Method and system for realizing fault detection of high and low voltage power distribution cabinet based on intelligent AI

Abstract

The invention relates to the technical field of remote monitoring, and discloses an intelligent AI-based fault detection method and system for a high-low voltage power distribution cabinet, and the method comprises the steps: constructing an LSTM-Attention hybrid network of a multi-modal data flow, extracting the time sequence characteristics of the multi-modal data flow, extracting the non-stationary signal frequency domain characteristics of the multi-modal data flow, and carrying out the fault detection of the multi-modal data flow. The method comprises the following steps: constructing a fault knowledge graph of a high-low voltage power distribution cabinet, analyzing a topological correlation feature vector of the high-low voltage power distribution cabinet, training a pre-constructed power distribution cabinet fault global model to obtain a trained power distribution cabinet fault global model, and analyzing a fault probability matrix of the high-low voltage power distribution cabinet by using the trained power distribution cabinet fault global model. And establishing a digital twinborn body of the high-low voltage power distribution cabinet, analyzing a fault evolution path of the high-low voltage power distribution cabinet by using the digital twinborn body, and analyzing a fault detection report of the high-low voltage power distribution cabinet. The maintenance efficiency of the high-low voltage power distribution cabinet can be improved.

Description

Technical Field

[0001] The present invention relates to a method and system for fault detection of high and low voltage distribution cabinets based on intelligent AI, belonging to the technical field of remote monitoring. Background Art

[0002] The fault detection of high and low voltage distribution cabinets refers to the process of using various detection technologies and methods to monitor, diagnose, and analyze various possible faults that may occur during the operation of high and low voltage distribution cabinets in real-time or regularly. Its purpose is to timely detect potential or existing faults, ensure the safe and stable operation of the distribution cabinets, prevent accidents, and improve the reliability and efficiency of power supply.

[0003] The traditional fault detection of high and low voltage distribution cabinets mainly relies on manual regular inspections and automated systems based on simple threshold alarms. This method is difficult to monitor in real-time, has limited diagnostic capabilities for complex faults, and cannot predict the fault evolution path, making it impossible to timely identify distribution cabinet faults, thus reducing the maintenance efficiency of the distribution cabinets. Summary of the Invention

[0004] The present invention provides a method and system for fault detection of high and low voltage distribution cabinets based on intelligent AI, and its main purpose is to improve the maintenance efficiency of high and low voltage distribution cabinets.

[0005] To achieve the above object, a method for fault detection of high and low voltage distribution cabinets based on intelligent AI provided by the present invention includes:

[0006] Collect the electrical parameters, mechanical states, and thermal imaging data of the high and low voltage distribution cabinets, and construct a multi-modal data stream of the electrical parameters, mechanical states, and thermal imaging data;

[0007] Construct an LSTM-Attention hybrid network for the multi-modal data stream, extract the temporal features of the multi-modal data stream through the LSTM-Attention hybrid network, and use a preset wavelet packet transform function to extract the non-stationary signal frequency domain features of the multi-modal data stream;

[0008] Obtain the historical fault data of the high and low voltage distribution cabinets, construct a fault knowledge graph of the high and low voltage distribution cabinets according to the historical fault data, and analyze the topological correlation feature vectors of the high and low voltage distribution cabinets through the fault knowledge graph;

[0009] Train a pre-constructed global power distribution cabinet fault model using the timing characteristics, the non-stationary signal frequency domain characteristics, and the topological correlation feature vector to obtain a trained global power distribution cabinet fault model, and use the trained global power distribution cabinet fault model to analyze the fault probability matrix of the high and low voltage power distribution cabinets, where the global power distribution cabinet fault model refers to a model for analyzing the fault states of high and low voltage power distribution cabinets;

[0010] Establish a digital twin of the high and low voltage power distribution cabinets, analyze the fault evolution path of the high and low voltage power distribution cabinets using the digital twin based on the fault probability matrix, and analyze the fault detection report of the high and low voltage power distribution cabinets according to the fault evolution path.

[0011] Optionally, constructing the multi-modal data stream of the electrical parameters, mechanical states, and thermal imaging data includes:

[0012] Preprocess the electrical parameters, mechanical states, and thermal imaging data to obtain processed electrical parameters, processed mechanical states, and processed thermal imaging data;

[0013] Define the main time reference for the processed electrical parameters, processed mechanical states, and processed thermal imaging data;

[0014] Perform time synchronization calibration on the processed electrical parameters, processed mechanical states, and processed thermal imaging data according to the main time reference to obtain calibrated electrical parameters, calibrated mechanical states, and calibrated thermal imaging data;

[0015] Integrate the calibrated electrical parameters, calibrated mechanical states, and calibrated thermal imaging data into a preset data framework to obtain integrated multi-modal data;

[0016] Annotate the integrated multi-modal data to obtain the multi-modal data stream.

[0017] Optionally, constructing the LSTM-Attention hybrid network of the multi-modal data stream includes:

[0018] Establish multiple input channels for the multi-modal data stream;

[0019] Construct the LSTM layer of the multiple input channels;

[0020] Determine the Attention mechanism of the LSTM layer;

[0021] Construct the LSTM-Attention hybrid network of the multi-modal data stream through the multiple input channels, the LSTM layer, and the Attention mechanism.

[0022] Optionally, extracting the non-stationary signal frequency domain features of the multi-modal data stream by using a preset wavelet packet transform function includes:

[0023] Analyze the data stream characteristics of the multi-modal data stream;

[0024] Determine the function parameters of the wavelet packet transform function according to the data stream characteristics;

[0025] Decompose the multi-modal data stream through the function parameters to obtain decomposition coefficients;

[0026] Analyze the Pearson coefficient and instantaneous energy gradient of the multi-modal data stream according to the decomposition coefficients;

[0027] Calculate the band energy of the multi-modal data stream through the Pearson coefficient;

[0028] Calculate the band entropy value of the multi-modal data stream according to the instantaneous energy gradient;

[0029] Determine the non-stationary signal frequency domain features of the multi-modal data stream based on the stream band energy and band entropy value.

