Intelligent risk early warning method, device and equipment for power distribution network and medium

Abstract

The invention relates to the technical field of data processing, and discloses an intelligent risk early warning method, device and equipment for a power distribution network, and a medium. Historical risk monitoring data of the power distribution network under multiple dimensions are fused through a graph convolutional network and a variational auto-encoder; a variational recurrent neural network and a long-short term memory network are trained in combination with historical risk fault data of the power distribution network, and an attention mechanism is introduced in the training process to generate a risk assessment model; acquiring real-time risk monitoring data of the power distribution network under multiple dimensions to extract multi-dimensional real-time fusion features and inputting the multi-dimensional real-time fusion features into the risk assessment model for processing to obtain a real-time risk assessment level so as to further process the multi-dimensional real-time fusion features through a multi-criterion decision analysis method and a fuzzy inference system; the target risk assessment level is obtained, the level is compared with the risk early warning threshold value, if the level exceeds the threshold value, early warning is triggered, and the accuracy of power distribution network risk assessment is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an intelligent risk early warning method, device, equipment and medium for a distribution network. Background Art

[0002] With the gradual popularization of distribution network automation, the operation and maintenance of power systems have become increasingly complex, and risk assessment of power grids has become increasingly important.

[0003] Traditional power grid risk assessment methods typically rely on static data, such as historical equipment failure records and manually set rule bases. This method cannot effectively respond to real-time changes in the environment and equipment status, resulting in the inability to promptly detect potential operational risks when faced with emergencies or complex environmental changes. This in turn makes the assessment results inaccurate and unable to provide accurate early warning information.

[0004] It can be seen that how to improve the accuracy of traditional power grid risk assessment solutions to provide accurate early warning has become a technical problem that needs to be urgently solved by those skilled in the art. Summary of the Invention

[0005] The present invention provides an intelligent risk early warning method, device, equipment and medium for distribution network, which solves the problem of how to improve the accuracy of traditional power grid risk assessment schemes to provide accurate early warning.

[0006] To solve the above technical problems, the first aspect of the present invention provides an intelligent risk early warning method for a distribution network, comprising:

[0007] Obtain historical risk monitoring data of the distribution network in multiple dimensions and its corresponding historical risk fault data, and fuse the historical risk monitoring data in multiple dimensions through a graph convolutional network and a variational autoencoder to obtain multi-dimensional historical fusion features;

[0008] Using the multi-dimensional historical fusion features and the historical risk failure data as training samples, training a variational recurrent neural network and a long short-term memory network, and introducing an attention mechanism into the trained model to generate a risk assessment model;

[0009] Acquire real-time risk monitoring data of the distribution network under the multiple dimensions and the real-time scenarios to which they belong, and extract multi-dimensional real-time fusion features corresponding to the real-time risk monitoring data to input them into the risk assessment model for processing, thereby obtaining a real-time risk assessment level;

[0010] Based on the real-time risk assessment level, the multi-dimensional real-time fusion features are processed through a multi-criteria decision analysis method and a fuzzy reasoning system to obtain a target risk assessment level for comparison with a risk warning threshold determined according to the real-time scenario, and an early warning is triggered based on the comparison result.

[0011] As one of the preferred solutions, historical risk monitoring data in multiple dimensions includes historical equipment status data and historical meteorological environment data; among them,

[0012] The historical risk monitoring data under multiple dimensions are fused through a graph convolutional network and a variational autoencoder to obtain multi-dimensional historical fusion features, including:

[0013] Various devices in the distribution network are used as nodes, and based on each of the nodes, the association relationship between the historical device status data and the historical meteorological environment data is used as an edge to construct a heterogeneous graph;

[0014] Extracting node embedding features of the heterogeneous graph through the graph convolutional network to obtain a historical fusion feature matrix;

[0015] Based on the maximum likelihood function, the variational autoencoder is used to reduce the dimension of the historical risk monitoring data under the multiple dimensions to obtain a historical monitoring feature dimension reduction matrix;

[0016] The historical fusion feature matrix and the historical monitoring feature dimensionality reduction matrix are fused through a dynamic weighting mechanism to obtain a multi-dimensional historical fusion feature.

[0017] As one preferred solution, the historical risk failure data includes historical scenarios of historical risk monitoring data of corresponding dimensions; wherein,

[0018] Before obtaining the real-time risk monitoring data of the distribution network in the multiple dimensions and the real-time scenarios to which it belongs, the method includes:

[0019] Based on the multi-task learning method, a multi-task learning network model is constructed, which includes a shared layer and several task-specific output layers, and a loss function is defined for each task;

[0020] Using the historical scenarios as labels to label the historical risk failure data corresponding to each dimension, and obtaining historical labeled samples as a training set;

[0021] Using the training set to train the multi-task learning network model, and optimizing each of the loss functions, so as to extract historical basic features common to each scene in the training set through the trained multi-task learning network model;

[0022] Several preset classification models are trained respectively using the historical basic features as base classifiers, and the Bayesian optimization method is used to tune the hyperparameters of the base classifiers, so as to construct a scene classification model in combination with the multi-task learning network model after training.

[0023] As one preferred solution, the scene classification model includes a shared layer, several task-specific output layers, a base classifier and an integration layer; wherein,

[0024] The obtaining of the real-time risk monitoring data of the distribution network in the multiple dimensions and the real-time scenarios to which it belongs includes:

[0025] Acquire real-time risk monitoring data of the distribution network under the multiple dimensions, and input the real-time risk monitoring data into the shared layer and the plurality of task-specific output layers for processing, thereby obtaining real-time general features and a plurality of preliminary prediction results;

[0026] Selecting a high-confidence preliminary prediction result from a plurality of the preliminary prediction results and fusing it with the real-time general feature to obtain a real-time basic feature;

[0027] The real-time basic features are classified by the base classifier to obtain a number of prediction results which are input as new features into the integration layer for processing to obtain the real-time scene.

[0028] As one of the preferred solutions, the risk assessment model includes an input layer, a VRNN hidden layer, an LSTM layer and an attention layer, wherein the VRNN hidden layer is a variational recurrent neural network with a hidden vector added; wherein,

[0029] The extracting of multi-dimensional real-time fusion features corresponding to the real-time risk monitoring data and inputting them into the risk assessment model for processing to obtain a real-time risk assessment level includes:

[0030] The real-time risk monitoring data under multiple dimensions are fused through the graph convolutional network and the variational autoencoder to obtain multi-dimensional real-time fusion features and input them into the VRNN hidden layer through the input layer;

[0031] An encoder in the variational recurrent neural network is used to quantify the potential state distribution of the multi-dimensional real-time fusion feature at each moment to obtain a plurality of latent vectors. Based on the latent vectors, a decoder in the variational recurrent neural network is used to reconstruct the multi-dimensional real-time fusion feature to generate a plurality of reconstructed features containing temporal dynamic information.

