A method for condition detection and fault diagnosis of high-voltage circuit breakers

Abstract

This invention relates to the field of power equipment monitoring technology and discloses a method for condition detection and fault diagnosis of high-voltage circuit breakers. The method involves collecting real-time condition data of the high-voltage circuit breaker and preprocessing it to obtain standardized data; extracting time-domain, frequency-domain, and time-frequency features from the standardized data to form feature vectors; analyzing the feature vectors using anomaly detection algorithms to identify abnormal state patterns; based on the abnormal state patterns and combined with historical fault records, applying causal reasoning methods to mine the causal relationship between abnormal states and faults, and determining key fault indicators; establishing a health benchmark model based on the key fault indicators, calculating the deviation between the current condition data and the health benchmark model; evaluating the deviation to obtain the fault risk level; and using a prediction model based on the fault risk level and key fault indicators to estimate the fault probability of the high-voltage circuit breaker in the future time period and generating a condition detection report.

Description

Technical Field

[0001] This invention relates to the field of power equipment monitoring technology, specifically a method for condition detection and fault diagnosis of high-voltage circuit breakers. Background Technology

[0002] Current high-voltage circuit breaker condition monitoring primarily relies on periodic maintenance and threshold alarms. Existing technologies largely depend on single-parameter thresholds for circuit breaker condition assessment, failing to effectively integrate and analyze multi-source data such as vibration, current, and temperature. Feature extraction methods are simplistic, lacking systematic joint mining of time-domain, frequency-domain, and time-frequency-domain features. Anomaly detection mechanisms are rigid, based on fixed rules, unable to identify complex anomaly patterns. Fault diagnosis is experience-based, lacking quantitative analysis of the causal relationship between abnormal states and fault types. Health assessment standards are standardized, failing to consider individual equipment differences and operational history. Risk warnings are delayed, largely based on current state assessments, lacking the ability to predict future fault probabilities. Existing methods need to address key technical challenges such as multi-source data fusion, intelligent anomaly detection, causal reasoning, and risk prediction.

[0003] Traditional circuit breaker fault diagnosis methods suffer from significant shortcomings in accuracy and foresight. Data acquisition frequency is low, and key state change features are easily lost. Preprocessing methods are crude, and noise interference affects data quality. Time-domain feature extraction has limited dimensions, failing to capture dynamic waveform changes. Frequency-domain analysis lacks resolution, making it difficult to identify weak fault features. Time-frequency analysis is computationally complex, resulting in low real-time processing efficiency. Anomaly detection algorithms have fixed sensitivity, leading to both false alarms and false negatives. Historical fault records are incomplete, resulting in a weak data foundation for causal relationship mining. Causal inference methods are linearized, lacking the ability to identify complex nonlinear relationships. Health benchmark models are static and cannot adapt to equipment performance degradation. Deviation calculation indicators are singular, failing to comprehensively reflect health status. Risk assessment level classification is crude, failing to quantify risk severity. Predictive models lack sufficient training data, resulting in limited long-term prediction accuracy. Summary of the Invention

[0004] The purpose of this invention is to provide a method for condition detection and fault diagnosis of high-voltage circuit breakers to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for condition detection and fault diagnosis of high-voltage circuit breakers, the method comprising:

[0006] Real-time status data of high-voltage circuit breakers are collected, and the real-time status data is preprocessed to obtain standardized data.

[0007] Multi-dimensional features, including time-domain features, frequency-domain features, and time-frequency features, are extracted from the standardized data to form a feature vector;

[0008] The abnormal state patterns of the high-voltage circuit breaker are identified by analyzing the feature vector using an anomaly detection algorithm.

[0009] Based on the aforementioned abnormal state pattern, combined with historical fault records, causal reasoning methods are applied to mine the causal relationship between abnormal states and faults, and to determine key fault indicators.

[0010] Based on the aforementioned key fault indicators, a health baseline model for the high-voltage circuit breaker is established, and the deviation between the current state data and the health baseline model is calculated.

[0011] The deviation is evaluated to obtain the fault risk level of the high-voltage circuit breaker;

[0012] Based on the aforementioned fault risk level and key fault indicators, a predictive model is used to estimate the fault probability of the high-voltage circuit breaker in the future time period and generate a condition monitoring report.

