Equipment fault prediction method in power transmission and transformation system

Abstract

The invention discloses an equipment fault prediction method in a power transmission and transformation system, and relates to the technical field of power system operation and maintenance, and the fault prediction method comprises the following steps: data collection: collecting the data of power transmission and transformation equipment through a sensor network, and forming a multi-source heterogeneous data set; data preprocessing: performing abnormal value elimination, missing value filling and standardization processing on the multi-source heterogeneous data set; feature extraction: mining associated features and time-space evolution laws of data of different dimensions by adopting a multi-modal fusion network based on an attention mechanism; performing fault prediction, and constructing a dynamic threshold early warning model in combination with an equipment aging curve; and correcting the result to generate a final fault prediction report. The equipment fault prediction method in the power transmission and transformation system solves the problems that traditional prediction depends on single data, the fixed threshold value false alarm is high, and the advance is insufficient, improves the fault prediction accuracy and reliability, and is suitable for operation and maintenance early warning of power transmission and transformation equipment such as transformers and circuit breakers.

Description

Technical Field

[0001] This invention relates to the field of power system operation and maintenance technology, specifically to a method for predicting equipment faults in power transmission and transformation systems. Background Technology

[0002] The power transmission and transformation system is a core component of the power system, undertaking the key functions of power transmission and voltage transformation. It connects power plants and users, ensuring the safe and efficient transmission of electricity. The system mainly includes transmission lines (such as high-voltage overhead lines and cables) and substations (including transformers, switchgear, etc.). Through transformers, the low-voltage electricity output from the power plant is raised to high voltage to achieve long-distance, low-loss transmission, and then stepped down to the low voltage that users can use. Its operation relies on real-time monitoring by the dispatching system and has overload protection and fault isolation functions. It widely serves industrial production, residential life and other fields, and is an important infrastructure for maintaining the stability of social power supply.

[0003] In existing technologies, such as the equipment fault prediction method in power transmission and transformation systems (application number 202410289593.1), the technical solution includes the following steps: Step 1: Equipment monitoring; Step 2: Data processing; Step 3: Data analysis; Step 4: Data retrieval; Step 5: Data comparison; Step 6: Data evaluation; Step 7: Fault prediction. In this invention, during the operation of a power transmission and transformation system, the monitoring module monitors the operating status of key equipment in real time, the data analysis module analyzes the potential fault modes and trends of each piece of equipment monitored by the monitoring module, and the data retrieval module retrieves historical fault information stored in the database module and the data analyzed by the data analysis module, compares the results with the analyzed data, and evaluates and judges the comparison results to complete the equipment fault prediction.

[0004] As the core link in power transmission, the stable operation of equipment such as transformers, circuit breakers, and insulators in the power transmission and transformation system directly affects the safety of the power grid. Traditional fault prediction methods have the following defects: they rely on a single data source, such as only using electrical quantity monitoring, ignoring key influencing factors such as environment and vibration, resulting in one-sided predictions; they use fixed thresholds for early warning, without considering dynamic factors such as equipment aging and changes in operating conditions, which easily leads to false alarms or missed alarms; and the feature extraction methods are static, unable to capture the spatiotemporal evolution of equipment status, resulting in insufficient prediction lead time.

[0005] In view of this, in-depth research was conducted on the above issues, which led to the creation of this case.

[0006] To address the aforementioned issues, an innovative design was developed based on existing equipment fault prediction methods in power transmission and transformation systems. Summary of the Invention

[0007] The purpose of this invention is to provide a method for predicting equipment faults in power transmission and transformation systems, in order to solve the problems mentioned in the background art, such as reliance on a single data source, use of fixed thresholds for early warning, and insufficient prediction lead time.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A method for predicting equipment faults in a power transmission and transformation system, the method comprising the following steps: Step 1: Data acquisition. The operating parameters, environmental parameters, and historical fault records of power transmission and transformation equipment are collected through a sensor network to form a multi-source heterogeneous dataset. Step 2: Data preprocessing, including outlier removal, missing value imputation, and standardization of the multi-source heterogeneous dataset to construct time series samples; Step 3, feature extraction: A multimodal fusion network based on an attention mechanism is used to mine the correlation features and spatiotemporal evolution patterns of data from different dimensions, and output the device state feature vector. Step 4, Fault prediction: Construct a dynamic threshold early warning model by combining the equipment aging curve, input the equipment state feature vector into the model, and output the fault probability and prediction time window. Step 5: Result correction. An expert experience knowledge base is introduced to iteratively optimize the fault probability and prediction time window, generating the final fault prediction report.