[0032] Each of the reconstructed features is input into the LSTM layer for processing, and the hidden state at each moment is generated and input into the attention layer for linear transformation. Each transformation result is mapped to a nonlinear space through an activation function, and a dot product operation is performed with the corresponding weight vector to obtain multiple attention scores to quantify the contribution of each of the hidden states to the final risk assessment result. The corresponding hidden states are weightedly summed according to each contribution to obtain the final risk assessment result and its corresponding real-time risk assessment level.

[0033] As one preferred solution, based on the real-time risk assessment level, the multi-dimensional real-time fusion features are processed by a multi-criteria decision analysis method and a fuzzy inference system to obtain a target risk assessment level for comparison with a risk warning threshold determined according to the real-time scenario, and an early warning is triggered according to the comparison result, including:

[0034] Processing the real-time risk assessment level and the multi-dimensional real-time fusion features by the hierarchical analysis method to obtain the risk weight of the real-time risk monitoring data of each dimension, and determining the risk warning threshold based on each risk weight and the real-time scenario;

[0035] A fuzzy inference system is used to perform a secondary risk assessment on the multi-dimensional real-time fusion features to obtain a target risk assessment level for comparison with the risk warning threshold, and an early warning is triggered based on the comparison result.

[0036] As one preferred solution, the real-time risk assessment level and the multi-dimensional real-time fusion features are processed by the analytic hierarchy process to obtain the risk weight of the real-time risk monitoring data of each dimension, and the risk warning threshold is determined based on each risk weight and the real-time scenario, including:

[0037] Based on the analytic hierarchy process, a target layer is constructed through the real-time risk assessment level, and a criterion layer is constructed using the multi-dimensional real-time fusion feature;

[0038] Compare each data in the criterion layer and the target layer in pairs to form a pairwise comparison matrix for eigenvalue decomposition, and normalize the decomposition results to obtain the risk weight of the real-time risk monitoring data in each dimension;

[0039] Based on the real-time scenario, the risk weight increment of the risk monitoring data of each dimension is determined, and based on each risk weight increment, multiple risk threshold adjustment factors are generated to perform weighted summation on each risk weight to obtain a risk warning threshold.

[0040] A second aspect of the present invention provides an intelligent risk early warning device for a distribution network, comprising:

[0041] A multi-dimensional feature fusion module is used to obtain historical risk monitoring data of the distribution network in multiple dimensions and its corresponding historical risk fault data, and fuse the historical risk monitoring data in multiple dimensions through a graph convolutional network and a variational autoencoder to obtain multi-dimensional historical fusion features;

[0042] An assessment model construction module is used to train a variational recurrent neural network and a long short-term memory network using the multi-dimensional historical fusion features and the historical risk failure data as training samples, and introduce an attention mechanism into the trained model to generate a risk assessment model;

[0043] A real-time risk assessment module is used to obtain real-time risk monitoring data of the distribution network under multiple dimensions and the real-time scenarios to which it belongs, and to extract multi-dimensional real-time fusion features corresponding to the real-time risk monitoring data to input them into the risk assessment model for processing to obtain a real-time risk assessment level;

[0044] The risk warning trigger module is used to process the multi-dimensional real-time fusion features based on the real-time risk assessment level through a multi-criteria decision analysis method and a fuzzy reasoning system to obtain a target risk assessment level for comparison with the risk warning threshold determined according to the real-time scenario, and trigger an early warning based on the comparison result.

[0045] A third aspect of the present invention provides an electronic device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the intelligent risk warning method for the distribution network as described above is implemented.

[0046] A fourth aspect of the present invention provides a computer-readable storage medium, which includes a stored computer program, wherein when the device where the computer-readable storage medium is located executes the computer program, the intelligent risk warning method for the distribution network as described above is implemented.

[0047] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0048] (1) By integrating historical risk monitoring data from multiple dimensions and using advanced deep learning technology for data fusion and model training, we can more comprehensively capture the risk characteristics of the distribution network, thereby improving the accuracy of risk assessment. The introduction of the attention mechanism enables the model to focus more on key information, improves the generalization ability of the model, and enables it to maintain good performance in different scenarios.

[0049] (2) By acquiring the risk monitoring data of the distribution network in real time and quickly generating the risk assessment level, once the risk assessment level exceeds the warning threshold, an early warning can be triggered immediately, providing a strong guarantee for the safe operation of the distribution network; using multi-criteria decision analysis and fuzzy reasoning system to process the multi-dimensional real-time fusion features, more comprehensive and accurate risk assessment results can be obtained, providing decision makers with a more reliable decision basis. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the implementation. Obviously, the drawings described below are only some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 This is a flow chart of an intelligent risk early warning method for a distribution network provided by an embodiment of the present invention;

[0052] Figure 2 This is a structural diagram of an intelligent risk warning device for a distribution network provided by an embodiment of the present invention;

[0053] Figure 3 This is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings and embodiments. Obviously, the embodiments described are only some embodiments of the present invention, rather than all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0055] In the description of this application, the terms "first," "second," "third," etc. are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first," "second," "third," etc. may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0056] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be a communication between the two components. The terms "vertical", "horizontal", "left", "right", "up", "down" and similar expressions used herein are for illustrative purposes only, and do not indicate or imply that the device or component referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to the specific circumstances.

[0057] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this application have the same meanings as those commonly understood by those skilled in the art. The terms used in this specification are only for describing specific embodiments and are not intended to limit the present invention. Those skilled in the art will understand the specific meanings of the above terms in this application according to specific circumstances.

[0058] In one embodiment, if Figure 1 As shown, the first aspect of the present invention provides an intelligent risk early warning method for a distribution network, comprising:

[0059] S1. Obtain historical risk monitoring data of the distribution network in multiple dimensions and its corresponding historical risk fault data, and fuse the historical risk monitoring data in multiple dimensions through a graph convolutional network and a variational autoencoder to obtain multi-dimensional historical fusion features;

[0060] Specifically, the present invention collects risk monitoring data of the distribution network from various dimensions such as equipment status, meteorological environment, load fluctuation, and operation tickets; among them, equipment status data includes voltage, current, temperature, failure rate, etc., meteorological environment data includes wind speed, rainfall, temperature, etc., operation ticket information includes task type, operation sequence, operator, etc., load fluctuation data includes power load change, etc. The monitoring data under these various dimensions can reflect the fault situation of the distribution network; and risk fault data includes fault records, fault data scenarios, risk assessment levels, maintenance conditions, specific early warning measures, etc.

[0061] By collecting historical risk monitoring data and its corresponding historical risk fault data in multiple dimensions of the distribution network in the past preset period, and cleaning and processing them, such as removing outliers, and using the Z-Score normalization method to unify these numerical data into the range of zero mean and unit variance, it is convenient for subsequent processing and thus obtains more accurate results.

[0062] In one embodiment, the historical risk monitoring data in multiple dimensions includes historical equipment status data and historical meteorological environment data; wherein,

[0063] The historical risk monitoring data under multiple dimensions are fused through a graph convolutional network and a variational autoencoder to obtain multi-dimensional historical fusion features, including:

[0064] Various devices in the distribution network are used as nodes, and based on each of the nodes, the association relationship between the historical device status data and the historical meteorological environment data is used as an edge to construct a heterogeneous graph;

[0065] Extracting node embedding features of the heterogeneous graph through the graph convolutional network to obtain a historical fusion feature matrix;

[0066] Based on the maximum likelihood function, the variational autoencoder is used to reduce the dimension of the historical risk monitoring data under the multiple dimensions to obtain a historical monitoring feature dimension reduction matrix;

[0067] The historical fusion feature matrix and the historical monitoring feature dimensionality reduction matrix are fused through a dynamic weighting mechanism to obtain a multi-dimensional historical fusion feature.

