Power dispatching and optimizing method based on multi-granularity tree graph neural network

Abstract

The invention discloses a power system scheduling and optimizing method and system based on a multi-granularity tree diagram neural network. The method comprises the following steps: step 1, constructing a power grid state sample data set; 2, performing graph representation and feature extraction on the power grid data; step 3, constructing a hierarchical multi-granularity power grid information coarsening model; 4, constructing a power grid information graph disturbance model; step 5, constructing a tree neural network adaptive path selection model; and step 6, according to the multi-granularity tree neural network model, executing explanation and scheduling decisions of the power system. According to the method, graph network structure characteristics of a power system are considered, a power grid is modeled into graph data, and on the basis of an existing graph neural network model, a multi-granularity graph coarsening module, a learnable disturbance module and a self-adaptive path selection module are introduced to construct a tree-shaped model of a multi-layer structure; therefore, the state and representation of the power system of each granularity level, such as global and local scale in different ranges, are analyzed, and the prediction accuracy of the model and the interpretability of the decision are improved.

Description

Technical Field

[0001] The invention belongs to the field of power dispatching and neural network interpretability, and relates to a power dispatching and optimization method and system. Background Art

[0002] The dispatch of power systems is a complex and critical task that requires coordination and optimization of multiple factors while ensuring the safety and reliability of power supply. These factors include the status of power generation equipment, power demand forecasts at different load centers, and the operating conditions of transmission lines. With the expansion of the power system and the increase in the proportion of renewable energy, the imbalance between power supply and demand has intensified, making the dispatch of power systems more difficult. The complexity of the power grid stems from its multi-node, multi-variable, and highly dynamic characteristics. At the same time, its operating status is full of uncertainty when the power load changes and the power generation equipment is overhauled. The complexity of these problems requires the use of more reliable data analysis and dispatching technologies to optimize the operation and management of the power grid.

[0003] In the power system, the power grid can be naturally modeled as a graph, where nodes represent key equipment such as power stations, substations, and load centers, and edges represent transmission lines between these nodes. As a solution to the power dispatch problem, methods based on graph neural networks (GNNs) have been shown to have significant potential for analyzing the complex topological structure of power grids and predicting the state of power systems, and can effectively capture the complex connectivity and interaction relationships in the graph structure.

[0004] However, although GNN has achieved certain success in the optimization problem of power dispatch, its complex model structure is often regarded as a "black box" and it is difficult to provide an intuitive explanation for dispatch decisions. This lack of explainability has become a major bottleneck in power systems that require strict compliance and transparency. In addition, complex relationships in power systems often need to be modeled at multiple granularities, such as the operating status of a single power station, the balance of power generation and load in regions of different scales, and the power flow path across the entire network. When dispatching and making decisions, power dispatchers need to pay attention not only to the global structure and state of the power grid system, such as cross-regional power dispatch paths and load distribution characteristics, but also to take into account the local structure and state at different granularity levels. Therefore, it is urgent to develop a new solution that can not only give full play to the powerful ability of GNN in complex network structure modeling, but also have high interpretability at multiple granularities. By analyzing the power system at different scales and visualizing the decision path, the dispatch decision process can be presented in a clear and intuitive way, so as to meet the high requirements for safety, transparency and flexibility in modern power systems. Summary of the invention

[0005] In order to solve the above problems, the present invention provides a power dispatch decision method and system based on a multi-granularity dendrogram neural network.

[0006] The present invention considers the graph network structure characteristics of the power system, models the power grid as graph data, and introduces multi-granularity graph coarsening, learnable perturbation and adaptive path selection modules on the basis of the existing graph neural network model to build a multi-level tree model, thereby analyzing the power system status and representation at various granularity levels such as the global and local scales of different ranges, and improving the prediction accuracy of the model and the interpretability of the decision. The technical solution of the present invention is:

[0007] The power dispatch decision-making method based on multi-granularity dendrogram neural network includes the following steps:

[0008] S1. Construct a sample data set of power grid status;

