Digitalized GIS equipment fault intelligent monitoring management system

Abstract

The invention discloses a digital GIS equipment fault intelligent monitoring management system, and relates to the technical field of fault intelligent monitoring, and the system comprises an information retrieval module which takes attribute information as a constraint, and carries out the networking retrieval to obtain a sample fault event set; the event clustering module is used for obtaining a plurality of high-frequency fault event sets; the traceability analysis module is used for determining a high-frequency fault factor set and generating a high-frequency fault feature matrix; the fixed-point monitoring module is used for obtaining a real-time monitoring data set and constructing a monitoring data matrix; the similarity analysis module is used for performing similarity analysis on the high-frequency fault feature matrix and the monitoring data matrix to determine comprehensive similarity; and the fault early warning module is used for performing fault early warning on the target equipment. Through the method and the device, the technical problem that real-time and accurate monitoring and management are difficult due to the fact that the intelligent degree and the detection accuracy are insufficient in the prior art can be solved, and the technical effect of improving the automation degree and the detection accuracy is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of fault intelligent monitoring, and in particular to a digital GIS equipment fault intelligent monitoring and management system. Background Art

[0002] The main goal of GIS equipment fault monitoring is to monitor the operating status of GIS equipment in real time and accurately, prevent, discover and handle equipment failures in a timely manner, and ensure the safe and reliable operation of the power system.

[0003] At present, in the traditional monitoring and management process, manual inspections and interventions account for a large proportion, resulting in relatively low efficiency in fault detection and handling. For example, when a GIS device fails, it is often necessary to determine the fault type and location through manual inspections and manual analysis, which takes a lot of time and human resources. The traditional monitoring and management system has a low degree of automation in data collection, processing and analysis. It is difficult to monitor the operating status of GIS equipment in real time and accurately, and it is impossible to quickly identify and handle potential fault risks. GIS equipment is usually installed in complex environments, such as underground substations. The environment may have multiple factors such as temperature, humidity, and electromagnetic interference, which will affect the accuracy and reliability of the sensor, resulting in reduced detection accuracy.

[0004] In summary, the existing technology has the technical problem that it is difficult to monitor and manage GIS equipment fault monitoring and management in real time and accurately due to the obvious deficiencies in the intelligence level and detection accuracy. Summary of the invention

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

[0006] Therefore, the present invention provides a digital GIS equipment fault intelligent monitoring and management system to solve the technical problem in the prior art that GIS equipment fault monitoring and management has obvious deficiencies in intelligence and detection accuracy, resulting in difficulty in real-time and accurate monitoring and management.

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

[0008] In the first aspect, an embodiment of the present invention provides a digital GIS equipment fault intelligent monitoring and management system, which includes an information retrieval module for obtaining attribute information of a target GIS device, and using the attribute information as a constraint to retrieve a sample fault event set through an online search; an event clustering module for processing the sample fault event set using a CNN model and a clustering algorithm to obtain multiple high-frequency fault event sets; a source tracing analysis module for performing source tracing analysis on multiple high-frequency fault event sets in turn based on the accident tree principle, determining a high-frequency fault factor set, and generating a high-frequency fault feature matrix; a fixed-point monitoring module for configuring a multi-source sensor monitoring array based on the high-frequency fault factor set, and performing fixed-point monitoring of the target GIS device using the multi-source sensor monitoring array to obtain a real-time monitoring data set and construct a monitoring data matrix; a similarity analysis module for performing similarity analysis on the high-frequency fault feature matrix and the monitoring data matrix according to a fusion similarity comparison strategy to determine a comprehensive similarity; and a fault warning module for performing a fault warning on the target GIS device if the comprehensive similarity meets a preset similarity threshold.

[0009] As a preferred solution of the multi-source heterogeneous data synchronization and reconstruction system of the new energy station of the present invention, the information retrieval module is also used for:

[0010] Get detailed attribute information from GIS equipment;

[0011] Use attribute information as query criteria to search historical data in the networked fault event database.

[0012] As a preferred solution of the multi-source heterogeneous data synchronization and reconstruction system of the new energy station of the present invention, the event clustering module is also used for:

[0013] Based on the attribute information and historical data of the fault event, the convolutional neural network (CNN) model is selected for fine-tuning to extract the deep feature vector of the fault event;

[0014] The convolutional layer of the convolutional neural network CNN model introduces a depth-separable convolution, and the depth convolution expression is:

[0015]

[0016] in, represents the feature map of the deep convolution output, k is the channel index, i and j represent the row and column coordinates on the feature map, F is the size of the convolution kernel, m, n, represent the row index and column index of the filter inner loop, respectively. is the weight of the kth channel of the depth convolution kernel at position m, n, is the value of the kth channel of the input image at position (i+m, j+n), B k is the bias term;

[0017] The point-by-point convolution expression is:

[0018]

[0019] in, represents the final output feature map of point-by-point convolution, c represents the channel index of the input image,

[0020] Represents the value of the cth channel at position (i, j) in the output feature map of the deep convolutional layer, G c，k is the weight matrix of point-wise convolution, D k is the bias term;

[0021] For the activation layer of the convolutional neural network CNN model, the LeakyReLU function is selected to apply nonlinear changes to the output of the convolutional layer. The expression is:

[0022]

[0023] in, Represents the value at position (i, j) on the kth channel in the output feature map of the activation layer, and α represents a small positive number. Represents the k-th channel in the convolutional layer output feature map, position (i,

[0024] j) as the input of the activation function;

[0025] The FractionalMaxPooling algorithm is used in the pooling layer of the convolutional neural network CNN model, and the expression is:

[0026]

[0027] Among them, p k i，j represents the value at position (i, j) on the kth channel of the feature map, N is the total number of elements in the pooling area, A i k Represents the position of the original feature map A and the value on channel k. A is the feature map after convolution and activation. s is the distance the pooling window moves on the feature map. r m and c m are randomly selected row and column indices, and s is the step size;

[0028] Using the Dropout regularization technique in the fully connected layer of the convolutional neural network CNN model, the expression is:

[0029] FC l =(W l ·Mk)·Fl n(P)+B l ;

[0030] Among them, FC l represents the output of the lth fully connected layer, l represents the hierarchical number of the current layer in the neural network, and W l represents the weight matrix of the lth fully connected layer, Mk represents the mask matrix used in the Dropout operation, and Fl n (P) represents the process of converting P into a one-dimensional vector. P refers to the output of the previous layer. B l Represents the bias vector of the lth fully connected layer;

[0031] The output layer of the convolutional neural network CNN model is normalized using the Softmax function, and the expression is:

[0032]

[0033] Among them, Y j is the output layer, representing the probability of the jth category, FC L j represents the output of the last layer of the neural network, j represents the index of the category in the output layer of the neural network, L represents the number of layers of the neural network, and T represents the temperature parameter. Represents the exponential operation FC L j The value after dividing by the temperature parameter T is exponentially expressed, and e represents the base of the natural logarithm. Represents the exponential operation FC L k Divide the value by the temperature parameter T for exponentialization;

[0034] Calculate the failure ratio of each sample failure event subset to obtain multiple failure ratios;

[0035] A subset of sample fault events whose fault proportion is less than a preset proportion threshold is eliminated to obtain multiple high-frequency fault event sets.