[0009] Preferably, the operating parameters collected in step one include electrical quantity parameters, mechanical vibration parameters, and temperature parameters of the power transmission and transformation equipment. The electrical quantity parameters include voltage, current, and power factor. The mechanical vibration parameters include vibration frequency and amplitude. The temperature parameters include equipment body temperature and winding temperature. The environmental parameters include ambient temperature, humidity, and wind speed.

[0010] Using the above technical solution, the operating parameters and environmental parameters collected in step one form a multi-dimensional data system, which covers all factors affecting equipment operation compared to monitoring a single electrical quantity. For example, vibration parameters can detect potential mechanical jamming in advance, and environmental humidity can reflect the risk of accelerated insulation aging. This provides comprehensive data support for subsequent accurate feature extraction and fault prediction, avoiding prediction deviations caused by incomplete data.

[0011] Preferably, in step two, the isolated forest algorithm is used to remove outliers. The number of trees in the isolated forest algorithm is set to 100-200, and the sample subset size is set to 20-50. The improved KNN algorithm is used to fill in missing values. The number of neighbors in the improved KNN algorithm is set to 5-10. The standardization process maps the data to the [0,1] interval.

[0012] Using the above technical solution, step two employs the isolated forest algorithm to efficiently remove abnormal data such as sensor malfunctions, and the improved KNN algorithm with 5-10 neighbors fills in missing values. Combined with standardization in the [0,1] interval, the quality of time series sample data is ensured, providing reliable input for the feature extraction stage and reducing the interference of data noise on the prediction results.

[0013] Preferably, the multimodal fusion network based on the attention mechanism in step three includes a modal feature encoder, a cross-modal attention layer, and a time-series feature decoder. The modal feature encoder designs branch networks for different types of data. It uses a 1D convolutional neural network to extract features for electrical parameters, a Fourier transform combined with a convolutional neural network to extract features for mechanical vibration parameters, and a fully connected network to extract features for temperature parameters and environmental parameters.

[0014] Using the above technical solution, the modal feature encoder in step three designs dedicated branches for different data types, which can maximize the retention of the core information of each modal data, improve the single-modal feature representation capability compared with general feature extraction methods, and lay a high-quality foundation for cross-modal fusion.

[0015] Preferably, the cross-modal attention layer calculates the association weights of different modal features through a self-attention mechanism. The calculation of the association weights is based on the inter-modal mutual information entropy, and the formula for calculating the inter-modal mutual information entropy is as follows: I(X,Y)= Where X and Y are different modal feature datasets, p(x,y) is the joint probability density, p(x) and p(y) are the marginal probability densities, and the weight values ​​range from 0 to 1. For modal features with mutual information entropy greater than 0.6, their association weights are increased to achieve dynamic adaptive fusion of different modal features.

[0016] Using the above technical solution, the cross-modal attention layer calculates the correlation weights based on mutual information entropy. The weights of highly correlated features with mutual information entropy greater than 0.6 are increased to achieve dynamic adaptive fusion, avoiding the submergence of key information caused by traditional fixed weight fusion. For example, after the weights of the correlation features between environmental humidity and insulation aging are increased, the insulation fault identification rate is improved, enhancing the effectiveness of fused features.

[0017] Preferably, the temporal feature decoder uses a bidirectional long short-term memory network, the number of hidden layers of the memory network is set to 2-4 layers, the number of hidden layer neurons is set to 128-256, the Adam optimizer is used for training, the learning rate is set to 0.001-0.005, the number of training iterations is set to 100-200 times, and the output dimension is a device state feature vector of 64-128 dimensions.

[0018] Using the above technical solution, the bidirectional long short-term memory network in step three, combined with the Adam optimizer and 100-200 iterations of training, can deeply explore the spatiotemporal evolution of data, such as the temporal correlation between load fluctuations and equipment overheating, and output 64-128 dimensional equipment state feature vectors. Compared with static feature extraction methods, it improves the ability to capture temporal features and extends the lead time for fault prediction.

