Power grid fault analysis method and system

Abstract

The invention relates to the technical field of fault diagnosis, in particular to a power grid fault analysis method and system, and the method comprises the steps: constructing a fusion tensor, and carrying out the dimension reduction of the fusion tensor, and generating a fusion feature matrix; constructing a convolutional auto-encoder model to detect an abnormal fusion feature matrix; and performing power grid fault classification according to abnormal data of the abnormal feature matrix, and calculating a power grid equipment fault area based on a power grid fault classification result. The method has the beneficial effects that the power grid operation data is complemented by adopting a high-precision low-rank tensor complementation algorithm, and the completeness and accuracy of the data are ensured. By using Tucker decomposition and matrix compression technologies, key features are extracted from multi-dimensional data collected by a multi-source sensor, and a fusion feature matrix is constructed, so that the data processing efficiency and the feature expression capability are effectively improved. The application of the convolutional auto-encoder model enhances the detection precision of abnormal features, and the method based on improved weighted summation and hierarchical clustering improves the accuracy of fault classification and positioning.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method and system for analyzing power grid faults. Background Art

[0002] With the development of intelligent and digital power grids, power grid fault analysis technology based on multi-dimensional data has gradually received attention. Traditional power grid monitoring methods mainly rely on a single sensor to obtain limited operating data, which is difficult to fully reflect the operating status of the power grid. With the advancement of multi-source sensor technology, the types and dimensions of power grid operation data have increased significantly, covering physical quantities such as voltage, current, temperature, frequency, and network characteristics such as network delay, bandwidth utilization, and packet loss rate. By integrating multi-dimensional data, researchers are able to use data-driven methods to improve the detection and diagnosis accuracy of power grid faults. The machine learning-based fault diagnosis model proposed in recent years uses a large amount of historical data for training, which can achieve rapid identification and classification of abnormal conditions. These technologies have played an important role in improving the safety and stability of the power grid.

[0003] Existing technologies still have many shortcomings when processing multi-dimensional data. For example, the existing tensor completion algorithm has limited application effect on the sparsity problem of multi-dimensional data, and often cannot effectively fill in the missing data, resulting in incomplete and unreliable analysis results. In addition, the analysis of power grid fault characteristics often lacks systematicity, resulting in insufficient classification and positioning accuracy. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a power grid fault analysis method and system, which can solve the problems mentioned in the background technology.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a power grid fault analysis method, comprising constructing a fusion tensor, and performing dimensionality reduction processing on the fusion tensor to generate a fusion feature matrix;

[0009] Construct a convolutional autoencoder model to detect abnormal fusion feature matrix;

[0010] The power grid faults are classified according to the abnormal data of the abnormal feature matrix, the fault area of ​​the power grid equipment is calculated based on the results of the power grid fault classification, and a fault report is generated.

[0011] As a preferred solution of the power grid fault analysis method of the present invention, wherein: constructing the fusion tensor includes:

[0012] Obtain multi-dimensional running data to construct observation tensors;

[0013] Construct a multi-dimensional feature matrix based on the rate of change of the tensor;

[0014] Decompose the multidimensional feature matrix and construct a fused tensor based on the decomposition results.

[0015] As a preferred solution of the power grid fault analysis method of the present invention, wherein: acquiring multi-dimensional operation data to construct an observation tensor includes:

[0016] Perform data cleaning and normalization on the multi-dimensional operation data obtained through multi-source sensors;

[0017] The cleaned and normalized multi-dimensional running data is constructed into observation tensor X according to time series, spatial dimension and feature dimension obs ;

[0018] The stopping threshold is set by the preset criteria and the preset algorithm is used to observe the tensor X obs Complete the tensor X and set the constraint: subject to X Ω =X obs , where X Ω is a subset of the observation position set Ω.

[0019] As a preferred solution of the power grid fault analysis method of the present invention, wherein: decomposing the multidimensional feature matrix, and constructing a fusion tensor based on the decomposition result includes:

[0020] Decompose the multidimensional feature matrix into core tensors and factor matrices through Tucker decomposition;

[0021] The network delay, bandwidth utilization and packet loss rate data in the multi-dimensional operation data of the completed tensor X are synchronously sorted according to the time step to construct the communication network feature matrix C;

[0022] The communication network feature matrix C, the core tensor, and the factor matrix are combined through the tensor product formula to generate a fused tensor.

[0023] As a preferred solution of the power grid fault analysis method of the present invention, generating a fault report includes generating a fault report based on the results of power grid fault classification and power grid equipment fault area.

