A Transformer Cost-Sensitive Fault Diagnosis Method Considering Importance Level Classification

Abstract

The present invention discloses a cost-sensitive fault diagnosis method for transformers considering importance level division, including: (1) feature selection and data preprocessing; (2) construction of a fault diagnosis model; (3) division of the importance levels of grid transformers; (4) cost-sensitive fault diagnosis; The present invention mainly introduces a cost-sensitive fault diagnosis method for transformers considering grid-level information, introducing the division of the importance levels of transformers in the grid into cost-sensitive fault diagnosis. Since the misdiagnosis costs of transformers with different importance levels are different, a cost-sensitive matrix is constructed according to the importance level classification, and the parameter optimization of the fault diagnosis model is carried out respectively with the goal of minimizing the final misdiagnosis cost, realizing the cost-sensitive fault diagnosis of transformers considering importance level division, which is of great significance for reasonably arranging transformer maintenance and operation and reducing the operation risk of the power grid, etc.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment fault diagnosis, and particularly relates to a cost-sensitive fault diagnosis method for transformers considering the classification of importance levels. Background Art

[0002] With the development of China's national economy, the demand for electricity is increasing, which promotes the continuous expansion of the power grid scale. As an important power equipment in the modern power system, transformers play a huge role in fields such as long-distance power transmission and power distribution. However, the faults of the transformer itself will not only cause large-scale power grid outages, but also seriously affect the normal production and living order of local people, posing certain challenges to the stability maintenance work of local society. Therefore, it is of great significance to timely detect the fault symptoms of transformers and determine the fault types. Currently, in the aspect of transformer fault diagnosis, dissolved gas analysis (DGA) in oil is the main method. This method analyzes the content or ratio of dissolved gases in oil to judge the operating state of the transformer and detect its potential faults. Common DGA ratio methods include the IEC three-ratio method, Rogers ratio method, Dornenburg ratio method, non-coded ratio method, etc. However, due to the overly absolute ratio boundaries, misdiagnosis is likely to occur near the boundary values. Therefore, intelligent diagnosis methods are now widely used for transformer fault diagnosis.

[0003] However, current transformer fault diagnosis often only pursues the global diagnosis accuracy rate, while ignoring the problem that the misdiagnosis cost is different in actual situations. The cost of diagnosing a normal state as a fault state is the loss of certain equipment maintenance costs. However, if a fault state is diagnosed as a normal state, it will lead to risks such as untimely fault handling and chain faults, bringing huge hidden dangers to the safe and stable operation of the power system. Moreover, for more important transformers, the misdiagnosis cost is greater. Therefore, introducing the concept of cost sensitivity and incorporating the importance level of transformers in the power grid can overcome the shortcomings of traditional diagnosis methods that do not consider misdiagnosis costs and do not consider information at the power grid level, and is of great significance for reasonably arranging transformer maintenance and operation and reducing the operation risks of the power grid. Summary of the Invention

[0004] The technical problem to be solved by the present invention is as follows: By introducing a cost-sensitive method into transformer fault diagnosis, it overcomes the limitation of mainly focusing on the global accuracy rate while ignoring the misdiagnosis cost in the fault diagnosis process. And through the classification of the importance levels of transformers, a cost-sensitive matrix for transformers is established hierarchically, which overcomes the problem that existing fault diagnosis methods often only target specific equipment information and do not consider the differences in the impacts of faults of different equipment on the power grid level.

[0005] To achieve the above-mentioned invention objectives, the technical solution adopted by the present invention is specifically as follows: A cost-sensitive fault diagnosis method for transformers considering importance level classification, comprising the following steps:

[0006] Step 1: Establish a sample feature set by inputting the coding ratio and oil-gas concentration characteristics of the transformer; normalize the input features in the sample feature set and divide them into a training set and a test set;

[0007] Step 2: Construct a fault diagnosis model by introducing a regularization term. The fault diagnosis model is as follows:

[0008]

[0009] where: β 2 represents the structural risk, ε 2 is the total error of N training samples, representing the empirical risk, and C is the regularization coefficient, which balances the empirical risk and the structural risk;

[0010] Step 3: The fault diagnosis model analyzes the parameters in the training set and the test set and outputs the transformer fault type;

[0011] Step 4: Divide the importance level of the transformer based on the risk caused to the power grid after the transformer is taken out of operation;

[0012] Step 5: Construct a cost-sensitive fault diagnosis model for the transformer with the goal of minimizing the misdiagnosis cost according to the transformer fault type and the transformer importance level. Among them, the construction process of the cost-sensitive fault diagnosis model for the transformer is as follows:

[0013] Establish a cost-sensitive matrix Cost(i,j) according to different transformer importance levels;

[0014] Construct a cost function according to different cost-sensitive matrices. The construction of the cost function is to calculate the sum of the costs of all misdiagnosed types of transformers, as shown in the following formula:

[0015]

[0016] In the formula: Cost act is the overall misdiagnosis cost, Cost m (i,j) is the misdiagnosis cost of the mth sample, n is the number of samples, i represents the true category of the sample, and j represents the predicted category of the sample;

[0017] Optimize with the cost function as the optimization objective of minimizing the overall misdiagnosis cost of the fault diagnosis model and output the cost-sensitive fault diagnosis type of the transformer.

