An intelligent relay protection hidden danger detection method and system

Abstract

An intelligent relay protection hidden danger detection method and system, the method includes data collection, data preprocessing, fault hidden danger prediction, fault classification and positioning, and relay protection hidden danger detection. The present invention relates to the field of relay protection hidden danger detection technology, and specifically refers to an intelligent relay protection hidden danger detection method and system. This solution adopts a fine-grained relay protection hidden danger detection method from hidden danger detection to classification and positioning, and provides intelligent detection demand support by first detecting and predicting the probability of fault occurrence and the trend change of the probability of occurrence, and confirms the type and spatiotemporal location of the fault by further performing classification and positioning, thereby improving the overall efficiency, intelligence, and detection dimension of relay protection hidden danger detection; adopts a time-series convolutional bidirectional gated recurrent network combined with feature reconstruction optimization to predict fault hidden dangers; adopts a convolutional bidirectional gated recurrent fusion network combined with a position perception graph to classify and locate faults.

Description

Technical Field

[0001] The present invention relates to the technical field of relay protection hidden danger detection, and in particular to an intelligent relay protection hidden danger detection method and system. Background Art

[0002] The intelligent relay protection hidden danger detection method is a power system protection monitoring method based on artificial intelligence and big data analysis technology. Its core function is to identify and warn potential fault hazards or equipment abnormalities through real-time data collection and analysis of relay protection equipment, so as to help operation and maintenance personnel to promptly discover and deal with safety hazards in the power system. This method not only improves the automation level of the system and reduces the need for manual intervention, but also can predict possible equipment failures in advance and optimize protection strategies, thereby improving the reliability and safety of the power system. By introducing intelligent technology, the relay protection hidden danger detection method can more accurately and quickly identify the potential risks of power equipment, providing a strong guarantee for the stable operation of the power system.

[0003] However, among the existing intelligent relay protection hidden danger detection methods, there are two ideas: manual detection and automatic detection. The automatic detection of relay protection hidden dangers involves complex analysis and evaluation of fault types, time series and fault locations. Although the existing methods are relatively mature in the technology of fault classification detection, they lack the combination of time and space detection combining time and location. At the same time, the existing methods are difficult to provide prediction and detection support from the three dimensions of time, space and fault type, resulting in the technical problem that the degree of intelligence needs to be improved.

[0004] Among the existing fault hidden danger prediction methods, there is a technical problem that the existing methods often use single-stage hidden danger detection, and the results of their intelligent output also focus on the specific types of hidden dangers. However, this method is actually easy to ignore the various complex factors that cause hidden dangers, because the hidden dangers that may exist in the operation process of relay protection equipment are actually complex and changeable, involving not only electrical and physical characteristics, but also environmental changes, and even the influence of maintenance history and fault history. Therefore, the single-stage prediction of traditional methods is difficult to meet the technical problem of fine-grained detection requirements;

[0005] In the existing fault classification and positioning, there is a technical problem that traditional methods have difficulty in performing multi-output prediction and detection of time, space and type at the same time. At the same time, traditional single-stage detection methods are difficult to support multi-dimensional classification and positioning outputs, while traditional multi-stage detection methods are difficult to achieve specific detection in the time and space dimensions. Summary of the invention

[0006] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent relay protection hidden danger detection method and system. In the existing intelligent relay protection hidden danger detection methods, there are two ideas: manual detection and automatic detection. The automatic detection of relay protection hidden dangers involves complex analysis and evaluation of fault types, time series and fault locations. Although the existing methods are relatively mature in the technology of fault classification detection, they lack the combination of time and space detection combined with time and positioning. At the same time, the existing methods are difficult to provide predictive detection support from the three dimensions of time, space and fault type, resulting in a technical problem that the degree of intelligence needs to be improved. This solution creatively adopts The fine-grained relay protection hidden danger detection method from hidden danger detection to classification and positioning provides intelligent detection demand support by first detecting and predicting the probability of fault occurrence and the trend change of the probability of occurrence, and confirms the type and spatiotemporal location of the fault by further performing classification and positioning, thereby improving the overall efficiency, intelligence and detection dimension of relay protection hidden danger detection; in the existing fault hidden danger prediction methods, there is a problem that the existing methods often adopt single-stage hidden danger detection, and the results of their intelligent output also focus on the specific type of hidden danger, but this method is actually easy to ignore the various complex factors that lead to hidden dangers, because the hidden dangers that may exist in the operation process of relay protection equipment are actually complex and changeable, involving not only electrical and physical characteristics, but also environmental changes, and even maintenance history. The single-stage prediction of traditional methods is difficult to meet the needs of fine-grained detection due to the influence of the history of power supply and the history of faults. This solution creatively adopts a time-series convolutional bidirectional gated recurrent network combined with feature reconstruction optimization to predict fault hazards. Through feature reconstruction optimization, the connection between various complex features is enhanced, and the long-term and short-term relay protection feature analysis is realized through time-series convolution and bidirectional gated recurrent networks. While improving the prediction accuracy of fault hazard prediction, the fault hazard prediction itself is used as a pre-step, which is conducive to subsequent more precise predictions. At the same time, it can also be used as a separate functional module for macro-level regulation and assistance, thereby improving the intelligent performance of relay protection hazard detection; for the existing fault classification and positioning There is a technical problem that traditional methods have difficulty in performing multi-output prediction and detection of time, space and type at the same time. At the same time, traditional single-stage detection methods are difficult to support multi-dimensional classification and positioning output, while traditional multi-stage detection methods are difficult to achieve specific detection in the time and space dimensions. This solution creatively uses a convolutional bidirectional gated recurrent fusion network combined with a position perception map to classify and locate faults. It applies sensor encoding in the data acquisition stage to position sensing mapping, improves feature encoding and processing, and finally uses a convolutional bidirectional gated recurrent fusion network to realize spatial and temporal feature processing, thereby improving the ability of fault classification and positioning to output multi-dimensional information. At the same time, it also provides an exploration and practical experience for the intelligent direction of relay protection hidden danger detection.

[0007] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent relay protection hidden danger detection method, the method comprising the following steps:

[0008] Step S1: data collection;

[0009] Step S2: data preprocessing;

[0010] Step S3: Fault potential prediction;

[0011] Step S4: fault classification and location;

[0012] Step S5: Relay protection hidden danger detection.

[0013] Further, in step S1, the data collection is used to collect the original data set required for intelligent relay protection hidden danger detection, specifically, by collecting data during the operation of the relay protection equipment to obtain the original data set for relay protection hidden danger detection;

[0014] The relay protection hidden danger detection original data set specifically includes electrical data, relay action status data, environmental data, equipment operation history data and fault alarm signal data;

[0015] The relay action state data specifically includes relay switch state data and relay operation time data;

[0016] The environmental data specifically includes temperature data, humidity data and sensor location data;

[0017] The equipment operation history data specifically includes equipment operation time data, historical fault record data and historical maintenance record data.

