Virtual power plant control method based on average field game

Abstract

A virtual power plant control method based on average field game comprises the following steps: firstly, establishing a virtual power plant operation model which comprises a plurality of distributed resources, then, dividing the plurality of distributed resources in the virtual power plant operation model into a plurality of homogeneous clusters in a dynamic granulation mode, and then, establishing a plurality of distributed resources in the virtual power plant operation model; wherein each homogeneous cluster has a down-adjustable power range and an up-adjustable power range, and then the virtual power plant operation model obtains an average field term linear equation and a partial differential equation according to the down-adjustable power range and the up-adjustable power range, a standard cluster formed by a plurality of homogeneous clusters by using average field characteristics is defined according to an average field item linear equation and a partial differential equation, and then, the virtual power plant operation model performs interactive iteration by using the average field item linear equation and the partial differential equation to obtain an average field equilibrium point of the virtual power plant operation model, so that the average field equilibrium point of the virtual power plant operation model is obtained. And enabling each homogeneous cluster to operate by using an average field equilibrium point.

Description

TECHNICAL FIELD

[0001] The present application relates to a virtual power plant, in particular to a control method of a virtual power plant based on mean field game. BACKGROUND

[0002] Virtual power plant (VPP) is used to aggregate a large number of distributed energy resources (DERs), which can utilize potential flexibility to provide regulation or frequency modulation service instructions for the power grid, and is gradually applied. Since the distributed resources aggregated by the virtual power plant usually belong to different interest subjects respectively, and the income distribution within the virtual power plant involves the interaction between a large number of interest subjects, in addition, the distributed resources need to respond to the regulation instructions of the virtual power plant while pursuing their own maximum benefits, so the virtual power plant needs to quickly decompose the regulation instructions of the power grid. In this case, the game of the virtual power plant is used to describe the interaction process between a large number of distributed resources or between the distributed resources and the virtual power plant. However, when the number of distributed resources increases, the computational complexity of the game model of the virtual power plant will increase.

[0003] The traditional virtual power plant uses the framework of mean field game (MFG) to transform the non-cooperative game between a large number of distributed resources into the game between each resource and the overall resource, so as to achieve fair and efficient resource allocation. However, the traditional mean field game framework uses two partial differential equations of individual strategy and overall strategy for iteration, and the conditions and numerical stability required for analysis and solution of partial differential equations are not good, so the traditional virtual power plant is not easy to perform mean field game.

[0004] In addition, the traditional virtual power plant uses centralized solution, but due to the different regulation capabilities and willingness to participate between the distributed resources aggregated by the virtual power plant, the centralized solution causes its precision to be reduced and the speed to be slow, so that the interests of part of the distributed resources are sacrificed, and in the case of a large number of distributed resources, the virtual power plant needs to consume more calculation time, and cannot meet the requirements of the time interval of the power grid dispatching.

[0005] Therefore, it is necessary to develop a control method of a virtual power plant based on mean field game to solve the problems faced by the prior art. SUMMARY

[0006] The purpose of the present case is to provide a virtual power plant, wherein a virtual power plant operation model divides a plurality of distributed resources into a plurality of homogeneous clusters in a dynamic granulation manner, and a standard cluster formed by the average field characteristics is defined according to the average field term linear equation and the partial differential equation, and the interactive iteration is performed to obtain the average field equilibrium point of the virtual power plant operation model, so that each homogeneous cluster operates by using the average field equilibrium point. Therefore, the virtual power plant of the present case only uses a single partial differential equation and an average field term linear equation for game playing, and uses a dynamic granulation manner to distinguish a plurality of homogeneous clusters to introduce an information function, forms a mapping between distributed resources and conditional attributes, converts the division of physical entities into the operation of conditional attributes, improves the division accuracy between homogeneous clusters, and thus improves the calculation accuracy.

