Electric vehicle aggregation frequency modulation virtual power plant average field game regulation method and device

Abstract

The application relates to an electric vehicle aggregation frequency modulation virtual power plant average field game regulation method and device, which comprises the following steps: obtaining historical charging data, real-time charging data and space-time relations of electric vehicle charging piles; the historical charging data and the space-time relations of the charging piles are analyzed by using a multi-graph fusion neural network model to obtain a power baseline prediction value; the adjustable power is rolling evaluated and aggregated according to the real-time charging information of the electric vehicles; the electric vehicles are divided into combinations with the same external characteristics, and an average field game method is used to quickly decompose the frequency modulation instructions and economic benefits to different electric vehicle combinations. The method builds a multi-graph fusion neural network considering the space-time relations between the electric vehicle charging piles to predict the charging power baseline of large-scale electric vehicles, and combines a rolling evaluation model to describe the up and down adjustment ability of the virtual power plant of the aggregated large-scale electric vehicles to provide frequency modulation services near the charging power baseline, so that the power data accuracy is improved.

Description

Technical Field

[0001] The present application relates to the field of new energy power supply technology, and specifically to a virtual power plant average field game control method and device for tram aggregate frequency regulation. Background Art

[0002] With the widespread adoption of renewable energy in power systems, the uncertainty of renewable generation exacerbates the imbalance between power generation and load, posing significant challenges to power system frequency regulation. Trams are considered an important means of providing frequency regulation services in a distributed manner and a potential high-quality frequency regulation resource, offering fast response and large battery storage capacity. However, the varying response characteristics of trams due to their random driving behavior, unique driving preferences, and distinctive charging requirements make it difficult to aggregate trams and calculate their operating baselines and adjustability for frequency regulation.

[0003] Currently, virtual power plants (VPPs) are widely used to combine small-capacity, geographically dispersed distributed energy resources with traditional power plants to provide ancillary services to the power system. VPPs are therefore used to regulate the collective storage capacity and controllability of electric vehicles to maintain power balance and frequency stability, transforming potential charging burdens into power system frequency regulation resources. However, due to the high uncertainty and randomness of electric vehicle charging demand and heterogeneous response characteristics, VPPs struggle to accurately assess the total controllable capacity of electric vehicles. Furthermore, the large number of electric vehicles aggregated by a VPP often have different ownership rights, inevitably leading to conflicts of interest when the VPP decodes frequency regulation signals and distributes related benefits. Artificial intelligence (AI) methods, such as long short-term memory (LSTM) neural networks, are currently being used to assess the aggregate regulation capacity of electric vehicles. However, these AI-based assessment methods primarily focus on capturing the temporal features hidden in the electric vehicle charging demand sequence data while ignoring spatial features, resulting in inaccurate assessment results. Furthermore, the computational complexity increases exponentially with the number of participating electric vehicles, potentially even encountering the "curse of dimensionality" problem. Summary of the Invention

[0004] To overcome the above-mentioned deficiencies of the prior art, the present application provides a virtual power plant average field game control method and device for tram aggregate frequency regulation, which specifically adopts the following technical solutions:

[0005] A virtual power plant mean field game control method for tram aggregate frequency regulation comprises the following steps:

[0006] Obtain historical charging data, real-time charging data, and the spatiotemporal relationship of electric vehicle charging piles;

[0007] A multi-graph fusion neural network model is used to analyze the historical charging data and spatiotemporal relationships of electric vehicle charging piles to obtain power baseline prediction values ​​for different electric vehicle charging piles.

[0008] Based on the real-time charging information of trams, a rolling evaluation of the adjustable power of different tram charging piles is performed. The Minkowski sum is used to aggregate the adjustable power of different tram charging piles to obtain the adjustable power range of the virtual power plant.

[0009] All trams are divided into tram combinations with the same external characteristics. The mean field game method is used to quickly decompose the frequency regulation instructions and economic benefits into different tram combinations, and the optimal frequency regulation control strategy for the corresponding tram combinations is obtained:

[0010] To determine the optimal frequency regulation strategy and economic benefit distribution, a mean field game model is constructed between tram combinations; and to encourage trams to actively respond to frequency regulation demands, a dynamic incentive mechanism is constructed.

[0011] Based on the collective effect of the mean field on the control strategy of each tram combination, a mean field term including upward adjustment of power command and downward adjustment of power command is constructed;

[0012] The mean field term is combined with the dynamic incentive mechanism to obtain the relationship between the incentive price and the mean field term corresponding to the upward and downward adjustment of the power instruction.

[0013] Based on the relationship between incentive price and mean field term and mean field game model, the optimal frequency modulation control strategy considering mean field term is obtained;

[0014] The corresponding Hamilton-Jacobi-Bellman equation is obtained based on the optimal frequency modulation control strategy considering the mean field term;

[0015] By iterating the mean field term and the Hamilton-Jacobi-Bellman equation until the mean field equilibrium state is reached, the optimal frequency regulation control strategy obtained at this time is sent to the tram charging piles corresponding to the tram combination.

[0016] Optional: The step of analyzing the historical charging data of the electric vehicle charging pile and the spatiotemporal relationship of the charging piles by using the multi-graph fusion neural network model includes:

[0017] First, a multi-graph fusion neural network model is constructed and a network topology graph and a feature matrix of the electric vehicle charging pile are obtained; wherein the multi-graph fusion neural network model comprises at least a multi-graph fusion convolutional layer, a gated recurrent unit loop layer, and a fully connected layer;

[0018] Based on the network topology of the electric vehicle charging piles and the electric vehicle charging modes of different nodes, a charging demand similarity graph of the corresponding nodes is constructed;

[0019] Based on the network topology of the electric vehicle charging piles and the spatial proximity of different nodes, a geographical adjacency graph of the corresponding nodes is constructed;

[0020] The charging demand similarity graph and geographic adjacency graph of the same node are respectively input into the multi-graph fusion convolutional layer of the multi-graph fusion neural network model to obtain the output results of the corresponding nodes;

[0021] The output results of the charging demand similarity graph and the geographic adjacency graph are weighted and fused to obtain the fused output result;

[0022] The fused output results are fed into the gated recurrent unit loop layer for analytical calculation to minimize the difference between the actual charging demand and the predicted demand of each node;

[0023] Finally, the power baseline prediction values ​​of the electric vehicle charging piles at different nodes are output through the fully connected layer.