[0013] Preferably, the step of preprocessing the real-time status data to obtain standardized data includes:

[0014] The real-time status data is subjected to noise filtering processing, and high-frequency noise is removed using a wavelet transform algorithm to obtain denoised data;

[0015] The denoised data is filled with missing values ​​by applying an interpolation algorithm to fill in the missing points and obtain complete data.

[0016] The complete data is normalized to scale it to a uniform scale, resulting in standardized data.

[0017] Preferably, the step of extracting multi-dimensional features from the standardized data, including time-domain features, frequency-domain features, and time-frequency features, to form a feature vector includes:

[0018] Time-domain features, including mean, variance, and kurtosis, are calculated from the standardized data.

[0019] Frequency domain features are extracted from the standardized data, and Fourier transform is applied to obtain the spectral energy distribution;

[0020] Analyze the time-frequency characteristics from the standardized data and use short-time Fourier transform to obtain joint time-frequency characteristics;

[0021] The time-domain features, frequency-domain features, and time-frequency features are combined into a feature vector.

[0022] Preferably, the step of analyzing the feature vector using an anomaly detection algorithm to identify the abnormal state mode of the high-voltage circuit breaker includes:

[0023] The feature vectors are trained using the isolated forest algorithm to construct a normal state baseline;

[0024] The current feature vector is input into the Isolation Forest algorithm to calculate the anomaly score and identify outliers that deviate from the normal state.

[0025] Based on the anomaly scores, clustered anomaly points form anomaly state patterns.

[0026] Preferably, the application of causal reasoning methods to mine the causal relationship between abnormal states and faults, and to determine key fault indicators, includes:

[0027] Granger causality test was used to analyze the time series data of the abnormal state patterns and historical fault records to examine the causal relationship.

[0028] The conditional probability relationship between the abnormal state patterns and fault events is modeled using Bayesian networks.

[0029] Key nodes are extracted from Bayesian networks as key failure indicators.

[0030] Preferably, the step of establishing a health baseline model for the high-voltage circuit breaker and calculating the deviation between the current state data and the health baseline model includes:

[0031] The support vector regression algorithm is used to train the model on historical normal state data to fit a health benchmark model.

[0032] Project the current state data onto the health baseline model and calculate the sum of squared residuals as the deviation.

[0033] The deviation is averaged using a sliding window to obtain a smoothed deviation.

[0034] Preferably, the assessment of the deviation to obtain the fault risk level of the high-voltage circuit breaker includes:

[0035] Construct a deep neural network model, inputting the smoothness deviation and key fault indicators;

[0036] The deep neural network model is trained to output a fault risk level, which includes low risk, medium risk and high risk.

[0037] Cross-validation is used to optimize the parameters of a deep neural network model.

[0038] Preferably, estimating the failure probability of the high-voltage circuit breaker in a future time period using a predictive model includes:

[0039] The Long Short-Term Memory (LSTM) network model is used to train historical fault sequences to learn fault time patterns.

[0040] Input the current fault risk level and key fault indicators into the long short-term memory network model to predict the fault probability at future time points, set a probability threshold, and trigger an early warning mechanism.

[0041] Preferably, the generation of the status detection report includes:

[0042] Summarize abnormal state patterns, key failure indicators, failure risk levels, and failure probabilities; format report content, including tables and trend charts; and output the report to the user interface.

[0043] Preferably, the method further includes periodically updating the health baseline model and the prediction model:

[0044] Collect newly generated state data and fault records, retrain the support vector regression algorithm and long short-term memory network model, verify the model performance and iteratively optimize.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] Multi-dimensional feature vectors are formed by extracting time-domain, frequency-domain, and time-frequency features from standardized data. Time-domain features include statistics such as mean, variance, and peak value, characterizing the signal amplitude. Frequency-domain features are obtained by calculating the power spectral density using Fourier transform, revealing the signal's frequency composition. Time-frequency features are obtained using wavelet transform, reflecting the joint time-frequency distribution of the signal. The feature vector dimensions are optimized through principal component analysis, retaining key discriminative information. Feature extraction employs a sliding window mechanism to ensure temporal continuity. Feature standardization eliminates the influence of dimensions. Feature selection algorithms remove redundant features, reducing computational complexity. Multi-dimensional feature fusion provides a comprehensive state representation.

