Method and device for diagnosing mechanical defects of three-position switch of GIS (Gas Insulated Switchgear) equipment

Through the random forest model and genetic algorithm combined with principal component analysis, the characteristics of current, angular displacement and vibration signals are used to solve the accuracy of mechanical defect recognition in the three-station switch of GIS equipment, and the efficient identification of different types of defects is achieved.

CN118914830APending Publication Date: 2024-11-08CHONGQING UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410973217.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The prior art is difficult to accurately identify different types of mechanical defects in the three-station switch of GIS equipment, resulting in insufficient fault diagnosis.

Method used

The random forest identification model is used to combine genetic algorithms and principal component analysis, and the multi-dimensional characteristic dimensionality reduction of current signals, angular displacement signals and vibration signals is trained to distinguish different types of mechanical defects.

Benefits of technology

The accurate recognition rate of mechanical defects is improved, so that the random forest model can effectively distinguish different types of mechanical defects, and enhance the accuracy of fault diagnosis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118914830A_ABST
    Figure CN118914830A_ABST
Patent Text Reader

Abstract

The invention discloses a method and a device for diagnosing mechanical defects of a three-position switch of GIS equipment. The method comprises the following steps: acquiring a to-be-diagnosed characteristic value; performing feature dimension reduction on the to-be-diagnosed feature values to obtain principal component features; inputting the principal component features into a trained random forest identification model to obtain a mechanical defect diagnosis result; the diagnosis result of the mechanical defect indicates the type of the mechanical defect; the trained random forest identification model is obtained by training based on the multiple sample principal component features and the optimized random forest identification model; the optimized random forest identification model is obtained by optimizing the random forest identification model based on a genetic algorithm; the sample principal component features are obtained based on dimension reduction of a plurality of signal feature values; the plurality of signal characteristic values at least comprise a current signal characteristic value, an angular displacement signal characteristic value and a vibration signal characteristic value. The random forest model is trained through the multi-dimensional features, so that the random forest model can effectively distinguish different types of mechanical defects.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of GIS mechanical defect diagnosis, and in particular to a method and device for diagnosing mechanical defects of a three-position switch of a GIS device. Background Art

[0002] Three-position switches are prone to latent mechanical defects during manufacturing, installation and operation, which can easily cause the opening / closing stroke of gas insulated switchgear (GIS) to be inadequate, and is one of the main factors leading to GIS equipment failure.

[0003] At present, the monitoring and diagnosis capabilities of the internal state of the three-position switch are usually measured through vibration signals. By detecting the vibration signal of the three-position switch and extracting the characteristics of the vibration signal, such as frequency, amplitude, periodicity, etc., and comparing it with known fault modes, potential mechanical defects can be identified. However, since the accuracy of the vibration signal is affected by the installation position, environmental interference and signal processing methods, it may not be possible to effectively distinguish different types of mechanical defects.

[0004] Therefore, how to accurately identify different types of mechanical defects has become an urgent problem to be solved in this field. Summary of the invention

[0005] The present application provides a method and device for diagnosing mechanical defects of a three-position switch of a GIS device, the purpose of which is to accurately identify different types of mechanical defects.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] A method for diagnosing mechanical defects of a three-position switch of a GIS device, comprising:

[0008] Obtain the characteristic value to be diagnosed;

[0009] Performing feature dimension reduction on the feature value to be diagnosed to obtain principal component features;

[0010] The principal component features are input into a trained random forest identification model to obtain a diagnosis result of a mechanical defect; the diagnosis result of the mechanical defect indicates the type of mechanical defect; the trained random forest identification model is trained based on multiple sample principal component features and an optimized random forest identification model; the optimized random forest identification model is obtained by optimizing the random forest identification model based on a genetic algorithm; the sample principal component features are obtained based on dimensionality reduction of multiple signal eigenvalues; the multiple signal eigenvalues ​​at least include current signal eigenvalues, angular displacement signal eigenvalues ​​and vibration signal eigenvalues.

[0011] Optionally, the process of obtaining the trained random forest recognition model based on the principal component features of multiple samples and the optimized random forest recognition model training includes:

[0012] Obtain principal component features of multiple samples in the training set;

[0013] The principal component features of each sample in the training set are used as the input of the task loss function of the optimized random forest identification model, the diagnosis result of the mechanical defect is calculated, and the parameters of the task loss function are continuously adjusted until the accuracy of the diagnosis result of the mechanical defect is greater than the preset threshold, confirming that the optimized random forest identification model is successfully trained.

[0014] Optionally, also include:

[0015] Obtain principal component features of multiple samples in the validation set;

[0016] For each principal component feature of the sample in the validation set, inputting the principal component feature of the sample into the trained random forest recognition model to obtain a target diagnosis result of the mechanical defect;

[0017] When all of the target diagnosis results are consistent with the preset diagnosis results, it is determined that the trained random forest recognition model has passed the verification.

[0018] Optionally, the process of optimizing the random forest identification model based on a genetic algorithm to obtain the optimized random forest identification model includes:

[0019] Get the decision tree depth and number of decision trees of the random forest identification model;

[0020] Performing binary conversion on the decision tree depth and the number of decision trees respectively to obtain a converted decision tree depth and a converted number of decision trees;

[0021] Based on a preset number of the converted decision tree depths and the number of the converted decision trees, constructing an initialization population; the initialization population includes a plurality of individuals;

[0022] Using a preset fitness function, selecting the individual with the highest fitness from the initialization population as the parent of the next generation;

[0023] Cross the current parent generation with the next generation parent generation to obtain a crossed parent generation;

[0024] Performing mutation processing on the parent generation after the crossover to obtain new individuals;

[0025] After calculating the new individual using the preset fitness function to obtain the fitness of the new individual, returning to the step of selecting the individual with the highest fitness from the initialization population as the parent of the next generation using the preset fitness function until the maximum number of iterations is reached or the fitness converges;

[0026] Screening out the individual with the highest fitness from all the fitnesses, and performing decimal conversion on the converted decision tree depth and the converted decision tree number among the individual with the highest fitness, to obtain the target decision tree number and the target decision tree depth;

[0027] The decision tree depth of the random forest identification model is replaced with the target decision tree depth, and the number of decision trees is replaced with the target number of decision trees to obtain an optimized random forest identification model.

[0028] Optionally, the process of obtaining the sample principal component feature by dimensionality reduction based on multiple signal feature values ​​includes:

[0029] Obtaining current signal characteristic values, angular displacement signal characteristic values, and vibration signal characteristic values;

[0030] Constructing a sample characteristic matrix based on the current signal characteristic value, the angular displacement signal characteristic value, and the vibration signal characteristic value;

[0031] Normalizing the sample features in the sample feature matrix to obtain a standardized sample feature matrix;

[0032] Based on the standardized sample feature matrix, construct a covariance matrix;

[0033] Calculating the covariance matrix to obtain multiple eigenvalues;

[0034] Filtering standardized sample features corresponding to a preset number of eigenvalues ​​from the covariance matrix in descending order of the eigenvalues;

[0035] Based on the plurality of standardized sample features, construct a new vector matrix;

[0036] Generate a principal component feature matrix according to the new vector matrix and the standardized sample feature matrix;

[0037] Sorting the principal component features in the principal component feature matrix to obtain a principal component feature sequence;

[0038] The cumulative contribution value of each principal component feature in the principal component feature sequence is calculated, and the principal component features whose cumulative contribution values ​​are greater than 85% and whose eigenvalues ​​corresponding to the principal component features are greater than 1 are screened out from all the cumulative contribution values, and are marked as the sample principal component features.