[0009] Pandapower, a Python power system simulation tool, is used to construct a grid state sample data set. This data set is based on desensitized regional power grid data and is generated through a high-precision power dispatch simulation algorithm. It contains more than 5,000 simulation samples covering different working conditions. First, a grid simulation model is constructed that includes electrical characteristic parameters of each node in the grid, such as voltage level, load type, generator capacity, grid topology, such as substation connection relationship, transmission line parameters, and output information of each node under different load conditions. Then, based on typical daily load data, the output of different generators is randomly adjusted, and the power demand of load nodes is adjusted at the same time to ensure the balance between power generation and power consumption under different scenarios. Finally, the simulation data is screened through power flow calculation, and samples with converged power flow calculation and meeting the thermal stability constraints of the power system are included in the data set. Repeat the above steps many times, and the generated sample data set can be used to train and verify the graph neural network model for power grid dispatch optimization. This data set not only accurately simulates the dynamic behavior of the power system under different load conditions, but also ensures the wide applicability and representativeness of the data through rigorous power flow calculation and topological structure analysis, providing a basis for the training and verification of subsequent models.

[0010] S2.Graph representation and feature extraction of power grid data;

[0011] The various components of the power grid, such as power stations, substations, and load centers, are modeled as nodes of a graph, and the transmission lines between the nodes are regarded as edges of the graph, so that the power grid data is converted into a graph representation; the state of this graph can be represented as s = (A, H), where A is an adjacency matrix representing the node connection relationship, containing the connection information of N nodes, and H is a node feature matrix, with each row h i Represents the characteristics of node i (such as power generation, voltage, and other power characteristics).

[0012] In the feature extraction process, we use the Graph Attention Network (GAT) to model the relationship between nodes. First, calculate the correlation score c between nodes i and j ij , which reflects the strength of the association between the two:

[0013]

[0014] Among them, W 1 and W 2 The weight matrix used to transform the features, It represents the connection operation of features, and MLP is a multi-layer perceptron, which is used to generate correlation scores between nodes.

[0015] Next, by normalizing the correlation score, we get the attention weight β of node j to node i. ij :

[0016]

[0017] Among them, β ij It represents the relative importance of node j to node i in the process of information transmission.

[0018] The new feature representation p of node i i It is obtained by aggregating the information of its neighbor nodes:

[0019]

[0020] Among them, W g is the weight matrix used for feature transformation, σ is the activation function, and p i is the updated feature representation of node i.

[0021] This approach enables each node to dynamically adjust the way it aggregates information based on the importance of its neighbors, thereby better capturing the complex relationships and interdependencies between grid nodes when constructing a representation of the grid graph. The final node feature p i Combined with the information of neighbor nodes, it provides in-depth feature representation for the analysis of power grid characteristics under various states in power dispatching.

[0022] S3. Construct a hierarchical and multi-granular grid information coarsening model;

[0023] The present invention adopts a hierarchical graph coarsening module to construct the tree-like neural network model layer by layer. By gradually modeling the complex relationships in the power system at multiple granularities, the power grid graph structure is gradually simplified to obtain multi-granular structural information, such as the operating status of a single power station, the balance of power generation and load in areas of different degrees, and the power flow path across the entire network, so as to better cope with the different levels of information needs in power dispatching.

[0024] This module is based on the classic graph pooling strategy. It extracts the node embedding representation of each layer through the graph neural network (GNN) and performs graph coarsening on this basis. Specifically, we first extract the node embedding Z in each layer (l) , and its update rules are as follows:

[0025]

[0026] Among them, Z (0) =X is the input feature matrix, is an adjacency matrix containing self-loops, yes The degree matrix, W (l) is the weight matrix of the lth layer, and σ(·) is the activation function.

[0027] Next, we generate the cluster assignment matrix S through a multi-layer perceptron (MLP) (l) , which is used to merge nodes into coarse-grained clusters at each layer:

[0028] S (l) =softmax(MLP (l) (Z (l) ; Θ MLP )) (5)

[0029] where Θ MLP is a trainable parameter, Represents node v i The probability of belonging to cluster j.

