Electric vehicle charging load bearing improving method comprising distributed photovoltaic power distribution network

Abstract

The invention discloses an electric vehicle charging load bearing improving method containing a distributed photovoltaic power distribution network, relates to the photovoltaic field, and constructs a hybrid demand response electric vehicle ordered charging control method based on Markov chain prediction in order to maximize the utilization rate of renewable energy sources and consider the uncertainty of EV user demands. The method comprises the following steps: firstly, introducing a Markov chain into electric vehicle charging load prediction to obtain a charging load curve; charging time is divided into a peak, a flat and a valley; and peak load shifting and supply and demand balance are realized by adjusting the charging demand of the user through mixed demand response. Secondly, a deterministic optimization target model is established with the maximum user income, the maximum PVs income and power balance; and finally, the risk of the PV uncertainty to the optimization target model is quantified through the CVaR, the PV uncertainty is introduced into the optimization target model, and the model robustness is improved.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaics, and in particular to a method for increasing the charging load carrying capacity of electric vehicles containing a distributed photovoltaic distribution network. Background Art

[0002] With growing global environmental awareness and the development of renewable energy, electric vehicles (EVs) are gaining widespread attention and adoption as a clean, low-carbon means of transportation. However, the rise of EVs also presents a series of new challenges. Traditional random charging, where EVs are connected to the power grid whenever they need to charge, presents a number of problems. First, the lack of forecasting and scheduling strategies for charging demand leads to uneven load distribution. During certain periods of time, a large number of EVs may be charging simultaneously, resulting in significant fluctuations in the power system load. Second, random charging also fails to fully utilize renewable energy, resulting in energy waste and increased costs.

[0003] To address the challenges of traditional random charging methods, this paper proposes an electric vehicle charging optimization method that considers photovoltaic uncertainty. This method uses a Markov chain model to predict electric vehicle charging demand. Based on this prediction, it also incorporates hybrid demand response to flexibly adjust charging strategies to achieve optimal control of electric vehicle charging. This method balances system loads, improves energy utilization, and achieves a win-win situation for both users and photovoltaic charging stations. Summary of the Invention

[0004] To address the challenges of traditional random charging methods, this paper proposes a method for increasing the charging load capacity of electric vehicles (EVs) within a distributed photovoltaic distribution network. This method utilizes a Markov chain model to predict EV charging demand, establishing an EV charging load prediction model. Based on this prediction, a hybrid demand response approach is combined to flexibly adjust the charging strategy and establish an orderly charging control model for EVs. This optimizes EV charging control. This method balances system load, improves energy utilization, and achieves a win-win situation for both users and PV charging stations. This reduces grid load spikes, improves energy utilization, and enhances user and charging station revenue. This method provides an orderly charging control method.

[0005] The technical solution of the present invention is:

[0006] A method for increasing the charging load capacity of electric vehicles (EVs) in a distributed photovoltaic distribution network and an orderly charging control method for electric vehicles (EVs) and photovoltaic charging stations (PVs) is proposed. This method improves prediction accuracy by introducing a Markov chain into charging load forecasting. Based on this, a hybrid demand response-based orderly charging control model for EVs is established, aiming to achieve a balance between energy supply and demand in the power system and a win-win situation for charging stations and users through optimized EV scheduling. The method includes the following steps:

[0007] S1: Use K-means fuzzy algorithm to divide the charging time;

[0008] S2: Establish a charging load prediction model based on Markov chain;

[0009] S3: Establish an EV orderly charging control model based on hybrid demand response;

[0010] S4: Establish a robust orderly charging control model based on information gap decision theory;

[0011] S5: Use the ε-constrained and non-dominated sorting genetic algorithm II (NSGA-II) algorithms to jointly solve the multi-objective model;

[0012] Furthermore, in step S1, a day is divided into 24 time points, each time period is 1 hour apart, and the steps of using the K-means fuzzy algorithm to divide the charging time into peak, flat and valley are as follows:

[0013] S11: Initialization: Randomly select 3 initial cluster centers.

[0014] S12: Calculate the membership matrix: For each sample, use the Euclidean distance to calculate its distance from each cluster center, and group each sample into the cluster center closest to it;

[0015]

[0016] Where dist(i,j) represents the i-th sample x i and the jth cluster center c j The distance between i represents the characteristic value of the i-th sample; c j Represents the eigenvalue of the j-th cluster center.

[0017] S13: Find the mean of samples of each class as the new cluster center.

