Power system state estimation method and device, medium and equipment

Abstract

The invention discloses a power system state estimation method and device, a medium and equipment in the technical field of power system state estimation. The method comprises the following steps: acquiring power grid topological structure information in a power system, and SCADA data and PMU data of each equipment node; constructing an adjacency matrix according to the topological structure information of the power grid, and performing data preprocessing on the adjacency matrix to obtain a processed adjacency matrix; performing feature extraction on the SCADA data and the PMU data through an enhanced graph convolution module and the processed adjacent matrix to obtain SCADA features and PMU features; performing feature fusion on the SCADA features and the PMU features through a dense connection module to obtain fusion features; flattening the fusion features through a flattening layer to obtain flattened features; and mapping the flattened features into state estimation output of each equipment node in the power system by using a multi-layer full-connection network to obtain a state estimation value of each equipment node.

Description

Technical Field

[0001] The present invention relates to the technical field of power system state estimation, and particularly relates to a power system state estimation method, device, medium and equipment. Background Art

[0002] In recent years, in order to efficiently process exponentially growing power system data and ensure the stable and reliable operation of the power system, higher requirements are put forward for the accuracy and efficiency of state estimation. At present, two measurement systems are configured in the power system, namely the Supervisory Control and Data Acquisition (SCADA) system and the Wide-Area Measurement System (WAMS) based on Phasor Measurement Units (PMUs); the data refresh frequency of the SCADA system is usually 0.5 - 5Hz, with a large communication delay and low data accuracy; the PMU measurement has high accuracy and small delay, and the PMU update period is usually in milliseconds; at the same time, the device node voltage phasor and branch current phasor as state quantities can be directly measured by the PMU.

[0003] Due to technical and economic limitations, although the PMU has better performance than the SCADA in many dimensions such as timeliness, accuracy, and comprehensiveness, at the current stage, the PMU cannot completely replace the SCADA, and the two measurement systems of PMU and SCADA will coexist for a quite long period in the future; currently, in China's distribution network, PMUs are only configured at key device nodes, and measurement based only on PMUs cannot meet the observability requirements. Therefore, effectively fusing PMU data and SCADA data and making full use of measurement data to improve state estimation accuracy is a relatively reasonable method at the present stage; however, due to the large differences between the two sets of measurement data, directly mixing and applying the two sets of data without processing will not only not improve the state estimation accuracy, but will instead lead to a decrease in accuracy; therefore, how to propose an effective measurement data fusion strategy to cope with the data differences is a key task in the early stage of state estimation.

[0004] There are numerous device nodes and a large scale in the power system. Traditional state estimation methods such as the weighted least squares method need to complete complex mathematical calculations while processing a large amount of real-time measurement data, and it is difficult to adapt to the access of renewable energy and the increase in data volume in modern power systems; when existing methods process the graph structure data of the power system, they lack the ability to model the complex relationships between device nodes and have insufficient ability to fuse and process multi-source data, making it difficult to make full use of the advantages of different data sources. Summary of the Invention

[0005] The object of the present invention is to overcome the deficiencies in the prior art, and provide a power system state estimation method, device, medium and equipment, which can effectively extract the spatial features of the power system topology structure, can capture the long-distance dependence relationship between nodes, can better learn the complex features and dynamic characteristics of the power system, and can significantly improve the generalization ability of the model by fusing multi-source data, thereby effectively improving the accuracy and robustness of state estimation.

[0006] To solve the above technical problems, the present invention is implemented by the following technical solutions:

[0007] In a first aspect, the present invention provides a power system state estimation method, including:

[0008] Obtain the power grid topology structure information, SCADA data and PMU data of each equipment node in the power system;

[0009] Construct an adjacency matrix according to the power grid topology structure information, and perform data preprocessing on the adjacency matrix to obtain a processed adjacency matrix;

[0010] Input the processed adjacency matrix, SCADA data and PMU data into a trained power system state estimation model: extract features of the SCADA data and PMU data through an enhanced graph convolution module and the processed adjacency matrix to obtain SCADA features and PMU features; through a dense connection module, perform feature fusion on the SCADA features and PMU features to obtain fused features; flatten the fused features through a flattening layer to obtain flattened features; use a multi-layer fully connected network to map the flattened features to the state estimation outputs of each equipment node in the power system to obtain the state estimation values of each equipment node.

