Power System Transmission Line Fault Diagnosis Method Based on the Combination of New Variational Mode Decomposition and K-Nearest Neighbor Algorithm

Abstract

The present invention discloses a fault diagnosis method based on the combination of new variational mode decomposition and K-nearest neighbor algorithm, comprising the following steps: S1, fault simulation data acquisition; S2, fault data preprocessing; S3, connecting a trained new variational mode decomposition model module, a constructed new variational mode decomposition model module and an improved K-nearest neighbor algorithm recognition and classification module, which is used to construct a new variational mode decomposition model for the collected signal data; S4, connecting the improved K-nearest neighbor algorithm recognition and classification module with the trained new variational mode decomposition model module, which is used to identify the diagnosis result. The present invention can quickly and accurately detect the faults of the power system transmission lines under various complex environments, ensure the normal operation of the power system transmission lines, improve the scientific detection system of the power system industry, avoid the occurrence of major accidents, improve the safety of the transmission lines, minimize the fault-tolerant operation, and minimize the performance degradation and avoid dangerous situations, which has important significance.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis, and particularly relates to a fault diagnosis method for transmission lines in a power system based on the combination of a new variational mode decomposition and the K-nearest neighbor algorithm. Background Technique

[0002] In the power grid, power plants, substations, electrical equipment, transmission lines, etc. are all important components of the power system. The normal operation of these components is of great significance to the stable operation of the power grid. As a key link connecting the power grid, the transmission line shoulders the important function of transmitting and distributing the electric energy of the power grid and is also known as the "lifeline" of the power grid. At present, the high-voltage transmission lines in the power grid are the backbone of the entire transmission grid, and the high-voltage and extra-high-voltage transmission lines account for a large proportion in the entire power system.

[0003] The main purpose of the fault diagnosis of the transmission line in the power system is to quickly identify the fault components or switches, provide a reliable basis for the dispatcher, and quickly realize the fault recovery of the transmission line. Affected by factors such as weather and human beings, the faults of the transmission line are inevitable, and other faults may be triggered at the same time. Therefore, the fast and accurate fault diagnosis through artificial intelligence algorithms is the key to ensuring the safe and stable operation of the transmission line. For this reason, we propose a fault diagnosis method for transmission lines in a power system based on the combination of a new variational mode decomposition and the K-nearest neighbor algorithm to solve the problems mentioned in the above background technique. Summary of the Invention

[0004] The purpose of the present invention is to provide a fault diagnosis method for transmission lines in a power system based on the combination of a new variational mode decomposition and the K-nearest neighbor algorithm to solve the problems raised in the above background technique.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A fault diagnosis method for transmission lines in a power system based on the combination of a new variational mode decomposition and the K-nearest neighbor algorithm, including a fault simulation data acquisition module, a module for constructing a new variational mode decomposition model, a module for training the new variational mode decomposition model, and an improved K-nearest neighbor algorithm recognition and classification module, including the following steps:

[0006] S1. Fault simulation data acquisition: The fault simulation data acquisition module is connected to the module for constructing a new variational mode decomposition model for collecting the original signal dataset;

[0007] S2. Fault data preprocessing: The module for constructing a new variational mode decomposition model is respectively connected to the fault simulation data acquisition module and the module for training the new variational mode decomposition model for constructing a new variational mode decomposition model with the collected data and transmitting the new variational mode decomposition to the module for training the new variational mode decomposition model;

[0008] S3. The module for training a new variational mode decomposition model is connected to the module for constructing a new variational mode decomposition model and the improved K-nearest neighbor algorithm recognition and classification module, and is used to construct a new variational mode decomposition model from the collected signal data;

[0009] Training the new variational mode decomposition model includes variational mode decomposition processing optimized by a genetic algorithm, wavelet decomposition processing, principal component analysis for dimensionality reduction processing, training sample data sets, test sample data sets, constructing a model, and obtaining a model;

[0010] S4. The improved K-nearest neighbor algorithm recognition and classification module is connected to the module for training a new variational mode decomposition model and is used to identify the diagnostic results.

