A Parameter Optimization Algorithm of Support Vector Machine for Transformer Fault Diagnosis

Abstract

The present invention discloses a support vector machine parameter optimization algorithm for transformer fault diagnosis. The kernel function of the support vector machine adopts the radial basis kernel function. For the kernel function parameter η, the average Euclidean distance between different-class samples is first calculated as the reference value η<subgt;ref< / subgt>. Fix η = η<subgt;ref< / subgt> and perform logarithmic equidistant search for the penalty factor C in the interval [0.001, 1000], and use the cross-validation method to obtain C<subgt;ref< / subgt> with the highest classification accuracy. Set (η<subgt;ref< / subgt>, C<subgt;ref< / subgt>) as the support vector machine parameters, and use all samples to train the support vector machine to obtain support vectors. Then calculate the average Euclidean distance between different-class support vectors as η, fix η and perform logarithmic equidistant search and cross-validation for C again to obtain the optimal C. Finally, set (η, C) as the support vector machine parameters and use all samples to train the support vector machine. The efficiency of training the support vector machine model is greatly improved on the premise of ensuring the diagnostic accuracy, so that the model can be continuously updated as the fault data accumulates.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to a support vector machine parameter optimization technology for transformer fault diagnosis. Background Art

[0002] As a key device in power transmission and distribution, the transformer has an important impact on the safe and stable operation of the power system. Therefore, it is necessary to monitor the operating status of the transformer and implement a fault diagnosis strategy to detect and eliminate potential hazards in a timely manner to avoid serious damage. The three-ratio method based on dissolved gas analysis proposed by the International Electrotechnical Commission is one of the most widely used tools for power transformer fault diagnosis. However, due to the relatively rigid diagnostic mechanism of this method, misdiagnosis often occurs when diagnosing samples near the classification boundary value, bringing costly consequences to stakeholders.

[0003] In recent years, with the increasing maturity of artificial intelligence technology, scholars have tried to apply it to establish a complex non-linear correspondence relationship between transformer faults and dissolved gas data in oil. Among these technologies, the support vector machine has attracted much attention due to its solid theoretical foundation and outstanding learning ability. There are already many cases proving that transformer fault diagnosis based on the support vector machine can improve the diagnostic accuracy to a certain extent.

[0004] However, at the present stage, there is still a problem that the efficiency of the support vector machine parameter optimization algorithm is relatively low, resulting in the difficulty of the diagnostic model to be frequently updated with the accumulation of fault data and the difficulty of applying it to actual situations. Currently, the commonly used algorithms include the grid search method, optimization methods combined with heuristic algorithms (particle swarm algorithm, simulated annealing algorithm, genetic algorithm, immune algorithm, etc.), Bayesian optimization method, etc. These algorithms have their own advantages and disadvantages in terms of classification accuracy and training efficiency. Generally speaking, the higher the final classification accuracy of the parameter optimization algorithm, the longer the training time required, resulting in the difficulty of the transformer fault diagnosis model based on the support vector machine to be frequently updated with the accumulation of fault data. Summary of the Invention

[0005] Aiming at the deficiencies mentioned in the above technical background, the purpose of the present invention is to provide a support vector machine parameter optimization algorithm for transformer fault diagnosis.

[0006] The present invention can be realized by the following technical solutions:

[0007] A support vector machine parameter optimization algorithm for transformer fault diagnosis, the support vector machine kernel function adopts the radial basis kernel function, and each support vector machine in the diagnostic framework first calculates the average Euclidean distance between different-class samples as a reference value for the kernel function parameter η η ref .

[0008] Fix the kernel function parameter η =η ref And perform a logarithmic equidistant search for the penalty factor C in the interval [0.001, 1000], and use the cross-validation method to obtain the reference value of the penalty factor with the highest classification accuracy. C ref 。

[0009] Set ([ η ref , C ref ) as the support vector machine parameters, and use all the samples to train the support vector machine to obtain the support vectors.

[0010] Calculate the average Euclidean distance between different-class support vectors as the kernel function parameter η.

[0011] Fix the kernel function parameter η and perform a logarithmic equidistant search for the penalty factor C in the interval [0.001, 1000], and use the cross-validation method to obtain the penalty factor C with the highest classification accuracy.

[0012] Finally, set (η, C) as the support vector machine parameters, and use all the samples to train the support vector machine to obtain the fault diagnosis model.

[0013] Furthermore, each support vector machine in the diagnosis framework uses the radial basis kernel function. When using the radial basis kernel function for samples with significant clustering characteristics, the optimal kernel function parameter η should make the influence range of each support vector not exceed that of other different-class support vectors as much as possible. Therefore, the optimal η can be obtained by calculating the average Euclidean distance between different-class support vectors. However, since the support vectors can only be determined when the support vector machine training is completed, initially start from the special case where all samples are support vectors, and calculate the average Euclidean distance between different-class samples as the reference value. η ref 。

[0014] Furthermore, the optimal value of the penalty factor C is closely related to the positions of the outliers in a specific sample set, and it is difficult to directly calculate the optimal C through theoretical derivation. Therefore, the traditional logarithmic equidistant search method is still used. Fix the kernel function parameter η = η ref to optimize the penalty factor C. According to experience, the optimal C value is generally in the interval [0.001, 1000]. Logarithmize and equally divide it to obtain multiple trial values, assign them to C to construct a support vector machine, and select the trial value with the highest classification accuracy through the cross-validation method as the reference value C of the penalty factor. ref 。

[0015] Furthermore, the support vectors are obtained through training, and the average Euclidean distance between different-class support vectors is calculated as the final kernel function parameter η.