[0018] Further, in step S2, the data preprocessing is used to optimize the original data, specifically, data preprocessing is performed on the original data set of relay protection hidden danger detection to obtain an optimized data set of relay hidden danger detection, including the following steps:

[0019] Step S21: data denoising, specifically, applying a filtering algorithm to the time series data in the original data set of relay protection hidden danger detection, performing a data smoothing and denoising operation, and obtaining smoothed and optimized data;

[0020] Step S22: data cleaning, specifically processing missing values ​​and outliers on the smoothed optimized data, specifically using interpolation to process missing data, and identifying and processing outliers through data distribution analysis to obtain a cleaned optimized data set;

[0021] Step S23: data standardization, specifically, performing standardization and normalization operations on the cleaning optimization data set in sequence, wherein the standardization is used to process electrical data, and the normalization is used to process environmental data, to obtain a standard optimization data set;

[0022] Step S24: time feature optimization, specifically, performing a sliding window division operation on the time series data in the standard optimization data set, and obtaining time statistical features and periodic frequency features through time series alignment and time series conversion operations, and obtaining time feature optimization data by integrating the time statistical features and periodic frequency features;

[0023] Step S25: discrete data encoding, specifically, performing balanced optimization of category data by data oversampling, and performing label encoding operation on the balanced optimized discrete category data in the standard optimization data set to obtain discrete encoding feature data;

[0024] Step S26: integrating feature extraction, specifically the time feature optimization and the discrete data encoding, performing statistical feature extraction and frequency domain feature extraction to obtain a comprehensive feature data set;

[0025] Step S27: data optimization processing, specifically, performing data optimization processing through the data denoising, data cleaning, data standardization and integrated feature extraction to obtain a relay hidden danger detection optimization data set;

[0026] The relay hidden danger detection optimization data set specifically includes electrical characteristic data, relay action state characteristic data, environmental characteristic data, equipment operation history characteristic data and fault alarm signal characteristic data;

[0027] The electrical characteristic data, relay action state characteristic data and equipment operation history characteristic data are used to perform fault hidden danger prediction; the electrical characteristic data, environmental characteristic data and fault alarm signal characteristic data are used to perform fault classification and positioning.

[0028] Further, in step S3, the fault hidden danger prediction is used to predict the frequency and trend of fault hidden dangers occurring in the relay protection process, specifically, based on the relay hidden danger detection optimization data set, a temporal convolutional bidirectional gated recurrent network combined with feature reconstruction optimization is used to perform fault hidden danger prediction to obtain fault hidden danger prediction reference data;

[0029] The temporal bidirectional gated recurrent network combined with feature reconstruction optimization specifically includes a feature correlation optimization improvement operator, a temporal convolutional subnet and a bidirectional gated recurrent subnet;

[0030] The feature correlation optimization improvement operator is used to optimize the correlation between various features;

[0031] The time series convolution subnet is used to process short-term time series data and provide short-term probability prediction;

[0032] The bidirectional gated recurrent subnetwork is used to process long-term time series data and provide trend probability prediction;

[0033] The step of using a temporal convolutional bidirectional gated recurrent network combined with feature reconstruction optimization to predict potential fault hazards and obtain reference data for potential fault hazard prediction includes:

[0034] Step S31: feature correlation optimization, specifically, optimizing the feature data correlation in the relay hidden danger detection optimization data set by feature correlation expansion and reconstruction, and obtaining reconstructed correlation optimization feature data, specifically including the following steps:

[0035] Step S311: Calculate the correlation between the fault hidden danger prediction characteristics, which is used to calculate the correlation between the electrical characteristics and the relay state characteristics. The calculation formula is:

[0036] ;

[0037] In the formula, is the fault potential prediction feature correlation, which is used to represent the combined feature data The correlation between the target feature y and the fault potential prediction target feature y, x1 is the first feature data, which is used to represent the electrical feature data, and x2 is the second feature data, which is used to represent the relay action state feature data. is an electrical correlation feature, used to indicate the correlation between the first feature data x1 and the fault potential prediction target feature y. is the feature combination correlation, which is used to indicate the correlation between the first feature data x1 and the second feature data x2. is a relay state correlation feature, which is used to represent the correlation between the second feature data x2 and the fault hidden danger prediction target feature y, wherein the correlation parameter r specifically refers to the Pearson coefficient;

[0038] Step S312: feature correlation extension, which is used to extend the correlation calculation to all features. The calculation formula is:

[0039] ;

[0040] In the formula, is the extended electrical correlation feature, is the electrical correlation feature, m is the total number of features, a is the feature index, It is the ath feature combination correlation, which is used to represent the first feature data x1 and the ath feature data x a The correlation between It is the ath feature combined with the correlation degree, used to represent the ath feature data x a The correlation between y and the target feature y for fault potential prediction;

[0041] Among them, the specific value of m is 3. When a is 3, x a Used to indicate the historical characteristics of equipment operation;

[0042] Step S313: Reconstruction of associated features, used to perform comprehensive weighting of all features, the calculation formula is:

[0043] ;

[0044] In the formula, R is the reconstructed association optimization feature data, is the extended electrical correlation feature, It is an electrical correlation feature;

[0045] Step S32: constructing a temporal convolutional subnetwork, specifically constructing a standard temporal convolutional subnetwork including causal convolution, dilated convolution and residual block, and optimizing temporal feature extraction by extended convolution to construct the temporal convolutional subnetwork;

[0046] Step S33: constructing a bidirectional gated recurrent subnetwork, specifically constructing a standard gated recurrent unit including an update gate and a reset gate, and constructing a bidirectional gated recurrent subnetwork by setting a forward and backward bidirectional structure;

[0047] Step S34: Fault hidden danger prediction model training, specifically, performing fault hidden danger prediction model training through the feature correlation optimization, the construction of the time series convolutional subnet and the construction of the bidirectional gated recurrent subnet to obtain the fault hidden danger prediction model Model P ;

[0048] Step S35: Fault potential prediction, specifically by using the fault potential prediction model Model P , based on the electrical characteristic data, relay action state characteristic data and equipment operation history characteristic data in the relay hidden danger detection optimization data set, fault hidden danger prediction is performed to obtain fault hidden danger prediction reference data;

[0049] The fault hidden danger prediction reference data specifically includes probability data for predicting whether the fault hidden danger will occur, trend data for predicting whether the fault hidden danger will occur, and window time data for fault hidden danger warning.

