Abnormality identification method for power distribution network and electronic device

Abstract

This invention discloses a method and electronic device for anomaly identification in a distribution network. It relates to the field of distribution network state estimation technology. The method includes: acquiring monitoring data of the distribution network at the current moment; determining the target state value of the distribution network at the current moment using a target state estimation model based on the monitoring data, wherein the target state estimation model is obtained by optimizing and training a state estimation network based on historical monitoring data of the distribution network at multiple historical moments and the corresponding historical state values; determining whether there is an anomaly in the distribution network at the current moment based on the target state value; and, if there is an anomaly in the distribution network at the current moment, determining the anomaly area of ​​the distribution network at the current moment. This invention solves the technical problems of low accuracy in anomaly detection and low accuracy in identifying anomaly areas in distribution network anomaly identification due to insufficient measurement coverage and poor generalization ability of the anomaly identification method.

Description

Technical Field

[0001] This invention relates to the field of power distribution network condition estimation technology, and more specifically, to a method and electronic device for identifying anomalies in power distribution networks. Background Technology

[0002] With the large-scale integration of new loads such as distributed photovoltaics, energy storage systems, and electric vehicles into medium- and low-voltage distribution networks, the operating mode of distribution networks has gradually evolved from a unidirectional radial structure to an active network with bidirectional power flow, strong randomness, and high dynamism. Against this backdrop, accurately and in real-time acquiring state information such as voltage amplitude and phase angle of all network nodes has become the core foundation for realizing distribution network safety monitoring, fault diagnosis, and intelligent control. Related technologies have significant shortcomings in distribution network anomaly identification, mainly in the following aspects:

[0003] Constrained by construction costs and operational complexity, county and rural power distribution networks generally suffer from sparse measurement deployment, with overall measurement coverage below 30%, making it difficult to provide complete and reliable data on the network's operation. Furthermore, the continuous accumulation of model parameter mismatches due to conductor aging, tap changer operation, and equipment aging leads to discrepancies between the physical model and the real power grid, resulting in poor generalization ability of the physical model. Although some power distribution network anomaly identification methods can detect anomalies to a certain extent, insufficient measurement coverage and poor generalization ability result in low accuracy in anomaly detection and anomaly area identification.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This invention provides a method and electronic device for identifying anomalies in a power distribution network, which at least solves the technical problems of low accuracy in detecting anomalies in the power distribution network and low accuracy in identifying abnormal areas caused by insufficient measurement coverage and poor generalization ability of the anomaly identification method in related technologies.

[0006] According to one aspect of the present invention, an anomaly identification method for a distribution network is provided, comprising: acquiring monitoring data of the distribution network at the current moment, wherein the monitoring data includes at least: the voltage corresponding to each of multiple voltage nodes in the distribution network, and the active power and reactive power corresponding to each of multiple branches in the distribution network; determining the target state value of the distribution network at the current moment by employing a target state estimation model based on the monitoring data, wherein the target state estimation model is obtained by optimizing and training a state estimation network based on historical monitoring data of the distribution network at multiple historical moments and the corresponding historical state values; determining whether there is an anomaly in the distribution network at the current moment based on the target state value; and determining the anomaly area of ​​the distribution network at the current moment if there is an anomaly in the distribution network at the current moment.

[0007] According to another aspect of the present invention, an anomaly identification device for a distribution network is also provided, comprising: a monitoring data acquisition module, configured to acquire monitoring data of the distribution network at the current moment, wherein the monitoring data includes at least: the voltage corresponding to each of multiple voltage nodes in the distribution network, and the active power and reactive power corresponding to each of multiple branches in the distribution network; a target state value determination module, configured to determine the target state value of the distribution network at the current moment based on the monitoring data and using a target state estimation model, wherein the target state estimation model is obtained by optimizing and training a state estimation network based on historical monitoring data of the distribution network at multiple historical moments and the corresponding historical state values; an anomaly judgment module, configured to determine whether there is an anomaly in the distribution network at the current moment based on the target state value; and an anomaly region determination module, configured to determine the anomaly region of the distribution network at the current moment if there is an anomaly in the distribution network at the current moment.

[0008] According to another aspect of the present invention, a non-volatile storage medium is also provided, which stores multiple instructions adapted for a power distribution network anomaly identification method, any one of which can be loaded and executed by a processor.

[0009] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement any one of the distribution network anomaly identification methods.

[0010] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of any one of the distribution network anomaly identification methods.

[0011] In this embodiment of the invention, monitoring data of the distribution network at the current moment is acquired. This monitoring data includes at least: the voltages corresponding to multiple voltage nodes in the distribution network, and the active and reactive power corresponding to multiple branches in the distribution network. Based on the monitoring data, a target state estimation model is used to determine the target state value of the distribution network at the current moment. The target state estimation model is obtained by optimizing and training a state estimation network based on historical monitoring data of the distribution network at multiple historical moments and the corresponding historical state values. Based on the target state value, it is determined whether there is an anomaly in the distribution network at the current moment. If there is an anomaly in the distribution network at the current moment, the distribution network is determined to be abnormal. The method achieves the goal of accurately identifying abnormal areas in the distribution network at the current moment by using the monitoring data of the distribution network at the current moment and employing a pre-constructed target state estimation model. This improves the accuracy of anomaly detection and abnormal area identification in the distribution network, thereby solving the technical problems of low accuracy in anomaly detection and abnormal area identification caused by insufficient measurement coverage and poor generalization ability of anomaly identification methods in distribution network anomaly identification. Attached Figure Description

[0012] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0013] Figure 1 This is a flowchart of a distribution network anomaly identification method according to an embodiment of the present invention;

[0014] Figure 2 This is a flowchart of an optional method for determining the objective function of a power distribution network at multiple historical moments according to an embodiment of the present invention;

[0015] Figure 3 This is a flowchart of an optional anomaly identification method for a power distribution network according to an embodiment of the present invention;

[0016] Figure 4 This is a schematic diagram of an anomaly identification device for a power distribution network according to an embodiment of the present invention;

[0017] Figure 5 This is a schematic diagram of an electronic device for identifying anomalies in a power distribution network according to an embodiment of the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] According to an embodiment of the present invention, a method for anomaly identification in a power distribution network is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0021] Figure 1 This is a flowchart of a distribution network anomaly identification method according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0022] Step S102: Obtain the monitoring data of the distribution network at the current moment. The monitoring data includes at least: the voltage corresponding to each of the multiple voltage nodes in the distribution network, and the active power and reactive power corresponding to each of the multiple branches in the distribution network.