[0039] Optionally, the obtaining of the current signal characteristic value, the angular displacement signal characteristic value, and the vibration signal characteristic value includes:

[0040] The current signal of the driving motor of the operating mechanism of the three-position switch of the GIS equipment is extracted to obtain the mean, standard deviation, mean value of the rate of change, root mean square and skewness of the current signal;

[0041] The angular displacement travel signal of the main shaft of the operating mechanism of the three-position switch of the GIS equipment is extracted to obtain the angle difference, slope mean and curvature mean of the angular displacement signal;

[0042] The vibration signals on the mainboard of the operating mechanism of the three-position switch of the GIS equipment and the vibration signals on the driving motor are extracted respectively to obtain the frequency domain eigenvalues ​​and time domain eigenvalues ​​of the vibration signals.

[0043] Determine the mean, standard deviation, change rate mean, root mean square and skewness of the current signal as current signal characteristic values;

[0044] Determining the angle difference, slope mean and curvature mean of the angular displacement signal as characteristic values ​​of the angular displacement signal;

[0045] The frequency domain eigenvalue and the time domain eigenvalue of the vibration signal are determined as vibration signal eigenvalues.

[0046] A diagnostic device for mechanical defects of a three-position switch of GIS equipment, comprising:

[0047] An acquisition unit, used for acquiring a characteristic value to be diagnosed;

[0048] A dimension reduction unit, used for performing feature dimension reduction on the feature value to be diagnosed to obtain a principal component feature;

[0049] A diagnostic unit is used to input the principal component features into a trained random forest identification model to obtain a diagnosis result of a mechanical defect; the diagnosis result of the mechanical defect indicates the type of the mechanical defect; the trained random forest identification model is trained based on multiple sample principal component features and an optimized random forest identification model; the optimized random forest identification model is obtained by optimizing the random forest identification model based on a genetic algorithm; the sample principal component features are obtained based on dimensionality reduction of multiple signal eigenvalues; the multiple signal eigenvalues ​​at least include current signal eigenvalues, angular displacement signal eigenvalues, and vibration signal eigenvalues.

[0050] Optionally, the diagnosis unit is specifically used for:

[0051] Obtain principal component features of multiple samples in the training set;

[0052] The principal component features of each sample in the training set are used as the input of the task loss function of the optimized random forest identification model, the diagnosis result of the mechanical defect is calculated, and the parameters of the task loss function are continuously adjusted until the accuracy of the diagnosis result of the mechanical defect is greater than the preset threshold, confirming that the optimized random forest identification model is successfully trained.

[0053] Optionally, also include:

[0054] A feature acquisition unit, used to acquire principal component features of multiple samples in a validation set;

[0055] An input unit, for inputting each of the sample principal component features in the validation set into the trained random forest recognition model to obtain a target diagnosis result of a mechanical defect;

[0056] A determination unit is used to determine that the trained random forest recognition model passes the verification when all the target diagnosis results are consistent with the preset diagnosis results.

[0057] Optionally, the diagnosis unit is specifically used for:

[0058] Get the decision tree depth and number of decision trees of the random forest identification model;

[0059] Performing binary conversion on the decision tree depth and the number of decision trees respectively to obtain a converted decision tree depth and a converted number of decision trees;

[0060] Based on a preset number of the converted decision tree depths and the number of the converted decision trees, constructing an initialization population; the initialization population includes a plurality of individuals;

[0061] Using a preset fitness function, selecting the individual with the highest fitness from the initialization population as the parent of the next generation;

[0062] Cross the current parent generation with the next generation parent generation to obtain a crossed parent generation;

[0063] Performing mutation processing on the parent generation after the crossover to obtain new individuals;

[0064] After calculating the new individual using the preset fitness function to obtain the fitness of the new individual, returning to the step of selecting the individual with the highest fitness from the initialization population as the parent of the next generation using the preset fitness function until the maximum number of iterations is reached or the fitness converges;

[0065] Screening out the individual with the highest fitness from all the fitnesses, and performing decimal conversion on the converted decision tree depth and the converted decision tree number among the individual with the highest fitness, to obtain the target decision tree number and the target decision tree depth;

[0066] The decision tree depth of the random forest identification model is replaced with the target decision tree depth, and the number of decision trees is replaced with the target number of decision trees to obtain an optimized random forest identification model.

[0067] The technical solution provided by the present application obtains the characteristic value to be diagnosed; performs feature dimension reduction on the characteristic value to be diagnosed to obtain the principal component feature; inputs the principal component feature into the trained random forest identification model to obtain the diagnosis result of the mechanical defect; the diagnosis result of the mechanical defect indicates the type of mechanical defect; the trained random forest identification model is obtained by training based on multiple sample principal component features and the optimized random forest identification model; the optimized random forest identification model is obtained by optimizing the random forest identification model based on the genetic algorithm; the sample principal component feature is obtained by dimension reduction based on multiple signal characteristic values; the multiple signal characteristic values ​​at least include current signal characteristic values, angular displacement signal characteristic values ​​and vibration signal characteristic values. The random forest model is trained by multi-dimensional features, so that the random forest model can effectively distinguish different types of mechanical defects. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0069] Figure 1 A flow chart of a method for diagnosing mechanical defects of a three-position switch of a GIS device provided in an embodiment of the present application;

[0070] Figure 2 A flowchart of a training method for a random forest recognition model provided in an embodiment of the present application;

[0071] Figure 3 A flowchart of a verification method of a random forest identification model provided in an embodiment of the present application;

[0072] Figure 4 A flowchart of a method for optimizing a random forest identification model provided in an embodiment of the present application;

[0073] Figure 5 A flowchart of a method for reducing the dimensionality of a signal characteristic value provided in an embodiment of the present application;

[0074] Figure 6 A flowchart of a method for obtaining a characteristic value provided in an embodiment of the present application;

[0075] Figure 7 A schematic diagram of a diagnosis process of a mechanical defect of a three-position switch of a GIS device provided in an embodiment of the present application;

[0076] Figure 8 A schematic diagram of a principal component contribution rate provided in an embodiment of the present application;

[0077] Fig. 9 A schematic diagram of the architecture of a device for diagnosing mechanical defects of a three-position switch of a GIS device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0078] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0079] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0080] like Figure 1 As shown, it is a flow chart of a method for diagnosing mechanical defects of a three-position switch of a GIS device provided in an embodiment of the present application, comprising the following steps:

[0081] S101: Obtain characteristic values ​​to be diagnosed.

[0082] Among them, the characteristic value to be diagnosed is obtained so that the characteristic value to be diagnosed can be subsequently input into the trained random forest identification model, and the mechanical defect type of the characteristic value to be diagnosed is determined by the trained random forest identification model.

