Power distribution network operation state detection method and device and storage medium

Abstract

The invention discloses a power distribution network operation state detection method and device and a storage medium, and belongs to the field of power systems, and the method comprises the steps: obtaining the real-time index data of a to-be-detected power distribution network, and carrying out the calculation to obtain a multivariable dynamic interaction sequence; according to the multivariable dynamic interaction sequence, determining an inter-region preliminary coupling relationship; after graph structure representation is obtained through calculation according to the preliminary coupling relation between the regions, the graph structure representation is input into a graph neural network model, neighborhood node features can be effectively aggregated, indirect association and implicit influence paths which are difficult to recognize through a traditional method are found out, and the robustness is high. The technical problem that in the prior art, the dynamic coupling relation between the areas is ignored when the mutual influence between the different areas of the power grid is analyzed is solved. And finally, according to a graph neural network model output result, evaluating to obtain a high-risk coupling link, and generating an operation state detection result of the to-be-detected power distribution network, so that the detection precision of the operation state of the power distribution network can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of power systems, and in particular to a method, apparatus, terminal equipment, and storage medium for detecting the operating status of a power distribution network. Background Technology

[0002] In modern power systems, the power distribution network, as the core of energy transmission and distribution, is crucial for ensuring the reliability and stability of power supply through real-time monitoring and analysis of its operational status. With the widespread integration of new loads such as distributed energy sources and electric vehicles, the operating environment of the distribution network is becoming increasingly complex, with significantly enhanced dynamic characteristics. How to accurately capture and intuitively present key operational information in this complex and ever-changing power grid environment has become an important research topic in power systems. Traditional monitoring methods often struggle to handle the complex interrelationships between multiple variables, making it difficult for dispatchers to quickly pinpoint the root cause of system anomalies. Existing distribution network monitoring methods have significant shortcomings in handling dynamic interactions among multiple variables. Many technologies rely on static data analysis at a single moment, making it difficult to adapt to real-time fluctuations in load and power supply. Furthermore, existing methods often neglect the dynamic coupling relationships between different regions of the power grid when analyzing their mutual influence. This limitation prevents the system from comprehensively reflecting the overall operational status of the power grid, especially under high load or abnormal operating conditions, making it difficult to accurately identify potential operational risks. In dynamic monitoring of the distribution network, the core technical challenge lies in effectively characterizing the complex coupling relationships between variable nodes.

[0003] In power distribution networks, variables such as transmission power, voltage amplitude, and phase angle difference interact across different areas, forming a complex dynamic network. Traditional analysis methods struggle to analyze the real-time interactions of these variables across different areas, making it difficult for dispatchers to quickly determine which areas' operating conditions might trigger overall system instability. For example, during peak load periods, voltage fluctuations in one area may affect power transmission in neighboring areas through coupling relationships, but current technologies cannot readily visualize this cross-regional dynamic impact. Therefore, in complex scenarios with multi-variable dynamic interactions, existing power distribution network operating status detection methods suffer from low accuracy. Summary of the Invention

[0004] This invention provides a method for detecting the operating status of a power distribution network, which can solve the problem of low accuracy in the detection of the operating status of a power distribution network in the prior art.

[0005] The distribution network operation status detection method provided by this invention includes: Obtain real-time indicator data of the power distribution network under test; A multivariate dynamic interaction sequence is calculated based on real-time indicator data; Based on the multivariate dynamic interaction sequence, the preliminary coupling relationship between regions is determined; After calculating the graph structure representation based on the preliminary coupling relationship between regions, the graph structure representation is input into the graph neural network model so that the graph neural network model can calculate the enhanced dynamic network representation. Evaluation based on the enhanced dynamic network representation yields high-risk coupled links; Based on high-risk coupling links, the operational status detection results of the distribution network to be detected are generated.

[0006] Furthermore, a graph structure representation will be calculated based on the preliminary coupling relationships between regions, including: Inter-regional interactive data analysis was performed on the preliminary coupling relationship between regions to obtain the first analysis results; Based on the results of the first analysis, the correlation strength between regions is calculated, and an initial coupling relationship matrix is ​​generated. Based on the initial coupling relationship matrix, a graph structure data is constructed to obtain a graph structure representation.

[0007] Furthermore, the graph neural network model computes an enhanced dynamic network representation, including: The graph neural network model updates the representations of each node in the graph structure by aggregating the features of neighboring nodes, thus obtaining the feature aggregation result. Based on the feature aggregation results, calculate the inter-regional propagation effect, analyze the feature transmission path between nodes in the graph structure representation, and obtain the propagation effect matrix; Based on the propagation effect matrix, the nodes in a specific neighborhood of the graph structure representation are updated with weights to obtain the enhanced node representation. Based on the enhanced node representation and combined with the dynamic relationships between regions, the graph structure representation is updated to generate a preliminary dynamic network representation; Based on the initial dynamic network representation, cross-regional propagation features are extracted to generate an enhanced dynamic network representation.

[0008] Furthermore, based on the propagation effect matrix, the specific neighborhood nodes in the graph structure representation are updated with weights to obtain the enhanced node representation, including: Calculate each region in the propagation effect matrix and its corresponding N effect values; where N is a positive integer; When the effect size is greater than the first preset threshold, it is recorded as the first effect size; The region corresponding to the first effect value is designated as the first region; The neighboring nodes of the first region are denoted as specific neighboring nodes; By performing weighted updates on specific neighboring nodes, an enhanced node representation is obtained.

[0009] Furthermore, a multivariate dynamic interaction sequence is calculated based on real-time indicator data, including: The fast Fourier transform algorithm is used to perform time-series decomposition on real-time indicator data, extract the frequency components of each variable in the real-time indicator data, and generate a set of frequency features. When the frequency components of M variables in the frequency feature set are greater than the second preset threshold, the M variables are recorded as the first variable; where M is a positive integer; The first variable is processed by an autoregressive moving average model to obtain stationary time series data; Based on stationary time series data, the cross-correlation coefficients between each variable are calculated to obtain the dynamic interaction matrix between each variable. Principal component analysis was used to extract the main fluctuation features of the dynamic interaction matrix, resulting in a set of dimensionality-reduced eigenvectors. Calculate the variance contribution rate of the eigenvectors in the eigenvector set; When the variance contribution rate of the L feature vectors is lower than the third preset threshold, the L feature vectors are recorded as the first feature vector. Remove the first feature vector to obtain the optimized feature vector set; Based on the optimized feature vector set, a multivariate dynamic interaction sequence is generated.