[0030] Then, we use S (l) Generate the coarsened adjacency matrix A (l+1) and embedding matrix X (l+1) :

[0031]

[0032] Among them, N is the number of nodes in the current layer, K (l) is the number of clusters at layer l. In order to maintain the connectivity of the graph structure during the coarsening process, we introduce the edge prediction loss To constrain the coarsening process:

[0033]

[0034] Among them, A ij is the element in the original adjacency matrix, is the coarsened adjacency matrix. Through this hierarchical coarsening strategy, we can gradually extract information from the detail level of the power grid to a higher level of abstraction to meet different scheduling requirements.

[0035] S4. Construct a power grid information graph disturbance model;

[0036] Based on multi-granularity graph coarsening, the present invention introduces a learnable graph perturbation module to construct corresponding nodes in each level of the tree-like neural network model, aiming to improve the adaptability and robustness of the model to different operating states of the power grid by introducing tiny disturbances, and provide leaf nodes for the tree-like neural network model layer by layer.

[0037] This module generates a set of learnable perturbation matrices for each parent node So the coarsened allocation matrix Perform perturbations to generate multiple possible subgraph states:

[0038]

[0039] in, is the original cluster assignment matrix.

[0040] Based on the perturbed assignment matrix, we calculate the perturbed node embedding

[0041]

[0042] in, is the embedding representation of the kth parent node in the lth layer. In order to ensure that the perturbed embedding retains the original information and has diversity, we designed two regularization terms:

[0043] Similarity Regularization:

[0044]

[0045] Ensure that the perturbed embedding is close to the original embedding and preserve the structural information of the graph.

[0046] Diversity Regularization:

[0047]

[0048] Promoting the differences between different disturbances ensures that the model can cope with different power grid scenarios. Through these two regularization terms, we define the total disturbance loss:

[0049]

[0050] S5. Construct a tree-structured neural network adaptive path selection model;

[0051] The adaptive path selection module aims to dynamically select the optimal path from the root node to the leaf node and construct the routing paths between the nodes corresponding to the different levels of the tree-like neural network model, so as to provide the best decision-making solution for power dispatching.

[0052] We design a router for each non-leaf node in each layer to select the most informative path. Specifically, the router perturbs the embeddings generated by the module As input, the logits of path selection are generated through MLP:

[0053]

[0054] in, and is the weight matrix of the router, and is the bias term, and σ is the nonlinear activation function. Then, we calculate the probability distribution of path selection through softmax:

[0055]

[0056] Choose the path with the highest probability:

[0057]

[0058] To encourage diversity in path selection, an entropy-based regularization term is introduced

[0059]

[0060] This regularization term encourages the model to explore multiple possibilities when choosing a path, thereby enhancing the adaptability of the model.

[0061] S6. Execute power system interpretation and dispatch decision based on multi-granularity tree-like neural network model;

[0062] The present invention provides explanatory analysis from local to global through the multi-granularity paths generated by the adaptive path selection module, combined with node embedding at different levels.

[0063] For each grid instance G t , we calculate the probability of path selection in each layer and select the optimal path sequence To represent the multi-granularity explanation path from root to leaf. The specific calculation is as follows:

[0064]

[0065] Final Embed Used to generate forecast results for power dispatch:

[0066] h i =sofmax(f(Z final )) (19)

[0067] Among them, h i Represents the classification probability of the power grid status.

[0068] In the loss function design, we combined classification loss, edge prediction loss, perturbation regularization, and entropy regularization to ensure the multi-faceted performance of the model:

[0069]

[0070] Among them, α 1 , α 2 , α 3 Control the weights of different regularization terms respectively.

[0071] Through the above design, the present invention achieves efficient and explanatory decision-making capabilities in complex power dispatching scenarios. The model not only provides a variety of dispatching solutions, but also provides a clear explanation for each decision path, meeting the high requirements of the power system for transparency and compliance.

[0072] The electric power dispatching decision system based on multi-granularity tree graph neural network that implements the method of the present invention is characterized by comprising: a power grid state sample data set construction module, a power grid data graph representation and feature extraction module, a hierarchical multi-granularity power grid information coarsening module, a power grid information disturbance module, an adaptive tree model path selection module and a power system multi-granularity interpretation and dispatching decision module.

