Abstract
As the penetration of renewable energy continues to increase, the demand-side resources in the grid will become more and more important. Electric vehicles (EVs) account for a relatively large proportion of demand-side resources, but individual and social factors have been less considered in multifactorial studies affecting EV participation in demand response (DR), and the multiscenario DR process has not been adequately studied. Therefore, an EV demand response strategy considering the influence of multiple factors is proposed in this paper. Firstly, a multisource charging load characteristic model is constructed by analyzing the characteristics of EV charging behavior under multiple scenarios. Secondly, the DEMATEL-AISM method is used to analyze the degree of influence of personal and social factors on users’ charging behavior under complex social environments, and the dominant factors in each scenario are identified. Finally, based on the analysis of the dominant factors in multiple scenarios, an EV regulation strategy under the influence of multiple factors is developed to achieve peak shaving. The feasibility of the proposed method is verified through simulation cases. The simulation results reveal that the revenues of aggregators and users are improved, and the stability of the power system is enhanced.
1. Introduction
With the increasing penetration of renewable energy [1], the uncertainty of power output leads to a decrease in power system stability [2]. It is increasingly important for the demand-side resource of the power grid to participate in DR regulation as a flexible resource. As a critical dispatchable resource, EVs have great potential to participate in DR [3]. In 2021, global EV sales reached 6,311,200 units, an increase of 119.01% year-on-year. In the same year, China’s EV sales exceeded 3.3 million units, Europe exceeded 2.3 million units, and the United States reached 530,000 units. As China is the world’s largest market for EVs, the rational and orderly usage of the huge demand-side resources can solve several critical problems including the high cost of power system equipment upgrading, stability, and reliability of the power system [4]. Since the energy-using behavior of demand-side users is influenced by personal preferences and social factors, these factors influence the operation, scheduling, and planning of the power system [5]. The power system becomes more diversified and complex. So, it is urgent to explore the dominant factors affecting users’ charging characteristics, build a multiscene and multitemporal user behavior preference model, and make full use of the users’ response regulation mechanism to improve the stability of the power system [6].
DR strategy is a typical approach to utilize electric vehicles as a flexible resource for grid operation optimizations [7]. In order to achieve the objectives of peak shaving [8], load fluctuation reduction [9], voltage excursion reduction [10], and cost reduction [11], DR strategies use a variety of approaches, such as price-based regulation [12, 13], day-ahead contract-based regulation [14], or incentive-based regulation [15]. The user has the right to decide whether the EVs participate in DR. Thus, the user has a significant impact on the potential of EVs as a flexible resource. The EV user factor must be considered in the DR strategy. It is necessary to develop a DR strategy that can increase user participation and motivate users to participate extensively in regulation. This will improve users’ willingness to participate in DR and attract more users to participate.
There is still lack of research on the factors that influence user behavior. When analyzing users’ charging behavior preferences, existing research generally clusters their charging behavior and EV charging clusters are completely divided. The variability of charging behavior can be directly obtained. But the multiple internal and external factors affecting charging behavior have not been studied in depth [16, 17]. Recent research uses intelligent algorithms such as machine learning and deep learning to analyze the uncertainty of the user’s charging behavior [18–20], and the demand response strategy is optimized. However, this method requires a large amount of data and is only aimed at the analysis of charging characteristics, and the discussion of the multiple factors influencing charging behavior is still insufficient. To address these issues, multifactor modelling using the DEMATEL-AISM algorithm is conducted to investigate the causes of the variability in charging behavior.
There is still lack of research on the factors that influence user behavior. The majority of the existing studies set the participation of user response behavior subjectively or qualitatively, considering the regret psychology and irrational choices of users when considering user charging preferences [21–23]. These studies lay little emphasis on the coupling characteristics of internal and external factors. When analyzing the factors that influence users’ charging behavior, only the relationship between factors and user behavior was considered [24], and the relationship between factors and interfactors was not analyzed. The dominant factor among many factors was not identified. The intricate interaction between multiple factors and user behavior was not quantitatively analyzed. The current EV optimal dispatching strategy is still insufficient for the study of users’ response willingness [25] and fails to make full use of the influencing factors to maximize users’ dispatching willingness. Although the coupling problem of road, vehicle, station, and grid is taken into account in the simulation scenario construction [26–28] and some literature has also combined the traffic network with user factors to analyze the coupling between internal and external factors [29], in general, less emphasis is laid on the influence of factors such as the user’s charging area and time and date on the participation in DR. To address these issues, an EV demand response strategy that considers the influence of multiple factors is proposed, and the possibility of increasing EV participation in DR is achieved by studying the coupling relationship between each influencing factor and EV participation in DR.