[0024] Optionally, the step of performing a rolling evaluation of the adjustable power of different electric vehicle charging piles according to the real-time charging information of the electric vehicle includes:

[0025] Based on the power baseline prediction values ​​of different electric vehicle charging piles, a rolling assessment framework model and a virtual energy storage model for the corresponding electric vehicle are constructed respectively;

[0026] Based on the state of charge, off-grid time, minimum off-grid charging level and rated charging power of the tram's virtual energy storage model in a single assessment period, the tram is classified into trams that remain on the grid and trams that will go off the grid;

[0027] Based on different categories of trams, the adjustable total power range of the corresponding trams within the corresponding evaluation time is calculated respectively:

[0028] The adjustable power range for the trams that remain connected to the grid during the evaluation time is:

[0029] ;

[0030] The adjustable power range for the electric vehicles that will be off-grid during the evaluation time is:

[0031] ;

[0032] in The maximum value of the total adjustable power of the trams that remain connected to the grid during the evaluation period; The minimum adjustable total power of the tram that remains connected to the grid during the evaluation period; is the maximum adjustable total power of the tram that will be off-grid during the evaluation time; is the minimum adjustable total power of the trams that will be off-grid during the evaluation period; n is the number of trams that remain on-grid during the evaluation period; m is the number of trams that will be off-grid during the evaluation period; is the upper limit of the state of charge of the electric vehicle i that remains connected to the grid during the evaluation period; is the lower limit of the state of charge of the electric vehicle i that remains connected to the grid during the evaluation period; is the upper limit of the state of charge of the electric vehicle j that will be off the grid during the evaluation period; The lower limit of the state of charge of the off-grid vehicle j in period t during the evaluation time; To maintain the maximum charging power of the grid-connected electric vehicle in period t during the evaluation period; To maintain the minimum charging power of the grid-connected electric vehicle in period t during the evaluation period; is the maximum charging power of the off-grid electric vehicle in period t during the evaluation period; is the minimum charging power of the off-grid electric vehicle in period t during the evaluation period; is the upper limit of the charging power of the electric vehicle i that remains connected to the grid during the evaluation period in period t; is the lower limit of the charging power of the electric vehicle i that remains connected to the grid during the evaluation period in period t; is the upper limit of the charging power of the electric vehicle j that will be off-grid during the evaluation period; is the lower limit of the charging power of the electric vehicle j that will be off-grid during the evaluation time period t.

[0033] Optional: The step of adopting Minkowski and aggregating the adjustable powers of different electric vehicle charging piles includes:

[0034] The adjustable power range of the virtual power plant is obtained by aggregating the adjustable powers of different types of trams using Minkowski sum;

[0035] The embedded hypercube approximation method is used to reduce the dimension and time decouple the adjustable power range of the virtual power plant obtained by aggregation to obtain the adjustable power range of the virtual power plant.

[0036] Optionally, the step of dividing all trams into tram combinations having the same external characteristics includes:

[0037] According to the frequency regulation performance of the trams, all trams are randomly divided into a number of tram combinations with the same external characteristics according to the preset number of single tram combinations;

[0038] The Euclidean distance between the combinations is calculated based on the adjustable power range of the combination and the adjustable power range of a standard combination of trams with the same external characteristics;

[0039] When the Euclidean distance between the tram combination and the standard tram combination is less than the set value, the corresponding tram combination is output; otherwise, the trams are randomly recombined and the Euclidean distance between the combinations is repeatedly calculated until the Euclidean distance between the output tram combination and the standard tram combination is less than the set value, that is, they have the same external characteristics.

[0040] Optional: The mean field game model is:

[0041] ;

[0042] in represents all the trolley combinations participating at time t; represents the joint frequency regulation command strategy of all tram combinations, where the frequency regulation command strategy of the tram combination in period t includes upward power regulation command and downward power regulation command; Indicates the battery charge state of all tram combinations; is the utility function, which represents the satisfaction of the trolley combination participating in the mean field game.

[0043] Optional: The dynamic incentive mechanism includes:

[0044] ;

[0045] in is the incentive price for upward adjustment of power command during period t; is the incentive price for downward adjustment of power command during period t; is a piecewise linear incentive price function for upward regulation of power command; A piecewise linear incentive price function for downwardly regulating power commands; and are the positive parameters of the linear pricing strategy that adjusts the power command upward; and are the positive parameters of the linear pricing strategy that adjusts the power command downward; is the power load during period t; is the average daily power load before frequency regulation during period t; The ratio of the incentive price to the clearing price for upward adjustment of power orders; The ratio of the incentive price to the clearing price for downward adjustment of power orders; To adjust the clearing price of power orders upwards; is the clearing price for adjusting power orders downward.

[0046] Optional: The mean-field term is expressed as:

[0047] ;

[0048] in is the mean-field term for adjusting the power command upward; is the mean field term for downwardly regulating the power command; represents the number of trolley combinations participating at time t; represents all the trolley combinations participating at time t; Adjust the power command upward for period t; The power command is adjusted downward for the period t.

[0049] Optional: The step of obtaining the optimal frequency modulation control strategy considering the mean field term based on the relationship between the incentive price and the mean field term and the mean field game model includes:

[0050] The relationship between the incentive price and the mean field term is:

[0051] ;

[0052] in is the incentive price for upward adjustment of power command during period t; is the incentive price for downward adjustment of power command during period t; is a piecewise linear incentive price function for upward regulation of power command; A piecewise linear incentive price function for downwardly regulating power commands;

[0053] Substitute the relationship between the incentive price and the mean field term into the utility function of the mean field game model:

[0054] ;

[0055] in is the joint frequency control strategy of all tram combinations; y and z are sequence variables; An adjustment period for adjusting the power command upward; An adjustment period for adjusting the power command downward; Adjust power downward for deviations from expectations The penalty factor; Adjust power upwards for deviations from expectations The penalty factor; Deviating from the desired state of charge The penalty factor; is the state of charge during period t.

[0056] Furthermore, the present application also discloses a virtual power plant average field game control device for tram aggregate frequency regulation, the device comprising:

[0057] Parameter acquisition module, used to obtain historical charging data, real-time charging data and the spatiotemporal relationship of charging piles for electric vehicle charging piles;

[0058] The baseline prediction module is used to analyze the historical charging data and the spatiotemporal relationship of electric vehicle charging piles using a multi-graph fusion neural network model to obtain the power baseline prediction values ​​of different electric vehicle charging piles;

[0059] The power aggregation module is used to perform a rolling evaluation of the adjustable power of different electric vehicle charging piles based on the real-time charging information of the electric vehicles, and aggregate the adjustable power of different electric vehicle charging piles using Minkowski sum to obtain the adjustable power range of the virtual power plant;

[0060] The control output module is used to divide all trams into homogeneous clusters with the same external characteristics. It uses the mean field game method to quickly decompose the frequency regulation instructions and economic benefits into the homogeneous clusters of trams to obtain the optimal frequency regulation control strategy:

[0061] To determine the optimal frequency regulation strategy and economic benefit distribution, a mean field game model is constructed between tram combinations; and to encourage trams to actively respond to frequency regulation demands, a dynamic incentive mechanism is constructed.

