High-voltage switch cabinet partial discharge mode identification method based on improved extreme gradient lifting tree algorithm

Abstract

The invention relates to the technical field of high-voltage switch cabinet partial discharge identification, and discloses a high-voltage switch cabinet partial discharge mode identification method based on an improved extreme gradient boosting tree algorithm. The method comprises the following steps: firstly, preprocessing data, standardizing and coding the data, and dividing the data into a training set and a test set; then initializing a particle swarm algorithm, including setting parameters of the particle swarm algorithm, randomly initializing the speed and the position of particles, calculating a fitness value and updating the optimal position, assigning the position to the model to construct an improved extreme gradient lifting tree algorithm model, calculating accuracy and updating the optimal position and the global optimal position of each particle; and finally, updating speed and position judgment stopping conditions of each particle according to a formula, and if the number of iterations is reached, stopping iteration and outputting optimal parameters. According to the method, the classification efficiency and accuracy can be further improved, and an effective scheme is provided for efficiently identifying the partial discharge type of the high-voltage switch cabinet.

Description

Technical Field

[0001] The invention relates to the technical field of partial discharge recognition of high-voltage switch cabinets, and in particular to a partial discharge pattern recognition method of a high-voltage switch cabinet based on an improved extreme gradient boosting tree algorithm. Background Art

[0002] The high-voltage switchgear industry is developing rapidly, with not only a variety of products changing with each passing day, but also technological innovations attracting much attention. However, with the continuous improvement of the power distribution network, higher standards have been put forward for the performance of high-voltage switchgear. Once a switchgear fails, it may threaten the stability of the power distribution network and cause significant losses to users. Therefore, timely evaluation of the operating status of the switchgear, detection of potential faults, and taking appropriate maintenance measures are crucial to ensure the safe operation of the switchgear.

[0003] Switchgear equipment may have insulation problems due to a variety of factors such as manufacturing, transportation, installation defects, delayed maintenance or complex operating environment. During normal operation, uneven electric fields may cause partial discharge. If not handled in time, it will lead to deterioration of insulation performance and even cause serious insulation damage. Partial discharge is a key indicator to measure the degradation of insulation performance of electrical equipment, especially high-voltage switchgear. According to the causes and characteristics of partial discharge, it can be preliminarily classified into four types: tip discharge, suspended discharge, surface discharge and bubble discharge. Each type has different effects on electrical equipment. Therefore, the identification of partial discharge patterns in switchgear can not only accurately evaluate the insulation status of the equipment, but also has become an important technology for fault diagnosis and operation and maintenance of electrical equipment. Summary of the invention

[0004] The purpose of the present invention is to overcome the problem of low accuracy and efficiency of the existing partial discharge pattern recognition classification model in identifying a certain type of partial discharge, and to provide a method for partial discharge pattern recognition of high-voltage switch cabinets based on an improved extreme gradient boosting tree algorithm. According to the method described in the present invention, the classifier is first integrated and optimized by improving the extreme gradient boosting tree algorithm to increase the accuracy of pattern recognition, and then the particle swarm algorithm is used to update and improve the six hyperparameters of the extreme gradient boosting tree algorithm, further improving the classification efficiency and accuracy, and providing an effective solution for efficient identification of partial discharge types of high-voltage switch cabinets.

[0005] In order to achieve the above object, the present invention provides a method for identifying partial discharge patterns of high-voltage switch cabinets based on an improved extreme gradient boosting tree algorithm, the method comprising the following steps:

[0006] (1) First, data preprocessing is performed to standardize and encode the data and divide it into training and test sets;

[0007] (2) Then, the particle swarm algorithm is initialized, including setting the parameters of the particle swarm algorithm and randomly initializing the speed and position of the particles, calculating the fitness value and updating the optimal position, assigning the position to the model to construct an improved extreme gradient boosting tree algorithm model, and calculating the accuracy and updating the optimal position of each particle and the global optimal position;

[0008] (3) Finally, the speed and position of each particle are updated according to the formula to determine the stopping condition. If the number of iterations is reached, the iteration is terminated and the optimal parameters are output to optimize the parameters of the improved extreme gradient boosting tree algorithm.