[0019] Preferably, the equipment aging curve in step four is fitted using a Weibull distribution. During the fitting process, the equipment's operating years, cumulative load, and maintenance frequency are introduced as influencing factors. The shape parameter of the Weibull distribution is set to 1.5-3.0, and the scale parameter is dynamically adjusted according to the equipment type and operating environment.

[0020] Using the above technical solution, step four uses Weibull distribution to fit the equipment aging curve, and introduces influencing factors such as years of operation, cumulative load, and number of maintenance times. This can accurately quantify the trend of equipment performance degradation. For example, the calculation error of the aging degradation coefficient of a transformer that has been in operation for ten years is reduced. Compared with the prediction model that does not consider aging, it is more in line with the changes in the state of the equipment throughout its entire life cycle, and provides a scientific basis for setting dynamic thresholds.

[0021] Preferably, the dynamic threshold early warning model in step four includes a basic threshold module, an aging correction module, and an operating condition adaptation module. The basic threshold module sets an initial early warning threshold based on the equipment's factory technical parameters and power industry standards. The aging correction module calculates the aging attenuation coefficient according to the equipment's aging curve and corrects the initial early warning threshold. The operating condition adaptation module combines real-time environmental parameters and operating load rate to perform a secondary adjustment on the corrected threshold.

[0022] By adopting the above technical solution, the basic threshold module, aging correction module, and operating condition adaptation module of the dynamic threshold early warning model form a three-level threshold adjustment mechanism, which can avoid false alarms or missed alarms of fixed thresholds under equipment aging or harsh operating conditions, thereby improving the accuracy of early warning.

[0023] Preferably, the expert experience knowledge base in step five includes historical cases of power transmission and transformation equipment failures, and each case includes the equipment parameters, environmental conditions, failure type, and handling solution at the time of the failure.

[0024] Using the above technical solution, the expert experience knowledge base in step five contains complete fault cases, which can provide reference for real operation and maintenance scenarios, such as historical insulation fault cases under similar humidity and load, providing authoritative data support for the correction of prediction results and making up for the prediction limitations of pure algorithm models under complex and special working conditions.

[0025] Preferably, step five uses a Bayesian inference algorithm for iterative optimization, matching the prediction results with similar cases in the knowledge base, calculating the posterior probability, and correcting the initial fault probability.

[0026] Using the above technical solution, step five employs a Bayesian inference algorithm to calculate the posterior probability by matching similar cases in the knowledge base. This iteratively optimizes the initial fault probability, improving confidence compared to uncorrected predictions and reducing the risk of misjudgment due to algorithmic bias.

[0027] Compared with the prior art, the beneficial effects of the present invention are: the equipment fault prediction method in the power transmission and transformation system, 1. An adaptive multimodal data fusion algorithm based on attention mechanism to improve fault feature representation capability: This algorithm retains the core information of each data by extracting modality-specific features, and then dynamically calculates the correlation weight based on mutual information entropy. For highly correlated features, such as high temperature and partial discharge, humidity and insulation aging, the weight is adaptively increased to avoid the key information being submerged due to traditional fixed fusion. Compared with single-modal or static fusion methods, the fault identification of fused features is improved. It can accurately capture the implicit features of multiple types of faults, provide high-quality input for fault prediction, and improve the fault identification accuracy. 2. A dynamic threshold early warning model based on equipment aging curves ensures the adaptability and reliability of early warning systems. The model uses the aging curve fitted by the Weibull distribution as its core to calculate the equipment performance degradation coefficient. Combined with the basic threshold and real-time operating conditions, it realizes three-level dynamic adjustment of the threshold. Compared with the traditional fixed threshold, it can adapt to the entire life cycle status of the equipment and complex operating conditions, effectively solving the problem of false alarms or missed alarms caused by equipment aging and changes in operating conditions. This reduces the false alarm rate of early warning and extends the fault prediction lead time to 72 hours, allowing sufficient preparation time for operation and maintenance. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the fault prediction method of the present invention. Detailed Implementation

[0029] 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.