[0018]

[0019] S14: Judgment: If the three cluster centers no longer change, the iteration ends, otherwise return to S12 and minimize the square error:

[0020]

[0021] Furthermore, in step S2, the steps of establishing a charging load prediction model based on a Markov chain are as follows:

[0022] S21: determine the time points, divide the day into 24 time points, each time point contains 1 hour;

[0023] S22: State definition, discretize the EV power consumption into different states, divide the entire power consumption range into multiple intervals, and define a state for each interval {S n ,n=0,1,2,…};

[0024] S23: Calculate the transition probability based on the collected data and real-time temperature data. Calculate the transition probability of the user transitioning from one power consumption state to another. Based on the power consumption at the current time point, that is, the current state, calculate the number of transitions from this state to other states. Calculate the number of transitions from the current state to time t+1 in the historical data and divide it by the total number of transitions to the current state to obtain the transition probability:

[0025]

[0026] Where, P t,t+1 is the probability of transitioning from the state at time t to the state at time t+1.

[0027] S24: Continuously rolling forecasts to complete the forecast for the day;

[0028] S25: Select the mean absolute error (MAE) indicator for deviation evaluation. The expression is as follows:

[0029]

[0030] Where x i To predict power consumption; x 0i is the actual value.

[0031] S26: Count the user's daily power consumption.

[0032] S27: Calculate the user's charging load. After obtaining the user's power consumption, the EV charging demand can be calculated using the following formula:

[0033]

[0034] Where, I is the user's power consumption; T u E is the daily usage time of electric vehicles, r Charging load for users.

[0035] Furthermore, based on step S2, a hybrid demand response orderly charging control model is established. In step S3, the steps of establishing the hybrid demand response orderly charging control model are as follows:

[0036] S31: Optimization model based on price demand response is established with the goal of maximizing the benefits of EVs during the price demand response period:

[0037]

[0038] Where, ρ(t) is the electricity price at time t; d(t) is the electricity demand of users at time t after responding to the DR plan; ρ0 is the initial electricity price; d0 is the electricity demand of users who do not respond to the DR plan; t start is the starting time when the user starts to respond to the price demand response; t end End charging time for the user; t，l A is the battery life loss cost; t，l The cost of electricity loss.

[0039] The EV constraints are:

[0040] S min ≤S(t)≤S max (8)

[0041]

[0042] U min ≤U(t)≤U max (10)

[0043] Where S min is the minimum state of charge; S(t) is the state of charge at time t; S max is the maximum state of charge S 0，l is the initial battery state of charge; E bat is the battery capacity of the electric vehicle; S E，l : desired battery state of charge; U min is the minimum voltage; is the voltage at time t; U max is the maximum voltage.

[0044] S32: Calculate the power supply margin after price demand response:

[0045]

[0046] in:

[0047] P pv (t) = B(t) × S × η pv ×Δt (12)

[0048]

[0049] When A(t) is equal to 0.5, it is an equilibrium point, at which point the supply and demand balance can be achieved by price demand response alone.

[0050] When A(t) is greater than 0.5, it means that there are fewer charging users at time t. The larger the margin, the greater the incentive should be given to users to encourage them to charge at this time.

[0051] When A(t) is less than 0.5, it means that there are many users charging at this time. The smaller the margin, the smaller the incentive should be given to users, prompting them to choose to charge at another time.

[0052] Where A(t) is the power supply margin at time t 0≤A(t)≤1; P pv (t) is the amount of electricity charged by photovoltaic cells at time t; d(t) is the load demand of EV users at time t; B(t) is the light intensity at time t B(t)∈[0,1]; S is the area of ​​the photovoltaic panel; η pv is the efficiency of the photovoltaic panel; Δt is the time interval; ω is the amplitude (maximum light intensity); is the frequency, representing 24 hours a day; is the phase difference;

[0053] PVs constraints:

[0054] Spinning reserve constraints:

[0055] P up =P max -P pv，t ≥r1%·P pv，t (14)

[0056] P down =p pv，t -P min ≥r2%·P pv，t (15)

[0057] Where, P up is the upper spinning reserve capacity; P down is the lower spinning reserve capacity; P max is the maximum output power of PVs, that is, the maximum power that PVs can provide under the most favorable weather conditions; P min is the minimum output power of PVs, that is, the minimum power that PVs can still provide under the most unfavorable weather conditions; P pv，tis the actual power provided by PVs at time t; r1 / r2 are the upper and lower spinning reserve demand coefficients respectively.

[0058] The supply of PVs cannot exceed their rated limit:

[0059] g t，min ≤g t ≤g t ,max (16)

[0060] Where g t，min is the minimum power generation of PVs; g t，max is the maximum power generation of the photovoltaic charging station.

[0061] Distribution transformer capacity: The total load at all times of the day should not exceed the maximum load of the transformer:

[0062] P t -g t <P MTF (17)

[0063] P MTF =C MTF ·k·τ (18)

[0064] Where, P t is the charging load; P MTF is the maximum load capacity of the distribution transformer; C MTF is the rated capacity of the transformer; k is the power factor; τ is the efficiency.

