An Incentivized and Optimized Dynamic Mechanism for Demand Response for Managing Voltage in Distribution Networks

The voltage regulation in distribution networks is one of the major obstacles when increasing the penetration of distributed generators (DGs) such as solar photovoltaics (PV), especially during cloud transients, causing potential stress on network voltage regulations. Residential demand response (DR) is one of the cost-effective solutions for voltage management in distribution networks. However, the main barriers of DR implementation are the complexities of controlling a large number and different types of residential loads, satisfying customers’ preferences and providing them fair incentives while identifying the optimum DR implementation locations and sizing as well as cooperating with the existing network equipment for the effective voltage management in the networks. A holistic and practical approach of DR implementation is missing in the literature. This study proposes a dynamic fair incentive mechanism using a multi-scheme load control algorithm for a large number of DR participants coordinating with the existing network equipment for managing voltage at medium voltage (MV) networks. The multi-scheme load control is comprised of short-interval (10-minute) and long-interval (2-hour) DR schemes. The dynamic incentive rates are optimized based on the energy contribution of DR participating consumers, their influence on the network voltage and total power loss improvement. The proposed method minimizes the DR implementation cost and size, fairly incentivizes the consumers participating in the DR and priorities their consumption preferences while reduces the network power losses and DGs’ reactive power contributions to effectively manage the voltage in the MV networks. An improved hybrid particle swarm optimization algorithm (IHPSO) is proposed for the load controller to provide fast convergence and robust optimization results. The proposed approach is comprehensively tested using the IEEE 33-bus and IEEE 69-bus networks with several scenarios considering a large number of DR participants coordinated with the DGs and on-load tap changer (OLTC) in the networks.


I. INTRODUCTION
Over the recent years, the use of renewable energy sources 24 (RESs) in the form of distributed generators (DGs) have 25 The associate editor coordinating the review of this manuscript and approving it for publication was Lei Chen . increased considerably [1]. The conventional distribution sys-26 tems have not been designed with the consideration of bidi-27 rectional power flows from the RESs, which create major 28 challenges for distribution system operators to maintain the 29 system reliability and power quality within the standard 30 limits. Cloud-induced transients over the solar photovoltaic 31 voltage variation is higher than the reactive power. Therefore, 88 unlike the transmission networks, the voltage magnitude in 89 the distribution networks can be controlled by the active 90 power from the consumers through demand response (DR) 91 programs effectively. 92 Therefore, one of the promising means of utilizing the 93 existing infrastructure for managing network voltage is the 94 optimal control of end-users' loads through DR programs 95 [7], [8], [9] incorporated with home energy management 96 systems (HEMSs) [10], [11]. Utilities can communicate with 97 the consumers' HEMSs, which can switch ON and OFF 98 the DR participated appliances almost instantaneously and 99 enable them to react fast to maintain the network voltage 100 effectively [12]. It can postpone the investments on the gen-101 eration resources and network upgrades [13]. HEMS helps 102 utility for DR implementation by providing information such 103 as household appliances' real-time energy consumption sta-104 tus and consumption preferences set by consumers and by 105 receiving load control signals from the utility to control the 106 appliances. HEMSs are developed considerably over the past 107 years, and the smart grid technologies like smart metering 108 and appliances for load control via HEMS are becoming 109 more attractive for the modern distribution networks [11], 110 [12]. One of the main challenges in the implementation of 111 DR is how to fairly incentivize participants for encouraging 112 them to contribute on a DR program. The incentives to the 113 participating consumers in a DR program should not be fixed 114 or equal across all conditions in a network during an event 115 of voltage or thermal limits violations, namely defined as the 116 DR event in this paper. An incentive scheme should be fair 117 based on the DR participant consumers contribution in each 118 DR event and their locations within the network. A study 119 in [14] propose a fair incentive mechanism for the customers 120 to improve the power quality problems in the network. In the 121 following, the literature of the subject will be investigated in 122 more detail. 123 The growing penetration of electric vehicles into the dis-124 tribution network may create huge challenges, which require 125 proper optimized operation for the distribution network [15]. 126 A study in [16] developed a strategy for peak shaving in the 127 distribution network via electric vehicle aggregators which 128 in turn leads to cost reduction and oversaturation of the 129 distribution transformers. Authors in [17], [18], and [19] 130 propose an optimal planning for renewable generators imple-131 mentation and electric vehicle operations to minimize the 132 voltage deviations and power flow from the main grid as 133 well as minimize the power loss in the microgrids. Likewise, 134 in [20] and [21], demand response program in the presence 135 of an energy management scheme was provided to optimally 136 schedule the electric vehicles as well as adjust the household 137 appliances. The study in [22] developed a reward-based load 138 control algorithm to shave network peaks, where consumers' 139 and utility's profits are considered. It shows that the pro-140 posed reward-penalty can move forward the organized opera-141 tional characteristics and relieve the top bounce back without 142 forcing more costs on consumers. However, only a limited 143  The study in [6] provides a large percentage of real-time 180 balancing reserve for the MV network by aggregating electric 181 water heaters (EWHs) for load shifting while maintaining the 182 consumers' comfort levels. In [26], a virtual energy storage 183 system concept is proposed considering EVs and ACs to 184 cater for the comfort levels of consumers at different indoor 185 temperatures. However, these studies are limited to control of 186 few appliances. Multi-layers DR study in [27] uses only air 187 conditioner (AC), EWH, and cloth dryer to satisfy both utility 188 and consumer preferences. A load shedding optimization 189 technique is proposed in [2] for the utility to maintain their 190 network voltage considering a limited number of household 191 appliances. These studies consider only a few selected appli-192 ances from a limited number of DR consumers in the load 193 control, assuming all the consumers have similar appliances 194 with fixed kW ratings of appliances without any fair opti-195 mized incentive distribution. In reality, the appliances' power 196 ratings and their availability vary between the consumers and 197 may not be the same across all participating consumers in 198 a DR event. Therefore, a realistic approach considering the 199 variability of household appliances and their different kW 200 sizes for a large number of DR participants are yet to be 201 investigated in the load control algorithm.