[0062] Based on the collective effect of the mean field on the control strategy of each tram combination, a mean field term including upward adjustment of power command and downward adjustment of power command is constructed;

[0063] The mean field term is combined with the dynamic incentive mechanism to obtain the relationship between the incentive price and the mean field term corresponding to the upward and downward adjustment of the power instruction.

[0064] Based on the relationship between incentive price and mean field term and mean field game model, the optimal frequency modulation control strategy considering mean field term is obtained;

[0065] The corresponding Hamilton-Jacobi-Bellman equation is obtained based on the optimal frequency modulation control strategy considering the mean field term;

[0066] By iterating the mean field term and the Hamilton-Jacobi-Bellman equation until the mean field equilibrium state is reached, the optimal frequency control strategy obtained at this time is sent to the corresponding tram combination.

[0067] Beneficial effects

[0068] The technical solution of this application has the following beneficial effects:

[0069] (1) The virtual power plant mean field game control method of this application predicts the charging power baseline of large-scale electric vehicles by constructing a multi-graph fusion neural network (MGCN) that considers the spatiotemporal relationship between electric vehicle charging piles, and combines the aggregated virtual energy storage rolling evaluation model to describe the up and down adjustment capability of the virtual power plant of the aggregated large-scale electric vehicles to provide frequency regulation services near the charging power baseline. The prediction using the multi-graph fusion neural network fully considers the spatiotemporal relationship between electric vehicle charging piles, and has the advantages of high reliability compared to traditional prediction methods. In addition, the aggregated virtual energy storage rolling evaluation model has the characteristics of real-time correction compared to previous static evaluation methods, making the final power data of the aggregated large-scale electric vehicle VPP that can participate in frequency regulation more accurate.

[0070] (2) The virtual power plant mean field game control method of this application effectively decomposes the secondary frequency regulation signals between a large number of trams through the mean field game strategy and realizes the income distribution of frequency regulation services. This method transforms the traditional pairwise game into an interactive mean field game between individual trams and the mean field, breaking through the dimensionality curse problem caused by the increase in the number of game participants. In addition, the mean field game method of this method can greatly reduce the calculation time. The frequency regulation market needs to complete instructions quickly, which better meets the requirements of the frequency regulation market. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 This is a flow chart of the average field game control method for a virtual power plant in an embodiment of the present application.

[0072] Figure 2 This is a structural diagram of the gated recurrent unit recurrent layer in the multi-graph fusion neural network model in an embodiment of the present application.

[0073] Figure 3 This is a rolling evaluation framework diagram of the adjustable capacity of a virtual power plant that aggregates large-scale electric vehicles to participate in frequency regulation in an embodiment of the present application.

[0074] Figure 4 This is a diagram of the game process of VPPs participating in frequency modulation based on the mean field game for aggregating large-scale trams in an embodiment of the present application.

[0075] Figure 5 This is a structural diagram of the virtual power plant average field game control device in an embodiment of the present application.

[0076] Figure 6 This is a structural diagram of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0077] The present application will be further described below in conjunction with the accompanying drawings. The following examples are only used to more clearly illustrate the technical solutions of the present application and are not intended to limit the scope of protection of the present application. It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application.

[0078] Mean field games (MFGs) can be used to analyze the complex interactions between large numbers of homogeneous individuals and the entire population for resource allocation and energy scheduling. MFGs transform traditional pairwise interactive game methods into interactive games between individuals and a mean field, addressing the "curse of dimensionality." This application employs a mean field game approach to address large-scale tram participation in the frequency regulation market, resolving the issues of frequency regulation instructions and revenue distribution, enabling faster decomposition of frequency regulation instructions and rapid distribution of frequency regulation revenue.

[0079] Combine Figure 1As shown, the embodiment of the present application specifically discloses a virtual power plant average field game control method for tram aggregate frequency regulation, which includes the following steps:

[0080] S1: Obtain historical charging data, real-time charging data, and spatiotemporal relationships of electric vehicle charging piles;

[0081] S2: Use the multi-graph fusion neural network model to analyze the historical charging data and spatiotemporal relationship of electric vehicle charging piles to obtain the power baseline prediction value of different electric vehicle charging piles.

[0082] Specifically, the specific process of using the multi-graph fusion neural network model in step S2 to analyze the historical charging data of the electric vehicle charging pile and the spatiotemporal relationship of the charging pile is as follows:

[0083] (1) First, a multi-graph fusion neural network model is constructed and the network topology and feature matrix of the electric vehicle charging pile are obtained. The multi-graph fusion neural network model includes at least a multi-graph fusion convolutional layer, a gated recurrent unit recurrent layer, and a fully connected layer. The multi-graph fusion convolutional layer (GCN) is used to mine the spatial features hidden in the graph information and historical charging data. The historical charging data with spatial features is then input into the gated recurrent unit recurrent layer (GRU), which is used to capture the dynamic time characteristics of the time series through information transmission between adjacent units. The fully connected layer is used to output the power baseline prediction value of the electric vehicle charging demand.

[0084] In this embodiment, an unweighted graph is constructed To describe the network topology of electric vehicle charging piles. is the node graph of the charging pile, E is the network edge set of the charging pile, and N is the number of nodes. It is used to represent the connection relationship between electric vehicle charging piles. In addition, the characteristic matrix It is also used as the input data of the multi-graph fusion neural network model, and B is the number of features. As the input data at time t, the nonlinear mapping function f is formulated and adjusted in the learning process of the multi-graph fusion neural network model to obtain the following charging prediction sequence value:

[0085] ;

[0086] In the formula n is the length of the historical time series, R is the forecast length of the time series.

[0087] (2) Since some nodes in the network topology have similar electric vehicle charging modes, the network topology of the electric vehicle charging piles is combined with the electric vehicle charging modes of different nodes to construct the charging demand similarity graph of the corresponding nodes. , and calculate the time series similarity by dynamic time warping. Specifically, set two sequence elements and (n and m are the number of nodes of the sequence element respectively). The similarity between two sequence elements can be obtained by finding the path with the smallest sum of distances of all matching points in the distance matrix M:

[0088] ;

[0089] ;

[0090] ;

[0091] In the above formula is an element of the distance matrix M; and are the data of steps i and j in sequences y and z respectively; represents the distance between sequences Y and Z in steps i and j; is the final distance of the path obtained by repeatedly calculating the distance between sequences Y and Z. In general, The smaller the value, the higher the similarity.