[0065] S33: When A(t)≠0.5, the user is incentivized. According to the characteristics of the tanh function, the constructed incentive function is expressed as follows:

[0066] D(t)=|tanh(A(t)-0.5)| (19)

[0067] Furthermore, since there is uncertainty in the actual PV power generation, the present invention uses the information gap decision theory to deal with the adverse effects brought about by the PV uncertainty. In step S4, the steps of establishing a robust orderly charging control model based on the information gap decision theory are as follows:

[0068] S41: Establish a deterministic optimization target model:

[0069] f1=max(f PDDR +D(t)-(e k，l +A k，l )) (20)

[0070]

[0071]

[0072] Wherein, objective function f1 minimizes the user charging cost; objective function f2 maximizes the daily revenue of PVs; objective function f3 minimizes the difference between the electric vehicle charging load and the output power of the photovoltaic charging station.

[0073] S42: Use conditional value at risk to quantify the risk brought by PV uncertainty to the optimization target model, expressed as follows:

[0074]

[0075] in:

[0076]

[0077] Where, represents the mathematical expectation; α is the confidence level; ξ is the VaR value under the confidence level α; (x, ξ) is the distribution function of the loss function f(x, y) that does not exceed the boundary value ξ.

[0078] S43: The information gap decision theory model based on risk aversion is:

[0079] maxθ θ≥0 (25)

[0080]

[0081] Where F(Y,d) is the objective function; Y is the uncertain variable; d is the decision variable; (1-β θ )F0 is the bottom line value (that is, the worst case acceptable to the decision maker); The deviation between the actual value and the predicted value cannot exceed R(Y,d) is the model equality constraint; R(Y,d) is the model inequality constraint.

[0082] According to formulas (25)-(26), the PV fluctuation range is:

[0083]

[0084] Where, is the predicted value of PV power generation.

[0085] If the optimization result of the objective function f2 of the deterministic optimization model is maxf2. According to the information gap decision theory model, and assuming that the minimum acceptable return of the decision maker is (1-β Re )maxf2, then the robust model based on information gap decision theory is:

[0086]

[0087] Where, Re min is the optimization result of f2 under maximum disturbance.

[0088] This model shows that the fluctuation range of the actual photovoltaic power generation is When the optimization result is within 100%, the final optimization result can ensure that the income of the photovoltaic charging station is not lower than the expected worst result (1-β θ )F0.

[0089] Furthermore, the NSGA-II algorithm is a more efficient algorithm for solving multi-objective problems. By introducing an elite-preserving operator, the diversity of the solution set is ensured, making the Pareto solution set uniformly distributed. After obtaining the feasible solution set, fuzzy theory is used to obtain the optimal compromise solution. In step S5, the NSGA-II algorithm solves the multi-objective model as follows:

[0090] S51: Using the ε-constraint method, the three-objective optimization problem is transformed into a series of two-objective optimization problems, where the objective that conflicts most with the other objectives is used as the constraint. Based on preliminary judgment, the PVs return conflicts most with the other objectives. To verify the accuracy of this judgment, the Spearman rank correlation coefficient is introduced. Taking variables x and y as an example, the formula for calculating the Spearman correlation coefficient can be expressed as:

[0091]

[0092] in:

[0093] d j =r xj -r yj (30)

[0094] Where, ρ xy is the Spearman correlation coefficient between x and y, and its value range is [-1,1]. xy A positive value indicates a harmonious relationship between the two, a negative value indicates conflict, and 0 indicates no correlation; r xj and r yj represents the rank value of the jth solution of x and y among J solutions; d j The Spearman correlation coefficient between the optimization objectives is calculated using the objective function values ​​corresponding to the three single-objective optimal solutions.

[0095] S52: The objective function with the most negative number is the most conflicting objective. After selecting a suboptimal objective, the four-objective optimization problem can be transformed into a three-objective optimization problem. Assuming f2 is the suboptimal objective and f1 and f3 are the main objectives, the transformation steps are as follows:

[0096] Calculate maxf2(x) and minf2(x) to determine the value range of f2, that is:

[0097] Δf2=f 2max -f 2min (31)

[0098] Divide Δf2 into q equal parts to obtain q+1 grid points. The value of the j-th grid point of f2 is expressed as f 2,j =f 2min +jΔf2 / q,j=0,1,…,q;

[0099] Transform the deterministic problem into the following form:

[0100]

[0101] Wherein, l0 is a constant whose value is within [0.001, 0.1]. The present invention sets l0 to 0.01; is an auxiliary variable of the selected suboptimal objective, which is introduced to find feasible solutions between adjacent grid points; 2,j Finally, NSGA-II is used to solve the multi-objective functions with the same optimization direction.

[0102] S53: Taking f1 as the primary objective, determine the maximum user benefit and obtain the corresponding PVs benefit and load fluctuation variance. Similarly, taking f2 and f3 as the primary objectives, respectively, obtain the optimal primary objective solution and other secondary objective values.