[0007] To achieve the above purpose, the present case provides a control method of a virtual power plant based on an average field game, wherein the control method comprises the following steps. First, a virtual power plant operation model is established, and the virtual power plant operation model comprises a plurality of distributed resources. Then, the plurality of distributed resources in the virtual power plant operation model are divided into a plurality of homogeneous clusters in a dynamic granulation manner, wherein each homogeneous cluster has a down-regulatable power range and an up-regulatable power range. Then, the virtual power plant operation model obtains an average field term linear equation and a partial differential equation according to the down-regulatable power range and the up-regulatable power range, and a standard cluster formed by the average field characteristics is defined according to the average field term linear equation and the partial differential equation. Then, the virtual power plant operation model performs interactive iteration by using the average field term linear equation and the partial differential equation to obtain the average field equilibrium point of the virtual power plant operation model, so that each homogeneous cluster operates by using the average field equilibrium point. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 The schematic diagram of the architecture of the virtual power plant of the present case is shown in

[0009] Figure 2 The control method of the virtual power plant shown in Figure 1 The flowchart of the control method shown in

[0010] Figure 3 The sub-step flowchart of step S3 of the control method shown in Figure 2 The valley filling power schematic diagram of different homogeneous clusters in the virtual power plant shown in

[0011] Figure 4A The peak clipping power schematic diagram of different homogeneous clusters in the virtual power plant shown in Figure 1 The valley filling power schematic diagram of different homogeneous clusters in the virtual power plant shown in

[0012] Figure 4B The peak clipping power schematic diagram of different homogeneous clusters in the virtual power plant shown in Figure 1 The peak clipping power schematic diagram of different homogeneous clusters in the virtual power plant shown in

[0013] Figure 5 The peak clipping power schematic diagram of different homogeneous clusters in the virtual power plant shown inFigure 1 Net benefit distribution of different homogeneous clusters in the virtual power plant shown.

[0014] Wherein, the reference numerals are explained as follows:

[0015] 1: Virtual power plant

[0016] 2: Virtual power plant operation model

[0017] 3: Distributed resource

[0018] A1, A2, A3, A4, A5, A c : Homogeneous cluster

[0019] A s : Standard cluster

[0020] S1-S4, M1-M7: Step DETAILED DESCRIPTION

[0021] Some typical embodiments embodying features and advantages of the present application will be described in detail in the following description. It should be understood that the present application can be varied in many different ways and that the description and drawings are to be taken as illustrative and not restrictive.

[0022] Please refer to Figure 1 , which is the architecture schematic diagram of the virtual power plant of the present application. As shown in the figure, the virtual power plant 1 of the present application has a virtual power plant operation model 2, wherein the virtual power plant operation model 2 contains an aggregated plurality of distributed resources 3, wherein the plurality of distributed resources 3 can include renewable energy sources, electric energy storage systems and temperature-controlled loads.

[0023] The virtual power plant operation model 2 and the plurality of distributed resources 3 have mutually signed contracts, and the contracts have operation characteristic information, such as operating power. The virtual power plant 1 uses the convex set representation of the operating power feasible region in the T-dimensional space of each distributed resource 3 in a time interval, as shown in equation (1),

[0024]

[0025] Wherein, is the operating power feasible region, is a preset coefficient matrix in the virtual power plant operation model 2, K is the number of distributed resources 3 constrained in a time interval in the scheduling time, is a preset constant matrix in the virtual power plant operation model 2, T is the number of time intervals in the scheduling time, is the power vector of the corresponding distributed resource 3, is the state of charge phasor of the corresponding distributed resource 3.

[0026] The virtual power plant operation model 2 further maps the operation power feasible region to the aggregated power feasible region by using the Minkowski sum and the Euclidean sum of point sets, as shown in equation (2),

[0027]

[0028] wherein Ω VPP is the aggregated power feasible region, is the Euclidean sum of point sets, n is the number of distributed resources 3, is the operation power of the corresponding distributed resource 3, P VPP is the operation power of the virtual power plant 1.

[0029] Since the operation model of part of the distributed resources 3 in the virtual power plant operation model 2 is time-varying, the flexibility of the aggregated power feasible region has time coupling, and in the spatial geometry, the aggregated power feasible region can be abstractly represented as a high-dimensional aggregated power polyhedron. In order to avoid the difficulty of obtaining the accurate operation power boundary, the virtual power plant operation model 2 of the virtual power plant 1 of the present case uses the embedded right-angle pyramid method to approximately solve the operation power boundary at different time periods, so as to realize the time decoupling of the aggregated power adjustment, as shown in equation (3),

[0030]

[0031] wherein t = 1, 2, …, T, is the operation power adjustable range of the virtual power plant 1 at t time under a given confidence level a, is the initial power of the virtual power plant 1 under a given confidence level a (i.e. the minimum value of the operation power), is the maximum value of the operation power of the virtual power plant 1 under a given confidence level a.

