CN113378652B - Disturbance classification method based on EWT-MPE-PSO-BP - Google Patents

Disturbance classification method based on EWT-MPE-PSO-BP Download PDF

Info

Publication number
CN113378652B
CN113378652B CN202110559254.7A CN202110559254A CN113378652B CN 113378652 B CN113378652 B CN 113378652B CN 202110559254 A CN202110559254 A CN 202110559254A CN 113378652 B CN113378652 B CN 113378652B
Authority
CN
China
Prior art keywords
neural network
pso
disturbance
ewt
function
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110559254.7A
Other languages
Chinese (zh)
Other versions
CN113378652A (en
Inventor
王宝华
薛凯
蒋海峰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN202110559254.7A priority Critical patent/CN113378652B/en
Publication of CN113378652A publication Critical patent/CN113378652A/en
Application granted granted Critical
Publication of CN113378652B publication Critical patent/CN113378652B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction

Abstract

The application relates to a power quality disturbance classification method based on an EWT-MPE-PSO-BP neural network, firstly, the EWT carries out accurate modal decomposition with anti-noise performance on disturbance signals of different types to obtain modal components with different frequencies; then, introducing a time scale concept, and optimizing the traditional permutation entropy to be better suitable for the problem of a complex system; then, PSO is introduced to optimize the BP neural network, the problem of searching the minimum error in BP is converted into the problem of searching the optimal position of PSO, the defect of slow convergence speed in the BP neural network is improved, and the working efficiency of the BP network is improved; and finally, taking the extracted characteristic quantity as the input of the optimized neural network, and obtaining a final electric energy quality disturbance classification result through multiple times of training. The method provided by the application effectively solves the problems of inaccurate detection of the disturbance signal and low speed of the classification process, and has high accuracy and high working efficiency.