[0036] As a preferred solution of the multi-source heterogeneous data synchronization and reconstruction system of the new energy station of the present invention, the source tracing analysis module is also used for:

[0037] The random forest model is used to analyze the high-frequency fault event set;

[0038] Based on the analysis results, the fault factors with high frequency of occurrence in the high-frequency fault event set are counted;

[0039] From the statistical results, a preset frequency threshold is set, and the fault factor whose factor frequency exceeds the preset threshold is selected as the first fault factor;

[0040] The screened first fault factors are summarized, and multiple initial high-frequency fault factor sets are analyzed in sequence and integrated to form a large set including all high-frequency fault factors;

[0041] De-duplication is performed on the integrated high-frequency fault factor set to remove duplicate fault factors;

[0042] After deduplication, a set of high-frequency fault factors is obtained.

[0043] As a preferred solution of the multi-source heterogeneous data synchronization and reconstruction system of the new energy station of the present invention, the fixed-point monitoring module is also used for:

[0044] Analyze the high-frequency fault factor set;

[0045] After analysis, interference factors of data collection are determined, wherein the interference factors at least include ambient temperature, ambient humidity, and electromagnetic interference;

[0046] Collect interference factors based on interference factors and determine real-time interference factors;

[0047] Optimizing and correcting the real-time monitoring data set according to the real-time interference factor to obtain an optimized monitoring data set, and constructing a monitoring data matrix based on the optimized monitoring data set;

[0048] Processing the monitoring data set according to a preset data processing strategy to obtain a standard monitoring data set, wherein the preset data processing strategy includes redundancy elimination, missing supplementation and error correction;

[0049] Optimizing and correcting the standard monitoring data set according to the real-time interference factors to obtain an optimized monitoring data set;

[0050] Collecting and obtaining a sample training data set, wherein the sample training data set includes a sample interference factor, a sample monitoring data set, and a sample optimization monitoring data set;

[0051] Taking sample interference factors and sample monitoring data sets as input and sample optimization monitoring data sets as output, supervised training of BP neural network is performed based on sample training data sets to obtain the optimization correction model;

[0052] The real-time interference factors and the standard monitoring data set are input into the optimization correction model, and the optimized monitoring data set is output.

[0053] As a preferred solution of the multi-source heterogeneous data synchronization and reconstruction system of the new energy station of the present invention, the similarity analysis module is also used for:

[0054] Based on the optimized monitoring data set, a similarity comparison operator set is obtained;

[0055] Obtaining a similarity comparison operator set, wherein the similarity comparison operator set at least includes a cosine similarity operator, a Euclidean distance operator, a Manhattan distance operator, and a Pearson coefficient operator;

[0056] Perform credibility analysis on the similarity comparison operators in the similarity comparison operator set in turn, set credibility weights based on the analysis results, and obtain multiple credibility weights;

[0057] Generate a fusion similarity comparison strategy based on multiple trust weights and multiple similarity comparison operators

[0058] Based on multiple similarity comparison operators, a similarity analysis is performed on the high-frequency fault feature matrix and the monitoring data matrix to obtain multiple similarity coefficients;

[0059] Based on multiple credible weights, multiple similarity coefficients are weighted and fused to obtain the comprehensive similarity.

[0060] As a preferred solution of the multi-source heterogeneous data synchronization and reconstruction system of the new energy station of the present invention, the fault warning module is also used for:

[0061] comparing the comprehensive similarity with a preset similarity threshold;

[0062] When the comprehensive similarity exceeds the preset similarity threshold, the system makes a warning decision and generates a warning signal;

[0063] Based on the alarm signal, the system will send the signal to relevant operation and maintenance personnel and decision makers through the predetermined communication channel;

[0064] After receiving the early warning signal, the operation and maintenance team and decision makers will take immediate action based on the recommendations in the report and provide feedback on the results.

[0065] The beneficial effects of the present invention are as follows: the information retrieval module accurately captures equipment attribute information, retrieves historical fault events online, and provides detailed data for subsequent analysis. The event clustering module uses the CNN model to efficiently identify high-frequency fault modes, greatly improving the accuracy and response speed of fault prediction. The source tracing analysis module deeply analyzes the root cause of the fault, constructs a fault feature matrix, and strengthens preventive maintenance. The fixed-point monitoring module optimizes the sensor layout according to the fault factor to achieve accurate and efficient real-time monitoring. The similarity analysis module integrates multiple algorithms to ensure the high sensitivity and accuracy of the early warning system. The fault warning module automatically triggers the early warning mechanism, notifies maintenance in a timely manner, and significantly reduces the risk and loss of faults. Overall, this system has significantly improved the operation and maintenance efficiency and safety level of GIS equipment through intelligent data analysis and management, reflecting major technological breakthroughs and application value in the field of fault warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0067] Figure 1 This is a flow chart of the intelligent monitoring and management system for digital GIS equipment faults in Example 1.

[0068] Figure 2 This is a flowchart of file name event clustering in Example 1. DETAILED DESCRIPTION

[0069] 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.

[0070] 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.

[0071] 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.

[0072] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a digital GIS equipment fault intelligent monitoring and management system, including the following steps:

[0073] S1. Obtain the attribute information of the target GIS equipment, and use the attribute information as a constraint to retrieve the sample fault event set through online retrieval.

[0074] Furthermore, detailed attribute information can be obtained from GIS devices;

[0075] Use attribute information as query conditions to search for historical data in the networked fault event database;

[0076] Among them, the number of sample failure events meets the preset constraints;

[0077] Specifically, the attribute information of the target GIS equipment is obtained, such as the equipment model, manufacturer, installation location, etc.; with the attribute information as a constraint, relevant sample fault events are retrieved from the networked fault event database to ensure that the number of retrieved sample fault events meets the preset constraint conditions (such as the lower limit of quantity, time range, etc.);

[0078] It should be noted that the information retrieval module is the foundation of the entire system. Its core task is to accurately and comprehensively collect the attribute information of the target GIS equipment, and based on this, screen out highly relevant historical fault cases from the huge networked fault event database. This process first requires us to deeply explore the detailed attributes of the equipment, including but not limited to key parameters such as equipment model, manufacturer information, installation location, and operating environment. Subsequently, these attribute information are converted into precise query conditions, and advanced retrieval technology is used to efficiently search the database to find sample fault events that match the characteristics of the target equipment. In order to ensure the effectiveness and reliability of the analysis, the system also sets strict preset constraints, such as the lower limit of the number of samples and the requirements of the time span, to ensure that the selected fault events are not only sufficient in number, but also meet the optimal standards in terms of timeliness and relevance. The efficient operation of this module provides a solid data foundation for subsequent fault pattern identification, root cause analysis, and early warning mechanisms, and is the key first step in achieving accurate fault early warning.