[0092] The elements of the adjacent matrix can be expressed as:

[0093] ;

[0094] In the above formula and From the node and Time data series; express and the distance between them; is the maximum distance; It is a restriction The threshold parameter for size.

[0095] (3) Based on the network topology of the electric vehicle charging piles and the spatial proximity of different nodes, a geographic adjacency graph of the corresponding nodes is constructed. The adjacency matrix can be constructed based on the spatial proximity of the nodes. The elements of can be represented as follows:

[0096] ;

[0097] In the above formula, d is the node and The distance between s and are the thresholds of graph connectivity, respectively.

[0098] (4) The charging demand similarity graph and geographic adjacency graph of the same node are respectively input into the multi-graph fusion convolution layer of the multi-graph fusion neural network model to obtain the output results of the corresponding nodes. Specifically, the filter in the Fourier domain is constructed to capture the features between nodes by stacking multiple convolution layers as follows:

[0099] ;

[0100] In the above formula and Respectively l The output and input data matrices of the layer; is the additive self-connection matrix, is the identity matrix; yes degree matrix of ; Is a trainable weight matrix; represents a non-linear activation function.

[0101] (5) The output results of the charging demand similarity graph and the geographic adjacency graph are weighted and fused to obtain the fused output result. Specifically, the calculation results of the charging demand similarity graph and the geographic adjacency graph in the future and The fusion is as follows:

[0102] ;

[0103] In the above formula It is the fusion result used to support full network connectivity; and The output results are and The trainable weight parameters of .

[0104] (6) The fused output results are input into the gated recurrent unit loop layer for analytical calculation to minimize the difference between the actual charging demand and the predicted demand of each node; the gated recurrent unit loop layer (GRU) of this embodiment can retain the historical information of the tram change trend while capturing the time dependency of the current moment charging information. The specific internal structure of the GRU layer is as follows Figure 2 shown.

[0105] Through iterative training of the gated recurrent unit cycle layer, the actual charging demand can be and forecast demand The difference between loss Minimize:

[0106] ;

[0107] ;

[0108] In the above formula, k is the number of large-scale electric vehicle charging piles; Forecasting demand for virtual power plants.

[0109] (7) Finally, the power baseline prediction values ​​of the electric vehicle charging piles at different nodes are output through the fully connected layer.

[0110] S3: Perform a rolling evaluation of the adjustable power of different electric vehicle charging piles based on the real-time charging information of the electric vehicle, and use the Minkowski sum to aggregate the adjustable power of different electric vehicle charging piles to obtain the adjustable power range of the virtual power plant.

[0111] Specifically, the specific process of performing a rolling evaluation of the adjustable power of different electric vehicle charging piles according to the real-time charging information of the electric vehicle in step S3 is as follows:

[0112] (1) Based on the power baseline prediction values ​​of different electric vehicle charging piles, a rolling evaluation framework model and a virtual energy storage model of the corresponding electric vehicle are constructed.

[0113] In view of the fact that the time of the tram's disconnection and connection to the grid is arbitrary, and the grid frequency regulation requires high accuracy and real-time clearing of frequency regulation results, it is necessary to conduct a rolling assessment of the tram's adjustable capacity every day and make continuous corrections. In this embodiment, according to the requirements of grid frequency regulation, the assessment cycle is set to be once every 5 minutes, and each assessment lasts for the next hour. The specific rolling assessment framework is as follows: Figure 3 As shown in the figure is the state of charge (SOC) value expected by the tram user; is the initial state of charge (SOC) value when the tram is connected to the grid; Off-grid times set for tram users; and Respectively represent the lower limit and upper limit of the charging power at time t; and Represents the power values ​​that can be adjusted up and down respectively; is the rolling duration, and T is the evaluation duration of a single rolling evaluation.

[0114] In addition, the virtual energy storage model of the tram can be described as follows:

[0115] ;

[0116] In the above formula and are the SOC (state of charge of the battery, i.e. the ratio of the remaining capacity to the battery capacity) of the electric vehicle at time t and t-1 respectively; is the electric vehicle charging power at time t, is the charging efficiency; is the battery capacity of the electric vehicle; and Respectively represent the lower and upper limits of the electric vehicle’s charging power; and They represent the lower and upper limits of the electric vehicle SOC value respectively.

[0117] (2) Based on the state of charge, off-grid time, minimum off-grid charging level and rated charging power of the electric vehicle virtual energy storage model in a single evaluation period, the electric vehicle is classified into electric vehicles that remain on the grid and electric vehicles that will go off the grid. When the electric vehicle is connected to the grid through a charging pile, the corresponding charging pile will transmit data such as SOC value, off-grid time, minimum off-grid charging level and rated charging power to the VPP. Based on the above data in a single evaluation period [ , ] Electric vehicles are divided into two categories: those that remain connected to the grid and those that will leave the grid. For electric vehicles i that remain connected to the grid during the evaluation period, in order to reach the user's desired energy level before leaving Goal, electric vehicle i needs to be at the end of the evaluation cycle Reach the minimum SOC value , as shown below:

[0118] ;

[0119] In the above formula and Represent the maximum charging power of electric vehicle i and the set off-grid time respectively; the above formula represents the SOC value of the electric vehicle if At the end of the evaluation period When greater than or equal to the minimum SOC value , then the charging pile connected to the tram will charge at the maximum power Charging can achieve the SOC value set by the user at the set off-grid time. Otherwise, it will not be able to meet the user's demand for SOC when they are off-grid.

[0120] Furthermore, the upper and lower limits of the charging power for vehicle i that remains connected to the grid during the evaluation period are modeled as follows:

[0121] ;

[0122] In the above formula and are the maximum and minimum SOC values ​​of tram i, respectively; is the initial state of charge (SOC) value of electric vehicle i when it is connected to the grid; and They represent the upper and lower limits of the SOC of the trolley i at time t that remains connected to the grid during the evaluation period; and denote the upper and lower limits of the charging power of trolley i at time t that remains connected to the grid during the evaluation period; is the charging efficiency; is the battery capacity of the electric vehicle; The end time of the evaluation period; The start time of the evaluation period.

[0123] For the electric vehicle j that will be off-grid within the evaluation time range, its adjustable power range can be expressed as:

[0124] ;

[0125] In the above formula is the set off-grid time of tram j that will be off-grid within the evaluation time range; is the initial state of charge (SOC) value of tram j when it is connected to the grid; are the upper and lower limits of the SOC of tram j at time t, respectively; and are the upper and lower limits of the charging power of tram j at time t; is the charging efficiency; is the battery capacity of the electric vehicle; The start time of the evaluation period.

