Active power distribution network partitioning method and system suitable for large-scale electric vehicle access

Abstract

The invention discloses an active power distribution network partitioning method and system suitable for large-scale electric vehicle access, and belongs to the technical field of power distribution network partitioning. According to the method, the active power distribution network accessed by the large-scale electric vehicle is partitioned, each node of the power distribution network is more reasonably partitioned, and the pressure of operation management of the power distribution network is reduced. According to the method, the partition time sequence weight matrix is utilized, the dynamic partition combination network is constructed, the network model can capture the structural features and the energy data features of the power distribution network, the nodes with potential consistency are divided into the same sub-region, and the dynamic adaptability of the power distribution network partition is improved by utilizing the multi-objective loss function. According to the method, large-scale active power distribution network partitioning of electric vehicle access can be realized, the reasonability and accuracy of power distribution network partitioning of electric vehicle access are improved, and the method has a wide development prospect.

Description

Technical Field

[0001] This invention belongs to the field of power distribution network zoning technology, and more specifically, relates to an active power distribution network zoning method and system suitable for large-scale electric vehicle access. Background Technology

[0002] Initially, power grid zoning was primarily aimed at addressing operational and management challenges arising from power grid expansion and electromagnetic loops. However, the development of various distributed energy sources and the large-scale integration of electric vehicles have presented new challenges to the stable operation of the power grid. Zoning the power grid not only optimizes its operation and management but also provides flexibility and operability for system expansion. Fully utilizing distributed energy sources and the voltage regulation capabilities of electric vehicles is a crucial foundation for the transition of modern power systems from centralized to distributed systems. However, traditional distribution network zoning methods, mainly based on static topology or simple electrical characteristics, struggle to adapt to the spatiotemporal dynamics of EV charging and discharging behavior. Furthermore, distribution networks face increasingly prominent issues such as voltage deviation, power imbalance, and equipment overload. Therefore, providing an effective active distribution network zoning method suitable for large-scale electric vehicle integration is of great significance in addressing these problems. Summary of the Invention

[0003] In response to the above-mentioned deficiencies or improvement needs of existing technologies, this invention proposes an active power distribution network zoning method and system suitable for large-scale electric vehicle access, thereby solving the problems of voltage deviation, power imbalance and equipment overload.

[0004] To achieve the above objectives, according to one aspect of the present invention, an active power distribution network zoning method suitable for large-scale electric vehicle access is provided, comprising:

[0005] Step 1: Obtain relevant data on the distribution network topology, perform preprocessing, and construct an initial partitioning model;

[0006] Step 2: Calculate the partition time-series weight matrix and construct a dynamic partition optimization model;

[0007] Step 3: Use the dynamic partitioning optimization model to achieve the final partitioning.

[0008] Preferably, the relevant data for obtaining the distribution network topology includes: the distribution network topology structure, power data and voltage data of each node, output data of distributed power sources, and charging demand data of electric vehicles.

[0009] Preferably, the pretreatment steps are as follows:

[0010] Step 1: Construct a composite feature matrix that includes electrical characteristics, spatiotemporal characteristics, and EV states:

[0011]

[0012] Among them, P G Q(t) represents the active power of the generator at time t. G (t) represents the reactive power of the generator at time t, P D (t) represents the active power of the load at time t, Q D V(t) represents the active power of the load at time t, V(t) is the magnitude of the node voltage at time t, and θ(t) is the phase angle of the node voltage at time t. The charging power of the electric vehicle at time t. To predict power, η EV (t) represents the battery charging efficiency; t = 1, 2, 3, ..., T;

[0013] Step 2: Standardize the composite feature matrix X using its standard deviation:

[0014]

[0015] Where, x * The data consists of standardized feature data, where N is the number of nodes and x... i The original feature data for node i. Let τ be the arithmetic mean of the feature data, and τ be the standard deviation of the feature data.

[0016] Preferably, the steps for constructing the initial partitioning model are:

[0017] Step 1: Initialize the relevant network layers of the initial partitioned network, such as the number of neurons Q in the input layer. in The number of neurons Q in the hidden layer H The number of neurons in the output layer is Q. O ;

[0018] Step 2: Initialize the input-to-hidden-layer connection weight matrix W of the initial partitioned network. in The matrix size is (Q in ×Q H );

[0019] Step 3: The input encoding module calculates the input data X using the following formula to compress and extract node features:

[0020]

[0021] Among them, H (l) Let f be the data representation learned by the l-th layer of the input encoding module, and f be the activation function. The weight matrix of the l-th layer in the input encoding module, This refers to the bias term of the l-th layer in the input encoding module;

[0022] Step 4: The output decoding module processes the extracted node features using the following formula to obtain the reconstructed data.