[0030] Please see Figure 1 The present invention provides a technical solution: A method for predicting equipment faults in power transmission and transformation systems, comprising the following steps: Step 1: Data acquisition. The operating parameters, environmental parameters, and historical fault records of power transmission and transformation equipment are collected through a sensor network to form a multi-source heterogeneous dataset. Step 2: Data preprocessing, including outlier removal, missing value imputation, and standardization of the multi-source heterogeneous dataset to construct time series samples; Step 3, feature extraction: A multimodal fusion network based on an attention mechanism is used to mine the correlation features and spatiotemporal evolution patterns of data from different dimensions, and output the device state feature vector. Step 4: Fault prediction. A dynamic threshold early warning model is constructed by combining the equipment aging curve. The equipment status feature vector is input into the model, and the fault probability and prediction time window are output. Step 5: Result correction. An expert knowledge base is introduced to iteratively optimize the failure probability and prediction time window, generating the final failure prediction report.

[0031] The operating parameters collected in Step 1 include electrical parameters, mechanical vibration parameters, and temperature parameters of the power transmission and transformation equipment. The electrical parameters include voltage, current, and power factor; the mechanical vibration parameters include vibration frequency and amplitude; the temperature parameters include the equipment body temperature and winding temperature; and the environmental parameters include ambient temperature, humidity, and wind speed. The operating parameters collected in Step 1, together with the environmental parameters, form a multi-dimensional data system. Compared with monitoring only electrical quantities, this system covers all factors that affect equipment operation. For example, vibration parameters can detect potential mechanical jamming in advance, and ambient humidity can reflect the risk of accelerated insulation aging. This provides comprehensive data support for subsequent accurate feature extraction and fault prediction, avoiding prediction deviations caused by incomplete data.

[0032] In step two, the Isolation Forest algorithm is used to remove outliers. The number of trees in the Isolation Forest algorithm is set to 100-200, and the sample subset size is set to 20-50. The improved KNN algorithm is used to fill in missing values. The number of neighbors in the improved KNN algorithm is set to 5-10. Standardization is performed to map the data to the [0,1] interval. Step two uses the Isolation Forest algorithm to efficiently remove abnormal data such as sensor malfunctions. The improved KNN algorithm with 5-10 neighbors fills in missing values. Combined with standardization in the [0,1] interval, the quality of time series sample data is ensured, providing reliable input for the feature extraction stage and reducing the interference of data noise on the prediction results.

[0033] Step three involves a multimodal fusion network based on an attention mechanism, comprising a modal feature encoder, a cross-modal attention layer, and a time-series feature decoder. The modal feature encoder designs branch networks for different data types: 1D convolutional neural networks are used to extract features for electrical parameters; Fourier transform combined with convolutional neural networks is used to extract features for mechanical vibration parameters; and fully connected networks are used to extract features for temperature and environmental parameters. The cross-modal attention layer calculates the association weights of different modal features through a self-attention mechanism. The calculation of these association weights is based on the intermodal mutual information entropy, and the formula for calculating the intermodal mutual information entropy is: I(X,Y)= Where X and Y are different modal feature datasets, p(x,y) is the joint probability density, and p(x) and p(y) are the marginal probability densities. The weight values ​​range from 0 to 1. For modal features with mutual information entropy greater than 0.6, their association weights are increased to achieve dynamic adaptive fusion of different modal features. The temporal feature decoder adopts a bidirectional long short-term memory network with 2-4 hidden layers and 128-256 neurons. It is trained using the Adam optimizer with a learning rate of 0.001-0.005 and 100-200 training iterations. The output dimension is a 64-128 dimensional device state feature vector. The modal feature encoder in step three designs dedicated branches for different data types to maximize the retention of the kernel of each modal data. The core information layer improves the single-modal feature representation capability compared to general feature extraction methods, laying a high-quality foundation for cross-modal fusion. The cross-modal attention layer calculates the correlation weights based on mutual information entropy, and increases the weights of highly correlated features with mutual information entropy greater than 0.6 to achieve dynamic adaptive fusion. This avoids the overloading of key information caused by traditional fixed-weight fusion. For example, after increasing the weights of the correlation feature between environmental humidity and insulation aging, the insulation fault identification rate is improved, enhancing the effectiveness of fused features. The bidirectional long short-term memory network in step three, combined with the Adam optimizer and 100-200 iterations of training, can deeply mine the spatiotemporal evolution law of data, such as the temporal correlation between load fluctuation and equipment overheating, and output 64-128 dimensional equipment state feature vectors. Compared with static feature extraction methods, it improves the temporal feature capture capability and extends the fault prediction lead time.