[0024] As a preferred solution of the power grid fault analysis method of the present invention, wherein: the multi-sensor includes voltage, current, temperature, frequency, bandwidth monitoring, delay monitoring and packet loss monitoring sensors;

[0025] Multi-dimensional operating data includes voltage, current, temperature, frequency, network delay, bandwidth utilization and packet loss rate data.

[0026] As a preferred solution of the power grid fault analysis method of the present invention, wherein: the preset criterion is a relative error stop criterion;

[0027] The default algorithm is a high-precision low-rank tensor completion algorithm.

[0028] In a second aspect, the present invention provides a power grid fault analysis method, comprising: constructing a generation module for constructing a fusion tensor, and performing dimensionality reduction processing on the fusion tensor to generate a fusion feature matrix;

[0029] Build a detection model to construct a convolutional autoencoder model to detect abnormal fusion feature matrix;

[0030] The calculation and generation module is used to classify power grid faults according to the abnormal data of the abnormal feature matrix, calculate the fault area of ​​the power grid equipment based on the results of the power grid fault classification, and generate a fault report.

[0031] In a third aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0032] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when the computer program is executed by a processor.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows: by using a high-precision low-rank tensor completion algorithm to complete the power grid operation data, the integrity and accuracy of the data are ensured. By using Tucker decomposition and matrix compression technology, key features are extracted from multi-dimensional data collected by multi-source sensors, and a fusion feature matrix is ​​constructed, which effectively improves the efficiency of data processing and the ability of feature expression. The application of the convolutional autoencoder model enhances the detection accuracy of abnormal features, while the method based on improved weighted summation and hierarchical clustering improves the accuracy of fault classification and location. Finally, the fault report is displayed in real time through a visual interface, which realizes the rapid identification, precise positioning and intuitive display of power grid faults, and significantly improves the intelligent monitoring level of the power grid and the response speed of fault handling. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0035] Figure 1 The flowchart of the power grid fault analysis method is shown in FIG.

[0036] Figure 2 A schematic diagram of the internal structure of a computer device. DETAILED DESCRIPTION

[0037] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0038] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0039] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0040] Example 1

[0041] Reference Figure 1 , which is the first embodiment of the present invention, and provides a power grid fault analysis method, which includes:

[0042] S1. Construct a fusion tensor and perform dimension reduction processing on the fusion tensor to generate a fusion feature matrix.

[0043] Furthermore, constructing the fused tensor includes,

[0044] Get multi-dimensional running data to build observation tensors.

[0045] Furthermore, obtaining multi-dimensional running data to construct observation tensors includes:

[0046] Perform data cleaning and normalization on the multi-dimensional operation data obtained through multi-source sensors;

[0047] The cleaned and normalized multi-dimensional running data is constructed into observation tensor X according to time series, spatial dimension and feature dimension obs ;

[0048] It should be noted that X obs ∈R I×J×T , where I is the number of sensor types, J is the number of spatial nodes, indicating the physical location of the sensor, and T is the time length, indicating the time series information of the data. Each observation is the measurement value of the i-th sensor at the j-th node and time t, and R is a real number set.

[0049] The stopping threshold is set by the preset criteria and the preset algorithm is used to observe the tensor X obs Complete the tensor X and set the constraint: subject to X Ω =X obs , where X Ω is a subset of the observation position set Ω.

[0050] Furthermore, the preset criterion is a relative error stopping criterion;

[0051] The default algorithm is a high-precision low-rank tensor completion algorithm.

[0052] It should be noted that the stopping threshold is set by the relative error stopping criterion, and the high-precision low-rank tensor completion algorithm is used to complete the positions of the observed tensor Xobs containing some missing values ​​to obtain the completed tensor X. The formula is:

[0053]

[0054] Where N is the number of modes of the tensor, corresponding to sensor type, spatial node and time dimension, respectively, (X) n is the expansion matrix of tensor X in the nth mode, ||(X) n || * is the nuclear norm, which represents the sum of the singular values ​​of the n-th mode expansion matrix and is used to maintain the low rank of the tensor. n is the weight hyperparameter of the expanded matrix in the nth mode, which controls the degree of low rank of the expanded matrix in each mode. X is the completed tensor. α is the regularization coefficient, which controls the strength of the retained constraint. is the matrix two norm, which represents the difference between the completed tensor X and the observed tensor X obs The difference in the observed position, Ω is the set of observed positions, x i，j，t is the tensor value of the i-th sensor at the j-th node and time t, is the measurement value of the i-th sensor at the j-th node and time t.