[0103] S54: Using the optimal value and the worst value in the above steps as the upper limit and the lower limit, respectively, calculate the fuzzy membership of each target:

[0104] For positive targets, the “bigger is better” transformation is adopted, which is given by the following formula:

[0105]

[0106] For negative targets, the “smaller the better” transformation is adopted, which is given by the following formula:

[0107]

[0108] The comprehensive membership can be obtained by linearly weighting the membership of the three objectives:

[0109]

[0110] Where, M(f i ), i=1,2,3 is the fuzzy membership of the i-th target; f i,max and f i,min is the maximum and minimum value of the i-th target; is the weight of the i-th objective; F is the comprehensive membership of the three objectives. The solution with the largest comprehensive membership among all solutions is regarded as the optimal compromise solution.

[0111] The beneficial effects of the present invention are:

[0112] 1. Balancing system loads. Traditional random charging methods are prone to load imbalance. However, electric vehicle charging optimization methods that consider photovoltaic uncertainty can achieve balanced distribution of charging loads by predicting charging demand and rationally deploying charging strategies, thereby reducing impact on the power grid and improving the stability of the power supply.

[0113] 2. Improve energy utilization. This method combines the Markov chain model and hybrid demand response technology. Based on the predicted charging demand, it can flexibly adjust the charging load when PV resources are limited, optimize energy management, and improve energy utilization efficiency.

[0114] 3. Achieve a win-win situation for users and PV charging stations. Hybrid demand response technology makes EV charging more flexible. It can adjust charging strategies based on PV power generation conditions, providing users with more convenient and efficient charging services. It can also improve the utilization rate and economic benefits of PV charging stations, achieving a win-win situation for users and PV charging stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0115] Figure 1 It is a flow chart of the present invention.

[0116] Figure 2 It is a system structure diagram.

[0117] Figure 3 It is a K-means fuzzy clustering division charging time flow chart. DETAILED DESCRIPTION

[0118] The present invention will be further described below with reference to the accompanying drawings.

[0119] Reference Figures 1 to 3 In this paper, we propose a method for improving the charging load capacity of electric vehicles in a distributed photovoltaic distribution network. The method for orderly charging control of electric vehicles is aimed at electric vehicles (EVs) and photovoltaic charging stations (PVs). By introducing Markov chains into charging load forecasting, the prediction accuracy is improved. On this basis, an orderly charging control model for EVs based on hybrid demand response is established, which aims to achieve a balance between energy supply and demand in the power system and a win-win situation for charging stations and users by optimizing EV scheduling. The method includes the following steps:

[0120] S1: Use K-means fuzzy algorithm to divide the charging time;

[0121] S2: Establish a charging load prediction model based on Markov chain;

[0122] S3: Establish an EV orderly charging control model based on hybrid demand response;

[0123] S4: Establish a robust orderly charging control model based on information gap decision theory;

[0124] S5: Solve the multi-objective model using the NSGA-II algorithm;

[0125] Furthermore, in step S2, the steps of establishing a charging load prediction model based on a Markov chain are as follows:

[0126] S21: Determine time points. Divide a day into 24 time points, each containing 1 hour;

[0127] S22: State definition. Discretize the EV's power consumption into different states. Divide the entire power consumption range into multiple intervals and define a state for each interval {S n ,n=0,1,2,…};

[0128] S23: Calculate the transition probability. Based on the collected data and real-time temperature data, calculate the transition probability of the user transitioning from one power consumption state to another. Based on the power consumption at the current time point, that is, the current state, calculate the number of transitions from this state to the other state. Calculate the number of transitions from the current state to the time t+1 in the historical data and divide it by the total number of transitions to the current state to obtain the transition probability:

[0129]

[0130] Where, P t,t+1 is the probability of transitioning from the state at time t to the state at time t+1.

[0131] S24: Continuously rolling forecasts to complete the forecast for the day;

[0132] S25: Select the mean absolute error (MAE) indicator for deviation evaluation. The expression is as follows:

[0133]

[0134] The smaller the MAE value, the smaller the prediction error of the model and the better the performance.

[0135] Where x i To predict power consumption; x 0i is the actual value.

[0136] S26: Count the user's daily power consumption.

[0137] S27: Calculate the user's charging load. After obtaining the user's power consumption, the EV's charging requirements can be calculated using the following formula (assuming the user consumes the same amount of power as required):

[0138]

[0139] Where, I is the user's power consumption; T u E is the daily usage time of electric vehicles, r Charging load for users.

[0140] Furthermore, based on step S2, a hybrid demand response orderly charging control model is established. In step S3, the steps of establishing the hybrid demand response orderly charging control model are as follows:

[0141] S31: Develop a price demand response optimization model with the goal of maximizing EV revenue during the price demand response period:

[0142]

[0143] Where, ρ(t) is the electricity price at time t; d(t) is the electricity demand of users at time t after responding to the DR plan; ρ0 is the initial electricity price; d0 is the electricity demand of users who do not respond to the DR plan; t start is the starting time when the user starts to respond to the price demand response; t end End charging time for the user; t,l A is the battery life loss cost; t,l The cost of electricity loss.