[0009] Preferably, in step (1), the data is initialized and an improved extreme gradient boosting tree algorithm model is constructed.

[0010] For a local discharge signal data set D containing n samples and m features, i ,y i )}, the output value of the integrated classifier model composed of K weak learners is expressed as follows:

[0011]

[0012] Here, f is the function representing each classification and regression tree.

[0013] Preferably, for each tree, function f(x) = ω q(x) , where ω∈R T ;q:R m →T represents the tree structure that maps samples to specific leaf nodes, T is the number of leaf nodes, and ω is the weight of the leaf node.

[0014] Preferably, the improved extreme gradient boosting tree algorithm model uses an additive training method by continuously adding trees and splitting features to find the optimal model structure. The predicted value when the tth tree is generated is as follows:

[0015]

[0016] The objective function of the model consists of two parts: the loss function l and the regularization term Ω:

[0017]

[0018] The loss function represents the predictive power of the model, and the regularization term controls the tree structure.

[0019] Preferably, the expression of the regularization term is:

[0020]

[0021] Among them, γ and λ are parameters that control complexity;

[0022] After the second-order Taylor expansion of the objective function, it is approximated as:

[0023]

[0024] in, are the first and second order partial derivatives of the loss function.

[0025] Preferably, the objective function is expressed as follows after expanding the regularization term and removing the constant term:

[0026]

[0027] Among them, I j is the index set of data points assigned to the jth leaf node;

[0028] By solving the ω that minimizes the objective function j , and get the optimal solution for each leaf node weight:

[0029]

[0030] The final formalized objective function is:

[0031]

[0032] Preferably, in step (2), particles are used to represent individuals in the population, which are distributed in a D-dimensional space. These particles move in the space at a speed V and are at a position X, representing a potential solution to the problem. In each iteration, the fitness of each particle is determined by calculating the fitness function.

[0033] Preferably, the individual optimal solution for each particle as well as the global optimal solution are found based on fitness, and then the speed and position are adjusted as follows:

[0034] V id =ωV id +C1random(0,1)(P id -X id )+C2random(0,1)(P gd -X id )

[0035] X id =X id +V id

[0036] Among them, C1 and C2 are learning factors, representing the influence of individuals and society on particle speed, respectively, and X id represents the current position of the i-th particle, P id represents the optimal position of the i-th particle, P gdrepresents the global optimal position, and ω is the inertia weight.

[0037] Preferably, in step (3), the hyperparameters that need to be optimized to improve the extreme gradient boosting tree algorithm include a learning rate eta, a minimum split loss gamma, a maximum depth of the tree max_depth, a minimum weight required for a child node and min_child_weight, a subsample ratio subsample, and a column sampling ratio colsample_bytree for building a tree, a total of six hyperparameters;

[0038] The position of each particle corresponds to the parameter to be optimized. The particles are located in a six-dimensional space. The speed and position of the i-th particle are expressed as follows:

[0039] V i =[V i,eta ,V i,gamma ,V i,max_depth ,V i,min_child_weight ,V i,subsample ,V i,colsample_bytree ]

[0040] P i =[P i,eta ,P i,gamma ,P i,max_depth ,P i,min_child_weight ,P i,subsample ,P i,colsample_bytree ]

[0041] Improved extreme gradient boosting tree algorithm model using particle positions as parameters for training.

[0042] Preferably, the particle swarm algorithm is initialized, including setting the parameters of the particle swarm algorithm and randomly initializing the speed and position of the particles, calculating the fitness value and updating the optimal position, assigning the position to the model to construct an improved extreme gradient boosting tree algorithm model, and calculating the accuracy and updating the optimal position and the global optimal position of each particle. The formula for updating the speed and position of each particle is as follows:

[0043] V id =ωV id +C1random(0,1)(P id -X id )+C2random(0,1)(P gd -X id )

[0044] X id =X id +V id

[0045] The loop iterates until the specified iteration parameter is reached, and the model outputs the optimal parameter.

[0046] Compared with the prior art, the present invention has the following advantages:

[0047] (1) An improved extreme gradient boosting tree algorithm is used as the classifier for partial discharge model identification and classification. The second-order derivative of the loss function with respect to the desired function is used for optimization to improve the prediction accuracy of the model.

