A method for tracing the source of power secondary equipment operation faults based on supervised learning

Abstract

The present application discloses a method for tracing the source of faults in the operation of power secondary equipment based on supervised learning. The method includes: establishing a data storage unit, storing fault types and constructing a fault propagation map, and using a relational database and data processing and integration tools to update the data in the data storage unit. After processing the stored data, the data set is divided to construct and verify a fault tracing model. By constructing a fault tracing credibility index and a fault propagation perception confidence interval, the accuracy of the fault tracing model and whether the data is unbalanced are judged. If the data is unbalanced, resampling and data enhancement are used for processing. The processed fault tracing model can effectively realize the tracing of the source of faults in the operation of power secondary equipment, improve the ability to handle faults, and provide guarantees for the stable operation of the power system.

Description

Technical Field

[0001] The present application relates to the technical field of power equipment data processing, and in particular to a method for tracing the source of power secondary equipment operation faults based on supervised learning. Background Art

[0002] Secondary power equipment, including relay protection devices, automatic devices, measuring instruments, and communications equipment, assists in monitoring, controlling, protecting, and regulating primary equipment in power systems. While not directly involved in the generation, transmission, and distribution of electrical energy, these devices are crucial for ensuring the safe and stable operation of power systems. They monitor the operating status of primary equipment in real time, acting promptly when an anomaly occurs to prevent further escalation and ensure the reliability and quality of power supply.

[0003] Fault tracing plays a critical role in the operation of secondary power equipment. Once a fault occurs, quickly and accurately identifying its source can significantly shorten outages, reduce economic losses, and ensure the normal operation of the power system. Fault tracing allows for in-depth analysis of the cause of the fault, summarizes lessons learned, and provides a basis for equipment maintenance, upgrades, and optimization, thereby improving the overall reliability and safety of the power system.

[0004] However, existing power secondary equipment fault tracing technology has several shortcomings. In terms of data processing, operations such as data augmentation are often performed indiscriminately on data with limited capacity. This not only consumes significant computing resources and time, resulting in low efficiency, but can also disrupt otherwise balanced data, affecting model accuracy. Furthermore, existing technologies lack in-depth analysis and utilization of fault propagation relationships, making it difficult to fully and accurately grasp the fault's propagation path and impact range. Consequently, the accuracy and speed of fault tracing remain insufficient to meet practical requirements. Summary of the Invention

[0005] In view of this, the present application provides a method for tracing the source of power secondary equipment operation faults based on supervised learning.

[0006] According to one aspect of the present application, a method for tracing the source of an operating fault of a power secondary device based on supervised learning is provided, the method comprising the following steps:

[0007] Step 1: Establish a data storage unit, which is used to store all known fault types in the power system, and establish a fault propagation map based on the association information between each fault type and each power secondary device, wherein the fault propagation map uses the power secondary device as a node;

[0008] Step 2: Process the data stored in the data storage unit, divide the processed data into a construction data set and a verification data set, establish a fault tracing model based on the construction data set, and verify the output results of the established fault tracing model based on the verification data set;

[0009] Step 3 includes: Step 3 (1): Constructing the fault tracing credibility index, the formula is as follows:

[0010]

[0011] Among them, N i represents the number of samples in the constructed data set of the i-th type of fault, β represents the preset sample size attenuation coefficient, ρ represents the preset fault traceability credibility index parameter value, ω represents the preset recall rate weight, R i The calculation formula is as follows:

[0012]

[0013] Among them, TP i FN represents the number of samples that actually belong to the i-th type of fault and are correctly predicted by the fault tracing model as the i-th type of fault. i represents the number of samples that are actually type i faults but are mistakenly predicted to be other fault types, P i The calculation formula is as follows:

[0014]

[0015] Among them, TP i It represents the number of samples that actually belong to the i-th type of fault and are correctly predicted by the fault tracing model as the i-th type of fault. i It represents the number of samples that do not actually belong to fault type i but are mistakenly predicted as type i fault by the fault tracing model;

[0016] Step 3 (2): Construct the fault propagation perception confidence interval, the formula is as follows:

[0017]

[0018] Among them, R i represents the recall rate, a represents the critical value of the time t distribution, N (i) represents the set of nodes directly connected to the i-th type of fault in the fault propagation graph, R ij represents the propagation probability from the i-th fault to the j-th fault calculated based on the historical data in the power system, λ represents the preset propagation influence coefficient, N i represents the number of samples in the constructed dataset of type i fault;

[0019] Step three (3): Determine whether the data in the constructed dataset of the type i fault needs data enhancement based on the fault tracing credibility index and the fault propagation perception confidence interval. When the following set is met, it indicates that the data in the constructed dataset of the type i fault needs data enhancement:

[0020]

[0021] Step 4: Based on the output of step 3, process the data of a type of fault that requires data enhancement. The processing methods include but are not limited to resampling, data enhancement, and adjusting the loss function.