1.1. Contributions and Organization
In order to study the multiple factors influencing users’ charging behavior in complex social environments, we explore the interaction among the factors and search for the dominant factors influencing users’ charging behavior. A user behavior analysis method and a charging dispatching strategy considering the influence of multiple factors are proposed in this paper. The contributions of this paper are summarized as follows:(1)The dominant influencing factors of user behavior under different scenarios of temporal and spatial partition are quantitatively analyzed, the influence degree of multiple factors under complex individual and social environments is quantified, and the effectiveness of the dominant factor analysis method proposed in this paper is verified through simulation cases.(2)A synergistic analysis strategy of dominant factor-multiple factors is proposed, which effectively promotes the precise participation of EVs in DR service and improves the participation and regulation accuracy of EV dispatch.
This paper is organized as follows: a multisource charging load characteristic model is constructed in Part II. Multiple factors are evaluated by the DEMATEL-AISM method in Part III. The regulation strategy considering the users’ charging behavior influenced by multiple factors is presented in Part IV. The accuracy of the proposed method and the superiority of the DR strategy are verified in Part V. The conclusions and future research are summarized in Part VI. The overall process block diagram is shown in Figure 1.
2. Multisource Charging Load Characteristic Model
In this section, the operation and charging characteristics of EV charging stations, charging piles, and EVs under different scenarios are analyzed. The coupling relationships between the load of charging stations and the external environment, the users’ charging behavior, and the station operation characteristics are sorted out. On the basis of the established relationship, the multisource charging characteristic model is constructed, which serves as the theoretical foundation for the DEMATEL-AISM method.
2.1. EV Charging Station Operation Characteristics
As an aggregate of centralized EV charging, the location of the region, the number of installed piles, the size, and other macroscopic characteristics of EV charging stations fundamentally determine the upper limit of EV DR potential. The variety of EV charging behavior in various conditions can be macroscopically understood by analyzing the total operational dynamics of the charging load.
The process of EV charging station operation characteristic analysis is as follows: (1) the daily load operation data of three EV charging stations of the same size in residential, commercial, and industrial areas are calculated. The correlation between functional area and EV charging behavior is analyzed; (2) weekdays and holidays are separated, and the correlation between dates and EV charging behavior is analyzed; (3) the standard deviation, the peak-to-valley difference, and other indicators of daily load are calculated. The necessity of regulating EV charging behavior is analyzed.
The formula for EV charging station operational characterization is as follows:
The fluctuation of load was measured by standard deviation σ:where yt is the load value at moment t and is the average value of load in one day.
The level of peak-to-valley difference of the load curve is measured by the maximum peak-to-valley difference percentage Dpv%:where is the peak value of the load curve at moment t and is the valley value of the load curve at moment t.
2.2. EV Electric Pile Usage Characteristics
The charging piles in EV charging stations are the connection carriers between EVs and the grid. Their electricity usage can reflect the variability of EV charging behavior in the same area and the correlation between temporal attributes and EV charging behavior.
The process of EV charging pile usage characteristic analysis is broken down as follows: (1) the pile usage in different scenarios according to temporal and spatial partition principles is analyzed. The pile idle rate is calculated, and the degree of adaptation to EV charging demand is evaluated; (2) the number of EV visits of individual piles in different scenarios is calculated. The EV’s single-day charging behavior is analyzed.
The characterization of EV charging pile usage is calculated as follows:where EPir is the station charging pile idle rate; d is the number of days; h is the number of hours; p is the number of charging piles; and is the usage hour of the j-th charging pile in day i.where EP is the average number of daily EV visits of a single charging pile and Nij is the number of EV visits of the j-th charging pile in day i.
2.3. EV Charging Characteristics
EV charging characteristics mainly include EV charging start time, EV initial state of charge (SOC), EV charging capacity, EV charging duration, and EV charging power. Each characteristic reflects the charging behavior of a single EV comprehensively. Establishing the correct EV charging characteristic model can effectively comprehend the time and location information in the process of EV participation in DR and improve the regulation effect. Meanwhile, formulating corresponding DR strategies according to the individual characteristic can promote EV users’ participation in regulation and effectively explore the regulation potential.
The main objectives of EV charging characteristics analysis are as follows: (1) all the indices in charging characteristics are calculated. These indices are obtained by fitting the actual charging data to their probability distribution functions; (2) the differences of each index in temporal and spatial partition scenarios are analyzed so as to obtain the charging characteristics of EV in different scenarios. The analysis result can provide data support and logical basis for developing EV user DR strategies in Part IV.