[0023]

[0024] in, To output the weight matrix of the l-th layer in the decoding module, This is the bias term for the l-th layer in the output decoding module;

[0025] Step 5: In each round of learning, the network parameters are continuously updated by minimizing the reconstruction loss using the following formula, and key node features are extracted:

[0026]

[0027] Among them, L recon For reconstruction loss;

[0028] Step 6: Process the data using the following formula to generate new feature representations for the nodes:

[0029]

[0030] Among them, Z l The node feature representation learned by the l-th graph convolutional layer. To add a self-looping adjacency matrix, for The corresponding degree matrix, W (l-1) Here is the weight matrix of the (l-1)th graph convolutional layer. α represents the node combined data generated by the (l-1)th layer encoding module and the graph convolutional layer, where α is the weight coefficient.

[0031] Step 7: Normalize the feature data and transform it into a probability distribution Z = {z ik}

[0032]

[0033] Step 8: Obtain the partition result identifier vector C = {c i}:

[0034]

[0035] Preferably, the specific steps for constructing a dynamic partitioning optimization model by calculating the partitioning time-series weight matrix are as follows:

[0036] Step 1: Initialize the relevant parameters of the dynamic partitioning ensemble network, including the number of neurons D in the input layer. in and the number of hidden layer neurons D HThe number of neurons in the output layer is D. O ;

[0037] Step 2: Initialize the partition time-series weight matrix W of the dynamic partitioning combined network. d The matrix size is (D in ×D H ):

[0038]

[0039] Among them, P EV,i The EV charging and discharging power time series for node i;

[0040] Step 3: Use augmented composite feature matrix As input data for the input encoding module;

[0041] Step 4: Extract node features to obtain the initial partitioning model;

[0042] Step 5: For the extracted node key features H (l) Perform clustering operations to obtain cluster centers μ. k ;

[0043] Step 6: Measure the key features H of the nodes using the following formula. (l) With cluster center μ k The distance is used to obtain the characteristic distribution Q = {q} ik}:

[0044]

[0045] Where, q ik Let ζ be the probability of assigning node i to partition k, and ζ be the temperature parameter.

[0046] Step 7: Calculate the KL divergence between the probability distribution and the characteristic distribution:

[0047]

[0048] Step 8: Design a multi-objective loss function with the objectives of minimizing power imbalance, voltage deviation, and reconstruction loss.

[0049]

[0050] Where, β 1，2，3,4 λ is the weighting coefficient, and λ is the regularization coefficient;

[0051] Power balance loss

[0052]

[0053] Where C is the partition identifier vector and T is the time step. The interaction power between partition k and the main network;

[0054] Voltage deviation loss

[0055]

[0056] Where M is the number of nodes in a single partition, Π is the indicator function, and V ref Rated voltage;

[0057] Preferably, the steps for implementing the final partition using the dynamic partitioning optimization model are as follows:

[0058] Step 1: Standardize the composite feature matrix X using its standard deviation;

[0059] Step 2: Initialize the current iteration count (iter) of the dynamic partitioning combination network, and set the maximum iteration count (iter). max Initialize the parameters of each layer of the network;

[0060] Step 3: Using the augmented composite feature matrix As the input to the input encoding module in the dynamic partitioning combined network, the deepest learned data H is processed. (3) Perform clustering to obtain cluster centers μ k Calculate the characteristic distribution Q = {q} ik};

[0061] Step 4: Using the adjacency matrix A and the augmented composite characteristic matrix As input to the graph convolutional network in the dynamic partitioning ensemble network, it extracts the features of the nodes. We obtain the probability distribution Z and the dynamic partitioning result identifier vector C;

[0062] Step 5: Calculate the multi-objective loss function and determine whether the number of iterations is greater than the maximum number of iterations. If not, backpropagate according to the gradient descent method to update the parameters of the dynamic partitioning combination network, and then return to execute step (3); if it is true, obtain the final partitioning result identifier vector to realize the partitioning of the distribution network nodes.