[0055] It should be further explained that using the observation tensor X obs Construct the initial tensor. When the observation position belongs to the observation position set, keep the measurement value of the i-th sensor at the j-th node and time t unchanged. When the missing position does not belong to the measurement position set, set the initial value to zero. Expand the initialized tensor into a matrix form along different modes n to obtain the expansion matrix of each mode. The purpose of tensor expansion is to convert the high-dimensional tensor into a matrix representation for subsequent calculation of the nuclear norm to perform low-rank constraints. In order to achieve high-precision low-rank tensor completion, the alternating direction optimization method is used in each iteration to update the expansion matrix of each mode and the elements at the missing position. Apply singular value decomposition to the expansion matrix of each mode into an orthogonal matrix and a diagonal matrix. Apply soft threshold operation to update the singular value according to the nuclear norm term in the objective function. Use the updated singular value matrix to reconstruct the expansion matrix. Keep the value of the observation tensor on the set of observation positions, that is, satisfy the constraint conditions. Re-estimate the position of the missing value according to the update result of the expansion matrix of the current mode at the missing position, so that the overall error of the tensor is minimized. Calculate the error value between the current iteration and the previous iteration, and judge the change of the objective function. If the change of the objective function is less than the set stop threshold, stop the iteration and get the final completed tensor.

[0056] It is better to set the constraint: subject to X Ω =X obs , where X Ω is a subset of the observation position set Ω, which is the observation tensor X obs A subset selection is used to ensure that the completion result is consistent with the known data at the observed position by selecting only the values ​​at the observed position. This selection and constraint setting help the algorithm focus on completing the missing values ​​without changing the observed values, ensuring that the completed tensor meets the low-rank characteristics and maintains the consistency of the observed data;

[0057] By keeping the observed data value at the observed position, ensuring that the completion result is consistent with the actual observed value, and preventing unnecessary changes to the known data, the authenticity of the completed data is guaranteed. This constraint condition limits the adjustment range of the optimization algorithm, so that the optimization process focuses on the missing data area, thereby improving the completion efficiency. The constraint condition effectively prevents excessive adjustment on the observed data and avoids the algorithm from overfitting the observed data, thereby ensuring that the completed tensor has good generalization ability;

[0058] The low-rank constraint of the tensor makes the completed tensor structure maintain the compactness and intrinsic correlation of the multi-dimensional data of the power grid and communication network. There is usually a strong correlation between the multimodal features of the power grid and communication data. For example, the voltage and current data of a specific spatial node in different time periods are highly correlated. These correlations can be retained through low-rank constraints. Low-rank constraints can also effectively remove noise. Power grid and communication data often contain noise. The low-rank structure helps to filter out irrelevant noise data and maintain the integrity of key fault features. Compared with the direct interpolation method, the low-rank constraint can avoid local data deviation. Other methods cannot fully capture the global characteristics of multi-dimensional data, which can easily lead to data inconsistency or over-smoothing when completing high-dimensional data. In power grid data, fault features often only appear in some observation locations and manifest as abnormal voltage and current fluctuations. In order not to lose these key fault information, the retention constraint terms must be used to ensure that the completion results are completely consistent with the observed data at known locations. The power grid observation data may contain key indicators of electrical equipment failures, such as voltage mutations, frequency fluctuations, etc. This information is an important basis for operation and maintenance analysis. If the retention constraint terms are not used in the completion process, the completion The full result may deviate from the observed data, resulting in distortion or loss of fault features. The retention constraint term can strictly maintain the authenticity of the observed data, so that the completion result accurately matches the original data at the known data points without introducing deviations. In contrast, if only relying on low-rank constraints or other methods (such as least squares method) may cause the completion result to deviate from the real data at the known data points and lose accuracy. Traditional low-rank tensor completion only focuses on low-rank constraints and ignores the retention of observed data, resulting in deviations at known data positions. This scheme adds retention constraints to ensure that while retaining the authenticity of the observed data, low-rank completion of the data is achieved. The characteristics of power grid and communication network data are complex and multimodal. The improved optimization objective function can better adapt to the characteristics of power grid fault data through the low-rank constraints of the multi-mode expansion matrix and the retention of observed data, avoiding the loss of key fault features due to smoothing during the completion process. Compared with other tensor completion methods (such as weighted interpolation method and mean filling method), this method can ensure the accuracy and reliability of the completed data, thereby providing higher-precision basic data support in power grid fault analysis, so that the analysis model can more accurately identify and locate faults.

[0059] Furthermore, the multi-sensor includes voltage, current, temperature, frequency, bandwidth monitoring, delay monitoring, and packet loss monitoring sensors;

[0060] Multi-dimensional operating data includes voltage, current, temperature, frequency, network delay, bandwidth utilization and packet loss rate data.