[0050] Further, in step S4, the fault classification and positioning is used to predict the specific type and location of the fault hidden danger in the relay protection process, specifically, based on the fault hidden danger prediction reference data and the relay hidden danger detection optimization data set, a convolutional bidirectional gated recurrent fusion network combined with a position perception graph is used to perform fault classification and positioning to obtain fault classification and positioning reference data;

[0051] The convolutional bidirectional gated recurrent fusion network combined with the position-aware graph specifically includes a feature-extended position-aware graph, an improved position-aware attention aggregation operator, a convolutional feature extraction subnet, and a bidirectional gated recurrent subnet;

[0052] The feature-expanded location perception map is used to enhance the location perception capability of relay fault hazards;

[0053] The improved location-aware attention aggregation operator is used to aggregate location-aware features and perform weighted aggregation;

[0054] The convolutional feature extraction subnet is used to extract classification features and position features from multidimensional features;

[0055] The bidirectional gated recurrent subnetwork is used to optimize the potential temporal features in the classification and localization process;

[0056] The step of using a convolutional bidirectional gated recurrent fusion network combined with a position perception graph to perform fault classification and location, and obtaining fault classification and location reference data, includes:

[0057] Step S41: constructing a feature-extended position-aware graph, specifically by constructing an initial position-aware mapping structure through node parameters and edge parameters, and obtaining a feature-extended position-aware graph by applying feature extension and position-aware attention initial calculation;

[0058] Step S42: improving position-aware attention aggregation, specifically by calculating a normalized attention coefficient and performing combined activation through a nonlinear function to calculate an activation enhancement node attention coefficient, and performing position-aware feature aggregation and multi-head attention mechanism optimization on the activation enhancement node attention coefficient to obtain improved position-aware attention aggregation feature data;

[0059] Step S43: constructing a convolutional feature extraction subnet, specifically constructing a standard convolutional feature extraction subnet including a convolutional layer and a pooling layer, and extracting classification and position features based on the improved position-aware attention aggregation feature data to obtain spatial classification and positioning feature data;

[0060] Step S44: constructing a bidirectional gated cyclic subnet, specifically executing step S33 to construct a bidirectional gated cyclic subnet, and optimizing the timing features according to the fault hidden danger prediction reference data and the feature data in the relay hidden danger detection optimization data set, constructing the bidirectional gated cyclic subnet, and performing feature output fusion of the convolution feature extraction subnet and the bidirectional gated cyclic subnet by applying network fusion, and obtaining classification prediction output and positioning prediction output by constructing classification calculation and positioning calculation;

[0061] Step S45: Fault classification and location model training, specifically, by constructing the feature-expanded location perception graph, improving the location perception attention aggregation, constructing the convolutional feature extraction subnet, and constructing the bidirectional gated recurrent subnet, to perform fault classification and location model training to obtain the fault classification and location model Model C ;

[0062] Step S46: Fault classification and location, specifically using the fault classification and location model Model C , performing fault classification and positioning according to the fault hidden danger prediction reference data and the relay hidden danger detection optimization data set to obtain fault classification and positioning reference data;

[0063] The fault classification and positioning reference data specifically include specific fault type data of the potential fault, probability data corresponding to the specific type of the potential fault, location prediction data of the potential fault, and specific time prediction data of the potential fault.

[0064] Further, in step S5, the relay protection hidden danger detection is used to perform comprehensive relay protection hidden danger detection by integrating the fault hidden danger trend frequency prediction and the fault hidden danger classification and positioning prediction. Specifically, the fault hidden danger prediction reference data and the fault classification and positioning reference data are obtained through the fault hidden danger prediction and the fault classification and positioning reference data, and the relay protection hidden danger comprehensive detection data is obtained by integrating the fault hidden danger prediction reference data and the fault classification and positioning reference data, and intelligent relay protection hidden danger detection is performed based on the relay protection hidden danger comprehensive detection data.

[0065] The present invention provides an intelligent relay protection hidden danger detection system, comprising a data acquisition module, a data preprocessing module, a fault hidden danger prediction module, a fault classification and positioning module and a relay protection hidden danger detection module;

[0066] The data acquisition module is used for data acquisition, and obtains the original data set of relay protection hidden danger detection through data acquisition, and sends the original data set of relay protection hidden danger detection to the data preprocessing module;

[0067] The data preprocessing module is used for data preprocessing, and obtains a relay hidden danger detection optimization data set through data preprocessing, and sends the relay hidden danger detection optimization data set to the fault hidden danger prediction module and the fault classification and positioning module;

[0068] The fault hidden danger prediction module is used for fault hidden danger prediction, obtains fault hidden danger prediction reference data through fault hidden danger prediction, and sends the fault hidden danger prediction reference data to the fault classification and positioning module and the relay protection hidden danger detection module;

[0069] The fault classification and positioning module is used for fault classification and positioning, obtains fault classification and positioning reference data through fault classification and positioning, and sends the fault classification and positioning reference data to the relay protection hidden danger detection module;

[0070] The relay protection hidden danger detection module is used for relay protection hidden danger detection. Through the relay protection hidden danger detection, comprehensive detection data of relay protection hidden dangers are obtained, and intelligent relay protection hidden danger detection is performed based on the comprehensive detection data of relay protection hidden dangers.

[0071] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0072] (1) In the existing intelligent relay protection hidden danger detection methods, there are two ideas: manual detection and automatic detection. The automatic detection of relay protection hidden dangers involves complex analysis and evaluation of fault types, time series and fault locations. Although the existing methods are relatively mature in the fault classification detection part, they lack the combination of time and location in spatiotemporal detection. At the same time, the existing methods are difficult to provide prediction and detection support from the three dimensions of time, space and fault type, resulting in the technical problem that the degree of intelligence needs to be improved. This solution creatively adopts a fine-grained relay protection hidden danger detection method from hidden danger detection to classification and positioning. By first detecting and predicting the probability of fault occurrence and the trend change of the probability of occurrence, it provides support for intelligent detection needs, and further confirms the type and spatiotemporal location of the fault by performing classification and positioning, thereby improving the overall efficiency, intelligence and detection dimension of relay protection hidden danger detection;

[0073] (2) In the existing fault hidden danger prediction methods, the existing methods often adopt single-stage hidden danger detection, and the results of their intelligent output also focus on the specific types of hidden dangers. However, this method is actually easy to ignore the various complex factors that cause hidden dangers. Because the hidden dangers that may exist in the operation process of relay protection equipment are actually complex and changeable, involving not only electrical and physical characteristics, but also environmental changes, and even the influence of maintenance history and fault history. Therefore, the single-stage prediction of traditional methods is difficult to meet the technical problem of fine-grained detection needs. This scheme creatively adopts a time-series convolutional bidirectional gated recurrent network combined with feature reconstruction optimization to predict fault hidden dangers. Through feature reconstruction optimization, the connection between various complex features is enhanced, and the long-term and short-term relay protection feature analysis is realized through time-series convolution and bidirectional gated recurrent networks. While improving the prediction accuracy of fault hidden danger prediction, the fault hidden danger prediction itself is used as a pre-step, which is conducive to subsequent more precise predictions. At the same time, it can also be used as a separate functional module for macro-level regulation and assistance, thereby improving the intelligent performance of relay protection hidden danger detection;