[0011] In the construction of the new variational mode decomposition model, the signal data set is processed by variational mode decomposition optimized by a genetic algorithm to obtain intrinsic mode functions, and then the intrinsic mode function signals are processed by wavelet decomposition, and finally dimensionality reduction processing is performed by principal component analysis to obtain the new variational mode decomposition model.

[0012] The acquisition of the intrinsic mode function signals by variational mode decomposition optimized by a genetic algorithm can decompose the signals into intrinsic mode function signals; the wavelet decomposition processing is to perform four-layer wavelet decomposition on the intrinsic mode function signals to obtain several wavelet components to form new intrinsic mode function components; the principal component analysis processing is to perform dimensionality reduction processing on the obtained new intrinsic mode function components.

[0013] The acquisition of the intrinsic mode function signals by variational mode decomposition optimized by a genetic algorithm can decompose the signals into intrinsic mode function signals; the wavelet decomposition processing is to perform four-layer wavelet decomposition on the intrinsic mode function signals to obtain several wavelet components to form new intrinsic mode function components; the principal component analysis processing is to perform dimensionality reduction processing on the obtained new intrinsic mode function components.

[0014] The construction of the new variational mode decomposition model includes sample data set classification, training sample data sets, and test sample data sets.

[0015] The sample data set classification is used to divide the obtained sample data into fault 1 samples, fault 2 samples... fault n samples; the training sample data sets include finally classifying the divided samples using the improved K-nearest neighbor algorithm, training sample data sets, test sample data sets, and output results.

[0016] The sample data set classification classifies the sample data set; the training sample data sets train the classified samples and divide the fault simulation data into a training set and a test set at a ratio of 90% to 10%.

[0017] The fault simulation data acquisition obtains the time series data of the fault through the data acquisition module, and every 1000 time series data are taken as a set of data.

[0018] The improved K-nearest neighbor algorithm recognition and classification can be better classified through the K-nearest neighbor algorithm optimized by the KD tree, because the KD tree can quickly find K nearest neighbors, which is used to reduce the search times, improve the work efficiency, and reduce the calculation amount.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: A power system transmission line fault diagnosis method based on the combination of new variational mode decomposition and K-nearest neighbor algorithm provided by the present invention. The present invention firstly proposes a method for processing non-linear and non-stationary signal fault signals, which has the characteristics of excellent performance such as quickly and accurately identifying different fault types under different conditions, and has effectiveness and superiority. The present invention combines the new variational mode decomposition with the improved K-nearest neighbor algorithm to construct a new power system transmission line fault diagnosis framework. Compared with the traditional combination of variational mode decomposition and support vector machine and the combination of mutual dimensionless and evidence theory, etc., the present invention has superiority and provides a new idea for the power system transmission line fault diagnosis. In short, the present invention can provide new and more means for power system industry practitioners to detect faults in the process of power system transmission lines, improve the discrimination ability and prevention ability of power system industry practitioners for accidents occurring during the operation of transmission lines, avoid the occurrence of major accidents caused by transmission line failures, ensure the safety of power system industry practitioners, and ensure the normal operation of transmission lines. Description of the Drawings

[0020] Figure 1 It is a structural schematic diagram of the power system transmission line fault diagnosis method based on the combination of new variational mode decomposition and K-nearest neighbor algorithm of the present invention. Detailed Embodiments

[0021] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0022] The present invention provides a Figure 1 power system transmission line fault diagnosis method based on the combination of new variational mode decomposition and K-nearest neighbor algorithm, including a fault simulation data acquisition module, a module for constructing a new variational mode decomposition model, a module for training a new variational mode decomposition model, and an improved K-nearest neighbor algorithm recognition and classification module, and includes the following steps:

[0023] S1. Fault simulation data acquisition: The fault simulation data acquisition module is connected to the module for constructing a new variational mode decomposition model, and is used to acquire the original signal dataset;