[0022] By means of the above technical solution, the present application provides a method for tracing the source of power secondary equipment operation faults based on supervised learning, which has the following beneficial effects:

[0023] Accurate determination and targeted processing: By constructing fault tracing credibility indicators and fault propagation perception confidence intervals, we can accurately determine whether the data for a particular fault type is imbalanced (if imbalanced, data augmentation is required). Rather than blindly augmenting all low-capacity data, we use the calculation results of these indicators and intervals to specifically identify the data that truly needs processing, avoiding unnecessary data processing operations and improving data processing efficiency.

[0024] Optimizing model performance: Enhancement is performed only on data identified as imbalanced. For example, the SMOTE algorithm (Synthetic Minority Oversampling Technique) is used to synthesize new minority class samples, effectively balancing the number of samples in each class within the dataset. This not only addresses model bias caused by data imbalance and improves the model's ability to identify minority class samples, but also prevents the risk of overfitting caused by excessive enhancement, resulting in improved fault tracing model performance and more accurate predictions.

[0025] Improve resource utilization efficiency: Accurately screening the data that needs to be enhanced avoids indiscriminate processing of all small-capacity data, reducing the waste of computing resources and time. While ensuring the accuracy of fault tracing, it rationally allocates resources, making the entire fault tracing system run more efficiently.

[0026] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0028] Figure 1 A flow chart of a method for tracing the source of an operating fault of a power secondary device based on supervised learning provided in an embodiment of the present application is shown;

[0029] Figure 2 A flow chart of another method for tracing the source of power secondary equipment operation faults based on supervised learning provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0030] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0031] In this embodiment, a method for tracing the source of power secondary equipment operation faults based on supervised learning is provided. Figure 1 As shown, the method includes:

[0032] Step 101: Establish a data storage unit, which is used to store all known fault types in the power system, and establish a fault propagation map based on the association information between each fault type and each power secondary device, wherein the fault propagation map uses the power secondary devices as nodes.

[0033] In the above embodiment of the present application, a data storage unit is established, and the data storage unit is used to store all known fault types in the power system. A fault propagation map is established based on the association information between each fault type and each power secondary device. In particular, there are many known fault types in the power system, and these fault types are closely associated with the power secondary devices. For example, the following information is provided about the fault types in the power system and their association with the power secondary devices:

[0034] 1. Fault Type

[0035] 1. Short circuit fault: refers to the abnormal short circuit between conductive parts at different points (including conductive parts to the ground).

[0036] Classification: including three-phase short circuit, two-phase short circuit, single-phase ground short circuit and two-phase short circuit grounding, etc.

[0037] Relationship with Secondary Equipment: Short-circuit faults can cause abnormal increases in current, potentially triggering relay protection devices (such as current protection and distance protection) to disconnect the faulty circuit. These protection devices are secondary power equipment used to monitor and control the operating status of the power system.

[0038] 2. Ground fault: refers to the insulation failure of equipment or lines that causes current to flow to the ground.

[0039] Classification: including metallic grounding fault and transition resistance grounding fault, etc.

[0040] Relationship with secondary equipment: Ground faults can also trigger the operation of relay protection devices, such as zero-sequence current protection. Furthermore, ground faults can affect the secondary voltage of voltage transformers, thereby affecting the normal operation of measurement and control circuits.

[0041] 3. Line break fault (also known as longitudinal fault): refers to the disconnection of one or two phases among the three phases.

[0042] Classification: including single-phase break and two-phase break, etc.

[0043] Relationship with secondary devices: A line break can cause current imbalance, potentially triggering line-break protection devices. It can also affect the accuracy of measurement circuits, which typically rely on secondary-side signals from current and voltage transformers.

[0044] 4. Overvoltage fault: refers to a sudden increase in system voltage that exceeds the tolerance range of the equipment.

[0045] Classification: including operational overvoltage, lightning overvoltage, etc.

[0046] Relationship with Secondary Equipment: Overvoltage faults can damage power equipment, including secondary equipment. For example, overvoltage can cause relays to malfunction or damage. Overvoltage can also affect the secondary voltage of voltage transformers, further impacting the proper functioning of measurement and control circuits.

[0047] 5. Other faults include transformer faults (such as phase-to-phase faults, turn-to-turn faults, etc.), generator faults (such as stator-to-phase faults, rotor faults, etc.) and busbar faults.

