Data-Based Orderly Charging Strategy Considering Users’ Charging Choices

Abstract

This work proposes a centralized data-based orderly charging strategy that considers the user’s charging choices. Three charging choices for different types of users are described. Then, a scheduling model of electric vehicles based on the time dimension is established. In this strategy, the optimization model not only considers the demand of the grid side and the user side, but also takes the driving data of electric vehicles as the driver. The grid-side optimization involves minimizing the equivalent load fluctuation, and the user-side is optimized to minimize the charging cost and maximize the charging electric quantity. The scheduling capabilities of the three charging strategies are analyzed based on a series of driving data of electric vehicles. The results show that the peak-valley difference and equivalent load fluctuation of the power grid in the data-based orderly charging strategy reduced by 22.2% and 22.7%, respectively, and the charging cost of users also reduced much more than the other two charging strategies. Additionally, the effect of users’ charging choices on the charging strategy is analyzed, and the results show that the orderly charging strategy that considers users’ charging choices can effectively decrease the scheduling deviation caused by users’ charging choices. It greatly improves the security and economy of the grid.

1. Introduction

The increase in car ownership means that the demand for oil is rising constantly, and the conflict between this demand and the decrease in fossil energy reserves is increasingly intensified [1,2,3,4,5]. On the other hand, road transport accounts for nearly one fifth of total emissions of greenhouse gas carbon dioxide, which increased by about 23 percent between 1990 and 2010 [6]. In response to the call for energy conservation and emission reduction, many countries are incentivizing car manufacturers to develop electric vehicles (EVs) more vigorously [7]. At the “New Revolution of the Automotive Industry” sub-forum of the China Development Forum 2021, Yongwei Zhang, Vice chairman and Secretary-General of the China Electric Vehicle 100 Committee, said that according to predictions, before and after 2030, the ownership of electric vehicles in China should be at the scale of 80 million.

In addition, in terms of electric vehicle charging, almost all electric vehicles start charging as soon as they are connected to the power grid. However, the charging characteristics of EVs are uncertain and random, which is bound to have a serious impact on the safety, stability and economy of the power grid when large-scale EVs are randomly charged [8,9,10,11,12]. Therefore, it is necessary to provide an appropriate and effective charging strategy that will allow EV users to conduct orderly charging.

The present literature on orderly charging strategies for EVs are mainly divided into two scheduling models: static scheduling and dynamic sensing scheduling.

The static scheduling strategy does not consider any dynamic factors of electric vehicles. Erol-Kantarci et al. [13] proposed a charging algorithm based on the prediction of electricity prices during charging. In their work, the real-time price information of charging stations is obtained to find the appropriate charging time at a low cost. During the charging process, the k-nearest neighbor classification algorithm is used to predict the price signal of the power grid. If the predicted price is greater than the threshold, the charging is delayed until a suitable charging time has been determined. Sortomme et al. [14] studied the influences of plug-in hybrid vehicles on distribution network overload, efficiency, power quality, voltage regulation and other performance, and proposed a collaborative charging solution for plug-in hybrid vehicles. Wu et al. [15] proposed centralized charging and load scheduling algorithms, which can be used by the aggregation to minimize the power cost. The algorithm uses a forecast of charging demand for the next day to determine each EV’s charging plan.

The dynamic sensing scheduling model considers some dynamic aspects of electric vehicles when forming charging plans. Korolko et al. [16] discussed the optimal issue of the charging of electric vehicles in an unregulated electricity market. In their work, an efficient and effective cutting plane algorithm for both uncertain and known electricity prices was developed to solve the charging optimization problem. Tao et al. [17] established an EV scheduling model on account of the electric vehicles driving data; the model considered the interests of both the user-side and the grid-side. However, the charging situation considered in this work is too idealized, and the influence of different electric vehicle users’ economic attributes is ignored.

Static scheduling [13,14,15] ignores the dynamic features of EV changing with time, and the established scheduling model is quite different from the actual situation. And in full accordance with the threshold charging or not, there is a good chance that EVs will not be able to obtain as much charge as they need. Distributed scheduling in some studies [18,19,20] can allow the penetration of a great many EVs in the programming process, but plenty of different types of EVs will be connected to the grid in the future. In order to achieve the optimal charging plan, as much EVs information needs to be collected as possible. In terms of the optimization model, existing research mostly concentrates on the economy or efficiency of the grid-side [21] or on the cost reduction of the user-side [22]. There are relatively few models that consider both grid-side and user-side benefits in the literature [23]. However, it is assumed that the user is completely controllable, ignoring the preferences and particularities of electric vehicle users, so the actual situation cannot be described very accurately.