[0068] The present invention fuses historical risk monitoring data in multiple dimensions by combining a graph convolutional network (GCN) with a variational autoencoder (VAE) to obtain multi-dimensional historical fusion features, specifically including:

[0069] In the initial stage of data fusion, feature extraction is performed based on GCN. That is, various devices in the distribution network are regarded as nodes. Based on these nodes, the association between historical device status data and historical meteorological environment data and other monitoring data is used as edges to construct a heterogeneous graph to intuitively reflect the mutual influence and association between device status and meteorological environment and other monitoring data. GCN is used to extract complex nonlinear features to enhance the correlation between multi-source data. Then, GCN aggregates the features of the nodes in the heterogeneous graph through multi-layer convolution operations to achieve feature propagation and fusion within the neighborhood. That is, node embedding features are extracted through two layers of GCN to form a historical fusion feature matrix. Among them, the forward propagation formula of GCN is as follows:

[0070]

[0071] Where H (l) is the feature matrix of the l-th layer node; is the normalized adjacency matrix of the heterogeneous graph; W (l) is the trainable weight matrix of the lth layer; σ is a nonlinear activation function (such as ReLU function).

[0072] In multi-dimensional data fusion scenarios, GCN can automatically learn the feature similarities between nodes and achieve comprehensive modeling of device status and environmental status.

[0073] VAE then approximates the distribution by maximizing the likelihood function, maps the input historical risk monitoring data into the latent space, and realizes data feature extraction and anomaly repair by minimizing the reconstruction error. The objective function of VAE is:

[0074]

[0075] Where x is the input data, i.e., historical risk monitoring data; z is the latent variable, i.e., latent vector; q φ (z|x) is the encoder distribution; p θ (x|z) is the decoder reconstruction distribution; D KL The Kullback-Leibler divergence measures the difference between the encoder output and the prior distribution. The VAE encoder compresses high-dimensional data into a low-dimensional representation, while the decoder reconstructs the original data from this low-dimensional representation. In this process, the VAE can also repair abnormal data, improving data accuracy and reliability. Through VAE processing, a historical monitoring feature reduction matrix is obtained, which contains the feature information after dimensionality reduction.

[0076] Finally, a dynamic weighting mechanism is used to concatenate the feature data after VAE dimensionality reduction with the embedded features extracted by GCN to generate multi-dimensional historical fusion features. The dynamic weighting mechanism adjusts the input data by weights, and the weights are dynamically modified according to key variables such as equipment operating status, load fluctuations, and meteorological conditions, thereby improving the accuracy and real-time performance of the data fusion results. The weight adjustment formula is:

[0077]

[0078] Where, is the initial weight; ΔW i (t) is the weight adjustment amount that changes over time.

[0079] The multi-dimensional historical fusion feature expression is:

[0080] X=[W eqp ·F equipment ,W atm ·F atmosphere ,W ctrl ·F controls ]

[0081] Where W eqp 、W atm and W ctrl They are historical equipment status data F equipment , historical meteorological environment dataF atmosphere and historical operation ticket information F controls The dynamic weights are continuously adjusted based on historical and real-time data to ensure the comprehensiveness and flexibility of the feature vector expression. It should be noted that the above formula is only an example of a multi-dimensional historical fusion feature expression. This formula should also include data and weights from other dimensions such as load fluctuations, which will not be elaborated here.

[0082] The present invention uses multi-source data fusion technology to introduce a feature extraction method based on graph convolutional networks, and fuses historical risk monitoring data of distribution networks in multiple dimensions to improve the correlation between multi-dimensional data. In the further step of data fusion, a variational autoencoder is introduced to reduce the dimensionality of the feature matrix, so that VAE can perform distributed learning and anomaly repair on complex multi-dimensional data through its encoder and decoder structure, effectively improving the stability and robustness of data fusion; a dynamic weight mechanism is adopted to ensure the real-time and adaptability of feature fusion, thereby improving the accuracy and flexibility of the risk prediction model.

[0083] S2. Using the multi-dimensional historical fusion features and the historical risk failure data as training samples, training a variational recurrent neural network and a long short-term memory network, and introducing an attention mechanism into the trained model to generate a risk assessment model;

[0084] Specifically, the present invention adopts a deep learning method to capture time series characteristics and the interaction between dimensions through a multi-layer structure, dynamically predicts the risk of input data through a variational recurrent neural network (VRNN), and combines a long short-term memory network (LSTM) to model and predict complex time series data (such as equipment state changes and load fluctuations), thereby effectively capturing time series characteristics and abnormal fluctuations to improve the sensitivity and accuracy of risk warnings. The specific training process of the risk assessment model is as follows:

[0085] The historical risk assessment levels in the historical risk failure data are used as labels to annotate the multi-dimensional historical fusion features as training samples, and a hybrid model architecture combining VRNN and LSTM is constructed, which includes an input layer, a VRNN hidden layer, an LSTM layer, an attention layer and an output layer. The VRNN hidden layer is a variational recurrent neural network with a hidden vector. VRNN can capture the temporal dependencies and potential variables of the data, while LSTM is good at dealing with long-term dependency problems. In this model, the VRNN hidden layer can be used to perform temporal modeling on the input training samples and extract the potential state. Then, the output of the VRNN hidden layer is used as the input of the LSTM layer to further capture the long-term dependency of the data, and the output of the LSTM layer is introduced into the attention layer to dynamically adjust the weights of the input data at different time steps, so that the model can pay more attention to key information when predicting and The output of the LSTM layer is weighted and summed to obtain a weighted feature representation for the final risk assessment. After training the model with historical risk failure data as labels and multi-dimensional historical fusion features as input, a suitable loss function (such as cross entropy loss, mean square error, etc.) is selected, and hyperparameters (such as learning rate, batch size, etc.) are adjusted according to the characteristics of the data. The model parameters are updated using the backpropagation algorithm and optimizer (such as Adam, SGD, etc.). During the training process, the performance of the model is regularly evaluated, such as indicators such as accuracy, recall rate, and F1 score. Based on the evaluation results, the model is then tuned, such as adjusting the model architecture, adding regularization terms, and adjusting the weight of the attention mechanism. After sufficient training and tuning, the final risk assessment model is obtained. The model can receive new multi-dimensional fusion features as input and output corresponding risk assessment results.