[0011] Preferably, construct a topological relationship graph according to the power grid topology structure information, and the topological relationship graph has the following expression: ; wherein, represents the set of equipment nodes, represents the set of connection edges between each equipment node; Construct an adjacency matrix according to the topological relationship graph which is used to represent the connection relationship between each equipment node; The data preprocessing of the adjacency matrix includes: Add self-loops to the adjacency matrix to obtain an adjacency matrix with self-loops added; Use the degree matrix of the adjacency matrix with self-loops added to normalize the adjacency matrix with self-loops added to obtain a processed adjacency matrix; Adding self-loops to the adjacency matrix is achieved through the following formula: ; where, represents the adjacency matrix with self-loops added, represents the adjacency matrix, represents the identity matrix; The degree matrix of the adjacency matrix with self-loops added is obtained through the following formula: ; where, represents the degree matrix of the adjacency matrix with self-loops added, represents the function to create a diagonal matrix, represents the row index of the equipment node in the power system, represents the column index of the equipment node in the power system, represents the th row and th column element of the adjacency matrix with self-loops added; Normalizing the adjacency matrix with self-loops added is achieved through the following formula: ; where, represents the processed adjacency matrix.

[0012] Preferably, the enhanced graph convolution module includes a linear transformation layer, a graph convolution layer, a multi-head attention mechanism, a residual connection layer, a batch normalization layer, a Dropout module, and a LeakyReLU layer; the data processing flow of the enhanced graph convolution module includes: Aligning the dimensions of the SCADA data and PMU data through the linear transformation layer to obtain the aligned PMU data and aligned SCADA data; Performing convolution processing on the aligned PMU data and aligned SCADA data through the graph convolution layer and the processed adjacency matrix to obtain the convolved PMU data and convolved SCADA data; Inputting the convolved PMU data and convolved SCADA data into the multi-head attention mechanism to obtain the attention output; Fusing the SCADA data and PMU data with the output of the multi-head attention mechanism respectively through the residual connection layer to obtain the SCADA data fusion result and PMU data fusion result; Sequentially inputting the SCADA data fusion result and PMU data fusion result into the batch normalization layer, Dropout module, and LeakyReLU layer for processing to obtain the SCADA features and PMU features; The dimension alignment is achieved through the following formula: ; Among them, represents a trainable weight matrix, , represents PMU data, represents SCADA data, , represents the aligned PMU data, represents the aligned SCADA data; The convolution process is implemented by the following formula: ; Among them, represents the processed adjacency matrix, , represents the PMU data after convolution, represents the SCADA data after convolution; The expression of the multi-head attention mechanism is as follows: ; ; ; ; ; ; ; Among them, represents the query matrix, represents the key matrix, represents the value matrix, represents the weight matrix of the query matrix, represents the weight matrix of the key matrix, represents the weight matrix of the value matrix, represents the splitting function, represents the split query matrix, represents the split key matrix, represents the split value matrix, represents the dimension of the key matrix, represents the transpose of the split key matrix, represents the attention score, represents the activation function, represents the attention probability, represents the context vector, represents the concatenation function, represents the context concatenation result, represents the weight matrix of the linear transformation, Represents the output of the multi-head attention mechanism; The expression of the residual connection layer is as follows: ; Where, , Represents the PMU data fusion result, Represents the SCADA data fusion result; The expression of the batch normalization layer is as follows: ; Where, Represents the output of the batch normalization layer, Represents the batch normalization operation; The expression of the Dropout module is as follows: ; Where, Represents the output of the Dropout module, Represents the Dropout operation; The expression of the LeakyReLU layer is as follows: ; Where, , Represents the PMU feature, Represents the PMU feature, Represents the LeakyReLU activation function.