[0083] S102: Perform feature dimensionality reduction on the feature values ​​to be diagnosed to obtain principal component features.

[0084] Optionally, the principal component analysis method can be used to perform feature dimension reduction on the characteristic values ​​to be diagnosed to obtain the principal component features.

[0085] It should be noted that the principal component analysis method is used to perform feature dimensionality reduction on the eigenvalues ​​to be diagnosed. Specifically, the eigenvalues ​​to be diagnosed are normalized to obtain standardized eigenvalues ​​to be diagnosed; a covariance matrix to be diagnosed is constructed based on the standardized eigenvalues ​​to be diagnosed; the covariance matrix to be diagnosed is calculated to obtain multiple eigenvalues ​​to be diagnosed; based on the multiple eigenvalues ​​to be diagnosed, the standardized eigenvalues ​​to be diagnosed corresponding to the eigenvalues ​​are determined; a new vector matrix is ​​constructed based on the standardized eigenvalues ​​to be diagnosed; and the principal component characteristics are determined based on the new vector matrix and the standardized eigenvalues ​​to be diagnosed.

[0086] S103: Input the principal component features into the trained random forest identification model to obtain the diagnosis results of mechanical defects.

[0087] Among them, the diagnosis result of the mechanical defect indicates the type of mechanical defect; the trained random forest identification model is trained based on multiple sample principal component features and the optimized random forest identification model; the optimized random forest identification model is optimized by the random forest identification model based on the genetic algorithm; the sample principal component features are obtained based on the dimensionality reduction of multiple signal eigenvalues; the multiple signal eigenvalues ​​at least include current signal eigenvalues, angular displacement signal eigenvalues ​​and vibration signal eigenvalues.

[0088] Optionally, the types of mechanical defects include but are not limited to: slight jamming of the three-position switch of the GIS equipment, severe jamming of the three-position switch of the GIS equipment, stalling of the motor of the three-position switch of the GIS equipment, and decoupling of the transmission of the three-position switch of the GIS equipment.

[0089] Optionally, in another embodiment of the present application, a specific implementation method of obtaining a trained random forest recognition model based on the principal component features of multiple samples and the optimized random forest recognition model training is as follows: Figure 2 As shown, the following steps are included:

[0090] S201: Obtain principal component features of multiple samples in a training set.

[0091] The sample principal component features include but are not limited to: "electrical current-mechanical" fusion feature vector.

[0092] S202: The principal component features of each sample in the training set are used as the input of the task loss function of the optimized random forest identification model, the diagnosis result of the mechanical defect is calculated, and the parameters of the task loss function are continuously adjusted until the accuracy of the diagnosis result of the mechanical defect is greater than the preset threshold, confirming that the training of the optimized random forest identification model is successful.

[0093] It is understandable that when the accuracy of the diagnosis results of mechanical defects is not greater than the preset threshold, it is necessary to adjust the parameters of the task loss function until the accuracy of the diagnosis results of mechanical defects is greater than the preset threshold, at which time the training of the optimized random forest identification model is completed.

[0094] Optionally, after step S202, in order to ensure the accuracy of the identification result of the optimized random forest identification model, it is also necessary to verify the trained optimized random forest identification model. Therefore, in another embodiment of the present application, a verification method for the random forest identification model is provided, such as Figure 3 As shown, the following steps are included:

[0095] S301: Obtain principal component features of multiple samples in a validation set.

[0096] The sample principal component features include but are not limited to: "electrical current-mechanical" fusion feature vector.

[0097] It should be noted that in order to ensure the consistency of feature dimensionality reduction, the validation set uses the same principal component direction as the training set so that feature dimensionality reduction can be performed in the same projection space.

[0098] S302: For each sample principal component feature in the validation set, the sample principal component feature is input into the trained random forest recognition model to obtain the target diagnosis result of the mechanical defect.

[0099] Among them, the target diagnosis result of mechanical defects indicates the diagnosis result of the principal component characteristics of the samples in the validation set.

[0100] It is understandable that since the random forest identification model has been trained and the parameters in the model have been determined using the training set, when the "current-mechanical" fusion feature vector of the validation set is input into the model, the model will classify the validation set into one of these mechanical defects.

[0101] S303: When all target diagnosis results are consistent with preset diagnosis results, it is determined that the trained random forest identification model has passed verification.

[0102] Optionally, if any target diagnosis result is inconsistent with the preset diagnosis result, it means that the trained random forest identification model has not passed the verification, and the parameters of the trained random forest identification model need to be adjusted so that the trained random forest identification model passes the verification.

[0103] Optionally, in another embodiment of the present application, a specific implementation of the optimized random forest identification model obtained by optimizing the random forest identification model based on a genetic algorithm is as follows: Figure 4 As shown, the following steps are included:

[0104] S401: Obtain the decision tree depth and number of decision trees of the random forest identification model.

[0105] Among them, the random forest model (RF) is an integrated learning algorithm that improves the overall model accuracy and robustness by combining the predictions of multiple decision trees. Multiple data subsets are obtained from the original data set using replacement sampling. A decision tree is built for each data subset, and a small number of features are randomly selected when splitting nodes, and then the best splitting feature is found from these random features. Finally, each decision tree independently makes a prediction and gives a classification result, and the category with the most occurrences is selected as the final prediction result through a voting mechanism.

[0106] The random forest model makes decisions by building a decision tree. The basic steps start from the root node, traverse all relevant feature attributes in turn, and point to the corresponding branch path based on these attribute values, and finally reach the leaf node. The category identified by the leaf node is the prediction result. When building a decision tree, a top-down growth strategy is usually adopted. Each iteration will be divided based on a certain feature attribute until it cannot be further subdivided. When the number of decision trees accumulates to a certain extent, although the diagnostic accuracy of the model tends to be stable and improved, it also leads to a decrease in the diagnostic speed. Therefore, it is necessary to find a number of decision trees that can guarantee both model performance and diagnostic efficiency. Deeper decision trees will consider more details of each feature and increase the complexity of the model. Models with higher complexity are more sensitive to training data, but have poor generalization ability on unseen data. Shallower trees will lead to underfitting and cannot adapt well to training data. Therefore, it is also particularly important to determine a more reasonable decision tree depth.

[0107] Specifically, the initial decision tree depth and number of decision trees of the random forest identification model are first obtained so that the initial decision tree depth and number of decision trees can be adjusted later.

[0108] S402: performing binary conversion on the decision tree depth and the number of decision trees respectively to obtain a converted decision tree depth and a converted number of decision trees.

[0109] It should be noted that the genetic algorithm cannot directly operate on the specific solution data in the solution space, so it is necessary to implement the encoding step first to map these solutions into a data structure in the form of a genotype string that can be processed in the genetic space. Therefore, it is necessary to convert the decision tree depth and the number of decision trees into binary, that is, convert the decimal decision tree depth and the number of decision trees into binary decision tree depth and the number of decision trees.

[0110] For example, the number 12 can be represented in the form 01100.

[0111] S403: constructing an initialization population based on a preset number of converted decision tree depths and a preset number of converted decision trees.