[0048] Association with secondary equipment: These faults may also trigger relay protection devices and affect the normal operation of measurement and control circuits. For example, a transformer fault may cause the differential protection device to operate; a generator fault may trigger loss of excitation protection or stator ground fault protection.

[0049] 2. Correlation Information between Power Secondary Equipment and Various Fault Types

[0050] 1. Relay protection device

[0051] Function: Monitor the operating status of the power system, quickly cut off the fault circuit when a fault occurs, and protect the safe and stable operation of power equipment and power systems.

[0052] Associated fault types: Associated with almost all types of faults, such as short circuit faults, ground faults, line break faults, and overvoltage faults.

[0053] 2. Measurement circuit

[0054] Function: Used to measure the voltage, current and other parameters of the power system and provide necessary operating information to operators.

[0055] Related fault types: These are related to short circuit faults, ground faults, line break faults, and overvoltage faults. These faults may cause abnormal signals on the secondary side of the voltage transformer or current transformer in the measurement loop, thus affecting the accuracy of the measurement results.

[0056] 3. Control loop

[0057] Function: Used to control the closing and opening operations of circuit breakers, disconnectors and other equipment in the power system.

[0058] Related fault types: Related to short circuit faults, ground faults, and disconnection faults. These faults may cause relays in the control loop to malfunction or be damaged, thus affecting the normal operation of the equipment.

[0059] 4. Signal loop

[0060] Function: Used to provide position and status signals of important equipment such as circuit breakers and disconnectors.

[0061] Associated fault type: Associated with various fault types. When a fault occurs, the signal circuit will send a corresponding signal to notify the operator.

[0062] Therefore, there is a close correlation between various fault types in the power system and the power secondary equipment. Understanding this correlation information helps operators better monitor and control the operating status of the power system, ensuring the safe and stable operation of the power system.

[0063] Optionally, the association information includes connection relationships, dependency relationships, and fault propagation history data. In step 101, a fault propagation graph is further established in the following manner:

[0064] Step 1011: collecting connection relationships, dependency relationships, and fault propagation history data of power secondary equipment through the power system.

[0065] Step 1012: Create nodes in a preset graph database to represent power secondary equipment and edges to represent fault propagation paths. Use the declarative query language of the preset graph database to evaluate the propagation probability on each fault propagation path, and then use the structure established by visualization of the preset graph database to obtain a fault propagation map.

[0066] In the above embodiment of the present application, a graph database such as Neo4j and a declarative query language such as Cypher query language are preset. Specifically, the connection relationship, dependency relationship and fault propagation history data of the power secondary equipment are collected through the power system. Then, nodes are created in Neo4j to represent the power secondary equipment and edges to represent the fault propagation paths. Using Neo4j's Cypher query language, the propagation probability on each fault propagation path can be easily evaluated, and the graph structure can be visualized using Neo4j;

[0067] When constructing the visualization map structure here, the data sources are SCD (substation configuration description) files, IED (intelligent electronic device) configuration parameters, SCADA system topology diagrams, etc. IEC61850 parsing tools are used to extract the connection relationships between devices, and the GIS system provides the physical location and electrical connection information of the secondary power equipment.

[0068] In particular, Neo4j is a high-performance NoSQL graph database, and Cypher is its declarative query language. When using the Neo4j graph database and its Cypher query language to process and evaluate the propagation path of power secondary equipment faults in power systems, the following key steps and concepts can be included:

[0069] 1. Data Collection

[0070] Collect data on the connection relationships, dependencies, and fault propagation history of power secondary equipment from the power system. This data forms the basis for subsequent analysis and modeling.

[0071] 2. Create a graph model in Neo4j:

[0072] In the Neo4j graph database, create nodes to represent power secondary equipment.

[0073] Create edges (Relationships) to represent fault propagation paths between devices. These edges can contain additional information such as fault propagation probability and propagation direction.

[0074] 3. Using Cypher query language:

[0075] Cypher is Neo4j's declarative query language that allows users to write complex queries to find and manipulate graph data in an intuitive way.

[0076] The Cypher query language allows users to easily evaluate the propagation probability of each fault propagation path. For example, a query can be written to find all fault propagation paths from a specific device and calculate the sum of the propagation probabilities along these paths.

[0077] 4. Visualization of graph structure:

[0078] Neo4j provides powerful visualization tools that allow users to intuitively view and analyze graph data in a graphical manner.

[0079] Through Neo4j's visual graph structure, users can more easily understand the connection relationships, dependencies, and fault propagation paths between devices, thereby making more informed decisions.

[0080] In summary, using the Neo4j graph database and Cypher query language to process and analyze the fault propagation paths of power secondary equipment in power systems allows power system operators to more effectively identify and manage potential fault risks, improving system reliability and safety.