[0073] In the process of power grid dispatching decision-making, first, by using the power system simulation tool, a power grid simulation model is built according to the electrical characteristics and topological structure of the power grid, and the flow calculation is performed to screen the data to generate a sample data set of power grid status suitable for training and verification; then, the power grid data is modeled as graph data, in which the nodes represent power stations, substations, and load centers, and the edges represent the transmission lines between these nodes; a graph convolutional network (GCN) is used to extract features from the graph data, and a high-level feature representation of the power grid status is extracted by capturing the complex interactions between the nodes; based on these graph features, a multi-granularity graph coarsening module is used to gradually compress the graph structure of the power grid, and a multi-level graph representation from fine-grained to coarse-grained is generated by gradually modeling the complex relationships in the power system at multiple granularities, such as the operating status of a single power station, the balance of power generation and load in regions of different ranges, and the power grid status of the entire network. The power flow path is constructed to ensure that the model can take into account both local details and global structure. At the same time, different disturbances are added to the power grid state through the learnable graph perturbation module, and the corresponding nodes in each level of the tree neural network model are constructed to simulate the power grid operation under various working conditions and improve the robustness and adaptability of the model. The adaptive path selection module constructs the routing paths between the nodes corresponding to the different levels of the tree neural network model, and generates the most reasonable power dispatching plan by dynamically selecting the optimal path from the root to the leaf. A detailed explanation is provided for the dispatching plan based on the multi-granularity interpretation module to help users understand the basis of the dispatching decision. Finally, the performance of the model is evaluated, including indicators such as accuracy, robustness and interpretability, and it is applied to the fields of real-time dispatching, load management and emergency response of the power grid, demonstrating the practical utility of the model in power grid dispatching and effectively realizing the optimized management and reliability improvement of power grid dispatching.

[0074] The advantages of the present invention are as follows: by modeling the power grid data as graph data and using graph convolutional networks to extract high-level features, the complex topological relationships in the power grid are effectively captured, and the accuracy of scheduling decisions is improved; in addition, the present invention combines multi-granularity graph coarsening with adaptive path selection, and gradually models the complex relationships in the power system at multiple granularities to generate multi-level graph representations from fine to coarse granularity, such as the operating status of a single power station, the balance of power generation and load in regions of different degrees, and the power flow path across the entire network, ensuring that the model can take into account both local details and global structures, achieving multi-level decision-making from local to global, and making the scheduling plan more interpretable and transparent; at the same time, the learnable graph perturbation module further enhances the adaptability of the model under actual working conditions, provides a more flexible and robust solution for power scheduling, and effectively supports the efficient management and safe operation of modern power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 is a flow chart of the method of the present invention.

[0076] Figure 2 This is a general schematic diagram of the method of the present invention. In the figure, gradient green is used to represent the hierarchical structure of the graph neural network from shallow to deep with different depths; orange, purple, green, blue and other colors are used to represent the substructures with different independent information in the graph result of each layer of the graph neural network or the node embedding structure; orange, blue and gray with different depths are used to represent the feature matrix, adjacency matrix, node embedding matrix, clustering allocation matrix, etc. in the graph result of each layer of the graph neural network or the node embedding structure. DETAILED DESCRIPTION

[0077] The technical solution of the present invention is explained and described clearly and completely below in conjunction with the accompanying drawings.

[0078] Example 1

[0079] This embodiment relates to a power dispatch decision method based on a multi-granularity dendrogram neural network, comprising the following steps:

[0080] S1. Construct a sample data set of power grid status;

[0081] Pandapower, a Python power system simulation tool, is used to construct a grid state sample data set. This data set is based on desensitized regional power grid data and is generated through a high-precision power dispatch simulation algorithm. It contains more than 5,000 simulation samples covering different working conditions. First, a grid simulation model is constructed that includes electrical characteristic parameters of each node in the grid, such as voltage level, load type, generator capacity, grid topology, such as substation connection relationship, transmission line parameters, and output information of each node under different load conditions. Then, based on typical daily load data, the output of different generators is randomly adjusted, and the power demand of load nodes is adjusted at the same time to ensure the balance between power generation and power consumption under different scenarios. Finally, the simulation data is screened through power flow calculation, and samples with converged power flow calculation and meeting the thermal stability constraints of the power system are included in the data set. Repeat the above steps many times, and the generated sample data set can be used to train and verify the graph neural network model for power grid dispatch optimization. This data set not only accurately simulates the dynamic behavior of the power system under different load conditions, but also ensures the wide applicability and representativeness of the data through rigorous power flow calculation and topological structure analysis, providing a basis for the training and verification of subsequent models.