[0063] According to another aspect of the present invention, an active power distribution network zoning system suitable for large-scale electric vehicle access is provided, comprising:

[0064] Initial model construction module: used to acquire relevant data on the distribution network topology, perform preprocessing, and construct the initial partition model;

[0065] Distribution network dynamic zoning optimization module: calculates the zoning time-series weight matrix and constructs a dynamic zoning optimization model;

[0066] Distribution network partitioning module: The final partitioning is achieved using a dynamic partitioning optimization model.

[0067] Preferably, the initial model construction module is used to acquire relevant data on the distribution network topology, perform preprocessing, and construct the initial partition model. The specific steps are as follows:

[0068] Step 1: Obtain relevant data on the distribution network topology, including: distribution network topology structure, power data and voltage data of each node, output data of distributed power sources, and charging demand data of electric vehicles;

[0069] Step 2: Construct a composite feature matrix that includes electrical characteristics, spatiotemporal characteristics, and EV states:

[0070]

[0071] Among them, P G Q(t) represents the active power of the generator at time t. G (t) represents the reactive power of the generator at time t, P D (t) represents the active power of the load at time t, Q D V(t) represents the active power of the load at time t, V(t) is the magnitude of the node voltage at time t, and θ(t) is the phase angle of the node voltage at time t. The charging power of the electric vehicle at time t. To predict power, η EV (t) represents the battery charging efficiency; t = 1, 2, 3, ..., T;

[0072] Step 3: Standardize the composite feature matrix X using its standard deviation:

[0073]

[0074] Where, x * The data consists of standardized feature data, where N is the number of nodes and x... i Let x be the original feature data of node i, x be the arithmetic mean of the feature data, and μ be the standard deviation of the feature data.

[0075] Step 4: Initialize the relevant network layers of the initial partitioned network, such as the number of neurons Q in the input layer. in The number of neurons Q in the hidden layer H The number of neurons in the output layer is Q. O ;

[0076] Step 5: Initialize the input-to-hidden-layer connection weight matrix W of the initial partitioned network. in The matrix size is (Q in ×Q H );

[0077] Step 6: The input encoding module calculates the input data using the following formula to compress and extract node features:

[0078]

[0079] Among them, H (l) Let f be the data representation learned by the l-th layer of the input encoding module, and f be the activation function. The weight matrix of the l-th layer in the input encoding module, The deviation of the l-th layer in the input encoding module;

[0080] Step 7: The output decoding module processes the extracted node features using the following formula to obtain the reconstructed data.

[0081]

[0082] in, To output the weight matrix of the l-th layer in the decoding module, This is the deviation of the l-th layer in the output decoding module;

[0083] Step 8: In each round of learning, continuously update the network parameters by minimizing the reconstruction loss using the following formula:

[0084]

[0085] Among them, L recon For reconstruction loss;

[0086] Step 9: Process the data using the following formula to generate new feature representations for the nodes:

[0087]

[0088] Among them, Z l The node feature representation learned by the l-th graph convolutional layer. To add a self-looping adjacency matrix, for The corresponding degree matrix, W (l-1) Here is the weight matrix of the (l-1)th graph convolutional layer. α represents the node combined data generated by the (l-1)th layer encoding module and the graph convolutional layer, where α is the weight coefficient.

[0089] Step 10: Normalize the feature data and transform it into a probability distribution Z = {z ik}:

[0090]

[0091] Step 11: Obtain the partition result identifier vector C = {c i}:

[0092]

[0093] Preferably, the implementation steps of the distribution network dynamic zoning optimization module to calculate the zoning time-series weight matrix and construct the dynamic zoning optimization model are as follows:

[0094] Step 1: Initialize the relevant parameters of the dynamic partitioning ensemble network, including the number of neurons D in the input layer. in and the number of hidden layer neurons D H The number of neurons in the output layer is D. O The number of partitions is k;

[0095] Step 2: Initialize the partition time-series weight matrix W of the dynamic partitioning combined network. d The matrix size is (D in ×D H ):

[0096]

[0097] Among them, P EV,i The EV charging and discharging power time series for node i;

[0098] Step 3: Use augmented composite feature matrix As input data for the input encoding module;

[0099] Step 4: Perform steps 3-8 of claim 4;

[0100] Step 5: For the extracted node key features H (l) Perform clustering operations to obtain cluster centers μ. k ;

[0101] Step 6: Measure the key features H of the nodes using the following formula. (l) With cluster center μ k The distance is used to obtain the characteristic distribution Q = {q} ik}:

[0102]

[0103] Where, q ik Let ζ be the probability of assigning node i to partition k, and ζ be the temperature parameter.