[0126] (3) Calculate the adjustable total power range of each tram within the corresponding evaluation time based on different categories of trams:

[0127] The adjustable power range for the trams that remain connected to the grid during the evaluation time is:

[0128] ;

[0129] The adjustable power range for the electric vehicles that will be off-grid during the evaluation time is:

[0130] ;

[0131] in The maximum value of the total adjustable power of the trams that remain connected to the grid during the evaluation period; The minimum adjustable total power of the tram that remains connected to the grid during the evaluation period; is the maximum adjustable total power of the tram that will be off-grid during the evaluation time; is the minimum adjustable total power of the trams that will be off-grid during the evaluation period; n is the number of trams that remain on-grid during the evaluation period; m is the number of trams that will be off-grid during the evaluation period; is the upper limit of the state of charge of the electric vehicle i that remains connected to the grid during the evaluation period; is the lower limit of the state of charge of the electric vehicle i that remains connected to the grid during the evaluation period; is the upper limit of the state of charge of the electric vehicle j that will be off the grid during the evaluation period; The lower limit of the state of charge of the off-grid vehicle j in period t during the evaluation time; To maintain the maximum charging power of the grid-connected electric vehicle in period t during the evaluation period; To maintain the minimum charging power of the grid-connected electric vehicle in period t during the evaluation period; is the maximum charging power of the off-grid electric vehicle in period t during the evaluation period; is the minimum charging power of the off-grid electric vehicle in period t during the evaluation period; is the upper limit of the charging power of the electric vehicle i that remains connected to the grid during the evaluation period in period t; is the lower limit of the charging power of the electric vehicle i that remains connected to the grid during the evaluation period in period t; is the upper limit of the charging power of the electric vehicle j that will be off-grid during the evaluation period; is the lower limit of the charging power of the electric vehicle j that will be off-grid during the evaluation time period t.

[0132] Furthermore, the process of aggregating the adjustable powers of different electric vehicle charging piles using Minkowski sum in step S3 includes:

[0133] (1) The Minkowski sum is used to aggregate the adjustable powers of different types of trams to obtain the adjustable power range of the virtual power plant; since the feasible adjustment flexibility domain of the two types of trams is a high-dimensional polyhedron, the Minkowski sum is used to aggregate the two types of tram clusters to obtain the VPP adjustable range , as shown below:

[0134] ;

[0135] In the above formula is the aggregate power of the VPP; and These are the spatially feasible and adjustable domains of the two types of trams; and are the charging power vectors of the two types of tram clusters within the evaluation time range.

[0136] (2) Then, the embedded hypercube approximation method is used to reduce the dimension and time decouple the adjustable power range of the aggregated virtual power plant to obtain the adjustable power range of the virtual power plant:

[0137] ;

[0138] ;

[0139] In the above formula is the aggregate power domain of the VPP; and They represent the upper and lower limits of the adjustable power at time t respectively; yes time dimension.

[0140] After embedding the hypercube approximation, the total adjustable power domain of VPP is: , can be formulated as follows:

[0141] .

[0142] S4: Divide all trams into tram combinations with the same external characteristics, and use the mean field game method to quickly decompose the frequency regulation instructions and economic benefits into different tram combinations to obtain the optimal frequency regulation control strategy for the corresponding tram combinations:

[0143] Specifically, the specific process of dividing all the electric vehicles into electric vehicle combinations with the same external characteristics in step S4 includes:

[0144] (1) All trams are randomly divided into a number of tram combinations with the same external characteristics according to the preset number of single tram combinations according to the frequency regulation performance of the trams; since the incentives received by the tram combinations should depend on their contribution to the frequency regulation service they provide, the upward and downward regulation powers are defined as conditional attributes and , as shown below:

[0145] ;

[0146] in and is the downward and upward regulation power of the random tram combination.

[0147] And with the conditional attribute and Corresponding attribute value range The calculation is as follows:

[0148] ;

[0149] Further from the conditional attributes and Information mapping to the maximum attribute value of the tram combination It can be expressed as follows:

[0150] ;

[0151] In the above formula and are the upper and lower limits of the power of the random combination of trams.

[0152] (2) The Euclidean distance between the adjustable power range of the tram combination and the adjustable power range of the standard tram combination with the same external characteristics is calculated; among them, the random tram combination and a standard tram combination with the same external characteristics The difference between The measurements are as follows:

[0153] ;

[0154] In the above formula represents the Euclidean distance between the randomly combined trolley combinations and the standard trolley combination; U For all trams; It is a number infinitely close to 0.

[0155] (3) When the Euclidean distance between the tram combination and the standard tram combination is less than the set value, the corresponding tram combination is output; otherwise, the trams are randomly recombined and the Euclidean distance between the combinations is repeatedly calculated until the Euclidean distance between the output tram combination and the standard tram combination is less than the set value, that is, they have the same external characteristics.

[0156] Finally, the information of the tram combination with the same external characteristics is expressed as follows:

[0157] ;

[0158] In the above formula is the SOC value of the tram combination with the same external characteristics at time t; and They represent the upward and downward power values ​​of the tram combination with the same external characteristics at time t; and They represent the upper and lower limits of the power regulation for a standard tram respectively; and are the lower and upper limits of the SOC of the standard tram combination; is the value of the initial SOC of a standard electric vehicle.

[0159] Furthermore, in step S4, the mean field game method is used to quickly decompose the frequency modulation instructions and economic benefits into different tram combinations. The specific method is as follows:

[0160] (1) To determine the optimal frequency regulation command strategy and economic benefit distribution, a mean field game model is constructed between the tram combinations; and to encourage the trams to actively respond to the frequency regulation command demand, a dynamic incentive mechanism is constructed; wherein the mean field game model is:

[0161] ;

[0162] in represents all the trolley combinations participating at time t; represents the joint frequency regulation command strategy of all tram combinations, where the frequency regulation command strategy of the tram combination in period t includes upward power regulation command and downward power regulation command; Indicates the battery charge state of all tram combinations; is the utility function, which represents the satisfaction of the trolley combination participating in the mean field game.

[0163] The dynamic incentive mechanism includes:

[0164] ;

[0165] in is the incentive price for upward adjustment of power command during period t; is the incentive price for downward adjustment of power command during period t; is a piecewise linear incentive price function for upward regulation of power command; A piecewise linear incentive price function for downwardly regulating power commands; and are the positive parameters of the linear pricing strategy that adjusts the power command upward; and are the positive parameters of the linear pricing strategy that adjusts the power command downward; is the power load during period t; is the average daily power load before frequency regulation during period t; The ratio of the incentive price to the clearing price for upward adjustment of power orders; The ratio of the incentive price to the clearing price for downward adjustment of power orders; To adjust the clearing price of power orders upwards; is the clearing price for adjusting power orders downward.