[0047] Anomaly detection algorithms are used to analyze feature vectors and identify abnormal state patterns. The anomaly detection algorithm employs the isolated forest algorithm, which is suitable for high-dimensional feature spaces. Abnormal state patterns are identified through feature space density deviation. Pattern recognition considers the correlation between features to avoid false positives. The detection threshold is dynamically adjusted based on operational history to balance sensitivity and specificity. Pattern classification distinguishes anomaly types, providing guidance for diagnosis. Real-time detection frequency is synchronized with data acquisition to ensure timely warnings. Detection results are visualized to assist in manual confirmation.

[0048] Key failure indicators (KCIs) are determined using causal inference methods based on anomaly patterns and historical failure records. Causal inference employs a causal discovery algorithm to infer causal relationships from data. Historical failure records contain information such as failure type and occurrence time. Causal relationship mining analyzes the temporal sequence and statistical dependencies between anomalies and failures. Key failure indicators are selected based on predictive capabilities. Indicator weights are allocated according to causal strength, highlighting key influencing factors. Causal network construction demonstrates the relationships between multiple factors. The inference results are interpretable, providing a basis for decision-making.

[0049] A health baseline model is established based on key failure indicators to calculate the deviation from the baseline status. The health baseline model is trained using historical normal data. The baseline value considers individual equipment differences and the operating environment. Deviation calculation uses a distance metric to quantify the difference between the current state and the baseline. Deviation is standardized for easy comparison between indicators. Dynamic thresholds are adjusted according to equipment aging to adapt to performance degradation. The health assessment integrates deviations from multiple indicators to comprehensively reflect the equipment's condition. Attached Figure Description

[0050] Figure 1 This is a schematic diagram illustrating the working principle of the high-voltage circuit breaker condition detection and fault diagnosis method described in this invention.

[0051] Figure 2 A flowchart for preprocessing real-time status data;

[0052] Figure 3 Flowchart for abnormal state pattern recognition;

[0053] Figure 4 The time-series distribution of the detection scores for abnormal conditions of high-voltage circuit breakers;

[0054] Figure 5 This represents the causal conditional probability distribution of key fault indicators for high-voltage circuit breakers. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Please see Figure 1 This invention provides a method for condition detection and fault diagnosis of high-voltage circuit breakers. The method includes: collecting real-time condition data of the high-voltage circuit breaker and preprocessing the real-time condition data to obtain standardized data; extracting multi-dimensional features from the standardized data, including time-domain features, frequency-domain features, and time-frequency features, to form a feature vector; analyzing the feature vector using an anomaly detection algorithm to identify abnormal state patterns of the high-voltage circuit breaker; based on the abnormal state patterns and combined with historical fault records, applying causal reasoning methods to mine the causal relationship between abnormal states and faults, and determining key fault indicators; establishing a health benchmark model for the high-voltage circuit breaker based on the key fault indicators, and calculating the deviation between the current condition data and the health benchmark model; evaluating the deviation to obtain the fault risk level of the high-voltage circuit breaker; and using a prediction model to estimate the fault probability of the high-voltage circuit breaker in a future time period based on the fault risk level and key fault indicators, and generating a condition detection report.

[0057] Example 1: See Figure 2 In specific implementation, the high-voltage circuit breaker condition detection and fault diagnosis method involves preprocessing real-time condition data to obtain standardized data, including noise filtering of real-time condition data and using wavelet transform algorithms to remove high-frequency noise to obtain denoised data. In some embodiments, the wavelet transform algorithm uses discrete wavelet transform for multi-scale decomposition and uses threshold processing to filter out high-frequency noise components. Optionally, missing value imputation uses linear interpolation algorithms to fill missing points, imputing based on the trend of adjacent data points to obtain complete data. It can be understood that normalization scaling scales the data to a uniform scale, using the min-max normalization method, with the formula:

[0058]

[0059] in: Represents the original data points. This represents the minimum value in the dataset. This represents the maximum value in the dataset. This represents the normalized data value. Normalization adjusts the data range to between zero and one.