[0112] The initialization population includes multiple individuals.

[0113] It should be noted that initializing model parameters is the starting step of the training process, and the choice of population size has a significant impact on the efficiency of the process. If the initial population size is set too small, the number of individuals in each generation will be limited, which will easily lead to local optimal solutions and make it difficult to find the global optimal solution; on the contrary, if the population size is too large, the calculation burden of the fitness value of each generation will increase, thereby reducing the overall calculation efficiency. Therefore, it is crucial to select the initial population size appropriately.

[0114] Optionally, in the embodiment of the present application, the number of individuals in the initialization population can be set to 50.

[0115] S404: Using a preset fitness function, select the individual with the highest fitness from the initialization population as the parent of the next generation.

[0116] Optionally, according to the principle that the fitness value must be a non-negative number, the fitness function is usually constructed directly based on the objective function of the actual problem. Therefore, in the embodiment of the present application, the preset fitness function can be the average accuracy.

[0117] It can be understood that the fitness of each individual in the initialization population is calculated using the preset fitness function, and the individual with the highest fitness is selected from the initialization population as the parent of the next generation.

[0118] It should be noted that the individual with the highest fitness can be selected from the initialized population as the parent of the next generation through the roulette wheel selection method.

[0119] S405: Cross the current parent generation with the next generation parent generation to obtain a crossed parent generation.

[0120] It is understandable that in order to maintain the diversity of the population and promote the evolution of the population towards the optimal solution, the current parent generation is crossed with the next generation parent generation, that is, the parameters of the current parent generation and the next generation parent generation (such as decision tree depth, number of decision trees) are exchanged to obtain the crossed parent generation.

[0121] S406: Perform mutation processing on the parent generation after crossover to obtain new individuals.

[0122] It is understandable that in order to maintain the diversity of the population and prevent falling into the local optimal solution due to the crossover operation, the crossover process adopts a method of randomly changing the value of a parameter in the population individuals with a lower probability, that is, mutating the parent generation after the crossover to obtain new individuals.

[0123] It should be noted that by mutating the parent generation after crossover, the initialized population can maintain the breadth of exploration in the search space and effectively avoid early convergence.

[0124] S407: Calculate the new individual using the preset fitness function to obtain the fitness of the new individual.

[0125] The new individual is calculated using a preset fitness function to obtain the fitness of the new individual, and then the process returns to step S404 until the maximum number of iterations is reached or the fitness converges.

[0126] It can be understood that based on the adaptive evaluation mechanism of the population individuals, the genetic algorithm sequentially traverses and implements a cyclic iteration of parameter selection, crossover and mutation operations for each individual. This process will continue until the objective function reaches a near-optimal state or the preset maximum number of iterations is met, at which time the genetic algorithm execution terminates.

[0127] S408: Screening out the individual with the highest fitness from all fitnesses, and performing decimal conversion on the converted decision tree depth and the converted decision tree quantity in the individual with the highest fitness, to obtain the target decision tree quantity and the target decision tree depth.

[0128] It can be understood that the individuals with the highest fitness are selected from all fitnesses. At this time, the decision tree depth and the number of decision trees contained in the individuals are the optimal decision tree depth and the number of decision trees in the random forest identification model. The optimal converted decision tree depth and the number of converted decision trees are converted into decimals to obtain the target number of decision trees and the target decision tree depth, so that the random forest identification model can be optimized according to the target number of decision trees and the target decision tree depth.

[0129] S409: Replacing the decision tree depth of the random forest identification model with the target decision tree depth, and replacing the number of decision trees with the target number of decision trees, to obtain an optimized random forest identification model.

[0130] Among them, the decision tree depth and the number of decision trees of the random forest identification model were replaced, thereby improving the diagnostic accuracy of the random forest identification model.

[0131] Optionally, in another embodiment of the present application, a specific implementation method of obtaining the sample principal component features based on dimensionality reduction of multiple signal feature values ​​is as follows: Figure 5 As shown, the following steps are included:

[0132] S501: Acquire current signal characteristic values, angular displacement signal characteristic values, and vibration signal characteristic values.

[0133] Optionally, a global feature extraction method may be used to obtain the current signal characteristic value, the angular displacement signal characteristic value, and the vibration signal characteristic value.

[0134] Optionally, in another embodiment of the present application, the specific implementation of step S501 is as follows: Figure 6 As shown, the following steps are included:

[0135] S601: extract the current signal of the driving motor of the operating mechanism of the three-position switch of the GIS equipment to obtain the mean, standard deviation, mean value of the rate of change, root mean square and skewness of the current signal.

[0136] Among them, the mean value of the current signal can be used as an indicator to measure the stability of the signal. Under different operating conditions, the stability of the current signal varies. Therefore, its mean value will also be different. The specific expression of the mean value is shown in formula (1).

[0137]

[0138] In formula (1), F1 is the mean value of the current signal, x i is a time series signal, x is the signal amplitude corresponding to each sampling point, i=1,2,…,N; N is the total number of sampling points.

[0139] Optionally, the standard deviation of the current signal can reflect the degree of discreteness of the current signal data distribution. In the analysis of the current signal of the drive motor, it can quantify the severity of the waveform fluctuation. The specific expression of the standard deviation is shown in formula (2).

[0140]

[0141] In formula (2), F2 is the standard deviation of the current signal and μ is the signal mean.

[0142] The mean value of the rate of change of the current signal is used to characterize the change of the load borne by the drive motor. The specific expression of the mean value of the rate of change is shown in formula (3).

[0143]

[0144] In formula (3), F3 is the mean value of the rate of change of the current signal, t i is a timing signal, and t is the signal time corresponding to each sampling point.

[0145] Optionally, the root mean square of the current signal is similar to the mean value and can also reflect the stability of the current signal, but the root mean square can eliminate the influence of negative components and is more suitable for reflecting the signal energy size. The specific expression of the root mean square is shown in formula (4).

[0146]

[0147] In formula (4), F4 is the root mean square of the current signal.

[0148] Among them, skewness is a measure of the direction and degree of skewness of the statistical data distribution, and is a numerical characteristic of the degree of asymmetry of the statistical data distribution. By definition, skewness is the third-order standardized moment of the sample. The specific manifestation of skewness is shown in formula (5).

[0149]

[0150] In formula (5), σ is the signal standard deviation.

[0151] S602: Extract the angular displacement travel signal of the main shaft of the operating mechanism of the three-position switch of the GIS equipment to obtain the angle difference, slope mean and curvature mean of the angular displacement signal.

[0152] The angle difference of the angular displacement signal represents the difference between the end angle and the start angle of the main shaft of the operating mechanism mainboard. The angle difference is proportional to the travel distance of the moving contact in the air chamber of the three-position switch. The specific expression of the angle difference is shown in formula (6).

[0153] F6=min(x i )-max(x i )(6)

[0154] In formula (6), F6 is the angle difference of the angular displacement signal.

[0155] Optionally, the slope mean of the angular displacement signal represents the rotation rate of the main shaft of the operating mechanism mainboard, which is proportional to the travel rate of the moving contact in the three-position switch air chamber. The specific expression of the slope mean is shown in formula (7).