3. DEMATEL-AISM-Based Multifactor Assessment Method
User behavior in the DR process is varied and complex. Consequently, it is difficult to evaluate the various factors that affect the user’s charging characteristics in different scenarios. Utilizing the system analysis techniques of graph theory and matrices, the DEMATEL algorithm is a solution to identify multifactor relationships for complex system problems [30]. It can identify the causal linkages between factors and the significance of each factor in the system. The Adversarial Interpretive Structure Modeling Method (AISM) incorporates the concept of game conflict based on the Interpretative Structural Modeling Method (ISM). AISM can perform a simple analysis of the coupling relationship between the system elements and obtain the simplest hierarchical topology diagram of the system.
The DEMATEL-AISM analysis method is selected as the analysis tool for this paper because it can effectively analyze the interactions between multiple factors affecting users’ charging behavior. The dominant, significant, and minor system factors are identified. The degree of influence among the factors is assessed precisely. The influencing factors analyzed in this paper include both internal and external multiple perspectives.
3.1. DEMATEL Model
First, each internal and external factor in the integrated system that affects users’ charging behavior is determined. The interrelationships among the factors are quantified by a fuzzification matrix. Since each user perceives different levels of internal and external influencing factors, after establishing the fuzzy influence range between the factors, the perceived levels of the respective internal and external factors of various users are simulated using the Monte Carlo method. As a result, the multifactor direct influence matrix M that influences each user’s charging is obtained. M is defined as follows:where M is a 10 × 10 direct influence matrix and the element mhk within the matrix is the degree of direct influence of factor on factor , mhh = mkk = 0, h, k = 1, 2, ..., n.
Second, the normative influence matrix N is obtained by normalization. The normative base value Maxvar is the maximum value of the row and column sum and defined as follows:
The matrix N is self-multiplied to represent the indirect influences among the elements, and all indirect influences are summed to obtain the integrated influence matrix T. T is defined as follows:where thk is the degree of integrated influence of factor on factor and I is the unit matrix.
The next step is to solve the influence degree, influenced degree, centrality degree, and cause degree of each factor of the system based on the matrix T.(1)Influence degree D: the integrated influence value of factor on all other elements. It is defined as follows:(2)Influenced degree C: the integrated influence value of factor by all other elements. It is defined as follows:(3)Centrality degree Cd: the central position and importance of factor in the integrated system affecting user charging, reflecting the degree of influence of the factor on charging behavior. It is defined as follows:(4)Reason degree Rd: the root reason affecting user charging, reflecting the degree of influence of factor on the system in affecting the user charging system. It is defined as follows:(5)Decision weights : the distance vector DC,R is calculated for each factor according to equation (12), and normalisation equation (13) is used to calculate the decision weights for each factor. The decision weight indicates the influence degree of the factor on the participation of EVs in DR. The higher the weight, the more the factor can influence the charging behavior of EVs, and the more effective of this factor in regulating the charging load of EVs.
3.2. AISM Model
The ISM topological hierarchy diagram is a UP topological hierarchy diagram that prioritises outcomes, while the AISM model builds on this with a DOWN topological hierarchy diagram that contrasts with it. This paired topological hierarchy diagram approach is able to explain the priority and causality of multiple factors in the system. Due to the inconsistency of the opposing topological hierarchy diagrams, the AISM model is used as an analytical model where the priorities of the elements in the system need to be analyzed.
Based on the matrix T , the intercept λ is obtained by normalizing the factor mean and the overall standard deviation σ2 of the system with the following equations:
Based on the intercept λ, the relation matrix A is calculated from the matrix T with the following rules:
The self-multiplication matrix B is calculated from the matrix A and the reachable matrix R is obtained by the concatenation method with the following equations:
According to the matrix R, two hierarchical extraction methods of result-first and cause-first are used to construct UP-type and DOWN-type hierarchical systems, and the factors affecting users’ charging behaviors in each hierarchical layer of the integrated system are determined. Then, the point reduction is performed by the matrix R. The loops that exist in the system that affect each other are grouped into one factor for analysis to obtain the point reduction reachable matrix R′; then, the duplicate edges in R′ are deleted for edge reduction. The skeleton matrix S′ can be calculated according to the following equation:
The matrix S′ is incremented to obtain the general skeleton matrix S. Finally, the corresponding values in the matrix T are substituted to obtain the matrix TS with influence values. Thus, the user charging behavior system with influence values is obtained.
