Capacitor fault prediction and diagnosis system based on loss data

Abstract

The invention, which relates to the technical field of capacitor fault diagnosis, discloses a capacitor fault prediction and diagnosis system based on loss data, comprising a data acquisition module, an analysis module, a diagnosis module and a decision module. The data acquisition module is responsible for acquiring capacitance parameter data of a capacitor, collecting internal parameters of the capacitor through a capacitance sensor array, and transmitting the capacitance parameter data to the data preprocessing module; the data preprocessing module is used for further preprocessing the capacitance parameter data transmitted from the data acquisition module; according to the method, the percentage loss proportion and the residual service life RUL of the capacitor are calculated through the analysis unit based on the screened optimization characteristic data, different strategies are further executed in a graded mode, the loss trend of the capacitor is effectively calculated by analyzing the internal parameters C, ESR and tan delta of the capacitor, potential faults are recognized in advance, and the fault diagnosis accuracy is improved. Therefore, the accuracy of capacitor fault prediction is improved, and long-term stable operation of equipment is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of capacitor fault diagnosis, and specifically to a capacitor fault prediction and diagnosis system based on loss data. Background Art

[0002] With the complexity of the power system and the continuous improvement of the requirements for power quality, capacitors, as important power equipment, are widely used in the power system for reactive power compensation, power factor regulation, and voltage stability enhancement. However, during long-term operation, capacitors may malfunction due to multiple factors such as environmental factors, electrical load fluctuations, and material aging of themselves, which may in turn affect the operation safety and stability of the entire power system.

[0003] Traditional capacitor fault detection methods rely on manual inspections, regular test measurements, and switch quantity monitoring, which are inefficient and prone to missing potential fault hazards, resulting in the failure to detect capacitor faults in a timely manner during the operation of the system, causing equipment damage and large-scale power outages.

[0004] In view of the above technical deficiencies, a solution is proposed herein. Summary of the Invention

[0005] In view of the deficiencies of the prior art, the present invention provides a capacitor fault prediction and diagnosis system based on loss data.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A capacitor fault prediction and diagnosis system based on loss data, including: a data acquisition module, an analysis module, a diagnosis module, and a decision module;

[0007] The data acquisition module is responsible for obtaining capacitor parameter data of the capacitor, collecting internal parameters of the capacitor through a capacitor sensor array, and transmitting the capacitor parameter data to the data preprocessing module;

[0008] The data preprocessing module further preprocesses the capacitor parameter data transmitted from the data acquisition module. The data processing includes signal denoising and data alignment, and the processed capacitor parameter data is transmitted to the analysis module;

[0009] The analysis module includes a feature selection unit and an analysis unit. The feature selection unit performs feature selection based on the data from the data preprocessing module and transmits the optimized feature data set to the prediction and analysis unit; the prediction and analysis unit constructs a long short-term memory network (LSTM) prediction model based on the filtered and optimized feature data, and the calculated percentage loss ratio and predicted remaining useful life (RUL) of the capacitor are transmitted to the diagnosis module;

[0010] The diagnosis module classifies the operating status of the capacitor based on the percentage loss ratio and the remaining useful life (RUL) provided by the analysis module. The capacitor health status is divided into four levels;

[0011] The decision-making module executes specific maintenance strategies according to the health status classification results provided by the diagnosis module. The maintenance strategies include preventive maintenance mechanisms, predictive replacement strategies, and emergency replacement strategies.

[0012] The data acquisition module extracts the parameters of the capacitor operating status, obtains the internal parameters C, ESR, and tanδ of the capacitor through the capacitor sensor array, monitors the capacitor temperature T through the temperature sensor thermistor NTC, measures the working voltage U and current I through the Hall effect sensor. The capacitor sensor array uses a high-precision LCR digital bridge with an accuracy of 0.01%, and the sampling frequency is set to 1 kHz; the capacitor parameter data includes the internal parameters C, ESR, and tanδ of the capacitor, the capacitor temperature T, the working voltage U, and the current I. The capacitor parameter data is transmitted to the data preprocessing module through a communication protocol.