[0086] S3. Acquire the real-time risk monitoring data of the distribution network under the multiple dimensions and the real-time scenarios to which they belong, and extract the multi-dimensional real-time fusion features corresponding to the real-time risk monitoring data to input them into the risk assessment model for processing to obtain a real-time risk assessment level;

[0087] In one embodiment, the historical risk failure data includes historical scenarios of historical risk monitoring data of corresponding dimensions; wherein,

[0088] Before obtaining the real-time risk monitoring data of the distribution network in the multiple dimensions and the real-time scenarios to which it belongs, the method includes:

[0089] Based on the multi-task learning method, a multi-task learning network model is constructed, which includes a shared layer and several task-specific output layers, and a loss function is defined for each task;

[0090] Using the historical scenarios as labels to label the historical risk failure data corresponding to each dimension, and obtaining historical labeled samples as a training set;

[0091] Using the training set to train the multi-task learning network model, and optimizing each of the loss functions, so as to extract historical basic features common to each scene in the training set through the trained multi-task learning network model;

[0092] Several preset classification models are trained respectively using the historical basic features as base classifiers, and the Bayesian optimization method is used to tune the hyperparameters of the base classifiers, so as to construct a scene classification model in combination with the multi-task learning network model after training.

[0093] Specifically, the present invention adopts the multi-task learning (MTL) method to simultaneously model the scenarios in which historical risk fault data are located in each dimension, so as to realize dynamic identification and classification prediction of complex operation scenarios. The core idea of multi-task learning is to enable the multi-task learning network model to share parameters between different tasks by simultaneously learning the feature representations of multiple tasks in the shared layer of the model. The model includes a shared layer and several task-specific output layers; wherein, the shared layer can extract basic features common to all scenarios, such as equipment status features, load fluctuation features, etc.; the task-specific output layer can further extract refined features for specific scenarios (such as emergency repairs or bad weather) to improve the classification accuracy of the model under specific tasks; then define a loss function for each task, which can be determined according to actual needs, such as cross entropy loss, mean square error, etc. Therefore, the MTL modeling formula is as follows:

[0094]

[0095] Where, L total is the total loss function, which represents the sum of the weighted losses of multiple tasks; M is the total number of tasks (which can also be understood as the total number of scenarios, such as emergency repair work, daily maintenance, severe weather and other different scenarios); λ m is the loss weight of task m, which can be dynamically adjusted according to the importance of the task; L m (θ) is the loss function of task m; θ is the model parameter.

[0096] The introduction of multi-task learning enables the model to adapt to the classification requirements of multiple different scenarios at the same time, avoiding the need to train independent models for each scenario, reducing training resource consumption, and improving the overall robustness of the model.

[0097] Then, the historical scenarios are used as labels to annotate the corresponding historical risk failure data under each dimension to form a historical annotated sample set; among them, the classification scenarios include emergency repair scenarios: higher weights are needed to pay attention to equipment failure rates and load fluctuations; daily maintenance scenarios: focus on historical risk weights and operation records; severe weather scenarios: increase the weight of the impact of meteorological data on risk assessment, etc.; and the historical annotated sample set is used as a training set to train the multi-task learning network model, and each loss function is optimized during the training process, so that the model can accurately extract the common historical basic features of each scenario in the training set and output preliminary prediction results; then an ensemble learning strategy is adopted, by combining multiple basic classifiers such as Random Forest (RF), XGBoost and Support Vector Machine (SVM). Machine, SVM), etc., to achieve more stable classification results, that is, through the historical basic features extracted by the shared layer, these classification models are trained as base classifiers, and the output results of these classification models are input as new features into the logistic regression model for integration, and the final results are generated for comparison with the labels. Based on the comparison results, the Bayesian optimization method is used to tune the hyperparameters of the base classifier to improve the classification performance. The scenario classification model is constructed by combining the multi-task learning network model after training and the tuned base classifier, which includes a shared layer, several task-specific output layers, a base classifier and an integration layer. The model can receive real-time risk monitoring data and output real-time scenario classification results.

[0098] During the scene classification process, the present invention introduces a multi-task learning method to model multiple scenes simultaneously, so that it can optimize multiple related tasks in the same model at the same time through shared feature representation, improve the model's generalization ability for different scenes, and effectively avoid the model's overfitting problem in scenarios with insufficient data; and adopts a Stacking ensemble learning strategy to significantly improve the accuracy and stability of scene classification by combining multiple basic classifiers; the constructed model helps to improve the safety and reliability of the distribution network and reduce the probability and impact of failures.

[0099] In one embodiment, obtaining the real-time risk monitoring data of the distribution network in the multiple dimensions and the real-time scenarios to which they belong includes:

[0100] Acquire real-time risk monitoring data of the distribution network under the multiple dimensions, and input the real-time risk monitoring data into the shared layer and the plurality of task-specific output layers for processing, thereby obtaining real-time general features and a plurality of preliminary prediction results;

[0101] Selecting a high-confidence preliminary prediction result from a plurality of the preliminary prediction results and fusing it with the real-time general feature to obtain a real-time basic feature;

[0102] The real-time basic features are classified by the base classifier to obtain a number of prediction results which are input as new features into the integration layer for processing to obtain the real-time scene.

[0103] Specifically, the present invention collects risk monitoring data of the distribution network in multiple dimensions in real time (the type is the same as the historical risk monitoring data, only the time is different) through various sensors deployed in the distribution network (such as temperature sensors, current sensors, voltage sensors, etc.) and from various storage systems or clouds and databases, and cleans the collected real-time data to remove noise and outliers, and then standardizes the data to ensure that the data in different dimensions have the same scale, and inputs the preprocessed real-time risk monitoring data into the shared layer and several task-specific output layers of the scenario classification model respectively, so that the shared layer extracts features from the input real-time risk monitoring data to obtain real-time general features, which reflect the common information of risk failures in different dimensions, and enables each task-specific output layer to make preliminary predictions on the input real-time risk monitoring data to obtain several preliminary prediction results, which reflect the specific information of risk failures in different dimensions.

[0104] Then, based on factors such as the model's output probability and historical prediction accuracy, a confidence assessment is conducted on each preliminary prediction result to determine its reliability. Results with high confidence (such as the top five) are selected from several preliminary prediction results as the basis for subsequent processing. The selected preliminary prediction results are fused with real-time general features to obtain real-time basic features. These features integrate the common and specific information of risk failures in different dimensions, providing a more comprehensive basis for subsequent classification processing.

[0105] Finally, the real-time basic features are input into the pre-trained base classifier for classification processing, and several prediction results are output. These results reflect the possibility of risk failure in different scenarios. The prediction results output by the base classifier are input into the integration layer as new features for processing. The integration layer obtains the final real-time scene classification result by combining the prediction results of different base classifiers. The integration process of the integration layer can be expressed as follows:

[0106] C=f meta (h1(X),h2(X),...,h n (X))

[0107] Where C is the final scene classification result; h i (X) is the output result of the i-th basic model; f meta is a meta-classifier (usually logistic regression or other lightweight classifiers).

[0108] In another embodiment, the hierarchical classification of scenarios includes emergency repair scenarios, routine maintenance, severe weather, and emergency scenarios, etc., supporting the model to automatically switch between different scenarios to ensure the accuracy and pertinence of risk assessment. Scenario classification is achieved through a machine learning classification model, and its classification formula is as follows:

[0109]

[0110] in, is a set of scenarios (e.g., emergency repair, maintenance, and special weather conditions); P(c|X) is the output probability of the scenario classification model. The scenario classification model is trained using a support vector machine (SVM) using a historically labeled sample set consisting of historical risk failure data corresponding to each dimension, using historical scenarios as labels.