[0032] Next, the virtual power plant operation model 2 divides the plurality of distributed resources 3 into a plurality of homogeneous clusters in a dynamic granulation manner, and the plurality of distributed resources 3 in each homogeneous cluster have the same response characteristics. Specifically, the detailed information of the homogeneous cluster containing the plurality of distributed resources 3 can be defined as a four-dimensional information system, as shown in equation (4),

[0033] S = (U, D, V, f)... (4)

[0034] wherein U = {DER1, DER2, …, DER n} is the domain of the distributed resources 3 in the homogeneous cluster, D is the physical attribute set of the distributed resources 3 in the homogeneous cluster, is the value set of the distributed resources 3 in the homogeneous cluster, is the conditional attribute D mThe value of f: U×D→V is the information function of distributed resource 3 within the homogeneous cluster, representing the mapping relationship between each distributed resource 3 and its corresponding condition attribute. The virtual power plant operation model 2 in this case uses dynamic granularization to distinguish multiple homogeneous clusters to introduce information functions, forming a mapping between distributed resource 3 and condition attributes. This transforms the partitioning of physical entities into the operation of condition attributes, improving the partitioning accuracy between homogeneous clusters and thus improving the accuracy of subsequent calculations.

[0035] Take one of the multiple homogeneous clusters, cluster A. c For example (e.g.) Figure 1 The homogeneous cluster A1 is shown. c It has an adjustable power range and an adjustable power range, as shown in equation (5).

[0036]

[0037] Where D1 and D2 are homogeneous clusters A c Conditional attributes, For homogeneous cluster A c The adjustable power range of the polymerization power. For homogeneous cluster A c The adjustable power range of the polymerization power.

[0038] Combining equation (5) with equation (3), the set of values ​​V can be calculated as shown in equation (6).

[0039]

[0040] in, The overall aggregated power baseline for virtual power plant 1.

[0041] In addition, homogeneous cluster A c The set of values ​​for each conditional attribute in the function is shown in equation (7).

[0042]

[0043] in, This represents the upper limit of polymerization power. This represents the lower limit of polymerization power. For homogeneous cluster A c The baseline for polymerization power.

[0044] In the virtual power plant operation model 2, besides the selected homogeneous cluster A... c The remaining homogeneous clusters can be used to form a standard cluster A based on their mean-field properties. s Among them, the selected homogeneous cluster A c With standard cluster A shas an indistinguishable relationship, as shown in formula (8),

[0045]

[0046] wherein Euc[f(A c , D), f(A s , D)] is the Euclidean distance between the homogeneous cluster A c and the standard cluster A s .

[0047] According to the above description, the plurality of distributed resources 3 are divided into a homogeneous cluster A c , and the expression is as shown in formula (9),

[0048]

[0049] wherein E d , and are the charge amount, valley filling power and peak clipping power of the corresponding homogeneous cluster A c respectively; and are the upper limit of the valley filling power and the upper limit of the peak clipping power of the standard cluster A s respectively, and are the upper limit of the charge amount and the lower limit of the charge amount of the corresponding homogeneous cluster A c respectively.

[0050] In the real-time peak regulation market (PRM), the virtual power plant 1 distributes the calculation results to all homogeneous clusters, and at the same time, performs peak regulation according to the control of the virtual power plant 1, and in order to seek the best peak regulation instruction and economic benefit distribution, a game is formed between all homogeneous clusters. Since the total number of homogeneous clusters is large, the peak value of a single homogeneous cluster has little influence on other arbitrary homogeneous clusters, so the virtual power plant operation model 2 in the virtual power plant 1 of the present case uses the Mean Field Game (MFG) framework to game between an arbitrary single homogeneous cluster A c and a standard cluster A s formed by the average field characteristics of the remaining other homogeneous clusters. In the present case, the virtual power plant operation model 2 obtains the mean field term (MFT) linear equation and the partial differential equation according to the down-regulatable power range and the up-regulatable power range , and the standard cluster A s formed by the average field characteristics of the remaining plurality of homogeneous clusters except the selected homogeneous cluster A c .According to the definition of the mean-field term linear equation and the partial differential equation, the following will be further described for the mean-field term linear equation and the partial differential equation.