[0013] The data preprocessing module preprocesses the capacitor parameter data. The data processing includes signal denoising and data alignment; the signal denoising includes using wavelet transform (WaveletDenoising) to remove high-frequency noise from the ESR and tanδ data, using adaptive Kalman filtering (AdaptiveKalmanFiltering, AKF) to smooth the signal of the capacitance C, and using a Butterworth low-pass filter with a cut-off frequency of 0.1 Hz to filter out environmental interference signals for the temperature data. The data alignment includes using linear interpolation (LinearInterpolation) for time alignment and transmitting the processed capacitor parameter data to the analysis module.

[0014] The analysis module includes a feature selection unit and an analysis unit;

[0015] The feature selection unit extracts the key features in the capacitor loss data, reduces data redundancy by screening high-influence features, improves the calculation efficiency and classification accuracy of the XGBoost classification model, calculates the contribution degree of each feature to the XGBoost classification result through SHAP (Shapley Additive Explanations) feature importance analysis, screens out the core parameters affecting the capacitor health status, and removes features with a contribution degree lower than a set threshold such as 0.1 to optimize the model input feature set. The calculation method of the SHAP value is as follows: where D is the SHAP importance value of feature i, F is the set of all features, f(S) is the model output containing subset S. If the SHAP value of a certain feature is less than the threshold 0.1, then this feature is deleted to reduce the model calculation complexity.

[0016] The feature selection unit combines historical maintenance data and uses principal component analysis (PCA) to reduce the dimension of high-dimensional feature data and reduce redundant features. The PCA calculation process is as follows: Standardize the feature matrix X: where: μ is the mean value and σ is the standard deviation; Calculate the covariance matrix C: Calculate the eigenvalue decomposition: C = VAV T ; Select the first k features so that the cumulative variance contribution rate: where λ is the eigenvalue and k represents the number of selected features; Finally, transmit the optimized feature data set to the prediction analysis unit.

[0017] Based on the screened and optimized feature data, the analysis unit constructs a long short-term memory (LSTM) network prediction model to calculate the capacitance percentage loss ratio and calculate its remaining useful life (RUL). The long short-term memory (LSTM) network prediction model constructs two layers of LSTM units, each containing 128 and 64 neurons, uses the ReLU activation function, and through the MSE loss function and the Adam optimization method, improves the model training efficiency. Based on the loss time series data, it learns the long-term attenuation pattern of the capacitor. The XGBoost classification model is used and trained through gradient boosting trees (GBTs) to process the screened historical capacitor loss data and variables, classify and predict the health state of the capacitor to obtain the specific percentage loss ratio and the remaining useful life (RUL) of the capacitor. The calculation formula is as follows: where C h is the critical capacitance when the capacitor reaches failure, and λ is the loss rate parameter. The percentage loss ratio and the RUL calculation results calculated by the prediction analysis unit are transmitted to the diagnosis module.

[0018] The diagnosis module classifies the operating state of the capacitor based on the percentage loss ratio and the remaining useful life (RUL) calculation results provided by the analysis module. The health state classification includes four levels, namely, healthy state capacitor with 0%-30% loss, slightly attenuated state capacitor with 30%-60% loss, severely attenuated state capacitor with 60%-90% loss, and about-to-fail state capacitor with ≥90% loss;

[0019] When the capacitor is in a healthy state, the operating temperature, voltage, and current are within the normal range. The system adopts a periodic monitoring mechanism, sets the monitoring period to every 6 months, regularly detects the changes in ESR and tanδ. If the increase in ESR exceeds 5%, the system adjusts the monitoring frequency. When it is detected that the loss rate accelerates, the system upgrades to the slightly attenuated state monitoring mode;

[0020] When the capacitor is in a mildly degraded state, the regulation strategy includes a dynamic operating environment optimization mechanism that adjusts the operating voltage of the capacitor, reduces the high-temperature operating time, and mitigates the risk of increased losses. At the same time, the monitoring period is adjusted to every three months, and the LSTM prediction model is used to calculate the change trends of ESR and capacitance in the next year. If the predicted loss may exceed 60% within six months, the system upgrades to the severely degraded state monitoring mode;

[0021] When the capacitor is in a severely degraded state, the regulation strategy includes a short-term optimization mechanism and a fault prediction mechanism. The short-term optimization mechanism adopts an active temperature control strategy, including air cooling and liquid cooling, to reduce the operating temperature, slow down the loss rate, adjust the load current distribution, reduce the inrush current, and reduce the occurrence of capacitor overload. The fault prediction mechanism uses an exponential decay model to calculate the remaining useful life (RUL) of the capacitor. If the RUL is less than six months, the system upgrades to the near-failure state monitoring mode;

[0022] When the capacitor is in the near-failure state, the regulation strategy includes an emergency replacement strategy and a redundant switching mechanism. The emergency replacement strategy triggers a replacement plan and executes the replacement task through the decision-making module. If the redundant switching mechanism detects that the capacitor state reaches the near-failure level, the system enables the standby capacitor and switches to the redundant circuit.