[0082] S2.Graph representation and feature extraction of power grid data;

[0083] The various components of the power grid, such as power stations, substations, and load centers, are modeled as nodes of a graph, and the transmission lines between the nodes are regarded as edges of the graph, so that the power grid data is converted into a graph representation; the state of this graph can be represented as s = (A, H), where A is an adjacency matrix representing the node connection relationship, containing the connection information of N nodes, and H is a node feature matrix, with each row h i Represents the characteristics of node i (such as power generation, voltage, and other power characteristics).

[0084] In the feature extraction process, we use the Graph Attention Network (GAT) to model the relationship between nodes. First, calculate the correlation score c between nodes i and j ij , which reflects the strength of the association between the two:

[0085]

[0086] Among them, W 1 and W 2 The weight matrix used to transform the features, It represents the connection operation of features, and MLP is a multi-layer perceptron, which is used to generate correlation scores between nodes.

[0087] Next, by normalizing the correlation score, we get the attention weight β of node j to node i. ij :

[0088]

[0089] Among them, β ij It represents the relative importance of node j to node i in the process of information transmission.

[0090] The new feature representation p of node i i It is obtained by aggregating the information of its neighbor nodes:

[0091]

[0092] Among them, W g is the weight matrix used for feature transformation, σ is the activation function, and p i is the updated feature representation of node i.

[0093] This approach enables each node to dynamically adjust the way it aggregates information based on the importance of its neighbors, thereby better capturing the complex relationships and interdependencies between grid nodes when constructing a representation of the grid graph. The final node feature p i Combined with the information of neighbor nodes, it provides in-depth feature representation for the analysis of power grid characteristics under various states in power dispatching.

[0094] S3. Construct a hierarchical and multi-granular grid information coarsening model;

[0095] The present invention adopts a hierarchical graph coarsening module to construct the tree-like neural network model layer by layer. By gradually modeling the complex relationships in the power system at multiple granularities, the power grid graph structure is gradually simplified to obtain multi-granular structural information, such as the operating status of a single power station, the balance of power generation and load in areas of different degrees, and the power flow path across the entire network, so as to better cope with the different levels of information needs in power dispatching.

[0096] This module is based on the classic graph pooling strategy. It extracts the node embedding representation of each layer through the graph neural network (GNN) and performs graph coarsening on this basis. Specifically, we first extract the node embedding Z in each layer (l) , and its update rules are as follows:

[0097]

[0098] Among them, Z (0) =X is the input feature matrix, is an adjacency matrix containing self-loops, yes The degree matrix, W (l) is the weight matrix of the lth layer, and σ(·) is the activation function.

[0099] Next, we generate the cluster assignment matrix S through a multi-layer perceptron (MLP) (l) , which is used to merge nodes into coarse-grained clusters at each layer:

[0100] S (l) =softmax(MLP (l) (z (l) ; Θ MLP )) (5)

[0101] where Θ MLP is a trainable parameter, Represents node v i The probability of belonging to cluster j.

[0102] Then, we use S (l) Generate the coarsened adjacency matrix A (l+1) and embedding matrix X (l+1) :

[0103]

[0104] Among them, N is the number of nodes in the current layer, K (l)is the number of clusters at layer l. In order to maintain the connectivity of the graph structure during the coarsening process, we introduce the edge prediction loss To constrain the coarsening process:

[0105]

[0106] Among them, A ij is the element in the original adjacency matrix, is the coarsened adjacency matrix. Through this hierarchical coarsening strategy, we can gradually extract information from the detail level of the power grid to a higher level of abstraction to meet different scheduling requirements.

[0107] S4. Constructing a power grid information graph disturbance model;

[0108] Based on multi-granularity graph coarsening, the present invention introduces a learnable graph perturbation module to construct corresponding nodes in each level of the tree-like neural network model, aiming to improve the adaptability and robustness of the model to different operating states of the power grid by introducing tiny disturbances, and provide leaf nodes for the tree-like neural network model layer by layer.