[0023] Optionally, acquiring monitoring data can lay the foundation for subsequent high-precision distribution network state estimation. Among the monitoring data, voltage is a variable describing the steady-state operation of the distribution network, directly determining power flow and voltage distribution; while the active and reactive power of branches serve as a bridge connecting node states and network topology. After acquiring real-time monitoring data, standardization is performed to eliminate the impact of differences in data magnitude. Standardized monitoring data can be obtained through the following methods. ,in, This represents the monitoring data at the current moment. This represents the historical average of the monitoring data. The historical standard deviation represents the historical data. The historical mean and historical standard deviation are calculated based on the monitoring data corresponding to multiple historical moments. t represents the index of the current moment.

[0024] Step S104: Based on the monitoring data, the target state estimation model is used to determine the target state value of the distribution network at the current time. The target state estimation model is obtained by optimizing and training the state estimation network based on the historical monitoring data of the distribution network at multiple historical times and the corresponding historical state values.

[0025] Optionally, based on standardized monitoring data, a target state estimation model optimized and trained using historical monitoring data is employed to determine the target state value of the distribution network at the current moment. This target state estimation model uses a deep learning framework, employing monitoring data and their corresponding historical state values ​​from multiple historical moments of the distribution network as training samples to perform end-to-end optimization of the state estimation network structure. This allows the state estimation network to automatically learn the implicit expressions of power propagation laws, topological coupling relationships, and physical constraints among nodes. During training, the target state estimation model does not merely fit the data distribution; instead, it maps multiple voltage nodes in the distribution network to multiple nodes in the state estimation network, and maps the normalized admittance values ​​of multiple branches in the distribution network to the weights of multiple edges in the state estimation network. This creates a one-to-one correspondence between the state estimation network structure and the distribution network topology, ensuring that the state estimation network continuously approximates the physical feasible region of the real distribution network during the learning process. Even in the case of missing monitoring data or noise pollution during the inference phase, it can still output state estimation results that satisfy Kirchhoff's laws and conform to historical operational statistics. The method for determining the target state value in this implementation can overcome the rigid dependence on measurement redundancy in related technologies, and at the same time avoid the physical infeasibility problems such as voltage exceeding limits or power imbalance caused by ignoring physical laws in pure data-driven models. It significantly improves the accuracy and generalization ability of state estimation, and provides a high-quality and reliable state input foundation for subsequent anomaly detection.

[0026] In an optional embodiment, when the target state estimation model includes a graph convolutional layer and an output layer, the target state estimation model is used to determine the target state value of the distribution network at the current time based on monitoring data. This includes: determining multiple input features and multiple adjacent feature sets of the distribution network at the current time based on monitoring data, wherein the multiple input features and multiple adjacent feature sets correspond one-to-one with multiple voltage nodes, and the multiple adjacent feature sets are used to indicate the degree of feature correlation between the corresponding voltage node and multiple target voltage nodes, and the multiple target voltage nodes represent the voltage nodes among the multiple voltage nodes that are connected to the corresponding voltage node through branches; performing convolution processing on the multiple input features and multiple adjacent feature sets based on the graph convolutional layer to determine the target node features of the multiple voltage nodes at the current time; and performing mapping processing on the target node features of the multiple voltage nodes at the current time based on the output layer to obtain the target state values ​​of the multiple voltage nodes at the current time.

[0027] Optionally, in a target state estimation model with a graph convolutional layer and an output layer, the process of determining the target state value of the distribution network based on the current monitoring data essentially encodes the physical topology and electrical operating state of the distribution network into graph-structured data, and achieves state inference through a hierarchical feature propagation mechanism. Specifically, firstly, based on the real-time standardized monitoring data, a corresponding input feature is constructed for each voltage node. This input feature consists of the standardized monitoring data of the corresponding voltage node; if the voltage node has no monitoring data, it is filled with zero values ​​and a measurement availability marker is added. Next, an adjacency feature set is generated for each node. This adjacency feature set accurately reflects the electrical association strength of the neighboring nodes (i.e., multiple target voltage nodes) directly connected to the voltage node through branches. Any element in this adjacency feature set is the normalized admittance of the corresponding branch. Subsequently, a graph convolutional layer aggregates and nonlinearly transforms the input features of each voltage node with its neighboring feature sets, enabling each voltage node to gradually absorb electrical information from its multi-hop neighbors. This achieves the progressive propagation and fusion of the state of nodes without monitoring data at a remote location from neighboring nodes with monitoring data, overcoming the limitation of relying solely on local monitoring data in related technologies. Finally, the output layer maps the high-dimensional target node features learned by graph convolution for each voltage node to the final target state value, completing the end-to-end reconstruction from local features to global state. The target state values ​​of multiple voltage nodes at the current time can be obtained in the following way. ,in, This represents the target state value of any voltage node among multiple voltage nodes at the current moment. This represents the input features of the last graph convolutional layer at the current time step. This represents the target node characteristics of any voltage node at the current moment. This represents the bias characteristics of the output layer. This embodiment calculates the target state value using a graph-based information aggregation method, which significantly reduces the computational complexity of the model. Furthermore, it maintains stable convergence even in extremely sparse scenarios with insufficient monitoring data coverage, effectively improving the accuracy and real-time performance of state estimation.

[0028] In an optional embodiment, when multiple graph convolutional layers are included, convolution processing is performed on multiple input features and multiple adjacent feature sets based on the graph convolutional layers to determine the target node features of multiple voltage nodes at the current time. This includes: based on multiple graph convolutional layers, performing convolution processing on multiple input features and multiple adjacent feature sets in the following manner to determine the target node features of any voltage node among the multiple voltage nodes at the current time: starting from the first graph convolutional layer among the multiple graph convolutional layers, repeating the following operations until the convolution processing of the last graph convolutional layer is completed: based on multiple input features output from the previous graph convolutional layer of any graph convolutional layer, candidate node features corresponding to each of the multiple voltage nodes, and any adjacent feature set from the multiple adjacent feature sets, obtaining the candidate node features of any voltage node in any graph convolutional layer; determining the candidate node features of the last graph convolutional layer obtained when the convolution processing of the last graph convolutional layer is completed as the target node features of any voltage node at the current time; and obtaining the target node features of multiple voltage nodes at the current time by adopting the method of obtaining the target node features of any voltage node at the current time.