[0048] (2) The particle swarm algorithm is used to optimize the hyperparameters of the improved extreme gradient boosting tree algorithm model, overcoming the shortcomings of relying on experience to adjust parameters and more accurately performing partial discharge pattern recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Flowchart for improving the extreme gradient boosting tree algorithm model. DETAILED DESCRIPTION

[0050] The specific embodiments of the present invention are described in detail below. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0051] The present invention proposes a method for local discharge pattern recognition of high-voltage switch cabinet based on improved extreme gradient boosting tree algorithm. First, data preprocessing is performed to standardize the data. Then, the particle swarm algorithm is initialized, including setting the parameters of the particle swarm algorithm and randomly initializing the speed and position of the particles, calculating the fitness value and updating the optimal position, assigning the position to the model to construct an improved extreme gradient boosting tree algorithm model, and calculating the accuracy and updating the optimal position and global optimal position of each particle. Finally, the speed and position of each particle are updated according to the formula to determine the stop condition. If the number of iterations is reached, the iteration is terminated and the optimal parameters are output. The present invention not only improves the pattern recognition accuracy compared with the traditional model, but also can be combined with the performance characteristic parameters of different devices to be extended to the local discharge pattern recognition of other electrical equipment.

[0052] Specifically, Figure 1 As shown, the method for identifying partial discharge patterns of high-voltage switch cabinets based on the improved extreme gradient boosting tree algorithm of the present invention comprises the following steps:

[0053] (1) First, data preprocessing is performed to standardize and encode the data and divide it into training and test sets;

[0054] (2) Then, the particle swarm algorithm is initialized, including setting the parameters of the particle swarm algorithm and randomly initializing the speed and position of the particles, calculating the fitness value and updating the optimal position, assigning the position to the model to construct an improved extreme gradient boosting tree algorithm model, and calculating the accuracy and updating the optimal position of each particle and the global optimal position;

[0055] (3) Finally, the speed and position of each particle are updated according to the formula to determine the stopping condition. If the number of iterations is reached, the iteration is terminated and the optimal parameters are output to optimize the parameters of the improved extreme gradient boosting tree algorithm.

[0056] In step (1), the data is initialized and the improved extreme gradient boosting tree algorithm model is built.

[0057] For a local discharge signal data set D containing n samples and m features, i ,y i )}, the output value of the integrated classifier model composed of K weak learners is expressed as follows:

[0058]

[0059] where f is a function representing each classification and regression tree.

[0060] For each tree, function f(x) = ω q(x) , where ω∈R T ;q:R m →T represents the tree structure that maps samples to specific leaf nodes, T is the number of leaf nodes, and ω is the weight of the leaf node. The improved extreme gradient boosting tree algorithm model uses an additive training method by continuously adding trees and splitting features to find the optimal model structure. Therefore, the predicted value when the tth tree is generated is as follows:

[0061]

[0062] The objective function of the model consists of two parts: the loss function l and the regularization term Ω:

[0063]

[0064] The loss function represents the predictive ability of the model, and the regularization term controls the tree structure. The expression of the regularization term is:

[0065]

[0066] Among them, γ and λ are parameters that control complexity. After the second-order Taylor expansion of the objective function, it can be approximated as:

[0067]

[0068] in, are the first and second order partial derivatives of the loss function.

[0069] The objective function expression after expanding the regularization term and removing the constant term is:

[0070]

[0071] Among them, I j is the set of indices of the data points assigned to the j-th leaf node.

[0072] By solving the ω that minimizes the objective function j , we can get the optimal solution for each leaf node weight:

[0073]

[0074] The final formalized objective function is:

[0075]

[0076] In step (2), the particle swarm algorithm is initialized, where particles are used to represent individuals in the population, and they are distributed in a D-dimensional space. These particles move in space at a speed V and are at a position X, representing a potential solution to the problem. In each iteration, the fitness of each particle is determined by calculating the fitness function. The individual optimal solution of each particle and the global optimal solution can be found based on the fitness. Then, the speed and position are adjusted as follows:

[0077] V id =ωV id +C1random(0,1)(P id -X id )+C2random(0,1)(P gd -X id )

[0078] X id =X id +V id

[0079] Among them, C1 and C2 are learning factors, representing the influence of individuals and society on particle speed, respectively, and X id represents the current position of the i-th particle, P id represents the optimal position of the i-th particle, P gd represents the global optimal position, ω is the inertia weight

[0080] In step (3), the particle swarm algorithm is used to optimize and improve the hyperparameters of the extreme gradient boosting tree algorithm.