[0061] It should be noted that the use of multi-source sensors for data collection can comprehensively monitor various operating parameters of the power grid, covering important indicators such as voltage, current, and temperature. This diversity helps to better understand the overall state of the power grid, thereby improving the accuracy of fault detection. Data cleaning and normalization are key steps to improve data quality. By cleaning the data, noise and inconsistency can be eliminated to ensure the reliability of the basic data for analysis. This process not only improves the integrity of the data, but also provides clean input for subsequent analysis. Normalization can make data with different characteristics at the same order of magnitude, thereby eliminating the adverse effects caused by dimensional differences and making model training more efficient. The construction of observation tensors allows multi-dimensional data to be organized in a more efficient form. It is convenient for subsequent analysis and calculation. Through the combination of time series, spatial dimensions and feature dimensions, it can comprehensively reflect the operating status of the power grid and provide rich information for fault detection. The application of high-precision low-rank tensor completion algorithm greatly enhances the system's ability to process missing data. Through this algorithm, the missing values ​​in the observation tensor can be effectively completed to ensure the integrity and accuracy of the data. This completion process not only improves the availability of data, but also optimizes the prediction performance of subsequent models. The use of relative error stopping criteria is set to ensure the efficiency of the completion algorithm. By dynamically adjusting the stopping threshold, the calculation time can be reduced while ensuring the completion accuracy, and the overall processing efficiency can be improved, so that the system can respond quickly and adapt to different operating environments in practical applications.

[0062] Furthermore, a multi-dimensional feature matrix is ​​constructed according to the tensor change rate.

[0063] It should be noted that the finite difference method is used to calculate the time and space change rates R of the completed tensor X respectively. T , R S ,formula:

[0064]

[0065] In the formula, Δt is the time interval, Δ is the distance between adjacent nodes, and X i,j,t is the measurement value of the tensor X at the jth node and time t of the i-th sensor;

[0066] Use the feature enhancement matrix construction method to calculate the time and space change rate R of the completed tensor X T , R S Construct a multidimensional feature matrix M.

[0067] Furthermore, the multi-dimensional feature matrix is ​​decomposed, and a fusion tensor is constructed based on the decomposition results.

[0068] Furthermore, the multi-dimensional feature matrix is ​​decomposed, and the fusion tensor is constructed based on the decomposition results, including:

[0069] Decompose the multidimensional feature matrix into core tensors and factor matrices through Tucker decomposition;

[0070] The network delay, bandwidth utilization and packet loss rate data in the multi-dimensional operation data of the completed tensor X are synchronously sorted according to the time step to construct the communication network feature matrix C;

[0071] The communication network feature matrix C, the core tensor, and the factor matrix are combined through the tensor product formula to generate a fused tensor.

[0072] It should be noted that the core tensor is tensor-producted with the factor matrix to obtain a new intermediate tensor, and this intermediate tensor is tensor-producted with another factor matrix to continue merging features, and the merged features are combined with the last factor matrix to complete the tensor product of all modes to obtain the fused tensor D, which contains all the important information of the power grid feature matrix and the communication network feature matrix.

[0073] It should be further explained that the fusion tensor D is reduced in dimension using matrix compression technology to generate a fusion feature matrix L.

[0074] Preferably, by constructing a multidimensional feature matrix based on the rate of change of the tensor, an efficient method is provided for power grid fault analysis. Calculating the time and space change rate of the completed tensor helps to dynamically monitor the operation status of the power grid. This process can capture the changes in the power grid in a timely manner and provide accurate fault warning information. Through this dynamic monitoring, the power system can respond to potential faults more quickly, reduce the risk of power outages, and improve the overall reliability of the power grid. Using Tucker decomposition to analyze the multidimensional feature matrix can effectively extract the core features in the data. The process simplifies complex high-dimensional data into core tensors and factor matrices, making subsequent data processing more efficient. In the power system, the key factors affecting the stability of the power grid can be quickly identified to provide a basis for decision-making. The construction of the fusion tensor enhances the system's comprehensive understanding of the fault by integrating data from different sources, and combines communication characteristics such as network delay, broadband utilization and packet loss rate with power grid operation data. The combination of matrix compression technology and time-domain matrix data can more accurately assess the operating status of the power grid and its mutual influence with the communication network. This comprehensive analysis method helps to reveal the potential causes of faults, thereby achieving more accurate fault classification and location. The application of matrix compression technology further improves the efficiency of analysis. Through dimensionality reduction processing, the complexity of the data is reduced, making subsequent model training and fault detection more efficient. The amount of data in the power grid system is usually huge. After using matrix compression technology, it can reduce the demand for storage and computing resources while retaining important information. This is particularly important for real-time monitoring and fault response. The feature enhancement matrix construction method uses information on the rate of change of time and space to significantly improve the expressive power of the feature matrix. This method enhances the sensitivity to small changes and makes fault detection more forward-looking. This is of great significance for quickly identifying potential fault hazards during power grid operation, and helps to take preventive measures in advance to reduce the probability of accidents.