[0051] Firstly, the mean-field game model contains multiple parameters, such as the participant set (i.e. the number of homogeneous clusters at time t), the decision vector space (i.e. the joint control strategy of all homogeneous clusters), and the utility function (i.e. the preference degree of the homogeneous cluster in the game, also the net income), wherein the expression of the utility function is as formula (10),

[0052]

[0053] Wherein, u d is the single decision vector of a single homogeneous cluster A c , u -d is the joint decision vector of the remaining homogeneous clusters in the standard cluster A s except for the homogeneous cluster A c , and are the unit incentive price of the downward power and the unit incentive price of the upward power respectively, and are the actual downward power and the actual upward power of the corresponding homogeneous cluster A c at time t respectively, and are the downward power period and the upward power period respectively, is the deviation from the standard charge capacity, μ d is the penalty coefficient of the deviation from the standard charge capacity , and are the deviation from the standard downward power and the deviation from the standard upward power respectively, and are the penalty coefficients of the deviation from the standard downward power and the deviation from the standard upward power . The deviation from the standard charge capacity can be further calculated by the deviation from the standard downward power and the deviation from the standard upward power using formula (11), wherein the expression of the deviation from the standard downward power and the deviation from the standard upward power is as formula (11),

[0054]

[0055] And according to the adjustment capacity of the distributed resources in the homogeneous cluster A c and the market adjustment demand, a dynamic incentive mechanism is further formulated as formula (12),

[0056]

[0057] where h d and g d are piecewise linear functions describing the relationship between the adjustment amount and the incentive price, and are linear parameters reflecting the adjustment amount and the adjustment aggressiveness of the homogeneous cluster A c when it is up-regulated and down-regulated, respectively; ∏ vf and ∏ pf are the proportions of the total up-regulation and down-regulation incentives to the market clearing price, D(t) and are the load power and the average value of daily load at time period t, respectively.

[0058] In addition, the expression of the mean-field term in the mean-field game framework is as shown in equation (13),

[0059]

[0060] where y(t) is the impact parameter of the overall down-regulation power of the standard cluster A s on the corresponding homogeneous cluster A c , z(t) is the impact parameter of the overall up-regulation power of the standard cluster A s on the corresponding homogeneous cluster A c , N(t) is the number of the homogeneous clusters participating in the game at time t, is the actual down-regulation power of the corresponding homogeneous cluster A c at time t, is the actual up-regulation power of the corresponding homogeneous cluster A c at time t, where the actual down-regulation power of the corresponding homogeneous cluster A c at time t and the actual up-regulation power of the corresponding homogeneous cluster A c at time t need to satisfy the down-regulation power range and the up-regulation power range shown in equation (5), respectively, that is, the actual down-regulation power of the corresponding homogeneous cluster A c at time t needs to be located within the down-regulation power range , and the actual up-regulation power of the corresponding homogeneous cluster A c at time t needs to be located within the up-regulation power range .

[0061] When the mean-field term linear equation (13) is combined with the dynamic incentive mechanism (12), the relationship between the incentive price and the mean-field term can be represented as equation (14),

[0062]

[0063] Therefore, the utility function equation (10) can be further represented as equation (15),

[0064]

[0065] Further, the optimal control strategy can be defined using the mean-field term linear equation (13) as shown in equation (16),

[0066]

[0067] Next, the influence of the mean-field characteristic on each homogeneous cluster is shown using a partial differential equation, which can be but is not limited to the Hamilton-Jacobi-Bellman equation as shown in equation (17),

[0068]

[0069] where E d is the charge quantity of the corresponding homogeneous cluster A c h d and g d are piecewise linear functions of the corresponding homogeneous cluster A c with respect to the adjustment amount and the incentive price, and are the penalty coefficients for deviating from the standard down-regulation power and up-regulation power μ d is the penalty coefficient for deviating from the standard charge quantity v d is the net profit of the corresponding homogeneous cluster A c u d is the adjustment strategy of the corresponding homogeneous cluster A c .

[0070] And the optimal peak shaving control strategy of the virtual power plant operation model 2 can be obtained using the partial differential equation as shown in equation (18).

[0071]

[0072] The virtual power plant operation model 2 defines a single homogeneous cluster A c mapped to a standard cluster As control strategy, and the standard cluster A s mapped to a single homogeneous cluster A c control strategy, while the virtual power plant operation model 2 utilizes the mean field term linear equation and the partial differential equation to interactively iterate to obtain the mean field equilibrium point of the virtual power plant operation model 2, wherein the virtual power plant operation model 2 can utilize the Brouwer fixed point theorem and the contraction mapping theorem to confirm that the mean field term linear equation is an equation of approximate Nash equilibrium.

