and multimodal [51], there is no need to employ 407 individual best. Therefore, velocity factor at t+1 iteration is 408 calculated as follows [52]: To further enhace convergence, updated position can be 417 written in one step as follows [52]: where α&β values are from 0.1 to 0.4 and 0.1 to 0.7, respec-420 tively. α&β values which are used are 0.2 and 0.7 respec-421 tively.

422
The solution steps of APSO are discussed as follows:.

423
Step1: Start with reading input data such as buses, 424 branches, base values, and required accuracy.  Step3: Starting initialization with swarm velocity and loca-428 tion, constants, and iterations.

431
Step6: Comparing individual best from each particle. 432 Selecting the particle with lowest individual Pbest as Gbest. 433 Step7: Updating particle velocity and location.

434
Step8: Check if maximum number of generations (itera-435 tions) max is reached, then go to Step 9. Otherwise go to 436 step 7 for t=t+1.

437
Step9: The solution set is G best , and P best .

439
PSO was formulated by J. Kennedy in 1995 [53]. In every iter-440 ation of the PSO algorithm, the velocity and location of every 441 swarm particle are updated. PSO algorithm process is based 442 on swarm or population of P particles, which are randomly 443 located in the searching space S. Every particle i (from 1 to 444 P) is represented by the positions The mathe-446 matical expression for velocity and position is [52]: where w represents inertia weight, a 1 and a 2 represent accel-450 erating coefficients, n 1 and n 2 show random numbers from 451 0 to 1, X t+1 i,j shows i th particle position towards j th direction 452 during t th iteration. P best and G best represent positions for 453 personal and global best respectively. The minimum and 454 maximum limits for velocity V and position X are defined 455 as [−V max V max ] and [X min X max ] respectively. t ranges from 456 1 to maximum number of iterations t max .

457
The PSO steps for the solution are as follows:

462
Step4: Calculating the fitness of each swarm for P best , and 463 comparison of other swarms for G best .

464
Step5: Changing swarm position X and velocity V .

465
Step6: Changing P best and G best . Step8: The outcome is G best , P best, and its relevant position 469 X .    Step4: Applying the crossover with the probability to each 484 gene of the off-springs by randomly selecting two chromo-485 somes (i.e., parents) to generate two off-springs.

486
Step5: Applying the mutation with the probability. The 487 random number from the possible range is selected for the 488 mutated gene.

489
Step6: Evaluating the off-springs by calculating the objec-490 tive function for each chromosome of the off-springs.

491
Step7: Replacing the worst chromosomes from parents and 492 off-springs for selecting the population of the next generation. 493 Step8: Repeating steps 4 through 7 until the limit of itera-494 tions is attained. The best chromosome selected so far is the 495 optimal solution of the studied problem.

496
The flow chart of implemented GA algorithm is illustrated 497 in figure 3 [31].    Case-2 shows only EVs intergration, while case-3 incorpo-516 rates both DGs and EVs at the same time. Table 1    electricity tariff in $/kWh, I rate is interest rate, and I inf is 520 inflation rate.

522
The optimal placement of DGs and EVs with the proposed 523 BRO algorithm are shown in table 2 and table 3 below for 524 each simulation case. For case 2, optimal locations of EVs 525 is selected by all algorthms including BRO and other three 526 algorithms. The authors have run the algorithms on multiple 527 times and the results for optimal EVs' location is the same 528 bus.

529
The authors have used the same parameters and load data 530 of CIGRE 14-bus test system which is given in [43], [58], 531 [59], and [60] and the readers can see the details from these 532 references. The authors have taken 4 load cases in the context 533 to analyze the performance improvement. Positive Q (+Q) 534 mean to inject the reactive power into the distribution system. 535 While negative Q (−Q) mean to draw/absorb the reactive 536

565
In case-1 with active and reactive load (P-Q) as shown in 566 figure 6, voltage profile with APSO shows higher magnitude 567 at each bus (except bus-3) as compared to the other three algo-568 rithms. Meanwhile base case shows higher voltage at bus-3 as 569 compared to APSO. PSO performance is better than BRO and 570 GA. While performance of BRO shows better performance 571 than GA. GA shows poor performance as compared to other 572 three algorithms. Voltage magnitude at bus-3 is lowest for 573 three algorithms (except APSO) as compared to the base 574 case. The minimum voltage magnitudes for APSO, PSO, 575 BRO, and GA are 0.96258, 0.96248, 0.87065, and 0.86152, 576 respectively. The maximum voltage magnitudes for APSO, 577 PSO, BRO, and GA are same as 1.03.