[0109] This module generates a set of learnable perturbation matrices for each parent node So the coarsened allocation matrix Perform perturbations to generate multiple possible subgraph states:

[0110]

[0111] in, is the original cluster assignment matrix.

[0112] Based on the perturbed assignment matrix, we calculate the perturbed node embedding

[0113]

[0114] in, is the embedding representation of the kth parent node in the lth layer. In order to ensure that the perturbed embedding retains the original information and has diversity, we designed two regularization terms:

[0115] Similarity Regularization:

[0116]

[0117] Ensure that the perturbed embedding is close to the original embedding and preserve the structural information of the graph.

[0118] Diversity Regularization:

[0119]

[0120] Promoting the differences between different disturbances ensures that the model can cope with different power grid scenarios. Through these two regularization terms, we define the total disturbance loss:

[0121]

[0122] S5. Construct a tree-structured neural network adaptive path selection model;

[0123] The adaptive path selection module aims to dynamically select the optimal path from the root node to the leaf node and construct the routing paths between the nodes corresponding to the different levels of the tree-like neural network model, so as to provide the best decision-making solution for power dispatching.

[0124] We design a router for each non-leaf node in each layer to select the most informative path. Specifically, the router perturbs the embeddings generated by the module As input, the logits of path selection are generated through MLP:

[0125]

[0126] in, and is the weight matrix of the router, and is the bias term, and σ is the nonlinear activation function. Then, we calculate the probability distribution of path selection through softmax:

[0127]

[0128] Choose the path with the highest probability:

[0129]

[0130] To encourage diversity in path selection, an entropy-based regularization term is introduced

[0131]

[0132] This regularization term encourages the model to explore multiple possibilities when choosing a path, thereby enhancing the adaptability of the model.

[0133] S6. Execute power system interpretation and dispatch decision based on multi-granularity tree-like neural network model;

[0134] The present invention provides explanatory analysis from local to global through the multi-granularity paths generated by the adaptive path selection module, combined with node embedding at different levels.

[0135] For each grid instance G t, we calculate the probability of path selection in each layer and select the optimal path sequence To represent the multi-granularity explanation path from root to leaf. The specific calculation is as follows:

[0136]

[0137] Final Embed Used to generate forecast results for power dispatch:

[0138] h i =softmax(f(z final )) (19)

[0139] Among them, h i Represents the classification probability of the power grid status.

[0140] In the loss function design, we combined classification loss, edge prediction loss, perturbation regularization, and entropy regularization to ensure the multi-faceted performance of the model:

[0141]

[0142] Among them, α 1 , α 2 , α 3 Control the weights of different regularization terms respectively.

[0143] Through the above design, the present invention achieves efficient and explanatory decision-making capabilities in complex power dispatching scenarios. The model not only provides a variety of dispatching solutions, but also provides a clear explanation for each decision path, meeting the high requirements of the power system for transparency and compliance.

[0144] Example 2

[0145] This embodiment relates to a power dispatch decision system based on a multi-granularity dendrogram neural network, which is used to implement the method of embodiment 1, including:

[0146] A power grid state sample data set construction module, which constructs a power grid state sample data set;

[0147] A graph representation and feature extraction module for power grid data, which performs graph representation and feature extraction on power grid data;

[0148] Hierarchical multi-granularity power grid information coarsening module, building a hierarchical multi-granularity power grid information coarsening model;

[0149] Power grid information disturbance module, building a power grid information graph disturbance model;

[0150] Adaptive tree model path selection module, building a tree neural network adaptive path selection model;

[0151] The multi-granularity interpretation and dispatch decision module of the power system performs the interpretation and dispatch decision of the power system based on the multi-granularity tree-like neural network model.

[0152] The above modules respectively include the contents of S1 to S6 of the method of the present invention in sequence.

[0153] The contents described in the embodiments of this specification are merely an enumeration of the implementation forms of the inventive concept. The protection scope of the present invention should not be regarded as limited to the specific forms described in the embodiments. The protection scope of the present invention also extends to the equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A power dispatching and optimization method based on a multi-granularity dendrogram neural network includes the following steps: Step 1. Construct a grid status sample dataset; Step 2. Graph representation and feature extraction of power grid data; Step 3. Construct a hierarchical multi-granularity grid information coarsening model; Step 4. Construct a power grid information graph disturbance model; Step 5. Construct a tree-structured neural network adaptive path selection model; Step 6. Perform power system interpretation and dispatch decision based on the multi-granularity tree neural network model.