[0144] The EV constraints are:

[0145] S min ≤S(t)≤S max (5)

[0146]

[0147] U min ≤U(t)≤U max (7)

[0148] Where S min is the minimum state of charge; S(t) is the state of charge at time t; S max is the maximum state of charge S 0，l is the initial battery state of charge; E bat is the battery capacity of the electric vehicle; S E，l : desired battery state of charge; U minis the minimum voltage; is the voltage at time t; U max is the maximum voltage.

[0149] S32: Calculate the power supply margin after price demand response:

[0150]

[0151] in:

[0152] P pv (t) = B(t) × S × η pv ×Δt (8)

[0153]

[0154] When A(t) is equal to 0.5, it is an equilibrium point, at which point the charging of electric vehicles can achieve supply and demand balance by relying solely on the price demand response strategy.

[0155] When A(t) is greater than 0.5, it means that there are fewer charging users at time t. The larger the margin, the greater the incentive should be given to users to encourage them to charge at this time.

[0156] When A(t) is less than 0.5, it means that there are many users charging at this time. The smaller the margin, the smaller the incentive should be given to users, prompting them to choose to charge at another time.

[0157] Where A(t) is the power supply margin at time t 0≤A(t)≤1; P pv (t) is the amount of electricity charged by photovoltaic cells at time t; d(t) is the load demand of EV users at time t; B(t) is the light intensity at time t B(t)∈[0,1]; S is the area of ​​the photovoltaic panel; η pv is the efficiency of the photovoltaic panel; Δt is the time interval; ω is the amplitude (maximum light intensity); is the frequency, representing 24 hours a day; is the phase difference;

[0158] PVs constraints:

[0159] Spinning reserve constraints:

[0160] P up =P max -P pv，t ≥r1%·P pv，t (10)

[0161] P down =P pv，t -P min ≥r2%·P pv，t (11)

[0162] Where, Pup is the upper spinning reserve capacity; P down is the lower spinning reserve capacity; P max is the maximum output power of PVs, that is, the maximum power that PVs can provide under the most favorable weather conditions; P min is the minimum output power of PVs, that is, the minimum power that PVs can still provide under the most unfavorable weather conditions; P pv，t is the actual power provided by PVs at time t; r1 / r2 are the upper and lower spinning reserve demand coefficients respectively.

[0163] The supply of PVs cannot exceed their rated limit:

[0164] g t，min ≤g t ≤g t，max (12)

[0165] Where g t，min is the minimum power generation of PVs; g t，max is the maximum power generation of the photovoltaic charging station.

[0166] Distribution transformer capacity: The total load at all times of the day should not exceed the maximum load of the transformer:

[0167] P t -g t <P MTF (13)

[0168] P MTF =C MTF ·k·τ (14)

[0169] Where, P t is the charging load; P MTF is the maximum load capacity of the distribution transformer; C MTF is the rated capacity of the transformer; k is the power factor; τ is the efficiency.

[0170] S33: When A(t)≠0.5, the user is incentivized. According to the characteristics of the tanh function, the constructed incentive function is expressed as follows:

[0171] D(t)=|tanh(A(t)-0.5)| (15)

[0172] Furthermore, since there is uncertainty in the actual PV power generation, the present invention uses the information gap decision theory to deal with the adverse effects brought about by the PV uncertainty. In step S4, the steps of establishing a robust orderly charging control model based on the information gap decision theory are as follows:

[0173] S41: Establish a deterministic optimization target model:

[0174] f1=max(f PSDR +D(t)-(e k，l +A k，l )) (20)

[0175]

[0176]

[0177] Wherein, objective function f1 minimizes the user charging cost; objective function f2 maximizes the daily revenue of PVs; objective function f3 minimizes the difference between the electric vehicle charging load and the output power of the photovoltaic charging station.

[0178] S42: Use conditional value at risk to quantify the risk brought by PV uncertainty to the optimization target model, expressed as follows:

[0179]

[0180] in:

[0181]

[0182] Where, represents the mathematical expectation; α is the confidence level; ξ is the VaR value under the confidence level α; (x,ξ) is the distribution function of the loss function f(x,y) that does not exceed the boundary value ξ.

[0183] S43: The information gap decision theory model based on risk aversion is:

[0184] maxθ θ≥0 (20)

[0185]

[0186] Where F(Y,d) is the objective function; Y is the uncertain variable; d is the decision variable; (1-β θ )F0 is the bottom line value (that is, the worst case acceptable to the decision maker); The deviation between the actual value and the predicted value cannot exceed R(Y,d) is the model equality constraint; R(Y,d) is the model inequality constraint.

