A pricing method for electric vehicle charging peak load regulation considering fast and slow charging load characteristics

Abstract

The present invention provides a peak-shaving pricing method for electric vehicle charging that takes into account the characteristics of fast and slow charging loads. First, the distribution of fast-charging loads and slow-charging loads of electric vehicles in each time period of a day is statistically obtained; then the load transfer rate is calculated, and a functional relationship between the change in electricity price and the load transfer rate is constructed; finally, with the goal of minimizing the peak-to-valley difference of the net load of the power grid, a deep reinforcement learning algorithm is used to solve and obtain the fast-charging price and slow-charging price for each time period. The present invention incorporates the differences in user consumption psychology into the price adjustment factors to construct a connection model between charging load transfer and electricity price, more realistically depicting the actual load transfer brought about by changes in electricity prices; at the same time, it fully considers the load constraints of different scenarios and different charging methods, so that fast-charging users can achieve local load transfer, slow-charging users can achieve large-scale load transfer, and minimize the peak-to-valley difference of the net load of the power grid.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart grids, and in particular relates to a peak-shaving pricing method for electric vehicle charging that takes into account fast and slow charging load characteristics. Background Art

[0002] With the rapid development of the global electric vehicle market in recent years, the EV industry has become a key symbol of energy transition and sustainable development. As the number and penetration of EVs continue to grow, the gap between peak and off-peak electricity demand will become even more pronounced, posing a significant challenge to the power grid. Therefore, it is crucial to optimize electricity price adjustment strategies to encourage EV charging to occur at off-peak times.

[0003] Existing research on charging guidance to help the power grid reduce peak loads and fill valleys has the following shortcomings: ① In terms of charging price response, it ignores the fact that charging behavior of vehicle owners in response to electricity prices is spontaneous, and the relationship between electricity prices and guided charging load is ambiguous and random; ② In terms of charging scenarios and charging methods, most existing research only considers a single charging mode, ignoring the combined impact of different charging methods on the total charging load; ③ In terms of simulating user charging behavior, it does not consider the characteristics of slow charging access to the grid and the constraints of fast charging delays. In summary, existing research on electric vehicle charging guidance is more or less constrained by ideal conditions and has a significant gap with reality. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides an electric vehicle charging peak-shaving pricing method that takes into account the fast and slow charging load characteristics, so as to solve the problem of guiding electric vehicle peak charging by regulating charging electricity prices.

[0005] The present invention achieves the above technical objectives through the following technical means.

[0006] A pricing method for electric vehicle charging peak load regulation considering the characteristics of fast and slow charging loads:

[0007] Step 1: Statistically obtain the distribution of electric vehicle fast charging load and slow charging load at each time period within a day;

[0008] Step 2: Calculate the load transfer rate and construct a functional relationship between the electricity price change and the load transfer rate;

[0009] Step three: Based on the functional relationship obtained in step two, an objective function is constructed with the goal of minimizing the peak-to-valley difference of the net load of the power grid, and a deep reinforcement learning algorithm is used to solve it to obtain the fast charging price and slow charging price for each time period.

[0010] Furthermore, in step 2, the load transfer rate is calculated using the gradient descent method, where the load transfer rate includes the load transfer rate μ from peak to valley pv , Load transfer rate from peak hours to normal hours μ pf , Load transfer rate from normal time to valley time μ fv ;

[0011] The error function is constructed as:

[0012]

[0013] Where J is the error, L(t) is the theoretical value of the electric vehicle charging load in period t after guidance, and L′(t) is the actual value of the electric vehicle charging load after guidance;

[0014] The gradient update strategy is:

[0015]

[0016] Where α is the learning rate.

[0017] Furthermore, in step 2, the logistic function is used to express the functional relationship between electricity price and load transfer rate:

[0018]

[0019] Where μ is the load transfer rate, Δλ is the change in electricity price, and a, b, c, and d are the coefficients of the logistic function.

[0020] Furthermore, in step 2, the electricity price under the same load transfer rate is divided into optimistic electricity price and pessimistic electricity price, and the electricity price change Δλ is updated as follows:

[0021]

[0022] Where Δλ min and Δλ max are the pessimistic and optimistic price changes under the load transfer rate μ, μ max is the maximum load transfer rate, and k is the optimistic response membership.

