Traction transformer fault diagnosis method for optimizing support vector machine through novel multi-strategy sparrow search algorithm

Abstract

The embodiment of the invention discloses a traction transformer fault diagnosis method for optimizing a support vector machine through a novel multi-strategy sparrow search algorithm, and relates to the field of transformer fault diagnosis. From the perspective of optimizing parameters of a support vector machine, aiming at the problems of low diagnosis accuracy, poor stability and the like caused by the fact that the support vector machine (SVM) is easily influenced by optimal parameters in transformer fault diagnosis, a multi-strategy sparrow search algorithm (ISSA) is designed to optimize the SVM. Firstly, SPM chaotic mapping and refraction reverse learning are adopted to initialize a population, a sine dynamic self-adaptive weight is introduced into a discoverer, a tangent flight operator is introduced into an inserter, a Cauchy inverse operator is adopted for a warning person to improve an original position updating formula, an optimization parameter is assigned to an SVM, and finally the model is applied to fault diagnosis of the transformer. The new model designed by the invention has the advantage that the fault diagnosis accuracy of the traction transformer is greatly improved.

Description

Technical Field

[0001] The invention relates to the field of traction transformer fault diagnosis, and in particular to a method for optimizing a support vector machine using a novel multi-strategy sparrow search algorithm. Technical Background

[0002] With the rapid development of my country's economy, the construction of electrified railways has become more and more perfect. The traction transformer of the traction substation is a key equipment in the power supply system of the electrified railway. It is responsible for converting the high-voltage electric energy of the power system into low-voltage electric energy suitable for electric locomotives. The normal operation of the traction transformer is crucial to the safe and reliable operation of the electrified railway. Therefore, fault diagnosis and maintenance of the traction transformer has extremely high practical significance and application value.

[0003] With the development of science and technology, the fault diagnosis technology of traction transformers is gradually developing towards automation and intelligence. At present, there are many fault diagnosis technologies based on different principles, such as fault diagnosis based on waveform analysis, fault diagnosis based on energy analysis, fault diagnosis based on neural network, etc. SVM has significant advantages in the fault diagnosis of traction transformers in electrified railway traction substations. Its characteristics of processing high-dimensional data, strong generalization ability, applicability to small sample data, processing nonlinear problems, robustness and interpretability make it one of the important tools in the field of fault diagnosis.

[0004] Support vector machine is a pattern recognition method based on statistical learning theory, which shows unique advantages in solving small sample, nonlinear and high-dimensional pattern recognition problems. Some traditional fault diagnosis algorithms have large computational complexity or low accuracy when dealing with nonlinear problems. Compared with logistic regression and neural networks, support vector machine provides a clearer and more powerful way to learn complex nonlinear equations. This method fully and effectively clarifies linear inseparability and can obtain satisfactory results for different data sets. Although support vector machine has certain advantages in fault diagnosis, the accuracy of the model is still affected by parameters. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] Aiming at the problem that the support vector machine (SVM) is easily affected by the optimal parameters in the fault diagnosis of transformers, resulting in low diagnostic efficiency and poor stability, the present invention provides a multi-strategy sparrow algorithm (ISSA) for optimizing the parameters of the support vector machine. First, the population is initialized by using SPM chaotic mapping and refraction reverse learning, the sinusoidal dynamic adaptive weight is introduced in the discoverer, the tangent flight operator is introduced in the joiner, the original position update formula is improved by the Cauchy inverse operator for the alerter, the optimization parameters are assigned to the SVM, and finally the model is applied to the fault diagnosis of the transformer. Under the premise of avoiding the sparrow search algorithm from falling into the local optimum, the optimal parameters of the SVM are found, which effectively improves the accuracy of the fault diagnosis of the traction transformer.

[0007] (II) Technical solution

[0008] To achieve the above object, the present invention provides a traction transformer fault diagnosis method using a multi-strategy sparrow search algorithm to optimize SVM, comprising the following steps:

[0009] Step 1: Initialize ISSA parameters using SPM and refraction reverse learning.

[0010] Step 2: Use sinusoidal dynamic adaptive weights to update the discoverer's position.

[0011] Step 3: Use the tangent flight operator to update the joiner's position.

[0012] Step 4: Use the inverse Cauchy operator to update the position of the sentinel.

[0013] Step 5: Assign the optimization parameters obtained by ISSA to SVM.

[0014] Step 6: Use the new model to classify faults

[0015] (III) Beneficial effects

[0016] a. The present invention can reduce the local optimality of the sparrow search algorithm: the present invention adopts SPM chaotic mapping and refraction reverse learning initialization in the original sparrow search algorithm, introduces sinusoidal dynamic adaptive weights in the discoverer, and introduces tangent flight operator and Cauchy inverse operator in the joiner and the sentinel respectively to improve the sparrow search algorithm, avoid falling into the local optimal value, and greatly improve the global optimization effect.