[0024] S2. Fault data preprocessing: The module for constructing a new variational mode decomposition model is respectively connected to the fault simulation data acquisition module and the module for training a new variational mode decomposition model, and is used to construct a new variational mode decomposition model with the acquired data and transmit the new variational mode decomposition to the module for training a new variational mode decomposition model;

[0025] S3. The module for training a new variational mode decomposition model is connected to the module for constructing a new variational mode decomposition model and the module for identifying and classifying with an improved K-nearest neighbor algorithm, and is used to construct a new variational mode decomposition model with the acquired signal data;

[0026] Training a new variational mode decomposition model includes variational mode decomposition processing optimized by a genetic algorithm, wavelet decomposition processing, principal component analysis for dimensionality reduction processing, training sample dataset, test sample dataset, constructing a model, and obtaining a model;

[0027] S4. The module for identifying and classifying with an improved K-nearest neighbor algorithm is connected to the module for training a new variational mode decomposition model, and is used to identify the diagnostic result.

[0028] In the said module for constructing a new variational mode decomposition model, after the signal dataset undergoes variational mode decomposition processing optimized by a genetic algorithm to obtain the intrinsic mode function, the intrinsic mode function signal is then subjected to wavelet decomposition processing, and finally dimensionality reduction processing is performed through principal component analysis to obtain a new variational mode decomposition model.

[0029] The said obtaining of the intrinsic mode function signal by using variational mode decomposition optimized by a genetic algorithm can decompose the signal into the intrinsic mode function signal; the said wavelet decomposition processing is to perform four-layer wavelet decomposition on the intrinsic mode function signal to obtain several wavelet components to form new intrinsic mode function components; the said principal component analysis processing is to perform dimensionality reduction processing on the obtained new intrinsic mode function components.

[0030] The said obtaining of the intrinsic mode function signal by using variational mode decomposition optimized by a genetic algorithm can decompose the signal into the intrinsic mode function signal; the said wavelet decomposition processing is to perform four-layer wavelet decomposition on the intrinsic mode function signal to obtain several wavelet components to form new intrinsic mode function components; the said principal component analysis processing is to perform dimensionality reduction processing on the obtained new intrinsic mode function components.

[0031] The said module for constructing a new variational mode decomposition model includes sample dataset classification, training sample dataset, and test sample dataset.

[0032] The classification of the sample data set is used to divide the obtained sample data into sample of fault 1, sample of fault 2... sample of fault n; the training sample data set includes classifying the divided samples finally using the improved K-nearest neighbor algorithm, the training sample data set, the test sample data set, and the output result.

[0033] The classification of the sample data set classifies the sample data set; the training sample data set trains the classified samples, and divides the fault simulation data into a training set and a test set at a ratio of 90% to 10%.

[0034] The acquisition of the fault simulation data obtains the time series data of the fault through the data acquisition module, and every 1000 time series data is a set of data.

[0035] The improved K-nearest neighbor algorithm for recognition and classification can classify better through the K-nearest neighbor algorithm optimized by the KD tree, because the KD tree can quickly find K nearest neighbors, which is used to reduce the search times, improve work efficiency, and reduce the calculation amount.

[0036] The module for constructing a new variational mode decomposition model is respectively connected to the data acquisition module and the module for training a new variational mode decomposition model, and is used to transform the original signal into n different modal signals such as intrinsic mode function 1, intrinsic mode function 2, intrinsic mode function 3, intrinsic mode function 4... intrinsic mode function n through variational mode decomposition optimized by the genetic algorithm, then obtain new intrinsic mode function components through four-layer wavelet decomposition processing, and finally perform dimensionality reduction processing through principal component analysis to obtain a new variational mode decomposition model.

[0037] The module for training a new variational mode decomposition model is respectively connected to the module for constructing a new variational mode decomposition model and the module for recognition and classification using the improved K-nearest neighbor algorithm, and is used to divide the new variational mode decomposition signal into sample of fault 1, sample of fault 2... sample of fault n, then divide each kind of fault simulation data into a training set and a test set at a ratio of 90% to 10%, and go through the training sample data set, the test sample data set, construct a model, and obtain a model.