202
Toward this end, this study introduces a holistic multi-203 scheme load control strategy for managing multi-interval 204 voltage fluctuations in the MV networks and minimizing the 205 power loss with the following main contributions: 206 1) A dynamic incentive mechanism is proposed for fairly 207 rewarding the DR participating consumers based on 208 their energy contribution and their influence on the 209 network voltage and loss improvements. 2) The load control algorithm is developed to optimize 211 the DR participants' locations and support their con-212 sumption decisions to maintain their comfort levels by 213 considering appliances' switching status, disturbance 214 ratio and their fair interruption in the DR event.

215
3) An improved hybrid particle swarm optimization 216 (IHPSO) algorithm is proposed in the load controller to 217 provide fast and robust convergence in handling a large 218 number of DR participants and objective parameters 219 such as minimizing the network loss, DGs' reactive 220 power contribution, DR cost and sizing, fair incen-221 tive distribution and the participants' consumption 222 preferences.

223
The rest of the paper is organized as follows; Section II 224 presents the proposed objective function, dynamic fair incen-225 tive strategy, multi-scheme DR for voltage management and 226 consumer preference definitions and modelling. Section III 227 explains the solution approach for the multi-interval voltage 228 management. Section IV provides simulation results of the 229 proposed approach, and the relevant conclusion is presented 230 in Section V. The Penalty Total factor is a combination of voltage violation 289 penalty factor (Penalty Volt.violation ), power loss penalty fac-290 tor (Penalty Power loss ) and appliances' switching constraints 291 penalty factor (Penalty t switching ). These applied penalty factors 292 will be discussed in Sections II.D and IV. In the following, 293 the DR incentive rates and the associated contribution will be 294 calculated. Use DR participants based on their locations in the network 298 will have a higher influence on the network parameters (such 299 as bus voltages, line currents, and network loss) and are 300 tended to be interrupted more in a DR event than those 301 participants located comparatively less sensitive locations in 302 the same network [21], [22]. As a consequence, DR par-303 ticipants located in the buses with higher impacts on the 304 network parameters, contribute more to the voltage and loss 305 improvements than the other participants. If all the par-306 ticipating consumers are provided with an equal incentive 307 rate ($/kWh) (e.g., as considered in [12], [25], and [28]), 308 it implies a potential fairness issue on the incentive distri-309 bution between the participating consumers. To provide a 310 better balance between the contributions and the rewards to 311 the participating consumers, this study proposes a mechanism 312 of calculating incentive rate dynamically for each DR event, 313 which uses the location of the participating consumers in 314 the network, technical parameters of the network and the 315 time of the DR event. The calculated incentive rate of each 316 participant is mainly a combination of three components: 317 energy cost rate ($/kWh) based on the time of use (TOU) 318 tariff, voltage improvement cost and total loss improvement 319 cost, as shown in (2a). As seen in (2a), k t 1 , k t 2 and k t 3 are the 320 coefficient factors of energy cost, voltage improvement cost 321 and total power loss improvement cost, respectively, which 322 are optimized dynamically at the time of DR consumers' 323 participating in a DR event (discussed in Section IV). The 324 proposed algorithm will optimize these coefficients for each 325 DR bus based on the voltage and power loss sensitivities and 326 applied penalty factors for voltage and power loss violations 327 (as explained in Section IV) to determine the incentive rate 328 of each DR bus. The objective function in (1a) will try to 329 minimize the total cost by optimizing k t 1 , k t 2 and k t 3 values. 