[0010] Furthermore, based on the multivariate dynamic interaction sequence, preliminary coupling relationships between regions are determined, including: Interaction data is obtained from multivariate time series data, and the dynamic interaction features of the interaction data in the time dimension are extracted using the sliding window method to obtain dynamic interaction data. Calculate the Pearson correlation coefficient among variables in dynamic interactive data to determine the strength of the correlation among variables in dynamic interactive data; Based on the correlation strength, the matrix filling method is used to map the correlation strength to the matrix elements to obtain the initial correlation matrix; The coupling strength between regions is extracted from the initial correlation matrix to obtain the preliminary coupling relationship between regions.

[0011] Furthermore, based on high-risk coupling links, the operational status detection results of the distribution network under test are generated, including: A subset of critical paths is obtained based on high-risk coupling links; Clustering algorithms are used to group subsets of critical paths to obtain regional coupled clusters; Temporal projection processing is performed on the regional coupled cluster to map the dynamic changes within the regional coupled cluster to a two-dimensional plane coordinate system, thus determining the basis for the visualization layout. Color coding and edge weight rendering are overlaid on the basic visual layout to generate an interactive topology map; The operating status detection results of the distribution network under test are obtained based on the interactive topology diagram.

[0012] Another embodiment of the present invention provides a power distribution network operation status detection device, comprising: an acquisition module, a first calculation module, a determination module, a second calculation module, an evaluation module, and a detection module; The acquisition module is used to acquire real-time indicator data of the distribution network under test. The first calculation module is used to calculate a multivariate dynamic interaction sequence based on real-time indicator data; The determination module is used to determine the initial coupling relationship between regions based on a multivariate dynamic interaction sequence; After the second calculation module calculates the graph structure representation based on the preliminary coupling relationship between regions, it inputs the graph structure representation into the graph neural network model so that the graph neural network model can calculate the enhanced dynamic network representation. The evaluation module is used to evaluate based on the dynamic network representation to identify high-risk coupled links; The detection module is used to generate operational status detection results for the distribution network under test based on the enhanced high-risk coupling links.

[0013] Another embodiment of the present invention provides a terminal device, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps of the power distribution network operation status detection method provided by the present invention.

[0014] Another embodiment of the present invention provides a computer-readable storage medium item, including: a stored computer program, which, when the computer program is running, controls the device where the computer-readable storage medium is located to perform the steps of the power distribution network operation status detection method provided by the present invention.

[0015] The following benefits can be obtained by implementing the present invention: The technical solution of this invention acquires real-time index data of the distribution network under test and calculates a multivariate dynamic interaction sequence. This sequence determines the initial coupling relationship between regions, enabling direct extraction and quantification of dynamic interactions between regions from operational data, thus more realistically reflecting the actual operating state of the distribution network. Subsequently, after calculating a graph structure representation based on the initial coupling relationship between regions, this representation is input into a graph neural network model to generate an enhanced dynamic network representation. This effectively aggregates features of neighboring nodes, captures the "cross-regional propagation effect" between non-directly connected regions, and discovers indirect connections and hidden influence paths that are difficult to identify using traditional methods. This solves the technical problem of existing technologies neglecting the dynamic coupling relationship between regions when analyzing the mutual influence between different regions of the power grid, providing a more reliable data foundation for distribution network operation risk assessment. Finally, based on the output of the graph neural network model, high-risk coupling links are evaluated, generating the operation status detection results of the distribution network under test, effectively improving the detection accuracy of the distribution network operation status. Attached Figure Description

[0016] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a distribution network operation status detection method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a power distribution network operation status detection device provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0020] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0021] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0022] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0023] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0024] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0025] See Figure 1 To address the issue of low accuracy in existing power distribution network operation status detection methods, an embodiment of the present invention provides a power distribution network operation status detection method, comprising: S101: Obtain real-time indicator data of the distribution network under test.

[0026] In this embodiment, the real-time index data of the distribution network to be tested is obtained by: collecting real-time index data from each node of the distribution network through a sensor network; wherein, the real-time index data includes, but is not limited to: real-time data of transmission power, voltage amplitude and phase angle difference.

[0027] Specifically, wireless sensor network nodes are deployed in the distribution network. For example, node A is installed at the output end of the main transformer in the substation to collect real-time data with a transmission power of 150kW and a voltage amplitude of 10.5kV. At the same time, node B collects multivariable signals with a phase angle difference of -2.3 degrees at the midpoint of the feeder. The data is transmitted to the central server via the ZigBee protocol at a sampling rate of 1Hz to form a raw dataset containing timestamps. After receiving the data, the server applies the ARIMA(p,d,q) time series analysis algorithm. The parameters p=2, d=1, and q=1 are fitted to the model after the stationarity is confirmed by the ADF test. The specific process is to first eliminate the trend of the power series difference to obtain the residuals, and then optimize the parameters using the AIC criterion. The autocorrelation function (ACF) and partial autocorrelation function (PACF) plots show that the second-order lag is significantly decayed, thereby extracting fluctuation characteristics such as the periodic oscillation amplitude of the power series. Analyzing data collection - real-time data of multiple variables such as transmission power, voltage amplitude, and phase angle difference are collected from each node of the distribution network through a sensor network.

[0028] Furthermore, ARIMA (Autoregressive Integral Moving Average) is a widely used model in time series analysis, particularly suitable for handling non-stationary time series data. It consists of three parts: The AR (Autoregressive) part: models the current value by representing it as a linear combination of past values.

[0029] Part I (Difference): By performing a difference operation on the time series, it is made into a stationary series.

[0030] The MA (Moving Average) component models the current value by representing it as a linear combination of past errors.