[0023] The decision-making module executes specific maintenance strategies based on the grading results of the diagnostic module. The maintenance strategies include a preventive maintenance mechanism, a predictive replacement strategy, and an emergency replacement strategy to enable timely maintenance of the capacitor in different health states;

[0024] For capacitors with healthy state losses and mildly degraded capacitors, the preventive maintenance mechanism performs periodic health assessments by the system if accelerated losses are predicted;

[0025] For capacitors in the severely degraded state, the predictive replacement strategy combines the LSTM prediction analysis model to calculate the remaining useful life (RUL). If the RUL is less than six months, the system sends a replacement task to the operation and maintenance personnel, and the operation and maintenance personnel complete the capacitor replacement within the planned downtime window;

[0026] For capacitors in the near-failure state, when the system detects that the ESR rises sharply by more than 50%, the tanδ increases rapidly, or the capacitance drops by more than 30%, the emergency replacement strategy triggers an emergency replacement plan, shuts down, and notifies the operation and maintenance personnel to complete the capacitor replacement during the shutdown.

[0027] The present invention provides a capacitor fault prediction and diagnosis system based on loss data. Compared with the prior art, it has the following beneficial effects:

[0028] Based on the optimized feature data after screening, the analysis unit of the present invention calculates the percentage loss ratio and the remaining useful life (RUL) of the capacitor, and further executes different strategies at different levels. By analyzing the internal parameters C, ESR, and tanδ of the capacitor, the loss trend of the capacitor is effectively calculated, potential faults are identified in advance, thereby improving the accuracy of capacitor fault prediction and ensuring the long-term stable operation of the equipment.

[0029] According to the classification of the health status of the capacitor, the present invention formulates a preventive maintenance mechanism, a predictive replacement strategy, and an emergency replacement strategy. The intelligent decision-making module provides corresponding processing solutions for capacitors in different attenuation stages. By dynamically adjusting the monitoring frequency, measures can be taken in a timely manner when the capacitor shows mild attenuation, reducing the risk of increased loss, and starting the emergency replacement strategy when the capacitor reaches the state of impending failure, thereby effectively reducing the equipment downtime and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of the principle framework of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0032] Please refer to Figure 1 , the present application provides a capacitor fault prediction and diagnosis system based on loss data, including a data acquisition module, an analysis module, a diagnosis module, and a decision module;

[0033] The data acquisition module is responsible for obtaining the capacitor parameter data of the capacitor, collecting the internal parameters of the capacitor through a capacitor sensor array, and transmitting the capacitor parameter data to the data preprocessing module;

[0034] The data preprocessing module further preprocesses the capacitor parameter data transmitted from the data acquisition module. The data processing includes signal denoising and data alignment. The processed capacitor parameter data is transmitted to the analysis module;

[0035] The analysis module includes a feature selection unit and an analysis unit. The feature selection unit performs feature selection based on the data from the data preprocessing module and transmits the optimized feature data set to the prediction analysis unit; the prediction analysis unit constructs a long short-term memory network (LSTM) prediction model based on the screened optimized feature data, and the calculated percentage loss ratio and predicted remaining useful life (RUL) of the capacitor are transmitted to the diagnosis module;

[0036] Based on the percentage loss ratio and the remaining useful life (RUL) provided by the analysis module, the diagnosis module classifies the operating state of the capacitor, and the capacitor health state is divided into four levels;

[0037] Based on the health state classification result provided by the diagnosis module, the decision-making module executes specific maintenance strategies, including preventive maintenance mechanisms, predictive replacement strategies, and emergency replacement strategies.

[0038] The data acquisition module extracts the parameters of the capacitor operating state, obtains the internal parameters C, ESR, and tanδ of the capacitor through the capacitor sensor array, monitors the capacitor temperature T through the temperature sensor thermistor NTC, measures the working voltage U and current I through the Hall effect sensor. The capacitor sensor array uses a high-precision LCR digital bridge with an accuracy of 0.01%, and the sampling frequency is set to 1 kHz; the capacitor parameter data includes the internal parameters C, ESR, and tanδ of the capacitor, the capacitor temperature T, the working voltage U, and the current I, and the capacitor parameter data is transmitted to the data preprocessing module through the communication protocol.