578
In case-1 with reactive load (Q) as shown in figure 7, volt-579 age profile with GA shows higher magnitude at each bus as 580 VOLUME 10, 2022     algorithms except bus-6 when BRO outperform PSO. APSO 604 also shows mixed trend during the simulation, with better 605 performance than BRO and lower than PSO and GA. APSO 606 performance is lower than GA throughout the simulation, 607 while it shows slightly better performance than PSO at bus 6. 608 Voltage magnitude at bus-3 is lowest for all four algorithms as 609 compared to the base case. The minimum voltage magnitudes 610 for PSO, GA, APSO, and BRO are 1.0067, 1.0061, 0.99365, 611 and 0.99113, respectively. The maximum voltage magnitudes 612 for PSO, GA, APSO, and BRO are 1.0664, 1.0631, 1.0424, 613 and 1.0396, respectively.

614
In case-3 with active and reactive load (P+Q) as shown in 615 figure 10, voltage profile with GA shows higher magnitude 616 at each bus as compared to the other three algorithms. BRO 617 performance is lower among all three algorithms, except bus 618 13 when BRO and APSO have identical trend. than PSO and 619 APSO. PSO performance is better than APSO until bus 8 and 620 bus 12, 13, 14. PSO and APSO have same trend at bus 9, 621 11. APSO shows better results at bus 10 as compared to PSO. 622 Voltage magnitude at bus-3 is lowest for all four algorithms as 623 compared to the base case. The minimum voltage magnitudes 624 for PSO, GA, APSO, and BRO are 0.80183, 1.03, 0.99616, 625 and 1.0261, respectively. The maximum voltage magnitudes 626 for PSO, GA, APSO, and BRO are 1.0577, 1.2186, 1.0556, 627 and 1.0895, respectively.

628
In case-3 with active and reactive load (P-Q) as shown in 629 figure 11, voltage profile with PSO shows higher magnitude 630 at each bus as compared to the other three algorithms, except 631 bus 3 when PSO and BRO have same trend. GA performed 632 better than APSO from bus 3 to 6, while APSO shows better 633 performance than GA from bus 7 onward. BRO performace is 634 lowest among all algorithms. It is worth to note that voltage 635 profile with BRO is lower than base case for Buses 4 to 8, 636 11 to 12. GA also have lower voltage than base case for bus 637 11, while GA shows no improvement at bus 12 as compared 638 to base case. Voltage magnitude at bus-3 is lowest for all 639 four algorithms as compared to the base case. The mini-640 mum voltage magnitudes for PSO, GA, APSO, and BRO 641    Maximmum VDI is the maximum voltage deviation from the 666 base voltage. It is observed from simulatin results that the 667 maximum VDI is drastically reduced as compared to the base 668 case.