[0023] Furthermore, the objective function in step three is:

[0024]

[0025] Where L G (t) is the net load of the grid, including the electric vehicle charging load L(t) and other loads L ot (t).

[0026] Furthermore, step three includes the following constraints:

[0027] Constraint 1: Both fast charging and slow charging prices must be within their respective upper and lower limits:

[0028]

[0029] Where λ k is the fast charging price, λ k,min and λ k,max are the lower and upper limits of the fast charging price, respectively, m is the slow charging price, λ m,min and λ m,max are the lower and upper limits of the slow charging price respectively;

[0030] Constraint 2: The daily charging load value of electric vehicles after the guidance change is greater than or equal to the daily load value before the guidance change:

[0031]

[0032] Where L k (t) and L m (t) are respectively the fast charging load and slow charging load after guidance, L 0,k (t) and L 0,m (t) are the fast charge and slow charge loads before guidance, respectively;

[0033] Constraint three: The fast charging load and slow charging load are within their respective upper and lower limits:

[0034]

[0035] Where, and are the lower and upper limits of the fast charging load in period t, and are the lower and upper load limits of slow charging in period t respectively;

[0036] Constraint 4: The loss margin from electricity price adjustment must meet the economic quota for grid support:

[0037]

[0038]

[0039] Where Δλ k and Δλ m are the price adjustment ranges of fast charging and slow charging respectively, E is the reward for shifting peak power load, δ is the economic quota coefficient of grid support, ρ pv is the load transfer rate from peak to valley time, μ pf Load transfer rate from peak hours to normal hours, is the average charging load during the pre-peak period, and m is the unit reward obtained by reducing the peak power load.

[0040] Furthermore, the deep reinforcement learning algorithm is a DDPG algorithm.

[0041] Furthermore, in the DDPG algorithm, the state S is defined as the electric vehicle charging electricity price determined in each time period:

[0042]

[0043] Where λ k,1 to λ k,24 They represent the fast charging prices from the 1st to the 24th period of the day, λ m,1 to λ m,24 Respectively represent the slow charging prices from the 1st to the 24th period of the day;

[0044] The action space matrix A is defined as:

[0045]

[0046] Where ΔP k,1 to ΔP k,24 They represent the fast charging power change growth step from the 1st period to the 24th period of the day, ΔP m,1 to ΔP m,24 They represent the growth steps of slow charging power change from the 1st to the 24th period of a day.

[0047] Furthermore, in the DDPG algorithm, the agent adopts an ε-greedy strategy to obtain the next action:

[0048]

[0049] Where, ε and ε0∈[0,1], ε is a random number, ε0 is a fixed value; Q(S ′ ,ΔP) is in state S ′ Under the condition of charging power change ΔP, the expected value of reward R obtained; S ′ It represents the next state relative to the state S during the internal learning process of DDPG; ΔP′ represents the next step of charging power change relative to ΔP; η is the strategy; the agent adopts a greedy strategy to obtain the optimal strategy and uses the Bellman equation to update the strategy value function.

[0050] Furthermore, the operating environment of the DDPG algorithm is the relationship model between load transfer rate and electricity price, and the reward value is:

[0051] R(S, ΔP, S′)=-F

[0052] Where F is the objective function; the reward signal is scaled using the z-score normalization method so that the reward signal is scaled to a range of mean 0 and standard deviation 1. The reward is updated as follows:

[0053]

[0054] Where ω and σ are the mean and standard deviation of the reward signal, respectively.

[0055] The beneficial effects of the present invention are:

[0056] (1) The present invention provides a peak-shaving pricing method for electric vehicle charging that takes into account the characteristics of fast and slow charging loads. Compared with other pricing methods used to guide off-peak electricity consumption, the present invention incorporates the differences in user consumption psychology into the pricing factors, thereby constructing a connection model between charging load transfer and electricity prices, and more realistically depicting the actual load transfer brought about by changes in electricity prices.