[0039] The capacitor sensor array is used to accurately measure the internal parameters (C, ESR, tanδ) of the capacitor. The LCR digital bridge has an accuracy of up to 0.01%, and the sampling frequency can be set to 1 kHz, which can meet the requirements of high-precision and high-frequency measurement of capacitor parameters, Precision Sensing and Actuation Laboratory;

[0040] The temperature sensor (thermistor NTC) is used to monitor the temperature (T) of the capacitor. The NTC thermistor has temperature-sensitive characteristics and can provide accurate temperature measurement within a certain temperature range;

[0041] The Hall effect sensor is used to measure the current I and the working voltage U. The Hall effect sensor can detect the current and voltage in a non-contact manner, avoiding the errors and safety hazards that may be introduced by traditional contact measurements.

[0042] The data preprocessing module preprocesses the capacitor parameter data. The data processing includes signal denoising and data alignment; signal denoising includes using wavelet transform (WaveletDenoising) to remove high-frequency noise from the ESR and tanδ data, using adaptive Kalman filtering (AdaptiveKalmanFiltering, AKF) to smooth the signal for the capacitance C, and using a Butterworth low-pass filter with a cut-off frequency of 0.1 Hz to filter out environmental interference signals for the temperature data. Data alignment includes using linear interpolation (LinearInterpolation) for time alignment, and transmitting the processed capacitor parameter data to the analysis module.

[0043] Wavelet transform is used to denoise the data of the equivalent series resistance (ESR) and loss tangent (tanδ) of the capacitor. Wavelet transform can analyze signals at multiple resolutions, effectively remove high-frequency noise, and retain the main features of the signals; Adaptive Kalman filter is used to smooth the capacitance (C) signal. Adaptive Kalman filter can adjust the filtering parameters according to the dynamic changes of the system to improve the filtering accuracy.

[0044] The analysis module includes a feature selection unit and an analysis unit;

[0045] The feature selection unit extracts the key features in the capacitor loss data, reduces data redundancy by screening high-influence features, improves the calculation efficiency and classification accuracy of the XGBoost classification model, calculates the contribution degree of each feature to the XGBoost classification result through SHAP (Shapley Additive Explanations) feature importance analysis, screens out the core parameters affecting the capacitor health status, and removes the features with a contribution degree lower than the set threshold such as 0.1 to optimize the model input feature set. The calculation method of the SHAP value is as follows: Among them, D is the SHAP importance value of feature i, F is all feature sets, f(S) is the model output containing subset S. If the SHAP value of a certain feature is less than the threshold 0.1, then delete this feature to reduce the model calculation complexity.

[0046] The feature selection unit combines historical maintenance data and uses principal component analysis (PCA) to reduce the dimensionality of high-dimensional feature data and reduce redundant features. The PCA calculation process is as follows: Standardize the feature matrix X: Among them: μ is the mean value, and σ is the standard deviation; Calculate the covariance matrix C: Calculate the eigenvalue decomposition: C = VAV T ; Select the first k features so that the cumulative variance contribution rate: Among them, λ is the eigenvalue, and k represents the number of selected features; Finally, transfer the optimized feature data set to the prediction and analysis unit.

[0047] The analysis unit constructs a long short-term memory network (LSTM) prediction model based on the filtered and optimized feature data to calculate the percentage loss ratio of the capacitor and calculate its remaining useful life (RUL). The long short-term memory network (LSTM) prediction model constructs two layers of LSTM units, which respectively contain 128 and 64 neurons, uses the ReLU activation function, and improves the model training efficiency through the MSE loss function and the Adam optimization method. Based on the loss time series data, learn the long-term attenuation pattern of the capacitor. Adopt the XGBoost classification model and train it through the gradient boosting tree (GBT) to process the filtered capacitor loss historical data and variables, classify and predict the health status of the capacitor to obtain the specific percentage loss ratio and the remaining useful life (RUL) of the capacitor. The calculation formula is as follows: Among them, C h is the critical capacitance at which the capacitor fails, λ is the loss rate parameter, and the percentage loss ratio calculated by the prediction and analysis unit and the RUL calculation result are transmitted to the diagnosis module.