[0030] Optionally, calculating the band energy of the multi-modal data stream through the Pearson coefficient includes:

[0031] Determine the cross-modal coupling term of the multi-modal data stream according to the Pearson coefficient;

[0032] Analyze the band variance of the multi-modal data stream based on the corresponding decomposition coefficients of the multi-modal data stream;

[0033] Determine the dynamic weight coefficient of the multi-modal data stream according to the band variance;

[0034] Calculate the band energy of the multi-modal data stream based on the cross-modal coupling term and the dynamic weight coefficient.

[0035] Optionally, constructing the fault knowledge graph of the high-voltage and low-voltage power distribution cabinets according to the historical fault data includes:

[0036] Preprocess the historical fault data to obtain preprocessed fault data;

[0037] Use a preset BERT+BiLSTM model to extract the data entities of the preprocessed fault data;

[0038] Analyze the entity relationships of the data entities, where the entity relationships include static relationships and dynamic relationships;

[0039] Define the semantic network of the historical fault data;

[0040] Based on the entity relationship and the data entity, establish the fault knowledge graph of the high and low voltage power distribution cabinet by using the semantic network.

[0041] Optionally, the analysis of the topological association feature vector of the high and low voltage power distribution cabinet through the fault knowledge graph includes:

[0042] Mark the graph nodes corresponding to the fault knowledge graph as feature vectors;

[0043] Analyze the topological association information of the graph nodes;

[0044] Based on the topological association information, fuse the feature vectors to obtain the fused association feature vector;

[0045] Aggregate the fused association feature vector to obtain the topological association feature vector of the high and low voltage power distribution cabinet.

[0046] Optionally, the establishment of the digital twin of the high and low voltage power distribution cabinet includes:

[0047] Identify the physical structure parameters of the high and low voltage power distribution cabinet;

[0048] According to the physical structure parameters, establish the physical model of the high and low voltage power distribution cabinet;

[0049] Establish the operating electrical equation and operating thermal equation of the high and low voltage power distribution cabinet;

[0050] Integrate the operating electrical equation and operating thermal equation into the physical model to obtain the initial digital twin of the high and low voltage power distribution cabinet;

[0051] Simulate the working condition state of the initial digital twin;

[0052] Determine the digital twin of the high and low voltage power distribution cabinet through the working condition state.

[0053] Optionally, the analysis of the fault evolution path of the high and low voltage power distribution cabinet based on the fault probability matrix by using the digital twin includes:

[0054] Define the fault offset code and state vector of the high and low voltage power distribution cabinet;

[0055] According to the fault probability matrix and the fault offset code, establish the fault transition probability matrix of the high and low voltage power distribution cabinet;

[0056] Calculate the fault evolution path probability of the high and low voltage power distribution cabinet through the state vector and the fault transition probability matrix;

[0057] According to the probability of the fault evolution path, the digital twin is used to determine the fault evolution path of the high- and low-voltage power distribution cabinet.

[0058] To solve the above problems, the present invention also provides a fault detection system for high- and low-voltage power distribution cabinets based on intelligent AI. The system includes:

[0059] A multi-modal data stream construction module, configured to collect electrical parameters, mechanical states, and thermal imaging data of the high- and low-voltage power distribution cabinet, and construct a multi-modal data stream of the electrical parameters, mechanical states, and thermal imaging data;

[0060] A time-frequency domain feature extraction module, configured to construct an LSTM-Attention hybrid network for the multi-modal data stream, extract the time series features of the multi-modal data stream through the LSTM-Attention hybrid network, and extract the frequency domain features of the non-stationary signals of the multi-modal data stream by using a preset wavelet packet transform function;

[0061] A topological association feature extraction module, configured to obtain historical fault data of the high- and low-voltage power distribution cabinet, construct a fault knowledge graph of the high- and low-voltage power distribution cabinet according to the historical fault data, and analyze the topological association feature vector of the high- and low-voltage power distribution cabinet through the fault knowledge graph;

[0062] A fault probability matrix analysis module, configured to train a pre-constructed global power distribution cabinet fault model through the time series features, the frequency domain features of the non-stationary signals, and the topological association feature vector to obtain a trained global power distribution cabinet fault model, and analyze the fault probability matrix of the high- and low-voltage power distribution cabinet by using the trained global power distribution cabinet fault model, where the global power distribution cabinet fault model is a model for analyzing the fault state of the high- and low-voltage power distribution cabinet;

[0063] A fault evolution path analysis module, configured to establish a digital twin of the high- and low-voltage power distribution cabinet, analyze the fault evolution path of the high- and low-voltage power distribution cabinet by using the digital twin based on the fault probability matrix, and analyze the fault detection report of the high- and low-voltage power distribution cabinet according to the fault evolution path.

[0064] Compared with the problems described in the background art, firstly, by collecting the electrical parameters, mechanical states, and thermal imaging data of high- and low-voltage distribution cabinets and constructing a multi-modal data stream, the comprehensive monitoring of the operation status of the distribution cabinets is realized. By using the method of combining the LSTM-Attention hybrid network and the wavelet packet transform function, the time-series features and non-stationary signal frequency-domain features in the multi-modal data stream are effectively extracted, improving the accuracy of fault identification. Further, by constructing a fault knowledge graph of high- and low-voltage distribution cabinets, the topological correlation feature vectors of the distribution cabinets are deeply analyzed, providing a richer information dimension for fault diagnosis. Based on the time-series features, non-stationary signal frequency-domain features, and topological correlation feature vectors, the global fault model of the distribution cabinet obtained by training has higher prediction accuracy and generalization ability, and can accurately analyze the fault probability matrix of high- and low-voltage distribution cabinets. In addition, the established digital twin of high- and low-voltage distribution cabinets realizes the intuitive display and in-depth analysis of the fault evolution path, providing more intuitive and effective decision-making support for maintenance personnel. Finally, the fault detection report generated based on the fault evolution path not only details the fault information but also provides targeted repair and improvement suggestions, effectively guiding the maintenance work of high- and low-voltage distribution cabinets, reducing the fault incidence rate, and improving the operation efficiency and safety of the power system. Overall, this technical solution realizes the intelligent prediction, diagnosis, and analysis of high- and low-voltage distribution cabinet faults, providing a strong guarantee for the reliable operation of the power system. Therefore, the present invention can improve the maintenance efficiency of high- and low-voltage distribution cabinets. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 FIG. is a schematic flow chart of a method for detecting faults in high- and low-voltage distribution cabinets based on intelligent AI provided by an embodiment of the present invention;