2. The power dispatching and optimization method based on multi-granularity dendrogram neural network according to claim 1 is characterized in that: Step 1 specifically includes: The Python power system simulation tool Pandapower is used to construct a grid state sample dataset. This dataset is based on desensitized regional power grid data and is generated through a high-precision power dispatch simulation algorithm, containing more than 5,000 simulation samples covering different working conditions. First, a grid simulation model is constructed that includes electrical characteristic parameters of each node in the grid, such as voltage level, load type, generator capacity, grid topology, such as substation connection relationship, transmission line parameters, and output information of each node under different load conditions. Then, based on typical daily load data, the output of different generators is randomly adjusted, and the power demand of load nodes is adjusted at the same time to ensure the balance between power generation and power consumption under different scenarios. Finally, the simulation data is screened through power flow calculation, and samples with converged power flow calculation and meeting the thermal stability constraints of the power system are included in the dataset. The above steps are repeated many times, and the generated sample dataset can be used to train and verify the graph neural network model for power grid dispatch optimization. This dataset not only accurately simulates the dynamic behavior of the power system under different load conditions, but also ensures the wide applicability and representativeness of the data through rigorous power flow calculation and topological structure analysis, providing a basis for the training and verification of subsequent models.

3. The power dispatching and optimization method based on multi-granularity dendrogram neural network according to claim 1 is characterized in that: Step 2 specifically includes: The various components of the power grid, such as power stations, substations, and load centers, are modeled as nodes of a graph, and the transmission lines between the nodes are regarded as edges of the graph, so that the power grid data is converted into a graph representation; the state of this graph can be represented as s = (A, H), where A is an adjacency matrix representing the node connection relationship, containing the connection information of N nodes, and H is a node feature matrix, with each row h i Represents the characteristics of node i (such as power generation, voltage and other power characteristics); In the feature extraction process, the graph attention network (GAT) is used to model the relationship between nodes; first, the correlation score c between nodes i and j is calculated ij , which reflects the strength of the association between the two: c ij =MLP(W1·h i ⊕W2·h j ) (1) Among them, W1 and W2 are used to transform the weight matrix of the feature, ⊕ represents the connection operation of the feature, and MLP is a multi-layer perceptron used to generate the correlation score between nodes; Next, by normalizing the correlation score, we get the attention weight β of node j to node i. ij : Among them, β ij Indicates the relative importance of node j to node i in the process of information transmission; The new feature representation p of node i i It is obtained by aggregating the information of its neighbor nodes: Among them, W g is the weight matrix used for feature transformation, σ is the activation function, and p i is the updated feature representation of node i; This method enables each node to dynamically adjust the way information is aggregated according to the importance of its neighbors, so that when constructing the representation of the power grid graph, the complex relationships and interdependencies between power grid nodes can be better captured; the final node feature p i Combined with the information of neighbor nodes, it provides in-depth feature representation for the analysis of power grid characteristics under various states in power dispatching.

4. The power dispatching and optimization method based on multi-granularity dendrogram neural network according to claim 1 is characterized in that: Step 3 specifically includes: A hierarchical graph coarsening module is used to construct the tree-like neural network model layer by layer. The complex relationships in the power system are gradually modeled at multiple granularities to gradually simplify the power grid graph structure and obtain multi-granular structural information, such as the operating status of a single power station, the balance of power generation and load in different ranges, and the power flow path of the entire network, so as to better cope with the different levels of information needs in power dispatching. This module is based on the classic graph pooling strategy. It extracts the node embedding representation of each layer through the graph neural network (GNN) and performs graph coarsening on this basis. Specifically, the node embedding Z is first extracted in each layer. (l) , and its update rules are as follows: Among them, Z (0) =X is the input feature matrix, is an adjacency matrix containing self-loops, yes The degree matrix, W (l) is the weight matrix of the lth layer, σ(·) is the activation function; Next, the clustering assignment matrix S is generated by a multi-layer perceptron (MLP) (l) , which is used to merge nodes into coarse-grained clusters at each layer: S (l) =softmax(MLP (l) (WITH (l) ;Θ MLP )) (5) where Θ MLP is a trainable parameter, Represents node v i The probability of belonging to cluster j; Then, using S (l) Generate the coarsened adjacency matrix A (l+1) and embedding matrix X (l+1) : Among them, N is the number of nodes in the current layer, K (l) is the number of clusters in the lth layer; in order to maintain the connectivity of the graph structure during the coarsening process, the edge prediction loss is introduced To constrain the coarsening process: Among them, A ij is the element in the original adjacency matrix, is the coarsened adjacency matrix; through this hierarchical coarsening strategy, information can be gradually extracted from the detail level of the power grid to a higher level of abstraction to meet different scheduling requirements.