[0081] Optionally, step 101 further includes the following steps:

[0082] Step 1013: The data storage unit uses a relational database to store the fault type and the association information between each fault type and each power secondary device. The latest data is regularly extracted from the power system monitoring system through data processing and integration tools to update the relational database, the connection relationship and dependency relationship of the power secondary devices.

[0083] In the above embodiments of the present application, data processing and integration tools, such as ETL tools (such as Talend or Apache NiFi), the data storage unit uses a relational database (MySQL) to store the association information between fault types and equipment, and at the same time combines the ETL tool (Talend or Apache NiFi) to regularly (for example, weekly, monthly or semi-annual, etc.) extract the latest data from the monitoring system of the power grid, update the connection relationship and dependency relationship of the power secondary equipment in the database, thereby realizing the update and iteration of the fault propagation map.

[0084] In particular, ETL tools (such as Talend or Apache NiFi) are mainly used to implement data extraction (Extract), transformation (Transform) and loading (Load) operations, and are an important part of the data processing field.

[0085] Talend is an open source data integration tool that provides a wealth of data conversion and processing functions, such as data cleaning, data merging, and data filtering. Its features include:

[0086] Supports multiple data sources, including relational databases, big data platforms, cloud storage, and web services.

[0087] Provides a graphical interface that enables users to easily design, deploy, and run data integration jobs.

[0088] It provides powerful job scheduling and monitoring capabilities, which can automate the data integration process and monitor the execution of jobs in real time.

[0089] Apache NiFi is a data flow system based on process programming concepts that supports directed graphs of powerful and scalable data routing, transformation, and system mediation logic. Features of NiFi include:

[0090] Features a web-based user interface for design, control, feedback, and monitoring of data flows.

[0091] It is highly configurable in multiple quality of service dimensions, such as loss tolerance and guaranteed delivery, low latency and high throughput, priority-based queuing, etc.

[0092] It provides fine-grained data traceability for all received data and can record the data processing process in detail.

[0093] To this end, both Talend and Apache NiFi are powerful ETL tools that help users efficiently process and integrate data. Choosing the right tool depends on specific requirements, such as the type of data source, processing performance requirements, scalability, ease of use, and cost. In practice, users can choose the appropriate ETL tool based on their needs to achieve efficient data processing and integration.

[0094] Step 102: Process the data stored in the data storage unit, divide the processed data into a construction data set and a verification data set, establish a fault tracing model through the construction data set, and verify the output results of the established fault tracing model based on the verification data set.

[0095] In the above embodiment of the present application, data processing is performed on the data stored in the data storage unit. Specifically, feature vectors can be extracted, and then the feature vectors are divided into a construction data set and a verification data set. A fault tracing model is established through the construction data set, and then the output results of the established fault tracing model are verified according to the verification data set. At the same time, Python's Pandas library can be used for data cleaning, and the cleaning methods include but are not limited to processing missing values, outliers, data type conversion, etc. Then, the Scikit-learn library is used to perform feature engineering operations, and feature engineering includes but is not limited to feature selection, feature scaling, and polynomial feature generation.

[0096] It should be noted that in addition to basic missing value and outlier handling and data type conversion, attention should also be paid to the detection and processing of duplicate values ​​to ensure data consistency, including the unification of date formats and numerical units. For text data, preprocessing such as removal of spaces and punctuation, and case conversion should be performed. In terms of feature engineering, in addition to conventional feature selection, scaling, and polynomial feature generation, the generation of interactive features should also be considered to capture the potential relationships between features. For time series data, extracting time-related features can significantly improve model performance. Target encoding can be used as an effective method for processing categorical features, while feature importance assessment helps to accurately select the features that contribute most to the model. When the number of features is large, the use of dimensionality reduction techniques such as PCA can effectively reduce the computational burden and reduce the risk of overfitting.

[0097] Optionally, step 102 further includes the following steps:

[0098] In step 1021, a deep learning framework or the Scikit-learn machine learning library is used to construct a fault tracing model, wherein the fault tracing model includes but is not limited to a neural network, a decision tree, a random forest, and a support vector machine. The constructed fault tracing model is trained using the constructed dataset.

[0099] In the above embodiments of the present application, when constructing a fault tracing model, a deep learning framework / Scikit-learn machine learning library is used to construct the fault tracing model. The fault tracing model includes but is not limited to neural networks, decision trees, random forests, and support vector machines. The fault tracing model is trained using a constructed data set, and then the performance of the fault tracing model is evaluated using a verification data set.