[0066] Figure 2 FIG. is a schematic block diagram of a module for implementing the method for detecting faults in high- and low-voltage distribution cabinets based on intelligent AI provided by an embodiment of the present invention.

[0067] The implementation, functional features, and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0068] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0069] An embodiment of the present application provides a fault detection method for high-voltage and low-voltage power distribution cabinets based on intelligent AI. The execution subject of the fault detection method for high-voltage and low-voltage power distribution cabinets based on intelligent AI includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided in the embodiment of the present application. In other words, the fault detection method for high-voltage and low-voltage power distribution cabinets based on intelligent AI can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc.

[0070] Embodiment 1:

[0071] Referring to Figure 1 As shown, it is a schematic flow chart of a fault detection method for high-voltage and low-voltage power distribution cabinets based on intelligent AI provided by an embodiment of the present invention. In this embodiment, the fault detection method for high-voltage and low-voltage power distribution cabinets based on intelligent AI includes:

[0072] S1. Collect the electrical parameters, mechanical status, and thermal imaging data of the high-voltage and low-voltage power distribution cabinet, and construct a multi-modal data stream of the electrical parameters, mechanical status, and thermal imaging data.

[0073] It should be explained that the high-voltage and low-voltage power distribution cabinet refers to a device combination used in the power system to receive, distribute, control, protect, and measure high-voltage and low-voltage electric energy. The electrical parameters refer to various parameters related to the electrical performance of the power distribution cabinet, such as voltage, current, power, power factor, frequency, harmonic content, insulation resistance, and switch status, etc. The mechanical status refers to various information related to the mechanical performance and physical state of the power distribution cabinet, such as vibration, noise, temperature, humidity, switch position, wear degree, and lubrication status, etc. The thermal imaging data refers to data related to the surface temperature distribution of the power distribution cabinet obtained through thermal imaging technology, such as temperature distribution map, maximum temperature, temperature gradient, hot spot, and thermal trajectory, etc.

[0074] The present invention constructs a multi-modal data stream of the electrical parameters, mechanical status, and thermal imaging data, which can construct a time-synchronized calibrated multi-modal data stream, providing a solid data foundation for the intelligent monitoring, fault diagnosis, and preventive maintenance of high-voltage and low-voltage power distribution cabinets.

[0075] Specifically, the construction of the multi-modal data stream of the electrical parameters, mechanical status, and thermal imaging data includes:

[0076] Preprocess the electrical parameters, mechanical status, and thermal imaging data to obtain processed electrical parameters, processed mechanical status, and processed thermal imaging data;

[0077] Define the main time reference for the processed electrical parameters, processed mechanical status, and processed thermal imaging data;

[0078] According to the main time reference, perform time synchronization calibration on the processed electrical parameters, processed mechanical states, and processed thermal imaging data to obtain calibrated electrical parameters, calibrated mechanical states, and calibrated thermal imaging data;

[0079] Integrate the calibrated electrical parameters, calibrated mechanical states, and calibrated thermal imaging data into a preset data framework to obtain integrated multi-modal data;

[0080] Annotate the integrated multi-modal data to obtain the multi-modal data stream.

[0081] Among them, the processed electrical parameters refer to the preprocessed electrical parameter data, the processed mechanical states refer to the preprocessed mechanical state data, the processed thermal imaging data refer to the preprocessed thermal imaging data, the main time reference refers to a unified time standard for time synchronization, such as GPS time, NTP time, etc., the calibrated electrical parameters refer to the electrical parameter data after time synchronization according to the main time reference, the calibrated mechanical states refer to the mechanical state data after time synchronization according to the main time reference, the calibrated thermal imaging data refer to the thermal imaging data after time synchronization according to the main time reference, the integrated multi-modal data refer to the data set obtained by integrating the calibrated electrical parameters, mechanical states, and thermal imaging data into a preset data framework, and the multi-modal data stream refers to the continuous, real-time or batch data stream formed after annotating the integrated multi-modal data.

[0082] S2. Construct an LSTM-Attention hybrid network for the multi-modal data stream, extract the temporal features of the multi-modal data stream through the LSTM-Attention hybrid network, and use a preset wavelet packet transform function to extract the non-stationary signal frequency domain features of the multi-modal data stream.

[0083] The LSTM-Attention hybrid network constructed by the present invention for the multi-modal data stream can be used as the basis for subsequent temporal feature extraction.

[0084] Specifically, the construction of the LSTM-Attention hybrid network for the multi-modal data stream includes:

[0085] Establish multiple input channels for the multi-modal data stream;

[0086] Construct an LSTM layer for the multiple input channels;

[0087] Determine the Attention mechanism of the LSTM layer;

[0088] Construct the LSTM-Attention hybrid network of the multimodal data stream through the multi-level input channels, the LSTM layer, and the Attention mechanism.