[0060] In specific implementations, multi-dimensional features are extracted from standardized data, including time-domain features, frequency-domain features, and time-frequency features, forming a feature vector. In some embodiments, time-domain features include mean, variance, and peak factor. The mean reflects the central tendency of the data, the variance describes the dispersion of the data, and the peak factor characterizes the ratio of the signal peak value to the effective value. Optionally, the frequency-domain features are subjected to Fourier transform to obtain the spectral energy distribution. The Fourier transform converts the time-domain signal into a frequency-domain representation and calculates the energy of each frequency component. It can be understood that the time-frequency features use short-time Fourier transform to obtain joint time-frequency features. The short-time Fourier transform analyzes the signal by windowing and segmentation to obtain the time-frequency distribution matrix. Finally, the time-domain features, frequency-domain features, and time-frequency features are combined in sequence into a feature vector for subsequent analysis.

[0061] Example 2: See Figure 3In specific implementations, anomaly detection algorithms are used to analyze feature vectors to identify abnormal state patterns of high-voltage circuit breakers. This includes training the feature vectors using the Isolation Forest algorithm to construct a normal state baseline. The Isolation Forest algorithm simulates normal data distribution by constructing multiple isolated trees, where each tree randomly selects features and segmentation values ​​to isolate data points. In some embodiments, the training process of the Isolation Forest algorithm uses historical normal state data as input to generate an Isolation Forest model that can effectively distinguish between normal and abnormal data patterns. Optionally, when constructing the normal state baseline, parameters such as the number of trees and subsample size are set to optimize the model's ability to represent normal states. It can be understood that the Isolation Forest algorithm assesses the degree of anomaly by calculating the path length of data points in the trees; data points with shorter paths are more likely to be abnormal.

[0062] In practice, the current feature vector is input into the Isolation Forest algorithm to calculate anomaly scores and identify outliers that deviate from the normal state. The anomaly scores are derived based on the average path length of data points in the Isolation Forest, using the following formula:

[0063]

[0064] in: Indicates abnormal scores. Representing data points The expected value of the path length in a set of isolated trees. Indicates the given sample size The standardization constant is used to adjust the baseline for path length; in some embodiments, the anomaly score is calculated by traversing all isolated trees and aggregating the path length results, and the closer the score is to 1, the more likely the data point is to be an anomaly; optionally, a threshold is set when identifying anomalies, for example, data points with a score higher than 0.5 are marked as anomalies, and the threshold can be adjusted according to the actual application; it can be understood that the process of identifying anomalies involves comparing the deviation of the current feature vector from the normal state baseline, thereby capturing potential anomalies in the state of the high-voltage circuit breaker.

[0065] In specific implementation, abnormal points are clustered based on abnormal scores to form abnormal state patterns. Clustering algorithms such as K-means are used to group abnormal points to discover common abnormal patterns. Optionally, the clustering process first standardizes the abnormal scores and then applies a distance metric such as Euclidean distance to evaluate the similarity between abnormal points. In some embodiments, the clustering results generate multiple abnormal state pattern categories, each category representing a type of abnormal behavior of a high-voltage circuit breaker.

[0066] See Figure 4This graph is a core visualization result of the anomaly detection stage in the condition monitoring of high-voltage circuit breakers. It presents the anomaly scores at different time points in a line chart format, while also marking the 0.5 anomaly threshold line. The value of this graph lies in its ability to intuitively locate the periods when anomalies occur in high-voltage circuit breakers, clearly distinguishing the boundaries between normal and abnormal states. It not only verifies the ability of the Isolation Forest algorithm to identify anomaly patterns but also provides accurate data support for subsequent causal inference to uncover the correlation between anomalies and faults. It is one of the key visualization achievements for realizing real-time condition monitoring.