[0031] This embodiment further illustrates the differences between applying the ARIMA(p,d,q) time series analysis algorithm and applying the autoregressive model (AR) for data processing: (1) ARIMA(p,d,q) time series analysis algorithm: Data acquisition: Real-time multivariate data on transmission power, voltage amplitude, and phase angle difference are collected from each node of the distribution network through a sensor network to form a raw data set.

[0032] Difference operations (Part I): Difference operations are performed on the original data to eliminate trends. For example, a first-order difference (d=1) is performed on a power series to make the data stationary. The formula for the difference operation is: ; Where Xt is the data value at time t, and ΔXt is the value after difference; Model Fitting (AR and MA Parts): After confirming data stationarity (using the ADF test), an ARIMA model was used for fitting. Specific parameters were set to p=2, d=1, and q=1.

[0033] AR component (p=2): This indicates that the current value is related to the values ​​at the previous two time points. The model form is: ; Where c is a constant term, ϕ1 and ϕ2 are autoregressive coefficients, and ϵt is an error term.

[0034] The MA component (q=1) represents the correlation between the current value and the error at the previous time point. The model form is: ; Where μ is the mean and θ1 is the moving average coefficient.

[0035] (2) Autoregressive model (AR): The autoregressive model is a special case of the ARIMA model, containing only the AR component. It assumes that the current value is linearly correlated with several past values. In claim 2, the autoregressive model is used to process the stabilized data, and the specific steps are as follows: Stabilization: Data is stabilized through difference operations.

[0036] Model Fitting: An autoregressive model is used to fit the stationary data. For example, for a power series, assuming a lag order k=3, the model form is: ; Among them, ϕ1, ϕ2 and ϕ3 are autoregressive coefficients.

[0037] (3) Combined with the data processing flow: Data acquisition: Real-time data on transmission power, voltage amplitude, and phase angle difference are collected through a sensor network to form a raw dataset.

[0038] Difference operation: Perform a first-order difference operation on the original data to eliminate trends and make the data stable.

[0039] Model fitting: The differencing data were fitted using the ARIMA(2,1,1) model. The stationarity of the data was confirmed by the ADF test.

[0040] Optimize model parameters using the AIC criterion. Calculate the autocorrelation function (ACF) and partial autocorrelation function (PACF) plots to determine the lag order. For example, the ACF and PACF plots show a significant decay in lag 2, confirming p=2.

[0041] Extracting fluctuation features: Fluctuation features, such as the periodic oscillation amplitude of the power sequence, are extracted using the fitted ARIMA model. For example, the model fitting yields a periodic oscillation amplitude of 20kW peak-to-peak for the power sequence, and a daily standard deviation of 0.2kV for the voltage amplitude.

[0042] Generate dynamic interaction sequences: Based on the extracted fluctuation features, generate multivariate dynamic interaction sequences for subsequent coupling relationship analysis.

[0043] (4) Summary The ARIMA(p,d,q) model uses differencing to stationary data and combines autoregressive and moving average components for modeling, making it suitable for processing non-stationary time series data.

[0044] Autoregressive Model (AR): Contains only the autoregressive component and is suitable for modeling stationary data. ARIMA models are used to process collected multivariate real-time data, extract fluctuation characteristics, and generate dynamic interaction sequences, providing a foundation for subsequent coupling relationship analysis and visualization.

[0045] Further, an autoregressive model of VAR vectors is introduced to handle multivariate interactions. The lag order is set to k=3. The causal influence coefficient of power on voltage is verified to be 0.65 by Granger causality test. The analysis process includes constructing a cointegration vector error correction model (VECM) to capture long-term equilibrium relationships. For example, the cointegration vector [1,-0.8] of phase angle difference sequence and power represents the dynamic response of a power decrease of 0.8kW corresponding to a 1-degree phase angle change. Finally, a multivariate dynamic interaction sequence is generated. For example, the power at time t in the sequence is Pt=β1*Vt-1+β2*Δθt-2+εt, where β1=12.5 represents the positive contribution of voltage lag effect to power, and β2=-5.2 reflects the negative feedback mechanism of phase angle difference. This reveals the coupled evolution of voltage instability and power transmission caused by load fluctuations in the distribution network, providing a basis for subsequent prediction.

[0046] S102: Calculate the multivariate dynamic interaction sequence based on real-time indicator data.

[0047] In this embodiment, a multivariate dynamic interaction sequence is calculated based on real-time indicator data, including: The fast Fourier transform algorithm is used to perform time-series decomposition on real-time indicator data, extract the frequency components of each variable in the real-time indicator data, and generate a set of frequency features. When the frequency components of M variables in the frequency feature set are greater than the second preset threshold, the M variables are recorded as the first variable; where M is a positive integer; The first variable is processed by an autoregressive moving average model to obtain stationary time series data; Based on stationary time series data, the cross-correlation coefficients between each variable are calculated to obtain the dynamic interaction matrix between each variable. Principal component analysis was used to extract the main fluctuation features of the dynamic interaction matrix, resulting in a set of dimensionality-reduced eigenvectors. Calculate the variance contribution rate of the eigenvectors in the eigenvector set; When the variance contribution rate of the L feature vectors is lower than the third preset threshold, the L feature vectors are recorded as the first feature vector. Remove the first feature vector to obtain the optimized feature vector set; Based on the optimized feature vector set, a multivariate dynamic interaction sequence is generated.

[0048] For example, the second preset threshold is 0.5. When the frequency component of a variable is greater than 0.5 (normalized amplitude), it is recorded as the first variable and used for subsequent stationary time series processing. The third preset threshold is 0.05. When the variance contribution rate of the eigenvector is less than 5%, it is recorded as the first eigenvector and discarded.

[0049] In one specific embodiment, dimensionality reduction is achieved using Principal Component Analysis (PCA). PCA is a statistical method used to project high-dimensional data into a low-dimensional space while preserving as much of the main variation information of the original data as possible. The specific steps are as follows: Standardized data: The data in the dynamic interaction matrix are standardized so that the mean of each feature is 0 and the variance is 1.

[0050] Calculate the covariance matrix: Calculate the covariance matrix for the standardized data. The covariance matrix reflects the linear relationship between the variables.