[0089] Among them, the multi-level input channels refer to independent input paths designed for different types of data (such as electrical parameters, mechanical states, thermal imaging data), the LSTM layer refers to a special recurrent neural network (RNN) layer that can learn and remember long-term dependencies in time series data, the Attention mechanism refers to a technique that allows the model to assign different weights when processing information, enabling the model to focus on the most important parts of the information, and the LSTM-Attention hybrid network refers to a deep learning model that combines the LSTM layer and the Attention mechanism for processing multimodal time series data.

[0090] Optionally, the LSTM layer of the multi-level input channels can be implemented through a recurrent neural network (RNN).

[0091] The present invention can effectively utilize the LSTM-Attention hybrid network to extract the temporal features of the multimodal data stream by extracting the temporal features of the multimodal data stream through the LSTM-Attention hybrid network, and provide strong support for subsequent tasks such as fault detection and analysis. Among them, the temporal features refer to the feature representations extracted from the multimodal data stream through the LSTM-Attention hybrid network, which can reflect the changing laws and internal relationships of the data over time. Specifically, the temporal features are obtained by fusing the LSTM outputs after Attention weighting of the multimodal data stream.

[0092] The present invention can effectively utilize the wavelet packet transform function to extract the frequency domain features of the non-stationary signals in the multimodal data stream by using the preset wavelet packet transform function to extract the frequency domain features of the non-stationary signals in the multimodal data stream, and further provide support for subsequent tasks such as signal analysis and fault analysis.

[0093] Specifically, the utilization of the preset wavelet packet transform function to extract the frequency domain features of the non-stationary signals in the multimodal data stream includes:

[0094] Analyze the data stream characteristics of the multimodal data stream;

[0095] Determine the function parameters of the wavelet packet transform function according to the data stream characteristics;

[0096] Decompose the multimodal data stream through the function parameters to obtain decomposition coefficients;

[0097] Analyze the Pearson coefficient and instantaneous energy gradient of the multimodal data stream according to the decomposition coefficient;

[0098] Calculate the band energy of the multimodal data stream through the Pearson coefficient;

[0099] Calculate the band entropy value of the multimodal data stream according to the instantaneous energy gradient;

[0100] Based on the stream band energy and band entropy value, determine the frequency-domain characteristics of the non-stationary signal of the multimodal data stream.

[0101] Among them, the data stream characteristics refer to the statistical characteristics, change trends, periodicity, volatility and other characteristics of the multimodal data stream in the time series. The wavelet packet transform function refers to a mathematical tool for signal processing, which can provide time-frequency localization analysis and is applicable to the feature extraction of non-stationary signals. The function parameters refer to the parameters used in the wavelet packet transform, such as the selection of wavelet basis functions, decomposition levels, thresholds, etc. The decomposition coefficient refers to the coefficient obtained after wavelet packet transform, which represents the components of the signal at different frequency levels. The Pearson coefficient refers to the correlation degree of different modes of the multimodal data stream in the frequency band. The instantaneous energy gradient refers to the energy change rate of the signal at a certain moment or within a short time window. The stream band energy refers to the sum of the signal energy in each frequency band, which reflects the intensity of the signal in that frequency band. The band entropy value refers to an index that measures the complexity or uncertainty of the signal in each frequency band. The frequency-domain characteristics of the non-stationary signal refer to the set of characteristics obtained through frequency-domain analysis and capable of describing the characteristics of non-stationary signals.

[0102] Furthermore, the calculating the band energy of the multimodal data stream through the Pearson coefficient includes:

[0103] Determine the cross-modal coupling term of the multimodal data stream according to the Pearson coefficient;

[0104] Based on the corresponding decomposition coefficient of the multimodal data stream, analyze the band variance of the multimodal data stream;

[0105] Determine the dynamic weight coefficient of the multimodal data stream according to the band variance;

[0106] Based on the cross-modal coupling term and the dynamic weight coefficient, use the following formula to calculate the band energy of the multimodal data stream:

[0107]

[0108] Among them, represents the band energy of the m-th mode of the multimodal data stream in the k-th frequency band, αm，k Represents the dynamic weight coefficient, E m，k represents the fundamental frequency band energy of the mth mode of the multimodal data stream in the kth frequency band, θ m，n，k represents the interaction weight, cos(E m，k , E n，k ) represents the cross-modal coupling term, cos represents the cosine function, E n，k Represents the fundamental frequency band energy of the nth mode of the multimodal data stream in the kth frequency band.

[0109] Among them, the cross-modal coupling term refers to the quantification of the synergistic effect strength of different modal signals in the same frequency band, the frequency band variance refers to the energy fluctuation intensity of a single modal signal in a specific frequency band, the dynamic weight coefficient refers to the weight allocated according to the energy variance proportion of each mode in the frequency band, the frequency band energy refers to the comprehensive energy value of the multimodal signal in a specific frequency band, the basic frequency band energy refers to the traditional energy value of the single modal signal within the frequency band, and the interaction weight refers to the control of the contribution intensity of the cross-modal coupling term to the total energy.

[0110] S3. Obtain historical fault data of the high and low voltage distribution cabinets, construct a fault knowledge graph of the high and low voltage distribution cabinets based on the historical fault data, and analyze the topological association feature vectors of the high and low voltage distribution cabinets through the fault knowledge graph.

[0111] It should be explained that the historical fault data refers to the relevant records and information of all faults that occurred in high and low voltage distribution cabinets in the past period of time, including fault time, fault type, fault location, maintenance records and other information.

[0112] The present invention constructs a comprehensive, accurate and interactive high and low voltage distribution cabinet fault knowledge graph based on the historical fault data.

[0113] In detail, the construction of the fault knowledge graph of the high and low voltage distribution cabinet based on the historical fault data includes:

[0114] Preprocessing the historical fault data to obtain preprocessed fault data;

[0115] Extracting the data entity of the preprocessed fault data using a preset BERT+BiLSTM model;

[0116] Analyzing entity relationships of the data entities, wherein the entity relationships include static relationships and dynamic relationships;

[0117] defining a semantic network of the historical fault data;

[0118] Based on the entity relationship and the data entities, a fault knowledge graph of the high- and low-voltage power distribution cabinets is established using the semantic network.