[0111] The present invention adopts a multi-task learning network model. Through the structure of shared layers and task-specific output layers, the scheme can process multiple tasks simultaneously and extract common historical basic features and preliminary prediction results; further, by fusing high-confidence preliminary prediction results with real-time general features, more comprehensive real-time basic features are obtained, providing more accurate and comprehensive information for subsequent classification processing; combining machine learning and data analysis technology, it realizes the intelligent prediction and classification of distribution network risk faults, which helps to improve the intelligence level of the power grid and promote the development of the power grid towards a more intelligent and automated direction.

[0112] In one embodiment, extracting the multi-dimensional real-time fusion features corresponding to the real-time risk monitoring data and inputting them into the risk assessment model for processing to obtain a real-time risk assessment level includes:

[0113] The real-time risk monitoring data under multiple dimensions are fused through the graph convolutional network and the variational autoencoder to obtain multi-dimensional real-time fusion features and input them into the VRNN hidden layer through the input layer;

[0114] An encoder in the variational recurrent neural network is used to quantify the potential state distribution of the multi-dimensional real-time fusion feature at each moment to obtain a plurality of latent vectors. Based on the latent vectors, a decoder in the variational recurrent neural network is used to reconstruct the multi-dimensional real-time fusion feature to generate a plurality of reconstructed features containing temporal dynamic information.

[0115] Each of the reconstructed features is input into the LSTM layer for processing, and the hidden state at each moment is generated and input into the attention layer for linear transformation. Each transformation result is mapped to a nonlinear space through an activation function, and a dot product operation is performed with the corresponding weight vector to obtain multiple attention scores to quantify the contribution of each of the hidden states to the final risk assessment result. The corresponding hidden states are weightedly summed according to each contribution to obtain the final risk assessment result and its corresponding real-time risk assessment level.

[0116] Specifically, the data fusion process for real-time risk monitoring data is the same as that for historical risk monitoring data, so we will not elaborate on this here. The multi-dimensional real-time fusion features obtained by fusion are then input into the VRNN hidden layer through the input layer of the risk assessment model. This layer is modeled using the VRNN. The VRNN effectively handles complex nonlinear dynamic features and long-term dependencies by adding hidden variables to the standard recurrent neural network and performing distributed modeling of the state at each time step. Taking the current time step t as an example, the VRNN calculation is divided into the following steps:

[0117] The encoder calculates the input multi-dimensional real-time fusion feature x by the following formula t The potential state distribution of t :

[0118]

[0119] Where μ t and σ t are the mean and standard deviation respectively, and are obtained from the previous time step hidden state h t-1 and the current input.

[0120] The decoder generates the output distribution as follows:

[0121]

[0122] in, is the feature of the reconstructed time step t.

[0123] The hidden state update of VRNN is similar to that of traditional RNN and is expressed as follows:

[0124] h t =f(h t-1 ,x t ,z t )

[0125] Where f is the state transfer function of RNN, which is usually implemented using GRU (Gated Recurrent Unit) or LSTM.

[0126] Taking the above process as an example, the reconstructed features calculated at each moment are input into the LSTM layer for processing. The LSTM layer effectively solves the gradient vanishing and gradient exploding problems of traditional RNNs on long time series by introducing forget gates and input gates. The state update formula of LSTM is as follows:

[0127] Forget Gate:

[0128] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0129] Where W f is the weight matrix of the forget gate; h t-1 is the hidden state at the previous moment; x t is the input at the current moment; b f is the bias of the forget gate; σ is the Sigmoid activation function.

[0130] Input Gate:

[0131] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0132] Where W i is the weight matrix of the input gate; b i is the bias of the input gate.

[0133] Cell status update:

[0134] Update cell state C t It consists of two parts: the output f of the forget gate t Control to retain the cell state C at the previous moment t-1 The ratio of the input gate output i t Control the current state of candidate cells The addition ratio is expressed by the following formula:

[0135]

[0136] Where W C is the weight matrix of the candidate state; b C is the bias of the candidate state; tanh is the hyperbolic tangent activation function.

[0137] Output gate and hidden state update:

[0138] The output gate determines the hidden state h at the current moment tThe generation of , which is through the combination of the hidden state of the previous moment and the current input, through the Sigmoid activation function, generates a gating signal that controls the output ratio, which is expressed by the following formula:

[0139] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0140] h t =o t tanh(C t )

[0141] Where W o is the weight matrix of the output gate; b o is the bias of the output gate. The introduction of LSTM effectively improves the risk prediction ability of the model in multiple time steps.

[0142] In order to further improve the model's attention to key features, the attention mechanism is introduced. The hidden state at each moment is input into the attention layer. The hidden state is transformed linearly and the tanh activation function is applied to map the transformed result to the nonlinear space. The dot product of the transformed result and the weight vector is taken to obtain the attention score. Based on the score, the contribution of the hidden state at different time steps to the final risk assessment result is calculated, and the risk assessment weight is dynamically adjusted to perform weighted summation on the corresponding hidden state to obtain the final risk assessment result and its corresponding real-time risk assessment level and output it through the output layer. The process is expressed by the following formula:

[0143] e t =v T tanh(W h h t +b h )

[0144]

[0145] Where, e t is the attention score; v is the weight vector used to map the transformed result to a scalar; W h is the weight matrix, which represents the linear transformation of the hidden state; b h is the bias vector; α t is the attention weight at time step t, which is between 0 and 1; R is the final risk assessment result vector; b r 、W r are model parameters; It is a real-time risk assessment level (low, medium, high risk).

[0146] This paper introduces VRNN for time series modeling, effectively enhancing the model's ability to model complex dynamic features and improve risk prediction accuracy. It also combines LSTM and attention mechanisms to enable the model to dynamically select and fuse features, allocate attention according to task requirements, improve the robustness of the model, further improve the model's expressiveness on long time series, and enhance its sensitivity to sudden risks.

[0147] S4. Based on the real-time risk assessment level, the multi-dimensional real-time fusion features are processed using a multi-criteria decision analysis method and a fuzzy inference system to obtain a target risk assessment level for comparison with a risk warning threshold determined according to the real-time scenario, and an early warning is triggered based on the comparison result;

[0148] In one embodiment, step S4 includes:

[0149] Processing the real-time risk assessment level and the multi-dimensional real-time fusion features by the hierarchical analysis method to obtain the risk weight of the real-time risk monitoring data of each dimension, and determining the risk warning threshold based on each risk weight and the real-time scenario;

[0150] A fuzzy inference system is used to perform a secondary risk assessment on the multi-dimensional real-time fusion features to obtain a target risk assessment level for comparison with the risk warning threshold, and an early warning is triggered based on the comparison result.