Therefore, an orderly charging strategy based on driving data of EVs and considering users’ charging choices is proposed in this work. The proposal of this strategy is based on a large number of collected real-world driving data of EVs. This strategy not only considers the interests of the grid-side and the user-side, but also considers the user’s charging preference. And the strategy restrains the state of charge (SOC) when the EV leaves the grid, so that the user’s next trip will not be influenced. Moreover, this strategy relies on the driving data of EVs to constrain the time the vehicles take to connect to the power grid, which can be closer to the intention and actual situation of electric vehicle users.

Major contributions involve many aspects, as follows: (1) The ordered charging strategy is formulated with the consideration of users’ charging choices. (2) Based on the driving data, an EV scheduling model is set up to satisfy the needs of the grid-side and the user-side. (3) The effectiveness of the data-based orderly charging strategy and the effect of the user’s charging choice on the charging strategy are analyzed. The rest of this article is constituted as below: Section 2 represents the data sources used. Section 3 introduces the method considering the user’s charging choice. Section 4 establishes the mathematical model of the orderly charging strategy. Section 5 simulates the established orderly charging strategy through a case. Section 6 discusses and analyzes the simulation results from multiple perspectives, and Section 7 concludes the paper.

2. Data Description

The data source of the study is the battery electric vehicle driving data collected by the operation management system platform mentioned in the previous article [17]. The data are the same set of data used in reference [17], and the study in this manuscript is an improvement on the basis of reference [17]. The driving data of electric vehicles used in this case contains the driving information of 1000 electric vehicles in a whole year, including 743,712 vehicle trips and a 3,542,209 km driving distance.

Due to the huge amount of data collected by this platform, in order to reduce the amount and time of calculation, only some data items related to this work are extracted from this massive dataset, and irrelevant redundant data items are eliminated. Specifically, only the start and end times of each trip for each electric vehicle are extracted. Data items include the vehicle number, start time, end time, driving distance, power grid connection status, battery charging status and state of charge (SOC) (see

Table 1. Description of data items required for this work.

Data Item	Detailed Description
Vehicle number	The unique number of the vehicle
Trip number	Number each trip individually in order
Start time	The time when the trip begin
End time	The time when the trip end
Driving distance	Driving distance of
Power grid connection status	1—Connected	0—Not connected
Battery charging status	1—Charging	0—Uncharged
SOC	Residual SOC value of the vehicle

). It should be noted in particular that there is a difference between the power grid connection state and the charging state of electric vehicles: electric vehicles may not be charged when they are connected to the grid. In other words, when the power grid connection state of electric vehicles is one, the charging state can be zero or one.

After the screening and pre-processing of the original data of electric vehicles, the data needed for this work have been simplified, but we still have a huge amount of data. At present, electric vehicles can only be charged when they are stationary, so the charging operation of electric vehicles by users is more likely to happen after the end of a trip rather than in the middle of a trip. Therefore, it is necessary to set a threshold to separate each trip, and the two trips with a stopping duration below this threshold will be regarded as the same trip. This work studies the orderly charging strategy of electric vehicles under the condition of slow charging. According to the current power battery charging technology of electric vehicles and parking charging policies, when the parking time is less than 30 min, most EV users will not choose slow charging to recharge their EV. Therefore, 30 min of parking time is selected as the threshold to distinguish each trip. If the parking time of an EV is less than this threshold, it will be considered continuous driving, and the data content of this period will be combined into the same trip. This can greatly reduce the amount of data.

3. Charging Strategy Considering User’s Charging Choice

For the same scenario, such as the same remaining power, the same parking time, etc., users with different economic preferences may make different charging choices, including whether to charge, deciding between fast charge or slow charge, and choosing whether to obey the grid scheduling. The orderly charging strategy of EVs studied in this work considers the user’s choice. Different types of EV users select different charging options, which are crucial to determining the optimal solution for EV charging plans [24]. This section describes the charging choices for different types of users based on the literature [24].