[0081] The hyperparameters that need to be optimized to improve the extreme gradient boosting tree algorithm include the learning rate eta, the minimum split loss gamma, the maximum depth of the tree max_depth, the minimum weight and min_child_weight required for child nodes, the subsample ratio subsample, and the column sampling ratio colsample_bytree for building trees, a total of six hyperparameters. The position of each particle corresponds to the parameter to be optimized. Therefore, the particles are located in a six-dimensional space. The speed and position of the i-th particle can be expressed as follows:

[0082] V i =[V i,eta ,V i,gamma ,V i,max_depth ,V i,min_child_weight ,V i,subsample ,V i,colsample_bytree ]

[0083] P i =[P i,eta ,P i,gamma ,P i,max_depth ,P i,min_child_weight ,P i,subsample ,P i,colsample_bytree ]

[0084] Improved extreme gradient boosting tree algorithm model using particle positions as parameters for training.

[0085] Initialize the particle swarm algorithm, including setting the parameters of the particle swarm algorithm and randomly initializing the speed and position of the particles, calculating the fitness value and updating the optimal position, assigning the position to the model to build an improved extreme gradient boosting tree algorithm model, and calculating the accuracy and updating the optimal position of each particle and the global optimal position. The formula for updating the speed and position of each particle is as follows:

[0086] V id =ωV id +C1random(0,1)(P id -X id )+C2random(0,1)(P gd -X id )

[0087] X id =X id +V id

[0088] The loop iterates until the specified iteration parameter is reached, and the model outputs the optimal parameter.

[0089] Compared with the traditional method, this method first integrates and optimizes the classifier by improving the extreme gradient boosting tree algorithm, thereby increasing the accuracy of pattern recognition. Then, the particle swarm algorithm is used to update and improve the six hyperparameters of the extreme gradient boosting tree algorithm, thereby further improving the classification efficiency and accuracy. This provides an effective solution for efficient identification of partial discharge types in high-voltage switchgear.

[0090] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited thereto. Within the technical concept of the present invention, the technical solution of the present invention can be subjected to a variety of simple modifications, including the combination of various technical features in any other suitable manner, and these simple modifications and combinations should also be regarded as the contents disclosed by the present invention and belong to the protection scope of the present invention.

Claims

1. A method for partial discharge pattern recognition of high-voltage switchgear based on an improved extreme gradient boosting tree algorithm, characterized in that: The method comprises the following steps: (1) First, data preprocessing is performed to standardize and encode the data and divide it into training and test sets; (2) Then, the particle swarm algorithm is initialized, including setting the parameters of the particle swarm algorithm and randomly initializing the speed and position of the particles, calculating the fitness value and updating the optimal position, assigning the position to the model to construct an improved extreme gradient boosting tree algorithm model, and calculating the accuracy and updating the optimal position of each particle and the global optimal position; (3) Finally, the speed and position of each particle are updated according to the formula to determine the stopping condition. If the number of iterations is reached, the iteration is terminated and the optimal parameters are output to optimize the parameters of the improved extreme gradient boosting tree algorithm.

2. The method for identifying partial discharge patterns of high-voltage switch cabinets based on an improved extreme gradient boosting tree algorithm according to claim 1 is characterized in that: In step (1), initialize the data and build an improved extreme gradient boosting tree algorithm model. For a local discharge signal data set D containing n samples and m features, i ,y i )}, the output value of the integrated classifier model composed of K weak learners is expressed as follows: Here, f is the function representing each classification and regression tree.

3. The method for identifying partial discharge patterns of high-voltage switch cabinets based on an improved extreme gradient boosting tree algorithm according to claim 2 is characterized in that: For each tree, function f(x) = ω q(x) , where ω∈R T ;q:R m →T represents the tree structure that maps samples to specific leaf nodes, T is the number of leaf nodes, and ω is the weight of the leaf node.