[0051] Finding eigenvalues ​​and eigenvectors: Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors. Eigenvalues ​​represent the variance contribution rate of each principal component, and eigenvectors represent the projection of the data onto the directions of the principal components.

[0052] Principal components are selected by sorting the eigenvalues ​​by size and choosing the eigenvectors corresponding to the top k eigenvalues ​​as principal components. These principal components retain the main variation information of the original data.

[0053] Dimensionality reduction: Projecting the original data onto these principal components yields a set of dimensionality-reduced eigenvectors. This set of dimensionality-reduced eigenvectors is a low-dimensional representation of the data after PCA processing. Each eigenvector represents a principal component, reflecting the main direction of variation in the original data. With lower dimensionality, redundant information is removed, while retaining the main fluctuation characteristics. The high variance contribution rate of the eigenvectors indicates that these principal components can effectively explain the variation in the original data.

[0054] Furthermore, the dynamic interaction matrix is ​​a high-dimensional matrix that represents the dynamic interaction relationships between different variables. Each row or column represents a variable, and the elements in the matrix represent the correlation or interaction strength between variables. It may have high dimensionality and contain a large amount of redundant information.

[0055] Furthermore, after extracting the optimized feature vector set, the specific steps for generating the multivariate dynamic interaction sequence are as follows: Extracting an optimized feature vector set: After dimensionality reduction using PCA, an optimized feature vector set is obtained. The feature vectors in the optimized feature vector set retain the main fluctuation characteristics of the original data. Feature vectors with a variance contribution rate lower than a preset threshold are removed to ensure that the retained feature vectors can effectively reflect the main dynamic changes in the data.

[0056] Reconstructing Dynamic Interaction Data: Using an optimized set of feature vectors, the data in the original dynamic interaction matrix is ​​projected onto these principal components. The reconstructed data is represented as a linear combination of low-dimensional feature vectors, with each feature vector corresponding to a principal component.

[0057] Generate a dynamic interaction sequence: For each time point t, extract the corresponding optimized feature vector. Arrange these feature vectors in chronological order to form a multivariate dynamic interaction sequence. Each feature vector represents the dynamic state at a time point, reflecting the main fluctuation characteristics at that time point.

[0058] Determine the operating state sequence: By analyzing multivariate dynamic interaction sequences, determine the operating state of distribution network nodes at different points in time. The operating state can be stable, fluctuating, or abnormal, and is determined by the changing trend and magnitude of feature vectors.

[0059] Specific examples are as follows: Suppose the original dynamic interaction matrix M has dimensions n×m, where n represents the number of time points and m represents the number of variables. After dimensionality reduction using PCA, we obtain an optimized feature vector set V with dimensions n×k, where k is the number of principal components.

[0060] Extracting the optimized feature vector set: For each time point t, extract the corresponding feature vector vt. vt is the t-th row in the optimized feature vector set V.

[0061] Generate a dynamic interaction sequence: Arrange the feature vectors of all time points in chronological order to form a multivariate dynamic interaction sequence S. ; where each vt represents the main fluctuation characteristic at time point t.

[0062] Determine the sequence of running states: Analyze the changing trends and magnitudes of the feature vectors in sequence S to determine the operational status at each time point.

[0063] For example, if the feature vector changes significantly at a certain point in time, it may indicate that there is fluctuation or anomaly at that point in time.

[0064] S103: Determine the preliminary coupling relationship between regions based on the multivariate dynamic interaction sequence.

[0065] In this embodiment, the preliminary coupling relationship between regions is determined based on the multivariate dynamic interaction sequence, including: Interaction data is obtained from multivariate time series data, and the dynamic interaction features of the interaction data in the time dimension are extracted using the sliding window method to obtain dynamic interaction data. Calculate the Pearson correlation coefficient among variables in dynamic interactive data to determine the strength of the correlation among variables in dynamic interactive data; Based on the correlation strength, the matrix filling method is used to map the correlation strength to the matrix elements to obtain the initial correlation matrix; The coupling strength between regions is extracted from the initial correlation matrix to obtain the preliminary coupling relationship between regions.

[0066] Furthermore, the coupling strength between regions is extracted from the initial correlation matrix to obtain the preliminary coupling relationship between regions, specifically: When the coupling strength is greater than the fourth preset threshold, the relationship between the regions corresponding to that threshold is determined to be a strong coupling relationship. When the coupling strength is less than the fourth preset threshold, the relationship between the regions corresponding to that threshold is determined to be a normal coupling relationship. By organizing and combining the strong and ordinary coupling relationships, preliminary coupling relationships between regions can be obtained.

[0067] In addition, cluster analysis was used to group the regions based on the initial coupling relationship, and the clustering results between the regions were obtained. Based on the clustering results, a network analysis method is used to construct an inter-regional coupling network, calculate the degree of each node in the network, and determine the structure of the inter-regional coupling network. Key nodes are extracted from the coupled network structure, and the key influence areas of the coupling relationship between regions are obtained by using the node importance ranking method.

[0068] For example, the fourth preset threshold is 0.6. When the coupling strength is greater than 0.6, it is determined to be a strong coupling relationship; otherwise, it is a normal coupling relationship.

[0069] S104: After calculating the graph structure representation based on the preliminary coupling relationship between regions, the graph structure representation is input into the graph neural network model so that the graph neural network model can calculate the enhanced dynamic network representation.

[0070] In this embodiment, a graph structure representation will be calculated based on the preliminary coupling relationships between regions, including: Inter-regional interactive data analysis was performed on the preliminary coupling relationship between regions to obtain the first analysis results; Based on the results of the first analysis, the correlation strength between regions is calculated, and an initial coupling relationship matrix is ​​generated. Based on the initial coupling relationship matrix, a graph structure data is constructed to obtain a graph structure representation.