[0079] S2, event clustering module, is used to process the sample fault event set using the CNN model and apply the clustering algorithm to obtain multiple high-frequency fault event sets.

[0080] Furthermore, based on the attribute information and historical data of the fault event, the convolutional neural network (CNN) model is selected for fine-tuning to extract the deep feature vector of the fault event;

[0081] The CNN convolutional neural network model structure includes: input layer, convolution layer, activation layer, pooling layer, fully connected layer and output layer;

[0082] Input layer: responsible for preparing and formatting data so that it can flow smoothly into deeper layers of the network for feature extraction and learning;

[0083] Convolutional layer: extracts features from input data;

[0084] Activation layer: used to introduce nonlinearity, help the network learn complex mapping relationships, and improve the model's expressiveness and discrimination ability;

[0085] Pooling layer: mainly reduces the spatial size of the feature map, thereby reducing the amount of calculation and the number of parameters, and also has a certain ability to resist deformation;

[0086] Fully connected layer: integrates the features extracted by the convolutional layer and the pooling layer to learn higher-level abstract features;

[0087] Output layer: It is an extension of the fully connected layer and is used to generate the final prediction results.

[0088] Here, suppose we have an input image I with the size of H×W×C, where H is the height, W is the width, and C is the number of channels;

[0089] The convolutional layer receives the input image and scans the entire image through a set of filters F, expressed as:

[0090]

[0091] in represents the convolution layer, represents the value at position (i, j) on the kth channel of the feature map, F is the convolution kernel with a size of f×f and a depth of C, which is used to scan the entire input image, O represents the output of the convolution operation, that is, the feature map, B k Represents the bias term of the kth convolution kernel, which is often used to adjust the offset of the output feature map. m, n, and c represent the row index, column index, and channel index of the input image in the filter inner loop, respectively.

[0092] The activation layer applies nonlinear transformation to the output of the convolutional layer to increase the expressive power of the network. The expression is:

[0093]

[0094] Among them, A k i，j represents the activation layer, A represents the feature map after being processed by the activation function, and max represents the maximum value;

[0095] Upgrade the convolutional layer, activation layer, pooling layer, fully connected layer and input layer of the original CNN convolutional neural network model;

[0096] The depth-separable convolution is introduced into the convolution layer, which is divided into depth convolution and point-by-point convolution. The depth convolution expression is:

[0097]

[0098] in, The feature map representing the output of the deep convolution is the value at position (i, j) on the kth channel, where k is the channel index, i and j represent the row and column coordinates on the feature map, respectively, F is the size of the convolution kernel, m and n represent the row index and column index of the filter inner loop, respectively. is the weight of the kth channel of the depth convolution kernel at position m, n, is the value of the kth channel of the input image at position (i+m, j+n), B k is the bias term;

[0099] The point-by-point convolution expression is:

[0100]

[0101] in, represents the final output feature map of the point-by-point convolution, the value of the kth channel at position (i, j), and c represents the channel index of the input image. Represents the value of the cth channel at position (i, j) in the output feature map of the deep convolutional layer, G c，k is the weight matrix of the point-by-point convolution, connecting the input channel c and the output channel k, D k is the bias term;

[0102] For the activation layer, the LeakyReLU function is selected to apply nonlinear changes to the output of the convolutional layer to enhance the expression ability of the network. The expression is:

[0103]

[0104] in, Represents the value at position (i, j) on the kth channel in the output feature map of the activation layer. α represents a small positive number to avoid the "dead zone" problem of ReLU. Represents the value at position (i, j) on the kth channel in the output feature map of the convolutional layer as the input of the activation function. The entire formula is the expression of the LeakyReLU function;

[0105] The pooling layer usually follows the convolutional layer and the activation layer to reduce the spatial dimension of the feature map, reduce the amount of calculation and the number of parameters, and the expression is:

[0106]

[0107] Among them, P represents the output of the pooling operation, that is, the feature map after pooling, p k i，j Represents the value at position (i, j) on the kth channel of the feature map, and s represents the step size of the pooling window, which determines the moving distance of the pooling operation on the feature map;

[0108] The FractionalMaxPooling algorithm is used in the pooling layer to enhance the dimensionality reduction and anti-deformation capabilities. The expression is:

[0109]

[0110] Among them, p k i，jrepresents the value at position (i, j) on the kth channel of the feature map, N is the total number of elements in the pooling area, A i k Represents the position of the original feature map A and the value on channel k. A is the feature map after convolution and activation. s is the distance the pooling window moves on the feature map. r m and c m are randomly selected row and column indices, s is the step size, and this formula describes the calculation process of FractionalMaxPooling;

[0111] The fully connected layer flattens the feature map into a one-dimensional vector and performs a linear transformation through matrix multiplication and bias addition. The expression is:

[0112] FC l =W l ·Fl n (P)+B l ;

[0113] Among them, FC l represents the output of the lth fully connected layer, W l represents the weight matrix of the lth fully connected layer, B l Represents the bias vector of the lth fully connected layer, Fl n (P) represents the process of flattening the feature map P into a one-dimensional vector;

[0114] Using Dropout regularization technology in the fully connected layer helps prevent overfitting and improve the generalization ability of the model. The expression is:

[0115] FC l =(W l ·Mk)·Fl n (P)+B l ;

[0116] Among them, FC l represents the output of the lth fully connected layer, l represents the hierarchical number of the current layer in the neural network, and W l represents the weight matrix of the lth fully connected layer, Mk represents the mask matrix used in the Dropout operation, whose elements are set to 0 or 1 with a certain probability, and Fl n (P) represents the process of converting P into a one-dimensional vector. P refers to the output of the previous layer. B l Represents the bias vector of the lth fully connected layer;

[0117] The output layer is usually also a fully connected layer, and may be connected to a normalization layer at the end for classification tasks. The expression is:

[0118]

[0119] Among them, Y j represents the output layer, represents the probability of the jth category, L represents the index of the last fully connected layer, FC L j represents the output of the last fully connected layer, j represents the index of the category in the output layer of the neural network, L represents the number of layers of the neural network, and T represents the temperature parameter. represents the exponential function applied to the jth element of the last fully connected layer, and e represents the base of the natural logarithm;

[0120] The output layer is normalized using the Softmax function and the temperature parameter T is introduced, which can adjust the "softness" of the output distribution. It is particularly useful for classification tasks, especially in multi-category fault diagnosis. The expression is:

[0121]

[0122] Among them, Y j is the output layer, representing the probability of the jth category, FC L j represents the output of the last layer of the neural network, j represents the index of the category in the output layer of the neural network, L represents the number of layers of the neural network, and T represents the temperature parameter. Represents the exponential operation FC L j The value after dividing by the temperature parameter T is exponentially expressed, e represents the base of the natural logarithm, ∑ k represents the summation symbol, indicating that we want to sum all categories of FC L j Output is operated to traverse all possible categories, Represents the exponential operation FC L k Divided by the temperature parameter T, this formula clearly shows how the Softmax function converts the output of the last layer of the neural network into a probability distribution.