3.1. Range of Charging Choice for Electric Vehicle Users

The charging choice of users with different economic attributes can be reflected in the vehicle SOC, parking time and duration, electricity price information, and other data. Thus, the charging choice of users can be predicted through the collected driving data. EV users can choose to “obey” or “not obey” the grid scheduling before charging. Since the study in this work is an orderly charging strategy for the slow charging mode, the fast charging mode does not obey the grid scheduling by default. Electric vehicle users can choose between three charging options: (1) fast charging (FC); (2) slow charging that does not obey power grid scheduling (SC); (3) slow charging that obeys grid scheduling (OSC). Users choose the charging mode according to the parking duration of electric vehicles, and the function of the charging choice range is as follows:

$$F\left(D_c^{dep}\right)=\left\{\begin{array}{c}\alpha \in \left\{FC\right\} D_c^{dep}\le D_c^f\\ \beta \in \left\{FC,SC\right\} D_c^f<D_c^{dep}\le D_c^s\\ \gamma \in \left\{FC,SC,OSC\right\} D_c^{dep}>D_c^s\end{array}\right.\hspace{1em}\forall c\in C$$

(1)

where $D_c^{dep}$ is the charging duration of electric vehicle c; $D_c^f$ is the required duration for electric vehicle c to reach the desired SOC through fast charging; $D_c^s$ is the required duration for electric vehicle c to reach the desired SOC through slow charging; $\alpha$, $\beta$ and $\gamma$ are the user’s choice ranges in each case.

The duration required for EV c to reach the expected SOC through fast or slow charging depends on the remaining electric power of the vehicle’s battery and charging power, as shown below

$$D_c^f=\frac{\left({SOC}_c^{set}-{SOC}_c\right)\times E_c^{cap}}{P^f} \forall c\in C$$

(2)

$$D_c^s=\frac{\left({SOC}_c^{set}-{SOC}_c\right)\times E_c^{cap}}{P^s} \forall c\in C$$

(3)

where ${SOC}_c^{set}$ is the expected SOC value when leaving the power grid set by the user of EV c. ${SOC}_c$ is the current SOC value of EV c.$E_c^{cap}$ is the rated battery capacity of EV c. $P^f$ and $P^s$ represents the power of fast and slow charging, respectively.

3.2. Charging Mode Selection Based on Economic Preference

Behavioral economics theory believes that the user’s choice should be based on the actual decision-making process. In the same case, the choice of action depends on how the individual weighs profit versus risk. Therefore, charging mode selection of users is essentially a risk decision [25]. Different charging decisions will be made by EV users based on charging demand, charging cost and risk appetite.

This work divides users into three categories: radical, conservative, and balanced. The selection functions of three types of users are described by means of the method proposed in reference [24].

3.2.1. Radical Users

Radical users will seek the lowest cost. The charging selection function of radical users can be expressed as:

$$\beta =\left\{\begin{array}{c}SC C^{SC}<C^{FC}\\ FC else\end{array}\right.$$

(4)

$$\gamma =\left\{\begin{array}{c}FC C^{FC}<min\left\{C^{SC},\mu C_{min}^{OSC}\right\}\\ SC C^{SC}<min\left\{C^{FC},\mu C_{min}^{OSC}\right\}\\ OSC \mu C_{min}^{OSC}<min\left\{C^{FC},C^{SC}\right\}\end{array}\right.$$

(5)

$$\mu =\frac{C_{min}^{OSC}+\left(C_{avg}^{OSC}-C_{min}^{OSC}\right)\times {\mu }_{OSC}}{C_{min}^{OSC}}$$

(6)

where, $C^{FC}$ and $C^{SC}$ are the charging costs of fast charging and slow charging that do not obey scheduling, respectively. $C_{min}^{OSC}$ is the minimum cost of obeying scheduling slow charging. $\mu$ is the coefficient representing the degree of radicalization. ${\mu }_{OSC}$ is a parameter between 0 and 1.

Formula (4) indicates that when the charging choice range of EV users is $\beta$, there are two charging modes: FC and SC. Radical users will choose the SC if the charging cost of SC is less than that of FC; otherwise, the FC will be selected. The other selection functions are similar to Formula (4).

3.2.2. Conservative Users

Conservative users will seek steady returns. They will choose to obey the grid scheduling if the maximum cost of obedience is less than the cost of disobedience. The charging selection function of conservative users can be expressed as:

$$\gamma =\left\{\begin{array}{c}FC C^{FC}<min\left\{C^{SC},C_{max}^{OSC}\right\}\\ SC C^{SC}<min\left\{C^{FC},C_{max}^{OSC}\right\}\\ OSC C_{max}^{OSC}<min\left\{C^{FC},C^{SC}\right\}\end{array}\right.$$

(7)

where $C_{max}^{OSC}$ is the maximum cost of obeying the slow charging schedule.