[0071] In one specific embodiment, the graph neural network model computes an enhanced dynamic network representation, including: The graph neural network model updates the representations of each node in the graph structure by aggregating the features of neighboring nodes, thus obtaining the feature aggregation result. Based on the feature aggregation results, calculate the inter-regional propagation effect, analyze the feature transmission path between nodes in the graph structure representation, and obtain the propagation effect matrix; Based on the propagation effect matrix, the nodes in a specific neighborhood of the graph structure representation are updated with weights to obtain the enhanced node representation. Based on the enhanced node representation and combined with the dynamic relationships between regions, the graph structure representation is updated to generate a preliminary dynamic network representation; Based on the initial dynamic network representation, cross-regional propagation features are extracted to generate an enhanced dynamic network representation.

[0072] Specifically, based on the propagation effect matrix, weighted updates are performed on specific neighborhood nodes in the graph structure representation to obtain an enhanced node representation, including: Calculate each region in the propagation effect matrix and its corresponding N effect values; where N is a positive integer; When the effect size is greater than the first preset threshold, it is recorded as the first effect size; The region corresponding to the first effect value is designated as the first region; The neighboring nodes of the first region are denoted as specific neighboring nodes; By performing weighted updates on specific neighboring nodes, an enhanced node representation is obtained.

[0073] For example, the first preset threshold is 0.7. The first preset threshold is used as the effect value in the propagation effect matrix. When the propagation effect value is greater than 0.7, it is recorded as the first effect value, and the corresponding area is considered a high-influence area.

[0074] S105: Evaluation based on the enhanced dynamic network representation yields high-risk coupled links.

[0075] In this embodiment, high-risk coupled links are identified through evaluation based on the enhanced dynamic network representation, specifically: Calculate the node connection strength in the enhanced dynamic network representation; When the node connection strength is greater than the fifth preset threshold, the connection weights between nodes are calculated using graph theory algorithms to obtain a set of high-strength connections. Node interaction data is extracted from a set of high-strength connections, and data flow is simulated using an information propagation model to determine the propagation efficiency distribution. Based on the propagation efficiency distribution analysis of potential wave paths, the wave propagation direction is tracked through a random walk algorithm to obtain a set of wave paths; The coupling relationship of the fluctuation path set is analyzed. If the interaction frequency of the nodes in the path of the fluctuation path set is higher than the seventh preset threshold, high-risk coupling links are obtained by clustering algorithm.

[0076] For example, the fifth preset threshold is 0.8. When the node connection strength is greater than 0.8, it is included in the high-strength connection set. The seventh preset threshold is 0.7. When the node interaction frequency in the fluctuating path is higher than 0.7 (normalized value), high-risk coupled links are identified by clustering algorithm.

[0077] Furthermore, based on the high-risk coupling links, a set of risky links is determined; The dynamic characteristics of the links are obtained from the risk link set, and the stability of the links is evaluated by time series analysis to obtain the unstable link subset; Network topology analysis is used to calculate the inter-link dependency strength of unstable link subsets and determine the priority of high-risk coupled links; Based on the priority of high-risk coupled links, generate link optimization configurations and determine the final network adjustment strategy.

[0078] Specifically, in the enhanced dynamic network representation, a complex network model with 100 nodes is first constructed, where nodes represent supply chain enterprises and edge weights represent connection strength. Embedding vectors are trained using a deep learning algorithm, a graph convolutional network (GCN), with an initial threshold set to 0.75 to capture the dynamic interaction relationships between nodes. For example, the weight between the enterprise supplier of node A and the manufacturer of node B is calculated to be 0.82. Exceeding the threshold triggers subsequent evaluation. Next, the SIR (Susceptible-Infected-Recovered) model was used to assess potential fluctuation paths in the information propagation simulation stage. The parameters included the infection rate β=0.3 and the recovery rate γ=0.1. The propagation was simulated starting from node A that exceeded the threshold, and the path probability was calculated by iterating 100 steps. The fluctuation intensity of the path from A to C was found to be 0.65, which represents the potential risk propagation link. The analysis process involved matrix multiplication to solve the propagation equation dI / dt=βSI-(γ+μ)I, where μ is the removal rate of 0.05, which quantifies the cumulative impact of the fluctuation path. When identifying high-risk coupled links, a variant of the PageRank algorithm is used to calculate link centrality with a threshold of 0.4. The simulated paths are sorted, and the link centrality score of ABC is identified as 0.52. This score exceeds the threshold and is therefore considered high-risk. By comparing the historical volatility of 0.12 with the current volatility of 0.28, it is confirmed that the coupling effect amplifies the risk by 20%. This forms a closed-loop decision support logic chain for the entire process, from representation construction to simulation evaluation to risk assessment, ensuring network stability through automated calculations without human intervention.

[0079] S106: Generate the operation status detection results of the distribution network to be detected based on the high-risk coupling links.

[0080] In this embodiment, based on high-risk coupling links, the operational status detection results of the distribution network to be detected are generated, including: A subset of critical paths is obtained based on high-risk coupling links; Clustering algorithms are used to group subsets of critical paths to obtain regional coupled clusters; Temporal projection processing is performed on the regional coupled cluster to map the dynamic changes within the regional coupled cluster to a two-dimensional plane coordinate system, thus determining the basis for the visualization layout. Color coding and edge weight rendering are overlaid on the basic visual layout to generate an interactive topology map; The operating status detection results of the distribution network under test are obtained based on the interactive topology diagram.

[0081] In this embodiment, a clustering algorithm is used to group the critical path subset to obtain regional coupled clusters, specifically: A graph segmentation algorithm is used to decompose the link data in high-risk coupled links to obtain a critical path subset; The K-means clustering algorithm is used to group the path features in the critical path subset to obtain similar interaction patterns; Based on similar interaction patterns, the coupling strength between paths is calculated to obtain regional coupled clusters.

[0082] Furthermore, if the number of paths in a regionally coupled cluster exceeds an eighth preset threshold, the cluster is then subjected to secondary clustering to obtain an optimized sub-cluster. By extracting the interaction pattern features of each group of clusters through the optimized sub-clusters, a pattern feature set is obtained; Based on the pattern feature set, the decision tree algorithm is used to classify the features and determine the stability of the regional coupled clusters; Based on the stability assessment results, classification labels for regionally coupled clusters are generated, and the final cluster division is obtained.