[0119] Among them, the preprocessed fault data refers to the data obtained by cleaning, formatting, normalizing, etc. the original historical fault data to eliminate noise, fill in missing values, and unify the data format, so as to improve the data quality and prepare for subsequent model training and knowledge graph construction. The BERT+BiLSTM model is a text processing model formed by combining a language representation model that can capture deep semantic information in the text and a bidirectional recurrent neural network that processes long-term dependencies in sequence data, and is used to extract data entities from the preprocessed fault data. The data entities refer to the information units with specific meanings extracted from the preprocessed fault data, such as equipment names, fault types, fault times, fault causes, etc. The static relationship refers to the relationship that does not change with time in the fault knowledge graph, such as the connection relationship between devices, the corresponding relationship between fault types and fault causes, etc. The dynamic relationship refers to the relationship that changes with time in the fault knowledge graph, such as the evolution of device states, the change in the frequency of fault occurrence, etc. The semantic network is a form used to represent knowledge, consisting of nodes (representing data entities) and edges (representing the relationships between entities). The fault knowledge graph is a graphical structure established based on the semantic network and used to represent the fault knowledge of high- and low-voltage power distribution cabinets, including data entities, static relationships and dynamic relationships between entities, and can intuitively display fault information and support applications such as fault diagnosis and predictive maintenance.

[0120] By analyzing the topological association feature vectors of the high- and low-voltage power distribution cabinets through the fault knowledge graph, the present invention can effectively utilize the fault knowledge graph to analyze the topological association feature vectors of the high- and low-voltage power distribution cabinets, providing strong support for the intelligent maintenance and management of the power distribution system.

[0121] Specifically, the analysis of the topological association feature vectors of the high- and low-voltage power distribution cabinets through the fault knowledge graph includes:

[0122] Marking the graph nodes corresponding to the fault knowledge graph as feature vectors;

[0123] Analyzing the topological association information of the graph nodes;

[0124] Based on the topological association information, fusing the feature vectors to obtain fused association feature vectors;

[0125] Aggregating the fused association feature vectors to obtain the topological association feature vectors of the high- and low-voltage power distribution cabinets.

[0126] Among them, the graph node refers to the basic unit in the fault knowledge graph, representing specific entities such as power distribution cabinets, equipment, fault points, maintenance records, etc. The feature vector refers to a set of numerical values used to represent the graph node, and these numerical values describe the characteristics of the node, such as the attributes, status, or degree of association with other nodes of the node. The topological association information refers to the structural relationship and connection mode described between graph nodes, including direct connections, indirect connections, connection strengths, etc. The fused association feature vector refers to a new feature vector obtained by combining the original feature vector of the node and its topological association information. The topological association feature vector refers to the fused association feature vector after further aggregation processing, which centrally reflects the topological structure and association characteristics of the node in the graph.

[0127] Optionally, the fusion of the feature vectors to obtain the fused association feature vector can be achieved through a fusion function (such as average value, weighted average, etc.).

[0128] S4. Train the pre-constructed global model of the power distribution cabinet fault through the time series feature, the frequency domain feature of the non-stationary signal, and the topological association feature vector to obtain a trained global model of the power distribution cabinet fault, and use the trained global model of the power distribution cabinet fault to analyze the fault probability matrix of the high and low voltage power distribution cabinets. Among them, the global model of the power distribution cabinet fault refers to a model used to analyze the fault state of the high and low voltage power distribution cabinets.

[0129] In the present invention, the pre-constructed global model of the power distribution cabinet fault is trained through the time series feature, the frequency domain feature of the non-stationary signal, and the topological association feature vector to obtain a trained global model of the power distribution cabinet fault, and a trained global model capable of accurately analyzing and predicting the faults of high and low voltage power distribution cabinets can be obtained. Among them, the global model of the power distribution cabinet fault refers to a model used to analyze the fault state of the high and low voltage power distribution cabinets. Specifically, the global model of the power distribution cabinet fault can be trained through a neural network.

[0130] The present invention uses the trained global model of the power distribution cabinet fault to analyze the fault probability matrix of the high and low voltage power distribution cabinets, improving the accuracy of subsequent fault evolution. Among them, the fault probability matrix refers to a matrix used to represent the probabilities of various fault types occurring in different components or units of a device or system.

[0131] Exemplarily, an example of the fault probability matrix:

[0132]

[0133] In the example of the fault probability matrix:

[0134] The probability of component A having a short circuit fault is 2%, the probability of an overload fault is 15%, and the probability of insulation damage is 3%.

[0135] The probability of component B having a short - circuit fault is 5%, the probability of an overload fault is 10%, and the probability of insulation damage is 20%.

[0136] The probability of component C having a short - circuit fault is 1%, the probability of an overload fault is 5%, and the probability of insulation damage is 15%.

[0137] S5. Establish a digital twin of the high - and low - voltage power distribution cabinet. Based on the fault probability matrix, use the digital twin to analyze the fault evolution path of the high - and low - voltage power distribution cabinet. According to the fault evolution path, analyze the fault detection report of the high - and low - voltage power distribution cabinet.

[0138] The present invention establishes a digital twin of the high - and low - voltage power distribution cabinet to improve the visibility of fault analysis, thereby increasing the timeliness of fault identification.

[0139] Specifically, the establishment of the digital twin of the high - and low - voltage power distribution cabinet includes:

[0140] Identify the physical structure parameters of the high - and low - voltage power distribution cabinet;

[0141] According to the physical structure parameters, establish a physical model of the high - and low - voltage power distribution cabinet;

[0142] Establish the operating electrical equations and operating thermal equations of the high - and low - voltage power distribution cabinet;

[0143] Integrate the operating electrical equations and operating thermal equations into the physical model to obtain the initial digital twin of the high - and low - voltage power distribution cabinet;

[0144] Simulate the working condition state of the initial digital twin;

[0145] Through the working condition state, determine the digital twin of the high - and low - voltage power distribution cabinet.