[0013] Preferably, the expression of the dense connection module is as follows: ; Where, Represents the fused feature, Represents the concatenation function, Represents the PMU feature, Represents the PMU feature.

[0014] Preferably, the expression of the flatten layer is as follows: ; Where, Represents the flattened feature, Represents the flatten operation function, Represents the fused feature.

[0015] Preferably, the multi-layer fully connected network uses the LeakyReLU activation function, and the data processing flow of the multi-layer fully connected network includes: Feature mapping is performed on the flattened feature according to the following formula to obtain the state variable of each device node: ; Among them, represents the state variable of the th device node, represents the trainable weight of the fully connected layer, represents the bias of the fully connected layer, represents the flattened feature in the th flattened feature element; Perform a non-linear transformation on the state variable of each device node according to the following formula to obtain the state estimation value of each device node: ; Among them, represents the state estimation value of the th device node, represents a constant.

[0016] Preferably, the power system state estimation model uses a mean square error loss function for model training, and the expression of the mean square error loss function is as follows: ; Among them, represents the loss of the power system state estimation model, represents the total number of training samples, represents the state of the th device node, represents the state estimation value of the th device node.

[0017] In a second aspect, the present invention provides a power system state estimation device, including:

[0018] A data acquisition module for: acquiring power grid topology structure information, SCADA data, and PMU data of each device node in the power system;

[0019] A data preprocessing module for: constructing an adjacency matrix according to the power grid topology structure information, and performing data preprocessing on the adjacency matrix to obtain a processed adjacency matrix;

[0020] A feature extraction module for: extracting features from the SCADA data and PMU data through an enhanced graph convolution module and the processed adjacency matrix to obtain SCADA features and PMU features;

[0021] A feature fusion module for: performing feature fusion on the SCADA features and PMU features through a dense connection module to obtain fusion features;

[0022] A feature flattening module, configured to: flatten the fused features through a flattening layer to obtain flattened features;

[0023] A state estimation module, configured to: map the flattened features to the state estimation outputs of each device node in the power system by using a multi-layer fully connected network to obtain the state estimation values of each device node.

[0024] In a third aspect, the present invention provides a computer-readable storage medium, on which computer instructions are stored. When the computer instructions are executed by a processor, the steps of any one of the power system state estimation methods in the first aspect are implemented.

[0025] In a fourth aspect, the present invention provides a computer device, including:

[0026] A memory, configured to store computer instructions;

[0027] A processor, configured to execute the computer instructions to implement the steps of any one of the power system state estimation methods in the first aspect.

[0028] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0029] 1. For the power system state estimation method provided by the present invention, by fusing PMU and SCADA data and using an enhanced graph convolutional Transformer model, the accuracy and robustness of power system state estimation are significantly improved; the multi-source data fusion enables the model to make full use of the advantages of the two data sources and capture the system state features more comprehensively; the graph convolutional network GCN effectively extracts the spatial features of the power system topology, and the multi-head attention mechanism captures the long-range dependencies between nodes, further improving the estimation accuracy;

[0030] 2. For the power system state estimation device provided by the present invention, by setting a data acquisition module, a data preprocessing module, a feature extraction module, a feature fusion module, a feature flattening module and a state estimation module, the power system state estimation is jointly realized. The modular design and calculation efficiency optimization enable the device to have high calculation efficiency, can quickly process large-scale data, has good generalization ability and adaptability, can better meet the actual needs of power system state estimation, and has practical significance and good application prospects;

[0031] 3. The computer-readable storage medium and computer device provided by the present invention can execute the steps of the power system state estimation method provided by the present invention. Description of the Drawings

[0032] Figure 1 It is a flowchart of the power system state estimation method according to an embodiment of the present invention;

[0033] Figure 2 Schematic structural diagram of a power system state estimation device provided according to an embodiment of the present invention;

[0034] Figure 3 Comparison diagram of reactive power prediction results between the power system state estimation method and the baseline method provided according to an embodiment of the present invention;

[0035] Figure 4 Comparison diagram of active power prediction results between the power system state estimation method and the baseline method provided according to an embodiment of the present invention. Specific embodiments

[0036] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific features in the embodiments of the present application and the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations on the technical solution of the present application. Without conflict, the technical features in the embodiments of the present application and the embodiments can be combined with each other.