[0100] In particular, deep learning frameworks and the Scikit-learn machine learning library are both tools for building and training machine learning models, but they differ in their functionality and applicable scenarios.

[0101] Deep learning frameworks are software libraries specifically designed for building and training deep learning models. Deep learning models typically consist of multi-layer neural networks capable of handling complex nonlinear problems. Deep learning frameworks provide a wide range of neural network components (such as convolutional layers, recurrent layers, and fully connected layers), optimization algorithms (such as SGD and Adam), and automatic differentiation mechanisms, enabling developers to efficiently build and train deep learning models.

[0102] Common deep learning frameworks include TensorFlow, PyTorch, Keras, etc. These frameworks are generally highly flexible and scalable and suitable for various deep learning applications such as image recognition, speech recognition, natural language processing, etc.

[0103] Scikit-learn is an open-source Python machine learning library that provides a wide range of machine learning algorithms and data processing tools. Unlike deep learning frameworks, Scikit-learn focuses primarily on traditional machine learning algorithms such as linear regression, logistic regression, support vector machines, decision trees, and random forests. These algorithms are generally suitable for processing smaller datasets and relatively simple tasks.

[0104] Scikit-learn is designed around simplicity, ease of use, and efficiency. It provides an easy-to-understand API and extensive documentation, enabling developers to quickly build and evaluate machine learning models. Furthermore, Scikit-learn offers a range of tools for data preprocessing, feature extraction, model evaluation, and more, making the machine learning process even more convenient.

[0105] For this reason, the deep learning framework and the Scikit-learn machine learning library differ in their functionality and applicable scenarios. The deep learning framework is suitable for building and training complex deep learning models, while Scikit-learn focuses primarily on traditional machine learning algorithms and data processing tools. In practical applications, the appropriate tools can be selected to build machine learning models based on the characteristics of the specific task and dataset in the above embodiments of this application.

[0106] Optionally, step 103 further includes the following steps:

[0107] Step 1031: Construct a fault tracing credibility index, the formula is as follows:

[0108]

[0109] Among them, N i represents the number of samples in the constructed data set of the i-th type of fault, β represents the preset sample size attenuation coefficient, ρ represents the preset fault traceability credibility index parameter value, ω represents the preset recall rate weight, R i The calculation formula is as follows:

[0110]

[0111] Among them, TP i FN represents the number of samples that actually belong to the i-th type of fault and are correctly predicted by the fault tracing model as the i-th type of fault. i represents the number of samples that are actually type i faults but are mistakenly predicted to be other fault types, P i The calculation formula is as follows:

[0112]

[0113] Among them, TP iIt represents the number of samples that actually belong to the i-th type of fault and are correctly predicted by the fault tracing model as the i-th type of fault. i It represents the number of samples that do not actually belong to fault type i but are mistakenly predicted as type i fault by the fault tracing model.

[0114] Step 1032: Construct a fault propagation perception confidence interval using the following formula:

[0115]

[0116] Among them, R i represents the recall rate, a represents the critical value of the time t distribution, N (i) represents the set of nodes directly connected to the i-th type of fault in the fault propagation graph, R ij represents the propagation probability from the i-th fault to the j-th fault calculated based on the historical data in the power system, λ represents the preset propagation influence coefficient, N i Represents the number of samples in the constructed dataset for type i fault.

[0117] Step 1033: Determine whether data in the constructed dataset for the i-th type of fault requires data enhancement based on the fault tracing credibility index and the fault propagation perception confidence interval. If the following conditions are met, data in the constructed dataset for the i-th type of fault requires data enhancement:

[0118]

[0119] In the above embodiment of the present application, the fault tracing credibility index parameter value is preset to adjust the parameters of the credibility index calculation formula, which can be set according to the specific application scenario or demand (credibility evaluation demand) to affect the final credibility evaluation result. The preset sample size attenuation coefficient β is, for example, 0.03, the preset recall rate weight ω is, for example, 0.7, the preset propagation influence coefficient λ is, for example, 0.2, and a is the critical value of the time t distribution. In the formula for constructing the fault propagation perception confidence interval, a is an important parameter used to determine the time range in the fault propagation perception confidence interval. This critical value is set based on the characteristics of the time t distribution, which affects the width and accuracy of the confidence interval. By reasonably setting the value of a, the fault propagation perception confidence interval can be made more consistent with the actual situation, thereby improving the reliability of fault tracing and prediction. To this end, by establishing a definition of the fault tracing credibility index and constructing the fault propagation perception confidence interval, the accuracy of the established fault tracing model is judged, thereby judging whether the data of a certain fault type is unbalanced, that is, whether data enhancement is required. To further explain, suppose there are 100 actual samples of a certain fault type, and the fault tracing model correctly predicts 80 of them (TP i=80), but 20 of them were incorrectly predicted as other categories (FN i =20). Then the recall rate (R i ) is 0.8. Meanwhile, 0.65 is the safety recall rate threshold. In power system reliability standards, it is usually required that the fault detection coverage of key equipment be no less than 65%, as per IEEE C37.119.