[0057] (2) The present invention fully considers the load constraints of different scenarios and different charging methods, enabling fast-charging users to achieve local load transfer and slow-charging users to achieve large-scale load transfer, increasing the load value of the power grid at night and noon, reducing the peak-to-valley difference of the power grid's net load, and achieving stable and economical operation of the power grid.

[0058] (3) This invention utilizes deep reinforcement learning to solve the problem, improving the accuracy and efficiency of the calculations. The potential impact of electricity prices on electric vehicle charging behavior is considered, ensuring that the calculated electricity prices are consistent with actual conditions. Based on a simulation of guiding 1,000 electric vehicles, the comparison of fast-charging load and electricity price, slow-charging load and electricity price, and grid net load before and after guidance demonstrates the effectiveness of the proposed guidance method. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 The distribution diagram of fast charging load and slow charging load before guidance;

[0060] Figure 2 To provide a distribution diagram of electric vehicle charging load and grid net load before guidance;

[0061] Figure 3 This is a graph showing the reward value versus the number of iterations in DDPG;

[0062] Figure 4 To provide a distribution diagram of electric vehicle charging load and grid net load after guidance;

[0063] Figure 5 The distribution diagram of fast charging load and slow charging load after guidance;

[0064] Figure 6 This is the fast charging price distribution map before and after guidance;

[0065] Figure 7 The slow charging price distribution diagram before and after guidance. DETAILED DESCRIPTION

[0066] The following describes embodiments of the present invention in detail. Examples of the illustrated embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0067] 1. Technical Solution

[0068] This paper first considers the slow charging characteristics of electric vehicles and the delayed charging characteristics of fast charging, obtains the corresponding slow charging and fast charging constraints, and then predicts the future fast charging load distribution based on Monte Carlo simulation, forming the cumulative impact of different charging methods on the total charging load. Secondly, it introduces user consumption psychology and constructs a charging price response model. Finally, it establishes a target optimization model for reducing net load, and solves the target model using the DDPG algorithm to obtain the optimal charging electricity price pricing scheme. The specific details are as follows:

[0069] Step 1: Based on the statistical data of electric vehicle fast-charging characteristics and slow-charging characteristics, the Monte Carlo method is used to repeatedly sample the uncertain parameters of electric vehicle travel in urban areas to obtain the distribution of electric vehicle fast-charging load and slow-charging load at different time periods throughout the day.

[0070] Step 2: Calculate the load transfer rate based on the gradient descent method and introduce user consumption psychology to construct the charging load response function relationship. The details are as follows:

[0071] S21, guide users to shift their charging time from peak hours to normal hours, peak hours to valley hours, and normal hours to valley hours. The guided electric vehicle charging load is represented by the following function:

[0072]

[0073] Where L(t) is the electric vehicle charging load in period t after guidance (theoretical value), L0(t) is the electric vehicle charging load in period t before guidance (obtained from step 1), To guide the average charging load during the pre-peak period (for example, a day is divided into 24 periods, and the peak period occupies 3 periods, then is the total load of these three periods divided by 3), is the average charging load before booting, μ pv is the load transfer rate from peak to valley time, μ pf is the load transfer rate from peak time to normal time, μ fvis the load transfer rate from normal time to valley time, T p 、T f 、T v They represent peak hours, normal hours, and off-peak hours respectively.

[0074] S22, using a gradient descent method to dynamically find the parameters of the load transfer rate.

[0075] The error function is constructed as follows:

[0076]

[0077] Where J is the error, L′(t) is the actual value of the electric vehicle charging load after guidance;

[0078] The gradient update strategy is:

[0079]

[0080] Where α is the learning rate.

[0081] Through the above repeated iterative fitting, the expression of load transfer rate can be obtained.