[0074] (3) In the existing fault classification and positioning, the traditional methods have difficulty in performing multi-output prediction and detection of time, space and type at the same time. At the same time, the traditional single-stage detection method is difficult to support multi-dimensional classification and positioning output, and the traditional multi-stage detection method is difficult to achieve specific detection in the time and space dimensions. This scheme creatively uses a convolutional bidirectional gated recurrent fusion network combined with a position perception map to perform fault classification and positioning. The sensor encoding in the data acquisition stage is applied to position perception mapping, and the encoding and processing of features are improved. Finally, the convolutional bidirectional gated recurrent fusion network is used to realize spatial and temporal feature processing, which improves the ability of fault classification and positioning to output multi-dimensional information. At the same time, it also provides an exploration and practical experience for the intelligent direction of relay protection hidden danger detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 A schematic diagram of a flow chart of an intelligent relay protection hidden danger detection method provided by the present invention;

[0076] Figure 2 A schematic diagram of an intelligent relay protection hidden danger detection system provided by the present invention;

[0077] Figure 3 This is a schematic diagram of the process of data preprocessing in step S2;

[0078] Figure 4 This is a schematic diagram of the process of predicting potential fault hazards in step S3;

[0079] Figure 5 This is a flow chart of fault classification and location in step S4.

[0080] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0081] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0082] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0083] Example 1, see Figure 1 The present invention provides an intelligent relay protection hidden danger detection method, which comprises the following steps:

[0084] Step S1: data collection;

[0085] Step S2: data preprocessing;

[0086] Step S3: Fault potential prediction;

[0087] Step S4: fault classification and location;

[0088] Step S5: Relay protection hidden danger detection.

[0089] By performing the above operations, in the existing intelligent relay protection hidden danger detection methods, there are two ideas: manual detection and automatic detection. The automatic detection of relay protection hidden dangers involves complex analysis and evaluation of fault types, time series and fault locations. Although the existing methods are relatively mature in the technology of fault classification detection, they lack the combination of time and space detection combined with time and positioning. At the same time, it is difficult for the existing methods to provide predictive detection support from the three dimensions of time, space and fault type, which leads to the technical problem that the degree of intelligence needs to be improved. This solution creatively adopts a fine-grained relay protection hidden danger detection method of hidden danger detection and classification positioning. By first detecting and predicting the probability of fault occurrence and the trend change of the probability of occurrence, it provides support for intelligent detection needs, and confirms the type and time and space location of the fault by further performing classification positioning, thereby improving the overall efficiency, intelligence and detection dimension of relay protection hidden danger detection.

[0090] Example 2, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment. In step S1, the data collection is used to collect the original data set required for intelligent relay protection hidden danger detection, specifically, the original data set for relay protection hidden danger detection is obtained by collecting data during the operation of the relay protection equipment;

[0091] The relay protection hidden danger detection original data set specifically includes electrical data, relay action status data, environmental data, equipment operation history data and fault alarm signal data;

[0092] Wherein, the electrical data and environmental data are collected by using a distributed sensor network;

[0093] Table 1 is a data sample table of the original data set of relay protection hidden danger detection, as shown in the table, the electrical data specifically includes current data, voltage data and sensor position data;

[0094] The relay action state data specifically includes relay switch state data and relay operation time data;

[0095] The environmental data specifically includes temperature data, humidity data and sensor location data;

[0096] The equipment operation history data specifically includes equipment operation time data, historical fault record data and historical maintenance record data;

[0097] Table 1 Sample data of the original data set for relay protection hidden danger detection

[0098]

[0099] The fault types specifically include overload fault, short circuit fault, ground fault and voltage loss fault;

[0100] The maintenance types include cleaning maintenance, overhaul maintenance and equipment replacement maintenance;

[0101] The alarm types specifically include overload alarm, short circuit alarm, sister alarm and equipment failure alarm.

[0102] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the data preprocessing is used to optimize the original data, specifically, the data preprocessing is performed on the original data set of relay protection hidden danger detection to obtain the relay hidden danger detection optimized data set, including the following steps:

[0103] Step S21: data denoising, specifically, applying a filtering algorithm to the time series data in the original data set of relay protection hidden danger detection, performing a data smoothing and denoising operation, and obtaining smoothed and optimized data;

[0104] The filtering algorithm specifically uses Kalman filtering and low-pass filtering in sequence;

[0105] Step S22: data cleaning, specifically processing missing values ​​and outliers on the smoothed optimized data, specifically using interpolation to process missing data, and identifying and processing outliers through data distribution analysis to obtain a cleaned optimized data set;

[0106] The interpolation method specifically adopts a polynomial interpolation method;

[0107] Step S23: data standardization, specifically, performing standardization and normalization operations on the cleaning optimization data set in sequence, wherein the standardization is used to process electrical data, and the normalization is used to process environmental data, to obtain a standard optimization data set;

[0108] The standardization specifically adopts a Z-score standardization operation;

[0109] The normalization specifically adopts a maximum and minimum normalization operation;

[0110] Step S24: time feature optimization, specifically, performing a sliding window division operation on the time series data in the standard optimization data set, and obtaining time statistical features and periodic frequency features through time series alignment and time series conversion operations, and obtaining time feature optimization data by integrating the time statistical features and periodic frequency features;

[0111] Step S25: discrete data encoding, specifically, performing balanced optimization of category data by data oversampling, and performing label encoding operation on the balanced optimized discrete category data in the standard optimization data set to obtain discrete encoding feature data;

[0112] Step S26: integrating feature extraction, specifically the time feature optimization and the discrete data encoding, performing statistical feature extraction and frequency domain feature extraction to obtain a comprehensive feature data set;

[0113] Step S27: data optimization processing, specifically, performing data optimization processing through the data denoising, data cleaning, data standardization and integrated feature extraction to obtain a relay hidden danger detection optimization data set;

[0114] The relay hidden danger detection optimization data set specifically includes electrical characteristic data, relay action state characteristic data, environmental characteristic data, equipment operation history characteristic data and fault alarm signal characteristic data;

[0115] The electrical characteristic data, relay action state characteristic data and equipment operation history characteristic data are used to perform fault hidden danger prediction; the electrical characteristic data, environmental characteristic data and fault alarm signal characteristic data are used to perform fault classification and positioning.