[0075] S2. Construct a convolutional autoencoder model to detect abnormal fusion feature matrix.

[0076] It should be noted that constructing a convolutional autoencoder model to detect abnormal fusion feature matrix refers to obtaining multi-dimensional running data with normal labels from scientific data sets for preprocessing to generate model training sets;

[0077] Build a convolutional autoencoder model, including an encoder and a decoder;

[0078] Set the model input format to fusion feature matrix;

[0079] Use the model training set to train the convolutional autoencoder model, and use the loss function and Adam optimizer to iteratively optimize the model parameters;

[0080] Input the fused feature matrix L into the trained convolutional autoencoder model to obtain the reconstructed feature matrix, and use the loss function to calculate the error feature matrix between the fused feature matrix and the reconstructed feature matrix;

[0081] The reconstruction error threshold is set based on the historical multi-dimensional operation data, the error feature matrix is ​​compared with the reconstruction error threshold, the element values ​​in the error feature matrix are screened out, and the abnormal feature matrix is ​​generated. If the element value is greater than the reconstruction error threshold, 1 is marked at the corresponding position of the abnormal feature matrix, otherwise it is marked with 0. The element value refers to the absolute value of the difference between each element of the input fusion feature matrix and the reconstruction feature matrix.

[0082] Preferably, by constructing a convolutional autoencoder model and using normal labels of multi-dimensional operating data for training, the accuracy and sensitivity of power grid fault detection are greatly improved. By collecting data with normal labels from scientific data sets for preprocessing to form a model training set, this process ensures that the features learned by the model can represent the normal operating mode. This data-driven approach enables the model to quickly adapt and effectively identify potential anomalies when faced with actual power grid operation conditions. The design of the convolutional autoencoder includes an encoder and a decoder. The encoder extracts data features through a convolutional layer, reducing the dimension while retaining key information, and the decoder is responsible for reconstructing the input data. This structure effectively captures the spatial relationship in the data and is particularly suitable for processing multi-dimensional power grid operation data. During the training process, the model parameters are iteratively optimized using the loss function and the Adam optimizer, which can quickly converge to the optimal solution, thereby improving the performance of the model. In order to improve the performance and robustness of the model, the fused feature matrix is ​​input into the trained convolutional autoencoder model to generate a reconstructed feature matrix. The error between the original data and the reconstructed data is then calculated through the loss function. This process can accurately reveal the limitations of model reconstruction and help identify feature variations that should not exist under normal operating conditions. By setting the reconstruction error threshold, the normal and abnormal feature matrices can be clearly distinguished, thereby generating an abnormal feature matrix. This strategy makes fault detection not only rely on historical data, but also dynamically adapt to new operating conditions, thereby improving the timeliness and accuracy of detection. In power grid management, the generation of abnormal feature matrices is an important basis for fault classification and location. By analyzing the abnormal features, potential fault areas and their causes can be quickly identified, thereby helping operation and maintenance personnel to take corresponding measures. This method not only improves the safety and stability of the power grid, but also lays the foundation for the intelligent management of the power system.

[0083] S3. Classify the power grid fault according to the abnormal data of the abnormal feature matrix, calculate the fault area of ​​the power grid equipment based on the result of the power grid fault classification, and generate a fault report.

[0084] It should be noted that the grid fault classification is performed according to the abnormal data of the abnormal feature matrix, and the calculation of the grid equipment fault area based on the result of the grid fault classification refers to scanning the position marked as 1 in the abnormal feature matrix and extracting the corresponding abnormal data from the fused feature matrix L;

[0085] Collect and preprocess the multi-dimensional operation data of historical normal labels, use the statistical control limit method to set the normal range for the multi-dimensional operation data of historical normal labels after preprocessing, compare the extracted abnormal data with the normal range, and retain the abnormal data that is larger than the normal range;

[0086] Normalize the retained abnormal data and use K-Means clustering to classify the fault types of the normalized abnormal data;

[0087] The locations of power grid equipment in the normalized abnormal data are taken as nodes. According to the actual physical connection relationship of the power grid, a topological structure is established and the adjacency matrix A is defined. A∈R N×N , where N is the total number of device nodes in the power grid, and the element A of the adjacency matrix ij , indicates whether node i and node j are directly connected, R is a real number set;