[0038] The data set acquisition module is connected to the module for constructing a new variational mode decomposition model, and is used to collect voltage and current signals of different fault types at the same position in the simulation model, and obtain that each set of data contains 1000 fault acquisition points. The improved K-nearest neighbor algorithm for recognition and classification mainly can classify better through the K-nearest neighbor algorithm optimized by the KD tree, because the advantage of the KD tree is that it can quickly find k nearest neighbors, which is used to reduce the search times, improve work efficiency, and reduce the calculation amount.

[0039] In the embodiment of the present invention, the acquisition of the fault simulation data set includes acquiring voltage and current signals with a sampling frequency of 1200, obtaining the time series data of the fault through the data acquisition module, and taking every 1000 time series data as a set of data. A new variational mode decomposition model is constructed, and the acquired data set is processed by variational mode decomposition optimized by the genetic algorithm. The variational mode decomposition optimized by the genetic algorithm can decompose the signal into intrinsic mode functions, and each intrinsic mode function can be decomposed to describe the dynamic characteristics of the original signal. By effectively decomposing the fault vibration signal into intrinsic mode functions, there is no problem of mode mixing. To solve the problem of data distortion, the present invention proposes to obtain new intrinsic mode function components by decomposing the intrinsic mode function signal through four-layer wavelet decomposition, and finally obtain a new variational mode decomposition model through principal component analysis for dimensionality reduction processing.

[0040] Furthermore, the new variational mode decomposition is used as a sample data set and is planned to be divided into fault 1 samples, fault 2 samples... fault n samples. Then, each type of fault simulation data is divided into a training set and a test set at a ratio of 90% to 10%. After the training sample data set, the test sample data set, the model is constructed, and the model is obtained.

[0041] The final improved K-nearest neighbor algorithm for recognition and classification mainly uses the K-nearest neighbor algorithm optimized by the KD tree for better classification. Because the advantage of the KD tree is that it can quickly find k nearest neighbors, which is used to reduce the search times, improve work efficiency, and reduce the amount of calculation.

[0042] The present invention can accurately and quickly detect power system faults in various complex environments. Accurate detection in various complex environments means constructing a new variational mode decomposition model and using variational mode decomposition to effectively decompose the signal into intrinsic mode functions. Each intrinsic mode function can be decomposed to describe the dynamic characteristics of the original signal, and there is no problem of mode mixing, which may lead to uncertain diagnostic results. Fast detection means constructing a new variational mode decomposition model using an optimized model of variational mode decomposition optimized by the genetic algorithm. It can overcome the problem of mode mixing. During the process of decomposing modal components, the parameters K and can be determined through continuous optimization, and then the combination of them can be optimized as a whole, which can effectively separate the modal components of the signal. Therefore, the present invention can diagnose various types of power system transmission line faults all-round and all-weather.

[0043] The obtained intrinsic mode function signal is obtained by processing the original signal data set through variational mode decomposition optimized by the genetic algorithm. The variational mode decomposition optimized by the genetic algorithm can decompose the signal into intrinsic mode functions, and each intrinsic mode function can be decomposed to describe the dynamic characteristics of the original signal. Perform variational mode decomposition on the original voltage and current signals, and optimize its parameters K and , the final intrinsic mode function components can be obtained. The present invention proposes to effectively decompose the fault vibration signal into intrinsic mode functions without the problem of mode mixing.

[0044] The variational mode decomposition optimized by the genetic algorithm is added to reconstruct the original vibration signal, which is decomposed into multiple intrinsic mode functions. The noise of the reconstructed signal is significantly reduced. Furthermore, to solve the problem of data distortion, four-layer wavelet decomposition is used to decompose the obtained intrinsic mode functions. Finally, a new variational mode decomposition model is obtained through dimensionality reduction processing by principal component analysis. This part of the simulation analysis verifies the effectiveness of the new variational mode decomposition. It emphasizes that the variational mode decomposition optimized by the genetic algorithm is more superior than the traditional variational mode decomposition, and that using four-layer wavelet decomposition can effectively solve the problem of data distortion and using principal component analysis for dimensionality reduction processing. It can solve the problem of unclear fault features by combining specific current-voltage characteristics into higher-level features without losing the most important information.