[0073] Please refer to Figure 2 , which is a flowchart of the control method of the virtual power plant shown in Figure 1 . As shown in the figure, first, step S1 is performed to establish a virtual power plant operation model 2, which includes a plurality of distributed resources. Next, step S2 is performed to divide the plurality of distributed resources in the virtual power plant operation model 2 into a plurality of homogeneous clusters in a dynamic granulation manner, wherein each homogeneous cluster has a down-regulatable power range and an up-regulatable power range. Next, step S3 is performed, and the virtual power plant operation model 2 obtains a mean field term linear equation and a partial differential equation according to the down-regulatable power range and the up-regulatable power range, and the plurality of homogeneous clusters are defined according to the standard cluster A s according to the mean field term linear equation and the partial differential equation. Next, step S4 is performed, and the virtual power plant operation model 2 utilizes the mean field term linear equation and the partial differential equation to interactively iterate to obtain the mean field equilibrium point of the virtual power plant operation model 2, so that each homogeneous cluster operates using the mean field equilibrium point.

[0074] As can be seen from the above, the virtual power plant operation model 2 of the virtual power plant 1 divides the plurality of distributed resources 3 into a plurality of homogeneous clusters in a dynamic granulation manner, and the plurality of homogeneous clusters are defined according to the standard cluster A s according to the mean field term linear equation and the partial differential equation, and are interactively iterated to obtain the mean field equilibrium point of the virtual power plant operation model 2, so that each homogeneous cluster operates using the mean field equilibrium point. Compared with the conventional virtual power plant which utilizes two partial differential equations to play a game, the virtual power plant 1 of the present case only utilizes a single partial differential equation and a mean field term linear equation to play a game, and utilizes a dynamic granulation manner to divide the plurality of homogeneous clusters to introduce an information function, to form a mapping between the distributed resources 3 and the conditional attributes, to convert the division of physical entities into the operation of conditional attributes, to improve the division accuracy between the homogeneous clusters, and thus to improve the calculation accuracy.

[0075] In this embodiment, since the number of homogeneous clusters in the virtual power plant operation model 2 is very large and very scattered, in order to improve the calculation efficiency, the mean field equilibrium point of the virtual power plant operation model 2 can be obtained by using the particle acceleration algorithm and the Mann iteration method to accelerate the iteration convergence, wherein the particle acceleration algorithm of the mean field equilibrium point is expressed as formula (19).

[0076]

[0077] The condition for iteration to converge to the mean field equilibrium point is shown in formula (20),

[0078] (||y (k+1) -y (k) || ∞ ≤δ)&&(||z (k+1) -z (k) || ∞ ≤δ)…(20)

[0079] wherein y (k) is the impact parameter of the overall down-regulation power of the standard cluster A s of this iteration on the corresponding homogeneous cluster A c , y (k+1) is the impact parameter of the overall down-regulation power of the standard cluster A s of the next iteration on the corresponding homogeneous cluster A c , ρ (k+1) is the coefficient sequence of the Mann iteration method, z (k) is the impact parameter of the overall up-regulation power of the standard cluster A s of this iteration on the corresponding homogeneous cluster A c , z (k+1 ) is the impact parameter of the overall up-regulation power of the standard cluster A s of the next iteration on the corresponding homogeneous cluster A c , and are an actual up-regulation power and an actual down-regulation power of the corresponding homogeneous cluster A c of the next iteration, N is the number of the plurality of homogeneous clusters participating in the game, and δ is the allowed error of the mean field term. In an embodiment, the coefficient sequence ρ (k+1) of the Mann iteration method can be e- 0.2k .