669
In case-1 with active load (P) as shown in figure 19, 670 maximum VDI for PSO is higher among four algorithms, 671 followed by GA and APSO. While BRO shows best per-672 formance among all algorithms with the lowest value of 673 MVDI. MVDI for PSO, GA, APSO, and BRO are 0.0679, 674 0.066, 0.0649, and 0.0626, respectively. In case-1 with active 675 and reactive load (P+Q), maximum VDI for BRO is higher 676 among four algorithms, followed by GA and PSO. While 677       In case-2 with active load (P) as shown in figure 20, 692 maximum VDI for four algorithms are same, which is 0.2798. 693 In case-3 with active load (P) as shown in figure 21, 694 maximum VDI for BRO is higher among four algorithms, 695 followed by PSO and APSO. While GA shows best perfor-696 mance among all algorithms with the lowest value of MVDI. 697 MVDI for BRO, PSO, APSO, and GA are 0.0374, 0.0353, 698 0.0344, and 0.0279, respectively. In case-3 with active and 699 reactive load (P+Q), maximum VDI for BRO is higher 700 among four algorithms, followed by GA and PSO. While 701 APSO shows best performance among all algorithms with 702 the lowest value of MVDI. MVDI for BRO, GA, PSO, and 703   BRO, APSO, PSO, and GA are 7.4219 kVar, 6.8282 kVar, 731 6.7819 kVar, and 6.7724 kVar, respectively. In case-1 with 732 active and reactive load (P+Q), active power loss for PSO 733 is higher among four algorithms, followed by GA and BRO. 734 While APSO shows best performance among all algorithms 735 with the minimum active power loss. Active power loss for 736 PSO, GA, BRO, and APSO are 15.2687 kW, 7.7606 kW, 737 4.4529 kW, and 4.2542 kW, respectively. Reactive power loss 738 for PSO is higher among four algorithms, followed by GA 739 and BRO. While APSO shows best performance among all 740 algorithms with the minimum reactive power loss. Reactive 741 power loss for PSO, GA, BRO, and APSO are 13.1145 kVar, 742 10.0438 kVar, 6.0705 kVar, and 5.6120 kVar, respectively. 743 In case-1 with active and reactive load (P-Q), active power 744 loss for GA is higher among four algorithms, followed by 745 BRO and APSO. While PSO shows best performance among 746 all algorithms with the minimum active power loss. Active 747 power loss for GA, BRO, APSO, and PSO are 7.5669 kW, 748 4.3804 kW, 2.7544 kW, and 2.7541 kW, respectively. Reac-749 tive power loss for GA is higher among four algorithms, 750 followed by BRO and APSO. While PSO shows best per-751 formance among all algorithms with the minimum reac-752 tive power loss. Reactive power loss for GA, BRO, APSO, 753 and PSO are 8.2863 kVar, 5.8267 kVar, 3.6131 kVar, and 754 3.6118 kVar, respectively. In case-1 with reactive load (Q), 755 active power loss for BRO is higher among four algorithms, 756 followed by PSO and GA. While APSO shows best perfor-757 mance among all algorithms with the minimum active power 758 loss. Active power loss for BRO, PSO, GA, and APSO are 759 13.0556 kW, 12.6698 kW, 12.5192 kW, and 12.5094 kW, 760 respectively. Reactive power loss for BRO is higher among 761 four algorithms, followed by GA and APSO. While PSO 762 shows best performance among all algorithms with the min-763 imum reactive power loss. Reactive power loss for BRO, 764 GA, APSO, and PSO are 13.5509 kVar, 13.2496 kVar, 765 13.2173 kVar, and 13.1872 kVar, respectively.   2.5323 kW, and 2.4667 kW, respectively. Reactive power loss 799 for GA is higher among four algorithms, followed by BRO 800 and PSO. While APSO shows best performance among all 801 algorithms with the minimum reactive power loss. Reactive 802 power loss for GA, BRO, PSO, and APSO are 5.5547 kVar, 803 4.912 kVar, 3.3357 kVar, and 3.1914 kVar, respectively. 804 In case-3 with reactive load (Q), active power loss for APSO 805 is higher among four algorithms, followed by PSO and GA. 806 While BRO shows best performance among all algorithms 807 with the minimum active power loss. Active power loss for 808 APSO, PSO, GA, and BRO are 57.0725 kW, 17 The obtained results are shown in table 4. Kt is total invest-817 ment cost. In case 1 with P type DGs, total cost of $49384142 818 shows that BRO algorithm performance is better. APSO is 819 the second best algorithm for DGs type of P+Q and Q with 820 total cost of $49386214 and $49346985, respectively. PSO 821 is the third best algorithm for DGs type of P-Q with the 822 total cost of $49389185. GA is not recommended in view 823 of cost-effective perspective of the proposed DGs integrated 824 distribution system.     The results of the proposed BRO algorithm for the optimal 835 locations of EVs and DGs in the distribution system are com-836 pared with the obtained results of three algorithms namely 837 GA, PSO, and APSO. The obtained results are shown in 838 table 5. It is observed that the proposed BRO algorithm per-839 formed well in most of the cases. First, it is also observed that 840 PSO performance (specially with P+Q and Q) is declining. 841 The possible reason for this effect is due to the result of 842 PSO trapping in the local optima without reaching the global 843 optimum point. However, BRO algorithm is better in all cases 844 except load type Q of all case 1 and case 3. APSO performed 845 better than PSO with the exception of load type Q of case 846 3 when power losses are significantly high. GA and PSO are 847 somewhat similar in performance. Second, the bus voltage 848 profiles obtained after integration of EVs and DGs (case 3) 849 are relatively higher and better than case 1 with DGs only and 850 case 2 with EVs only. This shows better performace of BRO 851 algorithm for the improvement of system voltage profiles dur-852 ing integrated EVs and DGs. It also validates superior voltage 853 stability with the proposed method. Moreover, power losses 854 are also reduced which reveals effectiveness of the imple-855 mented approach to find the best locations for EVs and DGs 856 VOLUME 10, 2022 with minimum active and reactive power losses. In this way, 857 BRO algorithm proved better in performance and its strenghs 858 can be used to replace the shortcomings of other algorithms.     DGs. It also supports the proposed method's higher volt-935 age stability. Additionally, power losses are decreased, 936 demonstrating the efficiency of the implemented strat-937 egy to identify the optimal sites for EVs and DGs with 938 the least amount of active and reactive power losses. 939 In this approach, the BRO algorithm demonstrated supe-940 rior performance, and its advantages can take the place 941 of the weaknesses of other algorithms.

943
The effectiveness of BRO for EVs and DGs placement vali-944 dates its further application as distribution system operators 945 for providing long-term, cost-effective, reliable, and cheap 946 services to users, while ensuring acceptable power quality 947 and voltage within limits. This is because the algorithm's 948 restrictions are quite similar to the planning horizon's require-949 ments.

950
The daylight fluctuation of RERs generation, the driving 951 priorities of EV drivers, distribution system uncertainnesses, 952 and the charging period of EVs can all be included in the 953 future scope of this research for the most optimal locations 954 of EVs and DGs in the distribution system. These steps can 955 be employed to assess the BRO algorithm's resilience.
considering intermittency of EVs and RERs using stochastic