[0067] Example 3: In specific implementation, causal reasoning methods are applied to uncover the causal relationship between abnormal states and faults, and to determine key fault indicators. This includes using Granger causality tests to analyze time-series data of abnormal state patterns and historical fault records to test the causal relationship. In some embodiments, Granger causality tests compare the goodness of fit of models containing and not containing potential causal variables by constructing autoregressive models, and use the F-test statistic to assess the significance of the causal relationship. In Granger causality tests, the specific implementation of the autoregressive model comparison includes constructing two models: a restricted model that only includes historical time-series data of the fault event itself as predictor variables, and an unrestricted model that includes both historical data of the fault event itself and historical data of the abnormal state pattern as predictor variables. By calculating the sum of squared residuals of the two models, the goodness of fit improvement of the unrestricted model relative to the restricted model is evaluated, thereby using the F-test statistic to determine whether the abnormal state pattern has a significant Granger causal relationship with the fault event. Optionally, the formula for calculating the F-test statistic is:

[0068]

[0069] in: This represents the F-statistic value. This represents the sum of squared residuals of the constrained model. This represents the sum of squared residuals of an unconstrained model. Indicates the lag order. This represents the total number of time series samples. It can be understood that the Granger causality test helps identify whether anomaly patterns statistically precede the occurrence of failure events, thus providing a quantitative basis for causal relationships.

[0070] In specific implementations, Bayesian networks are used to model the conditional probability relationship between abnormal state patterns and fault events. The Bayesian network uses a directed acyclic graph to represent the dependency structure between variables, with nodes corresponding to abnormal state patterns and fault event types, and edges representing conditional probability distributions. In some embodiments, Bayesian network parameter learning uses historical fault record data, and the conditional probability table is calculated using the maximum likelihood estimation method. Optionally, key nodes are extracted from the Bayesian network as key fault indicators, and the selection of key nodes is based on the conditional probability value between the node and the fault event or the centrality measure in the network.

[0071] See Figure 5 This chart is a core visualization result of the causal reasoning stage in the condition monitoring and fault diagnosis method for high-voltage circuit breakers, which involves uncovering the causal relationship between anomalies and faults. It presents the causal conditional probabilities of five key fault indicators in the form of a horizontal bar chart. The indicators covered in the chart include abnormal vibration, current fluctuation, temperature rise, mechanical wear, and insulation aging. The conditional probabilities of each indicator are quantified using Granger causality tests and Bayesian network modeling: mechanical wear has the highest conditional probability, indicating the strongest causal association with the fault event; followed by abnormal vibration, current fluctuation, and temperature rise; insulation aging has a relatively low conditional probability. The value of this chart lies in its intuitive differentiation of the influence weight of different fault indicators on high-voltage circuit breaker faults, helping technicians quickly identify core risk factors. It provides key input indicators for the subsequent construction of health benchmark models and lays a quantitative foundation for the accurate assessment of fault risk levels, serving as an important intermediate visualization result connecting anomaly detection and fault diagnosis.

[0072] Example 4: In specific implementation, a health baseline model of the high-voltage circuit breaker is established and the deviation between the current state data and the health baseline model is calculated. This includes using a support vector regression algorithm to train historical normal state data to fit the health baseline model. The support vector regression algorithm maps the input data to a high-dimensional space through a kernel function and finds the optimal hyperplane to model normal state behavior. In some embodiments, the training process of the support vector regression algorithm uses a radial basis function kernel, and sets regularization parameters and kernel coefficients to minimize prediction error. Optionally, the current state data is projected onto the health baseline model, and the sum of squared residuals is calculated as the deviation by comparing the actual observed values ​​with the model prediction values. The formula is:

[0073]

[0074] in: Indicates the degree of deviation. The first element representing the current state data One observation value, This indicates that the support vector regression model is effective for the first... Predicted values ​​for each observation. This represents the total number of observations. It can be understood that the deviation is processed by a sliding window average to obtain a smoothed deviation. The sliding window average process uses a fixed window size to calculate the moving average of continuous deviation values, thereby reducing the impact of random fluctuations.