3.2.3. Balanced Users

Balanced users are somewhere in between the above two types of users. The charging selection function of balancing users can be expressed as:

$$\gamma =\left\{\begin{array}{c}FC C^{FC}<min\left\{C^{SC},C_{avg}^{OSC}\right\}\\ SC C^{SC}<min\left\{C^{FC},C_{avg}^{OSC}\right\}\\ OSC C_{avg}^{OSC}<min\left\{C^{FC},C^{SC}\right\}\end{array}\right.$$

(8)

where $C_{avg}^{OSC}$ is the average cost of obeying the slow charging schedule.

4. Mathematical Formulations

The mathematical formulations of the orderly charging strategy of EVs are established in this section. At the present stage, under the condition that the power grid only charges EVs, the more concentrated the charging time of EVs, the more power the EVs need, and the greater the impact on the power grid. Therefore, a reasonable charging strategy should be designed to meet the charging demands of large-scale electric vehicles.

The ordered charging strategy optimization model described in this section is a scheduling model of electric vehicles. It can better meet the needs of the power grid-side and user-side by dynamically adjusting electric vehicles, whether they recharge or not, in each time period. The optimization object is a multi-objective optimization problem with multiple indexes and constraints.

4.1. Optimizing Target

The optimization objective of this study is to consider the benefits of both the power grid side and the user side. However, the optimization of the grid side goal will inevitably limit the user-side charging quantity, so that the two aspects of the goal cannot achieve the optimal result at the same time. Trade-offs and compromises have to be made between each goal, so that all the objective functions come as close as possible to achieving the optimal result. In this work, the linear weighting method is adopted to simplify the multi-objective optimization problem into a single-objective optimization problem. In addition, due to the large difference in the size of the calculated values of three objectives, standardized processing is needed. The expression is as follows:

$$F_{min}={\omega }_1{F}_1^s+{\omega }_2{F}_2^s+{\omega }_3{F}_3^s$$

(9)

$$F_1^s=\frac{F_1-min{F}_1}{max{F}_1-min{F}_1}$$

(10)

$$F_2^s=\frac{F_2-{minF}_2}{max{F}_2-min{F}_2}$$

(11)

$$F_3^s=\frac{max{F}_3-F_3}{max{F}_3-min{F}_3}$$

(12)

where ${\omega }_1$, ${\omega }_2$ and ${\omega }_3$ are weight factors. $F_1^s$, $F_2^s$ and $F_3^s$ are the standardized indicators of the target functions $F_1$, $F_2$ and $F_3$, respectively.

In this work, it is assumed that the power grid-side and the user-side are equally important to the optimization goal, so the values of the weight factors are 1/3. $F_1$, $F_2$ and $F_3$ represent the optimization targets of the power grid-side and the user-side, respectively.

4.1.1. Grid-Side

Smoothing the load fluctuation of the regional power grid as the optimization goal of an orderly charging strategy for electric vehicles can save on power grid operation and investment costs and improve the utilization rate of equipment.

This work studies the charging schedule within 1 day and divides that day into T periods. The grid-side objective function is:

$$F_{1min}=\sqrt{\frac{1}{T}{\sum }_{t\in T}{\left[P_s\left(t\right)-P_{avg}\right]}^2}$$

(13)

where $P_s\left(t\right)$ is the equivalent load power of the grid system at time t. $P_{avg}$ is the average daily load power of the grid system.

$$P_{avg}=\frac{1}{T}{\sum }_{t\in T}{P}_s\left(t\right)$$

(14)

$$P_s\left(t\right)=P_{load}\left(t\right)+P_{ev}\left(t\right) \forall t\in T$$

(15)

$$P_{ev}\left(t\right)={\sum }_{c\in C}\left[x_c\left(t\right)\times P_c^{ch}\right] \forall t\in T$$

(16)

where $P_{load}\left(t\right)$ is based on load power with no electric vehicle loads. $P_{ev}\left(t\right)$ is the charging power of EVs. $x_c\left(t\right)$ are used as decision variables; they determine the charging state of vehicle c during time period t. $P_c^{ch}$ is the charging power of electric vehicle c.