[0018] Furthermore, the fault diagnosis model also includes an optimization module for the precision rate in the transformer fault diagnosis process, comprising the following steps:

[0019] The optimization module optimizes the weight vector and bias of the fault diagnosis model through an adaptive differential evolution method;

[0020] The optimization module optimizes the search and development capabilities of the fault diagnosis model through the following formula;

[0021]

[0022]

[0023] where: F is the mutation factor, CR is the crossover factor, F max is the maximum value of the set mutation factor, F min is the minimum value of the set mutation factor, CR max is the maximum value of the set crossover factor, CR min is the minimum value of the set crossover factor, T is the current iteration number, T m is the maximum iteration number;

[0024] The transformer cost-sensitive fault diagnosis model performs target optimization on the optimization module through a cost function to minimize the misdiagnosis cost.

[0025] Beneficial effects

[0026] 1. The present invention introduces a cost-sensitive mechanism into transformer fault diagnosis, and establishes a cost-sensitive fault diagnosis model with the goal of minimizing the misdiagnosis cost. This method can improve the diagnostic accuracy of high misdiagnosis cost categories and overcome the limitation of traditional fault diagnosis methods that only focus on the global accuracy and ignore the misdiagnosis cost.

[0027] 2. The proposed solution of the present invention considers the difference in misdiagnosis costs of transformers with different importance levels, classifies the importance levels of transformers and establishes cost-sensitive fault diagnosis models respectively, which can further improve the diagnostic accuracy of high misdiagnosis cost categories of high-importance-level transformers, conforms to the actual engineering application, and is of great significance for reasonably arranging transformer maintenance and operation and reducing the operation risk of the power grid. Brief description of the drawings

[0028] Figure 1 is the flowchart of the present invention. Detailed implementation manners

[0029] The following makes a detailed description of the present invention in conjunction with the drawings.

[0030] As Figure 1 shown, the present invention provides a transformer cost-sensitive fault diagnosis method considering importance level classification, including the following steps:

[0031] Step 1: Establish a sample feature set based on the coding ratio and oil-gas concentration characteristics of the input transformer; perform normalization processing on the input features in the sample feature set and divide them into a training set and a test set; where:

[0032] Feature selection and data preprocessing. The content and proportional relationship of dissolved gases in oil can reflect the operating status of the transformer and provide a basis for transformer fault diagnosis. Extract the concentration data of dissolved gases H2, CH4, C2H6, C2H4, and C2H2 in oil, and select ratios according to the IEC three-ratio method, Rogers ratio method, Dornenburg ratio method, non-coding ratio method, etc. Take the coding ratio and oil-gas concentration as input features, and encode the fault types (normal, medium and low temperature overheating, high temperature overheating, partial discharge, low-energy discharge, high-energy discharge) as output features. Then, perform maximum-minimum normalization on the input feature data and divide it into a training set and a test set;

[0033] Step 2: Construct a fault diagnosis model by introducing a regularization term. The fault diagnosis model is:

[0034]

[0035] where: β 2 represents the structural risk, ε 2 is the total error of N training samples, representing the empirical risk, and C is the regularization coefficient, which balances the empirical risk and the structural risk;

[0036] Construction of the fault diagnosis model. Build a fault diagnosis model of the extreme learning machine. However, since the standard extreme learning machine is a learning method that minimizes empirical risk, the generalization ability of the model is poor and it is prone to overfitting. Therefore, the extreme learning machine is improved by introducing a regularization term to reduce the structural risk of the extreme learning machine and improve the generalization ability.

[0037] The fault diagnosis model also includes an optimization module for the precision rate in the transformer fault diagnosis process, including the following steps:

[0038] The optimization module optimizes the weight vector and bias of the fault diagnosis model through the adaptive differential evolution method;

[0039] The optimization module optimizes the search and development capabilities of the fault diagnosis model through the following formula;

[0040]

[0041]

[0042] where: F is the mutation factor, CR is the crossover factor, F max is the set maximum value of the mutation factor, Fmin is the minimum value of the set mutation factor, CR max is the maximum value of the set crossover factor, CR min is the minimum value of the set crossover factor, T is the current iteration number, T m is the maximum iteration number;

[0043] The cost-sensitive fault diagnosis model of the transformer optimizes the target of the optimization module by minimizing the misdiagnosis cost through a cost function.

[0044] In the process of transformer fault diagnosis, the weight vector and bias are randomly assigned. The uncertainty of the parameters will affect the accuracy of the final fault diagnosis result. Therefore, it is necessary to optimize the parameters of the fault diagnosis model. The weight vector and bias of the fault diagnosis model are optimized by the adaptive differential evolution method. The optimization module establishes an optimization model by adaptively updating the mutation factor and crossover factor, improves the search ability and development ability of the optimization model, and then improves the accuracy of the fault diagnosis model.