[0082] S23 uses the Logistic function to express the relationship between the electricity price and the resulting user charging load transfer. The function expression is as follows:

[0083]

[0084] Where μ is the load transfer rate, which has the same meaning as μ pv 、μ pf 、μ fv For example, when solving the load transfer from peak to valley, μ pv Substituting μ into the above formula, and solving other cases in the same way, Δλ is the change in electricity price (in each time period), and a, b, c, and d are the variable coefficients of the logistic function, which are used to reflect the user's consumption psychology. The value of a reflects the range of change in the transfer rate, and the value of c reflects the sensitivity of different charging methods to electricity prices. The above expression of the relationship between electricity price and load transfer function can be transformed into:

[0085]

[0086] S24, the electricity price under the same load transfer rate is divided into optimistic electricity price and pessimistic electricity price. Specifically, under the electricity price response user charging transfer mechanism, since users follow the voluntary principle, the user's response behavior is ambiguous; assuming that the electricity price change of the load transfer rate μ is responded to by users with optimistic consumption psychology, it is called the optimistic electricity price change, and the electricity price change of the load transfer rate μ is responded to by users with pessimistic consumption psychology, it is called the pessimistic electricity price change. Then the actual electricity price change setting will be between the optimistic electricity price change and the pessimistic electricity price change. Based on this, the electricity price change Δλ is updated as follows:

[0087]

[0088] Where Δλ min and Δλ max are the pessimistic and optimistic price changes under the load transfer rate μ (i.e., the lower and upper limits of the price change), μ max is the maximum load transfer rate that can be brought about by the change in electricity price, and k is the optimistic response membership, which indicates the probability that the user meets the optimistic electricity price.

[0089] Step 3: Analyze the economic feasibility of peak load regulation of the power grid, limit the fluctuation of electricity prices, build a charging guidance model with the goal of minimizing the difference between net load peak and valley, and use deep reinforcement learning algorithm to solve it. The details are as follows:

[0090] S31. Establish an economic model for peak-load shifting to limit pricing margins. The direct economic benefits of peak-load shifting are reflected in the incentives for shifting peak loads. The calculation formula is as follows:

[0091]

[0092] Where m is the unit reward obtained by reducing peak power load.

[0093] S32, formulate an objective function with the goal of minimizing the peak-valley load difference after the electric vehicle charging load is connected to the grid. The objective function is as follows:

[0094]

[0095] Where L G (t) is the net load of the power grid, L G (t)=L(t)+L ot (t), including electric vehicle charging load L(t) and other loads L ot (t), other loads L ot (t) is a known quantity in the present invention and is obtained through statistical survey means.

[0096] S33, set constraints:

[0097] 1) Both fast charging and slow charging prices need to be within their respective upper and lower limits:

[0098]

[0099] Where λ k is the fast charging price, λ k,min and λ k,max are the lower and upper limits of the fast charging price, respectively, m is the slow charging price, λ m,min and λ m,max They are the lower and upper limits of the slow charging price respectively.

[0100] 2) The daily charging load value of electric vehicles after the guidance change is greater than or equal to the daily load value before the guidance change:

[0101]

[0102] Where L k (t) and L m (t) are respectively the fast charging load and slow charging load after guidance, and L k (t)+L m (t)=L(t),L 0,k (t) and L 0,m (t) are the fast charge and slow charge loads before guidance, and L 0,k (t)+L 0,m (t) = L0(t).

[0103] 3) The fast charging load and slow charging load are within their respective upper and lower limits:

[0104]

[0105] Where, and are the lower and upper limits of the fast charging load within the t period (the specific values ​​are set based on actual statistics), and They are the lower and upper load limits of slow charging within period t (specific values ​​are set based on actual statistics).

[0106] 4) The loss from the electricity price adjustment meets the economic quota of grid support:

[0107]

[0108] Where Δλ k and Δλ m They are the price adjustment range of fast charging price and slow charging price (adjusted electricity price - original electricity price), and their meanings are the same as Δλ in S23 or S24, that is, when solving the fast charging price, Δλ is used. kSubstitute Δλ into the slow charging price. m Substituting Δλ, E is the reward for shifting peak power load, obtained from S31, and δ is the economic quota coefficient of grid support.

[0109] S34, build a DDPG (Deep Deterministic Policy Gradient) model based on the above objective function and constraints, where the state S is defined as the electric vehicle charging electricity price determined in each time period:

[0110]

[0111] Where λ k,1 and λ m,1 They are the fast charging price and slow charging price for the first period of the day, and so on. k,24 and λ m,24 They are the fast charging price and slow charging price for the 24th period of the day respectively.