330 Equations (2b)-(2c) and (2d)-(2e) are used to calculate volt-331 age and total loss improvement factors of each selected DR 332 bus for rate design, respectively. bus, which is obtained from inverse Jacobian matrix [9]. The quently the total DR cost [9]. It is important to mention that 366 sensitivity analysis is conducted with respect to active power 367 changes from participating consumers, as many policies at 368 the moment recognize energy contribution of small-scale 369 consumers in DR programs [9]. Therefore, to identify the The proposed load control algorithm is implemented into two 385 DR schemes i.e., 10-minute and 2-hour schemes for handling 386 the short and long intervals of voltage variations in the MV 387 networks, respectively. Household appliances are categorized 388 based on their operation cycles to use in each DR scheme, 389 as discussed below. A maximum of 10-minute load control is considered in this 392 scheme for the short duration cloud movements. The candi-393 date appliances for this DR scheme are AC and EWH. These 394 devices can be interrupted for a maximum of 10-minute of 395 their control cycle to avoid consumer discomfort and rebound 396 effect of DR. Thus, they can be interrupted multiple times to 397 compensate for the fast variations of voltage in the network 398 due to the cloud transients. To minimize the consumer dis-399 comfort and DR rebound effect, once the load control signal 400 is sent to these devices, another signal will not be sent to these 401 devices for the next 10 minutes.

432
The consumption preferences of each DR appliance are 433 defined as consumption restriction and priority. During a DR 434 event, these appliances will not be switched ON or OFF.

435
In Table 1 inconveniences will be minimized.

467
To comply with the appliance control priorities and mini-468 mize the excessive switching disturbances on a consumer's 469 appliances, the average disturbance ratio (ADR) factor is 470 proposed in this paper. ADR represents the ratio of the total 471 demand change ( P) to the total number of disturbed appli-472 ances of i th consumer at t th time interval, as in: where P t ADR(i) is the sum of kW demand change, N A is the 476 total number of appliances of a consumer participating in a 477 DR event, P n,t i is the rated kW demand of the n th appliance. 478 All the parameters are for the i th consumer at t th time interval 479 during the DR event. ADR is treated as a technical constraint 480 during the optimisation process in the objective function (in 481 Section II.A). The associated penalty factor for ADR t i pre-482 sented in (5), is added into the objective function (1a). 483 According to (5), the penalty factor of ADR t i is maximum 486 when 1 greater than or equal to ADR t i to exclude the corre-487 sponding switching solution from the load control optimisa-488 tion search space. M is a large number (i.e. = 10 2 ) which is 489 implemented in order to eliminate the unappropriated switch-490 ing solutions. If ADR t i value is larger than or equal to 2 , the 491 penalty factor is zero to relax the ADR t i constraint. If ADR t i 492 is confined within 1 and 2 , a descending linear equation 493 is implemented to provide a linear relationship between the 494 value of ADR for each participant and the penalty factor. 495 To clearly understand the ADR t i constraint, let us consider a 496 consumer have a 2kW washing machine and 1kW pool pump, 497 which current switching status are OFF and ON, respectively. 498 During a DR event, if the optimization algorithm decides 499 to switch on the washing machine but keep the switching 500 status of pool pump unchanged (which is ON), according 501 to (4a) and (4b), the ADR value will be 2 (=2/1). However, 502 if the optimization algorithm decides to switch on the washing 503 machine and switch off the pool pump at the same time, the 504 ADR value will be 0.5 (|2−1|/2). Higher the ADR value, less 505 penalty factor will be added into the objective function in (1a). 506 The 1 and 2 values are user defined based on the required 507 optimisation output and maximum and minimum kW ranges 508 of the participated appliances. In this study we considered 1 509 and 2 values are 2 and 0.5, respectively. As a result, the ADR t i 510 constraint in (4a) helps the load control algorithm by selecting 511 the large (kW) available appliances of the DR participant to 512 be controlled in a DR event and thus, reduces the random 513 switching of the appliances. 514 Further, to limit an excessive amount of appliances control 515 for some participated consumers in a DR event, an additional 516 constraint called appliance fair interruption (AFI) 517 96364 VOLUME 10, 2022 is added into the optimization process. As stated before 518 (in Section I), some consumers are usually interrupted more 519 than others due to their sensitive locations in the network.