[0016] Advantages of the present invention:

[0017] The present invention can greatly improve the efficiency of training the support vector machine model on the premise of ensuring the diagnostic accuracy, so that the diagnostic model can be continuously updated as the fault data accumulates. Description of the drawings

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings;

[0019] Figure 1 is the transformer fault diagnosis framework;

[0020] Figure 2 is the implementation flowchart of the support vector machine parameter optimization algorithm for transformer fault diagnosis. Detailed implementation manners

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0022] The present invention will be further described below with reference to the drawings.

[0023] Figure 1 As shown, it is the transformer fault diagnosis framework, where SVM in the figure is the English abbreviation of the support vector machine. Since the support vector machine is a binary classifier, when applied to multi-classification problems, multiple support vector machines need to be combined and used, and the specific fault type to which the sample belongs is determined through multiple dichotomy methods.

[0024] Figure 2 As shown, it is the implementation flowchart of the support vector machine parameter optimization algorithm for transformer fault diagnosis.

[0025] Step 1, each support vector machine in the diagnosis framework adopts a radial basis kernel function, and the functional formula is as follows:

[0026]

[0027] Where η is the kernel function parameter, which is used to control the influence range of support vectors. When using the radial basis kernel function for samples with significant clustering characteristics, the optimal η should be such that the influence range of each support vector does not exceed that of other different-class support vectors as much as possible. Thus, the optimal η can be obtained by calculating the average Euclidean distance between different-class support vectors. However, since the support vectors can only be determined when the support vector machine training is completed, initially start from the special case where all samples are support vectors. Suppose there are m1 positive-class samples and m2 negative-class samples in the training set, then the reference value of the kernel function parameter η ref can be calculated according to the following formula:

[0028]

[0029] In the formula d ij is the Euclidean distance from the i-th positive-class support vector to the j-th negative-class support vector.

[0030] Step 2, fix the kernel function parameter η = η ref , and optimize the penalty factor C. Since the optimal value of C is closely related to the positions of outliers in a specific sample set and it is difficult to directly calculate the optimal C through theoretical derivation, the traditional logarithmic equal-spacing search method is still adopted. According to experience, the optimal value of C is generally in the interval [0.001, 1000]. Logarithmize and equally divide it to obtain multiple trial values, assign them to C to construct a support vector machine, and select the trial value with the highest classification accuracy rate through the cross-validation method as the reference value of the penalty factor C ref .

[0031] Step 3, set ( η ref , C ref ) as the support vector machine parameters, and use all samples to train the support vector machine to determine the support vectors.

[0032] Step 4, suppose after the training in Step 3, n1 positive-class support vectors and n2 negative-class support vectors are obtained, and calculate the average Euclidean distance between different-class support vectors as the final kernel function parameter η according to the following formula:

[0033]

[0034] Step 5, fix the kernel function parameter η and perform logarithmic equal-spacing search on the penalty factor C in the interval [0.001, 1000], and use the cross-validation method to obtain the trial value with the highest classification accuracy rate as the final penalty factor C.

[0035] Step 6: Set (η, C) as the parameters of the support vector machine and use all samples to train the support vector machine. After completing the above parameter optimization process for each support vector machine in the diagnosis framework, a transformer fault diagnosis model is obtained.

[0036] In the description of this specification, the descriptions referring to the terms "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0037] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and the descriptions in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed.

Claims

1. A transformer fault diagnosis method based on support vector machines, characterized in that, Including the following steps: Collect the data of dissolved gases in transformer oil and input it as a sample; The radial basis kernel function is adopted as the kernel function of the support vector machine, and the average Euclidean distance between different types of samples is calculated as the reference value of the kernel function η ref , , where m1 is the number of positive class samples in the training set, and m2 is the number of negative class samples in the training set d ij is the Euclidean distance from the i-th positive class support vector to the j-th negative class support vector in the training set Fix the kernel function parameters η = η ref , logarithmically and equally search for the penalty factor C within the interval [0.001, 1000], and obtain the reference value of the penalty factor with the highest classification accuracy through the cross-validation method C ref ; Set (η ref , C ref ) as the support vector machine parameters, and use all the samples to train the support vector machine to obtain support vectors; Calculate the average Euclidean distance between heterogeneous support vectors as the final kernel function parameter η , and re-optimize to obtain the final penalty factor C; Set (η, C) as the support vector machine parameters to construct a transformer fault diagnosis model and diagnose the transformer fault type in real time.

2. The method for transformer fault diagnosis based on support vector machine according to claim 1, characterized in that Each support vector machine in the diagnostic framework uses a radial basis kernel function. When using the radial basis kernel function for samples with significant clustering characteristics, the optimal kernel function parameter η should ensure that the influence range of each support vector does not exceed that of other heterogeneous support vectors. Therefore, the optimal η can be obtained by calculating the average Euclidean distance between heterogeneous support vectors. However, since the support vectors can only be determined when the support vector machine training is completed, initially, starting from the special case where all samples are support vectors, the average Euclidean distance between heterogeneous samples is calculated as a reference value η ref 。

Images