[0048] LSTM is a special type of Recurrent Neural Network (RNN) that can learn and remember long-term dependencies, configured as follows:

[0049] Network structure: It contains two layers of LSTM units. The first layer has 128 neurons and the second layer has 64 neurons;

[0050] Activation function: The ReLU (Rectified Linear Unit) activation function is used to enhance the non-linear expression ability of the model;

[0051] Loss function: The mean squared error (MSE) loss function is adopted to measure the difference between the predicted value and the true value;

[0052] Optimization algorithm: The Adam optimizer is used, which combines the advantages of momentum and adaptive learning rate to accelerate the convergence speed of the model.

[0053] Based on the percentage loss ratio and the remaining useful life (RUL) calculation result provided by the analysis module, the diagnosis module classifies the operating state of the capacitor. The health state classification includes four levels, namely, healthy state capacitor with 0%-30% loss, slightly attenuated state capacitor with 30%-60% loss, severely attenuated state capacitor with 60%-90% loss, and near-failure state capacitor with ≥90% loss;

[0054] When the capacitor is in a healthy state, the operating temperature, voltage, and current are within the normal range. The system adopts a periodic monitoring mechanism with a monitoring period set to every 6 months. Regularly detect the changes in ESR and tanδ. If the increase in ESR exceeds 5%, the system adjusts the monitoring frequency. When it detects that the loss rate accelerates, the system upgrades to the slightly attenuated state monitoring mode;

[0055] When the capacitor is in the slightly attenuated state, the control strategy includes a dynamic working environment optimization mechanism to adjust the working voltage of the capacitor, reduce the high-temperature operation time, and reduce the risk of increased loss. At the same time, the monitoring period is adjusted to every 3 months. Use the LSTM prediction model to calculate the future 1-year trends of ESR and capacitance changes. If it is predicted that the loss may exceed 60% within 6 months, the system upgrades to the severely attenuated state monitoring mode;

[0056] When the capacitor is in a severely attenuated state, the regulation strategy includes a short-term optimization mechanism and a fault prediction mechanism. The short-term optimization mechanism adopts an active temperature control strategy, including air cooling and liquid cooling, to reduce the operating temperature, slow down the loss rate, adjust the load current distribution, reduce the inrush current, and reduce the occurrence of capacitor overload. The fault prediction mechanism uses an exponential decay model to calculate the remaining useful life (RUL) of the capacitor. If the RUL is less than 6 months, the system is upgraded to the near-failure state monitoring mode;

[0057] When the capacitor is in the near-failure state, the regulation strategy includes an emergency replacement strategy and a redundant switching mechanism. The emergency replacement strategy triggers a replacement plan and executes the replacement task through a decision-making module. If the redundant switching mechanism detects that the capacitor state reaches the near-failure level, the system enables the standby capacitor and switches to the redundant circuit.

[0058] The decision-making module executes specific maintenance strategies according to the grading results of the diagnostic module. The maintenance strategies include a preventive maintenance mechanism, a predictive replacement strategy, and an emergency replacement strategy, enabling timely maintenance of the capacitor in different health states;

[0059] For capacitors with healthy state loss and mild attenuation, the preventive maintenance mechanism performs periodic health assessments by the system if it predicts accelerated loss;

[0060] For capacitors in the severely attenuated state, the predictive replacement strategy combines the LSTM prediction analysis model by the system to calculate the remaining useful life (RUL). If the RUL is less than 6 months, the system sends a replacement task to the operation and maintenance personnel, and the operation and maintenance personnel complete the capacitor replacement within the planned downtime window;

[0061] For capacitors in the near-failure state, the emergency replacement strategy triggers an emergency replacement plan when the system detects that the ESR rises sharply by >50%, the tanδ increases rapidly, or the capacitance drops by more than 30%, shuts down, and notifies the operation and maintenance personnel to complete the capacitor replacement during the shutdown.

[0062] Furthermore, the present invention formulates a preventive maintenance mechanism, a predictive replacement strategy, and an emergency replacement strategy according to the health state classification of the capacitor, provides corresponding treatment solutions for capacitors in different attenuation stages through an intelligent decision-making module, can take timely measures when the capacitor shows mild attenuation by dynamically adjusting the monitoring frequency, reduce the risk of increased loss, and start the emergency replacement strategy when the capacitor reaches the near-failure state, thereby effectively reducing the equipment downtime and maintenance costs.