The flow chart of the DEMATEL-AISM method is shown in Figure 2. Firstly, the direct influence matrix M is constructed based on the multiple factors affecting user charging behavior, and the integrated influence matrix T is obtained through normalization. Secondly, multiple evaluation indicators are obtained based on the matrix T, and the AISM model is introduced. Finally, the users’ charging behavior system containing the influence values is obtained, which provides data support for DR strategy development considering the influence of multiple factors in Part IV.
4. Regulating Strategy of Users’ Charging Behavior under the Influence of Multiple Factors
Based on the DEMATEL-AISM method, the coupling relationship among the factors in the integrated user charging system is analyzed. A DR strategy that integrates multiple factors and focuses on regulating the dominant factors is formulated. As a result, the coupling relationship in the integrated user charging system is fully utilized, the participation of users in DR is increased, and the load stability of the grid is improved.
4.1. EV Charging Strategy I: Disorderly Charging
The user’s vehicle usage time and driving habits are not affected by the peak and valley price. The user acknowledges the charging time required to fully charge the battery when they choose the start charging time. This disorderly charging strategy is named as Strategy I. The required charging time is calculated as follows:where is the charging time of the e-th EV, Se is the charging volume required for the e-th EV, Pe is the charging power of the e-th EV, and η is the charging efficiency, which is taken as 0.923.
4.2. EV Charging Strategy II: Price Incentive Charging
If the EV is charging in the valley price period, the time period between the grid entry and the valley ends is defined as Δt1. If the required charging time is less than Δt1, the user starts charging at any time point that the EV can be fully charged in the valley period. If the required charging time is longer than or equal to Δt1, the user starts charging at the grid entry.
If the EV is charging in the peak price period, the time period between the next nearest valley starts and ends is defined as Δt2. If the required charging time is less than Δt2, the user starts charging at any time point so that the EV can be fully charged in the valley period. If the required charging time is longer than or equal to Δt2, the user starts charging at the beginning of the valley period. The possibility of price-incentivized users participating in DR Ye is defined as 70% [31]; otherwise, the user chooses disorderly charging. The strategy is named as Strategy II. The charging start time for the four cases is as follows:where is the start time of the e-th EV charging, is the EV entry time, R1 is a random number between [0, 1], Tvs is the start time of the valley period, Tve is the end time of the valley period, is the valley price period, and Tp is the peak price period.
4.3. EV Charging Strategy III: Multifactor Influenced Charging
Based on the DEMATEL-AISM method proposed in Part III, the dominant factors affecting users’ charging behavior are obtained, and the influence degree of the remaining factors on the dominant factors is determined. The multiple factors with the highest influence degree on the dominant factors are selected, and their influence values are calculated. Consequently, the user’s charging willingness under different scenarios is obtained, which is taken as the possibility Ye of users being guided by the DR strategy. At the same time, the regulated users’ charging start time is correlated with Ye. The EV charging start time and the possibility of user participation in DR Ye are taken as the optimization variables. The strategy is named as Strategy III.where Ye is the possibility of the e-th EV to participate in DR, yn is the influence value of the n-th factor, ωn is the influence degree of the n-th factor, yelse is the influence value possessed by the else factors, and ωelse is the influence degree of else factors.
Since the possibility of user participation in DR by price incentive Ye in Strategy II is set as 70%, the base value of the possibility of user participation in DR under different scenarios in Strategy III is set as 0.7. The possibility of user participation in DR is adjusted up or down according to different scenarios. The specific possibility values of user participation in DR are summarized in Appendix A.
5. Simulation Cases
5.1. Load Characteristic Analysis
5.1.1. EV Charging Station Operation Characteristic Analysis
In simulation, three EV charging stations at the same scale in residential, commercial, and industrial regions are chosen. The daily load curves of these stations are estimated and generated using the actual charging load data in a metropolitan area of China from 2020 to 2021 [32]. The curves are shown in Figure 3.
According to Figure 3, the following information can be obtained: (1) the regional and temporal attributes are closely related to the load characteristics of EV stations. The charging load in residential and commercial areas has obvious peak and valley characteristics, and the Dpv% are 95.94% and 95.23%, respectively. The industrial area does not have obvious peak-valley characteristics, and the Dpv% is 33.02%, which is related to the implementation of the two-shift system in the industrial area; (2) the peak load in the residential area is around 17 : 00, and the valley is around 5 : 00; the peak load in the commercial area is around 20 : 00, and the valley is around 5 : 00. The peak and valley periods of both loads overlap with the peak and valley periods of the base load of the grid. The peak-valley difference of the grid is enlarged by EV integration; (3) date attributes have a close relationship with EV station load. The EV charging load on holidays is significantly higher than that on weekdays. The peak load on holidays in residential areas is increased by 47.6% compared with that on weekdays. The load fluctuation on holidays is also higher than that on weekdays. The σ on holidays in residential areas is increased by 69.05% compared with that on weekdays, so there is an urgent need to systematically regulate EV charging behavior for different dates and regions. In summary, the EV charging behavior has an obvious correlation with the area, date and time period.