[0083] For example, the eighth preset threshold is 10. When the number of paths in the regional coupled cluster exceeds 10, a second clustering is performed to optimize the sub-cluster.

[0084] Specifically, when processing high-risk coupled link data, the first step is to extract a subset of critical paths from the overall links through network graph analysis. For example, given a graph containing 100 high-risk links, where nodes represent system components and edges represent interaction dependencies, Dijkstra's algorithm is used to calculate the single-source shortest path from the source node to all high-risk nodes. The threshold is set to paths with a risk score greater than 0.7, and 15 subsets of critical paths are extracted. For each high-risk coupled link analyzed, the critical path subset needs to be extracted by first identifying links with a risk value greater than 0.8. The average path length is 5 nodes, and the total risk scores are 0.85, 0.92, etc. The analysis process includes initializing the distance array to infinity, with only the source node being 0, and then updating the shortest path through node-by-node relaxation operations until there are no more updates, thereby ensuring that the extracted subset covers 80% of the cumulative risk. Subsequently, these 15 path subsets were input into a K-means clustering algorithm for grouping, with a preset k value of 3. The path vectors were measured using Euclidean distance (each path was converted into a node embedding vector with a dimension equal to 10 of the node features). The iterative process started with randomly initializing 3 centroids, calculating the distance from each path to the centroid, and reassigning cluster labels until the sum of squares within each cluster changed to less than 0.01, resulting in three clusters: Cluster 1 contained 5 paths with an average similarity of 0.78, representing a dense database interaction pattern; Cluster 2 had 6 paths with a similarity of 0.82, representing a dense API call pattern; and Cluster 3 had 4 paths with a similarity of 0.75, reflecting a file transfer pattern. Analysis showed an inter-cluster variance of 0.15, validating the effectiveness of the grouping. Finally, by identifying similar interaction patterns through these clusters, regional coupled clusters are formed. For example, the path of cluster 1 is mapped to the system regional graph, and the subgraphs with a node overlap coefficient greater than 0.6 within the cluster are calculated to generate 3 regional clusters with a total coverage of 45 nodes. The risk concentration is increased to 0.88, which facilitates subsequent monitoring and optimization. The entire process is automated through the graph database Neo4j to ensure the logical continuity from extraction to clusters and avoid isolated analysis.

[0085] In one specific embodiment, temporal projection processing is performed on the region-coupled cluster to map the dynamic changes within the region-coupled cluster to a two-dimensional plane coordinate system, thereby determining the basis for the visualization layout. Specifically: Principal component analysis was used to extract the main features of regional coupled clusters, resulting in a set of feature vectors; the dataset of regional coupled clusters includes time series and spatial distribution information. For the feature vector set, the t-SNE algorithm is applied for dimensionality reduction and mapped to a two-dimensional plane coordinate system to obtain a two-dimensional coordinate point set; Based on the two-dimensional coordinate point set, calculate the Euclidean distance between each point. If the Euclidean distance between two points is less than the ninth preset threshold, then these two points are assigned to the same sub-cluster, and the sub-cluster partitioning result is obtained. Based on the sub-cluster partitioning results, the time series variation trend of each sub-cluster is obtained, and the moving average method is used for smoothing to obtain the smoothed trend series. Based on the smoothed trend sequence, the dynamic change rate of each sub-cluster in the time dimension is calculated to obtain the rate distribution set; For the set of rate distributions, the k-means clustering algorithm is applied to group the sub-clusters according to their dynamic rate of change, resulting in a set of grouped clusters. Based on the grouped cluster set, a visual layout in a two-dimensional plane coordinate system is generated, and color coding is used to represent different dynamic change rates to determine the basis of the visual layout.

[0086] For example, the ninth preset threshold is 0.5. In a two-dimensional plane projection, when the Euclidean distance between two points is less than 0.5 (normalized coordinates), they are classified into the same sub-cluster.

[0087] Specifically, assuming the cluster contains 100 nodes, each node has data points with a time series length of T=365 days, collected from temperature and humidity indicators of an IoT sensor network; the input data matrix X is a standardized matrix of 100×365 with a mean of 0 and a variance of 1. Then, the covariance matrix C=X^TX / (100-1) is calculated, the eigenvalues ​​and eigenvectors are solved, and the eigenvectors corresponding to the two largest eigenvalues ​​are selected as the projection basis to form the projection matrix P (2×365). Finally, the two-dimensional coordinates Y=PX^T are used, where the x-coordinate of Y ranges from -5.2 to 4.8, and the y-coordinate ranges from -3.1 to 3.9, realizing the initial visualization layout basis for dynamic changes. Next, Kullback-Leibler divergence (KL-divergence) optimization is performed on the node positions on the two-dimensional plane to minimize the difference in time series distribution before and after projection, ensuring the fidelity of the layout. For example, the KL divergence D_KL(q||p)=∑q(t)log(q(t) / p(t)) of the original high-dimensional distribution q(t) and the projected two-dimensional distribution p(t) is calculated. The divergence value is 0.45 under the initial layout. Through the iterative gradient descent algorithm θ_new=θ_old-α∇D_KL, where the learning rate α=0.01, the divergence is reduced to 0.12 after 50 iterations. The coupling strength between nodes, such as the Pearson correlation coefficient r=0.78 between nodes i and j, is mapped to the Euclidean distance d=1 / r≈1.28, forming a compact cluster layout. Meanwhile, a time-series smoothing filter with business relevance is incorporated. A Gaussian filter with σ=2.5 is used to perform convolution smoothing on the Y coordinate sequence, Y_smoothed=Y*G(σ), to eliminate noise interference. For example, the original y coordinate fluctuation standard deviation std=1.2 is reduced to 0.3 after smoothing, thus improving layout stability. Finally, the attractive and repulsive forces between nodes were simulated using the force-directed algorithm Fruchterman-Reingold, F_attr=c^2 / d_ij*r_ij (c=node coupling threshold 0.5) and F_rep=-k^2 / d_ij*r_ij (k=ideal distance 1.0). After 100 iterations of temperature cooling T=T0*(1-iter / max_iter) from T0=10 to 0, the final visual layout was generated. The density of the central cluster nodes reached 0.85, and the spacing between the peripheral nodes was evenly distributed between 2.1 and 3.5, ensuring an intuitive two-dimensional presentation of dynamic changes, which is convenient for subsequent anomaly detection and analysis.