5. The power dispatching and optimization method based on multi-granularity dendrogram neural network according to claim 1, characterized in that: Step 4 specifically includes: Based on the multi-granularity graph coarsening, a learnable graph perturbation module is introduced to construct the corresponding nodes in each layer of the tree-like neural network model. The aim is to improve the adaptability and robustness of the model to different operating states of the power grid by introducing small perturbations, and provide leaf nodes for the tree-like neural network model layer by layer. This module generates a set of learnable perturbation matrices for each parent node. So the coarsened allocation matrix Perform perturbations to generate multiple possible subgraph states: in, is the original cluster assignment matrix; Based on the perturbed assignment matrix, calculate the perturbed node embedding in, is the embedding representation of the kth parent node in the lth layer; to ensure that the perturbed embedding retains both the original information and diversity, two regularization terms are designed: Similarity Regularization: Ensure that the perturbed embedding is close to the original embedding and preserve the structural information of the graph; Diversity Regularization: Promote the differences between different disturbances to ensure that the model can cope with different power grid scenarios; through these two regularization terms, the total disturbance loss is defined:

6. The power dispatching and optimization method based on multi-granularity dendrogram neural network according to claim 1 is characterized in that: Step 5 specifically includes: The adaptive path selection module aims to dynamically select the optimal path from the root node to the leaf node and construct the routing paths between the nodes corresponding to different levels of the tree-like neural network model, thereby providing the best decision-making solution for power dispatching; Design a router for each non-leaf node in each layer to select the most informative path; specifically, the router perturbs the embedding generated by the module As input, the logits of path selection are generated through MLP: in, and is the weight matrix of the router, and is the bias term, σ is the nonlinear activation function; then, the probability distribution of path selection is calculated by softmax: Choose the path with the highest probability: To encourage diversity in path selection, an entropy-based regularization term is introduced This regularization term encourages the model to explore multiple possibilities when choosing a path, thereby enhancing the adaptability of the model.

7. The power dispatching and optimization method based on multi-granularity dendrogram neural network according to claim 1 is characterized in that: Step 6 specifically includes: The multi-granularity paths generated by the adaptive path selection module, combined with different levels of node embedding, provide interpretative analysis from local to global. For each grid instance G t , calculate the probability of path selection in each layer and select the optimal path sequence To represent the multi-granularity explanation path from root to leaf; the specific calculation is as follows: Final Embed Used to generate forecast results for power dispatch: Among them, h i Represents the classification probability of the power grid status; In the loss function design, classification loss, edge prediction loss, perturbation regularization and entropy regularization are combined to ensure the multi-faceted performance of the model: Among them, α1, α2, and α3 control the weights of different regularization terms respectively.

8. The power system dispatching and optimization system based on multi-granularity tree graph neural network is characterized by: include: A power grid state sample data set construction module, which constructs a power grid state sample data set; A graph representation and feature extraction module for power grid data, which performs graph representation and feature extraction on power grid data; Hierarchical multi-granularity power grid information coarsening module, building a hierarchical multi-granularity power grid information coarsening model; Power grid information disturbance module, building a power grid information graph disturbance model; Adaptive tree model path selection module, building a tree neural network adaptive path selection model; The multi-granularity interpretation and dispatch decision module of the power system performs the interpretation and dispatch decision of the power system based on the multi-granularity tree-like neural network model.

Images