[0034] In step four, the equipment aging curve is fitted using a Weibull distribution. During the fitting process, the equipment's operating years, cumulative load, and maintenance frequency are introduced as influencing factors. The shape parameter of the Weibull distribution is set to 1.5-3.0, and the scaling parameter is dynamically adjusted according to the equipment type and operating environment. The dynamic threshold early warning model in step four includes a basic threshold module, an aging correction module, and an operating condition adaptation module. The basic threshold module sets the initial early warning threshold based on the equipment's factory technical parameters and power industry standards. The aging correction module calculates the aging decay coefficient based on the equipment aging curve and corrects the initial early warning threshold. The operating condition adaptation module combines real-time environmental parameters... The threshold is adjusted a second time based on the number of data and operating load rate. Step four uses Weibull distribution to fit the equipment aging curve, and introduces influencing factors such as operating years, cumulative load, and maintenance frequency. This can accurately quantify the equipment performance degradation trend. For example, the calculation error of the aging degradation coefficient of a transformer that has been running for ten years is reduced. Compared with the prediction model that does not consider aging, it is more in line with the changes in the equipment's status throughout its entire life cycle, providing a scientific basis for setting dynamic thresholds. The basic threshold module, aging correction module, and operating condition adaptation module of the dynamic threshold early warning model form a three-level threshold adjustment mechanism, which can avoid false alarms or missed alarms of fixed thresholds under equipment aging or severe operating conditions, thereby improving the accuracy of early warning.

[0035] Step 5's expert experience knowledge base contains historical cases of power transmission and transformation equipment failures. Each case includes equipment parameters, environmental conditions, failure type, and handling plan at the time of the failure. Step 5 uses a Bayesian inference algorithm for iterative optimization, matching the prediction results with similar cases in the knowledge base, calculating the posterior probability, and correcting the initial failure probability. The expert experience knowledge base in Step 5 contains complete failure cases, which can provide reference for real operation and maintenance scenarios, such as historical insulation failure cases under similar humidity and load conditions. This provides authoritative data support for correcting the prediction results and makes up for the limitations of pure algorithm models in complex and special working conditions. Step 5 uses a Bayesian inference algorithm to calculate the posterior probability by matching similar cases in the knowledge base and iteratively optimizes the initial failure probability, improving the confidence of the prediction results compared to the uncorrected prediction results and reducing the risk of decision-making errors caused by algorithm bias.

[0036] Working principle: When using this invention, Data support phase: The operation parameters of power transmission and transformation equipment (and environmental parameters) are collected through sensor networks and combined with historical fault records to form a multi-source heterogeneous dataset; outliers are removed by isolated forest, missing values ​​are filled by improved KNN and [0,1] standardization is performed to output high-quality time series samples, providing data input for the two core technologies; Multimodal fusion stage: Initiate an attention-based multimodal data adaptive fusion algorithm: The modal feature encoder extracts single-modal features of each data through dedicated branches, the cross-modal attention layer calculates the association weights based on mutual information entropy and dynamically fuses them, the bidirectional long short-term memory network mines the spatiotemporal patterns of the fused features, and finally outputs the device state feature vector; Dynamic threshold prediction stage: Start the dynamic threshold early warning model that combines the equipment aging curve: fit the equipment aging curve with Weibull distribution and calculate the aging attenuation coefficient; the basic threshold module sets the initial threshold, the aging correction module corrects the threshold with the attenuation coefficient, and the operating condition adaptation module makes secondary adjustments based on real-time parameters; input the equipment state feature vector into the model and output the failure probability and prediction time window. Results optimization phase: Introduce an expert experience knowledge base, match similar fault cases using Bayesian inference algorithm, correct the initial prediction results, and generate a high-confidence final fault prediction report.