[0037] It should be noted that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after.

[0038] Embodiment 1:

[0039] An embodiment of the present invention discloses a power system state estimation method. Referring to Figure 1 as shown, the method specifically includes the following steps:

[0040] S1, obtaining the grid topology structure information, SCADA data and PMU data of each equipment node in the power system;

[0041] S2, constructing an adjacency matrix according to the grid topology structure information, and performing data preprocessing on the adjacency matrix to obtain a processed adjacency matrix;

[0042] S3, inputting the processed adjacency matrix, SCADA data and PMU data into a trained power system state estimation model: extracting features of the SCADA data and PMU data through an enhanced graph convolution module and the processed adjacency matrix to obtain SCADA features and PMU features; performing feature fusion on the SCADA features and PMU features through a dense connection module to obtain fused features; flattening the fused features through a flattening layer to obtain flattened features; using a multi-layer fully connected network to map the flattened features to the state estimation outputs of each equipment node in the power system to obtain the state estimation values of each equipment node.

[0043] Specifically, in step S1, obtain the PMU data and SCADA data of each device node in the power system:

[0044]

[0045] Among them, represents the batch size, represents the number of device nodes, represents the number of PMU input features, represents the number of SCADA input features.

[0046] In step S2, construct a topological relationship graph according to the grid topological structure information. The expression of the topological relationship graph is as follows: ;

[0047] Among them, represents the set of device nodes, represents the set of connection edges between each device node; According to the topological relationship graph construct an adjacency matrix, and the adjacency matrix is used to represent the connection relationship between each device node.

[0048] The data preprocessing of the adjacency matrix includes: Add self-loops to the adjacency matrix to obtain an adjacency matrix with self-loops added; Use the degree matrix of the adjacency matrix with self-loops added to normalize the adjacency matrix with self-loops added to ensure the stability of feature propagation, and obtain the processed adjacency matrix.

[0049] Adding self-loops to the adjacency matrix is achieved through the following formula:

[0050] Among them, represents the adjacency matrix with self-loops added, represents the adjacency matrix, represents the identity matrix.

[0051] The degree matrix of the adjacency matrix with self-loops added is obtained through the following formula:

[0052]

[0053] Among them, represents the degree matrix of the adjacency matrix with self-loops added, which is used to normalize the adjacency matrix to ensure the stability of feature propagation; represents the function of creating a diagonal matrix, Indicates the row index of the device node in the power system, Indicates the column index of the device node in the power system, Indicates the row and

[0054] The normalization of the adjacency matrix with self-loops added is achieved through the following formula:

[0055] where, Indicates the processed adjacency matrix.

[0056] In step S3, the power system state estimation model includes an enhanced graph convolution module, a dense connection module, a flattening layer, and a multi-layer fully connected network; the enhanced graph convolution layer and the processed adjacency matrix are used to extract local and long-distance dependence features from the SCADA data and PMU data to obtain SCADA features and PMU features; the dense connection structure is used to fuse the SCADA features and PMU features processed by the enhanced graph convolution; the flattening layer is used to flatten the fused features; and the flattened features are mapped to the state estimation outputs of each device node of the power grid through a multi-layer fully connected network.