[0151] The present invention is based on the Multi-Criteria Decision Making (MCDM) method, which comprehensively analyzes multiple risk factors such as equipment status, load fluctuations, and meteorological environment to achieve risk warning in complex environments. MCDM quantifies the impact of different risk factors and sorts multi-dimensional data for decision-making, thereby optimizing the accuracy of risk assessment and early warning mechanisms. Within the MCDM framework, the Analytic Hierarchy Process (AHP) and Fuzzy Inference System (FIS) are further introduced to dynamically adjust risk warning thresholds, ensuring that the early warning mechanism can be adaptively adjusted under different environmental conditions to effectively reduce false alarm and missed alarm rates. This allows on-site operators to obtain accurate risk warnings and avoidance suggestions in advance before risks occur, thereby significantly improving operational safety and power system reliability.

[0152] In one embodiment, the real-time risk assessment level and the multi-dimensional real-time fusion features are processed by the analytic hierarchy process to obtain the risk weight of the real-time risk monitoring data of each dimension, and the risk warning threshold is determined based on each risk weight and the real-time scenario, including:

[0153] Based on the analytic hierarchy process, a target layer is constructed through the real-time risk assessment level, and a criterion layer is constructed using the multi-dimensional real-time fusion feature;

[0154] Compare each data in the criterion layer and the target layer in pairs to form a pairwise comparison matrix for eigenvalue decomposition, and normalize the decomposition results to obtain the risk weight of the real-time risk monitoring data in each dimension;

[0155] Based on the real-time scenario, the risk weight increment of the risk monitoring data of each dimension is determined, and based on each risk weight increment, multiple risk threshold adjustment factors are generated to perform weighted summation on each risk weight to obtain a risk warning threshold.

[0156] Specifically, when constructing a hierarchical model based on the hierarchical analysis method, the present invention sets the target layer as the real-time risk assessment level, which is the ultimate goal of decision-making, and the criterion layer is composed of multi-dimensional real-time fusion features; then, the dimensional features in the criterion layer are compared pairwise to judge their importance relative to the target layer (real-time risk assessment level), forming a pairwise comparison matrix A = [a ij ], which is represented by the following formula:

[0157]

[0158] Where a ij is the importance ratio of factor i to factor j, which is usually determined by expert evaluation based on a 9-level scaling method, while the diagonal elements are always 1 and the matrix satisfies symmetry;

[0159] The matrix is then decomposed into eigenvalues to obtain the eigenvectors and normalized to obtain the risk weights of the real-time risk monitoring data in each dimension, that is, the weights of each risk factor. The incremental risk weights of the risk monitoring data in each dimension are then determined based on the real-time scenario. For example, in severe weather or high-load scenarios, the weight of meteorological data is increased:

[0160] λ atm =λ atm +β weather

[0161] Where λ atm is the weight of the current meteorological factor on the risk assessment threshold, and the initial value can be preset based on historical data; β weather The incremental meteorological risk weight represents the incremental risk weight due to deteriorating weather conditions (such as typhoons, heavy rains, etc.). It can be dynamically calculated using the relationship model between real-time monitored weather data (such as wind speed and rainfall) and risk weights:

[0162] β weather=k1·wind_speed+k2·rain_intensity

[0163] Where wind_speed is the real-time wind speed; rain_intensity is the real-time rainfall; k1 and k2 are empirical coefficients obtained by fitting historical data.

[0164] For example, in the emergency repair scenario, the weight of the equipment failure rate is increased:

[0165] λ eqp =λ eqp +β repair

[0166] Where λ eqp is the weight of the current equipment operating status on the risk assessment threshold, and its initial value is set by the failure statistics data during normal equipment operation; β repair The incremental risk weight of the equipment in the fault repair scenario indicates the incremental weight due to an increase in the equipment failure rate or the risk of the repair operation. The failure rate is dynamically obtained through real-time equipment monitoring data:

[0167] β repair =k·fault_rate

[0168] Where fault_rate is the current fault rate (for example, the ratio of the number of faulty devices to the total number of devices); k is an empirical coefficient obtained by fitting historical data.

[0169] Then, based on each risk weight increment, multiple risk threshold adjustment factors are generated to perform weighted summation on the corresponding risk weights to obtain the risk warning threshold, which is expressed as follows:

[0170] λ t =λ0+Δλ(t)

[0171]

[0172] Where λ t is the risk threshold adjustment factor at the current time t, which represents the threshold state of the current risk assessment system and is dynamically adjusted with real-time monitoring data; λ0 is the initial threshold; Δλ(t) is the risk weight increment that is dynamically adjusted over time; λ 总 is the risk warning threshold; W i is the weight of each risk factor.

[0173] The present invention uses the hierarchical analysis method to comprehensively and systematically analyze all aspects of the problem, while allowing decision makers to adjust the weights and importance of factors according to actual conditions, with a high degree of flexibility; by integrating multi-dimensional features in real time and constructing a criterion layer, as well as determining risk weight increments and adjusting risk thresholds based on real-time scenarios, the method can achieve real-time risk assessment and early warning, and promptly reflect risk changes, so that the final risk warning threshold can provide a scientific reference basis for decision makers, helping to formulate effective risk management strategies and response measures.

[0174] The fuzzy inference system adopted in the present invention is based on fuzzy logic theory and models complex nonlinear systems through fuzzy rules. The main steps include fuzzification, inference rules and defuzzification: the multi-dimensional real-time fusion features and their corresponding risk weight input variables are fuzzified and fuzzy sets are defined, such as equipment status: low load, medium load, high load; meteorological data: clear, cloudy, bad weather; risk weight: low risk, medium risk, high risk, etc.; and a fuzzy rule base is established through expert experience or historical data, such as: IF the equipment status is "high load" AND the meteorological data is "bad weather", THEN the risk level is "high"; IF the equipment status is "medium load" AND the meteorological data is "cloudy", THEN the risk level is "medium", etc.; then the risk level under different input conditions is calculated based on the fuzzy rules, and the inference formula is as follows:

[0175] R=max(min(x1,x2,x3))

[0176] Where x1, x2, and x3 are the fuzzy set values of equipment status, meteorological data, and load fluctuation in the multi-dimensional real-time fusion features, respectively.

[0177] Finally, the fuzzy output is converted into a specific risk level (low risk, medium risk or high risk):

[0178]

[0179] Where μ i is the membership degree of each element in the fuzzy set; R i is the corresponding risk level value.

[0180] After obtaining the risk prediction results, they are compared with the risk warning threshold, and a warning is triggered when the threshold is exceeded. This invention introduces the MCDM framework, integrates multi-dimensional risk factors, and improves the comprehensiveness and sensitivity of the risk warning mechanism. It also combines AHP to achieve dynamic weighting and dynamic threshold adjustment of each risk factor, ensuring the accuracy of the warning mechanism in different scenarios. Through FIS, fuzzy risk assessment is achieved under complex conditions, significantly reducing the false alarm and missed alarm rates, and improving the system's adaptability and warning accuracy.