[0037] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

[0038] 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 predicting equipment failure in a power transmission and distribution system, characterized by: The fault prediction method comprises the following steps: Step one, data collection, collecting the operating parameters, environmental parameters and historical fault records of the power transmission and transformation equipment through a sensor network to form a multi-source heterogeneous data set; Step two, data preprocessing, performing outlier rejection, missing value filling and standardization processing on the multi-source heterogeneous data set to construct a time series sample; Step three, feature extraction, using a multi-modal fusion network based on an attention mechanism to mine the correlation features and spatio-temporal evolution law of different dimensional data and output a device state feature vector; Step four, fault prediction, constructing a dynamic threshold warning model in combination with the device aging curve, inputting the device state feature vector into the model, and outputting a fault probability and a prediction time window; Step five, result correction, introducing an expert experience knowledge base to iteratively optimize the fault probability and the prediction time window to generate a final fault prediction report.

2. The method of claim 1, wherein: The operating parameters collected in step one include electrical quantity parameters, mechanical vibration parameters and temperature parameters of the power transmission and transformation equipment, the electrical quantity parameters include voltage, current and power factor, the mechanical vibration parameters include vibration frequency and amplitude, and the temperature parameters include device body temperature and winding temperature, and the environmental parameters include environmental temperature, humidity and wind speed.

3. The method of claim 1, wherein: In step two, the isolated forest algorithm is used for outlier rejection, the number of trees of the isolated forest algorithm is set to 100-200, the size of the sample subset is set to 20-50, the improved KNN algorithm is used for missing value filling, the number of neighbors of the improved KNN algorithm is set to 5-10, and the standardization processing maps the data to the [0, 1] interval.

4. The method of claim 1, wherein: In step three, the multi-modal fusion network based on the attention mechanism includes a modal feature encoder, a cross-modal attention layer and a time series feature decoder, the modal feature encoder designs branch networks for different types of data, 1D convolutional neural network is used to extract features from electrical quantity parameters, Fourier transform combined with convolutional neural network is used to extract features from mechanical vibration parameters, and fully connected network is used to extract features from temperature parameters and environmental parameters.

5. The method of claim 4, wherein: The cross-modal attention layer calculates the correlation weight of different modal features through a self-attention mechanism, the calculation of the correlation weight is based on inter-modal mutual information entropy, the inter-modal mutual information entropy calculation formula is: I(X,Y) = I(X) + I(Y) - I(X,Y) where X and Y are different modal feature data sets, p(x, y) is the joint probability density, p(x) and p(y) are the marginal probability densities, the weight value ranges from 0 to 1, for modal features with mutual information entropy greater than 0.6, the correlation weight is increased to realize dynamic adaptive fusion of different modal features.

6. The method of claim 5, wherein: The time series feature decoder uses a bidirectional long short-term memory network, the number of hidden layers of the memory network is set to 2-4 layers, the number of hidden layer neurons is set to 128-256, the Adam optimizer is used for training, the learning rate is set to 0.001-0.005, the training iteration number is set to 100-200, and the output dimension is a 64-128 dimensional device state feature vector.

7. The method of claim 1, wherein: The equipment aging curve in the fourth step is fitted by using a Weibull distribution, and during the fitting process, the equipment operation time, cumulative load and maintenance frequency are introduced as influencing factors, the shape parameter of the Weibull distribution is set to 1.5-3.0, and the scale parameter is dynamically adjusted according to the type of the equipment and the operation environment.

8. The method of claim 7, wherein: The dynamic threshold early warning model in the fourth step comprises a basic threshold module, an aging correction module and a working condition adaptation module, the basic threshold module sets an initial early warning threshold based on the factory technical parameters of the equipment and the power industry standards, the aging correction module calculates an aging attenuation coefficient according to the equipment aging curve, and corrects the initial early warning threshold, and the working condition adaptation module adjusts the corrected threshold again in combination with real-time environmental parameters and the operation load rate.

9. The method of claim 1, wherein: The expert experience knowledge base in the fifth step contains power transmission and transformation equipment fault history cases, and each case includes the equipment parameters, environmental conditions, fault types and treatment schemes when the fault occurs.

10. The method of claim 9, wherein: The fifth step adopts a Bayesian inference algorithm for iterative optimization, matches the prediction result with similar cases in the knowledge base, calculates the posterior probability, and corrects the initial fault probability.

Images