[0120] Step 104: Based on the output result of step 3, the data of the type of fault that requires data enhancement is processed, and the processing methods include but are not limited to resampling, data enhancement, and adjustment of the loss function.

[0121] In the above embodiment of the present application, based on the output of step 103, the data of a certain fault type with data imbalance is processed, and the processing methods include but are not limited to resampling, data augmentation, and adjusting the loss function. When it is necessary to process data of a certain fault type with data imbalance, resampling and adjusting the loss function are two commonly used strategies. The following are specific operation methods for these two strategies:

[0122] Regarding resampling, resampling is a method of balancing a dataset by changing the number of samples of each type in the dataset. For fault types with data imbalance, the following two resampling methods can be used:

[0123] 1. Oversampling:

[0124] Method: Increase the number of minority class (i.e., unbalanced fault type) samples by copying existing samples, generating new samples using interpolation methods, or applying some generative models (such as SMOTE) to synthesize new samples.

[0125] Advantages: It can directly increase the number of minority class samples, thereby balancing the data set.

[0126] Disadvantages: It may lead to overfitting, especially when copying existing samples; the generated new samples may not be realistic enough, affecting model performance.

[0127] 2. Undersampling:

[0128] Method: Reduce the number of majority class (i.e. balanced fault type) samples. This can be done by randomly deleting majority class samples or using some clustering algorithm to select representative samples.

[0129] Advantages: It can reduce the total size of the dataset and speed up training.

[0130] Disadvantages: Useful information in the majority class samples may be lost, affecting the model's ability to identify the majority class.

[0131] In practical applications, the appropriate resampling method can be selected according to the specific situation of the dataset and the requirements of the model. In addition, it is also possible to consider combining oversampling and undersampling (such as SMOTE+Tomek Links) to further optimize the dataset.

[0132] Adjusting the loss function is a method of balancing the dataset by changing the weight of the loss function during model training. For data imbalance fault types, the following strategies can be adopted:

[0133] 1. Category weight adjustment:

[0134] Method: Assign different weights to samples of different categories. In the loss function, give more weight to minority class samples and less weight to majority class samples.

[0135] Advantages: It enables the model to pay more attention to minority class samples during training and improves the model's ability to recognize minority classes.

[0136] Disadvantages: The weight value needs to be adjusted manually, and the choice of weight value has a great impact on model performance.

[0137] 2. Custom loss function:

[0138] Method: Design a custom loss function based on the characteristics of the dataset and the requirements of the model. For example, you can combine methods such as class weight adjustment, cross entropy loss, and focal loss to optimize the loss function.

[0139] Advantages: It can adapt more flexibly to the characteristics of the dataset and the requirements of the model.

[0140] Disadvantages: Designing a custom loss function requires high expertise and experimental experience.

[0141] To this end, in practical applications, we can choose an appropriate loss function adjustment method based on the specific situation of the data set and the requirements of the model. In addition, we can also combine other techniques (such as feature selection, model integration, etc.) to further optimize model performance.

[0142] Optionally, for data that needs data enhancement, step 104 further includes the following steps:

[0143] Step 1041 : For data that needs data enhancement, use the synthetic minority class oversampling technology to perform data enhancement.

[0144] In the above embodiment of the present application, for the fault type with data imbalance, the synthetic minority oversampling technology (SMOTE) is used for data enhancement. The SMOTE algorithm synthesizes new minority class samples by analyzing the similarities between minority class samples to increase the number of minority classes, thereby balancing the data set. This method is based on the K nearest neighbor algorithm. For each minority class sample, its K nearest neighbor samples are found, and new samples are synthesized by linear interpolation between these neighbor samples. This effectively solves the data imbalance problem and improves the model's recognition ability for minority class samples. When applying the SMOTE algorithm for data enhancement, special attention should be paid to selecting a suitable K value. A K value that is too large may cause the synthesized samples to be too generalized and differ greatly from the original samples, while a K value that is too small may cause the synthesized samples to be too concentrated in certain specific areas, resulting in overfitting.

[0145] In particular, the synthetic minority oversampling technique (SMOTE) is an improved scheme based on the random oversampling algorithm, which aims to solve the class imbalance problem in machine learning.