[0029] Optionally, in a graph convolutional layer with multiple layers, a multi-layer progressive message-passing mechanism is used for the convolutional processing of multiple input features and adjacent feature sets. This allows the input features of each voltage node to diffuse layer by layer from the local neighborhood to the global topology, achieving deep modeling of the electrical characteristics of the distribution network. Specifically, starting from the first graph convolutional layer, the input features of each voltage node and the features of its directly connected neighboring nodes are weighted and aggregated, undergoing a nonlinear transformation within the graph convolutional layer to output the candidate node features of that voltage node in this layer. These candidate features do not only reflect direct neighbor information but are high-dimensional representations that integrate topological weighting and nonlinear mapping. Subsequently, this candidate feature serves as the input to the next graph convolutional layer, continuing to interact with neighboring node information at greater distances. With each additional layer, the range of information that a voltage node can perceive expands, thereby capturing electrical connections and power propagation paths across multiple branches in the deep network. This process iterates continuously through multiple graph convolutional layers until the last convolutional layer completes processing. The final output target node features incorporate the accumulated electrical influences from the one-hop neighborhood to the multi-hop topology of the entire network, fully expressing the global location and operating state of voltage nodes in complex distribution network structures. By independently executing this multi-layer propagation process for each voltage node, the target node features of all voltage nodes in the distribution network at the current moment are obtained, forming a unified high-dimensional state representation. This multi-graph convolutional layer structure can significantly enhance the expressive ability of the target state estimation model for state inference in sparse monitoring data environments. Even if a voltage node has no monitoring data, its target node features can still effectively absorb the monitoring data of distant voltage nodes through multi-hop propagation, avoiding the limitation of single-layer graph convolution focusing only on the neighborhood and ignoring the correlation of distant nodes. At the same time, multi-layer nonlinear transformation can endow the target state estimation model with strong nonlinear fitting capabilities, enabling the target state estimation model to adapt to complex operating conditions such as sudden changes in distributed power output and dynamic load fluctuations. Compared with single-layer mapping or fixed weight methods in related technologies, the method in this embodiment can not only satisfy physical consistency in the state estimation results, but also have adaptive generalization capabilities for atypical operating scenarios.

[0030] In an optional embodiment, the candidate node features of any voltage node in any graph convolutional layer are obtained based on multiple input features output from the previous graph convolutional layer, candidate node features corresponding to each of the multiple voltage nodes, and any adjacent feature set from multiple adjacent feature sets. This includes obtaining the candidate node features of any voltage node in any graph convolutional layer based on multiple input features output from the previous graph convolutional layer, candidate node features corresponding to each of the multiple voltage nodes, and any adjacent feature set, in the following manner:

[0031] ;

[0032] in, This represents the candidate node features of any voltage node in any graph convolutional layer. Indicates the index of any voltage node. This represents the index of any graph convolutional layer. This represents the activation function. This represents the input feature corresponding to any voltage node among the multiple input features output from the previous convolutional layer of any graph. This represents the candidate node features of any voltage node in the output of the previous graph convolutional layer for any given graph convolutional layer. It represents any element in any set of adjacent features. Indicates the index of any target voltage node. This represents the total number of target voltage nodes. This represents the input feature corresponding to any target voltage node among the multiple input features output from the previous convolutional layer of any graph. This represents the candidate node features of any target voltage node in the output of the previous graph convolutional layer of any graph convolutional layer. This represents the bias feature of the previous convolutional layer in any given graph.

[0033] Optionally, the graph convolution operation described in the above formula can achieve hierarchical modeling of the electrical characteristics of the distribution network. Here, the bias feature is a learnable parameter automatically optimized through backpropagation during the training of the state estimation network. Its function is to introduce a global offset into the feature transformation of voltage nodes in each graph convolutional layer to compensate for the bias before the nonlinear activation of the state estimation network. This convolution operation process makes each graph convolution layer no longer a simple feature averaging, but a message passing process that integrates physical structure, electrical weights, and nonlinear learning. Through multi-layer stacking, each voltage node can not only perceive the electrical state of its direct neighbors, but also indirectly obtain the power disturbance impact across the entire distribution network through cascading propagation.

[0034] In an optional embodiment, before determining the target state value of the distribution network at the current moment using a target state estimation model based on monitoring data, the method further includes: determining the objective function of the distribution network at multiple historical moments based on historical monitoring data of the distribution network at multiple historical moments and the corresponding historical state values, wherein the objective function represents the deviation between the operating state of the distribution network at the corresponding historical moment and the preset operating state; optimizing and training the state estimation network with the minimum function value of the objective function of the distribution network at multiple historical moments as the optimization objective, until a predetermined number of iterations is reached; and obtaining the target state estimation model based on the optimized state estimation network obtained after reaching the predetermined number of iterations.

[0035] Optionally, before using the target state estimation model to determine the target state value of the distribution network at the current moment, firstly, an objective function is constructed for each historical moment using monitoring data collected from the distribution network at multiple historical moments and their corresponding historical state values. Then, with the minimization of the objective function value at each historical moment as the optimization objective, the state estimation network is jointly trained end-to-end. The training process continuously adjusts the parameters of the state estimation network through backpropagation, enabling the network to automatically learn and approximate the historical optimal state distribution when facing similar operating modes. Finally, after reaching a preset number of iterations, the optimal state estimation network is locked, forming a target state estimation model with generalization capabilities. Compared to related technologies that rely on manually setting pseudo-measurement weights or static model parameters, this embodiment adopts a data-driven approach to adaptively learn the optimal estimation strategy, which can significantly improve the generalization ability of the target state estimation model in unseen operating scenarios. At the same time, since the training samples cover multiple historical operating conditions, the obtained target state estimation model has strong robustness to real-world operating conditions such as sparsity, noise, and abrupt changes. In the actual online inference stage, it can quickly output physically consistent and high-precision full-network state without iterative calculation, providing a stable and reliable state foundation for subsequent anomaly detection.

[0036] In one alternative embodiment, Figure 2 This is a flowchart of an optional method for determining the objective function of a distribution network at multiple historical moments according to an embodiment of the present invention, such as... Figure 2 As shown, based on historical monitoring data of the distribution network at multiple historical moments and the corresponding historical state values, the objective function of the distribution network at multiple historical moments is determined, including:

[0037] Step S202: Based on the historical monitoring data of the distribution network at multiple historical moments, determine the voltage error, active power error and reactive power error of the distribution network at multiple historical moments.