[0116] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the fault hidden danger prediction is used to predict the frequency and trend of fault hidden dangers in the relay protection process. Specifically, based on the relay hidden danger detection optimization data set, a temporal convolutional bidirectional gated recurrent network combined with feature reconstruction optimization is used to predict fault hidden dangers and obtain fault hidden danger prediction reference data;

[0117] The temporal bidirectional gated recurrent network combined with feature reconstruction optimization specifically includes a feature correlation optimization improvement operator, a temporal convolutional subnet and a bidirectional gated recurrent subnet;

[0118] The feature correlation optimization improvement operator is used to optimize the correlation between various features;

[0119] The time series convolution subnet is used to process short-term time series data and provide short-term probability prediction;

[0120] The bidirectional gated recurrent subnetwork is used to process long-term time series data and provide trend probability prediction;

[0121] The step of using a temporal convolutional bidirectional gated recurrent network combined with feature reconstruction optimization to predict potential fault hazards and obtain reference data for potential fault hazard prediction includes:

[0122] Step S31: feature correlation optimization, specifically, optimizing the feature data correlation in the relay hidden danger detection optimization data set by feature correlation expansion and reconstruction, and obtaining reconstructed correlation optimization feature data, specifically including the following steps:

[0123] Step S311: Calculate the correlation between the fault hidden danger prediction characteristics, which is used to calculate the correlation between the electrical characteristics and the relay state characteristics. The calculation formula is:

[0124] ;

[0125] In the formula, is the fault potential prediction feature correlation, which is used to represent the combined feature data The correlation between the target feature y and the fault potential prediction target feature y, x1 is the first feature data, which is used to represent the electrical feature data, and x2 is the second feature data, which is used to represent the relay action state feature data. is an electrical correlation feature, used to indicate the correlation between the first feature data x1 and the fault potential prediction target feature y. is the feature combination correlation, which is used to indicate the correlation between the first feature data x1 and the second feature data x2. is a relay state correlation feature, which is used to represent the correlation between the second feature data x2 and the fault hidden danger prediction target feature y, wherein the correlation parameter r specifically refers to the Pearson coefficient;

[0126] Step S312: feature correlation extension, which is used to extend the correlation calculation to all features. The calculation formula is:

[0127] ;

[0128] In the formula, is the extended electrical correlation feature, is the electrical correlation feature, m is the total number of features, a is the feature index, It is the ath feature combination correlation, which is used to represent the first feature data x1 and the ath feature data x a The correlation between It is the ath feature combined with the correlation degree, used to represent the ath feature data x a The correlation between y and the target feature y for fault potential prediction;

[0129] Among them, the specific value of m is 3. When a is 3, x a Used to indicate the historical characteristics of equipment operation;

[0130] Step S313: Reconstruction of associated features, used to perform comprehensive weighting of all features, the calculation formula is:

[0131] ;

[0132] In the formula, R is the reconstructed association optimization feature data, is the extended electrical correlation feature, It is an electrical correlation feature;

[0133] Step S32: constructing a temporal convolutional subnetwork, specifically constructing a standard temporal convolutional subnetwork including causal convolution, dilated convolution and residual block, and optimizing temporal feature extraction by extended convolution to construct the temporal convolutional subnetwork;

[0134] Step S33: construct a bidirectional gated recurrent subnet, specifically, construct a standard gated recurrent unit including an update gate and a reset gate, and construct a bidirectional gated recurrent subnet by setting a forward and backward bidirectional structure. The calculation formula is:

[0135] ;

[0136] In the formula, h t is the output hidden state of the bidirectional gated recurrent subnetwork, BiGRU(·) is the bidirectional gated recurrent subnetwork function, X t It is the input feature data, which is used to represent the reconstructed associated optimized feature data after the features are extracted by the temporal convolutional subnet. is the forward gated recurrent unit output hidden state, is the output hidden state of the backward gated recurrent unit, and t is the time index;

[0137] Step S34: Fault hidden danger prediction model training, specifically, performing fault hidden danger prediction model training through the feature correlation optimization, the construction of the time series convolutional subnet and the construction of the bidirectional gated recurrent subnet to obtain the fault hidden danger prediction model Model P ;

[0138] Step S35: Fault potential prediction, specifically by using the fault potential prediction model Model P , based on the electrical characteristic data, relay action state characteristic data and equipment operation history characteristic data in the relay hidden danger detection optimization data set, fault hidden danger prediction is performed to obtain fault hidden danger prediction reference data;

[0139] The fault hidden danger prediction reference data specifically includes probability data for predicting whether the fault hidden danger will occur, trend data for predicting whether the fault hidden danger will occur, and window time data for fault hidden danger warning.

[0140] By performing the above operations, in the existing fault hidden danger prediction methods, there is a problem that the existing methods often use single-stage hidden danger detection, and the results of its intelligent output also focus on the specific types of hidden dangers. However, this method is actually easy to ignore the various complex factors that cause hidden dangers, because the hidden dangers that may exist in the operation process of relay protection equipment are actually complex and changeable, involving both electrical and physical characteristics, and may also involve environmental changes, and even the influence of maintenance history and fault history. Therefore, the single-stage prediction of traditional methods is difficult to meet the technical problem of fine-grained detection needs. This scheme creatively uses a time-series convolutional bidirectional gated recurrent network combined with feature reconstruction optimization to predict fault hidden dangers. Through feature reconstruction optimization, the connection between various complex features is enhanced, and the long-term and short-term relay protection feature analysis is realized through time-series convolution and bidirectional gated recurrent networks. While improving the prediction accuracy of fault hidden danger prediction, the fault hidden danger prediction itself is used as a pre-step, which is conducive to subsequent more refined predictions. At the same time, it can also be used as a separate functional module for macro-level regulation and assistance, thereby improving the intelligent performance of relay protection hidden danger detection.

[0141] Example 5, see Figure 1 , Figure 2 and Figure 5 This embodiment is based on the above embodiment. In step S4, the fault classification and positioning is used to predict the specific type and location of the fault hidden danger in the relay protection process. Specifically, based on the fault hidden danger prediction reference data and the relay hidden danger detection optimization data set, a convolutional bidirectional gated recurrent fusion network combined with a position perception graph is used to perform fault classification and positioning to obtain fault classification and positioning reference data;

[0142] The convolutional bidirectional gated recurrent fusion network combined with the position-aware graph specifically includes a feature-extended position-aware graph, an improved position-aware attention aggregation operator, a convolutional feature extraction subnet, and a bidirectional gated recurrent subnet;

[0143] The feature-expanded location perception map is used to enhance the location perception capability of relay fault hazards;

[0144] The improved location-aware attention aggregation operator is used to aggregate location-aware features and perform weighted aggregation;

[0145] The convolutional feature extraction subnet is used to extract classification features and position features from multidimensional features;

[0146] The bidirectional gated recurrent subnetwork is used to optimize the potential temporal features in the classification and localization process;

[0147] The step of using a convolutional bidirectional gated recurrent fusion network combined with a position perception graph to perform fault classification and location, and obtaining fault classification and location reference data, includes:

[0148] Step S41: constructing a feature-extended position-aware graph, specifically by constructing an initial position-aware mapping structure through node parameters and edge parameters, and obtaining a feature-extended position-aware graph by applying feature extension and position-aware attention initial calculation;

[0149] The feature expansion specifically refers to obtaining high-dimensional feature data by constructing a shared parameter matrix to perform feature high-dimensional expansion;

[0150] The calculation formula for the initial calculation of the position-aware attention is:

[0151] ;

[0152] In the formula, is the initial value of position-aware attention, which is used to represent the attention relationship between the first node parameter i and the second node parameter j. i is the first node parameter, which is used to represent the environmental feature data in the relay hidden danger detection optimization data set. j is the second node parameter, which is used to represent the fault alarm signal characteristics. T is the transposed matrix of the weight values ​​of the position-aware attention mechanism, W is the shared parameter matrix, and h i is the characteristic data corresponding to the first node parameter i, h j is the characteristic data corresponding to the second node parameter j, is the relationship matrix between the first node parameter i and the second node parameter j, specifically represented by the Euclidean distance;

[0153] Step S42: improving position-aware attention aggregation, specifically by calculating a normalized attention coefficient and performing combined activation through a nonlinear function to calculate an activation enhancement node attention coefficient, and performing position-aware feature aggregation and multi-head attention mechanism optimization on the activation enhancement node attention coefficient to obtain improved position-aware attention aggregation feature data;

[0154] The normalized attention coefficient is specifically obtained by calculating the position-aware attention initial value using a softmax classifier function;

[0155] The activation enhancement node attention coefficient is specifically obtained by activating the position-aware attention initial value using a LeakyReLU activation function and calculating the normalized attention coefficient;

[0156] The perception feature aggregation is specifically to perform weighted summation of neighbor node features according to the activation enhancement node attention coefficient to obtain perception feature aggregation update node feature data;

[0157] The multi-head attention mechanism optimization specifically introduces a multi-head self-attention mechanism to optimize the perception feature aggregation update node feature data to obtain improved position perception attention aggregation feature data;

[0158] Step S43: constructing a convolutional feature extraction subnet, specifically constructing a standard convolutional feature extraction subnet including a convolutional layer and a pooling layer, and extracting classification and position features based on the improved position-aware attention aggregation feature data to obtain spatial classification and positioning feature data;

[0159] Step S44: constructing a bidirectional gated cyclic subnet, specifically executing step S33 to construct a bidirectional gated cyclic subnet, and optimizing the timing features according to the fault hidden danger prediction reference data and the feature data in the relay hidden danger detection optimization data set, constructing the bidirectional gated cyclic subnet, and performing feature output fusion of the convolution feature extraction subnet and the bidirectional gated cyclic subnet by applying network fusion, and obtaining classification prediction output and positioning prediction output by constructing classification calculation and positioning calculation;

[0160] The calculation formula of the feature output fusion is:

[0161] ;

[0162] Where F is the fused feature output, is the S-type activation function, W F is the fusion weight, X C is the spatial classification and positioning feature data, H BG is the optimized timing feature extracted by the bidirectional gated recurrent subnetwork, b F is the fusion bias term;

[0163] The calculation formula for the classification calculation is:

[0164] Y class =softmax(F);

[0165] Where Y class is the classification prediction output, softmax(·) is the softmax classifier function, and F is the fused feature output;

[0166] The calculation formula for the positioning calculation is:

[0167] Y location =Argmax(F);

[0168] Where Y location is the classification prediction output, Argmax(·) is the maximum confidence function of the position, and F is the fused feature output;

[0169] Step S45: Fault classification and location model training, specifically, by constructing the feature-expanded location perception graph, improving the location perception attention aggregation, constructing the convolutional feature extraction subnet, and constructing the bidirectional gated recurrent subnet, to perform fault classification and location model training to obtain the fault classification and location model Model C ;

[0170] Step S46: Fault classification and location, specifically using the fault classification and location model Model C , performing fault classification and positioning according to the fault hidden danger prediction reference data and the relay hidden danger detection optimization data set to obtain fault classification and positioning reference data;

[0171] The fault classification and positioning reference data specifically include specific fault type data of the potential fault, probability data corresponding to the specific type of the potential fault, location prediction data of the potential fault, and specific time prediction data of the potential fault.

[0172] By performing the above operations, in view of the technical problem that in the existing fault classification and positioning, the traditional methods have difficulty in simultaneously performing multi-output prediction and detection of time, space and type, and the traditional single-stage detection method is difficult to support multi-dimensional classification and positioning output, while the traditional multi-stage detection method is difficult to achieve specific detection in the time and space dimensions, this scheme creatively adopts a convolutional bidirectional gated recurrent fusion network combined with a position perception map to perform fault classification and positioning, applies sensor encoding in the data acquisition stage to position perception mapping, improves feature encoding and processing, and finally uses a convolutional bidirectional gated recurrent fusion network to realize spatial and temporal feature processing, thereby improving the ability of fault classification and positioning to output multi-dimensional information. At the same time, it also provides an exploration and practical experience for the intelligent direction of relay protection hidden danger detection.

[0173] Example 6, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S5, the relay protection hidden danger detection is used to comprehensively perform relay protection hidden danger detection by comprehensively predicting the trend frequency of fault hidden dangers and predicting the classification and positioning of fault hidden dangers. Specifically, the fault hidden danger prediction reference data and the fault classification and positioning reference data are obtained through the fault hidden danger prediction and the fault classification and positioning reference data, and the relay protection hidden danger comprehensive detection data is obtained by comprehensively analyzing the fault hidden danger prediction reference data and the fault classification and positioning reference data, and intelligent relay protection hidden danger detection is performed based on the relay protection hidden danger comprehensive detection data.

[0174] Embodiment 7, see Figure 1 and Figure 2, this embodiment is based on the above embodiment, and the present invention provides an intelligent relay protection hidden danger detection system, including a data acquisition module, a data preprocessing module, a fault hidden danger prediction module, a fault classification and positioning module and a relay protection hidden danger detection module;

[0175] The data acquisition module is used for data acquisition, and obtains the original data set of relay protection hidden danger detection through data acquisition, and sends the original data set of relay protection hidden danger detection to the data preprocessing module;

[0176] The data preprocessing module is used for data preprocessing, and obtains a relay hidden danger detection optimization data set through data preprocessing, and sends the relay hidden danger detection optimization data set to the fault hidden danger prediction module and the fault classification and positioning module;

[0177] The fault hidden danger prediction module is used for fault hidden danger prediction, obtains fault hidden danger prediction reference data through fault hidden danger prediction, and sends the fault hidden danger prediction reference data to the fault classification and positioning module and the relay protection hidden danger detection module;

[0178] The fault classification and positioning module is used for fault classification and positioning, obtains fault classification and positioning reference data through fault classification and positioning, and sends the fault classification and positioning reference data to the relay protection hidden danger detection module;

[0179] The relay protection hidden danger detection module is used for relay protection hidden danger detection. Through the relay protection hidden danger detection, comprehensive detection data of relay protection hidden dangers are obtained, and intelligent relay protection hidden danger detection is performed based on the comprehensive detection data of relay protection hidden dangers.