[0123] Calculate the failure ratio of each sample failure event subset to obtain multiple failure ratios;

[0124] Eliminate a subset of sample fault events whose fault proportion is less than a preset proportion threshold, and obtain multiple high-frequency fault event sets;

[0125] Specifically, a large number of sample fault events are collected, including various types of faults, and clustering algorithms (such as K-means, hierarchical clustering, etc.) are used to cluster the fault events. The purpose of clustering is to group similar fault events together so that different fault types can be identified. Through clustering, multiple subsets of sample fault events can be formed, each of which represents a specific fault type.

[0126] Then, for each sample fault event subset, the ratio of the number of fault events contained in it to the total number of fault events is calculated, which reflects the proportion of this type of fault in the overall faults, and an overview of the frequency of occurrence of various fault types is obtained;

[0127] Next, a preset percentage threshold is set to determine the high-frequency fault type, that is, the fault type that occurs frequently and needs to be paid special attention to. The preset percentage threshold is a custom judgment threshold used to determine high-frequency faults. The sample fault event subset with a fault percentage less than the preset threshold is eliminated, indicating that the fault type has a low frequency of occurrence and may not be the focus of maintenance and management work. After eliminating low-frequency faults, what remains is the high-frequency fault event set, which is the fault type that needs to be paid special attention to and solved;

[0128] By effectively identifying high-frequency fault types, it helps operation and maintenance personnel locate and solve problems more accurately, thereby improving the operating efficiency and reliability of equipment;

[0129] It should be noted that the event clustering module can not only automatically identify and distinguish different types of failure modes, but also further screen out the most common and critical problem areas, providing a strong basis for maintenance decisions and resource allocation.

[0130] S3, source tracing analysis module, is used to perform source tracing analysis on multiple high-frequency fault event sets in sequence based on the accident tree principle, determine the high-frequency fault factor set, and generate a high-frequency fault feature matrix.

[0131] Furthermore, the random forest model is used to analyze the high-frequency fault event set;

[0132] Based on the analysis results, the fault factors with high frequency of occurrence in the high-frequency fault event set are counted;

[0133] From the statistical results, a preset frequency threshold is set, and the fault factor whose factor frequency exceeds the preset threshold is selected as the first fault factor;

[0134] The screened first fault factors are summarized, and multiple initial high-frequency fault factor sets are analyzed in sequence and integrated to form a large set including all high-frequency fault factors;

[0135] De-duplication is performed on the integrated high-frequency fault factor set to remove duplicate fault factors;

[0136] After deduplication is completed, a set of high-frequency fault factors is obtained;

[0137] Specifically, an accident tree model is constructed to represent the relationship between fault events and their possible causes. The source analysis is performed on each high-frequency fault event set to determine the high-frequency fault factors, and the high-frequency fault factors are organized into a high-frequency fault feature matrix.

[0138] Furthermore, a first high-frequency fault event set is selected from multiple high-frequency fault event sets; based on the fault tree principle, a plurality of first high-frequency fault events in the first high-frequency fault event set are traced and analyzed to determine a plurality of first fault factors and a plurality of factor frequencies; a first fault factor whose factor frequency is greater than a preset frequency threshold is set as a first high-frequency fault factor to obtain a first initial high-frequency fault factor set; multiple initial high-frequency fault factor sets are analyzed in sequence, and multiple initial high-frequency fault factor sets are deduplicated to obtain a high-frequency fault factor set;

[0139] Specifically, from multiple high-frequency fault event sets, a random one (or one selected based on importance, priority, etc.) high-frequency fault event set is selected as the analysis object;

[0140] Then, the accident tree analysis is a method used to determine the cause of system failure. It starts from the undesirable failure event (top event) and analyzes the various factors (intermediate events and basic events) that may cause the event step by step, thereby constructing a logical tree diagram. For multiple failure events in the selected first high-frequency failure event set, the accident tree principle is applied, starting from each failure event (as the top event), all possible causes (intermediate events and basic events) that lead to the failure are analyzed, that is, the source tracing analysis is performed; multiple first failure factors are determined to obtain the direct or indirect causes of the high-frequency failure events;

[0141] Next, for each first fault factor, the frequency of occurrence in the traceability analysis is counted, a preset frequency threshold is set, and the first fault factor with a frequency greater than the threshold is defined as a first high-frequency fault factor, and the factor frequency corresponding to the fault factor is compared with the preset frequency threshold, and the first high-frequency fault factors are formed into a first initial high-frequency fault factor set;

[0142] Next, the remaining high-frequency fault event sets are analyzed repeatedly to obtain multiple initial high-frequency fault factor sets, which are deduplicated because different high-frequency fault event sets may contain the same fault factors. Deduplication can ensure the uniqueness and accuracy of the high-frequency fault factor set. The final high-frequency fault factor set contains all important and frequently occurring fault factors, and is the key to preventing and controlling faults.

[0143] By identifying the high-frequency fault factors that have the greatest impact on system stability, it provides strong data support for subsequent fault prevention, monitoring and response;

[0144] It should be noted that the source tracing analysis module can not only reveal the nature of high-frequency fault events, but also guide subsequent fault prevention and system optimization work, contributing to improving the overall system reliability and safety.

[0145] S4, fixed-point monitoring module, is used to configure a multi-source sensor monitoring array based on a high-frequency fault factor set, and use the multi-source sensor monitoring array to perform fixed-point monitoring of the target GIS equipment, obtain a real-time monitoring data set, and construct a monitoring data matrix.