[0104] Step 7: Calculate the KL divergence between the probability distribution and the characteristic distribution:

[0105]

[0106] Step 8: Design a multi-objective loss function with the objectives of minimizing power imbalance, voltage deviation, and reconstruction loss.

[0107]

[0108] Where, β 1，2，3,4 λ is the weighting coefficient, and λ is the regularization coefficient;

[0109] Power balance loss

[0110]

[0111] Where C is the partition identifier vector and T is the time step. The interaction power between partition k and the main network;

[0112] Voltage deviation loss

[0113]

[0114] Where M is the number of nodes in a single partition, Π is the indicator function, and V ref Rated voltage;

[0115] Preferably, the distribution network partitioning module uses a dynamic partitioning optimization model to achieve the final partitioning, and the implementation steps are as follows:

[0116] Step 1: Standardize the composite feature matrix X using its standard deviation;

[0117] Step 2: Initialize the current iteration count (iter) of the dynamic partitioning combination network, and set the maximum iteration count (iter). max Initialize the parameters of each layer of the network;

[0118] Step 3: Using the augmented composite feature matrix As the input to the input encoding module in the dynamic partitioning combined network, the deepest learned data H is processed. (3) Perform clustering to obtain cluster centers μ k Calculate the characteristic distribution Q = {q} ik};

[0119] Step 4: Using the adjacency matrix A and the augmented composite characteristic matrix As input to the graph convolutional network in the dynamic partitioning ensemble network, it extracts the features of the nodes. We obtain the probability distribution Z and the dynamic partitioning result identifier vector C;

[0120] Step 5: Calculate the multi-objective loss function and determine whether the number of iterations is greater than the maximum number of iterations. If not, backpropagate according to the gradient descent method to update the parameters of the dynamic partitioning combination network, and then return to execute step (3); if it is true, obtain the final partitioning result identifier vector to realize the partitioning of the distribution network nodes.

[0121] According to another aspect of the invention, a computer-readable storage medium is provided having program instructions stored thereon, which, when executed by a processor, implement an active power distribution network zoning method suitable for large-scale electric vehicle access as described above.

[0122] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0123] The active distribution network zoning method of this invention, applicable to large-scale electric vehicle (EV) access, can effectively zonify the distribution network with large-scale EV access using a dynamic zoning network model. It uses a zoning time-series weight matrix and a dynamic zoning network to capture the spatiotemporal correlation of EV charging power, allowing the zoning strategy to be adjusted in real time according to charging demand. The design of a multi-objective loss function enables real-time matching between the net power within a zoning area and the power exchanged with the main grid, avoiding local overload. Voltage deviation loss reduces the risk of overvoltage. This method reduces the risk of voltage exceeding limits when a high proportion of EVs are connected to the grid, improves the resilience of the distribution network, and compensates for the shortcomings of traditional zoning methods in terms of dynamic adaptability, efficiency, and real-time performance. It provides important support for the transformation of modern power systems from centralized to distributed systems. Attached Figure Description

[0124] Figure 1 This is a device connection diagram provided in an embodiment of the present invention;

[0125] Figure 2 This is a flowchart illustrating an active power distribution network zoning method suitable for large-scale electric vehicle access provided by an embodiment of the present invention;

[0126] Figure 3 This is a schematic diagram of the structure of an active power distribution network zoning system suitable for large-scale electric vehicle access, provided by an embodiment of the present invention. Detailed Implementation

[0127] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0128] like Figure 1 The diagram illustrates a system connection for an active power distribution network zone suitable for large-scale electric vehicle access, according to an embodiment of the present invention. The system includes:

[0129] Data acquisition layer: responsible for collecting data in real time from sensors and electric vehicle charging station monitoring equipment;

[0130] Data processing layer: Implements data preprocessing;

[0131] Data analysis layer: used to extract node features and build the initial partitioned network model;

[0132] Optimization Execution Layer: Used to optimize the initial partitioning model and perform the final dynamic partitioning operation;

[0133] like Figure 2 As shown in the figure, an active power distribution network zoning method applicable to large-scale electric vehicle access according to an embodiment of the present invention includes the following steps:

[0134] Step S1: Obtain relevant data on the distribution network topology, including: distribution network topology structure, power data and voltage data of each node, output data of distributed power sources, and charging demand data of electric vehicles;

[0135] Step S2: Construct a composite feature matrix that includes electrical characteristics, spatiotemporal characteristics, and EV state:

[0136]

[0137] Among them, P G Q(t) represents the active power of the generator at time t. G (t) represents the reactive power of the generator at time t, P D (t) represents the active power of the load at time t, Q D V(t) represents the active power of the load at time t, V(t) is the magnitude of the node voltage at time t, and θ(t) is the phase angle of the node voltage at time t. The charging power of the electric vehicle at time t. To predict power, η EV (t) represents the battery charging efficiency; t = 1, 2, 3, ..., T;

[0138] Step S3: Standardize the composite feature matrix X using its standard deviation:

[0139]

[0140] Where, x * The data consists of standardized feature data, where N is the number of nodes and x... i The original feature data for node i. Let τ be the arithmetic mean of the feature data, and τ be the standard deviation of the feature data.

[0141] Step S4: Calculate the partition time-series weight matrix W of the dynamic partitioning combined network. d The matrix size is (D in ×D H ):

[0142]

[0143] Among them, P EV,i The EV charging and discharging power time series for node i;

[0144] Step S5.1: Use the augmented composite feature matrix As input data for the input encoding module;

[0145] Step S5.2: Extract node features:

[0146]

[0147] Step S5.3: For the extracted node key features H (l) Perform clustering operations to obtain cluster centers μ. k ;

[0148] Step S5.4: Measure the key features H of the node using the following formula. (l) With cluster center μ k The distance is used to obtain the characteristic distribution Q = {q} ik}:

[0149]

[0150] Where, q ik Let ζ be the probability of assigning node i to partition k, and ζ be the temperature parameter.

[0151] Step S6.1: Using the adjacency matrix A and the augmented composite characteristic matrix As input to the graph convolutional network in the dynamic partitioning ensemble network;

[0152] Step S6.2: Extract new feature representations of nodes:

[0153]

[0154] Among them, Z l The node feature representation learned by the l-th graph convolutional layer. To add a self-looping adjacency matrix, for The corresponding degree matrix, W (l-1) This is the weight matrix of the (l-1)th graph convolutional layer;

[0155] Step S6.3: Generate a comprehensive node data representation:

[0156]

[0157] in, α represents the node combined data generated by the (l-1)th layer encoding module and the graph convolutional layer, where α is the weight coefficient.

[0158] Step S6.4: Normalize the feature data and transform it into a probability distribution Z = {z ik}

[0159]

[0160] Step S7: Process the extracted node features to obtain the reconstructed data.

[0161]

[0162] in, To output the weight matrix of the l-th layer in the decoding module, This is the bias term for the l-th layer in the output decoding module;

[0163] Step S8: Calculate the multi-objective loss function:

[0164]

[0165] Where, β 1，2，3,4 λ is the weighting coefficient, and λ is the regularization coefficient;

[0166] Power balance loss

[0167]

[0168] Where C is the partition identifier vector and T is the time step. The interaction power between partition k and the main network;

[0169] Voltage deviation loss

[0170]

[0171] Where M is the number of nodes in a single partition, Π is the indicator function, and V ref Rated voltage;

[0172] Reconstruction loss

[0173]

[0174] Among them, L recon For reconstruction loss;

[0175] KL divergence between probability distribution and characteristic distribution:

[0176]

[0177] Step S9: Determine if the number of iterations is greater than the maximum number of iterations. If not, perform backpropagation according to the gradient descent method to update the parameters of the dynamic partitioning combination network, and then return to execute step S5.

[0178] Step S10: If true, obtain the final partition result identifier vector to realize the partitioning of the distribution network nodes.

[0179]

[0180] like Figure 3 As shown, in another embodiment of the present invention, an active power distribution network zoning system suitable for large-scale electric vehicle access is also provided, comprising:

[0181] Initial model construction module 201: Used to acquire relevant data of the distribution network topology, perform preprocessing, and construct the initial partition model;

[0182] Distribution Network Dynamic Partition Optimization Module 202: Calculates the partition time-series weight matrix and constructs a dynamic partition optimization model;

[0183] Distribution network partitioning module 203: The final partitioning is achieved by using a dynamic partitioning optimization model.

[0184] The specific implementation methods of each module can be referred to the description of the above method embodiments, and the embodiments of the present invention will not be repeated.