[0187] According to formulas (20) and (21), the PV fluctuation range is:

[0188]

[0189] Where, is the predicted value of PV power generation.

[0190] If the optimization result of the objective function f2 of the deterministic optimization model is maxf2. According to the information gap decision theory model, and assuming that the minimum acceptable return of the decision maker is (1-β Re )maxf2, then the robust model based on information gap decision theory is:

[0191]

[0192] Where, Re min is the optimization result of f2 under maximum disturbance.

[0193] This model shows that when the fluctuation range of the actual photovoltaic power generation is When the optimization result is within 100%, the final optimization result can ensure that the income of the photovoltaic charging station is not lower than the expected worst result (1-β θ )F0.

[0194] Furthermore, the NSGA-II algorithm is a more efficient algorithm for solving multi-objective problems. By introducing an elite-preserving operator, the diversity of the solution set is ensured, making the Pareto solution set uniformly distributed. After obtaining the feasible solution set, fuzzy theory is used to obtain the optimal compromise solution. In step S5, the NSGA-II algorithm solves the multi-objective model as follows:

[0195] S51: Using the ε-constraint method, the three-objective optimization problem is transformed into a series of two-objective optimization problems, where the objective that conflicts most with the other objectives is used as the constraint. Based on preliminary judgment, the PVs return conflicts most with the other objectives. To verify the accuracy of this judgment, the Spearman rank correlation coefficient is introduced. Taking variables x and y as an example, the formula for calculating the Spearman correlation coefficient can be expressed as:

[0196]

[0197] in:

[0198] d j =r xj -r yj (25)

[0199] Where, ρ xy is the Spearman correlation coefficient between x and y, and its value range is [-1,1]. xy A positive value indicates a harmonious relationship between the two, a negative value indicates conflict, and 0 indicates no correlation; r xj and r yj represents the rank value of the jth solution of x and y among J solutions; d j The Spearman correlation coefficient between the optimization objectives is calculated using the objective function values ​​corresponding to the three single-objective optimal solutions.

[0200] S52: The objective function with the most negative number is the most conflicting objective. After selecting a suboptimal objective, the four-objective optimization problem can be transformed into a three-objective optimization problem. Assuming f2 is the suboptimal objective and f1 and f3 are the main objectives, the transformation steps are as follows:

[0201] Calculate maxf2(x) and minf2(x) to determine the value range of f2, that is:

[0202] Δf2=f 2max -f 2min (26)

[0203] Divide Δf2 into q equal parts to obtain q+1 grid points. The value of the j-th grid point of f2 is expressed as f 2,j =f 2min +jΔf2 / q,j=0,1,…,q;

[0204] Transform the deterministic problem into the following form:

[0205]

[0206] Wherein, l0 is a constant whose value is within [0.001, 0.1]. The present invention sets l0 to 0.01; is an auxiliary variable of the selected suboptimal objective, which is introduced to find feasible solutions between adjacent grid points; 2,j Finally, NSGA-II is used to solve the multi-objective functions with the same optimization direction.

[0207] S53: Taking f1 as the primary objective, determine the maximum user benefit and obtain the corresponding PVs benefit and load fluctuation variance. Similarly, taking f2 and f3 as the primary objectives, respectively, obtain the optimal primary objective solution and other secondary objective values.

[0208] S54: Using the optimal value and the worst value in the above steps as the upper limit and the lower limit, respectively, calculate the fuzzy membership of each target:

[0209] For positive targets, the “bigger is better” transformation is adopted, which is given by the following formula:

[0210]

[0211] For negative targets, the “smaller the better” transformation is adopted, which is given by the following formula:

[0212]

[0213] The comprehensive membership can be obtained by linearly weighting the membership of the three objectives:

[0214]

[0215] Where, M(f i ), i=1,2,3 is the fuzzy membership of the i-th target; f i,max and f i,min is the maximum and minimum value of the i-th target; is the weight of the i-th objective; F is the comprehensive membership of the three objectives. The solution with the largest comprehensive membership among all solutions is regarded as the optimal compromise solution.

[0216] The contents described in the embodiments of this specification are merely an enumeration of the implementation forms of the inventive concept. The scope of protection of the present invention should not be regarded as limited to the specific forms described in the examples. The scope of protection of the present invention also includes equivalent technical means that can be thought of by those skilled in the art based on the inventive concept.

Claims

1. A method for increasing the charging load of electric vehicles in a distributed photovoltaic distribution network, characterized in that: The following steps are involved: S1: Use K-means fuzzy algorithm to divide the charging time; S2: Establish a charging load prediction model based on Markov chain; S3: Establish an EV orderly charging control model based on hybrid demand response; S4: Establish a robust orderly charging control model based on information gap decision theory; S5: Use ε-constrained and non-dominated sorting genetic algorithm II algorithms to jointly solve the multi-objective model.