[0156]

[0157] In formula (7), F7 is the slope mean of the angular displacement signal.

[0158] Among them, the mean curvature of the angular displacement signal represents the smoothness of the rotation stroke curve of the main shaft of the operating mechanism mainboard. This characteristic parameter can better reflect the jamming of the mechanism during the switching operation. The specific expression of the mean curvature is shown in formula (8).

[0159]

[0160] In formula (8), x,,(t i ) is the second-order derivative of x(ti), x,(t i ) is the first-order derivative of x(ti).

[0161] S603: Extract vibration signals on the mainboard of the operating mechanism of the three-position switch of the GIS equipment and the vibration signals on the driving motor respectively to obtain frequency domain eigenvalues ​​and time domain eigenvalues ​​of the vibration signals.

[0162] Among them, the vibration signal includes the vibration signal on the main board of the operating mechanism and the vibration signal on the driving motor. The frequency domain characteristics of the vibration signal include: the maximum value, absolute mean, root mean square, kurtosis, and skewness of the vibration signal. The time domain characteristic values ​​of the vibration signal include: the center of gravity frequency and peak factor of the vibration signal.

[0163] Optionally, the maximum value of the vibration signal is related to the state of the opening and closing action of the three-position switch. When the three-position switch is in an abnormal state, its opening and closing state will change, and the impact intensity of each transmission component inside the operating mechanism will change. The specific expression form of the maximum value is shown in formula (9).

[0164] F9=max(x i |)(9)

[0165] In formula (9), F9 is the maximum value of the vibration signal.

[0166] Optionally, the absolute mean of the vibration signal can be used as an indicator to measure the stability of the vibration signal. Under different operating conditions, the stability of the vibration signal is different, and accordingly, the absolute mean of the vibration signal will also change. The specific expression of the absolute mean is shown in formula (10).

[0167]

[0168] In formula (10), the absolute mean value of the vibration signal.

[0169] Optionally, the root mean square of the vibration signal is similar to the absolute mean. The root mean square of the vibration signal can also characterize the stability characteristics of the signal, but its advantage is that it can strengthen the effective signal component and weaken the error signal. It is particularly suitable for showing the signal energy size. The specific form of the root mean square is shown in formula (11).

[0170]

[0171] In formula (11), F 11 is the root mean square of the vibration signal.

[0172] Optionally, the kurtosis of the vibration signal is a fourth-order statistical indicator that reveals the distribution characteristics of the vibration signal and is mainly used to describe the peak degree of the vibration waveform. The specific manifestation of the kurtosis is shown in formula (12).

[0173]

[0174] In formula (12), F 12 is the kurtosis of the vibration signal.

[0175] Optionally, the skewness of the vibration signal is a measure of the direction and degree of skewness of the statistical vibration signal data distribution, and is a digital feature of the degree of asymmetry of the statistical data distribution. The upper skewness is defined as the third-order standardized moment of the sample. The specific manifestation of the skewness is shown in formula (13).

[0176]

[0177] In formula (13), F 13 is the skewness of the vibration signal.

[0178] Optionally, the centroid frequency of the vibration signal represents the centroid or balance point of the area distribution under the power spectrum curve of the vibration signal. It represents the frequency region where the energy of the vibration signal is most concentrated. The specific expression of the centroid frequency is shown in formula (14).

[0179]

[0180] In formula (14), x f (i) and f(i) are frequency sequence signals, xf and f represent the amplitude and frequency value of the frequency sequence signal corresponding to each sampling point, respectively. 14 is the center frequency of the vibration signal.

[0181] Optionally, the peak factor of the vibration signal is a parameter that measures the relative size between the peak amplitude of the signal waveform and its effective value. The larger the peak factor, the higher the peak value of the vibration signal waveform relative to its average energy, the sharper the waveform, and the more high-frequency components it contains. The specific expression of the peak factor is shown in formula (15).

[0182]

[0183] In formula (15), F 15 is the peak factor of the vibration signal.

[0184] S604: Determine the mean, standard deviation, change rate mean, root mean square, and skewness of the current signal as characteristic values ​​of the current signal.

[0185] S605: Determine the angle difference, slope mean, and curvature mean of the angular displacement signal as characteristic values ​​of the angular displacement signal.

[0186] S606: Determine the frequency domain eigenvalue and the time domain eigenvalue of the vibration signal as the vibration signal eigenvalue.

[0187] S502: constructing a sample feature matrix based on the current signal eigenvalues, the angular displacement signal eigenvalues, and the vibration signal eigenvalues.

[0188] It can be understood that a sample characteristic matrix is ​​constructed based on the current signal eigenvalues, the angular displacement signal eigenvalues ​​and the vibration signal eigenvalues. Specifically, the sample characteristic matrix (i.e., the "current-mechanical" eigenvector) is constructed in the order of the current signal eigenvalues, the angular displacement signal eigenvalues ​​and the vibration signal eigenvalues.

[0189] Optionally, the construction size of the sample feature matrix includes but is not limited to: [1×22].

[0190] It should be noted that, in order to improve the accuracy of defect identification, the dimension of the sample feature matrix is ​​reduced by using the principal component analysis method, that is, the dimension of the sample feature matrix is ​​reduced by steps S503 to S510.

[0191] S503: Normalize the sample features in the sample feature matrix to obtain a standardized sample feature matrix.

[0192] The specific implementation process of normalizing the sample features in the sample feature matrix is: using the normalization formula to normalize the sample features in the sample feature matrix. The specific expression of the normalization formula is shown in formula (16).

[0193]

[0194] In formula (16), is the standardized sample feature, m is the number of sample features, n is the number of feature parameters extracted from the sample features, and the dimension of the sample feature matrix is ​​n×m. is the mean of the eigenvalues ​​of each dimension, S j is the variance of the eigenvalues ​​of each dimension, h ij is the jth feature of the i-th sample.

[0195] S504: Construct a covariance matrix based on the standardized sample feature matrix.

[0196] The specific expression of the covariance matrix is ​​shown in formula (17) and formula (18).

[0197]

[0198] In formula (17) and formula (18), Σ is the covariance matrix, Features and Features The covariance of is the normalized feature x i, is the normalized feature x j , Features The mean of Features The mean of .

[0199] S505: Calculate the covariance matrix to obtain multiple eigenvalues.

[0200] Optionally, the covariance matrix may be calculated by a method for calculating matrix eigenvalues ​​and eigenvectors in linear algebra to obtain multiple eigenvalues.

[0201] S506: Filtering out standardized sample features corresponding to a preset number of eigenvalues ​​from the covariance matrix in descending order of eigenvalues.

[0202] Optionally, the preset number can be set according to actual conditions and is not specifically limited here.

[0203] It should be noted that, in the order of eigenvalues ​​from large to small, standardized sample features corresponding to a preset number of eigenvalues ​​are screened out from the covariance matrix. Specifically, multiple eigenvalues ​​are sorted in the order of eigenvalues ​​from large to small to obtain an eigenvalue sequence; the eigenvalue sequence includes the serial number of each eigenvalue; in the order of serial numbers from small to large, a preset number of eigenvalues ​​are screened out; and standardized sample features corresponding to a preset number of eigenvalues ​​are screened out from the covariance matrix.