[0146] Furthermore, interference factors of data collection are analyzed and determined, wherein the interference factors at least include ambient temperature, ambient humidity, and electromagnetic interference;

[0147] Collect interference factors based on interference factors and determine real-time interference factors;

[0148] The real-time monitoring data set is optimized and corrected according to the real-time interference factors to obtain an optimized monitoring data set, and a monitoring data matrix is ​​constructed based on the optimized monitoring data set;

[0149] Processing the monitoring data set according to a preset data processing strategy to obtain a standard monitoring data set, wherein the preset data processing strategy includes redundancy elimination, missing supplementation and error correction;

[0150] The standard monitoring data set is optimized and corrected according to the real-time interference factors to obtain an optimized monitoring data set;

[0151] Collecting and obtaining a sample training data set, wherein the sample training data set includes a sample interference factor, a sample monitoring data set, and a sample optimization monitoring data set;

[0152] Taking sample interference factors and sample monitoring data sets as input and sample optimization monitoring data sets as output, supervised training of BP neural network is performed based on sample training data sets to obtain the optimization correction model;

[0153] The real-time interference factors and the standard monitoring data set are input into the optimization correction model, and the optimized monitoring data set is output;

[0154] Specifically, appropriate sensors and monitoring equipment are selected according to the high-frequency fault factor set, a multi-source sensor monitoring array is constructed, and the multi-source sensor monitoring array is deployed on the target GIS equipment to conduct fixed-point monitoring, collect real-time monitoring data, and organize it into a monitoring data matrix;

[0155] Furthermore, the interference factors of data collection are analyzed and determined, wherein the interference factors are user-defined or determined based on sample data analysis, and at least include ambient temperature, ambient humidity, and electromagnetic interference; interference factors are collected based on the interference factors to determine real-time interference factors; the real-time monitoring data set is optimized and corrected based on the real-time interference factors to obtain an optimized monitoring data set, and a monitoring data matrix is ​​constructed based on the optimized monitoring data set.

[0156] Specifically, temperature changes may affect the performance of the device and the accuracy of the sensor, and therefore are an important interference factor in data collection. High or low humidity may cause the performance of the internal circuit components of the device to degrade or be damaged, thus affecting the accuracy of data collection. Electromagnetic interference may come from surrounding electrical equipment or radio waves, which may interfere with the signal transmission of the sensor and cause data distortion.

[0157] Then, after determining the interference factors, collect real-time data of the interference factors, which can be achieved by installing corresponding sensors or monitoring equipment. Use temperature and humidity sensors to measure the ambient temperature and humidity around the equipment in real time. The monitoring of electromagnetic interference requires the use of special electromagnetic interference detection instruments, or indirectly evaluate by analyzing the noise level of the sensor signal;

[0158] Next, obtain the data of real-time interference factors and optimize and correct the real-time monitoring data set. According to the change of ambient temperature, adjust the parameters of the sensor or data acquisition system to compensate for the influence of temperature on equipment performance and sensor accuracy. According to the change of ambient humidity, perform humidity correction on the data to eliminate the influence of humidity on data acquisition. For electromagnetic interference, use more complex signal processing technology to filter or eliminate noise, so as to restore the original signal. After optimization and correction, the optimized monitoring data set is obtained. The optimized monitoring data is organized into a matrix form, namely the monitoring data matrix. The rows of the matrix represent different times or sampling points, and the columns represent different monitoring parameters (such as temperature, humidity, current, voltage, etc.).

[0159] By monitoring the data matrix, you can easily view and analyze the operating status and performance of the equipment at different time points;

[0160] Furthermore, the monitoring data set is processed according to a preset data processing strategy to obtain a standard monitoring data set, wherein the preset data processing strategy includes redundancy elimination, missing supplementation and error correction; the standard monitoring data set is optimized and corrected according to real-time interference factors to obtain an optimized monitoring data set;

[0161] Specifically, use hash functions or similarity algorithms to detect duplicate or similar data, set a threshold, and when the similarity between two data exceeds the threshold, they are considered redundant, retain one of the data, and delete the other redundant data. If the missing data has a clear pattern or trend, interpolation methods (such as linear interpolation, polynomial interpolation, etc.) can be used to estimate the missing values. If the missing data has no obvious pattern, statistical methods (such as mean, median, mode, etc.) or machine learning models (such as decision trees, K-nearest neighbors, etc.) can be used to predict the missing values. Use the threshold method to detect values ​​outside the normal range and mark them as errors or anomalies. If the pattern or cause of the error is known, the data can be corrected directly. If the pattern of the error is not certain, statistical methods (such as box plots, IQR rules, etc.) or machine learning models (such as anomaly detection algorithms) can be used to identify and correct the error;

[0162] Then, the original monitoring data set is processed by applying strategies such as redundant elimination, missing supplementation, and error correction in turn. A standard monitoring data set that has been preliminarily processed is obtained. After obtaining the standard monitoring data set, the impact of real-time interference factors (such as ambient temperature, humidity, electromagnetic interference, etc.) on the data is considered, and the corresponding correction model or algorithm is used to optimize and correct the standard monitoring data set according to the value of the real-time interference factor. For example, multiple regression, neural network and other models can be used to establish a mapping relationship between interference factors and monitoring data, and the data is corrected. After optimization and correction, an optimized monitoring data set is obtained, which has not only undergone preliminary data processing, but also considered the impact of real-time interference factors, so it is more accurate and reliable;

[0163] After obtaining the original monitoring data set, in order to ensure the accuracy and completeness of the data, it is necessary to process the data according to the preset data processing strategy and optimize and correct it by considering the real-time interference factors;

[0164] Furthermore, a sample training data set is collected, wherein the sample training data includes a sample interference factor, a sample monitoring data set, and a sample optimized monitoring data set; the sample interference factor and the sample monitoring data set are used as input, and the sample optimized monitoring data set is used as output, and the BP neural network is supervised and trained based on the sample training data set to obtain an optimized correction model; the real-time interference factor and the standard monitoring data set are input into the optimized correction model, and the optimized monitoring data set is output;

[0165] Specifically, real-time interference factor data related to GIS equipment monitoring, such as ambient temperature, ambient humidity, electromagnetic interference, etc., are collected to obtain the original monitoring data of GIS equipment under different interference factor conditions. For each sample monitoring data set, known methods or expert knowledge are applied to optimize and correct it, and the optimized monitoring data is obtained as the training target of the neural network;

[0166] Then, the sample interference factors and the sample monitoring data set are used as the input of the neural network. The number of neurons in the input layer should match the number of features of the input data. One or more hidden layers are designed, each containing a certain number of neurons. The number of hidden layers and the number of neurons in each layer usually need to be adjusted according to the actual problem. The number of neurons in the output layer should match the number of features of the sample optimization monitoring data set, and the output layer will output the optimized monitoring data. Using the back propagation algorithm and the gradient descent method, the weights and bias terms of the neural network are iteratively adjusted so that the output of the neural network gradually approaches the sample optimization monitoring data set. During the training process, cross-validation and other techniques are used to prevent overfitting, and the training effect is optimized by adjusting hyperparameters such as the learning rate and batch size. After sufficient iterative training, the neural network will learn to predict the optimized monitoring data from the sample interference factors and the sample monitoring data set. Save the trained neural network as an optimized correction model;

[0167] Next, according to the input real-time interference factors and the standard monitoring data set, an optimized monitoring data set is output;

[0168] By using BP (back propagation) neural network to correct interference factors when building and optimizing the intelligent monitoring and management system for GIS equipment failures;

[0169] It should be noted that the fixed-point monitoring module can provide high-quality, high-precision real-time monitoring data, providing strong support for the health management and fault diagnosis of GIS equipment.