[0185] In another embodiment of the present invention, a computer-readable storage medium is also provided, on which program instructions are stored, which, when executed by a processor, implement the active distribution network zoning method applicable to large-scale electric vehicle access as described above.

[0186] It should be noted that, depending on the implementation needs, the various steps / components described in this application can be broken down into more steps / components, or two or more steps / components or parts of the operation of steps / components can be combined into new steps / components to achieve the purpose of this invention.

[0187] The methods described above according to the invention can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code originally stored on a remote recording medium or a non-transitory machine-readable medium and subsequently stored on a local recording medium, downloaded via a network. Thus, the methods described herein can be stored as software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., RAM, ROM, flash memory, etc.) capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the processing methods described herein. Furthermore, when a general-purpose computer accesses the code used to implement the processing shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for performing the processing shown herein.

[0188] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for zoning an active power distribution network suitable for large-scale electric vehicle access, characterized in that, The method includes the following steps: Step 1: Obtain relevant data on the distribution network topology, perform preprocessing, and construct an initial partitioning model; Step 2: Calculate the partition time-series weight matrix and construct a dynamic partition optimization model; Step 3: Use the dynamic partitioning optimization model to achieve the final partitioning.

2. The active power distribution network zoning method applicable to large-scale electric vehicle access according to claim 1, characterized in that, In step one, the relevant data of the distribution network topology are obtained, including: the distribution network topology structure, the power data and voltage data of each node, the output data of distributed power sources, and the charging demand data of electric vehicles.

3. The active distribution network zoning method applicable to large-scale electric vehicle access according to claim 1, characterized in that, In step one, the preprocessing steps are as follows: Step 1: Construct a composite feature matrix that includes electrical characteristics, spatiotemporal characteristics, and EV states: Among them, P G Q(t) represents the active power of the generator at time t. G (t) represents the reactive power of the generator at time t, P D (t) represents the active power of the load at time t, Q D V(t) represents the active power of the load at time t, V(t) is the magnitude of the node voltage at time t, and θ(t) is the phase angle of the node voltage at time t. The charging power of the electric vehicle at time t. To predict power, η EV (t) represents the battery charging efficiency; t = 1, 2, 3, ..., T; Step 2: Standardize the composite feature matrix X using its standard deviation: Where, x * The data consists of standardized feature data, where N is the number of nodes and x... i The original feature data for node i. Let τ be the arithmetic mean of the feature data, and τ be the standard deviation of the feature data.

4. The active distribution network zoning method applicable to large-scale electric vehicle access according to claim 1, characterized in that, In step one, the steps to construct the initial partitioning model are as follows: Step 1: Initialize the relevant network layers of the initial partitioned network, such as the number of neurons Q in the input layer. in The number of neurons Q in the hidden layer H The number of neurons in the output layer is Q. O ; Step 2: Initialize the input-to-hidden-layer connection weight matrix W of the initial partitioned network. in The matrix size is (Q in ×Q H ); Step 3: The input encoding module calculates the input data X using the following formula to compress and extract node features: Among them, H (l) Let f be the data representation learned by the l-th layer of the input encoding module, and f be the activation function. The weight matrix of the l-th layer in the input encoding module, This refers to the bias term of the l-th layer in the input encoding module; Step 4: The output decoding module processes the extracted node features using the following formula to obtain the reconstructed data. in, To output the weight matrix of the l-th layer in the decoding module, This is the bias term for the l-th layer in the output decoding module; Step 5: In each round of learning, the network parameters are continuously updated by minimizing the reconstruction loss using the following formula, and key node features are extracted: Among them, L recon For reconstruction loss; Step 6: Process the data using the following formula to generate new feature representations for the nodes: Among them, Z l This represents the node feature representation learned by the l-th graph convolutional layer. To add a self-looping adjacency matrix, for The corresponding degree matrix, W (l-1) Here is the weight matrix of the (l-1)th graph convolutional layer. α represents the node combined data generated by the (l-1)th layer encoding module and the graph convolutional layer, where α is the weight coefficient. Step 7: Normalize the feature data and transform it into a probability distribution Z = {z ik } Step 8: Obtain the partition result identifier vector C = {c i }:

5. The active power distribution network zoning method applicable to large-scale electric vehicle access according to claim 1, characterized in that, In step two, the specific steps for constructing the dynamic partitioning optimization model by calculating the partitioning time-series weight matrix are as follows: Step 1: Initialize the relevant parameters of the dynamic partitioning ensemble network, including the number of neurons D in the input layer. in and the number of hidden layer neurons D H The number of neurons in the output layer is D. O ; Step 2: Initialize the partition time-series weight matrix W of the dynamic partitioning combined network. d The matrix size is (D in ×D H ): Among them, P EV,i The EV charging and discharging power time series for node i; Step 3: Use augmented composite feature matrix As input data for the input encoding module; Step 4: Extract node features to obtain the initial partitioning model; Step 5: For the extracted node key features H (l) Perform clustering operations to obtain cluster centers μ. k ; Step 6: Measure the key features H of the nodes using the following formula. (l) With cluster center μ k The distance is used to obtain the characteristic distribution Q = {q} ik }: Where, q ik Let ζ be the probability of assigning node i to partition k, and ζ be the temperature parameter. Step 7: Calculate the KL divergence between the probability distribution and the characteristic distribution: Step 8: Design a multi-objective loss function with the objectives of minimizing power imbalance, voltage deviation, and reconstruction loss. Where, β 1，2，3,4 λ is the weighting coefficient, and λ is the regularization coefficient; Power balance loss Where C is the partition identifier vector and T is the time step. The interaction power between partition k and the main network; Voltage deviation loss Where M is the number of nodes in a single partition, Π is the indicator function, and V ref This is the rated voltage.

6. The active distribution network zoning method applicable to large-scale electric vehicle access according to claim 1, characterized in that, In step three, the steps to achieve the final partitioning using the dynamic partitioning optimization model are as follows: Step 1: Standardize the composite feature matrix X using its standard deviation; Step 2: Initialize the current iteration count (iter) of the dynamic partitioning combination network, and set the maximum iteration count (iter). max Initialize the parameters of each layer of the network; Step 3: Using the augmented composite feature matrix As the input to the input encoding module in the dynamic partitioning combined network, the deepest learned data H is processed. (3) Perform clustering to obtain cluster centers μ k Calculate the characteristic distribution Q = {q} ik }; Step 4: Using the adjacency matrix A and the augmented composite characteristic matrix As input to the graph convolutional network in the dynamic partitioning ensemble network, it extracts the features of the nodes. We obtain the probability distribution Z and the dynamic partitioning result identifier vector C; Step 5: Calculate the multi-objective loss function and determine whether the number of iterations is greater than the maximum number of iterations. If not, backpropagate according to the gradient descent method to update the parameters of the dynamic partitioning combination network, and then return to execute step (3); if it is true, obtain the final partitioning result identifier vector to realize the partitioning of the distribution network nodes.

7. An active power distribution network zoning optimization system suitable for large-scale electric vehicle access, characterized in that, include: Initial model construction module: used to acquire relevant data on the distribution network topology, perform preprocessing, and construct the initial partition model; Distribution network dynamic zoning optimization module: calculates the zoning time-series weight matrix and constructs a dynamic zoning optimization model; Distribution network partitioning module: The final partitioning is achieved using a dynamic partitioning optimization model.