5.1.2. EV Charging Pile Usage Characteristic Analysis
In this section, the usage characteristic of charging piles in the previously mentioned three EV stations is analyzed. Two indices are calculated including the idle rate of charging piles in the stations and the number of EVs connected to a single electric pile per day. The simulation results are shown in Table 1.
The following information can be obtained from Table 1: (1) EV users mostly charge in residential areas. EV charging behavior has regional characteristics; (2) within the same functional area, the number of EVs connected to charging stations is obviously different. EV charging behavior has date difference characteristics; (3) the proportion of EV charging at night is lower than that of daytime. EV charging behavior has time difference characteristics; (4) the EV idle rate is high, basically higher than 90%. So, there is great potential for EV charging regulation.
5.1.3. EV Charging Characteristics
In this section, the charging characteristics of EVs in the abovementioned three EV stations are analyzed. The EVs start charging time, and the amount of a single charge is counted. The results of three stations are shown in Figure 4.
The following information can be obtained from Figure 4: (1) the EVs start charging moment curve has the fitting trend of a normal distribution, combined with the normal distribution fitting curve proposed in the literature [31], the probability distribution function (PDF) of the EVs start charging moment in different regions is set, see Appendix B for details; (2) the peak period of EVs start charging moment in different regions has obvious distinction, the peak in residential area is 17 : 00, the peak in commercial area is 19 : 00, and the peak in industrial area is 12 : 00; (3) the number of EVs charged in a single day in different regions has a quantitative difference. The ratio of EVs charged in a single day in a residential area, commercial area, and industrial area is 4 : 2 : 4, and different EV numbers need to be set according to regions in the following simulation.
According to Figure 4, EV single charge capacity has the fitting trend of lognormal distribution, so the PDF of EV single charge capacity in different functional areas is set according to Appendix B.
Referring to the simulation result in Section 4.1 and literature [31], the following simulation scenarios are set up: (1) only private EVs are considered. The total number of charging vehicles per day is 1000. The ratio is 4 : 2 : 4 in residential, industrial, and commercial areas, and there is only one charging per day; (2) the number of charging piles in the charging station meets the vehicle charging demand; (3) the EV battery capacity is set as 80 kWh; and (4) a two-stage peak and valley electricity price is used. The peak price periods are [8, 23]. The peak and valley price table [33] is shown in Appendix C.
5.2. DEMATEL-AISM Analysis
The range of each matrix M element is shown in Appendix D. The influence factors of the user in the integrated influence system are calculated through Monte Carlo simulations. The calculated results of each factor in various simulations are analyzed to obtain the evaluation index of each factor in the integrated system, as shown in Table 2.
From the indices D and C in Table 2, the external factors, such as the function of area, time, and date, have a large influence on the rest of the factors. The external conditions are proved to be in an important position in the integrated influence system and have a large influence on rest of the factors. So, it is verified that the temporal and spatial partition method proposed in this paper is necessary.
Users’ internal factors are easily influenced by external conditions. The charging behavior of each user in different scenarios is determined by the users themselves. Users’ sensitivity to electricity prices and anxiety about electricity are closely related to the charging price and EV SOC, so adjusting the charging price can effectively guide users’ charging behavior. The indices Cd and Rd show that price sensitivity and power anxiety are the key factors affecting the user’s charging behavior, and the rest of the internal and external factors influence the user’s charging behavior effectively by affecting these two key factors. Therefore, it is necessary to develop a regulation strategy considering users’ price sensitivity and power anxiety to effectively regulate their charging behavior.
From Cd and Rd in Table 2, the weights of each factor in the integrated system affecting users’ charging behavior are obtained and shown in Table 3.
According to Table 3, in the integrated system affecting users’ charging behavior, users’ price sensitivity, users’ willingness to charge, and users’ functional area have the greatest influence on users’ charging behavior, reaching 16.94%, 14.87%, and 14.3%, respectively. Charging station scale and charging power have the least influence on users’ charging behavior, with only 2.95% and 2.34%. Since the functional area difference exists objectively and the users’ willingness to charge is related to their charging experience and charging habits, which can hardly be changed in a short period, the regulation mainly focuses on the users’ price sensitivity and power anxiety.
The UP-type and DOWN-type topological hierarchy with influence values can be obtained by combining the matrix R and the matrix TS as shown in Figure 5.