[0166] In this embodiment, based on the dynamic incentive mechanism, efforts are made to develop the optimal frequency control strategy. , in order to maximize the economic benefits of frequency regulation services, while taking into account its status and frequency incentive price. Therefore, the above game problem can be formulated as:

[0167] ;

[0168] Among them, the above express The Nash equilibrium point.

[0169] ;

[0170] in is the joint frequency decision of the trolley combinations except for the single trolley combination player; the vector and Represents power down Incentive price and upward power adjustment The incentive price, and It is a period of adjustment, both downward and upward; is the deviation from the desired state of charge The penalty factor; and Deviates from the expected downward adjustment of power and the desired upward power regulation Penalty factor. Further based on the clearing results and , we can calculate and ,Right now:

[0171] .

[0172] (2) Based on the collective effect of the mean field on the control strategy of each tram combination, a mean field term is constructed, including upward and downward power command adjustments. When the mean field game involves large-scale tram combinations, the impact of the frequency control strategy of a single tram combination on other combinations can be ignored. In contrast, the collective effect of the mean field on the control strategy of each combination is significant. Specifically, the mean field term (MFT) is expressed as:

[0173] ;

[0174] in is the mean-field term for adjusting the power command upward; is the mean field term for downwardly regulating the power command; represents the number of trolley combinations participating at time t; represents all the trolley combinations participating at time t; Adjust the power command upward for period t; The power command is adjusted downward for the period t.

[0175] (3) Combining the mean field term with the dynamic incentive mechanism, we can obtain the relationship between the incentive price and the mean field term corresponding to the upward and downward adjustment of the power instruction:

[0176] ;

[0177] in is the incentive price for upward adjustment of power command during period t; is the incentive price for downward adjustment of power command during period t; is a piecewise linear incentive price function for upward regulation of power command; is a piecewise linear incentive price function that adjusts the power command downward.

[0178] (4) Based on the relationship between the incentive price and the mean field term and the mean field game model, the optimal frequency modulation control strategy considering the mean field term is obtained; specifically, by substituting the relationship between the incentive price and the mean field term into the utility function of the mean field game model, we can obtain:

[0179] ;

[0180] in is the joint frequency control strategy of all tram combinations; y and z are sequence variables; An adjustment period for adjusting the power command upward; An adjustment period for adjusting the power command downward; Adjust power downward for deviations from expectations The penalty factor; Adjust power upwards for deviations from expectations The penalty factor; Deviating from the desired state of charge The penalty factor; is the state of charge during period t.

[0181] Considering the mean field term and Optimal frequency control strategy for tram combinations Rephrased as follows:

[0182] .

[0183] (5) Based on the optimal frequency modulation control strategy considering the mean field term, the corresponding Hamilton-Jacobi-Bellman (HJB) equation is obtained; for example, the HJB equation for the mean field term at time t is defined as:

[0184] ;

[0185] ;

[0186] in, is the value function based on the Bellman optimality principle, sup{•} is the optimistic value, Yes The result after derivation.

[0187] (6) By iterating the mean field term and the Hamilton-Jacobi-Bellman equation until the mean field equilibrium state is reached, the optimal frequency control strategy obtained at this time is sent to the tram charging piles corresponding to the tram combination.

[0188] In summary, the mean field term (MFT) represents the mapping from the individual control strategies of the tram combination to the average field effect, which is expressed as the operator The Hamilton-Jacobi-Bellman (HJB) equation describes the mapping from the average field effect to a single frequency control strategy, expressed as the operator . Combined Figure 4 As shown, the operator and Alternate implementation to achieve mean field equilibrium.

[0189] Further, such as Figure 5 As shown, the present application also discloses a virtual power plant average field game control device for tram aggregate frequency regulation, the device comprising:

[0190] Parameter acquisition module, used to obtain historical charging data, real-time charging data and the spatiotemporal relationship of charging piles for electric vehicle charging piles;

[0191] The baseline prediction module is used to analyze the historical charging data and the spatiotemporal relationship of electric vehicle charging piles using a multi-graph fusion neural network model to obtain the power baseline prediction values ​​of different electric vehicle charging piles;

[0192] The power aggregation module is used to perform a rolling evaluation of the adjustable power of different electric vehicle charging piles based on the real-time charging information of the electric vehicles, and aggregate the adjustable power of different electric vehicle charging piles using Minkowski sum to obtain the adjustable power range of the virtual power plant;

[0193] The control output module is used to divide all trams into homogeneous clusters with the same external characteristics. It uses the mean field game method to quickly decompose the frequency regulation instructions and economic benefits into the homogeneous clusters of trams to obtain the optimal frequency regulation control strategy:

[0194] To determine the optimal frequency regulation strategy and economic benefit distribution, a mean field game model is constructed between tram combinations; and to encourage trams to actively respond to frequency regulation demands, a dynamic incentive mechanism is constructed.

[0195] Based on the collective effect of the mean field on the control strategy of each tram combination, a mean field term including upward adjustment of power command and downward adjustment of power command is constructed;

[0196] The mean field term is combined with the dynamic incentive mechanism to obtain the relationship between the incentive price and the mean field term corresponding to the upward and downward adjustment of the power instruction.

[0197] Based on the relationship between incentive price and mean field term and mean field game model, the optimal frequency modulation control strategy considering mean field term is obtained;

[0198] The corresponding Hamilton-Jacobi-Bellman equation is obtained based on the optimal frequency modulation control strategy considering the mean field term;

[0199] By iterating the mean field term and the Hamilton-Jacobi-Bellman equation until the mean field equilibrium state is reached, the optimal frequency control strategy obtained at this time is sent to the corresponding tram combination.

[0200] The device provided in the embodiment of the present application can achieve Figure 1 To avoid repetition, the various processes implemented in the method embodiment will not be described here.

[0201] like Figure 6 As shown, the embodiment of the present application also provides an electronic device, including a processor and a memory, a program or instruction stored in the memory and capable of running on the processor, and when the program or instruction is executed by the processor, the following is achieved: Figure 1 The various processes of the method embodiment shown in the figure can achieve the same technical effect. To avoid repetition, they will not be described here.

[0202] The embodiment of the present application also provides a readable storage medium on which a program or instruction is stored, and when the program or instruction is executed by the processor, the above Figure 1 The various processes of the method embodiments described above can achieve the same technical effects, and will not be described again here to avoid repetition.

[0203] The present application also provides a computer program product including computer instructions, which, when executed by a processor, implement the above Figure 1 The various processes of the method embodiments described above can achieve the same technical effects, and will not be described again here to avoid repetition.

[0204] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The above-mentioned serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments.