8. The active power distribution network zoning optimization system suitable for large-scale electric vehicle access according to claim 7, characterized in that, The initial model construction module is used to obtain relevant data on the distribution network topology, perform preprocessing, and construct the initial partition model. The specific steps are as follows: Step 1: Obtain relevant data on the distribution network topology, including: distribution network topology structure, power data and voltage data of each node, output data of distributed power sources, and charging demand data of electric vehicles; Step 2: Construct a composite feature matrix that includes electrical characteristics, spatiotemporal characteristics, and EV states: Among them, P G Q(t) represents the active power of the generator at time t. G (t) represents the reactive power of the generator at time t, P D (t) represents the active power of the load at time t, Q D V(t) represents the active power of the load at time t, V(t) is the magnitude of the node voltage at time t, and θ(t) is the phase angle of the node voltage at time t. The charging power of the electric vehicle at time t. To predict power, η EV (t) represents the battery charging efficiency; t = 1, 2, 3, ..., T; Step 3: Standardize the composite feature matrix X using its standard deviation: Where, x * The data consists of standardized feature data, where N is the number of nodes and x... i The original feature data for node i. is the arithmetic mean of the feature data, and μ is the standard deviation of the feature data; Step 4: Initialize the relevant network layers of the initial partitioned network, such as the number of neurons Q in the input layer. in The number of neurons Q in the hidden layer H The number of neurons in the output layer is Q. O ; Step 5: Initialize the input-to-hidden-layer connection weight matrix W of the initial partitioned network. in The matrix size is (Q in ×Q H ); Step 6: The input encoding module calculates the input data using the following formula to compress and extract node features: Among them, H (l) Let f be the data representation learned by the l-th layer of the input encoding module, and f be the activation function. The weight matrix of the l-th layer in the input encoding module, The deviation of the l-th layer in the input encoding module; Step 7: The output decoding module processes the extracted node features using the following formula to obtain the reconstructed data. in, To output the weight matrix of the l-th layer in the decoding module, This is the deviation of the l-th layer in the output decoding module; Step 8: In each round of learning, continuously update the network parameters by minimizing the reconstruction loss using the following formula: Among them, L recon For reconstruction loss; Step 9: Process the data using the following formula to generate new feature representations for the nodes: Among them, Z l The node feature representation learned by the l-th graph convolutional layer. To add a self-looping adjacency matrix, for The corresponding degree matrix, W (l-1) Here is the weight matrix of the (l-1)th graph convolutional layer. α represents the node combined data generated by the (l-1)th layer encoding module and the graph convolutional layer, where α is the weight coefficient. Step 10: Normalize the feature data and transform it into a probability distribution Z = {z ik }: Step 11: Obtain the partition result identifier vector C = {c i }:

9. The active power distribution network zoning optimization system suitable for large-scale electric vehicle access according to claim 7, characterized in that, The steps for the distribution network dynamic zoning optimization module to calculate the zoning time-series weight matrix and construct the dynamic zoning optimization model are as follows: Step 1: Initialize the relevant parameters of the dynamic partitioning ensemble network, including the number of neurons D in the input layer. in and the number of hidden layer neurons D H The number of neurons in the output layer is D. O The number of partitions is k; Step 2: Initialize the partition time-series weight matrix W of the dynamic partitioning combined network. d The matrix size is (D in ×D H ): Among them, P EV,i The EV charging and discharging power time series for node i; Step 3: Use augmented composite feature matrix As input data for the input encoding module; Step 4: Extract node features to obtain the initial partitioning model; Step 5: For the extracted node key features H (l) Perform clustering operations to obtain cluster centers μ. k ; Step 6: Measure the key features H of the nodes using the following formula. (l) With cluster center μ k The distance is used to obtain the characteristic distribution Q = {q} ik }: Where, q ik Let ζ be the probability of assigning node i to partition k, and ζ be the temperature parameter. Step 7: Calculate the KL divergence between the probability distribution and the characteristic distribution: Step 8: Design a multi-objective loss function with the objectives of minimizing power imbalance, voltage deviation, and reconstruction loss. Where, β 1，2，3,4 λ is the weighting coefficient, and λ is the regularization coefficient; Power balance loss Where C is the partition identifier vector and T is the time step. The interaction power between partition k and the main network; Voltage deviation loss Where M is the number of nodes in a single partition, Π is the indicator function, and V ref This is the rated voltage.

10. The active power distribution network zoning optimization system suitable for large-scale electric vehicle access according to claim 7, characterized in that, The distribution network partitioning module uses a dynamic partitioning optimization model to achieve the final partitioning. The implementation steps are as follows: Step 1: Standardize the composite feature matrix X using its standard deviation; Step 2: Initialize the current iteration count (iter) of the dynamic partitioning combination network, and set the maximum iteration count (iter). max Initialize the parameters of each layer of the network; Step 3: Using the augmented composite feature matrix As the input to the input encoding module in the dynamic partitioning combined network, the deepest learned data H is processed. (3) Perform clustering to obtain cluster centers μ k Calculate the characteristic distribution Q = {q} ik }; Step 4: Using the adjacency matrix A and the augmented composite characteristic matrix As input to the graph convolutional network in the dynamic partitioning ensemble network, it extracts the features of the nodes. We obtain the probability distribution Z and the dynamic partitioning result identifier vector C; Step 5: Calculate the multi-objective loss function and determine whether the number of iterations is greater than the maximum number of iterations. If not, backpropagate according to the gradient descent method to update the parameters of the dynamic partitioning combination network, and then return to execute step (3); if it is true, obtain the final partitioning result identifier vector to realize the partitioning of the distribution network nodes.

Images