[0180] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0181] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0182] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent relay protection hidden danger detection method, characterized in that: The method comprises the following steps: Step S1: data collection, obtaining a relay protection hidden danger detection original data set by collecting relay protection equipment operation process data; the relay protection hidden danger detection original data set specifically includes electrical data, relay action state data, environmental data, equipment operation history data and fault alarm signal data; Step S2: data preprocessing, performing data preprocessing on the original data set for relay protection hidden danger detection to obtain an optimized data set for relay hidden danger detection; The relay hidden danger detection optimization data set specifically includes electrical characteristic data, relay action state characteristic data, environmental characteristic data, equipment operation history characteristic data and fault alarm signal characteristic data; The electrical characteristic data, relay action state characteristic data and equipment operation history characteristic data are used to perform fault hidden danger prediction; the electrical characteristic data, environmental characteristic data and fault alarm signal characteristic data are used to perform fault classification and positioning; Step S3: Fault hidden danger prediction is used to predict the frequency and trend of fault hidden dangers in the relay protection process. Specifically, based on the relay hidden danger detection optimization data set, a temporal convolutional bidirectional gated recurrent network combined with feature reconstruction optimization is used to predict fault hidden dangers and obtain fault hidden danger prediction reference data, which specifically includes the following steps: Step S31: feature correlation optimization; Step S32: constructing a time series convolutional subnet; Step S33: constructing a bidirectional gated recurrent subnet; Step S34: training a fault potential prediction model; Step S35: predicting fault potential; In step S31, the feature correlation degree optimization specifically includes the following steps: Step S311: Calculate the correlation between the fault hidden danger prediction characteristics, which is used to calculate the correlation between the electrical characteristics and the relay state characteristics. The calculation formula is: ； In the formula, is the fault potential prediction feature correlation, which is used to represent the combined feature data The correlation between the target feature y and the fault potential prediction target feature y, x1 is the first feature data, which is used to represent the electrical feature data, and x2 is the second feature data, which is used to represent the relay action state feature data. is an electrical correlation feature, used to indicate the correlation between the first feature data x1 and the fault potential prediction target feature y. is the feature combination correlation, which is used to indicate the correlation between the first feature data x1 and the second feature data x2. is a relay state correlation feature, which is used to represent the correlation between the second feature data x2 and the fault hidden danger prediction target feature y, wherein the correlation parameter r specifically refers to the Pearson coefficient; Step S312: feature correlation extension, which is used to extend the correlation calculation to all features. The calculation formula is: ； In the formula, is the extended electrical correlation feature, is the electrical correlation feature, m is the total number of features, a is the feature index, It is the ath feature combination correlation, which is used to represent the first feature data x1 and the ath feature data x a The correlation between It is the ath feature combined with the correlation degree, used to represent the ath feature data x a The correlation between y and the target feature y for fault potential prediction; Among them, the specific value of m is 3. When a is 3, x a Used to indicate the historical characteristics of equipment operation; Step S313: Reconstruction of associated features, used for comprehensive weighting of all features, the calculation formula is: ； In the formula, R is the reconstructed association optimization feature data, is the extended electrical correlation feature, It is an electrical correlation feature; Step S4: Fault classification and positioning, which is used to predict the specific type and location of the potential fault in the relay protection process. Specifically, based on the potential fault prediction reference data and the potential relay fault detection optimization data set, a convolutional bidirectional gated recurrent fusion network combined with a position perception graph is used to perform fault classification and positioning to obtain the fault classification and positioning reference data, which specifically includes the following steps: Step S41: construct feature-extended location-aware graph; Step S42: improve location-aware attention aggregation; Step S43: construct convolutional feature extraction subnet; Step S44: construct bidirectional gated recurrent subnet; Step S45: fault classification and localization model training; Step S46: fault classification and localization Step S5: Relay protection hidden danger detection, obtaining relay protection hidden danger comprehensive detection data, and performing intelligent relay protection hidden danger detection based on the relay protection hidden danger comprehensive detection data.

2. The intelligent relay protection hidden danger detection method according to claim 1 is characterized in that: In step S1, the electrical data specifically includes current data, voltage data and sensor position data; The relay action state data specifically includes relay switch state data and relay operation time data; The environmental data specifically includes temperature data, humidity data and sensor location data; The equipment operation history data specifically includes equipment operation time data, historical fault record data and historical maintenance record data.

3. The intelligent relay protection hidden danger detection method according to claim 2 is characterized in that: In step S2, the data preprocessing is used to optimize the original data, specifically, data preprocessing is performed on the original data set of relay protection hidden danger detection to obtain an optimized data set of relay hidden danger detection, including the following steps: Step S21: data denoising, specifically, applying a filtering algorithm to the time series data in the original data set of relay protection hidden danger detection, performing a data smoothing and denoising operation, and obtaining smoothed and optimized data; Step S22: data cleaning, specifically processing missing values ​​and outliers on the smoothed optimized data, specifically using interpolation to process missing data, and identifying and processing outliers through data distribution analysis to obtain a cleaned optimized data set; Step S23: data standardization, specifically, performing standardization and normalization operations on the cleaning optimization data set in sequence, wherein the standardization is used to process electrical data, and the normalization is used to process environmental data, to obtain a standard optimization data set; Step S24: time feature optimization, specifically, performing a sliding window division operation on the time series data in the standard optimization data set, and obtaining time statistical features and periodic frequency features through time series alignment and time series conversion operations, and obtaining time feature optimization data by integrating the time statistical features and periodic frequency features; Step S25: discrete data encoding, specifically, performing balanced optimization of category data by data oversampling, and performing label encoding operation on the balanced optimized discrete category data in the standard optimization data set to obtain discrete encoding feature data; Step S26: integrating feature extraction, specifically the time feature optimization and the discrete data encoding, performing statistical feature extraction and frequency domain feature extraction to obtain a comprehensive feature data set; Step S27: data optimization processing, specifically, performing data optimization processing through the data denoising, the data cleaning, the data standardization and the integrated feature extraction to obtain a relay hidden danger detection optimized data set.