[0112] The action space matrix A is defined as follows:

[0113]

[0114] Where ΔP k,1 and ΔP m,1 They are the total power change growth step of fast charging and the total power change growth step of slow charging in the first period of the day, and so on, ΔP k,24 and ΔP m,24 These are the total power change increments for fast charging and slow charging during the 24th period of the day, respectively. Note: The charging load for a period is equal to the charging power multiplied by the duration of that period.

[0115] In DDPG, the agent adopts the ε-greedy strategy to obtain the next action, which is expressed as follows:

[0116]

[0117] Where, ε and ε0∈[0,1], ε is a random number, ε0 is a fixed value; Q(S ′ ,ΔP) is in state S ′ Under the condition of charging power change ΔP, the expected value of reward R obtained; S ′ It represents the next state relative to state S during the internal learning process of DDPG; ΔP has the same meaning as ΔP k,1 , ΔP m,1 The power change growth step is equal. For example, when solving the fast charging power change in the first period, ΔP k,1Substituting ΔP, ΔP′ represents the next power change step relative to the current ΔP; η is the strategy; the agent adopts a greedy strategy to obtain the optimal strategy and uses the Bellman equation to update the strategy value function.

[0118] Set the training environment and rewards. The operating environment is the relationship model between load transfer rate and electricity price. Set the reward value as follows:

[0119] R(S, ΔP, S′)=-F

[0120] Where F is the objective function in S32.

[0121] The reward signal is scaled using the z-score normalization method so that the reward signal is scaled to a range of mean 0 and standard deviation 1. The reward is updated as follows:

[0122]

[0123] Where ω and σ are the mean and standard deviation of the reward signal, respectively, which can be obtained by statistical calculation of historical reward signals.

[0124] 2. Application Testing

[0125] Assuming the total energy storage capacity of electric vehicle batteries is 100kW·h, the rated fast-charging power of charging stations is 60kW, the rated slow-charging power of private vehicles is 7kW, and the charging efficiency is 97%. Private cars account for 60%, online ride-hailing and taxis account for 23.9%, and other vehicles account for 16.1%. The fast and slow charging percentages of various types of vehicles are shown in Table 1 below:

[0126] Table 1: Fast and slow charging ratio

[0127] Model Fast charging slow charge private car 14.8％ 85.2％ Ride-hailing and taxis 24.9％ 75.1％ Other vehicles 41.1％ 58.9％

[0128] In the load transfer rate response electricity price model, the c value under the fast charging mode is 0.4, and the c value under the slow charging mode is 0.15. The settings of other user consumption psychological parameters are shown in Table 2:

[0129] Table 2: Consumer psychology parameters

[0130] a b d Optimistic electricity prices 1 0 0.1 Pessimistic electricity prices 1.04 -0.0036 0.1

[0131] Taking the total number of electric vehicles in the future as 229 million as an example, it is calculated that the charging load of electric vehicles accounts for about 25.95% of the total load of the power grid. Using the Monte Carlo sampling method, 1,000 electric vehicles were surveyed and sampled. The day was divided into 24 periods with an interval of 1 hour. The distribution of fast charging load and slow charging load was predicted as follows: Figure 1As shown in the figure. Among them, slow charging accounts for a large proportion and is concentrated in the period of 18:00-6:00; fast charging load is relatively uniform during the period. The total charging load and the net load of the power grid are shown in the figure. Figure 2 The peak values ​​of charging load and total grid load overlap, making the "peak upon peak" phenomenon of grid load more obvious and exacerbating the peak-to-valley difference of total grid load.