4.1.2. User-Side

The main interests of EV users in charging are the cost and the charging quantity. Thus, minimizing the charging cost of electric vehicle users is one of the optimization objectives, and another goal is to maximize the charging quantity of electric vehicles. Suppose c refers to the vehicle label and k refers to the trip label, then the objective functions are

$$F_{2min}={\sum }_{c\in C}{\sum }_{t\in T}{H}_c\left(t\right)$$

(17)

$$F_{3max}={\sum }_{c\in C}{\sum }_{k\in K}{E}_{c,k}$$

(18)

where $H_c\left(t\right)$ is the charging cost of EV c in time period t. $E_{c,k}$ represents the additional mileage (km) gained from charging after trip k of vehicle c.

The electricity cost of EV users is related to the charging quantity and the unit price of electricity purchased in time period t. In addition, considering the combination of orderly charging and electricity price, this work assumes that the electricity price incurred when obeying power grid scheduling is different from that incurred when not obeying power grid scheduling.

$$H_c\left(t\right)={price}_{ch}\left(t\right)\times P_c^{ch}\times ∆t\times x_c\left(t\right) \forall c\in C, \forall t\in T$$

(19)

$${price}_{ch}\left(t\right)=m_1{price}_{FC}\left(t\right)+m_2{price}_{SC}\left(t\right)+m_3{price}_{OSC}\left(t\right)$$

(20)

where ${price}_{ch}\left(t\right)$ is the electricity price during time period t. ${price}_{FC}\left(t\right)$ is the fast charging price during time period t. ${price}_{SC}\left(t\right)$ is the slow charging price that does not obey power grid scheduling during time period t. ${price}_{OSC}\left(t\right)$ is the slow charging price that obeys power grid scheduling during time period t. $m_1$, $m_2$ and $m_3$ are binary parameters.

The additional mileage depends on the charging quantity and the average driving power consumption of the vehicle. The formula is as follows:

$$E_{c,k}=\frac{24}{T}\times \frac{P_c^{ch}}{r_c}\times {\sum }_{t={eh}_{c,k}}^{{sh}_{c,k+1}}{x}_c\left(t\right) \forall c\in C, \forall k\in K, \forall t\in T$$

(21)

where $r_c$ is the average power consumption of vehicle c in normal driving mode (kWh/km). ${eh}_{c,k}$ is the time period when vehicle c finishes trip k. ${sh}_{c,k+1}$ is the time period when vehicle c starts trip k + 1.

$$x_c\left(t\right),m_1,m_2,m_3\in \left\{\mathrm{0,1}\right\}\hspace{1em}\forall c\in C, \forall t\in T$$

(22)

Equation (22) indicates that $x_c\left(t\right)$ is a binary variable (0 or 1).

4.2. Constraint

The proposed optimization model also takes into account the constraints of battery capacity, the state of the vehicle, and the battery SOC requirements of the user after recharging.

Limited by the maximum capacity of the power battery, the sum of the recharged electric quantity during the charging period and the remaining electric quantity of the battery before charging cannot exceed the maximum battery capacity.

$$E_{c,k}\le R_c-R_{c,k} \forall c\in C, \forall k\in K$$

(23)

$$R_{c,k}=R_{c,k-1}+E_{c,k-1}-d_{c,k} \forall c\in C, \forall k\in K$$

(24)

$$R_{c,1}=R_c-d_{c,1} \forall c\in C$$

(25)

where $R_c$ is the maximum battery capacity of vehicle c. $R_{c,k}$ is the remaining mileage of vehicle c battery at the end of trip k.$d_{c,k}$ represents the mileage of vehicle c during trip k.

When the battery of a battery electric vehicle runs out, the vehicle cannot drive any additional distance, so the remaining battery power cannot be negative.

$$R_{c,k}\ge 0 \forall c\in C, \forall k\in K$$

(26)

Suppose electric vehicle users connect their vehicles to the grid after each trip directly and take them off the grid to start the next trip. The vehicle may be recharged only when it is connected to the power grid, and the vehicle state constraint is

$$x_c\left(t\right)=0, { sh}_{c,k}\le t\le e{h}_{c,k} \forall c\in C, \forall k\in K$$

(27)

Before starting next trip, the battery SOC of the vehicle should meet the demand, and the constraint condition is

$$E_{c,k}\ge {SOC}_c^{set}\times R_c-R_{c,k} \forall c\in C, \forall k\in K$$

(28)

When charging during parking, users can only choose one charging mode, so the corresponding price can only be one.

$$m_1+m_2+m_3=1$$

(29)

5. Case Study

In this section, a case is used to solve the orderly charging strategy optimization model considering the charging choices of users based on the driving data of electric vehicles, and the results of the solution model are analyzed.