[0063] Some of the data in the above formulas are numerically calculated after removing their dimensions, and the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

[0064] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those 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 capacitance fault prediction and diagnosis system based on loss data, characterized in that, It includes: A data acquisition module, an analysis module, a diagnosis module, and a decision-making module; The data acquisition module is responsible for obtaining the capacitance parameter data of the capacitor, collecting the internal parameters of the capacitor through a capacitor sensor array, and transmitting the capacitance parameter data to the data preprocessing module; The data preprocessing module further preprocesses the capacitance parameter data transmitted from the data acquisition module. The data processing includes signal denoising and data alignment. The processed capacitance parameter data is transmitted to the analysis module; The analysis module includes a feature selection unit and an analysis unit. The feature selection unit performs feature selection based on the data from the data preprocessing module and transmits the optimized feature data set to the prediction analysis unit. The prediction analysis unit constructs a long short-term memory network (LSTM) prediction model based on the filtered optimized feature data, and the calculated percentage loss ratio and predicted remaining useful life (RUL) of the capacitor are transmitted to the diagnosis module; The diagnosis module classifies the operating state of the capacitor based on the percentage loss ratio and remaining useful life (RUL) provided by the analysis module. The capacitor health state is divided into four grades; The decision-making module executes specific maintenance strategies according to the health state classification results provided by the diagnosis module. The maintenance strategies include a preventive maintenance mechanism, a predictive replacement strategy, and an emergency replacement strategy.

2. The capacitance fault prediction and diagnosis system based on loss data according to claim 1, wherein The data acquisition module extracts the parameters of the capacitor operating state, obtains the internal parameters C, ESR, and tanδ of the capacitor through a capacitor sensor array, monitors the capacitor temperature T through a temperature sensor thermistor NTC, measures the working voltage U and current I through a Hall effect sensor. The capacitor sensor array uses a high-precision LCR digital bridge with an accuracy of 0.01%, and the sampling frequency is set to 1 kHz. The capacitance parameter data includes the internal parameters C, ESR, and tanδ of the capacitor, the capacitor temperature T, the working voltage U, and the current I. The capacitance parameter data is transmitted to the data preprocessing module through a communication protocol.

3. The capacitance fault prediction and diagnosis system based on loss data according to claim 1, wherein The data preprocessing module performs data preprocessing on the capacitance parameter data. The data processing includes signal denoising and data alignment. The signal denoising includes using wavelet transform (WaveletDenoising) to remove high-frequency noise from the ESR and tanδ data, using adaptive Kalman filtering (AdaptiveKalmanFiltering, AKF) to smooth the signal of the capacitance C, and using a Butterworth low-pass filter with a cut-off frequency of 0.1 Hz to filter out environmental interference signals for the temperature data. The data alignment includes using linear interpolation (LinearInterpolation) for time alignment and transmitting the processed capacitance parameter data to the analysis module.

4. The capacitance fault prediction and diagnosis system based on loss data according to claim 1, characterized in that The analysis module includes a feature selection unit and an analysis unit; The feature selection unit extracts key features from the capacitance loss data, reduces data redundancy by screening high-influence features, improves the computational efficiency and classification accuracy of the XGBoost classification model, calculates the contribution degree of each feature to the XGBoost classification result through SHAP (Shapley Additive Explanations) feature importance analysis, screens out the core parameters affecting the capacitance health state, and removes features with a contribution degree lower than a set threshold such as 0.1 to optimize the model input feature set. The calculation method of the SHAP value is as follows: Where D is the SHAP importance value of feature i, F is the set of all features, f(S) is the model output containing subset S. If the SHAP value of a certain feature is less than the threshold 0.1, then this feature is deleted to reduce the model computational complexity.

5. The capacitance fault prediction and diagnosis system based on loss data according to claim 1, wherein The feature selection unit combines historical maintenance data and uses principal component analysis (PCA) to reduce the dimension of high-dimensional feature data and reduce redundant features. The PCA calculation process is as follows: Standardize the feature matrix X: where μ is the mean and σ is the standard deviation; Calculate the covariance matrix C: Calculate the eigenvalue decomposition: C = VAV T ; Select the first k features so that the cumulative variance contribution rate: where λ is the eigenvalue and k represents the number of selected features; Finally, transfer the optimized feature data set to the prediction analysis unit.