[0080] Please refer to Figure 3 , which is Figure 2The flow chart of the sub-step of step S3 of the control method. First, step Ml is performed to initialize the initial down-regulation power and the initial up-regulation power. Next, step M2 is performed to set the iteration order, the maximum iteration number and the iteration parameter, wherein the iteration parameter is the maximum convergence error between any two iterations, in other words, when the error between two iterations is less than or equal to the iteration parameter, then the iteration converges. Next, step M3 is performed to confirm whether the iteration order is less than or equal to the maximum iteration number. When the confirmation result of step M3 is yes, i.e. the iteration order is less than or equal to the maximum iteration number, step M4 is performed to calculate the actual up-regulation power and the actual down-regulation power of the next iteration by using the partial differential equation. Next, step M5 is performed to calculate the standard cluster A s of the whole down-regulation power of the corresponding homogeneous cluster A c of the first impact parameter and the whole up-regulation power of the corresponding homogeneous cluster A c of the second impact parameter. Next, step M6 is performed to confirm whether the first impact parameter of the next iteration minus the first impact parameter of the current iteration is less than or equal to the iteration parameter, and to confirm whether the second impact parameter of the next iteration minus the second impact parameter of the current iteration is less than or equal to the iteration parameter. When the confirmation result of step M6 is yes, i.e. the first impact parameter of the next iteration minus the first impact parameter of the current iteration is less than or equal to the iteration parameter, and the second impact parameter of the next iteration minus the second impact parameter of the current iteration is less than or equal to the iteration parameter, step M7 is performed to output the actual up-regulation power and the actual down-regulation power to adjust the corresponding homogeneous cluster A c . When the confirmation result of step M3 is no, i.e. the iteration order is greater than the maximum iteration number, step M7 is directly performed. When the confirmation result of step M6 is no, i.e. the first impact parameter of the next iteration minus the first impact parameter of the current iteration is greater than the iteration parameter, or the second impact parameter of the next iteration minus the second impact parameter of the current iteration is greater than the iteration parameter, then the next iteration is performed, and step M3 is re-performed.

[0081] According to the above description, the virtual power plant operation model 2 of the virtual power plant 1 of the present case decomposes the multi-body problem of a large number of distributed resources 3 into single-body problems for distributed calculation, and calculates the same by using the particle acceleration algorithm and the Mann iteration method to effectively accelerate the solution process of finding the game equilibrium point, and can achieve fair and reasonable distribution of benefits within the virtual power plant 1, and can effectively achieve the requirements of the power grid dispatch for time intervals.

[0082] Please refer to Figure 4A and Figure 4B , wherein Figure 4A is the valley filling power diagram of different homogeneous clusters in the virtual power plant shown in Figure 1 , and Figure 4B isFigure 1 Fig. 6 shows a diagram of the peak shaving power of different homogeneous clusters in the virtual power plant. In the diagram, according to the current peak shaving market, the peak shaving period is set to [10:00, 21:00], the valley filling period is set to [1:00, 7:00]∪[24:00, 24:45], and the linear parameters of the adjustment aggressiveness of the up-regulation and down-regulation of the homogeneous clusters are set to monotonically linearly increase with the number d of the different homogeneous clusters, respectively, i.e., the incentive price thereof. Meanwhile, the time interval is set to 15 minutes, and the total number of the homogeneous clusters is set to 100. As shown in Fig. 6, the growth trend of the incentive price and the peak shaving power is the same, i.e., the method of the mean field game proposed in the present case has incentive compatibility, and can well incentivize the homogeneous clusters to participate in peak shaving. When the incentive price rises, the homogeneous clusters will actively respond to the peak shaving instruction of the virtual power plant 1, and the incentive price will also increase with the improvement of the peak shaving performance until it reaches the upper limit. In the time dimension, the change trend of the peak shaving power and the incentive price of the different homogeneous clusters is similar, in other words, the virtual power plant 1 in the present case can effectively divide all the distributed resources into homogeneous clusters with similar external regulation characteristics in a dynamic granulation manner. Figure 4A 4B As shown in Fig. 6, the growth trend of the incentive price and the peak shaving power is the same, i.e., the method of the mean field game proposed in the present case has incentive compatibility, and can well incentivize the homogeneous clusters to participate in peak shaving. When the incentive price rises, the homogeneous clusters will actively respond to the peak shaving instruction of the virtual power plant 1, and the incentive price will also increase with the improvement of the peak shaving performance until it reaches the upper limit. In the time dimension, the change trend of the peak shaving power and the incentive price of the different homogeneous clusters is similar, in other words, the virtual power plant 1 in the present case can effectively divide all the distributed resources into homogeneous clusters with similar external regulation characteristics in a dynamic granulation manner.