[0057] The feature propagation rule of the enhanced graph convolution module is as follows:

[0058]

[0059] where, Indicates the feature output matrix of the th layer, Indicates the feature output matrix of the th layer, Indicates the trainable weight matrix of the

[0060] In this embodiment, the enhanced graph convolution module is used to extract the features of the PMU data and the SCADA data respectively. The enhanced graph convolution module includes a linear transformation layer, a graph convolution layer, a multi-head attention mechanism, a residual connection layer, a batch normalization layer, a Dropout module, and a LeakyReLU layer; the data processing flow of the enhanced graph convolution module includes:

[0061] The SCADA data and the PMU data are dimensionally aligned through the linear transformation layer to obtain the aligned PMU data and the aligned SCADA data;

[0062] Perform convolution processing on the aligned PMU data and aligned SCADA data through the graph convolution layer and the processed adjacency matrix to obtain the convolved PMU data and convolved SCADA data;

[0063] Input the convolved PMU data and convolved SCADA data into the multi-head attention mechanism to obtain the attention output;

[0064] Fuse the SCADA data and PMU data with the output of the multi-head attention mechanism through the residual connection layer respectively to obtain the SCADA data fusion result and the PMU data fusion result;

[0065] Input the SCADA data fusion result and the PMU data fusion result into the batch normalization layer, Dropout module and LeakyReLU layer in sequence for processing to obtain the SCADA features and PMU features.

[0066] The dimension alignment is achieved through the following formula:

[0067]

[0068] where represents the trainable weight matrix, , represents the PMU data, represents the SCADA data, , represents the aligned PMU data, represents the aligned SCADA data.

[0069] The convolution processing is achieved through the following formula:

[0070]

[0071] where represents the processed adjacency matrix, , represents the convolved PMU data, represents the convolved SCADA data.

[0072] The expression of the multi-head attention mechanism is as follows:

[0073]

[0074]

[0075]

[0076]

[0077]

[0078]

[0079]

[0080] Among them, represents the query matrix, represents the key matrix, represents the value matrix, represents the weight matrix of the query matrix, represents the weight matrix of the key matrix, represents the weight matrix of the value matrix, represents the splitting function, represents the split query matrix, represents the split key matrix, represents the split value matrix, represents the dimension of the key matrix, represents the transpose of the split key matrix, represents the attention score, represents the activation function, applied The activation function converts the attention score into a probability distribution; represents the attention probability, represents the context vector, represents the concatenation function, represents the context concatenation result, represents the weight matrix of the linear transformation, represents the output of the multi-head attention mechanism.

[0081] The expression of the residual connection layer is as follows:

[0082]

[0083] Among them, , represents the PMU data fusion result, represents the SCADA data fusion result.

[0084] The expression of the batch normalization layer is as follows:

[0085]

[0086] Among them, represents the output of the batch normalization layer, represents the batch normalization operation.

[0087] The expression of the Dropout module is as follows:

[0088]

[0089] Among them, represents the output of the Dropout module, represents the Dropout operation, randomly discarding the outputs of some neurons during the training process.

[0090] The expression of the LeakyReLU layer is as follows:

[0091]

[0092] Among them, , represents the PMU feature, represents the PMU feature, represents the LeakyReLU activation function.

[0093] The expression of the dense connection module is as follows:

[0094]

[0095] Among them, represents the fused feature, represents the concatenation function, represents the PMU feature, represents the PMU feature.

[0096] The expression of the flattening layer is as follows:

[0097] ;

[0098] Among them, represents the flattened feature, represents the flattening operation function, represents the fused feature.

[0099] The multi-layer fully connected network uses the LeakyReLU activation function. The fully connected layer maps the flattened feature to the target output dimension (such as active and reactive power or voltage magnitude and phase angle); through the fully connected layer, the extracted features can be transformed into the final state estimation result, with high accuracy and robustness; the data processing flow of the multi-layer fully connected network includes:

[0100] Perform feature mapping on the flattened feature according to the following formula to obtain the state variables of each device node:

[0101]

[0102] Among them, represents the The state variables of each device node represent the trainable weights of the fully connected layer represent the bias of the fully connected layer represent the flattened features the th flattened feature element;

[0103] Perform a non - linear transformation on the state variables of each device node according to the following formula to obtain the state estimate of each device node:

[0104]

[0105] where, represent the state estimate of the th device node, represent a very small constant used to prevent the vanishing gradient.