[0045] Step 3: The fault diagnosis model analyzes the parameters in the training set and the test set and outputs the transformer fault type;

[0046] Step 4: Divide the importance level of the transformer based on the risk caused to the power grid after the transformer is taken out of service; Taking the IEEE RTS-79 system as an example, divide the importance level of the transformer based on the risk caused to the power grid after the transformer is taken out of service. Calculate the overvoltage risk, low voltage risk, line overload risk, transformer overload risk, and load loss risk, and determine the weights of each risk index through the analytic hierarchy process, and then calculate the comprehensive risk index to characterize the importance of the transformer in the power grid. Finally, according to the As Low As Reasonably Practicable (ALARP) principle, the importance level of the transformer is divided into generally important, moderately important, and extremely important.

[0047] Step 5: Construct a cost-sensitive fault diagnosis model of the transformer with the lowest misdiagnosis cost as the goal according to the transformer fault type and the transformer importance level. Among them: The construction process of the cost-sensitive fault diagnosis model of the transformer:

[0048] Establish a cost-sensitive matrix Cost(i,j) according to different transformer importance levels;

[0049] Cost-sensitive fault diagnosis. According to the different importance levels of transformers, the cost-sensitive matrices for generally important transformers, moderately important transformers, and extremely important transformers are respectively established for transformers of different importance levels. The construction of the cost-sensitive matrix is based on the cost differences between misdiagnoses of different fault types. The cost of diagnosing the normal state as the fault state is the loss of certain equipment maintenance costs. However, if the fault state is diagnosed as the normal state, it will lead to risks such as untimely fault handling and triggering of cascading faults, bringing huge hidden dangers to the safe and stable operation of the power system. Therefore, different misdiagnosis costs are set between different fault types. At the same time, the misdiagnosis cost of more important transformers is greater. Therefore, the cost-sensitive matrix Cost(i,j) is established according to the different importance levels of transformers;

[0050] Cost functions are respectively constructed according to different cost-sensitive matrices. The construction of the cost function is to calculate the sum of the costs of misdiagnosis types of all transformers, as shown in the following formula:

[0051]

[0052] In the formula: Cost act is the overall misdiagnosis cost, Cost m (i,j) is the misdiagnosis cost of the m-th sample, n is the number of samples, i represents the true category of the sample, and j represents the predicted category of the sample;

[0053] Finally, the cost function is used as the fitness calculation formula of the SADE model, and the optimization goal of the SADE model is to minimize the fitness, that is, to minimize the overall misdiagnosis cost, so as to optimize the parameters of the RELM model and realize the cost-sensitive fault diagnosis of transformers considering the importance equivalent division.

Claims

1. A cost-sensitive fault diagnosis method for transformers considering importance level classification, characterized in that It includes the following steps: Step 1: Establish a sample feature set based on the coding ratio of the input transformer and the oil-gas concentration characteristics; normalize the input features in the sample feature set and divide them into a training set and a test set; Step 2: Construct a fault diagnosis model by introducing a regularization term. The fault diagnosis model is: Among them: ||β|| 2 represents the structural risk, ||ε|| 2 is the total error of N training samples, representing the empirical risk, and C is the regularization coefficient, which balances the empirical risk and the structural risk; Step 3: The fault diagnosis model analyzes the parameters in the training set and the test set and outputs the transformer fault type; Step 4: Divide the transformer importance level based on the risk caused to the power grid after the transformer is taken out of operation; Step 5: Construct a cost-sensitive fault diagnosis model for the transformer with the lowest misdiagnosis cost as the goal according to the transformer fault type and the transformer importance level. Among them: The construction process of the cost-sensitive fault diagnosis model for the transformer: Establish a cost-sensitive matrix Cost(i,j) according to different transformer importance levels; Construct cost functions according to different cost-sensitive matrices. The construction of the cost function is to calculate the cost sum of all misdiagnosis types of transformers, as shown in the following formula: Where: Cost act is the overall misdiagnosis cost, Cost m (i, j) is the misdiagnosis cost of the m-th sample, n is the number of samples, i represents the true class of the sample, and j represents the predicted class of the sample; Use the cost function as the optimization goal of minimizing the overall misdiagnosis cost of the fault diagnosis model to optimize and output the cost-sensitive fault diagnosis type of the transformer.

2. The method for transformer cost-sensitive fault diagnosis considering importance level division according to claim 1, characterized in that: The fault diagnosis model also includes an optimization module for optimizing the precision rate in the transformer fault diagnosis process, including the following steps: The optimization module optimizes the weight vector and bias of the fault diagnosis model through the adaptive differential evolution method; The optimization module optimizes the search and development capabilities of the fault diagnosis model through the following formula; Where: F is the mutation factor, CR is the crossover factor, F max is the maximum value of the set mutation factor, F min is the minimum value of the set mutation factor, CR max is the maximum value of the set crossover factor, CR min is the minimum value of the set crossover factor, T is the current iteration number, T m is the maximum number of iterations; The cost-sensitive fault diagnosis model for the transformer optimizes the optimization module with the cost function as the minimum misdiagnosis cost.

Images