2. The method for increasing the charging load of electric vehicles in a distributed photovoltaic power distribution network according to claim 1, characterized in that: In step S1, the steps of dividing the charging time using the K-means fuzzy algorithm are as follows: S11: Initialization: Randomly select 3 initial cluster centers; S12: Calculate the membership matrix: For each sample, use the Euclidean distance to calculate its distance from each cluster center, and group each sample into the cluster center closest to it: Where dist(i,j) represents the i-th sample x i and the jth cluster center c j The distance between i represents the characteristic value of the i-th sample; c j Represents the eigenvalue of the jth cluster center; S13: Find the mean of samples of each class as the new cluster center: S14: Judgment: If the three cluster centers no longer change, the iteration ends, otherwise return to S12; minimize the square error:

3. The method for increasing the charging load capacity of electric vehicles with a distributed photovoltaic distribution network according to claim 2, characterized in that: In step S2, the steps of establishing a charging load prediction model based on a Markov chain are as follows: S21: determine the time points, divide the day into 24 time points, each time point contains 1 hour; S22: State definition, discretize the EV power consumption into different states, divide the entire power consumption range into multiple intervals, and define a state for each interval {S n ,n=0,1,2,…}; S23: Calculate the transition probability. Based on the collected data and the real-time temperature data, calculate the transition probability of the user transitioning from one power consumption state to another. Based on the power consumption at the current time point, that is, the current state, calculate the number of transitions from the current state to the other state. Calculate the number of transitions from the current state to the time t+1 in the historical data and divide it by the total number of transitions to the current state to obtain the transition probability: Where, P t,t+1 is the probability of transitioning from the state at time t to the state at time t+1; S24: Continuously rolling forecasts to complete the forecast for the day; S25: Select the mean absolute error (MAE) indicator for deviation evaluation, the expression is as follows: Where x i To predict power consumption; x 0i is the actual value; S26: Count the user's daily power consumption; S27: Calculate the user's charging load. After obtaining the user's power consumption, calculate the EV's charging demand using the following formula: Where, I is the user's power consumption; T u is the daily usage time of the electric vehicle, E r Charging load for users.

4. The method for increasing the charging load of electric vehicles in a distributed photovoltaic power distribution network according to claim 3, characterized in that: In step S3, the steps of establishing the EV orderly charging control model based on hybrid demand response are as follows: S31: Optimization model based on price demand response is established with the goal of maximizing EV revenue during the price demand response period: Where, ρ(t) is the electricity price at time t; d(t) is the electricity demand of users at time t after responding to the DR plan; ρ0 is the initial electricity price; d0 is the electricity demand of users who do not respond to the DR plan; t start The moment when the user starts responding to PSDR; t end End charging time for the user; t,l A is the battery life loss cost; t,l Cost of electricity loss; The EV constraints are: S min ≤S(t)≤S max (8) U min ≤U(t)≤U max (10) Where S min is the minimum state of charge; S(t) is the state of charge at time t; S max is the maximum state of charge S 0,l is the initial battery state of charge; E bat is the battery capacity of the electric vehicle; S E,l : desired battery state of charge; U min is the minimum voltage; is the voltage at time t; U max is the maximum voltage; S32: Calculate the power supply margin after price demand response: in: P pv (t)=B(t)×S×η pv ×Δt (12) Where A(t) is the power supply margin at time t 0≤A(t)≤1; P pv (t) is the amount of electricity charged by photovoltaic power at time t; d(t) is the load demand of EV users at time t; B(t) is the light intensity at time t B(t)∈[0,1]; S is the area of ​​the photovoltaic panel; η pv is the efficiency of the photovoltaic panel; Δt is the time interval; ω is the amplitude; is the frequency, representing 24 hours a day; is the phase difference; PVs constraints: Spinning reserve constraints: P up =P max -P pv,t ≥r1%·P pv,t (14) P down =P pv,t -P min ≥r2%·P pv,t (15) Where, P up is the upper spinning reserve capacity; P down is the lower spinning reserve capacity; P max is the maximum output power of PVs, that is, the maximum power that PVs can provide under the most favorable weather conditions; P min is the minimum output power of PVs, that is, the minimum power that PVs can still provide under the most unfavorable weather conditions; P pv,t is the actual power provided by PVs at time t; r1 / r2 are the upper and lower spinning reserve demand coefficients respectively; The supply of PVs cannot exceed their rated limit: g t,min ≤g t ≤g t ,max (16) Where g t,min is the minimum power generation of PVs, g t,max is the maximum power generation of the photovoltaic charging station; Distribution transformer capacity: The total load at all times of the day should not exceed the maximum load of the transformer: P t -g t <P MTF (17) P.S MTF (C MTF ·k·τ (18) Where, P t is the charging load; P MTF is the maximum load capacity of the distribution transformer; C MTF is the rated capacity of the transformer; k is the power factor; τ is the efficiency; S33: When A(t)≠0.5, the user is incentivized. According to the characteristics of the tanh function, the constructed incentive function is expressed as follows: D(t)=|tanh(A(t)-0.5)| (19).