6. The capacitance fault prediction and diagnosis system based on loss data according to claim 1, characterized in that Based on the filtered optimized feature data, the analysis unit constructs a long short-term memory network (LSTM) prediction model to calculate the capacitance percentage loss ratio and calculates its remaining useful life (RUL). For the long short-term memory network (LSTM) prediction model, two layers of LSTM units are constructed, each containing 128 and 64 neurons respectively. The ReLU activation function is used, and through the MSE loss function and the Adam optimization method, the model training efficiency is improved. Based on the loss time series data, the long-term attenuation pattern of the capacitance is learned. The XGBoost classification model is adopted and trained by gradient boosting trees (GBT) to process the filtered historical capacitance loss data and variables, and classify and predict the health state of the capacitance to obtain the specific percentage loss ratio and the remaining useful life (RUL) of the capacitance. The calculation formula is as follows: where C h is the critical capacitance when the capacitor reaches failure, λ is the loss rate parameter, and the percentage loss ratio and RUL calculation results calculated by the prediction and analysis unit are transmitted to the diagnosis module.

7. The capacitance fault prediction and diagnosis system based on loss data according to claim 1, wherein The diagnosis module classifies the operating state of the capacitor based on the percentage loss ratio and the remaining useful life (RUL) calculation results provided by the analysis module. The health state classification includes four levels, namely, capacitors in a healthy state with 0%-30% loss, capacitors in a slightly attenuated state with 30%-60% loss, capacitors in a severely attenuated state with 60%-90% loss, and capacitors in an impending failure state with ≥90% loss; When the capacitor is in a healthy state, the operating temperature, voltage, and current are within the normal range. The system adopts a periodic monitoring mechanism with a monitoring period set at every six months to regularly detect changes in ESR and tanδ. If the increase in ESR exceeds 5%, the system adjusts the monitoring frequency. When it detects an accelerating loss rate, the system upgrades to the monitoring mode for the slightly attenuated state; When the capacitor is in a slightly attenuated state, the control strategy includes a dynamic working environment optimization mechanism to adjust the working voltage of the capacitor, reduce the high-temperature operation time, and reduce the risk of increased loss. At the same time, the monitoring period is adjusted to every three months. The LSTM prediction model is used to calculate the change trends of ESR and capacitance in the next year. If it is predicted that the loss may exceed 60% within six months, the system upgrades to the monitoring mode for the severely attenuated state; When the capacitor is in a severely attenuated state, the control strategy includes a short-term optimization mechanism and a fault prediction mechanism. The short-term optimization mechanism adopts an active temperature control strategy, including air cooling and liquid cooling, to reduce the operating temperature, slow down the loss rate, adjust the load current distribution, reduce the impact current, and reduce the occurrence of capacitor overload. The fault prediction mechanism uses an exponential decay model to calculate the remaining useful life (RUL) of the capacitor. If the RUL is less than six months, the system upgrades to the monitoring mode for the impending failure state; When the capacitor is in an impending failure state, the control strategy includes an emergency replacement strategy and a redundant switching mechanism. The emergency replacement strategy triggers a replacement plan and the decision-making module executes the replacement task. If the redundant switching mechanism detects that the capacitor state reaches the impending failure level, the system enables the standby capacitor and switches to the redundant circuit.

8. The capacitance fault prediction and diagnosis system based on loss data according to claim 1, wherein The decision-making module executes specific maintenance strategies according to the classification results of the diagnosis module. The maintenance strategies include a preventive maintenance mechanism, a predictive replacement strategy, and an emergency replacement strategy to enable timely maintenance of the capacitor in different health states; The preventive maintenance mechanism is for capacitors with healthy state loss and slightly attenuated capacitors. The system performs periodic health assessments if it predicts an accelerating loss; The predictive replacement strategy is for capacitors in a severely attenuated state. The system combines the LSTM prediction analysis model to calculate the remaining useful life (RUL). If the RUL is less than six months, the system sends a replacement task to the operation and maintenance personnel, and the operation and maintenance personnel complete the capacitor replacement within the planned downtime window; The emergency replacement strategy is for capacitors in an impending failure state. When the system detects a sharp increase in ESR > 50%, a rapid increase in tanδ, or a capacitance decrease exceeding 30%, it triggers an emergency replacement plan, shuts down, and immediately notifies the operation and maintenance personnel to complete the capacitor replacement during the shutdown.

Images