[0132] In this test, the designed DDPG training parameters are shown in Table 3 below:

[0133] Table 3: DDPG training parameters

[0134] parameter value Number of iterations 500 Sampling time <![CDATA[1*10 -4 ]]> Action learning rate <![CDATA[1*10 -3 ]]> Critical Learning Rate <![CDATA[1*10 -3 ]]> Discount Factor 0.9 Mini-batch size 64 Experience replay area length <![CDATA[1*10 6 ]]> Update coefficient <![CDATA[1*10 -3 ]]>

[0135] As the number of training times changes, the reward function value obtained is as follows Figure 3 As shown in the figure, the reward value converges when the number of iterations reaches about 2500, and the optimal solution is 19305.19.

[0136] Finally, the charging electricity price of each period of fast charging and slow charging is adjusted by the method of the present invention, and the total charging load of electric vehicles and the net load of the power grid are distributed as follows: Figure 4 As shown in the figure, the standard deviation of the grid's net load was 7955.87 before the guidance, but it dropped to 5966.90 after the guidance. The peak-to-valley difference in the grid's net load that day dropped from 25740.25 to 19305.19, a decrease of approximately 25%. The peak load before the guidance was 27712.13, but after the guidance, it dropped to 23766.23, a decrease of approximately 14.24%. After the guidance, electric vehicle charging loads were concentrated during off-peak hours, increasing grid load values ​​at night and midday. This demonstrates that guiding electric vehicle charging significantly reduces peak loads and fills valleys in the grid.

[0137] The distribution of electric vehicle fast charging and slow charging load after guidance is as follows Figure 5 The figure clearly shows that fast charging increased the load during the 12:00-14:00 period and reduced the load in surrounding periods, shifting the peak load of 1286.58 at 11:00 to 2414.97 at 13:00, achieving a localized shift in the fast-charging load. Slow charging reduced the daytime load, concentrating the load between 23:00 and 6:00 at night. The maximum slow-charging shift rate, reaching 92.51%, was achieved during the 4:00 period, achieving a large-scale shift in the slow-charging load.

[0138] The fast charging and slow charging price changes solved by the present invention are as follows: Figure 6 and Figure 7As shown. The highest electricity prices for both fast charging and slow charging occur during the 17:00-20:00 period, with the fast charging price reaching 1.85 yuan / (kW·h) and the slow charging price reaching 0.78 yuan / (kW·h); this period is the peak period for residential electricity consumption, and by raising the charging price, users are guided to stagger charging during this peak period. The lowest electricity prices for fast charging occur during the 0:00-5:00 and 12:00-14:00 periods, and the lowest electricity prices for slow charging occur during the 22:00-8:00 period. The above periods are all low-peak periods for residential electricity consumption. The lowest slow charging price reaches 0.21 yuan / (kW·h) during the 3:00 period, while the fast charging price during this period is 0.71 yuan / (kW·h). When the difference between fast charging and slow charging prices is large, guiding users to choose slow charging during this period is conducive to eliminating the long peaks and valleys of the net load of the power grid at night. During the 12:00-14:00 period, the fast charging price is only 9.86% higher than the slow charging price. During this period, guiding users to choose fast charging will help eliminate the short-term peaks and valleys in the net load of the power grid during the afternoon.

[0139] In the description of the present invention, it should be understood that the terms "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0140] The present invention is not limited to the above-mentioned embodiments. Any obvious improvement, replacement or modification that can be made by those skilled in the art without departing from the essence of the present invention shall fall within the scope of protection of the present invention.

Claims

1. A peak-shaving pricing method for electric vehicle charging that considers fast and slow charging load characteristics, characterized by: Step 1: Statistically obtain the distribution of electric vehicle fast charging load and slow charging load at each time period within a day; Step 2: Calculate the load transfer rate and construct a functional relationship between the change in electricity price and the load transfer rate; where: The load transfer rate is calculated using the gradient descent method, where the load transfer rate includes the load transfer rate from peak to valley μ pv , Load transfer rate from peak hours to normal hours μ pf , Load transfer rate from normal time to valley time μ fv ; The error function is constructed as: Where J is the error, L(t) is the theoretical value of the electric vehicle charging load in period t after guidance, and L′(t) is the actual value of the electric vehicle charging load after guidance; The gradient update strategy is: Where α is the learning rate; The logistic function is used to express the functional relationship between electricity price and load transfer rate: Where μ is the load transfer rate, Δλ is the change in electricity price, and a, b, c, and d are the coefficients of the logistic function; The electricity price under the same load transfer rate is divided into optimistic electricity price and pessimistic electricity price, and the electricity price change Δλ is updated as follows: Where Δλ min and Δλ max are the pessimistic and optimistic price changes under the load transfer rate μ, μ max is the maximum load transfer rate, k is the optimistic response membership; Step three: Based on the functional relationship obtained in step two, an objective function is constructed with the goal of minimizing the peak-to-valley difference of the net load of the power grid, and a deep reinforcement learning algorithm is used to solve it to obtain the fast charging price and slow charging price for each time period.