[0146] Among them, the physical structure parameters refer to the parameters describing the physical characteristics of the high - and low - voltage power distribution cabinet, including dimensions, shapes, material properties, component layouts, connection methods, etc. The physical model refers to the virtual representation of the high - and low - voltage power distribution cabinet. The operating electrical equations refer to the mathematical equations describing the electrical behavior of the high - and low - voltage power distribution cabinet, including circuit equations, electromagnetic field equations, etc. The operating thermal equations refer to the mathematical equations describing the thermal behavior of the high - and low - voltage power distribution cabinet, including heat conduction equations, convection equations, radiation equations, etc. The initial digital twin refers to the preliminary virtual model obtained after integrating the operating electrical equations and operating thermal equations into the physical model. The working condition state refers to various working conditions that the high - and low - voltage power distribution cabinet may encounter during actual operation, such as different loads, ambient temperatures, humidities, etc. The digital twin refers to the virtual model that can accurately simulate the physical and behavioral characteristics of the high - and low - voltage power distribution cabinet after verification and adjustment.

[0147] Optionally, integrating the operating electrical equation and the operating thermal equation into the physical model to obtain the initial digital twin of the high and low voltage power distribution cabinet can be achieved through multi-physics field simulation technology.

[0148] Based on the fault probability matrix, the present invention analyzes the fault evolution path of the high and low voltage power distribution cabinet by using the digital twin, which can effectively utilize the digital twin to analyze the fault evolution path of the high and low voltage power distribution cabinet and provide strong support for fault prediction, prevention and maintenance.

[0149] Specifically, analyzing the fault evolution path of the high and low voltage power distribution cabinet by using the digital twin based on the fault probability matrix includes:

[0150] Define the fault deviation code and the state vector of the high and low voltage power distribution cabinet;

[0151] Establish the fault transition probability matrix of the high and low voltage power distribution cabinet according to the fault probability matrix and the fault deviation code;

[0152] Calculate the fault evolution path probability of the high and low voltage power distribution cabinet through the state vector and the fault transition probability matrix by using the following formula:

[0153]

[0154] where P(Path t→t+k ) represents the fault evolution path probability of the high and low voltage power distribution cabinet from time t to t + k, represents the transition probability from the state vector Z i to the state vector Z i+1 at time t + i, P represents the fault transition probability matrix, PPO action (Z i+1 |Z i ) represents the probability that the policy network selects an action to make the state vector Z i transition to the state vector Z i+1 , Z i represents the state vector at the i-th step, Z i+1 represents the state vector at the (i + 1)-th step, and k represents the step length of the fault evolution path;

[0155] Determine the fault evolution path of the high and low voltage power distribution cabinet by using the digital twin according to the fault evolution path probability.

[0156] Among them, the fault deviation code refers to the code used to uniquely identify various fault types in the high-voltage and low-voltage power distribution cabinets. The state vector refers to the fault state representing a specific component or system in the power distribution cabinet. The fault transition probability matrix is a square matrix, where each element represents the probability of transitioning from one fault state to another. The fault evolution path probability refers to the probability that the high-voltage and low-voltage power distribution cabinet evolves from a certain initial state to another state along a specific path. The transition probability refers to the probability of transitioning from state vector to state vector at time. The policy network selects actions, which refers to a decision-making model used to select actions to affect the transition of the state vector. Actions may include maintenance, component replacement, etc. The step size refers to the time interval in the fault evolution path, indicating the time required to transition from one state vector to the next. The fault evolution path refers to the process by which the high-voltage and low-voltage power distribution cabinet reaches the final fault state from the initial fault state through a series of state transitions.

[0157] Finally, analyzing the fault detection report of the high-voltage and low-voltage power distribution cabinet according to the fault evolution path of the present invention can effectively utilize the fault evolution path to analyze the fault detection report of the high-voltage and low-voltage power distribution cabinet, improving the operation reliability of the power distribution system. Among them, the fault detection report refers to a comprehensive document for fault detection, analysis, diagnosis, and recording of the high-voltage and low-voltage power distribution cabinet, including fault cause analysis, fault impact assessment, fault evolution path analysis, and maintenance and improvement suggestions.

[0158] Compared with the problems described in the background art, first, by collecting the electrical parameters, mechanical states, and thermal imaging data of high- and low-voltage distribution cabinets and constructing a multi-modal data stream, the comprehensive monitoring of the operating state of the distribution cabinets is realized. By using a method that combines the LSTM-Attention hybrid network and the wavelet packet transform function, the time-series features and the frequency-domain features of non-stationary signals in the multi-modal data stream are effectively extracted, improving the accuracy of fault identification. Further, by constructing a fault knowledge graph of high- and low-voltage distribution cabinets, the topological correlation feature vectors of the distribution cabinets are deeply analyzed, providing a richer information dimension for fault diagnosis. Based on the time-series features, the frequency-domain features of non-stationary signals, and the topological correlation feature vectors, the global fault model of the distribution cabinets trained has higher prediction accuracy and generalization ability, and can accurately analyze the fault probability matrix of high- and low-voltage distribution cabinets. In addition, the established digital twin of high- and low-voltage distribution cabinets realizes the intuitive display and in-depth analysis of the fault evolution path, providing more intuitive and effective decision-making support for maintenance personnel. Finally, the fault detection report generated based on the fault evolution path not only details the fault information but also provides targeted repair and improvement suggestions, effectively guiding the maintenance work of high- and low-voltage distribution cabinets, reducing the failure rate, and improving the operation efficiency and safety of the power system. Overall, this technical solution realizes the intelligent prediction, diagnosis, and analysis of faults in high- and low-voltage distribution cabinets, providing a strong guarantee for the reliable operation of the power system. Therefore, the present invention can improve the maintenance efficiency of high- and low-voltage distribution cabinets.