In this case, three charging scenarios are simulated and the results are compared and analyzed. The three charging scenarios are as follows: (1) An orderly charging strategy based on EVs driving data, e.g., the data-based orderly charging strategy proposed in this work. (2) An orderly charging strategy based on the driving data of non-electric vehicles (such as the American Household travel survey data), e.g., the model-based orderly charging strategy proposed in reference [26]. (3) Disordered charging mode, e.g., charging starts immediately when the EV is connected to the power grid.

The difference between “data-based” and “model-based” scenarios is that “data-based” scenarios take real-world driving data as the input of the optimization model, while “model-based” scenarios take certain probability distribution data as the input. The probability distribution used for the model-based scenario may come from fuel oil vehicle data, traffic flow data, questionnaire data, etc. Such a probability distribution has a certain reference value for electric vehicles, but there are also significant differences.

5.1. Problem Solving

The genetic algorithm has good robustness and is very convenient for solving discrete or complex nonlinear problems; it also has strong applicability for nonlinear, discontinuous complex systems or problems without analytic expressions. Figure 1 shows the solution process flow chart of the data-based orderly charging optimization model.

5.2. Parameter Selection

In addition to the driving data of electric vehicles described in Section 2, there are also some parameters whose values need to be specified.

Table 2. Mathematical model parameters for case study.

Parameters	Values	Units
Rc	250	km
rc	0.2	kWh/km
Ps	7	kW
Pf	24	kW
${\mu }_{OSC}$	1	/

shows the values of these parameters.

Some major assumptions should also be addressed: (1) All vehicles start the first trip fully charged. (2) All vehicles can be recharged immediately upon arrival at the charging station, and both slow and fast charging are always available at all locations. (3) The threshold value of 30 min can be changed according to the lithium-ion battery technology. (4) The expected SOC value of the user at the end of slow charging is 100%, and that of fast charging is 80%. (6) 1/3 of users are radical, 1/3 are conservative, and 1/3 are balanced.

6. Results and Discussions

Taking the classic daily load curve of grid in a certain area as a case, 1000 electric vehicles were simulated to participate in orderly charging. The input was the driving data of electric vehicles, taking half an hour as a scheduling cycle, and the output result was the charging status of each electric vehicle within each scheduling cycle.

The electricity price information in the simulation calculation was executed according to the actual charging price strategy of electric vehicles. Figure 2 shows the price information of slow charging for an area in Wuhan that does not obey power grid scheduling. As can be seen from the figure, the price of electricity in one day is divided into three grades: peak price (10, 11, 18~21 o’clock, 6 h in total), flat price (8, 9, 12~17, 22, 23 o’clock, 10 h in total), valley price (24~7 o’clock in next day, 8 h in total).

The price of charging that obeys power grid scheduling is lower than that of charging that does not obey grid scheduling, so the price of slow charging that obeys power grid scheduling is set to be 0.1 CNY/kWh lower than that of does not obey grid scheduling in the corresponding period. The fast charge price is set to be 0.1 CNY/kWh higher than the slow charge price and does not obey grid scheduling in the corresponding period.

6.1. Comparative Analysis of Data-Based and Model-Based Orderly Charging

6.1.1. Comparison of Three Charging Strategies

The load and time relationship between the data-based and model-based orderly charging strategies of EVs is obtained through simulation calculation, as shown in Figure 3. As can be seen from the figure, the solid blue line, namely the data-based orderly charging proposed in this work, has its load peak at 1:00~6:00, and the charging load is 500~600 kW. Another small peak occurs from 14:00 to 16:00, and its charging load is 200~250 kW. These two periods correspond to the valley load period of the power grid. The charging valley load of this orderly charging strategy appears in two periods, 10:00~13:00 and 18:00~22:00, respectively, and its charging load is less than 50 kW.

The model-based orderly charging strategy, the charging load curve represented by the dotted line in red, has a similar overall trend compared with the data-based orderly charging strategy, but the charging load is relatively flat, with the charging load below 200 kW throughout the day. The load peak of the model-based orderly charging strategy appears from 22:00 to 4:00 a.m. the next day; the charging load is 100~200 kW. The load valley appears around 18:00.