[0017] b. The present invention can effectively improve the classification accuracy of SVM: the ISSA algorithm is used to optimize the SVM model parameters, find the penalty factor C and the kernel function width parameter σ that are suitable for the model, and optimize the parameters of the two to improve the classification diagnosis accuracy of SVM. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0019] Figure 1 is a flow chart of the method described in an embodiment of the present invention;

[0020] Figure 2 Schematic diagram of population distribution of the SPM chaotic mapping method according to an embodiment of the present invention;

[0021] Figure 3 is a schematic diagram of population distribution of the refraction reverse learning method according to an embodiment of the present invention;

[0022] Figure 4 is a schematic diagram of a sinusoidal adaptive weighting method according to an embodiment of the present invention;

[0023] Figure 5 is a schematic diagram of the spiral perturbation trajectory method according to an embodiment of the present invention;

[0024] Figure 6 Schematic diagram of the tangent flight operator step method according to an embodiment of the present invention;

[0025] Figure 7 is a schematic diagram of the Cauchy inverse accumulation step size method according to an embodiment of the present invention;

[0026] Figure 8 It is a schematic diagram of the test effect of the improved SSA algorithm in a single peak function according to an embodiment of the present invention;

[0027] Fig. 9 It is a schematic diagram of the test effect of the improved SSA algorithm in a multi-peak function according to an embodiment of the present invention;

[0028] Fig.10 is a schematic diagram of the final fault classification diagnosis performed by the method according to the embodiment of the present invention; DETAILED DESCRIPTION

[0029] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0031] The embodiment of the present invention discloses a new method for traction transformer fault diagnosis, which adopts a multi-strategy sparrow search algorithm to optimize the SVM technology. From the perspective of optimizing SVM parameters, this method focuses on introducing a multi-strategy sparrow search algorithm to address the problem that SVM is easily affected by parameters. First, the SPM chaotic mapping and refraction reverse learning initialization are used to initialize the sparrow search algorithm population to increase population diversity. Sine dynamic adaptive weights are introduced into the discoverer to increase the globality of the optimization. Tangent flight operators and Cauchy inverse operators are introduced into the joiner and the alerter to improve the original sparrow search algorithm to avoid falling into the local optimum. Then, the parameters of the support vector machine are optimized, and finally the model is applied to the fault diagnosis of the transformer. In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0032] Combination Figure 1 The embodiment of the present invention discloses a novel multi-strategy sparrow search algorithm to optimize SVM traction transformer fault diagnosis method, comprising the following steps:

[0033] Step 1: Initialize ISSA parameters using SPM and refraction reverse learning.

[0034] Step 2: Use sinusoidal dynamic adaptive weights to update the discoverer's position.

[0035] Step 3: Use the tangent flight operator to update the joiner's position.

[0036] Step 4: Use the inverse Cauchy operator to update the position of the sentinel.

[0037] Step 5: Assign the optimization parameters obtained by ISSA to SVM.

[0038] Step 6: Use the new model to classify faults

[0039] The above steps are explained below with specific methods:

[0040] The population initialization expression in step 1) is as follows:

[0041] Inspired by the typical mapping function, a new type of chaotic mapping is designed. First, the SPM chaotic mapping is introduced to initialize the population to form N populations with globally uniformly distributed individuals. As shown in the following formula:

[0042]

[0043] Then, we use refraction reverse learning to find the reverse solution for the N individuals in the population. Finally, we merge the chaotic population, calculate the respective objective function values, and sort the fitness of these 2N individuals. We select the first N individuals with better fitness values ​​as the initial population. During the initialization process, we can avoid falling into the local optimum and perform global search, which can achieve better initialization. As shown in the following formula:

[0044] X(t+1)'=(X ub +X lb ) / 2+(X ub +X lb ) / (2KN)-X(t+1) / (KN) (2)

[0045] The resulting population initialization is more comprehensive and global, including both local and global individuals, such as Figure 2 , 3 shown.

[0046] The expression for updating the finder position using the sinusoidal adaptive weight in step 2 is as follows:

[0047]

[0048] The sinusoidal inertia weight ω allows the algorithm to have a smaller weight value in the early stage of operation, and the individual sparrows have a stronger local search ability to obtain the individual optimal solution. In the middle of the algorithm, with the changing characteristics of the sine function, the weight value gradually increases, and the search for the global optimal solution begins. In the later stage of the algorithm, the inertia weight gradually decreases, and a detailed local search is carried out to find the optimal solution in the area and the global optimal solution. Figure 4 shown.