[0045] The number of intrinsic mode functions can be obtained in different quantities by continuously optimizing the value of K according to the genetic algorithm. The selection of the number of intrinsic mode function decompositions in the present invention is made by optimizing and selecting the minimum sample entropy value through the genetic algorithm. The experimental results show that through the overall optimization of the parameters K and combination according to the genetic algorithm, the present invention selects the scheme with 6 intrinsic mode function decompositions. This scheme has good performance.

[0046] Wavelet decomposition processing means that since there is a certain degree of distortion in the fault data corresponding to each intrinsic mode function obtained from vibration monitoring data, wavelet decomposition is used to perform secondary denoising on each component, avoiding the problem of inaccurate fault diagnosis results caused by data distortion at a certain moment and improving the accuracy of fault diagnosis for the power system. Four-layer wavelet decomposition is performed on the intrinsic mode functions to obtain new intrinsic mode function components.

[0047] Principal component analysis processing is because the dimension of the generated new intrinsic mode function components is relatively high, so dimensionality reduction processing is performed on them. And it can solve the problem of unclear fault features by combining specific current-voltage characteristics into higher-level features without losing the most important information.

[0048] Since the amount of data collected in this experiment is small, the present invention uses an improved K-nearest neighbor algorithm for fault identification. Because the improved K-nearest neighbor algorithm has good classification performance for small sample data, its greatest advantage is that the method is simple, easy to understand, does not require parameter estimation and training, and is very suitable for the classification of rare events, and its application in fault diagnosis is also relatively extensive.

[0049] The training of the new variational mode decomposition model receives the new variational mode decomposition processed by the constructed new variational mode decomposition model, and then classifies the sample data set. The classification of the sample data set is divided into normal samples, fault 2 samples, fault 3 samples... fault n samples. The purpose is to better perform machine learning and obtain accurate results more quickly.

[0050] In the present invention, the time series is decomposed by variational mode decomposition optimized by the genetic algorithm for the original voltage and current signals to obtain the optimal number of intrinsic mode functions (intrinsic mode functions) containing fault feature information. Then, each intrinsic mode function is decomposed by four-layer wavelet decomposition to obtain new intrinsic mode function components, avoiding the problem of inaccurate fault diagnosis results caused by data distortion at a certain moment. Then, principal component analysis is used for dimensionality reduction processing, which can solve the problem of unobvious fault features by combining specific current and voltage features into higher-level features without losing the most important information. Finally, the improved K-nearest neighbor algorithm recognition uses the K-nearest neighbor algorithm recognition optimized by the KD tree to output the diagnosis result.

[0051] In summary, compared with the prior art, the present invention first proposes a method for processing non-linear and non-stationary signal fault signals, which has the excellent performance characteristics of quickly and accurately identifying different fault types under different conditions, and has effectiveness and superiority. The present invention first combines variational mode decomposition optimized by the genetic algorithm, wavelet decomposition, principal component analysis, and the improved K-nearest neighbor algorithm to construct a new fault diagnosis framework for the power system transmission line. Compared with the traditional variational mode decomposition and the latest published methods, the present invention has superiority and provides a new idea for intelligent fault diagnosis. In short, the present invention can provide new and more means for power system industry practitioners to detect the safe operation of power system transmission lines, improve the discrimination ability and prevention ability of power system industry practitioners for accidents occurring during the safe operation of transmission lines, avoid the occurrence of major accidents caused by faults in power system transmission lines, ensure the safety of power system industry practitioners, and ensure the safe and stable operation of the power system field.