[0083] Fig. 7 shows a diagram of the net income distribution of different homogeneous clusters in the virtual power plant. As shown in Fig. 7, the net income of the homogeneous clusters gradually increases with the increase of the number d of the different homogeneous clusters, and reaches the peak value of 4427.7 yuan when the number d of the homogeneous clusters is 72, at which time the incentive price of the homogeneous clusters for the valley filling demand is 0.034, the incentive price of the homogeneous clusters for the peak shaving demand is 0.052, and the net income is increased due to the increase of the peak shaving power and the corresponding incentive price with the increase of the number d of the different homogeneous clusters. When d∈[73, 100], the net income sharply decreases due to the limited income and peak shaving power that can be allocated by the virtual power plant 1. When the total output of the homogeneous clusters exceeds the peak shaving power, the homogeneous clusters will be economically punished, in other words, the penalty coefficient can effectively limit the response peak shaving power of the homogeneous clusters so as to not exceed the settlement peak shaving power of the virtual power plant 1. Figure 5 Figure 1 As shown in Fig. 7, the growth trend of the incentive price and the peak shaving power is the same, i.e., the method of the mean field game proposed in the present case has incentive compatibility, and can well incentivize the homogeneous clusters to participate in peak shaving. When the incentive price rises, the homogeneous clusters will actively respond to the peak shaving instruction of the virtual power plant 1, and the incentive price will also increase with the improvement of the peak shaving performance until it reaches the upper limit. In the time dimension, the change trend of the peak shaving power and the incentive price of the different homogeneous clusters is similar, in other words, the virtual power plant 1 in the present case can effectively divide all the distributed resources into homogeneous clusters with similar external regulation characteristics in a dynamic granulation manner.

[0084] ​​​The average calculation time of the mean field game and the traditional stochastic game under different numbers of homogeneous clusters is shown in Table 1. When the total number of homogeneous clusters is 100, the calculation time of the mean field game is 0.129% of the calculation time of the traditional stochastic game. As the total number of homogeneous clusters increases, the calculation time of the traditional stochastic game increases sharply, and when the total number of homogeneous clusters is 700, it exceeds 24 hours, which is no longer suitable for real-time income distribution of the electricity market. The calculation time of the mean field game slowly increases with the increase of the total number of homogeneous clusters, because the proposed distributed acceleration algorithm based on granular computing allows homogeneous clusters to calculate in parallel and speeds up the convergence.

[0085]

[0086]

[0087] In summary, the virtual power plant operation model of the virtual power plant divides multiple distributed resources into multiple homogeneous clusters in a dynamic granulation manner, and the standard cluster composed of the mean field characteristic is defined according to the mean field term linear equation and the partial differential equation, and iterates to obtain the mean field equilibrium point of the virtual power plant operation model, so that each homogeneous cluster operates using the mean field equilibrium point. Therefore, the virtual power plant only uses a single partial differential equation and a mean field term linear equation for game, and uses a dynamic granulation manner to distinguish multiple homogeneous clusters to introduce an information function, forming a mapping between distributed resources and conditional attributes, converting the division of physical entities into the operation of conditional attributes, improving the division accuracy between homogeneous clusters, thereby improving the calculation accuracy. In addition, the virtual power plant operation model of the virtual power plant decomposes the many-body problem of a large number of distributed resources 3 into a single-body problem for distributed calculation, and uses a granular acceleration algorithm and Mann iteration method to calculate and effectively accelerate the solution process of finding the game equilibrium point, and can achieve a fair and reasonable distribution of benefits within the virtual power plant, and can effectively meet the requirements of the power grid dispatch for time intervals.

Claims

1. A control method for a virtual power plant based on mean-field game theory, wherein the control method comprises: (a) Establish a virtual power plant operation model, which includes multiple distributed resources; (b) The multiple distributed resources in the virtual power plant operation model are dynamically granularized into multiple homogeneous clusters, each of which has a power range that can be adjusted down and a power range that can be adjusted up. (c) The virtual power plant operation model obtains a mean-field linear equation and a partial differential equation based on the adjustable power range and the adjustable power range. The standard cluster formed by the multiple homogeneous clusters using the mean-field characteristics is defined based on the mean-field linear equation and the partial differential equation. (d) The virtual power plant operation model uses the mean field linear equation and the partial differential equation to perform interactive iteration to obtain a mean field equilibrium point of the virtual power plant operation model, so that each homogeneous cluster can operate using the mean field equilibrium point.