The load curve of disorderly charging, namely the black dotted line, is consistent with the distribution of the parking time of electric vehicles. The load peak occurs at 12:00~16:00, and 22:00~2:00 the next day. The maximum charging load exceeds 700 kW. The load valley of disorderly charging appears at 6:00~10:00, and its charging load does not exceed 50 kW.

On the whole, compared with disorderly charging, the data-based orderly charging strategy delays the time by about 2 h to stagger the load peak of the power grid, and also controls the charging load. From 12:00 to 24:00, the charging load of the data-based orderly charging strategy is about 200 kW lower than that of disorderly charging. However, the high charging load is maintained for a long time between 1:00 and 6:00, which plays a role in filling the valley of the power grid.

6.1.2. Comparison of Three Charging Strategies Superimposed on Classical Daily Load of Power Grid

The data-based orderly charging strategy is superimposed on the classical daily load of the power grid and compared with the model-based orderly charging strategy and disordered charging. The results are shown in Figure 4. As can be seen from the figure, if electric vehicles adopt disorderly charging, it will obviously lead to the addition of peaks on two power grid peak loads. In addition, disorderly charging has little effect on grid valley filling, as most of the EVs have been completely charged before 4:00.

In contrast, two orderly charging strategies both control the charging time of EVs to different degrees and reduce the EV charging times on the peak load of the grid. In addition, from 0:00 to 6:00, both of the orderly charging strategies can improve the valley load of the grid, especially the data-based orderly charging strategy.

Further analysis shows that the model-based orderly charging strategy does not have many charging arrangements in the period at 2:00~6:00. This is because the probability distribution of charging initial SOC is the overall probability distribution in a day adopted by this charging strategy. However, the actual situation is that the probability distribution of charging the initial SOC of electric vehicles is different in different time periods of the day.

Table 3. Grid load under the three charging strategies.

	Peak Load/kW	Valley Load/kW	Peak-Valley Difference/kW	Equivalent Load Fluctuation/kW
Classic daily load of power grid	3758	949.4	2808.6	867.9
Data-based orderly charging	3772	1488.4	2283.6	670.2
Model-based orderly charging	3863	1089.4	2773.6	850.8
Disorderly charging	4003	1069	2934	866.7

shows the influence of three charging strategies on the grid load. As can be seen from the table, although disorderly charging has no great influence on the equivalent load fluctuation of the power grid, it increases the peak load of the power grid from 3758 kW to 4003 kW.

Among the three charging strategies, the data-based orderly charging strategy has the least influence on the peak load of the grid, though it improves the peak-valley difference the most, and the system equivalent load fluctuation can also be reduced from 867.9 kW to 670.2 kW.

Compared with model-based ordered charging, the data-based ordered charging strategy has a better effect on reducing the peak-valley difference and equivalent load fluctuation of the grid, which reduces the peak-valley difference and equivalent load fluctuation by 17.7% and 21.2%, respectively. Compared with disorderly charging, the data-based orderly charging strategy is more effective at improving the grid.

6.1.3. Cumulative Charging Cost of Three Charging Strategies

Figure 5 shows the cumulative charging cost curves of the data-based orderly charging, model-based orderly charging, and disorderly charging strategies. As can be seen from the figure, the total cost of disorderly charging is the largest, followed by data-based orderly charging, and model-based orderly charging costs the least.

The comparison of charging cost also needs to consider the electric quantity of charging.

Table 4. Total charging quantity and cost under three charging strategies.

Charging Strategy	Charging Quantity/kWh	Total Cost/yuan	Average Unit Cost/yuan/kWh
Data-based orderly charging	5176.6	4566.5	0.88
Model-based orderly charging	3111.5	3350.9	1.08
Disorderly charging	6216.4	7281.4	1.17

shows the charging electric quantity and corresponding cost under the three charging strategies. Although the total charging electric quantity of the data-based orderly charging strategy is 5176.6 kWh, slightly less than the 6216.4 kWh of disorderly charging, it has met the demand of EV users for charging capacity. In addition, the total cost of the model-based orderly charging strategy is less than that of the data-based orderly charging strategy, but it is caused by the small electric quantity of charging. Therefore, by comparing the average unit price of charging under three charging strategies, the cost of data-based orderly charging is the lowest, followed by model-based orderly charging, and that of disorderly charging is the highest.