[0181] In the embodiment of the present application, based on the problem of how to improve the accuracy of traditional power grid risk assessment schemes for accurate early warning, an intelligent risk early warning method for distribution networks is designed. By introducing a graph convolutional network for graph data modeling, deep fusion of multi-dimensional data is achieved to improve data correlation, and a variational autoencoder is used for feature distribution learning and anomaly repair to effectively improve the stability and robustness of data fusion. The dynamic weight mechanism ensures the real-time and adaptability of feature fusion, thereby improving the accuracy and flexibility of the risk prediction model. Then, a variational recurrent neural network is introduced for time series modeling to effectively enhance the model's modeling ability for complex dynamic features and improve risk prediction accuracy. The long short-term memory network and attention mechanism are combined to further improve the model's expressiveness on long time series and enhance its sensitivity to sudden risks. Finally, a multi-criteria decision analysis framework is introduced to integrate multi-dimensional risk factors to improve the comprehensiveness and sensitivity of the risk early warning mechanism. The hierarchical analysis method and fuzzy reasoning system are combined to realize dynamic weighting of each risk factor to ensure the accuracy of the early warning mechanism in different scenarios, thereby realizing risk fuzzy assessment and dynamic threshold adjustment under complex conditions, significantly reducing false alarm and omission rates, and improving the system's adaptability and early warning accuracy.

[0182] It should be noted that although the steps in the above flowchart are shown in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders.

[0183] In another embodiment, if Figure 2 As shown, the second aspect of the present invention provides an intelligent risk early warning device for a distribution network, comprising:

[0184] The multi-dimensional feature fusion module 10 is used to obtain historical risk monitoring data of the distribution network in multiple dimensions and its corresponding historical risk fault data, and fuse the historical risk monitoring data in multiple dimensions through a graph convolutional network and a variational autoencoder to obtain multi-dimensional historical fusion features;

[0185] An assessment model construction module 20 is configured to train a variational recurrent neural network and a long short-term memory network using the multi-dimensional historical fusion features and the historical risk fault data as training samples, and introduce an attention mechanism into the trained model to generate a risk assessment model;

[0186] A real-time risk assessment module 30 is configured to obtain real-time risk monitoring data of the distribution network under the multiple dimensions and the real-time scenarios to which it belongs, and extract multi-dimensional real-time fusion features corresponding to the real-time risk monitoring data to input into the risk assessment model for processing to obtain a real-time risk assessment level;

[0187] The risk warning trigger module 40 is used to process the multi-dimensional real-time fusion features based on the real-time risk assessment level through a multi-criteria decision analysis method and a fuzzy reasoning system to obtain a target risk assessment level for comparison with the risk warning threshold determined according to the real-time scenario, and trigger an early warning based on the comparison result.

[0188] It should be noted that each module in the above-mentioned intelligent risk warning device for distribution network can be fully or partially implemented by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules. For the specific definition of an intelligent risk warning device for distribution network, please refer to the definition of an intelligent risk warning method for distribution network above. The two have the same functions and effects and will not be repeated here.

[0189] A third aspect of the present invention provides an electronic device, comprising:

[0190] processor, memory, and bus;

[0191] The bus is used to connect the processor and the memory;

[0192] The memory is used to store operation instructions;

[0193] The processor is used to call the operation instruction, and the executable instruction enables the processor to perform operations corresponding to the intelligent risk warning method for distribution network as shown in the first aspect of the present application.

[0194] In an alternative embodiment, an electronic device is provided, such as Figure 3 As shown, Figure 3 The electronic device 5000 shown includes: a processor 5001 and a memory 5003. The processor 5001 and the memory 5003 are connected, for example, via a bus 5002. Optionally, the electronic device 5000 may further include a transceiver 5004. It should be noted that in actual applications, the number of transceivers 5004 is not limited to one, and the structure of the electronic device 5000 does not constitute a limitation on the embodiments of the present application.

[0195] Processor 5001 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, a transistor logic device, a hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 5001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like.

[0196] The bus 5002 may include a path for transmitting information between the above components. The bus 5002 may be a PCI bus or an EISA bus, etc. The bus 5002 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0197] The memory 5003 may be a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, or an EEPROM, a CD-ROM or other optical disk storage, an optical disc storage (including a compact disc, a laser disc, an optical disc, a digital versatile disc, a Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.

[0198] The memory 5003 is used to store application code for executing the solution of the present application, and the execution is controlled by the processor 5001. The processor 5001 is used to execute the application code stored in the memory 5003 to implement the content shown in any of the above method embodiments.

[0199] Among them, electronic devices include but are not limited to: mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc.

[0200] The fourth aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements an intelligent risk warning method for a distribution network shown in the first aspect of the present application.

[0201] Another embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer-readable storage medium is run on a computer, the computer can execute the corresponding contents of the aforementioned method embodiments.

[0202] In addition, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0203] In summary, the present invention relates to the field of data processing technology, and discloses an intelligent risk warning method, device, equipment and medium for distribution networks. The historical risk monitoring data of the distribution network in multiple dimensions are fused through a graph convolutional network and a variational autoencoder to obtain multi-dimensional historical fusion features to train a variational recurrent neural network and a long short-term memory network in combination with the historical risk fault data of the distribution network, and an attention mechanism is introduced during the training process to generate a risk assessment model; real-time risk monitoring data of the distribution network in multiple dimensions are obtained to extract multi-dimensional real-time fusion features and input them into the risk assessment model for processing to obtain a real-time risk assessment level to further process the multi-dimensional real-time fusion features through a multi-criteria decision analysis method and a fuzzy inference system to obtain a target risk assessment level, which is compared with a risk warning threshold. If the threshold is exceeded, an early warning is triggered, effectively improving the accuracy of the risk assessment of the distribution network.

[0204] Each embodiment in this specification is described in a progressive manner, and the parts that are directly the same or similar in each embodiment can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the various technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0205] The above-described embodiments merely represent several preferred implementations of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art could make several improvements and substitutions without departing from the technical principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be based on the scope of protection of the claims.

Claims

1. An intelligent risk early warning method for distribution network, characterized in that: include: Obtain historical risk monitoring data of the distribution network in multiple dimensions and its corresponding historical risk fault data, and fuse the historical risk monitoring data in multiple dimensions through a graph convolutional network and a variational autoencoder to obtain multi-dimensional historical fusion features; Using the multi-dimensional historical fusion features and the historical risk failure data as training samples, training a variational recurrent neural network and a long short-term memory network, and introducing an attention mechanism into the trained model to generate a risk assessment model; Acquire real-time risk monitoring data of the distribution network under the multiple dimensions and the real-time scenarios to which they belong, and extract multi-dimensional real-time fusion features corresponding to the real-time risk monitoring data to input them into the risk assessment model for processing, thereby obtaining a real-time risk assessment level; Based on the real-time risk assessment level, the multi-dimensional real-time fusion features are processed through a multi-criteria decision analysis method and a fuzzy reasoning system to obtain a target risk assessment level for comparison with a risk warning threshold determined according to the real-time scenario, and an early warning is triggered based on the comparison result.