[0038] Step S204: Based on voltage error, active power error and reactive power error, determine the physical losses of the distribution network at multiple historical moments, wherein the physical losses are used to indicate the deviation between the actual voltage drop and the preset voltage drop of the distribution network at the corresponding historical moment.

[0039] Step S206: Based on the historical monitoring data of the distribution network at multiple historical moments and the corresponding historical state values, determine the measurement loss of the distribution network at multiple historical moments. The measurement loss is used to indicate the deviation between the historical state value and the predicted state value of the distribution network at the corresponding historical moment. The predicted state value is predicted by the state estimation network based on the corresponding historical monitoring data during the optimization training process.

[0040] Step S208: Based on the node injection power corresponding to the historical state values ​​of the distribution network at multiple historical moments, determine the Bayesian prior loss of the distribution network at multiple historical moments. The Bayesian prior loss is used to indicate the deviation between the probability distribution corresponding to the node injection power of the distribution network at the corresponding historical moment and the preset probability distribution.

[0041] Step S210: Based on the measurement loss, physical loss and Bayesian prior loss of the distribution network at multiple historical moments, determine the objective function of the distribution network at multiple historical moments.

[0042] Optionally, in the training framework for constructing the target state estimation model, the objective function is not a single error metric, but a multi-dimensional collaborative optimization objective that integrates physical laws, monitoring data, and statistical priors. Specifically, firstly, based on the measured voltage, active power, and reactive power at historical moments, the errors between these values ​​and the theoretical voltage, theoretical active power, and theoretical reactive power derived from the target state estimation model are calculated, forming three types of basic error terms. From these, the physical loss is derived. This physical loss term ensures that the output of the target state estimation model closely approximates the basic physical laws of the distribution network during training. Even in areas where monitoring data is missing, it avoids estimation results that violate physical laws, such as power imbalance and voltage exceeding limits. This internalizes the distribution network's topology and impedance parameters as hard constraints of the network. The physical loss of the distribution network at any historical moment can be obtained as follows: ,in, This represents the voltage error of any one of the multiple branches in a distribution network at any historical moment. This represents the voltage of any voltage node among multiple voltage nodes in a distribution network at any historical moment. This represents the voltage of a specified voltage node at any historical moment, where i represents the index of any voltage node, j represents the index of the specified voltage node, any branch includes both any voltage node and the specified voltage node, L represents the total number of branches, and N represents the total number of voltage nodes. This represents the resistance of any branch at any historical moment. This represents the reactance of any branch at any historical moment. This represents the theoretical active power of any branch at any historical moment. This represents the theoretical reactive power of any branch at any historical moment. This represents the voltage phase angle of any voltage node at any historical moment. This represents the voltage phase angle of a specified voltage node at any historical moment. This represents the electrical conductance of any branch at any historical moment. This represents the electrical impedance of any branch at any historical moment. This represents the active power error at any voltage node at any historical moment. This represents the reactive power error at any voltage node at any historical moment. This represents the theoretical injected active power at any voltage node at any historical moment. This represents the theoretical reactive power injected at any voltage node at any historical moment. This represents the set of branches connected to any voltage node. This represents the active power of any branch at any historical moment at any voltage node. This represents the reactive power of any branch at any voltage node at any historical moment. By obtaining the physical losses of the distribution network at any historical moment, the measurement losses of the distribution network at multiple historical moments can be obtained. Secondly, the measurement loss is constructed, which is the weighted residual between the historical state values ​​and the predicted state values ​​predicted by the state estimation network. This measurement loss ensures that the target state estimation model can accurately fit the actual operating state, achieving precise data-driven mapping. The measurement loss of the distribution network at any historical moment can be obtained as follows: ,in, This represents any monitoring value in the historical monitoring data of the distribution network at any given historical moment, where m represents the index of any monitoring value, and M represents the total number of monitoring values ​​included in the historical monitoring data. This represents the predicted state value at any historical moment based on the historical state value of any monitoring value. The standard deviation of any monitoring value is represented by the standard deviation of the monitoring value. This represents the weighting coefficient for any monitored value. The measurement loss of the distribution network at any historical moment can be obtained, and the measurement loss of the distribution network at multiple historical moments can also be obtained. Furthermore, a Bayesian prior loss is introduced. This Bayesian prior loss embeds statistical prior knowledge such as load patterns and photovoltaic output characteristics inherent in the long-term operation of the distribution network into the training process. It provides reasonable probability guidance in voltage node regions without monitoring data, compensates for statistical uncertainties that physical constraints cannot cover, and avoids the target state estimation model falling into unreasonable state intervals. The Bayesian prior loss of the distribution network at any historical moment can be determined as follows: ,in, This represents the theoretical node injection power at any historical moment, representing the historical state value of any voltage node. T denotes the transpose. This represents the number of multiple Gaussian components. This represents the weight of any voltage node in any Gaussian component. This represents the mean of any Gaussian component. Denotes the covariance of any Gaussian component. Let k represent the Gaussian distribution density function of the mean and covariance of any Gaussian component, and k represent the index of any Gaussian component. Multiple Gaussian components can be determined from historical operating data of the distribution network. Specifically, this is achieved by acquiring historical operating data of each voltage node in the distribution network, including the time series of active power of the load. Time series of active power output of distributed photovoltaic power Extract historical daily load curves for each voltage node. , An index representing any historical day. This represents the total number of historical sampling moments within any given historical day; a clustering algorithm is used to perform cluster analysis on the historical day load curves to minimize the sum of squares within each cluster. To optimize the objective, among which, This represents the total number of clusters, where k represents the index of any cluster (corresponding to a Gaussian component). Represents a set of multiple historical days for any cluster. The center of any cluster is represented, and each cluster center represents the average temporal characteristics of that type of load. The Bayesian prior loss of the distribution network at any historical moment can be obtained by using the method of obtaining the Bayesian prior loss of the distribution network at multiple historical moments. The same clustering process is performed on the distributed photovoltaic (PV) output sequence to obtain several typical PV output patterns. Based on this, the load and PV output of each historical day are considered as a two-dimensional injected power vector, and it is assigned to the corresponding joint pattern category according to its load and PV cluster combination. Based on the historical samples of injected power of each voltage node under various joint patterns, the weight, mean, and covariance of each Gaussian component are determined. Since the joint distribution of load and PV output has strong non-Gaussianity and multi-modal characteristics, the complex distribution is decomposed into several local Gaussian subspaces through pre-partitioning of clusters, ultimately obtaining a mixture density function composed of multiple Gaussian components. Each component corresponds to the statistical distribution of node injected power under a typical operating mode, providing a solid data foundation and a probabilistic model with clear physical meaning for the construction of the Bayesian prior loss. Finally, the objective function is obtained by weighted fusion of measurement loss, physical loss, and Bayesian prior loss.