6.2. The Impact of Considering the User’s Charging Choices

Ideal orderly charging assumes that all electric vehicles can obey the scheduling of the grid. However, some users may disobey the charging scheduling of the power grid for subjective or objective reasons. This will cause a certain deviation between actual orderly charging and ideal orderly charging, and the deviation caused by considering the user’s charging choices is studied. The grid load curves of the data-based orderly charging strategy considering the user’s charging choices and the ideal orderly charging strategy are shown in Figure 6. In the figure, the red marked solid line represents the data-based orderly charging strategy considering the user’s charging choices; the blue dotted line represents the ideal data-based orderly charging strategy; the solid green line is the original load curve of the power grid.

As can be seen from the figure, the load curves of the two orderly charging strategies are slightly different, especially in the peak and valley periods of power grid load. In the peak load periods of 13:00 and 19:00, due to the consideration of the user’s charging choices, some EV users will not obey the scheduling of the grid, thus increasing the load peak. During the valley period of the grid load from 0:00 to 6:00, some EV users have been charged in the daytime, so there is not enough battery space to fill the valley of the grid. Although the effect on the grid of orderly charging considering the user’s choices is slightly worse than that of ideal orderly charging, the orderly charging strategy considering the user’s choices can be closer to the actual situation while optimizing the load characteristics of the grid.

6.3. Individual Electric Vehicles Analysis

In this section, an EV with different charging schedules under the two orderly charging strategies was selected to conduct a comparative analysis of the charging arrangements.

Table 5. Basic information of the individual EV.

Parameters	Values	Units
Driving distance	134	km
Time required for fast charging	1.12	h
Time required for slow charging	3.83	h
parking duration	16	h

lists the basic information about this EV.

Under the ideal orderly charging strategy, this EV will be charged according to a slow charging strategy that obeys grid scheduling, and the system will allocate the charging time according to the optimization model. However, in the orderly charging strategy considering the user’s charging choices, this EV needs to calculate charging costs in various modes, and the specific costs are calculated in

Table 6. Charging costs under different charging modes.

	Disobey Scheduling	Obey Scheduling
		Maximum	Average	Minimum
Fast charging cost/CNY	27.56	/	/	/
Slow charging cost/CNY	32.16	40.2	28.14	21.44

. Assuming that the user is a balanced type, the fast charging cost, the slow charging cost that does not obey scheduling, and the average of the slow charging cost that obeys scheduling are required to perform this calculation. As can be seen from Table 6, the cost of fast charging is the lowest, so the fast charging mode will be adopted by the optimization model considering the user’s charging choices.

Figure 7 shows the charging time arrangement of this EV according to the two orderly charging strategies. As shown in the figure, the charging time of ideal orderly charging is from 2:00 to 5:30, and considering user’s charging choices is from 12:00 to 13:00. There are obvious differences between ideal and actual scheduling both in charging time and mode. When the user base is large enough, the ideal orderly charging strategy will have a negative impact on the grid.

7. Conclusions

In this work, a centralized data-based orderly charging strategy that considers the user’s charging choices is proposed. In this strategy, the needs of the grid-side and the user-side are considered, and the optimization model is based on the driving data of electric vehicles. The main research results are as follows:

An orderly charging method that considers the user’s charging choices is introduced, and the charging time scheduling model of EVs is established. The optimization model can better meet the power demand of both the power grid-side and the user-side. The goal of the power grid-side is to minimize the equivalent load fluctuation of the system, and the goal of the user-side is to minimize the charging cost and maximize the electric quantity of charging.

Furthermore, the scheduling capabilities of three charging strategies, namely data-based orderly charging, model-based orderly charging, and disorderly charging, are analyzed based on a series of driving data of electric vehicles. The results show that the data-based orderly charging strategy reduces the equivalent load fluctuation and peak–valley difference of the power grid significantly, and greatly reduces the charging cost of users, making it superior to the other two charging strategies.

Additionally, the effect of the user’s charging choices on the charging strategy is analyzed, and the results show that the orderly charging strategy that considers the user’s charging choices can effectively reduce the scheduling deviation caused by the user’s charging choices.

Of course, the genetic algorithm is flawed, and the output of the optimization model may not be the global optimal solution. In the next step, the solving algorithm can be improved. In addition, the establishment of mathematical models can further integrate renewable energy generation and vehicle-to-grid (V2G) technology. Thus, electric vehicles could make the power grid more economical and environmentally friendly.