[0075] In specific implementation, the deviation is evaluated to obtain the fault risk level of the high-voltage circuit breaker. This includes constructing a deep neural network model, inputting smooth deviation and key fault indicators. The deep neural network model consists of multiple fully connected layers and activation functions to learn complex nonlinear relationships. In some embodiments, the deep neural network model is trained to output fault risk levels, which include low risk, medium risk, and high risk. The training process uses labeled historical data and minimizes the classification loss through backpropagation. Optionally, cross-validation is used to optimize the parameters of the deep neural network model. Cross-validation divides the dataset into multiple subsets and iterates training and validation to select the optimal hyperparameters, such as the learning rate and the number of hidden layer nodes. It can be understood, referring to Table 1, that the output layer of the deep neural network model uses a softmax function to generate the probability distribution of each risk level, and the final risk level is determined by the highest probability category.

[0076] Table 1: Example layer structures and parameter configurations of deep neural network models

[0077] Layer type Number of nodes Activation function illustrate Input layer 10 none Receiver smoothing deviation and key failure indicators Hidden layer 1 50 ReLU Learning feature representations Hidden layer 2 30 ReLU Further abstract features Output layer 3 Softmax Output the probabilities of low risk, medium risk, and high risk.

[0078] Example 5: In specific implementation, a predictive model is used to estimate the failure probability of a high-voltage circuit breaker in the future time period. This includes training a Long Short-Term Memory (LSTM) network model on historical fault sequences to learn fault time patterns. The LTM network model captures long-term dependencies in the time series through a gating mechanism, using historical fault records and corresponding state data as training samples. In some embodiments, the training process of the LTM network model adopts a time-stepping approach. The input sequence includes historical fault risk levels and key fault indicators, and the output is the probability of fault occurrence. The loss function uses binary cross-entropy to optimize the model parameters. Optionally, the current fault risk level and key fault indicators are input into the LTM network model to predict the fault probability at future time points. The model output probability value ranges between zero and one, as shown in the formula:

[0079]

[0080] in: Indicates the probability of failure. This represents the output layer weight matrix. This indicates that the Long Short-Term Memory network is in time step The hidden state vector, This represents the output layer bias term; it can be understood as setting a probability threshold to trigger an early warning mechanism, for example, generating a high-level alarm when the failure probability exceeds 0.7. The threshold can be adjusted according to the actual application scenario to balance the false alarm and false negative rates.

[0081] In practical implementation, the generated status monitoring report includes a summary of abnormal status patterns, key fault indicators, fault risk levels, and fault probabilities. The formatted report content is presented in the form of tables and trend charts. In some embodiments, the report generation process automatically integrates multi-source data and uses a template engine to populate the data into a predefined structure, including a classification list of abnormal status patterns and a numerical table of key fault indicators. Optionally, the trend chart displays the curves of change of fault risk level and fault probability over time, helping users intuitively identify the status evolution trend. It can be understood that the output of the report to the user interface is achieved through web services or a graphical interface, supporting real-time refresh and export functions to ensure that users can obtain high-voltage circuit breaker status information in a timely manner.

[0082] In practice, regularly updating the health baseline model and prediction model involves collecting newly generated state data and fault records, and retraining the support vector regression algorithm and long short-term memory network model. Optionally, the retraining process uses incremental learning or batch learning methods, selecting an appropriate strategy based on the amount of new data to maintain the model's adaptability to changes in equipment state. In some embodiments, model performance is verified and iteratively optimized through cross-validation and performance metric evaluation, such as using accuracy, recall, or mean squared error as evaluation criteria. It is understood that iterative optimization involves adjusting model hyperparameters and structure, such as the kernel function parameters of the support vector regression algorithm or the hidden layer size of the long short-term memory network model, to improve the accuracy and robustness of fault diagnosis.

[0083] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0084] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for condition detection and fault diagnosis of a high-voltage circuit breaker, characterized in that, include: Real-time status data of high-voltage circuit breakers are collected, and the real-time status data is preprocessed to obtain standardized data. Multi-dimensional features, including time-domain features, frequency-domain features, and time-frequency features, are extracted from the standardized data to form a feature vector; The abnormal state patterns of the high-voltage circuit breaker are identified by analyzing the feature vector using an anomaly detection algorithm. Based on the aforementioned abnormal state pattern, combined with historical fault records, causal reasoning methods are applied to mine the causal relationship between abnormal states and faults, and to determine key fault indicators. Based on the aforementioned key fault indicators, a health baseline model for the high-voltage circuit breaker is established, and the deviation between the current state data and the health baseline model is calculated. The deviation is evaluated to obtain the fault risk level of the high-voltage circuit breaker; Based on the aforementioned fault risk level and key fault indicators, a predictive model is used to estimate the fault probability of the high-voltage circuit breaker in the future time period and generate a condition monitoring report. The application of causal reasoning methods to uncover the causal relationship between abnormal states and faults, and to determine key fault indicators, includes: Granger causality test was used to analyze the time series data of the abnormal state patterns and historical fault records to examine the causal relationship. The conditional probability relationship between the abnormal state patterns and fault events is modeled using Bayesian networks. Key nodes are extracted from Bayesian networks as key failure indicators.