[0088] Use z-score to calculate the outlier value f of node i in the kth type of power grid fault ik ;

[0089] The weighted summation formula is:

[0090]

[0091] In the formula, S 0,i is the weighted sum to calculate the abnormal score of node i, ω k is the weight of the kth type of power grid fault, K is the total number of power grid fault categories, and f ik is the abnormal value of node i in the kth type of power grid fault;

[0092] The anomaly score S of node i is calculated using the improved weighted sum i , the formula is:

[0093]

[0094] Where K is the total number of grid fault categories, g(f ik ) is a nonlinear function, B is the total number of nodes in the power grid, h(f ik ) is a square root function, which is used to limit the excessive amplification of the abnormal values ​​of the adjacent nodes on the current node, and k is the index of the fault type, which is used to distinguish different fault types;

[0095] Preferably, there are some isolated nodes in the power grid. If the autocorrelation term is not added, when the node has no adjacent nodes, the adjacency term in the denominator will become invalid and cannot reasonably reflect the node's abnormal score. The autocorrelation term ensures that even in the absence of adjacent nodes, each node can calculate the abnormal score based on its own abnormality, thereby ensuring the universality and stability of the formula. The adjacency matrix reflects the connection relationship between nodes, so that the formula can spatially reflect the abnormal interaction between nodes according to the power grid topology. Using the adjacency matrix to extend the outliers to the adjacent areas of the nodes can effectively detect regional anomalies and improve the positioning accuracy of power grid faults. If the adjacency matrix is ​​not used and only the outliers of the node itself are relied on, regional faults cannot be accurately captured, and the detection ability of fault propagation and regional interaction is lost. The outliers may change dramatically, and a single linear operation is difficult to balance the impact of various abnormal features on the node score. The nonlinear function not only amplifies the impact of high outliers, but also controls the impact of outliers on the total score. The response amplitude is increased to maintain the stability of the model. The use of these adaptive functions can more flexibly adapt to the distribution of different types of abnormal data in the power grid and avoid calculation imbalance caused by extreme outliers. Traditional weighted summation or linear methods are usually unable to effectively handle isolated nodes because they often rely on the outliers of adjacent nodes for calculation. The improved formula adds node autocorrelation terms in the denominator so that the formula can still run normally without adjacent nodes. It is applicable to a variety of power grid topologies and improves the universality of calculation. Compared with the existing technology, the formula uses the adjacency matrix to incorporate the outliers of adjacent nodes into the calculation, which can reflect the spatial correlation of power grid nodes. In regional faults, the outliers of adjacent nodes will affect the scores of target nodes, thereby improving the detection ability of local anomalies. Traditional methods often ignore the influence of adjacent nodes and find it difficult to capture regional fault characteristics. By smoothing and controlling the influence of extreme outliers, the calculation stability is guaranteed and the sensitivity to the anomaly score is improved.

[0096] It should be further explained that the nonlinear function g(f ik ), the formula is:

[0097]

[0098] Square root function h(f ik ), the formula is:

[0099]

[0100] Use the percentile method to set the anomaly score threshold and compare the anomaly score threshold with the anomaly score S of node i. i For comparison, retain the abnormal score S of node i i The nodes whose scores are greater than the anomaly score threshold are marked as faulty nodes.

[0101] Use the hierarchical clustering method to set the maximum path distance threshold, and find the connection between the faulty nodes through the adjacency matrix. If A is satisfied, ik =1, it is judged to be the same fault area. If A ik =0, the Floyd-Warshall algorithm is used to calculate the shortest path distance between faulty nodes, and the shortest path distance is compared with the maximum path distance threshold. If the shortest path distance is less than or equal to the maximum path distance threshold, it is set as the same fault area; if the shortest path distance is greater than the maximum path distance threshold, it is set as an isolated fault area.