4. The method for identifying partial discharge patterns of high-voltage switch cabinets based on an improved extreme gradient boosting tree algorithm according to claim 3 is characterized in that: The improved extreme gradient boosting tree algorithm model uses an additive training method by continuously adding trees and splitting features to find the optimal model structure. The predicted value when the tth tree is generated is as follows: The objective function of the model consists of two parts: the loss function l and the regularization term Ω: The loss function represents the predictive power of the model, and the regularization term controls the tree structure.

5. The method for identifying partial discharge patterns of high-voltage switch cabinets based on an improved extreme gradient boosting tree algorithm according to claim 4 is characterized in that: The expression of the regularization term is: Among them, γ and λ are parameters that control complexity; After the second-order Taylor expansion of the objective function, it is approximated as: in, and are the first and second order partial derivatives of the loss function.

6. The method for identifying partial discharge patterns of high-voltage switch cabinets based on an improved extreme gradient boosting tree algorithm according to claim 5 is characterized in that: The objective function expression after expanding the regularization term and removing the constant term is: Among them, I j is the index set of data points assigned to the jth leaf node; By solving the ω that minimizes the objective function j , and get the optimal solution for each leaf node weight: The final formalized objective function is:

7. The method for identifying partial discharge patterns of high-voltage switch cabinets based on an improved extreme gradient boosting tree algorithm according to claim 1 is characterized in that: In step (2), particles are used to represent individuals in the population. They are distributed in a D-dimensional space. These particles move in the space at a speed V and are at position X, representing a potential solution to the problem. In each iteration, the fitness of each particle is determined by calculating the fitness function.

8. The method for identifying partial discharge patterns of high-voltage switch cabinets based on an improved extreme gradient boosting tree algorithm according to claim 7 is characterized in that: The individual optimal solution for each particle and the global optimal solution are found based on the fitness, and then the speed and position are adjusted as follows: V id =ωV id +C1random(0,1)(P id -X id )+C2random(0,1)(P gd -X id ) X id =X id +V id Among them, C1 and C2 are learning factors, representing the influence of individuals and society on particle speed, respectively, and X id represents the current position of the i-th particle, P id represents the optimal position of the i-th particle, P gd represents the global optimal position, and ω is the inertia weight.

9. The method for identifying partial discharge patterns of high-voltage switch cabinets based on an improved extreme gradient boosting tree algorithm according to claim 1, characterized in that: In step (3), the hyperparameters that need to be optimized to improve the extreme gradient boosting tree algorithm include the learning rate eta, the minimum split loss gamma, the maximum depth of the tree max_depth, the minimum weight required for child nodes and min_child_weight, the subsample ratio subsample, and the column sampling ratio colsample_bytree for building the tree, a total of six hyperparameters; The position of each particle corresponds to the parameter to be optimized. The particles are located in a six-dimensional space. The speed and position of the i-th particle are expressed as follows: V i =[V i,eta ,V i,gamma ,V i,max_depth ,V i,min_child_weight ,V i,subsample ,V i,colsample_bytree ] P i =[P i,eta ,P i,gamma ,P i,max_depth ,P i,min_child_weight ,P i,subsample ,P i,colsample_bytree ] Improved extreme gradient boosting tree algorithm model using particle positions as parameters for training.

10. The method for identifying partial discharge patterns of high-voltage switch cabinets based on an improved extreme gradient boosting tree algorithm according to claim 9, characterized in that: Initialize the particle swarm algorithm, including setting the parameters of the particle swarm algorithm and randomly initializing the speed and position of the particles, calculating the fitness value and updating the optimal position, assigning the position to the model to build an improved extreme gradient boosting tree algorithm model, and calculating the accuracy and updating the optimal position and global optimal position of each particle. The formula for updating the speed and position of each particle is as follows: V id =ωV id +C1random(0,1)(P id -X id )+C2random(0,1)(P gd -X id ) X id =X id +V id The loop iterates until the specified iteration parameter is reached, and the model outputs the optimal parameter.

Images