[0043] In one optional embodiment, the objective function of the distribution network at multiple historical moments is determined based on the measurement loss, physical loss, and Bayesian prior loss of the distribution network at multiple historical moments, including: determining the objective function of the distribution network at multiple historical moments based on the measurement loss, physical loss, and Bayesian prior loss of the distribution network at multiple historical moments in the following manner:

[0044] ;

[0045] in, This represents the objective function of the distribution network at multiple historical moments. This represents the measured losses of the distribution network at multiple historical moments. This represents the physical losses of the distribution network at multiple historical moments. This represents the Bayesian prior loss of the distribution network at multiple historical moments. This represents the weighting coefficient corresponding to the physical loss. This represents the weighting coefficients of the Bayesian prior loss.

[0046] Optionally, when constructing the training objective function of the target state estimation model, a unified, differentiable joint optimization objective with clear physical and statistical significance is formed by weighted fusion of measurement loss, physical loss, and Bayesian prior loss. Specifically, the measurement loss ensures that the target state estimation model faithfully reproduces the actual operating state within the observable region; the physical loss enables the state estimation network to automatically learn the physical feasibility boundary during training, avoiding unacceptable engineering outputs such as power imbalances or voltage exceedances; and the Bayesian prior loss effectively suppresses over-extrapolation and unreasonable fluctuations in the target state estimation model under sparse measurements. By introducing learnable weight coefficients, this objective function achieves a dynamic balance among the three information sources. This objective function jointly optimizes the parameters in the state estimation network during end-to-end training, ultimately forming an intelligent estimation model that deeply integrates physical mechanisms, measurement information, and historical statistics, thereby outputting highly accurate state estimation results.

[0047] Step S106: Based on the target state value, determine whether there is an anomaly in the distribution network at the current moment.

[0048] Optionally, the distribution network can be judged to determine whether there is an anomaly at the current moment based on the target state value. This judgment process can improve the sensitivity and accuracy of anomaly detection by upgrading the target state value from numerical output to intelligent diagnosis of the distribution network's state perception and anomaly cause analysis.

[0049] In one optional embodiment, determining whether the distribution network is abnormal at the current moment based on the target state value includes: obtaining a state matrix of the distribution network at the current moment based on the target state value and historical state values ​​of a predetermined number of historical moments prior to the current moment, wherein any element in the state matrix represents the state value of the corresponding voltage node at the corresponding moment; obtaining a covariance matrix of the distribution network at the current moment based on the state matrix, wherein the covariance matrix is ​​used to indicate the correlation of state changes among multiple voltage nodes in the time dimension; determining a target eigenvalue corresponding to the covariance matrix, wherein the target eigenvalue represents the eigenvalue greater than a preset first eigenvalue among multiple eigenvalues ​​obtained after eigenvalue decomposition of the covariance matrix; determining the probability of anomaly of the distribution network at the current moment based on the target eigenvalue and a preset second eigenvalue, wherein the preset second eigenvalue is greater than the preset first eigenvalue; determining that the distribution network is abnormal at the current moment if the probability of anomaly is greater than the preset probability; or determining that the distribution network is not abnormal at the current moment if the probability of anomaly is less than or equal to the preset probability.

[0050] Optionally, in determining whether an anomaly exists in the distribution network at the current moment, firstly, the state estimates of multiple voltage nodes in the distribution network at the current moment and at a predetermined number of historical moments prior to the current moment are stacked into a state matrix along the time dimension. This state matrix can completely depict the dynamic evolution trajectory of multiple voltage nodes within a time window. Subsequently, the state matrix is ​​standardized to eliminate the influence of differences in the magnitude of different nodes, and a covariance matrix is ​​calculated based on the standardized state matrix. Each element in the covariance matrix reflects the linear correlation of voltage fluctuations between two voltage nodes in the time series, thereby revealing the potential power propagation path and disturbance coupling relationship between voltage nodes. Any element in the state matrix can be standardized in the following way to obtain the standardized element. ,in, This represents the average voltage at any voltage node over a time window (including the current moment and a predetermined number of historical moments prior to the current moment). This represents the standard deviation of any voltage node within a time window. This represents a time window, where t represents the index of any point in the time window. This represents the voltage at any voltage node at any given time. The covariance matrix can be obtained as follows: ,in, Let T represent the standardized state matrix, and T denote the transpose. Based on this, eigenvalue decomposition is performed on the covariance matrix to obtain a set of eigenvalues ​​sorted by size. The eigenvalues ​​greater than a preset first eigenvalue are selected as target eigenvalues, which can be the largest eigenvalue. These target eigenvalues ​​indicate the presence of non-random structured anomalies in the distribution network. Furthermore, the ratio between the target eigenvalue and a preset second eigenvalue is determined as the anomaly probability. Finally, an anomaly is determined to exist in the distribution network only when the anomaly probability is greater than the preset probability. This embodiment's method for determining distribution network anomalies does not rely on any single-point residuals or manually set rules. Instead, it extracts the essence of the anomaly from the statistical structure of multi-node collaborative evolution. It can sensitively identify weak anomaly patterns that are difficult to capture by related technical methods, such as multi-node voltage cascading shifts, distributed power source cluster fluctuations, and concealed grounding faults. This significantly improves the early warning capability for early faults in the distribution network and provides a reliable foundation for subsequent accurate location of anomaly areas.

[0051] Step S108: If there is an anomaly in the distribution network at the current moment, determine the abnormal area of ​​the distribution network at the current moment.

[0052] Optionally, after determining that an anomaly exists in the distribution network at the current moment, further identifying the anomaly area can achieve a precise leap from global anomaly detection to topology-aware local anomaly clustering, improving anomaly location accuracy from the node level to the region level. Simultaneously, by strongly correlating the anomaly area with the physical topology of the distribution network, the engineering interpretability and operational feasibility of the anomaly location results can be ensured, thereby significantly improving the distribution network's autonomous diagnostic capabilities under complex disturbances.