[0159] Embodiment 2:

[0160] As Figure 2 shown, it is a functional module diagram of a fault detection system for high- and low-voltage distribution cabinets based on intelligent AI according to the present invention.

[0161] The fault detection system 200 for high- and low-voltage distribution cabinets based on intelligent AI according to the present invention can be installed in an electronic device. According to the functions realized, the fault detection system based on intelligent AI can include a multi-modal data stream construction module 201, a time-frequency domain feature extraction module 202, a topological correlation feature extraction module 203, a fault probability matrix analysis module 204, and a fault evolution path analysis module 205. The modules in the present invention can also be referred to as units, which refer to a series of computer program segments that can be executed by the processor of an electronic device and can complete fixed functions, and are stored in the memory of the electronic device.

[0162] In the embodiment of the present invention, the functions of each module / unit are as follows:

[0163] The multi-modal data stream construction module 201 is used to collect the electrical parameters, mechanical states, and thermal imaging data of high- and low-voltage distribution cabinets and construct a multi-modal data stream of the electrical parameters, mechanical states, and thermal imaging data;

[0164] The time-frequency domain feature extraction module 202 is configured to construct an LSTM-Attention hybrid network for the multimodal data stream, extract the temporal features of the multimodal data stream through the LSTM-Attention hybrid network, and extract the frequency domain features of the non-stationary signals of the multimodal data stream by using a preset wavelet packet transform function;

[0165] The topological association feature extraction module 203 is configured to obtain the historical fault data of the high-voltage and low-voltage power distribution cabinets, construct a fault knowledge graph of the high-voltage and low-voltage power distribution cabinets according to the historical fault data, and analyze the topological association feature vectors of the high-voltage and low-voltage power distribution cabinets through the fault knowledge graph;

[0166] The fault probability matrix analysis module 204 is configured to train a pre-constructed global power distribution cabinet fault model through the temporal features, the frequency domain features of the non-stationary signals, and the topological association feature vectors to obtain a trained global power distribution cabinet fault model, and analyze the fault probability matrix of the high-voltage and low-voltage power distribution cabinets by using the trained global power distribution cabinet fault model, where the global power distribution cabinet fault model is a model for analyzing the fault states of high-voltage and low-voltage power distribution cabinets;

[0167] The fault evolution path analysis module 205 is configured to establish a digital twin of the high-voltage and low-voltage power distribution cabinets, analyze the fault evolution path of the high-voltage and low-voltage power distribution cabinets by using the digital twin based on the fault probability matrix, and analyze the fault detection report of the high-voltage and low-voltage power distribution cabinets according to the fault evolution path.

[0168] Specifically, each module in the fault detection system 200 for high-voltage and low-voltage power distribution cabinets based on intelligent AI in the embodiments of the present invention adopts the same technical means as those in the above Figure 1 The fault detection method for high-voltage and low-voltage power distribution cabinets based on intelligent AI described therein, and can produce the same technical effects, which will not be elaborated here.

[0169] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention.

[0170] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A fault detection method for high and low voltage switchgear based on intelligent AI, characterized in that, The method includes: Collecting the electrical parameters, mechanical states, and thermal imaging data of high- and low-voltage switchgear cabinets, and constructing a multi-modal data stream of the electrical parameters, mechanical states, and thermal imaging data; Constructing an LSTM-Attention hybrid network for the multi-modal data stream, extracting the temporal features of the multi-modal data stream through the LSTM-Attention hybrid network, and using a preset wavelet packet transform function to extract the non-stationary signal frequency domain features of the multi-modal data stream; Obtaining the historical fault data of the high- and low-voltage switchgear cabinets, constructing a fault knowledge graph of the high- and low-voltage switchgear cabinets according to the historical fault data, and analyzing the topological correlation feature vectors of the high- and low-voltage switchgear cabinets through the fault knowledge graph; Training a pre-constructed global switchgear fault model through the temporal features, the non-stationary signal frequency domain features, and the topological correlation feature vectors to obtain a trained global switchgear fault model, and using the trained global switchgear fault model to analyze the fault probability matrix of the high- and low-voltage switchgear cabinets, where the global switchgear fault model refers to a model for analyzing the fault states of high- and low-voltage switchgear cabinets; Establishing a digital twin of the high- and low-voltage switchgear cabinets, analyzing the fault evolution path of the high- and low-voltage switchgear cabinets based on the fault probability matrix, and analyzing the fault detection report of the high- and low-voltage switchgear cabinets according to the fault evolution path.

2. The fault detection method for high and low voltage distribution cabinets based on intelligent AI as described in claim 1, wherein The constructing the multi-modal data stream of the electrical parameters, mechanical states, and thermal imaging data includes: Preprocessing the electrical parameters, mechanical states, and thermal imaging data to obtain processed electrical parameters, processed mechanical states, and processed thermal imaging data; Defining the main time reference for the processed electrical parameters, processed mechanical states, and processed thermal imaging data; Performing time synchronization calibration on the processed electrical parameters, processed mechanical states, and processed thermal imaging data according to the main time reference to obtain calibrated electrical parameters, calibrated mechanical states, and calibrated thermal imaging data; Integrating the calibrated electrical parameters, calibrated mechanical states, and calibrated thermal imaging data into a preset data framework to obtain integrated multi-modal data; Labeling the integrated multi-modal data to obtain the multi-modal data stream.

3. The fault detection method for high and low voltage power distribution cabinets based on intelligent AI according to claim 2, characterized in that, The constructing the LSTM-Attention hybrid network for the multi-modal data stream includes: Establishing multiple input channels for the multi-modal data stream; Constructing an LSTM layer for the multiple input channels; Determining the Attention mechanism of the LSTM layer; Constructing the LSTM-Attention hybrid network for the multi-modal data stream through the multiple input channels, the LSTM layer, and the Attention mechanism.