[0170] S5, a similarity analysis module, is used to perform similarity analysis on the high-frequency fault feature matrix and the monitoring data matrix according to the fusion similarity comparison strategy to determine the comprehensive similarity.

[0171] Furthermore, based on the optimized monitoring data set, a similarity comparison operator set is obtained;

[0172] Obtaining a similarity comparison operator set, wherein the similarity comparison operator set at least includes a cosine similarity operator, a Euclidean distance operator, a Manhattan distance operator, and a Pearson coefficient operator;

[0173] Perform credibility analysis on the similarity comparison operators in the similarity comparison operator set in turn, set credibility weights based on the analysis results, and obtain multiple credibility weights;

[0174] Generate a fusion similarity comparison strategy according to multiple trust weights and multiple similarity comparison operators;

[0175] Based on multiple similarity comparison operators, a similarity analysis is performed on the high-frequency fault feature matrix and the monitoring data matrix to obtain multiple similarity coefficients;

[0176] Based on multiple credible weights, multiple similarity coefficients are weighted and fused to obtain a comprehensive similarity;

[0177] Specifically, the pre-trained fusion similarity comparison strategy (including multiple similarity comparison operators and corresponding trust weights) is loaded, and the fusion similarity comparison strategy is applied to perform similarity analysis on the high-frequency fault feature matrix and the monitoring data matrix, and the comprehensive similarity is calculated based on the analysis results;

[0178] Furthermore, a similarity comparison operator set is obtained, wherein the similarity comparison operator set includes at least a cosine similarity operator, a Euclidean distance operator, a Manhattan distance operator, and a Pearson coefficient operator, credibility analysis is performed on the similarity comparison operators in the similarity comparison operator set in turn, credibility weights are set based on the analysis results, a plurality of credibility weights are obtained, and a fusion similarity comparison strategy is generated according to the plurality of credibility weights and the plurality of similarity comparison operators;

[0179] Specifically, it is used to calculate the cosine angle between two vectors to measure their similarity; it is often used to compare text or image feature vectors. The Euclidean distance operator calculates the straight-line distance between two points in multidimensional space and is suitable for similarity measurement of continuous numerical data. The Manhattan distance operator (also known as the city block distance) calculates the sum of the absolute values ​​of each dimension between two points in multidimensional space and is suitable for data sets with discrete features. The Pearson coefficient operator measures the linear correlation between two variables, with a value between -1 and 1, and is often used to evaluate the similarity between time series data or continuous variables.

[0180] Then, for each operator, analyze its performance under different data types, data distributions, and noise conditions. Use historical data or simulated data to test the performance of each operator, such as calculating accuracy, recall, F1 score, and other indicators. You can also consider the robustness (resistance to noise) and computational efficiency of the operator. According to the credibility analysis results, assign a weight to each operator. The weight can be a real number, indicating the importance of the operator in the fusion strategy. If an operator performs well in multiple tests, it can be given a higher weight. On the contrary, if the performance is poor, a lower weight is assigned. The weight setting can be based on expert knowledge, experimental results, or machine learning methods.

[0181] Next, multiple similarity comparison operators and their corresponding trust weights are combined to form a fused similarity comparison strategy. When calculating the similarity between two objects, multiple operators are weighted and fused. Each operator is used to calculate the similarity score, and then the score is multiplied by the corresponding weight. Finally, all weighted scores are added or averaged to obtain the final similarity score. Adjust and optimize according to specific needs, such as balancing the contributions of different operators by adjusting the weights.

[0182] By collecting similar comparison operators and evaluating the credibility of the operators in specific application scenarios, appropriate weights can be assigned to each operator;

[0183] Furthermore, based on multiple similarity comparison operators, similarity analysis is performed on the high-frequency fault feature matrix and the monitoring data matrix to obtain multiple similarity coefficients; based on multiple credible weights, multiple similarity coefficients are weighted and fused to obtain comprehensive similarity;

[0184] Specifically, the monitoring data matrix contains the characteristic data of historical high-frequency faults. Multiple similarity comparison operators are selected, such as cosine similarity, Euclidean distance (converted into similarity score), Manhattan distance (converted into similarity score) and Pearson correlation coefficient, etc., to calculate the similarity between the two matrices. For each similarity comparison operator, it is used for the high-frequency fault characteristic matrix and the monitoring data matrix to generate a similarity score or coefficient. For distance metrics (such as Euclidean distance and Manhattan distance), it is converted into a similarity score;

[0185] Then, based on historical data, expert knowledge or model testing, a trust weight is assigned to each similarity comparison operator, which reflects the effectiveness or reliability of the operator in a specific application scenario, ensuring that the sum of all weights is 1, so that weighted fusion can be performed in subsequent steps. The similarity score generated by each similarity comparison operator is multiplied by its corresponding trust weight, and then all weighted similarity scores are added together to obtain the comprehensive similarity;

[0186] Through comprehensive similarity, the high-frequency fault feature matrix that is most similar to the monitoring data matrix can be determined, so that the current equipment or system status may be predicted or identified;

[0187] It should be noted that the comprehensive similarity score can be used in a variety of decision-making scenarios, such as fault warning, fault type identification and fault severity assessment. When the score exceeds a certain threshold, the system can trigger the corresponding early warning mechanism or maintenance plan, thereby achieving effective monitoring and preventive maintenance of GIS equipment.

[0188] S6, a fault warning module, is used to issue a fault warning to the target GIS device if the comprehensive similarity meets a preset similarity threshold.

[0189] comparing the comprehensive similarity with a preset similarity threshold;

[0190] Furthermore, if the comprehensive similarity exceeds the preset similarity threshold, the system makes a warning decision and generates a warning signal;

[0191] This signal may include, but is not limited to: prediction of fault type, assessment of the likelihood of fault occurrence, and recommended emergency response measures;

[0192] Based on the alarm signal, the system will send the signal to relevant operation and maintenance personnel and decision makers through the predetermined communication channel;

[0193] Communication channels include: SMS, email, APP push, etc.;

[0194] After receiving the warning signal, the operation and maintenance team and decision makers will take immediate action based on the recommendations in the report and provide feedback on the results, such as dispatching maintenance teams, adjusting operating parameters, or implementing preventive maintenance plans to minimize the impact of potential failures. The feedback includes information such as whether the failure actually occurred and whether the measures taken are effective, which will be used to optimize future warning thresholds and strategies and improve the accuracy and efficiency of the warning system.

[0195] Specifically, set a suitable similarity threshold to determine whether there is a potential fault risk; when the comprehensive similarity exceeds the preset similarity threshold, the fault warning mechanism is triggered. Relevant personnel are notified to handle the problem through sound and light alarms, sending warning information, etc.

[0196] It should be noted that through such a closed-loop mechanism, the fault warning module can continuously learn and adapt, gradually improving its warning capability and response speed, and becoming an important line of defense to ensure the stable operation of GIS equipment.