[0205] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0206] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another device, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0207] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0208] In addition, all functional units in the embodiments of the present application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the above-mentioned integrated units can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0209] Those skilled in the art will understand that all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, and other media that can store program codes.

[0210] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a device (which can be a terminal or platform, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks or optical disks.

[0211] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A virtual power plant average field game control method for tram aggregate frequency regulation, characterized by: The steps include: Obtain historical charging data, real-time charging data, and the spatiotemporal relationship of electric vehicle charging piles; The multi-graph fusion neural network model is used to analyze the historical charging data and the spatiotemporal relationship of electric vehicle charging piles to obtain the power baseline prediction values ​​of different electric vehicle charging piles: Construct a multi-graph fusion neural network model and obtain the network topology and feature matrix of the electric vehicle charging pile; The multi-graph fusion neural network model comprises at least a multi-graph fusion convolutional layer, a gated recurrent unit loop layer, and a fully connected layer; Based on the network topology of the electric vehicle charging piles and the electric vehicle charging modes of different nodes, the charging demand similarity graph of the corresponding nodes is constructed. The elements of the adjacent matrix in the charging demand similarity graph are: ； in and From the node and Time data series; express and the distance between them; is the maximum distance; is the threshold parameter that limits the size of the charging demand similarity graph; Based on the network topology of the electric vehicle charging piles and the spatial proximity of different nodes, a geographic adjacency graph of the corresponding nodes is constructed. The elements of the adjacent matrix in the geographic adjacency graph are: ； where d is the node and The distance between s and are the thresholds of graph connectivity; The charging demand similarity graph and geographic adjacency graph of the same node are respectively input into the multi-graph fusion convolutional layer of the multi-graph fusion neural network model to obtain the output results of the corresponding nodes; The output results of the charging demand similarity graph and the geographic adjacency graph are weighted and fused to obtain the fused output result; The fused output results are fed into the gated recurrent unit loop layer for analytical calculation to minimize the difference between the actual charging demand and the predicted demand of each node; Output the power baseline prediction value of the electric vehicle charging pile at different nodes through the fully connected layer; Based on the real-time charging information of trams, a rolling evaluation of the adjustable power of different tram charging piles is performed. The Minkowski sum is used to aggregate the adjustable power of different tram charging piles to obtain the adjustable power range of the virtual power plant: Based on the power baseline prediction values ​​of different electric vehicle charging piles, a rolling assessment framework model and a virtual energy storage model for the corresponding electric vehicle are constructed respectively; Based on the state of charge, off-grid time, minimum off-grid charging level and rated charging power of the tram's virtual energy storage model in a single assessment period, the tram is classified into trams that remain on the grid and trams that will go off the grid; Based on different classifications of trams, the adjustable total power range of the corresponding tram within the corresponding evaluation time is calculated respectively; The adjustable power range of the virtual power plant is obtained by aggregating the adjustable powers of different types of trams using Minkowski sum; The embedded hypercube approximation method is used to reduce the dimension and time decouple the adjustable power range of the virtual power plant obtained by aggregation to obtain the adjustable power range of the virtual power plant; All trams are divided into tram combinations with the same external characteristics, and the mean field game method is used to quickly decompose the frequency regulation instructions and economic benefits into different tram combinations to obtain the optimal frequency regulation control strategy for the corresponding tram combinations. To determine the optimal frequency regulation strategy and economic benefit distribution, a mean field game model is constructed between tram combinations; and to encourage trams to actively respond to frequency regulation demands, a dynamic incentive mechanism is constructed. Based on the collective effect of the mean field on the control strategy of each tram combination, a mean field term including upward adjustment of power command and downward adjustment of power command is constructed; The mean field term is combined with the dynamic incentive mechanism to obtain the relationship between the incentive price and the mean field term corresponding to the upward and downward adjustment of the power instruction. Based on the relationship between incentive price and mean field term and mean field game model, the optimal frequency modulation control strategy considering mean field term is obtained; The corresponding Hamilton-Jacobi-Bellman equation is obtained based on the optimal frequency modulation control strategy considering the mean field term; By iterating the mean field term and the Hamilton-Jacobi-Bellman equation until the mean field equilibrium state is reached, the optimal frequency control strategy obtained at this time is sent to the corresponding tram combination.

2. The virtual power plant average field game control method according to claim 1, characterized in that: The step of performing a rolling evaluation of adjustable power on different electric vehicle charging piles according to the real-time charging information of the electric vehicle comprises: The adjustable power range for the trams that remain connected to the grid during the evaluation time is: ； The adjustable power range for the electric vehicles that will be off-grid during the evaluation time is: ； in The maximum value of the total adjustable power of the trams that remain connected to the grid during the evaluation period; The minimum adjustable total power of the tram that remains connected to the grid during the evaluation period; is the maximum adjustable total power of the tram that will be off-grid during the evaluation time; is the minimum adjustable total power of the trams that will be off-grid during the evaluation period; n is the number of trams that remain on-grid during the evaluation period; m is the number of trams that will be off-grid during the evaluation period; is the upper limit of the state of charge of the electric vehicle i that remains connected to the grid during the evaluation period; is the lower limit of the state of charge of the electric vehicle i that remains connected to the grid during the evaluation period; is the upper limit of the state of charge of the electric vehicle j that will be off the grid during the evaluation period; The lower limit of the state of charge of the off-grid vehicle j in period t during the evaluation time; To maintain the maximum charging power of the grid-connected electric vehicle in period t during the evaluation period; To maintain the minimum charging power of the grid-connected electric vehicle in period t during the evaluation period; is the maximum charging power of the off-grid electric vehicle in period t during the evaluation period; is the minimum charging power of the off-grid electric vehicle in period t during the evaluation period; is the upper limit of the charging power of the electric vehicle i that remains connected to the grid during the evaluation period in period t; is the lower limit of the charging power of the electric vehicle i that remains connected to the grid during the evaluation period in period t; is the upper limit of the charging power of the electric vehicle j that will be off-grid during the evaluation period; is the lower limit of the charging power of the electric vehicle j that will be off-grid during the evaluation time period t.

3. The virtual power plant average field game control method according to claim 1, characterized in that: The step of dividing all the trams into tram combinations having the same external characteristics comprises: According to the frequency regulation performance of the trams, all trams are randomly divided into a number of tram combinations with the same external characteristics according to the preset number of single tram combinations; The Euclidean distance between the combinations is calculated based on the adjustable power range of the combination and the adjustable power range of a standard combination of trams with the same external characteristics; When the Euclidean distance between the tram combination and the standard tram combination is less than the set value, the corresponding tram combination is output; otherwise, the trams are randomly recombined and the Euclidean distance between the combinations is repeatedly calculated until the Euclidean distance between the output tram combination and the standard tram combination is less than the set value, that is, they have the same external characteristics.