[0053] In one optional embodiment, when there is an anomaly in the distribution network at the current moment, determining the abnormal area of ​​the distribution network at the current moment includes: when there is an anomaly in the distribution network at the current moment, determining the anomaly contribution of multiple voltage nodes at the current moment based on the feature vector corresponding to the target feature value; and determining the area composed of voltage nodes whose anomaly contribution is greater than a preset contribution threshold as the abnormal area.

[0054] Optionally, in determining the anomalous region, firstly, the contribution intensity of multiple voltage nodes to the anomalous region is quantified using the eigenvector corresponding to the target eigenvalue in the eigenvalue decomposition of the covariance matrix. Multiple elements in this eigenvector correspond one-to-one with multiple voltage nodes, and each element indicates the weight of the corresponding voltage node to the anomalous region. The anomalous contribution of any voltage node in the eigenvector corresponding to the target eigenvalue at the current time can be obtained as follows: ,in, This represents any element among multiple eigenvalues. Furthermore, by setting a preset contribution threshold, all voltage nodes with abnormal contributions exceeding this threshold are selected; these voltage nodes collectively constitute anomalous nodes. Since the distribution network has a radial topology, these high-contribution voltage nodes are often physically adjacent to each other or located on the same feeder branch; their spatial clustering implies the propagation path and impact range of the anomaly. Therefore, directly defining these high-contribution voltage nodes as anomalous regions allows for high-precision focusing on the anomaly source, avoiding misjudging distant, non-dominant nodes affected by disturbances as the core of the anomaly.

[0055] Through the above steps S102 to S108, the target state value of the distribution network at the current moment can be obtained by using the constructed target state estimation model based on the monitoring data of the distribution network at the current moment, thereby detecting anomalies in the distribution network. When anomalies are detected in the distribution network at the current moment, the abnormal area can be accurately determined, thereby improving the technical effect of improving the accuracy of anomaly detection and the accuracy of abnormal area identification in the distribution network. This solves the technical problems of low accuracy of anomaly detection and low accuracy of abnormal area identification in distribution network anomaly identification caused by insufficient measurement coverage and poor generalization ability of anomaly identification methods.

[0056] Based on the above embodiments and optional embodiments, the present invention proposes an optional implementation method. Figure 3 This is a flowchart of an optional distribution network anomaly identification method according to an embodiment of the present invention, such as... Figure 3 As shown, the method includes:

[0057] S1: Construct a target state estimation model for the distribution network. Specifically, this includes: determining the objective function of the distribution network at multiple historical moments based on historical monitoring data of the distribution network at multiple historical moments and the corresponding historical state values; optimizing the state estimation network by minimizing the function value of the objective function of the distribution network at multiple historical moments to determine the target state estimation model. The specific implementation process is the same as the aforementioned embodiment and will not be repeated here.

[0058] S2: Obtain monitoring data of the distribution network at the current moment, specifically including: obtaining monitoring data such as the voltage of each of the multiple voltage nodes of the distribution network at the current moment, the active power and reactive power of multiple branches, etc., and performing standard processing on the monitoring data to form standardized monitoring data. The specific implementation process is the same as the aforementioned embodiment, and will not be repeated here.

[0059] S3: Determine the target state value of the distribution network at the current moment. Specifically, this includes: inputting the standardized monitoring data obtained in step S1 into multiple graph convolutional layers in the target state estimation model constructed in step S2 for convolution processing; finally, through the output layer in the target state estimation model, mapping the target node features of multiple voltage nodes output by the last graph convolutional layer at the current moment to obtain the target state values ​​of multiple voltage nodes at the current moment. The specific implementation process is the same as in the aforementioned embodiment and will not be repeated here.

[0060] S4: Determine whether there is an anomaly in the distribution network at the current moment. Specifically, this includes: determining the covariance matrix of the distribution network at the current moment based on the target state value obtained in step S3 and the historical state values ​​of a predetermined number of historical moments before the current moment; performing eigenvalue decomposition on the covariance matrix to determine the target eigenvalue; further obtaining the anomaly probability of the distribution network at the current moment; and finally determining whether there is an anomaly in the distribution network at the current moment by comparing the anomaly probability with the preset probability. The specific implementation process is the same as in the aforementioned embodiment and will not be repeated here.

[0061] S5: Determine the abnormal area of ​​the distribution network at the current moment. Specifically, this includes: if there is an anomaly in the distribution network at the current moment, determine the anomaly contribution of multiple voltage nodes at the current moment to determine the abnormal area. The specific implementation process is the same as the aforementioned embodiment, and will not be repeated here.

[0062] This embodiment also provides an anomaly identification device for a power distribution network, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that performs a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware implementations, or a combination of software and hardware, are also possible and contemplated.

[0063] According to an embodiment of the present invention, an apparatus embodiment for implementing the above-described anomaly identification method for power distribution networks is also provided. Figure 4 This is a schematic diagram of the structure of a power distribution network anomaly identification device according to an embodiment of the present invention, as shown below. Figure 4 As shown, the above-mentioned power distribution network anomaly identification device includes: a monitoring data acquisition module 400, a target state value determination module 402, an anomaly judgment module 404, and an anomaly area determination module 406, wherein:

[0064] The monitoring data acquisition module 400 is used to acquire the monitoring data of the distribution network at the current moment. The monitoring data includes at least: the voltage corresponding to each of the multiple voltage nodes in the distribution network, and the active power and reactive power corresponding to each of the multiple branches in the distribution network.

[0065] The target state value determination module 402 is connected to the monitoring data acquisition module 400. It is used to determine the target state value of the distribution network at the current time based on the monitoring data and the target state estimation model. The target state estimation model is obtained by optimizing and training the state estimation network based on the historical monitoring data of the distribution network at multiple historical times and the corresponding historical state values.

[0066] The anomaly detection module 404 is connected to the target state value determination module 402 and is used to determine whether there is an anomaly in the distribution network at the current moment based on the target state value.

[0067] The abnormal area determination module 406 is connected to the abnormal judgment module 404 and is used to determine the abnormal area of ​​the distribution network at the current moment when there is an abnormality in the distribution network at the current moment.

[0068] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0069] It should be noted that the monitoring data acquisition module 400, target state value determination module 402, anomaly judgment module 404, and anomaly area determination module 406 mentioned above correspond to steps S102 to S108 in the embodiments. The instances and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should be noted that the above modules, as part of the device, can run on a computer terminal.