[0204] S507: Construct a new vector matrix based on multiple standardized sample features.

[0205] Wherein, a new vector matrix is ​​constructed based on a plurality of standardized sample features, that is, a new vector matrix is ​​constructed based on standardized sample features corresponding to a preset number of eigenvalues.

[0206] S508: Generate a principal component feature matrix according to the new vector matrix and the standardized sample feature matrix.

[0207] Among them, according to the new vector matrix and the standardized sample feature matrix, the principal component feature matrix is ​​generated. Specifically, the standardized sample feature matrix is ​​mapped to the space constructed by the principal component feature vector, and the principal component feature matrix is ​​calculated. Specifically, the standardized sample feature matrix is ​​mapped to the space constructed by the principal component feature vector, and the specific expression of the principal component feature matrix is ​​calculated, as shown in formula (19).

[0208] Z=U T H * (19)

[0209] In formula (19), Z is the principal component feature matrix, U Tis the transpose of the new vector matrix, H * is the standardized sample feature matrix.

[0210] S509: Sort the principal component features in the principal component feature matrix to obtain a principal component feature sequence.

[0211] Optionally, the principal component features in the principal component feature matrix are sorted in descending order to obtain a principal component feature sequence sorted in descending order.

[0212] S510: Calculate the cumulative contribution value of each principal component feature in the principal component feature sequence, and select principal component features whose cumulative contribution values ​​are greater than 85% and whose corresponding eigenvalues ​​are greater than 1 from all the cumulative contribution values, and mark them as sample principal component features.

[0213] Among them, the cumulative contribution value of each principal component feature in the principal component feature sequence is calculated. Specifically, for each principal component feature, the ratio of the principal component feature to the sum of the principal component features is calculated to obtain the cumulative contribution value of the principal component feature.

[0214] It should be noted that the sample principal component features are split into the sample principal component features of the training set and the sample principal component features of the validation set according to a certain ratio, for example, the split is performed in a ratio of 7:3.

[0215] For a better Figure 1 For an explanation of the contents shown, see Figure 7 The schematic diagram of the diagnosis process of mechanical defects of a three-position switch of a GIS device is shown. First, the current signal eigenvalue, angular displacement signal eigenvalue and vibration signal eigenvalue are obtained; based on the current signal eigenvalue, angular displacement signal eigenvalue and vibration signal eigenvalue, a sample feature matrix (i.e., "current-mechanical" eigenvector) is constructed, and the sample feature matrix is ​​reduced in dimension by the principal component analysis method to obtain the sample principal component features (i.e., "current-mechanical" fusion eigenvector), and the sample principal component features are split into sample principal component features of a training set and sample principal component features of a validation set; a random forest model is optimized by a genetic algorithm to obtain an optimized random forest model; the optimized random forest model is trained by using the sample principal component features of the training set to obtain a trained random forest model; the trained random forest model is verified by using the sample principal component features of the validation set to obtain a verification result.

[0216] It should be noted that in order to verify the recognition effects and differences of different types of input feature parameters on the improved random forest diagnosis model, the following five types of input features were used to enter the model for calculation: (1) single current signal feature parameters, (2) single angular displacement signal feature parameters, (3) single vibration signal feature parameters, (4) "current-mechanical" signal joint feature parameters, (5) "current-mechanical" feature parameters after PCA dimensionality reduction. The same training set and validation set division ratios were used and 30 groups of experiments were repeated to obtain the recognition results.

[0217] In order to verify the effectiveness of the trained random forest model in the embodiment of the present application, the following is an explanation in combination with a specific application scenario, including processes A1 to A3.

[0218] A1: The current signals, angular displacement signals and vibration signals collected under five operating conditions of the three-position switch of GIS equipment, namely normal working condition, slight jamming, severe jamming, motor stalling and transmission decoupling, are selected. The “current-mechanical” feature vector results under different defect types obtained by the global feature extraction method are shown in Tables 1, 2 and 3.

[0219] Table 1

[0220]

[0221] Table 2

[0222]

[0223] Table 3

[0224]

[0225] It can be seen from Table 1, Table 2 and Table 3 that the mean and root mean square characteristics of the motor stall condition are significantly greater than the normal condition, and the mean and root mean square characteristics of the transmission decoupling condition are significantly less than the normal condition; the skewness characteristics of the motor stall condition are less than the normal condition, and the skewness characteristics of the transmission decoupling condition are significantly greater than the normal condition. The comparison and analysis of the two characteristics of the angle difference and the slope mean of the angular displacement signal show that the degree of jamming of the operating mechanism is inversely proportional to these two characteristic values, and the difference between these two characteristics is most obvious in the case of transmission decoupling; the comparison and analysis of the curvature mean characteristics of the angular displacement signal show that the angular displacement curve is relatively stable in the case of transmission decoupling. The maximum value of the vibration signal will increase first and then decrease with the increase of the jamming degree of the mechanism, and the maximum value of the vibration signal in the case of transmission decoupling is slightly smaller than the other cases; the absolute mean and root mean square of the vibration signal are slightly larger than the other cases in the case of motor stall condition, and slightly smaller than the other cases in the case of transmission decoupling.

[0226] A2: 500 sets of sample data were collected for five different operating conditions, and then integrated to form an overall characteristic parameter matrix of size [500×22]. Principal component analysis was performed on the constructed [500×22] scale characteristic parameter matrix to perform feature dimension reduction, and the eigenvalues ​​were arranged in descending order, and the principal components with cumulative contribution rates exceeding 85% and eigenvalues ​​greater than 1 were selected. The results show that the number of principal components required is 5. Its eigenvalues ​​= [10.126, 3.433, 3.008, 1.916, 1.176], and the contribution rates of each principal component from high to low are 46.026%, 15.605%, 13.674%, 8.71%, and 5.345%, respectively. The cumulative contribution rate is as follows Figure 8 After completing the principal component analysis and successfully reducing the dimension, a principal component feature matrix with a specification of [500×5] is obtained.

[0227] A3: Five different types of input feature vectors are used to input the model to obtain the diagnosis results. The average recognition accuracy and standard deviation of the recognition accuracy of the identification model based on single signal, "current-mechanical" signal and "current-mechanical" PCA reduced dimension feature vector are shown in Table 4.

[0228] Table 4

[0229]

[0230] It can be seen from Table 4 that the accuracy of the model using a single feature parameter input can only maintain about 80%. The random forest diagnosis model constructed by using PCA to fuse the feature vector of the signal after the "current-mechanical" dimension reduction can achieve an average recognition accuracy of 90.93% and 91.58% for different working conditions of disconnectors and grounding switches.

[0231] In summary, by constructing a mechanical-current feature vector and then reducing the dimension of the mechanical-current feature vector, we get the "current-mechanical" fusion feature vector. The optimized random forest model is trained and verified by the "current-mechanical" fusion feature vector. The random forest model is trained by multi-dimensional features, so that the random forest model can effectively distinguish different types of mechanical defects.