520
If this constraint is not included in the algorithm will create a 521 fairness issue regarding the number of appliance interruptions 522 in a DR event. To limit the excessive number of appliances 523 interruption for some location-based consumers, AFI will 524 be calculated and minimized by the proposed load control 525 algorithm in each DR event. AFI is defined using (6a) as the 526 total number of disturbed appliances for each consumer. The 527 associated penalty factor for AFI treated as a constraint in the 528 optimization, is shown in (6b).
where AiFI t i is AFI for i th consumer at t th time. wish. Consumers who signed for participation in any DR 564 scheme need to provide their lists of available DR appliances 565 to the utility. Consumers can predefine their consumption 566 preferences for each day through HEMSs. Furthermore, the 567 consumers have an opportunity to change their preferences 568 frequently before committing their participation in any DR 569 event. The information exchange with the consumers takes 570 less than a second using the technologies like WiMAX (with 571 a bit rate of 5 to 25Mbps) and ZigBee (with a bit rate of 572 250 kbps) [12]. Therefore, the total required time for step 2 is 573 less than 1 minute.

574
In step 3, participated consumers' information such as 575 appliances' current states, consumption preferences and pre-576 vious history of DR event participations are collected. Data 577 collection and processing in this stage are achieved within 578 2 minutes. The final step 4 optimizes the objective function to 579 calculate the optimum switching positions of the appliances, 580 DGs' reactive power contributions and fair dynamic incentive 581 rates to the participated consumers to calculate the total DR 582 cost. Once the optimum solution is obtained, the control 583 signals are sent to appliances and DGs' inverters to switch 584 ON/OFF, and provide reactive power support, respectively. 585 This stage is estimated to be performed within 2 minutes, 586 as the average computational time of the optimization process 587 for long-and short-interval schemes are around 50 and 20 sec-588 onds in the IEEE 33-bus system respectively using MATLAB 589 software on Intel CORE i7-2600 PC with a clock speed of 590   Therefore, an improved hybrid PSO (IHPSO) is proposed 646 in this study as a combination of MPSO and PS algorithms 647 to improve the optimisation performance and minimise the 648 computational time. Although the standard PSO combined 649 with PS is discussed in [39], we propose a hybrid model 650 of MPSO with PS in this paper with higher capabilities in 651 finding better solutions in a short period of time. In this hybrid 652 method, MPSO runs first to find the near global best location. 653 This global solution is provided to PS for further minimisa-654 tion of the objective function. The idea behind this strategy 655 is to let MPSO utilise its strength aggressively exploring 656 the search space to find near optimum solution, then let 657 PS utilises its strength to quickly find the global optimum 658 solution by searching locally around the solutions given by 659 MPSO. Fig. 2 shows the hybrid optimisation process with the 660 MPSO and PS algorithms in Step 4.