4. The fault detection method for high and low voltage power distribution cabinets based on intelligent AI according to claim 3, wherein The using the preset wavelet packet transform function to extract the non-stationary signal frequency domain features of the multi-modal data stream includes: Analyzing the data stream characteristics of the multi-modal data stream; Determining the function parameters of the wavelet packet transform function according to the data stream characteristics; Decomposing the multi-modal data stream through the function parameters to obtain decomposition coefficients; Analyze the Pearson coefficient and instantaneous energy gradient of the multimodal data stream according to the decomposition coefficient; Calculate the band energy of the multimodal data stream through the Pearson coefficient; Calculate the band entropy value of the multimodal data stream according to the instantaneous energy gradient; Based on the flow band energy and band entropy value, determine the non-stationary signal frequency domain characteristics of the multimodal data stream.

5. The fault detection method for high and low voltage switchgear based on intelligent AI according to claim 4, wherein, The calculating the band energy of the multimodal data stream through the Pearson coefficient includes: Determine the cross-modal coupling term of the multimodal data stream according to the Pearson coefficient; Analyze the band variance of the multimodal data stream based on the corresponding decomposition coefficient of the multimodal data stream; Determine the dynamic weight coefficient of the multimodal data stream according to the band variance; Calculate the band energy of the multimodal data stream based on the cross-modal coupling term and the dynamic weight coefficient.

6. The fault detection method for high and low voltage switchgear based on intelligent AI as claimed in claim 5, wherein The constructing the fault knowledge graph of the high-voltage and low-voltage power distribution cabinet according to the historical fault data includes: Preprocess the historical fault data to obtain preprocessed fault data; Use a preset BERT+BiLSTM model to extract the data entities of the preprocessed fault data; Analyze the entity relationships of the data entities, where the entity relationships include static relationships and dynamic relationships; Define the semantic network of the historical fault data; Based on the entity relationships and the data entities, use the semantic network to establish the fault knowledge graph of the high-voltage and low-voltage power distribution cabinet.

7. The fault detection method for high and low voltage switchgear based on intelligent AI according to claim 6, characterized in that, The analyzing the topological association feature vector of the high-voltage and low-voltage power distribution cabinet through the fault knowledge graph includes: Mark the graph nodes corresponding to the fault knowledge graph as feature vectors; Analyze the topological association information of the graph nodes; Based on the topological association information, fuse the feature vectors to obtain a fused association feature vector; Aggregate the fused association feature vector to obtain the topological association feature vector of the high-voltage and low-voltage power distribution cabinet.

8. The fault detection method for high and low voltage power distribution cabinets based on intelligent AI as claimed in claim 7, wherein The establishing the digital twin of the high-voltage and low-voltage power distribution cabinet includes: Identify the physical structure parameters of the high-voltage and low-voltage power distribution cabinet; Establish the physical model of the high-voltage and low-voltage power distribution cabinet according to the physical structure parameters; Establish the operating electrical equation and operating thermal equation of the high-voltage and low-voltage power distribution cabinet; Integrate the operating electrical equation and operating thermal equation into the physical model to obtain the initial digital twin of the high-voltage and low-voltage power distribution cabinet; Simulate the working condition state of the initial digital twin; Determine the digital twin of the high-voltage and low-voltage power distribution cabinet through the working condition state.

9. The fault detection method for high- and low-voltage switchgear based on intelligent AI as described in claim 8, wherein, The analyzing the fault evolution path of the high-voltage and low-voltage power distribution cabinet by using the digital twin based on the fault probability matrix includes: Define the fault offset code and state vector of the high-voltage and low-voltage power distribution cabinet; Establish the fault transition probability matrix of the high-voltage and low-voltage power distribution cabinet according to the fault probability matrix and the fault offset code; Calculate the fault evolution path probability of the high-voltage and low-voltage power distribution cabinet through the state vector and the fault transition probability matrix; Determine the fault evolution path of the high-voltage and low-voltage power distribution cabinet by using the digital twin according to the fault evolution path probability.

10. A fault detection system for high and low voltage switchgear based on intelligent AI, characterized in that, The system includes: A multi-modal data stream construction module, which is used to collect the electrical parameters, mechanical states, and thermal imaging data of high- and low-voltage distribution cabinets, and construct a multi-modal data stream of the electrical parameters, mechanical states, and thermal imaging data; A time-frequency domain feature extraction module, which is used to construct an LSTM-Attention hybrid network for the multi-modal data stream, extract the temporal features of the multi-modal data stream through the LSTM-Attention hybrid network, and use a preset wavelet packet transform function to extract the frequency domain features of the non-stationary signals of the multi-modal data stream; A topological correlation feature extraction module, which is used to obtain the historical fault data of the high- and low-voltage distribution cabinets, construct a fault knowledge graph of the high- and low-voltage distribution cabinets according to the historical fault data, and analyze the topological correlation feature vectors of the high- and low-voltage distribution cabinets through the fault knowledge graph; A fault probability matrix analysis module, which is used to train a pre-constructed global distribution cabinet fault model through the temporal features, the frequency domain features of the non-stationary signals, and the topological correlation feature vectors to obtain a trained global distribution cabinet fault model, and use the trained global distribution cabinet fault model to analyze the fault probability matrix of the high- and low-voltage distribution cabinets, where the global distribution cabinet fault model refers to a model for analyzing the fault states of high- and low-voltage distribution cabinets; A fault evolution path analysis module, which is used to establish a digital twin of the high- and low-voltage distribution cabinets, analyze the fault evolution path of the high- and low-voltage distribution cabinets based on the fault probability matrix by using the digital twin, and analyze the fault detection report of the high- and low-voltage distribution cabinets according to the fault evolution path.

Images