[0232] like Fig. 9 As shown, it is a schematic diagram of the architecture of a diagnostic device for mechanical defects of a three-position switch of a GIS device provided in an embodiment of the present application. The diagnostic device includes: an acquisition unit 100, a dimensionality reduction unit 200, and a diagnostic unit 300.

[0233] The acquisition unit 100 is used to acquire the characteristic value to be diagnosed.

[0234] The dimension reduction unit 200 is used to perform dimension reduction on the characteristic values ​​to be diagnosed to obtain principal component features.

[0235] The diagnosis unit 300 is used to input the principal component features into the trained random forest identification model to obtain the diagnosis result of the mechanical defect; the diagnosis result of the mechanical defect indicates the type of the mechanical defect; the trained random forest identification model is trained based on multiple sample principal component features and the optimized random forest identification model; the optimized random forest identification model is obtained by optimizing the random forest identification model based on a genetic algorithm; the sample principal component features are obtained based on the dimensionality reduction of multiple signal eigenvalues; the multiple signal eigenvalues ​​at least include current signal eigenvalues, angular displacement signal eigenvalues ​​and vibration signal eigenvalues.

[0236] The diagnosis unit 300 is specifically used to: obtain principal component features of multiple samples in a training set; use the principal component features of each sample in the training set as the input of the task loss function of the optimized random forest identification model, calculate the diagnosis result of the mechanical defect, and continuously adjust the various parameters of the task loss function until the accuracy of the diagnosis result of the mechanical defect is greater than a preset threshold, confirming that the optimized random forest identification model is successfully trained.

[0237] The diagnostic unit 300 is specifically used to: obtain the decision tree depth and the number of decision trees of the random forest identification model; perform binary conversion on the decision tree depth and the number of decision trees respectively to obtain the converted decision tree depth and the converted decision tree number; construct an initialization population based on a preset number of converted decision tree depths and the converted decision tree number; the initialization population includes multiple individuals; using a preset fitness function, select the individual with the highest fitness from the initialization population as the next generation parent; cross the current parent with the next generation parent to obtain the crossover parent; perform mutation processing on the crossover parent to obtain a new individual; and use the preset fitness function to The new individual is calculated, and after the fitness of the new individual is obtained, it returns to execute the step of using the preset fitness function to select the individual with the highest fitness from the initialized population as the parent of the next generation until the maximum number of iterations is reached or the fitness converges; the individual with the highest fitness is screened out from all fitnesses, and the converted decision tree depth and the converted number of decision trees in the individual with the highest fitness are converted into decimals to obtain the target number of decision trees and the target decision tree depth; the decision tree depth of the random forest identification model is replaced by the target decision tree depth, and the number of decision trees is replaced by the target number of decision trees to obtain the optimized random forest identification model.

[0238] The diagnostic unit 300 is specifically used to: obtain the current signal eigenvalues, the angular displacement signal eigenvalues ​​and the vibration signal eigenvalues; construct a sample feature matrix based on the current signal eigenvalues, the angular displacement signal eigenvalues ​​and the vibration signal eigenvalues; normalize the sample features in the sample feature matrix to obtain a standardized sample feature matrix; construct a covariance matrix based on the standardized sample feature matrix; calculate the covariance matrix to obtain multiple eigenvalues; filter out standardized sample features corresponding to a preset number of eigenvalues ​​from the covariance matrix in descending order of the eigenvalues; construct a new vector matrix based on multiple standardized sample features; generate a principal component feature matrix based on the new vector matrix and the standardized sample feature matrix; sort the principal component features in the principal component feature matrix to obtain a principal component feature sequence; calculate the cumulative contribution value of each principal component feature in the principal component feature sequence, and filter out the principal component features whose cumulative contribution values ​​are greater than 85% and whose eigenvalues ​​corresponding to the principal component features are greater than 1 from all the cumulative contribution values, and mark them as sample principal component features.

[0239] The diagnostic unit 300 is specifically used for: extracting the current signal of the driving motor of the operating mechanism of the three-position switch of the GIS equipment to obtain the mean, standard deviation, mean value of the rate of change, root mean square and skewness of the current signal; extracting the angular displacement stroke signal of the main shaft of the main board of the operating mechanism of the three-position switch of the GIS equipment to obtain the angle difference, slope mean and curvature mean of the angular displacement signal; extracting the vibration signal on the main board of the operating mechanism of the three-position switch of the GIS equipment and the vibration signal on the driving motor to obtain the frequency domain eigenvalues ​​and time domain eigenvalues ​​of the vibration signal; determining the mean, standard deviation, mean value of the rate of change, root mean square and skewness of the current signal as the current signal eigenvalues; determining the angle difference, slope mean and curvature mean of the angular displacement signal as the angular displacement signal eigenvalues; determining the frequency domain eigenvalues ​​and time domain eigenvalues ​​of the vibration signal as the vibration signal eigenvalues.

[0240] In summary, by constructing a mechanical-current feature vector and then reducing the dimension of the mechanical-current feature vector, we get the "current-mechanical" fusion feature vector. The optimized random forest model is trained and verified by the "current-mechanical" fusion feature vector. The random forest model is trained by multi-dimensional features, so that the random forest model can effectively distinguish different types of mechanical defects.

[0241] Preferably, combined Figure 8 As shown in the figure, the diagnostic device also includes: a feature acquisition unit, an input unit and a determination unit.

[0242] The feature acquisition unit is used to obtain the principal component features of multiple samples in the validation set.

[0243] The input unit is used to input the principal component features of each sample in the verification set into the trained random forest recognition model to obtain the target diagnosis result of the mechanical defect.

[0244] The determination unit is used to determine whether the trained random forest recognition model has passed the verification when all target diagnosis results are consistent with the preset diagnosis results.

[0245] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0246] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0247] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for diagnosing mechanical defects of a three-position switch of a GIS device, characterized in that: include: Obtain the characteristic value to be diagnosed; Performing feature dimension reduction on the feature value to be diagnosed to obtain principal component features; The principal component features are input into a trained random forest identification model to obtain a diagnosis result of a mechanical defect; the diagnosis result of the mechanical defect indicates the type of mechanical defect; the trained random forest identification model is trained based on multiple sample principal component features and an optimized random forest identification model; the optimized random forest identification model is obtained by optimizing the random forest identification model based on a genetic algorithm; the sample principal component features are obtained based on dimensionality reduction of multiple signal eigenvalues; the multiple signal eigenvalues ​​at least include current signal eigenvalues, angular displacement signal eigenvalues ​​and vibration signal eigenvalues.

2. The method according to claim 1, characterized in that The process of obtaining the trained random forest recognition model based on the principal component features of multiple samples and the optimized random forest recognition model training includes: Obtain principal component features of multiple samples in the training set; The principal component features of each sample in the training set are used as the input of the task loss function of the optimized random forest identification model, the diagnosis result of the mechanical defect is calculated, and the parameters of the task loss function are continuously adjusted until the accuracy of the diagnosis result of the mechanical defect is greater than the preset threshold, confirming that the optimized random forest identification model is successfully trained.