[0052] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A power system transmission line fault diagnosis method based on the combination of new variational mode decomposition and K-nearest neighbor algorithm, including a fault simulation data acquisition module, a module for constructing a new variational mode decomposition model, a module for training the new variational mode decomposition model, and an improved K-nearest neighbor algorithm recognition and classification module, characterized in that: It includes the following steps: S1. Fault simulation data acquisition: The fault simulation data acquisition module is connected to the module for constructing a new variational mode decomposition model, and is used to acquire the original signal dataset; the fault simulation data acquisition includes voltage and current signals with a sampling frequency of 1200, and the time series data of the fault is obtained through the data acquisition module. Every 1000 time series data is a set of data; S2. Fault data preprocessing: The module for constructing a new variational mode decomposition model is respectively connected to the fault simulation data acquisition module and the module for training a new variational mode decomposition model, and is used to construct a new variational mode decomposition model with the acquired data and transmit the new variational mode decomposition to the module for training a new variational mode decomposition model; In the construction of the new variational mode decomposition model, the signal dataset is processed by variational mode decomposition optimized by the genetic algorithm to obtain the intrinsic mode function, and then the intrinsic mode function signal is processed by wavelet decomposition. Finally, dimensionality reduction processing is performed through principal component analysis to obtain the new variational mode decomposition model; S3. The module for training a new variational mode decomposition model is connected to the module for constructing a new variational mode decomposition model and the module for identifying and classifying by the improved K-nearest neighbor algorithm, and is used to construct a new variational mode decomposition model with the acquired signal data; Training the new variational mode decomposition model includes variational mode decomposition processing optimized by the genetic algorithm, wavelet decomposition processing, principal component analysis dimensionality reduction processing, training sample dataset, test sample dataset, model construction, and model acquisition; S4. The module for identifying and classifying by the improved K-nearest neighbor algorithm is connected to the module for training a new variational mode decomposition model and is used to identify the diagnostic result; the improved K-nearest neighbor algorithm for identification and classification realizes better classification through the K-nearest neighbor algorithm optimized by the KD tree. Because the KD tree can quickly find K nearest neighbors, it is used to reduce the search times, improve the work efficiency, and reduce the calculation amount.

2. A fault diagnosis method for a power system transmission line based on the combination of new variational mode decomposition and K-nearest neighbor algorithm according to claim 1, characterized in that: The intrinsic mode function signal is decomposed into the intrinsic mode function signal by variational mode decomposition optimized by the genetic algorithm; the wavelet decomposition processing is to perform four-layer wavelet decomposition on the intrinsic mode function signal to obtain several wavelet components to form new intrinsic mode function components; the principal component analysis processing is to perform dimensionality reduction processing on the obtained new intrinsic mode function components.

3. A fault diagnosis method for a power system transmission line based on the combination of new variational mode decomposition and K-nearest neighbor algorithm according to claim 2, characterized in that: The construction of the new variational mode decomposition model includes sample dataset classification, training sample dataset, and test sample dataset.

4. A fault diagnosis method for a power system transmission line based on the combination of new variational mode decomposition and K-nearest neighbor algorithm according to claim 3, characterized in that: The sample dataset classification is used to divide the obtained sample data into fault 1 sample, fault 2 sample... fault n sample; the training sample dataset includes finally classifying the divided samples by the improved K-nearest neighbor algorithm, training sample dataset, test sample dataset, and output result.

5. A fault diagnosis method for a power system transmission line based on the combination of new variational mode decomposition and K-nearest neighbor algorithm according to claim 3, characterized in that: The sample dataset classification classifies the sample dataset; the training sample dataset trains the classified samples, and divides the fault simulation data into a training set and a test set at a ratio of 90% to 10%.

6. A fault diagnosis method for a power system transmission line based on the combination of new variational mode decomposition and the K-nearest neighbor algorithm according to claim 1, characterized in that: The fault simulation data acquisition obtains the time series data of the fault through the data acquisition module, and every 1000 time series data is a set of data.

Images