[0197] In summary, the present invention uses an information retrieval module to obtain the attribute information of the target GIS equipment, and uses the attribute information as a constraint to retrieve and obtain a sample fault event set through the Internet, wherein the number of sample fault events meets the preset constraint; an event clustering module is used to cluster the sample fault event set to obtain multiple high-frequency fault event sets; a source tracing analysis module is used to perform source tracing analysis on multiple high-frequency fault event sets in turn based on the accident tree principle, determine the high-frequency fault factor set, and generate a high-frequency fault feature matrix; a fixed-point monitoring module is used to configure a multi-source sensor monitoring array based on the high-frequency fault factor set, and use the multi-source sensor monitoring array to perform fixed-point monitoring on the target GIS equipment to obtain a real-time monitoring data set and construct a monitoring data matrix; a similarity analysis module is used to perform similarity analysis on the high-frequency fault feature matrix and the monitoring data matrix according to a fusion similarity comparison strategy to determine the comprehensive similarity; a fault warning module is used to perform a fault warning on the target GIS equipment if the comprehensive similarity meets the preset similarity threshold, thereby achieving the technical effect of improving the degree of automation and intelligence and the accuracy and reliability of detection.

[0198] Example 2, referring to Table 1, is the second example of the present invention. To further verify the advancement of the present invention, experimental simulation data of the digital GIS equipment fault intelligent monitoring and management system are provided.

[0199] A GIS device in actual operation was selected as the research object. The experiment aims to demonstrate the ability of the present invention in improving the accuracy and efficiency of fault warning by comparing with the existing technology.

[0200] First, the implementation of the information retrieval module: Detailed attribute information was obtained from the target GIS equipment, including equipment model, operating years, working environment, etc. Then, using this attribute information as query conditions, historical data was searched in the networked fault event database, and a total of 1500 relevant fault event records were collected;

[0201] Secondly, the implementation of the event clustering module: the above sample fault event set is processed using an improved CNN model. The model introduces depthwise separable convolution in the convolution layer, uses the LeakyReLU function in the activation layer, uses the FractionalMaxPooling algorithm in the pooling layer, and applies the Dropout regularization technology in the fully connected layer. The output layer adds the Softmax function of the temperature parameter T. Through the clustering algorithm, 10 high-frequency fault event sets are obtained, each of which contains different types of fault modes.

[0202] Next, the implementation of the source tracing analysis module: the source tracing analysis was conducted on 10 high-frequency fault event sets, and the key fault factors affecting the stability of the equipment were determined based on the random forest model, such as insulation material aging, gas pressure fluctuations, arc discharge, etc. By setting the frequency threshold, 20 high-frequency fault factors were screened out to form a high-frequency fault factor set; the implementation of the fixed-point monitoring module: based on the high-frequency fault factor set, a multi-source sensor monitoring array was configured to conduct fixed-point monitoring of the target GIS equipment for 30 consecutive days. The monitoring data set includes equipment operating status, environmental parameters, etc. By correcting the real-time interference factors, an optimized monitoring data matrix was constructed;

[0203] Furthermore, the implementation of the similarity analysis module: adopting the fusion similarity comparison strategy, the high-frequency fault feature matrix and the monitoring data matrix are analyzed for similarity. By calculating the cosine similarity, Euclidean distance, Manhattan distance and Pearson coefficient, multiple similarity coefficients are obtained, and then weighted fusion is performed according to the trust weight to determine the comprehensive similarity;

[0204] Finally, the implementation of the fault warning module: the similarity threshold is set to 0.8. When the comprehensive similarity exceeds this threshold, the system automatically triggers a fault warning. The warning signal includes fault type prediction, possibility assessment and recommended measures. The operation and maintenance personnel and decision makers are notified in real time through SMS, email and APP push.

[0205] The details are shown in Table 1 below:

[0206] Table 1 Experimental record of digital GIS equipment fault intelligent monitoring and management system

[0207]

[0208] It can be seen from the examples that the new energy station multi-source heterogeneous data synchronization and reconstruction system of the present invention has shown significant advantages in GIS equipment fault warning:

[0209] Attribute information completeness: This system reached 95%, much higher than the 80% of Comparison System 1 and 85% of Comparison System 2, which means that this system can understand the equipment status more comprehensively and provide a richer data foundation for subsequent fault event retrieval and analysis;

[0210] Number of fault events: This system collected 1,500 fault event records, while the comparison systems 1 and 2 collected 1,400 and 1,450 respectively. More fault event records help improve the accuracy and reliability of cluster analysis;

[0211] Number of high-frequency fault events: This system identified 10 high-frequency fault event sets, compared to 8 and 7 for systems 1 and 2, respectively. This shows that the event clustering algorithm of this system is more efficient and can locate frequently occurring fault modes more accurately.

[0212] Monitoring data integrity: The monitoring data integrity of this system reached 98%, which is higher than the 90% and 92% of the comparison system. This shows that the correction of interference factors in the monitoring process of this system is more effective, ensuring the quality of monitoring data;

[0213] Early warning accuracy: This system achieved an early warning accuracy of 92%, compared with 80% and 85% for systems 1 and 2, respectively. This directly reflects the innovation of the present invention in improving the accuracy and efficiency of fault early warning, which can significantly improve the operation and maintenance efficiency of GIS equipment and reduce downtime and economic losses caused by faults;

[0214] In summary, the present invention effectively improves the accuracy of GIS equipment fault warning through a series of innovative technical means, demonstrating its great potential and value in practical applications;

[0215] 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 digital GIS equipment fault intelligent monitoring and management system, characterized by: include, An information retrieval module is used to obtain the attribute information of the target GIS equipment, and to retrieve and obtain a sample fault event set online with the attribute information as a constraint; The event clustering module is used to process the sample fault event set using the CNN model and apply the clustering algorithm to obtain multiple high-frequency fault event sets; The source tracing analysis module is used to perform source tracing analysis on multiple high-frequency fault event sets in sequence based on the fault tree principle, determine the high-frequency fault factor set, and generate a high-frequency fault feature matrix; A fixed-point monitoring module is used to configure a multi-source sensor monitoring array based on the high-frequency fault factor set, and use the multi-source sensor monitoring array to perform fixed-point monitoring on the target GIS equipment, obtain a real-time monitoring data set, and construct a monitoring data matrix; The similarity analysis module is used to perform similarity analysis on the high-frequency fault feature matrix and the monitoring data matrix according to the fusion similarity comparison strategy to determine the comprehensive similarity; The fault warning module is used to issue a fault warning to the target GIS device if the comprehensive similarity meets a preset similarity threshold.

2. The digital GIS equipment fault intelligent monitoring and management system according to claim 1, characterized in that: The information retrieval module is also used for: Get detailed attribute information from GIS equipment; Use attribute information as query criteria to search historical data in the networked fault event database.