3. The method according to claim 2, characterized in that Also includes: Obtain principal component features of multiple samples in the validation set; For each principal component feature of the sample in the validation set, inputting the principal component feature of the sample into the trained random forest recognition model to obtain a target diagnosis result of the mechanical defect; When all of the target diagnosis results are consistent with the preset diagnosis results, it is determined that the trained random forest recognition model has passed the verification.

4. The method according to claim 1, characterized in that: The process of optimizing the random forest identification model based on a genetic algorithm to obtain the optimized random forest identification model includes: Get the decision tree depth and number of decision trees of the random forest identification model; Performing binary conversion on the decision tree depth and the number of decision trees respectively to obtain a converted decision tree depth and a converted number of decision trees; Based on a preset number of the converted decision tree depths and the number of the converted decision trees, constructing an initialization population; the initialization population includes a plurality of individuals; Using a preset fitness function, selecting the individual with the highest fitness from the initialization population as the parent of the next generation; Cross the current parent generation with the next generation parent generation to obtain a crossed parent generation; Performing mutation processing on the parent generation after the crossover to obtain new individuals; After calculating the new individual using the preset fitness function to obtain the fitness of the new individual, returning to the step of selecting the individual with the highest fitness from the initialization population as the parent of the next generation using the preset fitness function until the maximum number of iterations is reached or the fitness converges; Screening out the individual with the highest fitness from all the fitnesses, and performing decimal conversion on the converted decision tree depth and the converted decision tree number among the individual with the highest fitness, to obtain the target decision tree number and the target decision tree depth; The decision tree depth of the random forest identification model is replaced with the target decision tree depth, and the number of decision trees is replaced with the target number of decision trees to obtain an optimized random forest identification model.

5. The method according to claim 1, characterized in that The process of obtaining the sample principal component features based on dimensionality reduction of multiple signal feature values ​​includes: Obtaining current signal characteristic values, angular displacement signal characteristic values, and vibration signal characteristic values; Constructing a sample characteristic matrix based on the current signal characteristic value, the angular displacement signal characteristic value, and the vibration signal characteristic value; Normalizing the sample features in the sample feature matrix to obtain a standardized sample feature matrix; Based on the standardized sample feature matrix, construct a covariance matrix; Calculating the covariance matrix to obtain multiple eigenvalues; Filtering standardized sample features corresponding to a preset number of eigenvalues ​​from the covariance matrix in descending order of the eigenvalues; Based on the plurality of standardized sample features, construct a new vector matrix; Generate a principal component feature matrix according to the new vector matrix and the standardized sample feature matrix; Sorting the principal component features in the principal component feature matrix to obtain a principal component feature sequence; The cumulative contribution value of each principal component feature in the principal component feature sequence is calculated, and the principal component features whose cumulative contribution values ​​are greater than 85% and whose eigenvalues ​​corresponding to the principal component features are greater than 1 are screened out from all the cumulative contribution values, and are marked as the sample principal component features.

6. The method according to claim 5, characterized in that The obtaining of the current signal characteristic value, the angular displacement signal characteristic value and the vibration signal characteristic value comprises: The current signal of the driving motor of the operating mechanism of the three-position switch of the GIS equipment is extracted to obtain the mean, standard deviation, mean value of the rate of change, root mean square and skewness of the current signal; The angular displacement travel signal of the main shaft of the operating mechanism of the three-position switch of the GIS equipment is extracted to obtain the angle difference, slope mean and curvature mean of the angular displacement signal; The vibration signals on the mainboard of the operating mechanism of the three-position switch of the GIS equipment and the vibration signals on the driving motor are extracted respectively to obtain the frequency domain eigenvalues ​​and time domain eigenvalues ​​of the vibration signals. Determine the mean, standard deviation, change rate mean, root mean square and skewness of the current signal as current signal characteristic values; Determining the angle difference, slope mean and curvature mean of the angular displacement signal as characteristic values ​​of the angular displacement signal; The frequency domain eigenvalue and the time domain eigenvalue of the vibration signal are determined as vibration signal eigenvalues.

7. A diagnostic device for mechanical defects of a three-position switch of a GIS device, characterized in that: include: An acquisition unit, used for acquiring a characteristic value to be diagnosed; A dimension reduction unit, used for performing feature dimension reduction on the feature value to be diagnosed to obtain a principal component feature; A diagnostic unit is used to input the principal component features into a trained random forest identification model to obtain a diagnosis result of a mechanical defect; the diagnosis result of the mechanical defect indicates the type of the mechanical defect; the trained random forest identification model is trained based on multiple sample principal component features and an optimized random forest identification model; the optimized random forest identification model is obtained by optimizing the random forest identification model based on a genetic algorithm; the sample principal component features are obtained based on dimensionality reduction of multiple signal eigenvalues; the multiple signal eigenvalues ​​at least include current signal eigenvalues, angular displacement signal eigenvalues, and vibration signal eigenvalues.

8. The device according to claim 7, characterized in that The diagnostic unit is specifically used for: Obtain principal component features of multiple samples in the training set; The principal component features of each sample in the training set are used as the input of the task loss function of the optimized random forest identification model, the diagnosis result of the mechanical defect is calculated, and the parameters of the task loss function are continuously adjusted until the accuracy of the diagnosis result of the mechanical defect is greater than the preset threshold, confirming that the optimized random forest identification model is successfully trained.

9. The device according to claim 8, characterized in that Also includes: A feature acquisition unit, used to acquire principal component features of multiple samples in a validation set; An input unit, for inputting each of the sample principal component features in the validation set into the trained random forest recognition model to obtain a target diagnosis result of a mechanical defect; A determination unit is used to determine that the trained random forest recognition model passes the verification when all the target diagnosis results are consistent with the preset diagnosis results.

10. The device according to claim 7, characterized in that The diagnostic unit is specifically used for: Get the decision tree depth and number of decision trees of the random forest identification model; Performing binary conversion on the decision tree depth and the number of decision trees respectively to obtain a converted decision tree depth and a converted number of decision trees; Based on a preset number of the converted decision tree depths and the number of the converted decision trees, constructing an initialization population; the initialization population includes a plurality of individuals; Using a preset fitness function, selecting the individual with the highest fitness from the initialization population as the parent of the next generation; Cross the current parent generation with the next generation parent generation to obtain a crossed parent generation; Performing mutation processing on the parent generation after the crossover to obtain new individuals; After calculating the new individual using the preset fitness function to obtain the fitness of the new individual, returning to the step of selecting the individual with the highest fitness from the initialization population as the parent of the next generation using the preset fitness function until the maximum number of iterations is reached or the fitness converges; Screening out the individual with the highest fitness from all the fitnesses, and performing decimal conversion on the converted decision tree depth and the converted decision tree number among the individual with the highest fitness, to obtain the target decision tree number and the target decision tree depth; The decision tree depth of the random forest identification model is replaced with the target decision tree depth, and the number of decision trees is replaced with the target number of decision trees to obtain an optimized random forest identification model.

Citation Information

Cited By

  • Few-sample equipment state prediction method based on transfer learning

    CN120448978A