[0106] The power system state estimation model uses the mean square error loss function for model training, and the expression of the mean square error loss function is as follows:

[0107]

[0108] where, represent the loss of the power system state estimation model, which is used to guide the update of model parameters to ensure that the prediction results are as close as possible to the actual state; represent the total number of training samples, represent the state of the th device node, represent the state estimate of the th device node.

[0109] To verify the effectiveness of the power system state estimation method proposed in this embodiment, the MATLAB package MATPOWER is used for simulation, and the power flow calculation is carried out based on the typical example of the IEEE30 - bus power system; since the state variables do not fluctuate violently under the steady - state of the power system, in order to make the experimental results more robust, the measurement data sets of SCADA and PMU are constructed by adding Gaussian noise with different precisions to the true values. Under the steady - state environment, the load level is sampled based on uniform or normal distribution. Gaussian white noise with a standard deviation of 0.05 p.u is added to the SCADA measurement, and Gaussian white noise with a standard deviation of 0.02 p.u is added to the PMU. All data are generated by the Monte Carlo method; the baseline methods of this experiment include the weighted least - squares method (WLS) and the deep neural network (DNN). Hereinafter, this power system state estimation method is referred to as EnhancedGCN, and the experimental results are shown in Table 1 below.

[0110] Table 1 Comparison of State Estimation Results between the Power System State Estimation Method and Two Baseline Methods

[0111] Algorithm Mean Absolute Error Accuracy WLS 0.0264 89.7% DNN 0.0097 96.3% EnhancedGCN 0.0053 99.5%

[0112] Figure 3 The figure shows the comparison of reactive power prediction results between the power system state estimation method (EnhancedGCN) and the baseline methods Figure 4 The figure shows the comparison of active power prediction results between the power system state estimation method and the baseline methods. Combining Table 1 Figure 3 and Figure 4 According to the data, the mean absolute error (MAE) of the power system state estimation method is only 20% of that of WLS and 55% of that of DNN, indicating a significant reduction in its prediction error; the accuracy of the power system state estimation method is 3.2% higher than that of DNN and 9.8% higher than that of WLS. The power system state estimation method is significantly superior to WLS and DNN in terms of both mean absolute error and prediction accuracy, demonstrating its superiority in the complex power system state estimation task; its core advantage lies in combining graph structure modeling and deep learning techniques, and at the same time enhancing robustness through regularization and multi-modal fusion

[0113] The experimental results show that under the IEEE 30-node network structure, the power system state estimation method has significant advantages in terms of estimation accuracy, real-time performance, and adaptability compared with traditional methods, and can effectively cope with the challenges of renewable energy integration and increasing data volume in modern power systems, providing reliable technical support for the real-time monitoring and optimal scheduling of smart grids

[0114] In summary, compared with traditional state estimation methods, the power system state estimation method proposed in this embodiment can effectively extract the spatial features of the power system topology structure, capture the long-distance dependence relationships between nodes, better learn the complex features and dynamic characteristics of the power system, and can significantly improve the generalization ability of the model by fusing multi-source data, thereby effectively improving the accuracy and robustness of state estimation

[0115] Embodiment 2

[0116] Based on the same inventive concept as Embodiment 1, the present invention discloses a power system state estimation device. Referring to Figure 2 as shown, it includes:

[0117] A data acquisition module for: acquiring the grid topology structure information, SCADA data, and PMU data of each equipment node in the power system

[0118] A data preprocessing module, configured to: construct an adjacency matrix according to the power grid topology structure information, and perform data preprocessing on the adjacency matrix to obtain a processed adjacency matrix;

[0119] A feature extraction module, configured to: extract features from the SCADA data and PMU data through an enhanced graph convolution module and the processed adjacency matrix to obtain SCADA features and PMU features;

[0120] A feature fusion module, configured to: perform feature fusion on the SCADA features and PMU features through a densely connected module to obtain fused features;

[0121] A feature flattening module, configured to: flatten the fused features through a flattening layer to obtain flattened features;

[0122] A state estimation module, configured to: map the flattened features to state estimation outputs of each device node in the power system by using a multi-layer fully connected network to obtain state estimation values of each device node.