[0146] Class imbalance is a common challenge in machine learning and data mining. When the number of samples in one class significantly exceeds that in another, machine learning models tend to favor the majority class, resulting in poor prediction performance for the minority class. SMOTE (Sensitivity Modeling, or Smooth Motion Detection) is a technique developed to address this issue.

[0147] The basic idea of ​​SMOTE technology is to analyze minority samples and artificially synthesize new minority samples based on these samples, thereby increasing the number of minority samples and balancing the dataset. The specific process is as follows:

[0148] 1. Identify minority samples: First, identify the samples that belong to the minority class in the dataset.

[0149] 2. Calculate K nearest neighbors: For each minority sample, calculate its K nearest neighbors in the feature space using the Euclidean distance metric.

[0150] 3. Synthesize a new sample: Randomly select a neighbor and calculate the difference between the minority sample and this neighbor. Then, multiply this difference by a random number between 0 and 1 and add it to the feature vector of the minority sample to generate a new synthetic sample. This new sample lies on the line segment between the minority sample and its selected neighbor.

[0151] 4. Repeat generation: Repeat the above process until the desired level of class balance is achieved.

[0152] The advantages of SMOTE technology are as follows:

[0153] 1. Reduce the risk of overfitting: Compared with random oversampling, SMOTE reduces the risk of model overfitting by generating new synthetic samples instead of simply copying existing samples.

[0154] 2. Improve model performance: By balancing the dataset, SMOTE helps AI models better identify patterns and generalize across categories, thereby improving overall performance.

[0155] 3. Resource-efficient: SMOTE effectively amplifies the number of samples in the minority class without collecting additional data, making it a resource-efficient technique.

[0156] 4. Algorithm independence: SMOTE can be used with various AI algorithms, including decision trees, support vector machines, neural networks, etc.

[0157] The limitations of SMOTE technology are as follows:

[0158] Although SMOTE technology has many advantages, it also has some limitations in practical applications:

[0159] 1. Overfitting risk: If used improperly, such as generating too many synthetic samples, SMOTE may lead to overfitting.

[0160] 2. Potential noise: The synthetic samples generated by SMOTE may not accurately represent real-world data instances, thereby introducing noise that may negatively affect model performance.

[0161] 3. K value selection: The performance of SMOTE is affected by the choice of parameter K (i.e., the number of nearest neighbors). An inappropriate K value may lead to poor results.

[0162] 4. Distribution Marginalization: Because the distribution of minority samples determines their selectable neighbors, if a minority sample is at the edge of the distribution, the resulting synthetic sample may also be at the edge, blurring the class boundaries and increasing classification difficulty. To address the limitations of SMOTE, various improvements have been proposed, such as Borderline-SMOTE. These methods aim to address issues such as sample overlap and distribution marginalization, thereby improving the performance and applicability of SMOTE.

[0163] To this end, SMOTE technology is an effective method for addressing class imbalance. It generates new synthetic samples to balance the dataset, thereby improving the performance and fairness of AI models. However, in practical applications, it is necessary to be aware of its limitations and select appropriate parameters and improvement methods to optimize performance.

[0164] By applying the technical solution of this embodiment, it is possible to effectively trace the source of power secondary equipment operation faults, improve the fault handling capability, and provide a guarantee for the stable operation of the power system.

[0165] Furthermore, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, another method for tracing the source of power secondary equipment operation faults based on supervised learning is provided, such as Figure 2 As shown, the method includes the following steps:

[0166] Step 105 includes: Step 1051, calculating the category entropy value CE, the formula is as follows:

[0167]

[0168] Among them, N i represents the number of samples in the constructed dataset of type i fault, N z Indicates the total number of samples.

[0169] Step 1052: Calculate the confidence entropy value MCE. The formula is as follows:

[0170]

[0171] Among them, P i (x) represents the predicted probability that the sample x output by the fault tracing model in step 2 belongs to the i-th type of fault sample, D i represents the set of type i fault samples,

[0172] Step 1053: The formula model D is established by combining the category entropy calculation formula and the confidence entropy calculation formula, as follows:

[0173]

[0174] in, Indicates the preset parameter value of the sparsity of the balanced model. max(CE) and max(MCE) represent the maximum category entropy and maximum confidence entropy of all categories of fault types, respectively. CE i and MCE i They represent the category entropy value and confidence entropy value of the i-th type of fault respectively;

[0175] Step 106: When D>0.8, step 104 is triggered to perform data enhancement on the data of this type of fault.

[0176] In the above embodiments of the present application, The value can be set to 0.6. At the same time, there are two ways to judge whether the imbalanced data needs data enhancement, further improve the accuracy of the judgment and avoid misjudgment. When judging whether the imbalanced data needs data enhancement, the following rules need to be followed:

[0177]

[0178] It should be noted that after weighted calculation of the category entropy and confidence entropy, normalization is performed. The normalized entropy value range is (0, 1). Here, 0.8 is in the high range, indicating that data sparsity or model uncertainty has reached a significant level, and the table name needs to be enhanced.