2. The electric vehicle charging peak load pricing method according to claim 1, characterized in that: The objective function in step three is: Where L G (t) is the net load of the grid, including the electric vehicle charging load L(t) and other loads L ot (t).

3. The electric vehicle charging peak load pricing method according to claim 2, characterized in that: The solution in step 3 also includes the following constraints: Constraint 1: Both fast charging and slow charging prices must be within their respective upper and lower limits: Where λ k is the fast charging price, λ k,min and λ k,max are the lower and upper limits of the fast charging price, respectively, m is the slow charging price, λ m,min and λ m,ma are the lower and upper limits of the slow charging price respectively; Constraint 2: The daily charging load value of electric vehicles after the guidance change is greater than or equal to the daily load value before the guidance change: Where L k (t) and L , (t) are respectively the fast charging load and slow charging load after guidance, L 0,k (t) and L 0,m (t) are the fast charge and slow charge loads before guidance, respectively; Constraint three: The fast charging load and slow charging load are within their respective upper and lower limits: Where, and are the lower and upper limits of the fast charging load in period t, and are the lower and upper load limits of slow charging in period t respectively; Constraint 4: The loss margin from electricity price adjustment must meet the economic quota for grid support: Where Δλ k and Δλ m are the price adjustment ranges of fast charging and slow charging respectively, E is the reward for shifting peak power load, δ is the economic quota coefficient of grid support, μ pv is the load transfer rate from peak to valley time, μ pf Load transfer rate from peak hours to normal hours, is the average charging load during the pre-peak period, and m is the unit reward obtained by reducing the peak power load.

4. The electric vehicle charging peak load pricing method according to claim 3, characterized in that: The deep reinforcement learning algorithm is the DDPG algorithm.

5. The electric vehicle charging peak load pricing method according to claim 4, characterized in that: In the DDPG algorithm, the state S is defined as the electric vehicle charging electricity price determined in each period: Where λ k,1 to λ k,24 They represent the fast charging prices from the 1st to the 24th period of the day, λ m,1 to λ m,24 Respectively represent the slow charging prices from the 1st to the 24th period of the day; The action space matrix A is defined as: Where ΔP k,1 to ΔP k,24 They represent the fast charging power change growth step from the 1st period to the 24th period of the day, ΔP m,1 to ΔP m,24 They represent the growth steps of slow charging power change from the 1st to the 24th period of a day.

6. The electric vehicle charging peak load pricing method according to claim 5, characterized in that: In the DDPG algorithm, the agent adopts the ε-greedy strategy to obtain the next action: Where, ε and ε0∈[0,1], ε is a random number, ε0 is a fixed value; Q(S ′ ,ΔP) is in state S ′ Under the condition of charging power change ΔP, the expected value of reward R obtained; S ′ It represents the next state relative to the state S during the internal learning process of DDPG; ΔP′ represents the next step of charging power change relative to ΔP; η is the strategy; the agent adopts a greedy strategy to obtain the optimal strategy and uses the Bellman equation to update the strategy value function.

7. The electric vehicle charging peak load pricing method according to claim 6, characterized in that: The operating environment of the DDPG algorithm is the relationship model between load transfer rate and electricity price, and the reward value is: R(S, ΔP, S′)=-F Where F is the objective function; the reward signal is scaled using the z-score normalization method so that the reward signal is scaled to a range of mean 0 and standard deviation 1. The reward is updated as follows: Where ω and σ are the mean and standard deviation of the reward signal, respectively.

Images