5. The method for increasing the charging load capacity of electric vehicles with a distributed photovoltaic distribution network according to claim 4, characterized in that: The step S4 comprises: S41: Establish a deterministic optimization target model: f1=max(f PSDR +D(t)-(e k,l +A k,l )) (20) Where, the objective function f1 is to minimize the user charging cost; the objective function f2 is to maximize the daily revenue of PVs; the objective function f3 is to minimize the difference between the electric vehicle charging load and the output power of the photovoltaic charging station; S42: Use conditional value at risk to quantify the risk brought by PVs uncertainty to the optimization target model, expressed as follows: in: Where, represents the mathematical expectation; α is the confidence level; ξ is the VaR under the confidence level α (x,ξ) is the distribution function of the loss function f(x,y) not exceeding the boundary value ξ; S43: The information gap decision theory model based on risk aversion is: maxθ θ≥0 (25) Where F(Y,d) is the objective function; Y is the uncertain variable; d is the decision variable; (1-β θ )F0 is the bottom line value (that is, the worst case acceptable to the decision maker); The deviation between the actual value and the predicted value cannot exceed R(Y,d) is the model equality constraint; R(Y,d) is the model inequality constraint; According to formulas (25)-(26), the PV fluctuation range is: Where, is the predicted value of PV power generation; If the optimization result of the objective function f2 of the deterministic optimization model is maxf2, according to the information gap decision theory model, and assuming that the minimum acceptable return of the decision maker without interference is (1-β Re )maxf2, then the robust model based on information gap decision theory is: Where, Re min is the optimization result of f2 under maximum disturbance; This model indicates that when the fluctuation range of the actual photovoltaic power generation is When the optimization result is within 100%, the final optimization result can ensure that the income of the photovoltaic charging station is not lower than the expected worst result (1-β θ )F0.

6. A method for increasing the charging load capacity of electric vehicles with a distributed photovoltaic distribution network according to claim 5, characterized in that: The step S5 comprises: S51: Use the ε constraint method to transform the three-objective optimization problem into a series of two-objective optimization problems, where the objective with the greatest conflict with other objectives is used as a constraint. The Spearman rank correlation coefficient is introduced. Taking variables x and y as an example, the calculation formula of the Spearman correlation coefficient is expressed as: in: d j ＝r xj -r yj (30) Where, ρ xy is the Spearman correlation coefficient between x and y, ranging from -1 to 1; ρ xy A positive value indicates a harmonious relationship between the two, a negative value indicates conflict, and 0 indicates no correlation; r xj and r yj represents the rank value of the jth solution of x and y among J solutions; d j Represents the difference in rank values ​​of the j-th solution of two variables; the Spearman correlation coefficient between the optimization objectives is calculated using the objective function values ​​corresponding to the three single-objective optimal solutions; S52: The objective function with the most negative number is the most conflicting objective. After selecting a suboptimal objective, the three-objective optimization problem is transformed into a two-objective optimization problem. Assuming f2 is the suboptimal objective and f1 and f3 are the main objectives, the transformation steps are as follows: Calculate maxf2(x) and minf2(x) to determine the value range of f2, that is: Δf2=f 2max -f 2min (31) Divide Δf2 into q equal parts to obtain q+1 grid points. The value of the j-th grid point of f2 is expressed as f 2,j =f 2min +jΔf2 / q,j=0,1,…,q; Transform the deterministic problem into the following form: Where l0 is a constant whose value is within [0.001, 0.1], and l0 is set to 0.01; is an auxiliary variable of the selected suboptimal objective, which is introduced to find feasible solutions between adjacent grid points; 2,j Afterwards, NSGA-II is used to solve the multi-objective functions with the same optimization direction; S53: Taking f1 as the primary objective to determine the maximum user benefit, the corresponding PVs benefit and load fluctuation variance are obtained. Similarly, taking f2 and f3 as the primary objectives, the optimal primary objective solution and other secondary objective values ​​are obtained. S54: Using the optimal value and the worst value in the above steps as the upper limit and the lower limit, respectively, calculate the fuzzy membership of each target: For positive targets, the "bigger is better" transformation is adopted, which is given by the following formula: For negative targets, the "smaller the better" transformation is adopted, which is given by the following formula: The comprehensive membership is obtained by linearly weighting the membership of the three targets: Where, M(f i ), i=1,2,3 is the fuzzy membership of the i-th target; f i,max and f i,min is the maximum and minimum value of the i-th target; is the weight of the i-th objective; F is the comprehensive membership of the three objectives, and the solution with the largest comprehensive membership among all solutions is taken as the optimal compromise solution.

Images