[0102] By comparing the abnormal feature matrix with historical normal data, an efficient means of fault location and analysis is provided. The extracted abnormal feature matrix identifies potential fault points. Through the position marked as 1, the abnormal data in the fused feature matrix can be quickly located, laying the foundation for subsequent fault analysis. When preprocessing the multi-dimensional operating data of historical normal labels, the statistical control limit method is used to set the normal range to ensure that the data used reflects the actual operating status. This process not only improves the accuracy of abnormal data screening, but also provides a reliable basis for subsequent analysis. After normalizing the retained abnormal data, the K-Means clustering algorithm is used to classify the fault type. This method enables different fault types to be accurately identified, which is convenient for subsequent analysis and response. Normalization helps to eliminate the impact of different dimensions and makes the comparison between data more reasonable. By constructing the topological structure of the power grid and defining the adjacency matrix A, the connection relationship between devices can be clearly reflected. This structure It provides an important basis for the subsequent fault area judgment, so that the fault possibility of each node can be effectively evaluated through the node anomaly score. The z-score is used to calculate the outlier value of the node under a specific fault type. Combined with the improved weighted summation method, a more accurate anomaly score can be obtained. This score not only comprehensively considers the abnormal situation of the node itself, but also moderately considers the influence of adjacent nodes, effectively reducing the excessive amplification effect of the outlier value. After setting the anomaly score threshold, the faulty node can be quickly identified by comparing it with the node's anomaly score. This mechanism ensures that the system can respond and maintain in time when potential faults are discovered, reducing the impact of faults on the power grid. Through the hierarchical clustering method and the Floyd-Warshall algorithm combined with the adjacency matrix, the connection relationship between faulty nodes can be determined, and the same fault area can be clarified. This process makes fault location more accurate, especially in a complex power grid environment, which helps operation and maintenance personnel to take measures quickly to prevent the expansion of faults.

[0103] Furthermore, generating a fault report includes generating a fault report based on the results of grid fault classification and grid equipment fault area, wherein the fault report can be displayed by constructing a visual interface.

[0104] It should be noted that generating a fault report based on the results of power grid fault classification and power grid equipment fault area includes obtaining the fault type of each node from the classification of node anomaly score and fault type, sorting the fault nodes in the fault area from large to small, setting the maximum anomaly score of the fault node as the fault center node, generating a fault report, and converting the fault report into PDF format.

[0105] By obtaining the fault type of each node from the classification of node anomaly scores and fault types, the health status of power grid equipment can be systematically evaluated. This process enables operation and maintenance personnel to accurately identify which equipment is at risk of failure, thereby providing a scientific basis for subsequent maintenance decisions. After obtaining the fault type of each node, sorting the faulty nodes in the faulty area from large to small helps to clearly display the severity of the fault. This sorting not only makes the faulty nodes clear at a glance, but also helps operation and maintenance personnel quickly find the nodes that need to be processed first. Setting the maximum anomaly score of the faulty node as the fault center node can effectively focus on the most critical equipment and ensure the reasonable allocation and efficient use of resources. This method helps to reduce the response time of faults, improve the overall stability of the power grid, and generate The fault reporting process integrates all fault information in a structured manner, which is convenient for subsequent analysis and archiving. Through reports containing key information such as fault type, fault node, anomaly score, etc., operation and maintenance personnel can have a more comprehensive understanding of the current operating status of the power grid. This report not only provides support for decision-making, but also serves as a reference for daily operation and maintenance, helping to formulate targeted maintenance plans. Converting fault reports into PDF format ensures the readability and ease of sharing of the report. Documents in PDF format can be easily viewed on different devices and can maintain the original format, avoiding information loss due to software differences. This shareable feature enables fault information to be quickly transmitted to managers and technical teams at all levels, promoting cross-departmental collaboration and improving response speed.

[0106] Further, building a visualization interface to display the fault report means using a visualization tool Python's Dash framework to build a visualization interface, using a data visualization tool to display the fault report in real time, and using a map visualization tool to display the fault area;

[0107] Users who have passed real-name verification are allowed to view the information.

[0108] The visualization interface built with the Dash framework can quickly respond to user input and provide an intuitive interactive experience. Developers can combine multiple chart types, such as line charts, bar charts, and heat maps, to fully display fault data. This not only improves user operability, but also enhances the comprehensibility of data, allowing technicians and management to quickly grasp the situation of power grid faults. Using data visualization tools to display fault reports in real time can ensure that users obtain the latest information as soon as the fault occurs. This feature is crucial for the timely response of the power system, which can reduce delays in fault handling and improve the reliability and safety of the power grid. Displaying the fault area through map visualization tools can enable users to intuitively understand the geographical distribution of faults and their impact on the power grid. This spatial analysis can help the operation and maintenance team quickly locate the problem area, effectively guide on-site inspection and maintenance, and reduce the time and cost of fault recovery. Through real-name verification, the system can ensure that only authorized users can view sensitive fault data and reports. This measure not only protects the company's trade secrets, but also increases the security of data access, ensuring that the power company's operating information is not abused or leaked.