[0179] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0180] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0181] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for tracing the source of power secondary equipment operation faults based on supervised learning, characterized in that: The method comprises the following steps: Step 1: Establish a data storage unit, which is used to store all known fault types in the power system, and establish a fault propagation map based on the association information between each fault type and each power secondary device, wherein the fault propagation map uses the power secondary device as a node; Step 2: Process the data stored in the data storage unit, divide the processed data into a construction data set and a verification data set, establish a fault tracing model based on the construction data set, and verify the output results of the established fault tracing model based on the verification data set; Step 3 includes: Step 3 (1): Constructing the fault tracing credibility index, the formula is as follows: Among them, N i represents the number of samples in the constructed data set of the i-th type of fault, β represents the preset sample size attenuation coefficient, ρ represents the preset fault traceability credibility index parameter value, ω represents the preset recall rate weight, R i The calculation formula is as follows: Among them, TP i FN represents the number of samples that actually belong to the i-th type of fault and are correctly predicted by the fault tracing model as the i-th type of fault. i represents the number of samples that are actually type i faults but are mistakenly predicted to be other fault types, P i The calculation formula is as follows: Among them, TP i It represents the number of samples that actually belong to the i-th type of fault and are correctly predicted by the fault tracing model as the i-th type of fault. i It represents the number of samples that do not actually belong to fault type i but are mistakenly predicted as type i fault by the fault tracing model; Step 3 (2): Construct the fault propagation perception confidence interval, the formula is as follows: Among them, R i represents the recall rate, a represents the critical value of the time t distribution, N (i) represents the set of nodes directly connected to the i-th type of fault in the fault propagation graph, R ij represents the propagation probability from the i-th fault to the j-th fault calculated based on the historical data in the power system, λ represents the preset propagation influence coefficient, N i represents the number of samples in the constructed dataset of type i fault; Step three (3): Determine whether the data in the constructed dataset of the type i fault needs data enhancement based on the fault tracing credibility index and the fault propagation perception confidence interval. When the following set is met, it indicates that the data in the constructed dataset of the type i fault needs data enhancement: Step 4: Based on the output of step 3, process the data of a type of fault that requires data enhancement. The processing methods include but are not limited to resampling, data enhancement, and adjusting the loss function.

2. The method according to claim 1, characterized in that The associated information includes connection relationships, dependency relationships, and fault propagation history data. Step 1 also establishes a fault propagation graph in the following manner: Collect connection relationships, dependencies, and fault propagation history data of power secondary equipment through the power system; In the preset graph database, nodes are created to represent power secondary equipment, and edges are created to represent fault propagation paths. The propagation probability on each fault propagation path is evaluated using the declarative query language of the preset graph database. The established structure is visualized using the preset graph database to obtain a fault propagation map.

3. The method according to claim 2, characterized in that The data storage unit uses a relational database to store fault types and the association information between each fault type and each power secondary device. The latest data is regularly extracted from the power system monitoring system through data processing and integration tools to update the relational database, the connection relationship and dependency relationship of the power secondary devices.

4. The method according to claim 1, wherein In the step 2, a deep learning framework or Scikit-learn machine learning library is used to build a fault tracing model, wherein the fault tracing model includes but is not limited to a neural network, a decision tree, a random forest, and a support vector machine, and the constructed fault tracing model is trained using the constructed data set.

5. The method according to claim 1, wherein For data that needs data enhancement, synthetic minority class oversampling technology is used for data enhancement.

6. The method according to claim 1, characterized in that The following steps are also included: Step 5: Calculate the category entropy CE. The formula is as follows: Among them, N i represents the number of samples in the constructed dataset of type i fault, N z represents the total sample size; Calculate the confidence entropy value MCE, the formula is as follows: Among them, P i (x) represents the predicted probability that the sample x output by the fault tracing model in step 2 belongs to the i-th type of fault sample, D i represents the set of type i fault samples; The formula for calculating the comprehensive category entropy value and the confidence entropy value is used to establish the formula model D, as follows: in, Indicates the preset parameter value of the sparsity of the balanced model. max(CE) and max(MCE) represent the maximum category entropy and maximum confidence entropy of all categories of fault types, respectively. CE i and MCE i They represent the category entropy value and confidence entropy value of the i-th type of fault respectively; Step 6: When D > 0.8, step 4 is triggered to perform data enhancement on the data of this type of fault.

Images