2. The method for condition detection and fault diagnosis of high-voltage circuit breakers according to claim 1, characterized in that, The step of preprocessing the real-time status data to obtain standardized data includes: The real-time status data is subjected to noise filtering processing, and high-frequency noise is removed using a wavelet transform algorithm to obtain denoised data; The denoised data is filled with missing values ​​by applying an interpolation algorithm to fill in the missing points and obtain complete data. The complete data is normalized to scale it to a uniform scale, resulting in standardized data.

3. The method for condition detection and fault diagnosis of high-voltage circuit breakers according to claim 2, characterized in that, The step of extracting multi-dimensional features from the standardized data, including time-domain features, frequency-domain features, and time-frequency features, to form a feature vector includes: Time-domain features, including mean, variance, and kurtosis, are calculated from the standardized data. Frequency domain features are extracted from the standardized data, and Fourier transform is applied to obtain the spectral energy distribution; Analyze the time-frequency characteristics from the standardized data and use short-time Fourier transform to obtain joint time-frequency characteristics; The time-domain features, frequency-domain features, and time-frequency features are combined into a feature vector.

4. The method for condition detection and fault diagnosis of high-voltage circuit breakers according to claim 3, characterized in that, The step of analyzing the feature vector using an anomaly detection algorithm to identify the abnormal state mode of the high-voltage circuit breaker includes: The feature vectors are trained using the isolated forest algorithm to construct a normal state baseline; The current feature vector is input into the Isolation Forest algorithm to calculate the anomaly score and identify outliers that deviate from the normal state. Based on the anomaly scores, clustered anomaly points form anomaly state patterns.

5. The method for condition detection and fault diagnosis of high-voltage circuit breakers according to claim 1, characterized in that, The establishment of a health baseline model for the high-voltage circuit breaker and the calculation of the deviation between the current state data and the health baseline model include: The support vector regression algorithm is used to train the model on historical normal state data to fit a health benchmark model. Project the current state data onto the health baseline model and calculate the sum of squared residuals as the deviation. The deviation is averaged using a sliding window to obtain a smoothed deviation.

6. The method for condition detection and fault diagnosis of high-voltage circuit breakers according to claim 5, characterized in that, The assessment of the deviation to obtain the fault risk level of the high-voltage circuit breaker includes: Construct a deep neural network model, inputting the smoothness deviation and key fault indicators; The deep neural network model is trained to output a fault risk level, which includes low risk, medium risk and high risk. Cross-validation is used to optimize the parameters of a deep neural network model.

7. The method for condition detection and fault diagnosis of high-voltage circuit breakers according to claim 6, characterized in that, The method of estimating the failure probability of high-voltage circuit breakers in the future time period using a predictive model includes: The Long Short-Term Memory (LSTM) network model is used to train historical fault sequences to learn fault time patterns. Input the current fault risk level and key fault indicators into the long short-term memory network model to predict the fault probability at future time points, set a probability threshold, and trigger an early warning mechanism.

8. The method for condition detection and fault diagnosis of high-voltage circuit breakers according to claim 7, characterized in that, The generated status detection report includes: Summarize abnormal state patterns, key failure indicators, failure risk levels, and failure probabilities; format report content, including tables and trend charts; and output the report to the user interface.

9. The method for condition detection and fault diagnosis of high-voltage circuit breakers according to claim 8, characterized in that, The method also includes periodically updating the health baseline model and the prediction model: Collect newly generated state data and fault records, retrain the support vector regression algorithm and long short-term memory network model, verify the model performance and iteratively optimize.

Images