2. The intelligent risk early warning method for distribution network according to claim 1, characterized in that: Historical risk monitoring data in multiple dimensions includes historical equipment status data and historical meteorological environment data; among them, The historical risk monitoring data under multiple dimensions are fused through a graph convolutional network and a variational autoencoder to obtain multi-dimensional historical fusion features, including: Various devices in the distribution network are used as nodes, and based on each of the nodes, the association relationship between the historical device status data and the historical meteorological environment data is used as an edge to construct a heterogeneous graph; Extracting node embedding features of the heterogeneous graph through the graph convolutional network to obtain a historical fusion feature matrix; Based on the maximum likelihood function, the variational autoencoder is used to reduce the dimension of the historical risk monitoring data under the multiple dimensions to obtain a historical monitoring feature dimension reduction matrix; The historical fusion feature matrix and the historical monitoring feature dimensionality reduction matrix are fused through a dynamic weighting mechanism to obtain a multi-dimensional historical fusion feature.

3. The intelligent risk early warning method for distribution network according to claim 1, characterized in that: The historical risk failure data includes historical scenarios of historical risk monitoring data of corresponding dimensions; wherein, Before obtaining the real-time risk monitoring data of the distribution network in the multiple dimensions and the real-time scenarios to which it belongs, the method includes: Based on the multi-task learning method, a multi-task learning network model is constructed, which includes a shared layer and several task-specific output layers, and a loss function is defined for each task; Using the historical scenarios as labels to label the historical risk failure data corresponding to each dimension, and obtaining historical labeled samples as a training set; Using the training set to train the multi-task learning network model, and optimizing each of the loss functions, so as to extract historical basic features common to each scene in the training set through the trained multi-task learning network model; Several preset classification models are trained respectively using the historical basic features as base classifiers, and the Bayesian optimization method is used to tune the hyperparameters of the base classifiers, so as to construct a scene classification model in combination with the multi-task learning network model after training.

4. The intelligent risk early warning method for distribution network according to claim 3, characterized in that: The scene classification model includes a shared layer, several task-specific output layers, a base classifier and an integration layer; wherein, The obtaining of the real-time risk monitoring data of the distribution network in the multiple dimensions and the real-time scenarios to which it belongs includes: Acquire real-time risk monitoring data of the distribution network under the multiple dimensions, and input the real-time risk monitoring data into the shared layer and the plurality of task-specific output layers for processing, thereby obtaining real-time general features and a plurality of preliminary prediction results; Selecting a high-confidence preliminary prediction result from a plurality of the preliminary prediction results and fusing it with the real-time general feature to obtain a real-time basic feature; The real-time basic features are classified by the base classifier to obtain a number of prediction results which are input as new features into the integration layer for processing to obtain the real-time scene.

5. The intelligent risk early warning method for distribution network according to claim 1, characterized in that: The risk assessment model includes an input layer, a VRNN hidden layer, an LSTM layer and an attention layer, wherein the VRNN hidden layer is a variational recurrent neural network with a hidden vector added; wherein, The extracting of multi-dimensional real-time fusion features corresponding to the real-time risk monitoring data and inputting them into the risk assessment model for processing to obtain a real-time risk assessment level includes: The real-time risk monitoring data under multiple dimensions are fused through the graph convolutional network and the variational autoencoder to obtain multi-dimensional real-time fusion features and input them into the VRNN hidden layer through the input layer; An encoder in the variational recurrent neural network is used to quantify the potential state distribution of the multi-dimensional real-time fusion feature at each moment to obtain a plurality of latent vectors. Based on the latent vectors, a decoder in the variational recurrent neural network is used to reconstruct the multi-dimensional real-time fusion feature to generate a plurality of reconstructed features containing temporal dynamic information. Each of the reconstructed features is input into the LSTM layer for processing, and the hidden state at each moment is generated and input into the attention layer for linear transformation. Each transformation result is mapped to a nonlinear space through an activation function, and a dot product operation is performed with the corresponding weight vector to obtain multiple attention scores to quantify the contribution of each of the hidden states to the final risk assessment result. The corresponding hidden states are weightedly summed according to each contribution to obtain the final risk assessment result and its corresponding real-time risk assessment level.

6. The intelligent risk early warning method for distribution network according to claim 1, characterized in that: Based on the real-time risk assessment level, the multi-dimensional real-time fusion features are processed by a multi-criteria decision analysis method and a fuzzy inference system to obtain a target risk assessment level for comparison with a risk warning threshold determined according to the real-time scenario, and an early warning is triggered according to the comparison result, including: Processing the real-time risk assessment level and the multi-dimensional real-time fusion features by the analytic hierarchy process to obtain risk weights of the real-time risk monitoring data of each dimension, and determining risk warning thresholds based on the risk weights and the real-time scenarios; A fuzzy inference system is used to perform a secondary risk assessment on the multi-dimensional real-time fusion features to obtain a target risk assessment level for comparison with the risk warning threshold, and an early warning is triggered based on the comparison result.

7. The intelligent risk early warning method for distribution network according to claim 6, characterized in that: The real-time risk assessment level and the multi-dimensional real-time fusion feature are processed by the analytic hierarchy process to obtain the risk weight of the real-time risk monitoring data of each dimension, and the risk warning threshold is determined based on each risk weight and the real-time scenario, including: Based on the analytic hierarchy process, a target layer is constructed through the real-time risk assessment level, and a criterion layer is constructed using the multi-dimensional real-time fusion feature; Compare each data in the criterion layer and the target layer in pairs to form a pairwise comparison matrix for eigenvalue decomposition, and normalize the decomposition results to obtain the risk weight of the real-time risk monitoring data in each dimension; Based on the real-time scenario, the risk weight increment of the risk monitoring data of each dimension is determined, and based on each risk weight increment, multiple risk threshold adjustment factors are generated to perform weighted summation on each risk weight to obtain a risk warning threshold.

8. An intelligent risk warning device for distribution network, characterized in that: include: A multi-dimensional feature fusion module is used to obtain historical risk monitoring data of the distribution network in multiple dimensions and its corresponding historical risk fault data, and fuse the historical risk monitoring data in multiple dimensions through a graph convolutional network and a variational autoencoder to obtain multi-dimensional historical fusion features; An assessment model construction module is used to train a variational recurrent neural network and a long short-term memory network using the multi-dimensional historical fusion features and the historical risk failure data as training samples, and introduce an attention mechanism into the trained model to generate a risk assessment model; A real-time risk assessment module is used to obtain real-time risk monitoring data of the distribution network under multiple dimensions and the real-time scenarios to which it belongs, and to extract multi-dimensional real-time fusion features corresponding to the real-time risk monitoring data to input them into the risk assessment model for processing to obtain a real-time risk assessment level; The risk warning trigger module is used to process the multi-dimensional real-time fusion features based on the real-time risk assessment level through a multi-criteria decision analysis method and a fuzzy reasoning system to obtain a target risk assessment level for comparison with the risk warning threshold determined according to the real-time scenario, and trigger an early warning based on the comparison result.

9. An electronic device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method implements the intelligent risk early warning method for a distribution network as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the device where the computer-readable storage medium is located executes the computer program, the intelligent risk early warning method for the distribution network according to any one of claims 1 to 7 is implemented.

Images