[0123] For the specific function implementation of each of the above modules, refer to the relevant content in the method of Embodiment 1, which will not be elaborated.

[0124] Embodiment 3:

[0125] This embodiment provides a computer-readable storage medium, on which computer instructions are stored. When the computer instructions are executed by a processor, the steps of the power system state estimation method described in any one of Embodiment 1 are implemented.

[0126] Embodiment 4:

[0127] This embodiment provides a computer device, including:

[0128] A memory, configured to store computer instructions;

[0129] A processor, configured to execute the computer instructions to implement the steps of the power system state estimation method described in any one of the first aspect.

[0130] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0131] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device generate means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0132] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0133] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0134] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the spirit and scope protected by the present invention and the claims. These all fall within the protection scope of the present invention.

Claims

1. A method for estimating a power system state, characterized in that: include: Obtain the grid topology information, SCADA data and PMU data of each device node in the power system; Constructing an adjacency matrix according to the power grid topology information, and performing data preprocessing on the adjacency matrix to obtain a processed adjacency matrix; The processed adjacency matrix, SCADA data and PMU data are input into a trained power system state estimation model: the SCADA data and PMU data are feature extracted by an enhanced graph convolution module and the processed adjacency matrix to obtain SCADA features and PMU features; the SCADA features and PMU features are feature fused by a dense connection module to obtain fused features; the fused features are flattened by a flattening layer to obtain flattened features; the flattened features are mapped to state estimation outputs of each device node in the power system by a multi-layer fully connected network to obtain state estimation values ​​of each device node.

2. The method for estimating the state of a power system according to claim 1, characterized in that: A topological relationship diagram is constructed according to the power grid topological structure information. The expression is as follows: ； in, Represents a collection of device nodes. Represents the set of connection edges between each device node; According to the topological relationship diagram Constructing an adjacency matrix, wherein the adjacency matrix is ​​used to represent the connection relationship between each device node; The data preprocessing of the adjacency matrix includes: Adding self-loops to the adjacency matrix to obtain an adjacency matrix with self-loops added; Using the degree matrix of the adjacency matrix with the self-loop added, normalizing the adjacency matrix with the self-loop added, to obtain a processed adjacency matrix; Adding self-loops to the adjacency matrix is ​​achieved by the following formula: ； in, represents the adjacency matrix with self-loops added, represents the adjacency matrix, represents the identity matrix; The degree matrix of the adjacency matrix with self-loop added is obtained by the following formula: ； in, represents the degree matrix of the adjacency matrix with self-loops added, Represents a function for creating a diagonal matrix, Represents the row index of the equipment node in the power system, Represents the column index of the equipment node in the power system, The adjacency matrix with the self-loop added is represented by Line Elements of a column; Normalizing the adjacency matrix with self-loops added is done using the following formula: ； in, Represents the adjacency matrix after processing.