[0070] It should be noted that the optional or preferred implementation methods of this embodiment can be found in the relevant descriptions in the embodiments, and will not be repeated here.

[0071] The aforementioned distribution network anomaly identification device may also include a processor and a memory. The aforementioned monitoring data acquisition module 400, target state value determination module 402, anomaly judgment module 404, and anomaly area determination module 406 are all stored in the memory as program modules. The processor executes the aforementioned program modules stored in the memory to realize the corresponding functions.

[0072] The processor contains a core that retrieves the corresponding program modules from memory. One or more cores may be configured. Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory includes at least one memory chip.

[0073] According to an embodiment of this application, an embodiment of a non-volatile storage medium is also provided. Optionally, in this embodiment, the non-volatile storage medium includes a stored program, wherein, when the program runs, it controls the device where the non-volatile storage medium is located to execute any of the above-mentioned power distribution network anomaly identification methods.

[0074] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals, and the non-volatile storage medium includes stored programs.

[0075] Optionally, during program execution, the device containing the non-volatile storage medium is controlled to perform the following functions: acquire monitoring data of the distribution network at the current moment, wherein the monitoring data includes at least: the voltage corresponding to each of multiple voltage nodes in the distribution network, and the active power and reactive power corresponding to each of multiple branches in the distribution network; based on the monitoring data, use a target state estimation model to determine the target state value of the distribution network at the current moment, wherein the target state estimation model is obtained by optimizing and training the state estimation network based on historical monitoring data of the distribution network at multiple historical moments and the corresponding historical state values; based on the target state value, determine whether there is an anomaly in the distribution network at the current moment; if there is an anomaly in the distribution network at the current moment, determine the anomaly area of ​​the distribution network at the current moment.

[0076] According to an embodiment of this application, an embodiment of a processor is also provided. Optionally, in this embodiment, the processor is used to run a program, wherein the program executes any of the above-described methods for identifying anomalies in a power distribution network.

[0077] According to an embodiment of this application, an embodiment of a computer program product is also provided. Optionally, in this embodiment, the computer program product includes a computer program that, when executed by a processor, implements the steps of any of the above-described power distribution network anomaly identification methods.

[0078] Optionally, when the aforementioned computer program product is executed on a data processing device, it is suitable to execute an initialization program with the following method steps: acquiring monitoring data of the distribution network at the current moment, wherein the monitoring data includes at least: the voltage corresponding to each of multiple voltage nodes in the distribution network, and the active power and reactive power corresponding to each of multiple branches in the distribution network; based on the monitoring data, using a target state estimation model, determining the target state value of the distribution network at the current moment, wherein the target state estimation model is obtained by optimizing and training the state estimation network based on historical monitoring data of the distribution network at multiple historical moments and the corresponding historical state values; based on the target state value, determining whether there is an anomaly in the distribution network at the current moment; and if there is an anomaly in the distribution network at the current moment, determining the anomaly area of ​​the distribution network at the current moment.

[0079] like Figure 5 As shown, this embodiment of the invention provides an electronic device 10, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: acquiring monitoring data of the distribution network at the current moment, wherein the monitoring data includes at least: the voltage corresponding to each of multiple voltage nodes in the distribution network, and the active power and reactive power corresponding to each of multiple branches in the distribution network; based on the monitoring data, using a target state estimation model, determining the target state value of the distribution network at the current moment, wherein the target state estimation model is obtained by optimizing and training a state estimation network based on historical monitoring data of the distribution network at multiple historical moments and the corresponding historical state values; based on the target state value, determining whether there is an anomaly in the distribution network at the current moment; and if there is an anomaly in the distribution network at the current moment, determining the abnormal area of ​​the distribution network at the current moment.

[0080] The order of the above embodiments of the present invention is merely for description and does not represent the superiority or inferiority of the embodiments.

[0081] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0082] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of modules described above can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between modules, and may be electrical or other forms.

[0083] The modules described above as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0084] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0085] If the aforementioned integrated modules are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable non-volatile storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a non-volatile storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned non-volatile storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0086] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for anomaly identification in a power distribution network, characterized in that, include: Acquire monitoring data of the distribution network at the current moment, wherein the monitoring data includes at least: the voltage corresponding to each of the multiple voltage nodes in the distribution network, and the active power and reactive power corresponding to each of the multiple branches in the distribution network; Based on the monitoring data, a target state estimation model is used to determine the target state value of the distribution network at the current time. The target state estimation model is obtained by optimizing and training the state estimation network based on historical monitoring data of the distribution network at multiple historical times and the corresponding historical state values. Based on the target state value, determine whether there is an anomaly in the distribution network at the current moment; If the distribution network exhibits the anomaly at the current time, determine the abnormal area of ​​the distribution network at the current time.

2. The method according to claim 1, characterized in that, When the target state estimation model includes a graph convolutional layer and an output layer, determining the target state value of the distribution network at the current time based on the monitoring data using the target state estimation model includes: Based on the monitoring data, multiple input features and multiple adjacent feature sets of the distribution network at the current time are determined, wherein the multiple input features and multiple adjacent feature sets correspond one-to-one with the multiple voltage nodes, and the multiple adjacent feature sets are used to indicate the feature correlation degree between the corresponding voltage node and multiple target voltage nodes, and the multiple target voltage nodes represent the voltage nodes among the multiple voltage nodes that are connected to the corresponding voltage node through branches; Based on the graph convolutional layer, the multiple input features and the multiple adjacent feature sets are convolutionally processed to determine the target node features of the multiple voltage nodes at the current time. Based on the output layer, the target node features of the plurality of voltage nodes at the current time are mapped to obtain the target state values ​​of the plurality of voltage nodes at the current time.

3. The method according to claim 2, characterized in that, When the graph convolutional layer comprises multiple layers, the step of performing convolution processing on the multiple input features and the multiple adjacent feature sets based on the graph convolutional layer to determine the target node features of the multiple voltage nodes at the current time includes: Based on multiple graph convolutional layers, the convolutional processing is performed on the multiple input features and the multiple adjacent feature sets in the following manner to determine the target node features of any voltage node among the multiple voltage nodes at the current time: Starting from the first graph convolutional layer among the plurality of graph convolutional layers, the following operations are repeated until the convolutional processing of the last graph convolutional layer is completed: based on the plurality of input features output from the previous graph convolutional layer in any graph convolutional layer, the candidate node features corresponding to each of the plurality of voltage nodes, and any adjacent feature set in the plurality of adjacent feature sets, the candidate node features of any voltage node in any graph convolutional layer are obtained. The candidate node features of the last graph convolutional layer obtained when the convolution processing of the last graph convolutional layer is completed are determined as the target node features of any voltage node at the current time. The target node characteristics of the plurality of voltage nodes at the current time are obtained by using the method of obtaining the target node characteristics of any voltage node at the current time.