The UP-type hierarchical topology diagram is the first result. It infers the level of each influencing factor from the target factor. The DOWN-type hierarchical topology diagram is the first reason . It infers the level of each influencing factor from the underlying factor. Figure 5 is to show that the electricity price and the charging station scale are the active elements, so this integrated system affecting users’ charging behavior is topologically changeable. Charging power is an isolated element with little influence in the system.
Function of area, charging station scale, date, and time period are the causal factors, and price sensitivity and power anxiety are the result factors. Layer 1 and layer 2 are the intermediate factors, which indirectly affect the result factors in Figure 5. ω is the degree of influence of factor on factor , which indicates the degree of influence produced by changing factor on in Figure 5.
The influence degrees within the matrix T were calculated. Based on the sum of the integrated influence degrees of factor by other factors, the weight values of other factors influencing are calculated. The weight values are organized in Appendix E.
From the previous analysis, it can be concluded that the dominant factor affecting the characteristics of users’ charging behavior is the users’ perception of electricity price and power, so the degree of influence of each factor on these two factors is emphasized. According to Table E, date, time, functional area, and initial SOC have the largest influence on price sensitivity. These four account for 65.01%, indicating that users’ psychological decisions are closely related to the external environment. Among the factors affecting the power anxiety, the charging station scale factor accounts for 16.34%, with an influence degree of 0.1658, indicating that the scale of the charging station affects users’ charging decisions to a high degree. To relieve the power anxiety in a specific area, there is an urgent need to expand the charging station or build a new charging station to meet the charging demand of users.
Combined with Figure 5 and Table E, it can be seen that the influence weight of electricity price on user price sensitivity reaches 13.18% and the influence degree is 0.1872. The influence weight of power anxiety on price sensitivity is 9.89%, and the influence degree is 0.1346. The influence weight of initial SOC on power anxiety is 16.2%, and the influence degree is 0.1665. Due to the high weight of the electricity price factor on price sensitivity, it is easy to regulate charging behavior by price compensation. While EV initial SOC is related to user’s driving demand, user driving behavior is more difficult to be adjusted. So, the scheme of developing price compensation to promote EV participation in DR is proposed in the following section of this paper.
The dominant factor affecting the characteristics of users’ charging behavior is their own perceived degree of electricity price and SOC referring to the simulation results in Sections 5.1 and 5.2. Therefore, price compensation needs to consider the regional, date, and time differences. According to Figure 3, during holidays, residential areas are overloaded from 15 to 20 and commercial areas are overloaded from 18 to 22. Therefore, high price compensation should be adopted in the corresponding time periods; medium price compensation measures should be adopted in industrial areas from 8 to 16 to reduce the number of EVs that charge during daytime.
5.3. DR Strategy Analysis
According to the DR strategy considering the influence of multiple factors proposed in Part IV, the stability of EV daily charging load under different strategies is analyzed, and the economic benefits of aggregators and users under different strategies are also calculated to verify the effectiveness of the method proposed in this paper. The probability density functions of EV single charge capacity, EV start charge time, battery capacity, and charging power used in the simulation case are summarized in Appendix B.
5.3.1. Stability Analysis
EV start charging time is regulated based on the possibility of user participation in DR. The EV charging load under different DR strategies on holidays is calculated and superimposed with the basic load of the grid [34] to obtain the overall load. The overall load profile is shown in Figure 6.
The peak-to-valley difference Dpv and Dpv% was used to measure the peak-to-valley level of the load curve. The standard deviation was used as an indicator for the fluctuation of the load curve. The calculation results are shown in Table 4.
Figure 6 and Table 4 show that in the case of Strategy I, EV charging time is mostly from 17 to 21, resulting in the peak load rising from 13,951.6 kW to 17,575.3 kW and the Dpv% rising from 90.80% to 90.82%, so it is necessary to regulate the charging behavior of EV users. The latter three regulation strategies all effectively reduce the peak-to-valley difference level. The Dpv% of Strategy III is 66.96%, which is the lowest among three strategies without price compensation and is 3.68% lower than that of Strategy II. Dpv is 485.1 kW, which is also lower than that of Strategy II. After taking the price compensation mechanism into Strategy III, Dpv is 541.2 kW, which is lower than that of Strategy III. Dpv% is reduced by 7.59% compared with Strategy III. It is proven that adopting price compensation can better achieve the effect of peak shaving and valley filling.