3. The method for estimating the state of a power system according to claim 1, characterized in that: The enhanced graph convolution module includes a linear transformation layer, a graph convolution layer, a multi-head attention mechanism, a residual connection layer, a batch normalization layer, a Dropout module and a LeakyReLU layer; The data processing flow of the enhanced graph convolution module includes: The SCADA data and the PMU data are dimensionally aligned by the linear transformation layer to obtain aligned PMU data and aligned SCADA data; The aligned PMU data and the aligned SCADA data are convolved by the graph convolution layer and the processed adjacency matrix to obtain convolved PMU data and convolved SCADA data; Inputting the convolved PMU data and the convolved SCADA data into a multi-head attention mechanism to obtain an attention output; The SCADA data and the PMU data are respectively fused with the output of the multi-head attention mechanism through a residual connection layer to obtain a SCADA data fusion result and a PMU data fusion result; The SCADA data fusion result and the PMU data fusion result are sequentially input into a batch normalization layer, a Dropout module and a LeakyReLU layer for processing to obtain SCADA features and PMU features; The dimension alignment is achieved by the following formula: ； in, represents the trainable weight matrix, , Indicates PMU data, Represents SCADA data, , Indicates the PMU data after alignment, Indicates the aligned SCADA data; The convolution process is implemented by the following formula: ； in, represents the adjacency matrix after processing, , Represents the PMU data after convolution, Represents SCADA data after convolution; The expression of the multi-head attention mechanism is as follows: ； ； ； ； ； ； ； in, represents the query matrix, represents the key matrix, represents the value matrix, represents the weight matrix of the query matrix, represents the weight matrix of the bond matrix, represents the weight matrix of the value matrix, represents the partition function, represents the query matrix after segmentation, represents the key matrix after segmentation, represents the value matrix after segmentation, represents the dimension of the key matrix, represents the transpose of the key matrix after partitioning, represents the attention score, represents the activation function, represents the attention probability, represents the context vector, represents the concatenation function, Represents the context concatenation result, represents the weight matrix of the linear transformation, Represents the output of the multi-head attention mechanism; The expression of the residual connection layer is as follows: ； in, , represents the PMU data fusion result, Indicates the SCADA data fusion result; The expression of the batch normalization layer is as follows: ； in, represents the output of the batch normalization layer, Represents batch normalization operation; The expression of the Dropout module is as follows: ； in, represents the output of the Dropout module, Represents the Dropout operation; The expression of the LeakyReLU layer is as follows: ； in, , Indicates the PMU characteristics, Indicates the PMU characteristics, Represents the LeakyReLU activation function.

4. The method for estimating the state of a power system according to claim 1, characterized in that: The expression of the densely connected module is as follows: ； in, represents the fusion feature, represents the concatenation function, Indicates the PMU characteristics, Indicates PMU characteristics.

5. The method for estimating the state of a power system according to claim 1, characterized in that: The expression of the flattened layer is as follows: ； in, represents the flattened feature, represents the flattening operation function, Represents fusion features.

6. The method for estimating the state of a power system according to claim 1, characterized in that: The multi-layer fully connected network adopts the LeakyReLU activation function, and the data processing flow of the multi-layer fully connected network includes: The flattened features are mapped according to the following formula to obtain the state variables of each device node: ； in, Indicates The state variables of the device node, represents the trainable weights of the fully connected layer, represents the bias of the fully connected layer, Represents the flattened feature The A flattened feature element; The state variables of each device node are transformed nonlinearly according to the following formula to obtain the estimated state value of each device node: ； in, Indicates The estimated state of the device nodes, Represents a constant.

7. The method for estimating the state of a power system according to claim 1, characterized in that: The power system state estimation model uses a mean square error loss function for model training. The expression of the mean square error loss function is as follows: ； in, represents the loss of the power system state estimation model, represents the total number of training samples, Indicates The status of the device node. Indicates The estimated state of each device node.

8. A power system state estimation device, characterized in that: include: The data acquisition module is used to obtain the grid topology information, SCADA data of each device node and PMU data in the power system; A data preprocessing module is used to: construct an adjacency matrix according to the power grid topology information, and perform data preprocessing on the adjacency matrix to obtain a processed adjacency matrix; A feature extraction module is used to: extract features from the SCADA data and the PMU data by using an enhanced graph convolution module and the processed adjacency matrix to obtain SCADA features and PMU features; The feature fusion module is used to: perform feature fusion on the SCADA feature and the PMU feature through a dense connection module to obtain a fusion feature; A feature flattening module, used to: flatten the fused features through a flattening layer to obtain flattened features; The state estimation module is used to: use a multi-layer fully connected network to map the flattened features into state estimation outputs of each device node in the power system to obtain a state estimation value of each device node.

9. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instruction is executed by a processor, the steps of the power system state estimation method described in any one of claims 1-7 are implemented.

10. A computer device, characterized in that: include: Memory, for storing computer instructions; A processor, configured to execute the computer instructions to implement the steps of the power system state estimation method according to any one of claims 1 to 7.

Images