4. The method according to claim 3, characterized in that, The step of obtaining the candidate node features of any voltage node in any graph convolutional layer based on the multiple input features output from the previous graph convolutional layer, the candidate node features corresponding to each of the multiple voltage nodes, and any adjacent feature set from the multiple adjacent feature sets includes: Based on the multiple input features output from the previous graph convolutional layer of any graph convolutional layer, the candidate node features corresponding to each of the multiple voltage nodes, and any adjacent feature set, the candidate node features of any voltage node in any graph convolutional layer are obtained in the following manner: ； in, This represents the candidate node features of any voltage node in any graph convolutional layer. Indicates the index of any of the voltage nodes. This represents the index of any of the graph convolutional layers. This represents the activation function. This represents the input feature corresponding to any voltage node among the plurality of input features output by the preceding graph convolutional layer of any graph convolutional layer. The candidate node features represent the output of the preceding graph convolutional layer of any graph convolutional layer for any voltage node. This represents any element in any of the adjacent feature sets. Indicates the index of any target voltage node. This represents the total number of the multiple target voltage nodes. This represents the input feature corresponding to any target voltage node among the multiple input features output by the previous graph convolutional layer of any graph convolutional layer. The candidate node features represent the output of the previous graph convolutional layer of any graph convolutional layer for any target voltage node. This represents the bias feature of the previous convolutional layer of any given graph convolutional layer.

5. The method according to claim 1, characterized in that, Before determining the target state value of the distribution network at the current time using a target state estimation model based on the monitoring data, the method further includes: Based on the historical monitoring data of the distribution network at the multiple historical times and the corresponding historical state values, the objective function of the distribution network at the multiple historical times is determined, wherein the objective function represents the deviation between the operating state of the distribution network at the corresponding historical time and the preset operating state; The optimization objective is to minimize the function value of the objective function of the distribution network at the multiple historical times, and the state estimation network is trained in the optimization until a predetermined number of iterations is reached. The target state estimation model is obtained based on the optimized state estimation network obtained when the predetermined number of iterations is reached.

6. The method according to claim 5, characterized in that, The step of determining the objective function of the distribution network at each of the multiple historical time periods based on the historical monitoring data of the distribution network at each of the multiple historical time periods and the corresponding historical state values ​​includes: Based on the historical monitoring data of the distribution network at the various historical times, the voltage error, active power error and reactive power error of the distribution network at the various historical times are determined. Based on the voltage error, the active power error, and the reactive power error, the physical losses of the distribution network at the various historical times are determined, wherein the physical losses are used to indicate the deviation between the actual voltage drop of the distribution network at the corresponding historical time and the preset voltage drop. Based on the historical monitoring data of the distribution network at various historical times and the corresponding historical state values, the measurement loss of the distribution network at various historical times is determined, wherein the measurement loss is used to indicate the deviation between the historical state value and the predicted state value of the distribution network at the corresponding historical time; the predicted state value is predicted by the state estimation network based on the corresponding historical monitoring data during the optimization training process. Based on the node injection power corresponding to the historical state values ​​of the distribution network at the multiple historical times, the Bayesian prior loss of the distribution network at the multiple historical times is determined, wherein the Bayesian prior loss is used to indicate the deviation between the probability distribution corresponding to the node injection power of the distribution network at the corresponding historical time and the preset probability distribution. Based on the measurement loss, physical loss, and Bayesian prior loss of the distribution network at the various historical times, the objective function of the distribution network at the various historical times is determined.

7. The method according to claim 6, characterized in that, The step of determining the objective function of the distribution network at multiple historical moments based on the measurement loss, physical loss, and Bayesian prior loss of the distribution network at multiple historical moments includes: Based on the measurement loss, physical loss, and Bayesian prior loss of the distribution network at various historical moments, the objective function of the distribution network at various historical moments is determined as follows: ； in, This represents the objective function of the distribution network at each of the multiple historical moments. This represents the measured losses of the distribution network at the various historical times. This represents the physical losses of the power distribution network at the various historical moments. This represents the Bayesian prior loss of the power distribution network at each of the multiple historical moments. This represents the weighting coefficient corresponding to the physical loss. The weight coefficients of the Bayesian prior loss are represented.

8. The method according to claim 1, characterized in that, The step of determining whether there is an anomaly in the distribution network at the current time based on the target state value includes: Based on the target state value and the historical state values ​​of a predetermined number of historical times prior to the current time, the state matrix of the distribution network at the current time is obtained, wherein any element in the state matrix represents the state value of the corresponding voltage node at the corresponding time. Based on the state matrix, the covariance matrix of the distribution network at the current time is obtained, wherein the covariance matrix is ​​used to indicate the correlation of state changes among the multiple voltage nodes in the time dimension; Determine the target eigenvalue corresponding to the covariance matrix, wherein the target eigenvalue represents the eigenvalue that is greater than a preset first eigenvalue among the multiple eigenvalues ​​obtained after eigenvalue decomposition of the covariance matrix; Based on the target feature value and the preset second feature value, the probability of anomaly of the distribution network at the current time is determined, wherein the preset second feature value is greater than the preset first feature value; If the probability of an anomaly is greater than a preset probability, it is determined that the distribution network exhibits the anomaly at the current moment; or If the probability of an anomaly is less than or equal to the preset probability, it is determined that the distribution network does not have the anomaly at the current time.

9. The method according to any one of claims 1 to 8, characterized in that, When the anomaly exists in the distribution network at the current time, determining the abnormal region of the distribution network at the current time includes: When the distribution network has the anomaly at the current time, the anomaly contribution of the multiple voltage nodes at the current time is determined based on the feature vector corresponding to the target feature value. The region consisting of voltage nodes whose abnormal contribution is greater than a preset contribution threshold among the plurality of voltage nodes is determined as the abnormal region.

10. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the distribution network anomaly identification method according to any one of claims 1 to 9.

Images