According to Table 4, it can be seen that EV disorderly charging further aggravates the daily load fluctuation and the standard deviation increases by 32.7%. While adopting the DR strategy can effectively reduce the load fluctuation, the standard deviation decreases by 26.1% and 30.2%, respectively. Adding the price compensation mechanism based on Strategy III can better suppress the extreme conditions, such as rapid increase or decrease of load in a short period, the standard deviation decreases by 4.13% compared with Strategy III. It reduces the fluctuation of the grid load within one day by 4.13%. Price peaks and valleys are separately discussed, and load optimization during price peaks and valleys with different DR strategies is analyzed. Dpv% and standard deviation are calculated, and the results are shown in Table 5.
According to Table 5, it can be seen that after the DR strategy is adopted, the load in the valley hours is obviously optimized; Dpv% of Strategy III is reduced by 50.51% compared with disorderly charging. After the price compensation mechanism is adopted, the load in the valley hours is further optimized and Dpv% is further reduced by 15.39% compared with Strategy III. The grid base load is divided into regions in the ratio of 2 : 1 : 2 according to Section 5.1, and the load optimization of each region under different strategies is analyzed. The results are shown in Table 6.
According to Table 6, the peak shaving objectives of Strategy III are effectively achieved in both residential and commercial areas, with an average decrease of 13.4% compared to Dpv% of Strategy II. The best regulation effect in residential areas indicates that the possibility of participating in DR in this area is high. The effect of Strategy III in commercial areas is slightly worse than that of Strategy II, which may be caused by the low number of charging EVs in this area and the low enthusiasm of customers to participate in DR.
The follow-up content further explores the DR potential in the residential area. After adding the price compensation mechanism, the proportion of DR in the residential area and industrial area increases, generating new load peaks at night, resulting in Dpv expanding by 33.32% and 4.84%, respectively, compared to Strategy III. On the other hand, the load fluctuation in the commercial area decreases. So, the compensation price in the residential area needs to be reduced to prevent the generation of new load peaks in this area.
5.3.2. Economic Analysis
The cost of power purchase by aggregators and the cost of charging by users are calculated, and the results are shown in Table 7.
According to Table 7, it can be seen that the cost of aggregators and users is the highest in the case of disorderly charging. While the cost of aggregators and users is effectively reduced after adopting the DR strategy. The cost of electricity purchase for aggregators in Strategy III is reduced by 44.64% compared with Strategy I and 2.73% compared with Strategy II. The cost of charging for users is reduced by 33.38% compared with Strategy I and 2.35% compared with Strategy I. After adding the price compensation mechanism, the cost of power purchase was further reduced by 730.4 CNY and 895.2 CNY, respectively, compared with Strategy III.
The cost of power purchase and charging for different regions is shown in Table 8.
According to Table 8, the following conclusions can be obtained: (1) the cost of aggregators and users under Strategy III is effectively reduced in residential areas, which is 55.56% and 44.83% lower than Strategy I, respectively. (2) The cost reduction under Strategy III performs better in residential areas than in industrial and commercial areas. So, the best effect of increasing DR incentives is achieved in residential areas.
A load peak-valley difference incentive mechanism is added to evaluate the total profit obtained by the aggregator after adopting the DR strategy. The maximum and minimum values of the system load within each hour th are given by the following equation:
The incentive profit of peak-to-valley difference is as follows:where Fu is the base value of incentive profit of peak-to-valley difference. It is set as 20,000 CNY; u is the unit penalty cost of peak-to-valley difference. It is set as 0.5 CNY/kW [31].
The simulation results are shown in Table 9. According to Table 9, without the price compensation mechanism, the total profit of Strategy I is the lowest among the first three strategies because of the high peak-to-valley difference penalty cost due to the increased load fluctuation caused by disorderly charging. Strategy II and Strategy III effectively reduce the peak-to-valley difference penalty cost. Strategy III further achieves peak and valley reduction, so the total profit is the highest among the first three strategies. Total profit increases by 1022.4 CNY and 1649 CNY, respectively, compared with the other two strategies. After adding the price compensation mechanism, the aggregator gets the highest load peak-valley difference incentive because the load fluctuation is smoothed out best. However, the compensation price needs to be paid to the EV users participating in DR, and the total profit decreases by 13.02% compared with Strategy III. If the unit penalty cost of load peak-to-valley difference is increased, aggregators will be more inclined to participate in regulation to gain more profit due to load optimization.
In summary, under the temporal and spatial partition scenario, multiple factors affecting users’ charging behavior are regulated collaboratively and the possibility of EV participation in DR is increased. Based on effective reduction of peak-to-valley load difference and smoothing of daily load fluctuation, the profit gained from aggregators’ participation in regulation is increased, while the charging cost of users is also reduced.