661
The velocity and position updates of each MPSO parti-662 cle at iteration k to search for the optimal solution are as 663 follows: where V k i and X k i are velocity and position vectors of ith 668 particle at iteration k, respectively; γ is the constriction factor 669 coefficient; P best i is the best value vector of ith particle so 670 far; G best is the best value among P best i so far; and rand is 671 a random number generator uniformly distributed between 672 0 and 1. The constriction factor coefficient (γ ) is calculated 673 as shown in (8b). ϕ max and ϕ are constant values. In this study 674 ϕ max = 4.05 and ϕ =1 are considered [13].
In (8b), ∈ [0,1] is a coefficient that allows control of explo-677 ration versus exploitation propensities. For a bigger value 678 of coefficient k, particles desire more exploration and pre-679 venting explosion, derives slow convergence and searching 680 thoroughly the space before collapsing into a point. However, 681 for smaller values, particles care more exploitation and less 682 exploration. The mutation function is applied when G best is 683 not improving while the increasing of number of iterations. 684 The mutation function selects a particle randomly and then 685 adds a random perturbation to a randomly selected element of 686 the velocity vector of that particle by a mutation probability. 687 In this paper, if the G best after 11 iterations is not improving, 688 the mutation function with the mutation probability of 0.85 is 689 applied.

690
MPSO handovers its global best (G best ) location to PS as 691 an initial point (X0), which has a great influence on PS's 692 calculation results. PS utilizes a set of directions comprising a 693 ''pattern'' that it uses to search around the initial point (X0) to 694 find better points and ignores the rest of the search space. This 695  Table 1). Therefore, the number of cells (variables) for  voltage management method. The proposed approach is 740 tested on an IEEE 33-bus radial distribution test system 741 shown in Fig. 4. In this paper, the IEEE 33-bus distribution 742 system is modified with three large solar PV-based DGs with 743 a capacity of 1.22 MW each connected to buses 15, 29, 744 and 31. The optimal locations of the DG units in the 33-bus 745 distribution network as shown in Fig. 4 are identified by the 746 study in [41]. To analyze the DG power output, 1-minute 747 interval power production data is gathered for a 1.22 MW 748 PV system located at the University of Queensland's St Lucia 749 campus in Brisbane [42], [43]. In this study, it is assumed that 750 a total 90 consumers are available to participate in each DR 751 event and randomly distributed in the DR candidate buses. 752 In this study, the permitted boundary of voltage magnitudes 753 for all network buses is considered within ±5% of nominal 754 voltage [9]. In addition, the amount of reactive power and 755 active power from each DG unit can be obtained by consid-756 ering the limits of power factor (PF i ) within [−0.95, +0.95]. 757 1 and 2 are 0.5 kW and 2 kW, respectively (as discussed 758 in Section II.D). The total number of PSO particles is consid-759 ered 300 and a self-adaptive iteration size technique is taken 760 into account. The mutation probability for MPSO is consid-761 ered 0.85. As discussed previously, the proposed dynamic 762 fair incentive mechanism implemented through large-scale 763 consumer participation is a new approach compared to the 764 previous methods [44], [45].   Fig. 9. To compensate for the voltage drops due 861 to intermittent DG power generations, four DR events are 862 initiated as below, in which DR events 1 to 3 are short-interval 863 and DR event 4 is a long-interval DR:

872
Due to the short-interval (<10 minutes) variation of the power 873 generation from the DGs, the first three DR events are acti-874 vated using the 10-minute DR scheme by controlling ACs 875   from the DGs. The owner of the DGs can maximize the use 911 of the DGs' capacity to freely produce maximum amount 912 of active power to the grid depending on the weather con-913 ditions. The AFI (appliance fair interruption) is violated for 914 DR events 1 and 3 for some DR participants. It is due to those 915 particular 916 DR participants are located in the sensitive buses in the 917 network and required to control more appliances that other 918 participants to satisfy the voltage limit constraints. Therefore, 919 a small penalty factor (60) is added into the total objective 920   Table 4 presents the number of appliances controlled during 927 all DR events for a total 90 participated consumers in both 928 Case 1 and Case 2. It shows the switching status (ON/OFF) 929 of the participated appliances before and after each DR event. 930 The DR events in Cases 1 and 2 reduce the consumers' load 931 demand to compensate for the active power drops from the 932 DGs due to the clouding. The proposed algorithm determines 933 the number of appliances to be switched off in each DR 934 candidate bus to keep the voltage within the ±5% limits 935 and minimize the total power loss. The algorithm optimizes 936 the switching of these appliances based on the constraints 937 applied by the participated consumers on their consump-938 tion preferences and appliance switching (as explained in 939 Section II.D). As shown in Table 4, in all the DR events of 940 Case 1, the washing machine and dishwasher loads are not 941 switched off when they are operating (ON condition), as their 942 control preference is considered 3 (based on the definition 943 in Table 1) during the optimization process they cannot be 944 switched off while they are operating to avoid resetting the 945 control cycles of these appliances. Since, the Case 1 has 946 the long interval (1-2 hours) voltage problems as similar to 947 DR event 4 in Case 2, the short usage of loads such as AC 948 and EWH are not controlled in those DR events. Table 5 949 represents the randomly selected participated consumers' 950 appliances switching status before and during DR event 1 in 951 both Case 1 and Case 2. It shows that consumer consumption 952 preferences are prioritized in the proposed during voltage 953 management in the network. Furthermore, the variability of 954 kW demand (or rating) of each appliance and their participa-955 tion availability in each DR event are considered in the opti-956 mization algorithm to provide a realistic DR implementation 957 approach. A very few studies are available in the literature, 958 which include this flexibility in the algorithm, as discussed 959 in Section I. In this section, the proposed dynamic DR incentive method 962 for voltage control is compared with three different 963 incentive methods. Fig. 12(a) shows different optimized 964 incentive rates simulated for DR event 1 in Case 1, which 965 are ''Only power loss improvement rate'', ''Only voltage 966 VOLUME 10, 2022 can't be switched OFF in the DR event. It is due to the 1008 particular appliance can't be interrupted while it is running 1009 (e.g., washing machine and dishwasher) to avoid hardware 1010 owner's personal needs (e.g., the owner may need to charge the EV for traveling). Therefore, the optimized switching 1013 position for those appliances will not be changed (switching 1014 status is 1). The above-mentioned appliance's switching pref-  Table 1. Therefore, the optimized switching    proposed IHPSO method the total objective function cost and 1064 the standard deviation are the lowest among all methods and 1065 with a slight increase in the optimization time compared to 1066 the PSO and MPSO methods. The optimization time in the 1067 PSO and MPSO is less because these methods don't have any 1068 hybridization approach with another optimization method, 1069 as a result, the OF result is very high as compared to the 1070 proposed IHPSO method. Therefore, the proposed IHPSO 1071 algorithm improves the OF performance and the accuracy of 1072 the voltage control algorithm. 1073 Fig. 13 depicts the obtained global best solution curve 1074 during 100 iterations from all optimization methods. In all 1075 these PSO methods when Pbest and Gbest stopped updating 1076 for a period of time, it means a local optimal solution is found, 1077 as seen in Fig. 13. It can be observed that the classic PSO 1078 method stopped updating the optimization result after 50 th 1079 iteration and MPSO method after 40 th iteration. MPSO shows 1080 better performance than the classical PSO method due to a 1081 mutation function as equivalent to GA algorithm is added 1082 into MPSO. The proposed IHPSO method stopped updating 1083 at around 70 th iteration with a global best value of 11.9, while 1084 HPSO stopped at 80 th with the value of 112.4. The IHPSO 1085 outperforms the HPSO method by reducing the optimization 1086 time and objective function value. In IHPSO method, the GA 1087 mutation function and PS algorithm help the particle to jump 1088 out of the local optimum and enhance the chance of finding 1089 the global optimal solution.   coordination approaches, which are DR with DGs' reactive 1118 power, DR with OLTC (on-load tap changer located between 1119 bus 1 and 2) [46], and DR combined with DG and OLTC. 1120 As seen, from the optimized voltage profiles in Fig. 15, 1121 it can be seen that the DR with OLTC coordination approach 1122 provides slightly better voltage improvement at the far end 1123 buses as compared to other coordination approaches. How-1124 ever, it increases the voltage significantly at the buses closer 1125 to the substation. On the other hand, using the DR with DG 1126 and OLTC coordination approach provides a smooth variation 1127 of the bus voltages across the network compared to all other 1128 DR coordination approaches. 1129 Table 7 shows the different optimized variables obtained 1130 from each DR coordinated voltage management approach. 1131 As explained above, the voltage management approach 1132 (VMA) using DR with DG and OLTC provides better volt-1133 age management across all buses in the network. However, 1134 the optimized objective function cost associated with this 1135 solution is quite high compared to DR with DG solution, 1136 as shown in Table 7. It is due to an additional cost in the objec-1137 tive function in (1a) is added for each tap change of the 1138 OLTC. The optimized cost using DR with DG approach is the 1139 lowest amongst the three voltage management approaches. 1140 Therefore, DR with DG approach can be a suitable and 1141 work conditions.