4. The virtual power plant average field game control method according to claim 1, characterized in that: The mean field game model is: ; in represents all the trolley combinations participating at time t; represents the joint frequency regulation command strategy of all tram combinations, where the frequency regulation command strategy of the tram combination in period t includes upward power regulation command and downward power regulation command; Indicates the battery charge state of all tram combinations; is the utility function, which represents the satisfaction of the trolley combination participating in the mean field game.

5. The virtual power plant average field game control method according to claim 4 is characterized in that: The dynamic incentive mechanism includes: ； in is the incentive price for upward adjustment of power command during period t; is the incentive price for downward adjustment of power command during period t; is a piecewise linear incentive price function for upward regulation of power command; A piecewise linear incentive price function for downwardly regulating power commands; and are the positive parameters of the linear pricing strategy that adjusts the power command upward; and are the positive parameters of the linear pricing strategy that adjusts the power command downward; is the power load during period t; is the average daily power load before frequency regulation during period t; The ratio of the incentive price to the clearing price for upward adjustment of power orders; The ratio of the incentive price to the clearing price for downward adjustment of power orders; To adjust the clearing price of power orders upwards; is the clearing price for adjusting power orders downward.

6. The virtual power plant average field game control method according to claim 5, characterized in that: The mean field term is expressed as: ； in is the mean-field term for adjusting the power command upward; is the mean field term for downwardly regulating the power command; represents the number of trolley combinations participating at time t; represents all the trolley combinations participating at time t; Adjust the power command upward for period t; The power command is adjusted downward for the period t.

7. The virtual power plant average field game control method according to claim 6, characterized in that: The step of obtaining the optimal frequency modulation control strategy considering the mean field term based on the relationship between the incentive price and the mean field term and the mean field game model includes: The relationship between the incentive price and the mean field term is: ； in is the incentive price for upward adjustment of power command during period t; is the incentive price for downward adjustment of power command during period t; is a piecewise linear incentive price function for upward regulation of power command; A piecewise linear incentive price function for downwardly regulating power commands; Substitute the relationship between the incentive price and the mean field term into the utility function of the mean field game model: ； in is the joint frequency control strategy of all tram combinations; y and z are sequence variables; An adjustment period for adjusting the power command upward; An adjustment period for adjusting the power command downward; Adjust power downward for deviations from expectations The penalty factor; Adjust power upwards for deviations from expectations The penalty factor; Deviating from the desired state of charge The penalty factor; is the state of charge during period t.

8. A virtual power plant average field game control device for tram aggregate frequency regulation, characterized in that: The device comprises: Parameter acquisition module, used to obtain historical charging data, real-time charging data and the spatiotemporal relationship of charging piles for electric vehicle charging piles; The baseline prediction module is used to analyze the historical charging data and the spatiotemporal relationship of electric vehicle charging piles using a multi-graph fusion neural network model to obtain the power baseline prediction values ​​of different electric vehicle charging piles; Constructing a multi-graph fusion neural network model and obtaining a network topology map and a feature matrix of a tram charging pile; wherein the multi-graph fusion neural network model includes at least a multi-graph fusion convolutional layer, a gated recurrent unit loop layer, and a fully connected layer; Based on the network topology of the electric vehicle charging piles and the electric vehicle charging modes of different nodes, the charging demand similarity graph of the corresponding nodes is constructed. The elements of the adjacent matrix in the charging demand similarity graph are: ； in and From the node and Time data series; express and the distance between them; is the maximum distance; It is a restriction size threshold parameter; Based on the network topology of the electric vehicle charging piles and the spatial proximity of different nodes, a geographic adjacency graph of the corresponding nodes is constructed. The elements of the adjacent matrix in the geographic adjacency graph are: ； where d is the node and The distance between s and are the thresholds of graph connectivity; The charging demand similarity graph and geographic adjacency graph of the same node are respectively input into the multi-graph fusion convolutional layer of the multi-graph fusion neural network model to obtain the output results of the corresponding nodes; The output results of the charging demand similarity graph and the geographic adjacency graph are weighted and fused to obtain the fused output result; The fused output results are fed into the gated recurrent unit loop layer for analytical calculation to minimize the difference between the actual charging demand and the predicted demand of each node; Output the power baseline prediction value of the electric vehicle charging pile at different nodes through the fully connected layer; The power aggregation module is used to perform a rolling evaluation of the adjustable power of different electric vehicle charging piles based on the real-time charging information of the electric vehicle. It also uses the Minkowski sum to aggregate the adjustable power of different electric vehicle charging piles to obtain the adjustable power range of the virtual power plant: Based on the power baseline prediction values ​​of different electric vehicle charging piles, a rolling assessment framework model and a virtual energy storage model for the corresponding electric vehicle are constructed respectively; Based on the state of charge, off-grid time, minimum off-grid charging level and rated charging power of the tram's virtual energy storage model in a single assessment period, the tram is classified into trams that remain on the grid and trams that will go off the grid; Based on different classifications of trams, the adjustable total power range of the corresponding tram within the corresponding evaluation time is calculated respectively; The adjustable power range of the virtual power plant is obtained by aggregating the adjustable powers of different types of trams using Minkowski sum; The embedded hypercube approximation method is used to reduce the dimension and time decouple the adjustable power range of the virtual power plant obtained by aggregation to obtain the adjustable power range of the virtual power plant; The control output module is used to divide all trams into homogeneous clusters with the same external characteristics, and use the mean field game method to quickly decompose the frequency regulation instructions and economic benefits into the homogeneous clusters of trams to obtain the optimal frequency regulation control strategy; To determine the optimal frequency regulation strategy and economic benefit distribution, a mean field game model is constructed between tram combinations; and to encourage trams to actively respond to frequency regulation demands, a dynamic incentive mechanism is constructed. Based on the collective effect of the mean field on the control strategy of each tram combination, a mean field term including upward adjustment of power command and downward adjustment of power command is constructed; The mean field term is combined with the dynamic incentive mechanism to obtain the relationship between the incentive price and the mean field term corresponding to the upward and downward adjustment of the power instruction. Based on the relationship between incentive price and mean field term and mean field game model, the optimal frequency modulation control strategy considering mean field term is obtained; The corresponding Hamilton-Jacobi-Bellman equation is obtained based on the optimal frequency modulation control strategy considering the mean field term; By iterating the mean field term and the Hamilton-Jacobi-Bellman equation until the mean field equilibrium state is reached, the optimal frequency control strategy obtained at this time is sent to the corresponding tram combination.

Images