[0088] Furthermore, by overlaying color coding and edge weight rendering onto the basic visual layout, an interactive topology map is generated as follows: Data on inter-regional coupling relationships is obtained from a data source, and node and edge information is stored in a graph database to obtain the initial topology. For the initial topology, the association strength between each node is calculated, and the edge weight value is determined by weighted average method. If the weight exceeds the tenth preset threshold, it is marked as a strongly correlated edge; otherwise, it is marked as a weakly correlated edge, thus obtaining the set of edge weights after classification. From the set of edge weights after classification, a color coding scheme is generated by mapping strongly associated edges to dark colors and weakly associated edges to light colors. Based on the color coding scheme and edge weight set, and based on the visualization layout, an interactive topology graph is constructed. The force-directed layout algorithm is used to adjust the node positions to obtain the visualization topology structure. For the visualized topology, acquire overall operational status data, map status changes to node colors, and generate dynamic rendering effects. The interactive topology map is generated through dynamic rendering and real-time updates, responding to user interactions via an event-triggered mechanism. This interactive topology map effectively presents the coupling relationships between different areas and their impact on the overall operating status, significantly improving the accuracy of power distribution network operation status detection.

[0089] For example, the tenth preset threshold is 0.6. When the edge weight is greater than 0.6, it is marked as a strongly associated edge; otherwise, it is marked as a weakly associated edge.

[0090] Specifically, when processing data on coupling relationships between different regions, the initial topology data is first imported using information technology tools such as Python's NetworkX library. Nodes represent five regions A to E, and the initial edges are automatically extracted based on historical operation logs. For example, the coupling strength between A and B is 0.7, B and C is 0.5, C and D is 0.8, D and E is 0.6, and E and A is 0.4. Then, Dijkstra's algorithm is applied to calculate the shortest path to identify critical connection paths. For example, the shortest path length from A to D is 1.2. Potential bottlenecks are assessed by analyzing links on the path whose cumulative coupling value exceeds the threshold of 0.6. Next, a color encoding mechanism is overlaid, and the HSL color space algorithm is used to map the coupling strength to the hue value. The specific formula is hue=60*(intensity-0.1Analyzing visualization requirements-) / (0.9-0.1), so that the edge with an intensity of 0.7 is rendered as orange (hue=45°) and 0.5 is rendered as green (hue=15°). The gradient color spectrum is generated by Matplotlib's colormap function to ensure that the low coupling edge is highlighted in blue to show the abnormal risk. Edge weight rendering uses a linear scaling algorithm, where weight w_i = intensity_i * 10 + base thickness 2, so that the edge width of 0.8 intensity reaches 8 pixels. Combined with SVG export, scalable vector graphics are realized to support browser rendering. Subsequently, the overall operational status impact assessment was performed, and the PageRank algorithm was used to iteratively calculate the node centrality. The initial damping factor was 0.85. After 20 iterations, the impact scores for regions A were 0.25, B was 0.18, C was 0.22, D was 0.20, and E was 0.15. The contribution of each region to the system was quantified by the formula Total Impact = ∑(Centrality_j*Coupling_{i,j}). For example, the disturbance impact of region C on the overall state reached 0.35, which exceeded the warning line of 0.3 and triggered a red alarm. Finally, an interactive topology map is generated using the D3.js library. Users can trigger JavaScript events by hovering the mouse to dynamically highlight paths. For example, after clicking on edge AB, the algorithm recalculates the influence chain in real time, showing that the probability of disturbance propagation is 65%, and the delay of affecting region E is rendered in animation, ensuring a seamless integration of the visual layout from static topology to dynamic simulation, thereby revealing the causal chain that the unbalanced coupling leads to a 15% decrease in system stability.

[0091] This invention addresses the challenges of complex real-time data interaction among multiple nodes in distribution networks and the difficulty in accurately identifying fluctuation propagation paths. It collects multi-variable data such as transmission power, voltage amplitude, and phase angle difference, employs time-series analysis to extract fluctuation characteristics, constructs an initial correlation matrix, and calculates correlation coefficients to quantify the coupling strength between variables. Then, it utilizes a graph neural network to aggregate features of neighboring nodes, captures cross-regional propagation effects, and generates an enhanced dynamic network representation. This invention assesses high-risk coupling links through information propagation simulation, extracts critical path subsets, and clusters them into regional coupling clusters. Combining time-series projection and color coding, it generates an interactive topology map, intuitively presenting the coupling relationships between regions and their impact on the overall operating status. This invention effectively solves the challenges of dynamic interactive analysis and visualization in distribution networks, improves the accuracy of fluctuation path identification and risk prevention, and provides efficient technical support for distribution network operation optimization.

[0092] like Figure 2 As shown, based on the above method embodiments, corresponding apparatus embodiments are provided; An embodiment of the present invention provides a power distribution network operation status detection device, including: an acquisition module 201, a first calculation module 202, a determination module 203, a second calculation module 204, an evaluation module 205, and a detection module 206; The acquisition module 201 is used to acquire real-time indicator data of the distribution network to be tested; The first calculation module 202 is used to calculate a multivariate dynamic interaction sequence based on real-time indicator data; The determination module 203 is used to determine the preliminary coupling relationship between regions based on the multivariate dynamic interaction sequence; After the second calculation module 204 calculates the graph structure representation based on the preliminary coupling relationship between regions, it inputs the graph structure representation into the graph neural network model so that the graph neural network model can calculate the enhanced dynamic network representation. Evaluation module 205 is used to evaluate based on the dynamic network representation to obtain high-risk coupling links; The detection module 206 is used to generate the operation status detection results of the distribution network under test based on the enhanced high-risk coupling links.

[0093] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention, and can implement the distribution network operation status detection method provided by any of the above-described method embodiments of the present invention.

[0094] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0095] Based on the above embodiments of the power distribution network operation status detection method, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the power distribution network operation status detection method of any embodiment of the present invention.