[0109] In summary, the beneficial effect of a power grid fault analysis method of the present invention is to complete the power grid operation data by adopting a high-precision low-rank tensor completion algorithm, thereby ensuring the integrity and accuracy of the data. By using Tucker decomposition and matrix compression technology, key features are extracted from multi-dimensional data collected by multi-source sensors, and a fusion feature matrix is ​​constructed, which effectively improves the efficiency of data processing and the ability of feature expression. The application of the convolutional autoencoder model enhances the detection accuracy of abnormal features, while the method based on improved weighted summation and hierarchical clustering improves the accuracy of fault classification and location. Finally, by displaying the fault report in real time through a visual interface, the technical solution realizes the rapid identification, precise positioning and intuitive display of power grid faults, significantly improving the intelligent monitoring level of the power grid and the response speed of fault handling.

[0110] Example 2

[0111] This embodiment provides a power grid fault analysis system, which includes a construction generation module, which is used to construct a fusion tensor and perform dimensionality reduction processing on the fusion tensor to generate a fusion feature matrix;

[0112] Build a detection model to construct a convolutional autoencoder model to detect abnormal fusion feature matrix;

[0113] The calculation and generation module is used to classify power grid faults according to the abnormal data of the abnormal feature matrix, calculate the fault area of ​​the power grid equipment based on the results of the power grid fault classification, and generate a fault report.

[0114] The above-mentioned unit modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above-mentioned modules.

[0115] Example 3

[0116] This embodiment provides a computer device, which may be a terminal, and its internal structure diagram may be as follows: Figure 2 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a power grid fault analysis method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0117] The present embodiment also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the following steps are implemented: constructing a fused tensor, and performing dimensionality reduction processing on the fused tensor to generate a fused feature matrix; constructing a convolutional autoencoder model to detect an abnormal fused feature matrix; classifying power grid faults according to abnormal data of the abnormal feature matrix, calculating the fault area of ​​the power grid equipment based on the result of the power grid fault classification, and generating a fault report.

[0118] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for analyzing power grid faults, characterized in that: include, Constructing a fusion tensor, and performing dimensionality reduction processing on the fusion tensor to generate a fusion feature matrix; Construct a convolutional autoencoder model to detect abnormal fusion feature matrix; The power grid fault is classified according to the abnormal data of the abnormal feature matrix, the power grid equipment fault area is calculated based on the result of the power grid fault classification, and a fault report is generated.

2. The power grid fault analysis method according to claim 1, characterized in that: The construction of the fused tensor includes: Obtain multi-dimensional running data to construct observation tensors; Construct a multi-dimensional feature matrix based on the rate of change of the tensor; The multidimensional feature matrix is ​​decomposed, and the fusion tensor is constructed based on the decomposition result.

3. The power grid fault analysis method according to claim 2, characterized in that: The obtaining of multi-dimensional operation data to construct an observation tensor includes: Perform data cleaning and normalization on the multi-dimensional operation data obtained through multi-source sensors; The cleaned and normalized multi-dimensional running data is constructed into observation tensor X according to time series, spatial dimension and feature dimension obs ; The stopping threshold is set by a preset criterion and the preset algorithm is used to calculate the observed tensor X obs Complete the tensor X and set the constraint: subject to X Ω =X obs , where X Ω is a subset of the observation position set Ω.

4. The power grid fault analysis method according to claim 3, characterized in that: Decomposing the multidimensional feature matrix and constructing the fusion tensor based on the decomposition result includes: Decomposing the multidimensional feature matrix into a core tensor and a factor matrix by Tucker decomposition; The network delay, bandwidth utilization and packet loss rate data in the multi-dimensional operation data of the completed tensor X are synchronously sorted according to the time step to construct the communication network feature matrix C; The communication network feature matrix C, the core tensor and the factor matrix are combined through a tensor product formula to generate the fused tensor.

5. The power grid fault analysis method according to claim 4, characterized in that: The generating of the fault report includes generating the fault report based on the results of the grid fault classification and the grid equipment fault area.

6. The power grid fault analysis method according to claim 5, characterized in that: The multi-sensor includes voltage, current, temperature, frequency, bandwidth monitoring, delay monitoring and packet loss monitoring sensors; The multi-dimensional operation data includes voltage, current, temperature, frequency, network delay, bandwidth utilization and packet loss rate data.

7. The power grid fault analysis method according to any one of claims 3 to 6, characterized in that: The preset criterion is a relative error stopping criterion; The preset algorithm is a high-precision low-rank tensor completion algorithm.

8. A power grid fault analysis system, characterized in that: include: Constructing a generation module, used for constructing a fusion tensor, and performing dimensionality reduction processing on the fusion tensor to generate a fusion feature matrix; Build a detection model to construct a convolutional autoencoder model to detect abnormal fusion feature matrix; The calculation and generation module is used to classify power grid faults according to the abnormal data of the abnormal feature matrix, calculate the fault area of ​​power grid equipment based on the result of the power grid fault classification, and generate a fault report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images