2. The control method as described in claim 1, wherein in step (b), the representation of each homogeneous cluster is as follows: in, E d , and These are the corresponding charge, valley filling power, and peak shaving power of the homogeneous cluster, respectively. and These are the upper limits for valley filling power and peak shaving power for this standard cluster, respectively. and These represent the upper and lower limits of the charge capacity for the corresponding homogeneous cluster.

3. The control method as described in claim 1, wherein in step (c), the expression of the linear equation for the mean field term is as follows: in, For this partial differential equation, E d For the charge of each of these homogeneous clusters, h d and g d This is a piecewise linear function relating the adjustment amount and incentive price in the corresponding homogeneous cluster. To reduce power to avoid deviation from the standard, To increase power to deviate from the standard, and This refers to the penalty coefficients for reducing power and increasing power due to deviation from the standard. For deviations from the standard charge, μ is the penalty coefficient for such deviation, v d For the corresponding net profit of this homogeneous cluster, u d This refers to the adjustment strategy for the corresponding homogeneous cluster.

4. The control method as described in claim 3, wherein in step (c), the partial differential equation is expressed as follows: in, For this partial differential equation, E d For the charge of each of these homogeneous clusters, h d and g d This is a piecewise linear function relating the adjustment amount and incentive price in the corresponding homogeneous cluster. To reduce power to avoid deviation from the standard, To increase power to deviate from the standard, and This refers to the penalty coefficients for reducing power and increasing power due to deviation from the standard. For deviations from the standard charge, μ is the penalty coefficient for such deviation, v d For the corresponding net profit of this homogeneous cluster, u d This refers to the adjustment strategy for the corresponding homogeneous cluster.

5. The control method as described in claim 1, wherein in step (c), the virtual power plant operation model uses Brouwer's fixed point theorem and the compression mapping theorem to confirm that the linear equation of the mean field term is an equation with approximate Nash equilibrium.

6. The control method as described in claim 1, wherein in step (d), the mean field equilibrium point is obtained using a particle acceleration algorithm, wherein the particle acceleration algorithm expression for the mean field equilibrium point is as follows: in, y (k) y represents the first influence parameter of the standard cluster's overall power reduction in this iteration on the corresponding homogeneous cluster. (k+1) For the next iteration, the overall down-adjustment power of the standard cluster is a second influence parameter on the corresponding homogeneous cluster, ρ. (k+1) It is a coefficient sequence of the Mann iteration method, z (k) For the overall power increase of the standard cluster in this iteration, z represents a third influence parameter of the corresponding homogeneous cluster. (k+1) This is a fourth influence parameter for the overall power increase of the standard cluster in the next iteration on the corresponding homogeneous cluster. and For the next iteration, let N be the actual adjusted power and the actual decreased power of the homogeneous cluster, and let N be the number of the multiple homogeneous clusters participating in the game.

7. The control method as described in claim 6, wherein the coefficient sequence is e -0.2k .

8. The control method as claimed in claim 1, wherein step (c) further comprises the following steps: (c1) Initialize an initial down-regulation power and an initial up-regulation power; (c2) Set an iteration order, a maximum number of iterations, and an iteration parameter; (c3) Confirm whether the iteration order is less than or equal to the maximum number of iterations; (c4) If the confirmation result of step (c3) is yes, use the partial differential equation to calculate and obtain an actual adjustment power and an actual down-adjustment power for the next iteration; (c5) The mean-field linear equation is used to calculate a first influence parameter of the overall down-adjustment power of the standard cluster on the corresponding homogeneous cluster and a second influence parameter of the overall up-adjustment power on the corresponding homogeneous cluster for the next iteration. (c6) Confirm whether the first influence parameter of the next iteration minus the first influence parameter of the current iteration is less than or equal to the iteration parameter, and confirm whether the second influence parameter of the next iteration minus the second influence parameter of the current iteration is less than or equal to the iteration parameter; and (c7) If the confirmation result of step (c6) is yes, output the actual adjusted power and the actual decreased power to adjust the corresponding homogeneous cluster.

9. The control method as described in claim 8, wherein if the confirmation result of step (c3) is negative, step (c6) is executed directly; if the confirmation result of step (c6) is negative, the next iteration is executed, and step (c3) is re-executed.

10. The control method of claim 1, wherein the plurality of distributed resources within each homogeneous cluster have the same response characteristics.

Images