[0096] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0097] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0098] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0099] Based on the above-described method embodiments, another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the power distribution network operation status detection method described in any of the above-described method embodiments of the present invention.

[0100] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0101] The above description represents the 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 principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method of detecting an operating state of a power distribution network, characterized by, The method comprises the following steps: obtaining real-time index data of a power distribution network to be detected; calculating a multivariate dynamic interaction sequence according to the real-time index data; determining an initial inter-regional coupling relationship according to the multivariate dynamic interaction sequence; inputting a graph structure representation calculated according to the initial inter-regional coupling relationship into a graph neural network model to enable the graph neural network model to calculate an enhanced dynamic network representation; evaluating the enhanced dynamic network representation to obtain a high-risk coupling link; generating an operation state detection result of the power distribution network to be detected according to the high-risk coupling link.

2. The power distribution network operating state detection method of claim 1, wherein The graph structure representation calculated according to the initial inter-regional coupling relationship comprises the following steps: performing inter-regional interaction data analysis on the initial inter-regional coupling relationship to obtain a first analysis result; calculating an inter-regional correlation strength according to the first analysis result to generate an initial coupling relationship matrix; constructing graph structure data according to the initial coupling relationship matrix to obtain the graph structure representation.

3. The power distribution network operating state detection method of claim 2, wherein The graph neural network model calculating the enhanced dynamic network representation comprises the following steps: the graph neural network model updates each node representation in the graph structure representation by aggregating neighborhood node features to obtain a feature aggregation result; calculating an inter-regional propagation effect according to the feature aggregation result, analyzing a feature transmission path between nodes in the graph structure representation, and obtaining a propagation effect matrix; weighting and updating a specific neighborhood node in the graph structure representation according to the propagation effect matrix to obtain an enhanced node representation; updating the graph structure representation according to the enhanced node representation and combining the inter-regional dynamic relationship to generate a preliminary dynamic network representation; extracting a cross-regional propagation feature according to the preliminary dynamic network representation to generate the enhanced dynamic network representation.

4. The power distribution network operating state detection method of claim 3, wherein, The weighting and updating of the specific neighborhood node in the graph structure representation according to the propagation effect matrix to obtain the enhanced node representation comprises the following steps: calculating each region and its corresponding N effect values in the propagation effect matrix; wherein N is a positive integer; when the effect value is greater than a first preset threshold, it is recorded as a first effect value; the region corresponding to the first effect value is recorded as a first region; the neighborhood node of the first region is recorded as the specific neighborhood node; weighting and updating the specific neighborhood node to obtain the enhanced node representation.

5. The power distribution network operating state detection method of claim 4, wherein, The calculation of the multivariate dynamic interaction sequence according to the real-time index data comprises the following steps: using a fast Fourier transform algorithm to perform time series decomposition on the real-time index data, extract frequency components of each variable in the real-time index data, and generate a frequency feature set; when the frequency components of M variables in the frequency feature set are greater than a second preset threshold, the M variables are recorded as first variables; wherein M is a positive integer; performing autoregressive moving average model processing on the first variables to obtain stationary time series data; calculating the cross-correlation coefficients between the variables according to the stationary time series data to obtain a dynamic interaction matrix between the variables; extract main fluctuation characteristics of the dynamic interaction matrix by using a principal component analysis algorithm to obtain a feature vector set after dimension reduction; calculate variance contribution rates of feature vectors in the feature vector set; when variance contribution rates of L feature vectors are lower than a third preset threshold, mark the L feature vectors as first feature vectors; remove the first feature vectors to obtain an optimized feature vector set; generate the multivariate dynamic interaction sequence according to the optimized feature vector set.

6. The power distribution network operating state detection method of claim 5, wherein, The determining of the preliminary coupling relationship between regions according to the multivariate dynamic interaction sequence comprises: obtain interaction data from the multivariate time sequence, extract dynamic interaction characteristics of the interaction data in the time dimension by using a sliding window method to obtain dynamic interaction data; calculate Pearson correlation coefficients between variables in the dynamic interaction data to determine correlation strengths between variables in the dynamic interaction data; map the correlation strengths to matrix elements by using a matrix filling method according to the correlation strengths to obtain an initial correlation matrix; extract coupling strengths between regions from the initial correlation matrix to obtain the preliminary coupling relationship between regions.

7. The power distribution network operating state detection method of claim 6, wherein, The generating of the operation state detection result of the power distribution network to be detected according to the high-risk coupling link comprises: obtain a key path subset according to the high-risk coupling link; perform grouping processing on the key path subset by using a clustering algorithm to obtain a regional coupling cluster; perform time sequence projection processing on the regional coupling cluster to map dynamic changes in the regional coupling cluster to a two-dimensional plane coordinate system to determine a visualization layout basis; superimpose color coding and edge weight rendering on the visualization layout basis to generate an interactive topology diagram; obtain the operation state detection result of the power distribution network to be detected according to the interactive topology diagram.

8. A power distribution network operating state detection apparatus characterized by comprising: comprises: an acquisition module, a first calculation module, a determination module, a second calculation module, an evaluation module, and a detection module; the acquisition module is configured to acquire real-time index data of a power distribution network to be detected; the first calculation module is configured to calculate a multivariate dynamic interaction sequence according to the real-time index data; the determination module is configured to determine a preliminary coupling relationship between regions according to the multivariate dynamic interaction sequence; the second calculation module is configured to input a graph structure representation obtained according to the preliminary coupling relationship between regions into a graph neural network model to enable the graph neural network model to calculate an enhanced dynamic network representation; the evaluation module is configured to evaluate according to the dynamic network representation to obtain a high-risk coupling link; the detection module is configured to generate an operation state detection result of the power distribution network to be detected according to the enhanced high-risk coupling link.

9. A terminal device, comprising: comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and the processor implements the power distribution network operation state detection method in any one of claims 1-7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, comprises: A computer program product stored in the computer readable storage medium, wherein the computer program product, when executed, controls a device in which the computer readable storage medium is located to perform the power distribution network operating state detection method according to any one of claims 1-7.

Images