6. Conclusion
The multisource charging load characteristic model is developed in this paper. Each internal and external factor affecting user charging behavior is analyzed by the DEMATEL-AISM method. The dominant factor-multiple factors regulation strategy is proposed based on the analysis results, and the effectiveness of the proposed method is verified by simulation cases. Based on the simulation results, the following conclusions are obtained:(1)The dominant factors influencing users’ charging behavior are accurately analyzed using the DEMATEL-AISM method. The multiple factors influencing the degree in the complex social environment are calculated, and the coupling relationships between the influencing factors are quantified. Based on the simulation results, the participation of EVs in DR can be effectively increased by regulating the dominant factor.(2)The multifactor analysis based on the DEMATEL-AISM algorithm reveals that the dominant factor influencing EV charging is the user’s perception of price and SOC, so EV participation in DR can be effectively motivated by the price subsidy policy and charging station construction.(3)According to the simulation results, through the multifactor regulation strategy, the load fluctuation of the grid is reduced, the peak-to-valley difference is reduced, and the profits of aggregators are improved, proving that the dominant factor regulation method is more application-oriented.
On the basis of the research stated previously, the multiple factors affecting the user charging behavior will be further studied. The impact of different DR strategies on the voltage fluctuation, frequency fluctuations, and congestion of the grid will be studied.
Symbols and Abbreviations
| EV: | Electric vehicles |
| DR: | Demand response |
| DEMATEL-AISM: | Decision-making trial and evaluation laboratory-adversarial interpretive structure modeling |
| SOC: | State of charge |
| PDF: | Probability distribution function |
| : | The standard deviation of load fluctuation |
| : | The moment of load |
| : | The load value at moment t |
| : | The average value of load in one day |
| : | The maximum peak-to-valley difference percentage |
| : | The peak value of load curve at moment t |
| : | The valley value of load curve at moment t |
| : | The EV charging pile idle rate |
| : | The number of days |
| : | The number of hours |
| : | The number of charging piles |
| : | The usage hour of the j-th charging pile in day i |
| : | The average number of daily EV visits of a single charging pile |
| : | The number of EV visits of the j-th charging pile in day i |
| : | The element of matrix M with h-th row and k-th column |
| , : | The influence factor of user charging behavior |
| : | The element of matrix T with h-th row and k-th column |
| : | The intercept |
| : | The normalizing factor |
| : | The overall standard deviation of the system |
| : | The element of matrix A with h-th row and k-th column |
| M: | The direct influence matrix |
| N: | The normative influence matrix |
| T: | The integrated influence matrix |
| D, C: | The influence degree |
| Cd: | The centrality degree |
| Rd: | The reason degree |
| A: | The relation matrix |
| B: | The self-multiplication matrix |
| R: | The reachable matrix |
| R′: | The point reduction reachable matrix |
| S′: | The skeleton matrix |
| S: | The general skeleton matrix |
| I: | The unit matrix |
| : | The charging time of the e-th EV |
| : | The charging capacity required for the e-th EV |
| : | The charging power of the e-th EV |
| : | Is the charging efficiency |
| Δt1: | The time period between the EV entering grid and the valley ends |
| Δt2: | The time period between the next nearest valley starts and ends |
| : | The possibility of price-incentivized users participating in DR |
| : | The start time of the e-th EV charging |
| : | The e-th EV entry time |
| : | A random number between [0, 1] |
| : | The start time of the valley period |
| : | The end time of the valley period |
| : | The valley price period |
| : | The peak price period |
| : | The influence value of the n-th factor |
| : | The influence degree of the n-th factor |
| : | The influence value possessed by the else factors |
| : | The influence degree of else factors |
| : | The peak-to-valley difference |
| : | The hour in one day |
| : | The maximum values of the system load in the th period |
| : | The minimum values of the system load in the th period |
| : | The values of the system load in the th period |
| : | The base value of incentive profit of peak-to-valley difference |
| u: | The unit penalty cost of peak-to-valley difference |
| F: | The incentive profit of peak-to-valley difference. |
Data Availability
The data used to support the findings of this study are available from the corresponding author upon request.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Acknowledgments
This work was supported in part by the project (51977127) of National Natural Science Foundation of China.
Supplementary Materials
The supplementary materials contain the following: the degree of each factor influencing EV participation in demand response, the probability distribution function of the charging start moment and the single charge capacity of EVs in different functional areas, the probability distribution function of the charging power and battery capacity of EVs, the peak and valley price table, the range of the elements within the initial fuzzy matrix, and the coupling weight values between the factors. (Supplementary Materials)