3. The digital GIS equipment fault intelligent monitoring and management system as claimed in claim 2 is characterized in that: The event clustering module is also used for: Based on the attribute information and historical data of the fault event, the convolutional neural network (CNN) model is selected for fine-tuning to extract the deep feature vector of the fault event; The depth-separable convolution is introduced into the convolution layer of the convolutional neural network CNN model. The depth-separable convolution expression is: Among them, O i k ，j represents the feature map of the deep convolution output, k is the channel index, i and j represent the row and column coordinates on the feature map, F is the size of the convolution kernel, m, n, represent the row index and column index of the filter inner loop, respectively, and F m k ，n is the weight of the kth channel of the depth convolution kernel at position m, n, I i k +m，j+n is the value of the kth channel of the input image at position (i+m, j+n), B k is the bias term; The point-by-point convolution expression is: Among them, P i k ，j represents the final output feature map of point-by-point convolution, c represents the channel index of the input image, O i c ，j Represents the value of the cth channel at position (i, j) in the output feature map of the deep convolutional layer. G c，k is the weight matrix of point-wise convolution, D k is the bias term; For the activation layer of the convolutional neural network CNN model, the LeakyReLU function is selected to apply nonlinear changes to the output of the convolutional layer. The expression is: Among them, A i k ，j represents the value at position (i, j) on the kth channel of the output feature map of the activation layer, α represents a small positive number, O i k ，j Represents the k-th channel in the convolutional layer output feature map, position (i, The value at j) is used as the input of the activation function; The FractionalMaxPooling algorithm is used in the pooling layer of the convolutional neural network CNN model, and the expression is: Among them, p k i，j represents the value at position (i, j) on the kth channel of the feature map, N is the total number of elements in the pooling area, A i k Represents the position of the original feature map A and the value on channel k. A is the feature map after convolution and activation. s is the distance the pooling window moves on the feature map. r m and c m are randomly selected row and column indices, and s is the step size; Using the Dropout regularization technique in the fully connected layer of the convolutional neural network CNN model, the expression is: FC l =(W l ·Mk)·Fl n (P)+B l ; Among them, FC l represents the output of the lth fully connected layer, l represents the hierarchical number of the current layer in the neural network, and W l represents the weight matrix of the lth fully connected layer, Mk represents the mask matrix used in the Dropout operation, and Fl n (P) represents the process of converting P into a one-dimensional vector. P refers to the output of the previous layer. B l Represents the bias vector of the lth fully connected layer; The output layer of the convolutional neural network CNN model is normalized using the Softmax function, and the expression is: Among them, Y j is the output layer, representing the probability of the jth category, FC L j represents the output of the last layer of the neural network, j represents the index of the category in the output layer of the neural network, L represents the number of layers of the neural network, and T represents the temperature parameter. Represents the exponential operation FC L j The value after dividing by the temperature parameter T is exponentially expressed, and e represents the base of the natural logarithm. Represents the exponential operation FC L k Divide the value by the temperature parameter T for exponentialization; Calculate the failure ratio of each sample failure event subset to obtain multiple failure ratios; A subset of sample fault events whose fault proportion is less than a preset proportion threshold is eliminated to obtain multiple high-frequency fault event sets.

4. The digital GIS equipment fault intelligent monitoring and management system as claimed in claim 3 is characterized in that: The traceability analysis module is also used for: The random forest model is used to analyze the high-frequency fault event set; Based on the analysis results, the fault factors with high frequency of occurrence in the high-frequency fault event set are counted; From the statistical results, a preset frequency threshold is set, and the fault factor whose factor frequency exceeds the preset threshold is selected as the first fault factor; The screened first fault factors are summarized, and multiple initial high-frequency fault factor sets are analyzed in sequence and integrated to form a large set including all high-frequency fault factors; De-duplication is performed on the integrated high-frequency fault factor set to remove duplicate fault factors; After deduplication, a set of high-frequency fault factors is obtained.

5. The digital GIS equipment fault intelligent monitoring and management system as claimed in claim 4 is characterized in that: The fixed-point monitoring module is also used for: Analyze the high-frequency fault factor set; After analysis, interference factors of data collection are determined, wherein the interference factors at least include ambient temperature, ambient humidity, and electromagnetic interference; Collect interference factors based on interference factors and determine real-time interference factors; The real-time monitoring data set is optimized and corrected according to the real-time interference factors to obtain an optimized monitoring data set, and a monitoring data matrix is ​​constructed based on the optimized monitoring data set.

6. The digital GIS equipment fault intelligent monitoring and management system according to claim 4, characterized in that: The fixed-point monitoring module is also used for: Processing the monitoring data set according to a preset data processing strategy to obtain a standard monitoring data set, wherein the preset data processing strategy includes redundancy elimination, missing supplementation and error correction; The standard monitoring data set is optimized and corrected according to the real-time interference factors to obtain an optimized monitoring data set.

7. The digital GIS equipment fault intelligent monitoring and management system according to claim 4, characterized in that: The fixed-point monitoring module is also used for: Collecting and obtaining a sample training data set, wherein the sample training data set includes a sample interference factor, a sample monitoring data set, and a sample optimization monitoring data set; Taking sample interference factors and sample monitoring data sets as input and sample optimization monitoring data sets as output, supervised training of BP neural network is performed based on sample training data sets to obtain the optimization correction model; The real-time interference factors and the standard monitoring data set are input into the optimization correction model, and the optimized monitoring data set is output.

8. The digital GIS equipment fault intelligent monitoring and management system according to claim 5 is characterized in that: The similarity analysis module is also used to: Based on the optimized monitoring data set, a similarity comparison operator set is obtained; Obtaining a similarity comparison operator set, wherein the similarity comparison operator set at least includes a cosine similarity operator, a Euclidean distance operator, a Manhattan distance operator, and a Pearson coefficient operator; Perform credibility analysis on the similarity comparison operators in the similarity comparison operator set in turn, set credibility weights based on the analysis results, and obtain multiple credibility weights; A fusion similarity comparison strategy is generated according to multiple trust weights and multiple similarity comparison operators.

9. The digital GIS equipment fault intelligent monitoring and management system according to claim 5, characterized in that: The similarity analysis module is also used for: Based on multiple similarity comparison operators, a similarity analysis is performed on the high-frequency fault feature matrix and the monitoring data matrix to obtain multiple similarity coefficients; Based on multiple credible weights, multiple similarity coefficients are weighted and fused to obtain the comprehensive similarity.

10. The digital GIS equipment fault intelligent monitoring and management system according to claim 6, characterized in that: The fault warning module is also used for: comparing the comprehensive similarity with a preset similarity threshold; When the comprehensive similarity exceeds the preset similarity threshold, the system makes a warning decision and generates a warning signal; Based on the alarm signal, the system will send the signal to relevant operation and maintenance personnel and decision makers through the predetermined communication channel; After receiving the early warning signal, the operation and maintenance team and decision makers will take immediate action based on the recommendations in the report and provide feedback on the results.

Images