[0049] The step 3) uses the tangent flight operator to update the joiner position, as shown in the following formula:

[0050]

[0051] In formula (4) The present invention combines the two methods. Since the position update will only move according to the previous direction, the search space is not comprehensive enough, and some spaces may not be traversed. The tangent flight operator and the spiral strategy are combined to traverse the entire problem space in a more comprehensive search method, effectively improving the search efficiency and avoiding omissions. Figure 5 , Figure 6 .

[0052] The step 4) introduces the Cauchy inverse operator in the sentinel to update the position, as shown in the following formula:

[0053]

[0054] In formula (5) The inverse Cauchy operator is used to consider that after the sentinel finds the natural enemy, it may escape from the optimal position and move closer to the worst position. The present invention proposes a method of combining the inverse Cauchy operator with a variable spiral. The step length of the inverse Cauchy operator is relatively even, and occasionally a large step length may occur. With the addition of the inverse Cauchy cumulative distribution operator, the overall step length is rapidly reduced, and the optimal fitness value is reached more quickly. The inverse Cauchy operator increases the ability to search locally uniformly near the optimal value, thereby increasing the probability of the sparrow moving closer to the optimal position. The improved formula indicates that if the sentinel sparrow is a sparrow in the optimal position, it will escape to a random position between the optimal position and the worst position. In order to avoid falling into the local optimum, the sentinel sparrow will escape away from the optimal position and move closer to the worst position, and will disturb within a small range close to the worst position. Figure 7 shown.

[0055] After the two best ranking positions of the sparrow individuals are determined in step 6), the optimal parameter vector (C, σ) is assigned to the SVM classifier. As shown in the following formula:

[0056]

[0057] When the SVM model is used for classification diagnosis, the classification effect is related to the penalty factor C and the kernel function width parameter σ, which have a great impact on the SVM model classification. Therefore, it is necessary to obtain the optimal parameters to improve the classification accuracy. Figure 8 , 9 shown.

[0058] In step 6), the optimal parameters are input into the SVM trainer to perform training prediction and then obtain the classification accuracy. Fig.10 shown.

Claims

1. A novel multi-strategy sparrow search algorithm optimized SVM traction transformer fault diagnosis method, characterized in that: The steps include: Step S1: Initialize ISSA parameters using SPM and refraction reverse learning. Step S2: Use sinusoidal dynamic adaptive weights to update the finder's position. Step S3: Use the tangent flight operator to update the joiner's position. Step S4: Use the inverse Cauchy operator to update the sentinel position. Step S5: Assign the optimization parameters obtained by ISSA to SVM. Step S6: Use the new model to perform fault classification diagnosis.

2. The design method according to claim 1, characterized in that: The expression for population initialization in step S1 is: SPM chaotic mapping is introduced to initialize the population, forming globally uniformly distributed individuals in N populations. Then, refraction reverse learning is used to find the reverse solution for the N population individuals. Finally, the chaotic populations are merged, the respective objective function values ​​are calculated, and the fitness of these 2N individuals is sorted. The first N individuals with better fitness values ​​are selected as the initial population. The initialization process can avoid falling into the local optimum and perform global search, which can achieve better initialization.

3. The design method according to claim 1, characterized in that: In step S2, the expression for updating the population finder position according to the improved formula is: The dynamic adaptive weight is added to the formula for updating the position of the discoverer. The value changes nonlinearly with the number of iterations, which can avoid the situation where the discoverer function is prone to falling into local extreme values ​​in the later stages of iteration.

4. The design method according to claim 1, characterized in that: In step S3, the expression for updating the position of the population joiner according to the improved formula is: In the position update formula, the update direction of the spiral ψ is to better traverse the local search space, and the tangent flight operator can search the global space. The combination of the two traverses the entire problem space in a more comprehensive search manner, effectively improving the search efficiency and avoiding omissions.

5. The design method according to claim 1, characterized in that: In step S4, the expression for updating the position of the population sentinel according to the improved formula is: In order to solve the problem that the sentinel may escape from the optimal position and move closer to the worst position after discovering the natural enemy, the Cauchy inverse operator is combined with the variable spiral. The step size of the Cauchy inverse operator is more average and the optimal fitness value is reached more quickly, so that the sentinel can find the optimal position faster when escaping.

6. The design method according to claim 1, characterized in that: After the two best ranking positions of the sparrow individuals are determined in step S5, the optimal parameter vector (C, σ) is assigned to the SVM classifier. SVM parameter expression: The optimal segmentation hyperplane of the SVM model is related to the penalty factor C and the kernel function width parameter σ, so it is necessary to optimize the parameters of the two to improve the classification and diagnosis accuracy of the SVM.

7. The design method according to claim 1, characterized in that: In step S6, after the improved sparrow search algorithm is used for optimization, a new fault diagnosis model is established to perform fault classification diagnosis.

Images