4. The intelligent relay protection hidden danger detection method according to claim 3 is characterized in that: In step S3, the temporal bidirectional gated recurrent network combined with feature reconstruction optimization specifically includes a feature correlation optimization improvement operator, a temporal convolutional subnet and a bidirectional gated recurrent subnet; The feature correlation optimization improvement operator is used to optimize the correlation between various features; The time series convolution subnet is used to process short-term time series data and provide short-term probability prediction; The bidirectional gated recurrent subnetwork is used to process long-term time series data and provide trend probability prediction; The step of using a temporal convolutional bidirectional gated recurrent network combined with feature reconstruction optimization to predict potential fault hazards and obtain reference data for potential fault hazard prediction includes: Step S31: feature correlation optimization, specifically, optimizing the correlation of feature data in the relay hidden danger detection optimization data set by feature correlation expansion and reconstruction, to obtain reconstructed correlation optimization feature data; Step S32: constructing a temporal convolutional subnetwork, specifically constructing a standard temporal convolutional subnetwork including causal convolution, dilated convolution and residual block, and optimizing temporal feature extraction by extended convolution to construct the temporal convolutional subnetwork; Step S33: constructing a bidirectional gated recurrent subnetwork, specifically constructing a standard gated recurrent unit including an update gate and a reset gate, and constructing a bidirectional gated recurrent subnetwork by setting a forward and backward bidirectional structure; Step S34: Fault hidden danger prediction model training, specifically, performing fault hidden danger prediction model training through the feature correlation optimization, the construction of the time series convolutional subnet and the construction of the bidirectional gated recurrent subnet to obtain the fault hidden danger prediction model Model P ; Step S35: Fault potential prediction, specifically by using the fault potential prediction model Model P , based on the electrical characteristic data, relay action state characteristic data and equipment operation history characteristic data in the relay hidden danger detection optimization data set, fault hidden danger prediction is performed to obtain fault hidden danger prediction reference data; The fault hidden danger prediction reference data specifically includes probability data for predicting whether the fault hidden danger will occur, trend data for predicting whether the fault hidden danger will occur, and window time data for fault hidden danger warning.

5. The intelligent relay protection hidden danger detection method according to claim 4 is characterized in that: In step S4, the convolutional bidirectional gated recurrent fusion network combined with the position-aware graph specifically includes a feature-extended position-aware graph, an improved position-aware attention aggregation operator, a convolutional feature extraction subnet, and a bidirectional gated recurrent subnet; The feature-expanded location perception map is used to enhance the location perception capability of relay fault hazards; The improved location-aware attention aggregation operator is used to aggregate location-aware features and perform weighted aggregation; The convolutional feature extraction subnet is used to extract classification features and position features from multidimensional features; The bidirectional gated recurrent subnetwork is used to optimize the potential temporal features in the classification and localization process; The step of using a convolutional bidirectional gated recurrent fusion network combined with a position perception graph to perform fault classification and location, and obtaining fault classification and location reference data, includes: Step S41: constructing a feature-extended position-aware graph, specifically by constructing an initial position-aware mapping structure through node parameters and edge parameters, and obtaining a feature-extended position-aware graph by applying feature extension and position-aware attention initial calculation; Step S42: improving position-aware attention aggregation, specifically by calculating a normalized attention coefficient and performing combined activation through a nonlinear function to calculate an activation enhancement node attention coefficient, and performing position-aware feature aggregation and multi-head attention mechanism optimization on the activation enhancement node attention coefficient to obtain improved position-aware attention aggregation feature data; Step S43: constructing a convolutional feature extraction subnet, specifically constructing a standard convolutional feature extraction subnet including a convolutional layer and a pooling layer, and extracting classification and position features based on the improved position-aware attention aggregation feature data to obtain spatial classification and positioning feature data; Step S44: constructing a bidirectional gated cyclic subnet, specifically executing step S33 to construct a bidirectional gated cyclic subnet, and optimizing the timing features according to the fault hidden danger prediction reference data and the feature data in the relay hidden danger detection optimization data set, constructing the bidirectional gated cyclic subnet, and performing feature output fusion of the convolution feature extraction subnet and the bidirectional gated cyclic subnet by applying network fusion, and obtaining classification prediction output and positioning prediction output by constructing classification calculation and positioning calculation; Step S45: Fault classification and location model training, specifically, by constructing the feature-expanded location perception graph, improving the location perception attention aggregation, constructing the convolutional feature extraction subnet, and constructing the bidirectional gated recurrent subnet, to perform fault classification and location model training to obtain the fault classification and location model Model C ; Step S46: Fault classification and location, specifically using the fault classification and location model Model C , performing fault classification and positioning according to the fault hidden danger prediction reference data and the relay hidden danger detection optimization data set to obtain fault classification and positioning reference data; The fault classification and positioning reference data specifically include specific fault type data of the potential fault, probability data corresponding to the specific type of the potential fault, location prediction data of the potential fault, and specific time prediction data of the potential fault.

6. The intelligent relay protection hidden danger detection method according to claim 5 is characterized in that: In step S5, the relay protection hidden danger detection is used to perform comprehensive relay protection hidden danger detection by integrating the fault hidden danger trend frequency prediction and the fault hidden danger classification and positioning prediction. Specifically, the fault hidden danger prediction reference data and the fault classification and positioning reference data are obtained through the fault hidden danger prediction and the fault classification and positioning reference data, and the relay protection hidden danger comprehensive detection data is obtained by integrating the fault hidden danger prediction reference data and the fault classification and positioning reference data, and intelligent relay protection hidden danger detection is performed based on the relay protection hidden danger comprehensive detection data.

7. An intelligent relay protection hidden danger detection system, used to implement an intelligent relay protection hidden danger detection method as described in any one of claims 1 to 6, characterized in that: It includes data acquisition module, data preprocessing module, fault hidden danger prediction module, fault classification and positioning module and relay protection hidden danger detection module; The data acquisition module is used for data acquisition, and obtains the original data set of relay protection hidden danger detection through data acquisition, and sends the original data set of relay protection hidden danger detection to the data preprocessing module; The data preprocessing module is used for data preprocessing, and obtains a relay hidden danger detection optimization data set through data preprocessing, and sends the relay hidden danger detection optimization data set to the fault hidden danger prediction module and the fault classification and positioning module; The fault hidden danger prediction module is used for fault hidden danger prediction, obtains fault hidden danger prediction reference data through fault hidden danger prediction, and sends the fault hidden danger prediction reference data to the fault classification and positioning module and the relay protection hidden danger detection module; The fault classification and positioning module is used for fault classification and positioning, obtains fault classification and positioning reference data through fault classification and positioning, and sends the fault classification and positioning reference data to the relay protection hidden danger detection module; The relay protection hidden danger detection module is used for relay protection hidden danger detection. Through the relay protection hidden danger detection, comprehensive detection data of relay protection hidden dangers are obtained, and intelligent relay protection hidden danger detection is performed based on the comprehensive detection data of relay protection hidden dangers.

Images