Description

Disturbance classification method based on EWT-MPE-PSO-BP
Technical Field
The application relates to the field of electric energy quality analysis and control, in particular to the field of electric energy quality disturbance classification based on feature extraction and pattern recognition.
Background
Along with the increasing requirements of China on power systems, not only is the power load supply required to have higher stability, but also the power quality is required to have higher reliability. The power quality in the power system shows different characteristics due to the different operating characteristics of the power equipment, the time and the position of the fault, and the like, so that different types of power quality disturbance are generated. Common power quality disturbances mainly include harmonics, voltage imbalances, voltage fluctuations and flicker, transient pulses and oscillations, voltage sags, breaks, etc. The method has the advantages of comprehensively and deeply analyzing the electric energy quality problem, rapidly and accurately identifying the disturbance type of the electric energy quality, timely finding out the reason of the disturbance type, and having important significance for the management and the treatment of the electric energy quality.
The power quality disturbance classification mainly comprises two parts, namely feature extraction and pattern recognition. The electric energy quality disturbance signal parameter detection and the characteristic analysis are the basis of disturbance classification, and the characteristic information with stability and difference is extracted through the analysis of an effective signal processing method, so that a powerful support is provided for electric energy quality disturbance classification. The current main electric energy quality disturbance detection method mainly comprises a time domain analysis method, a frequency domain analysis method and a time-frequency analysis method, wherein the time-frequency analysis method is suitable for non-stationary signals and gives consideration to time-frequency information, and is the main flow analysis method of the current electric energy quality disturbance analysis. The main time-frequency analysis methods include short-time fourier Transform (Short Time Fourier Transform, STFT), wavelet Transform (Wavelet Transform, WT), S Transform and Hilbert yellow Transform (HHT). However, STFT is relatively weak when processing multi-scale signals, the WT redundancy is relatively high, the ST resolution is relatively fixed, and HHT has a mode aliasing phenomenon, which all need to be improved to some extent.
The power quality mode recognition method mainly comprises a decision tree, a support vector machine (Support Vector Machine, SVM), a neural network, an expert control system and the like. The disturbance characteristic value obtained by analyzing and processing the disturbance signal is input into a classifier, so that the classification process of the power quality disturbance signal can be completed. Wherein, when the decision tree processes more questions at a higher level, the attribute weight is not credible, the SVM kernel function selects the questions to be complex, the neural network is easy to fall into the problem of the local optimal solution of the algorithm, the training time is long, and the problem of fitting is solved.
Disclosure of Invention
The application aims to provide an electric energy quality disturbance classification method based on an EWT-MPE-PSO-BP neural network.
The technical solution for realizing the purpose of the application is as follows: a power quality disturbance classification method based on feature extraction and pattern recognition comprises the following steps:
step 1, decomposing an electric energy quality disturbance signal through EWT (adaptive WT), and adaptively obtaining modal components with different frequencies;
step 2, calculating the EWT modal components through MPE to obtain characteristic values of different disturbance types, namely, the characteristic values can be used as the basis for distinguishing the disturbance types;
and step 3, introducing a PSO algorithm to optimize the BP neural network, dividing the power quality disturbance signal into a training data set and a test data set, respectively calculating characteristic values of the power quality disturbance signal as input of the BP neural network, training the optimized BP neural network, and inputting a test sample to obtain a classification result.
Compared with the prior art, the application has the remarkable advantages that: 1) EWT is simple and convenient to calculate, high in decomposition precision, strong in self-adaptation and free of modal aliasing, and is an excellent method for processing signals in time and frequency. 2) The arrangement entropy algorithm is optimized through the multi-scale algorithm, so that the feature extraction of the complex problem is more scientific. 3) The BP neural network is optimized through PSO, so that the convergence speed of the neural network serving as a classifier is higher when complex problems are handled, and the working efficiency is improved.
Drawings
FIG. 1 is a flow chart of a PSO optimized BP neural network
FIG. 2 is a flow chart of a method for classifying power quality disturbances based on EWT-MPE-PSO-BP
FIG. 3 is a modal component diagram of the EWT processing the voltage sag signal
FIG. 4 is a block diagram of a BP neural network according to the application
Detailed Description
The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.
In order to solve the problem of accuracy of the traditional electric energy quality disturbance signal comprehensive classification method, the disturbance classification method based on the EWT-MPE-PSO-BP introduces the EWT into a classification flow, decomposes the disturbance signal through the EWT with good performance, extracts the characteristic quantity by combining with an MPE algorithm, simultaneously introduces PSO to optimize the BP neural network, and inputs the characteristic quantity as an input quantity into the optimized neural network to obtain a classification result. The method comprises the following steps:
step 1 decomposition of disturbance Signal Using EWT
The power quality disturbance signal often contains noise and various frequency waveforms, and proper decomposition thereof to extract the main disturbance for analysis is the basis of feature extraction. The EWT integrates wavelet transformation, empirical mode decomposition and Fourier transformation, can carry out self-adaptive division on signals, and extracts the amplitude-frequency mode of Fourier transformation by constructing a quadrature filter bank so as to realize the self-adaptive decomposition of the signals. The EWT has a series of advantages of simple calculation, high decomposition precision, strong self-adaptability, no modal aliasing phenomenon and the like, so that the EWT has wider application range and better application effect than other time-frequency analysis methods.
Using EWT for disturbance signal decomposition comprises the sub-steps of:
step 1.1 Spectrum segmentation and boundary determination
The power quality disturbance signal f (t) can be expressed as the sum of a series of amplitude modulation and frequency modulation signals, and different modal components f k (t) is characterized by being centered at a certain frequency in the Fourier spectrum, so that an adaptive empirical wavelet filter bank can be used for analysis and extraction.
Performing Fourier transform on F (t) to obtain a frequency spectrum F (omega), and setting the maximum value sequence of F (omega) asFor->The amplitude is arranged in a descending order, and meanwhile, a reference value M is set M +α(M 1 -M M ),α∈(0,1)，/>The number of the maximum values larger than the reference value is N. Obtain the mode numberAfter the number N, F (omega) is divided into N segments, wherein the frequency omega of two adjacent maximum values larger than the reference value in F (omega) is divided into N segments n And omega n+1 Mid-point of (i.e. boundary frequency ω of each segment) n Where n=1, 2,..n-1, the frequency ω at both ends of the spectrum 0 ＝0，ω N ＝π。
Step 1.2 construction of a Scale function and a wavelet function
Empirical wavelet scale functionAnd empirical wavelet function->According to boundary frequency omega n The construction is carried out:
wherein, beta (x) satisfies:
the value of γ is the ratio of the filter bandwidth to the cut-off frequency, and can be expressed as:
step 1.3 empirical wavelet transform
The power quality disturbance signal f (t) is subjected to inner product with the two functions obtained in the step 1.2 to obtain an empirical wavelet approximation coefficientAnd empirical wavelet detail coefficients->
In the method, in the process of the application,and->Complex conjugate of empirical scale function and empirical wavelet function, respectively, +.>And->Respectively fourier transforms thereof; />Is the fourier transform of the disturbance signal f (t).
f (t) empirical mode function f obtained by decomposition k (t)：
In phi 1 (t) and ψ k (t) empirical scale functions and empirical wavelet functions, respectively.
Step 2 feature calculation Using MPE
The extraction of the power quality disturbance characteristics is the basis for classifying disturbance, the complexity of characteristic values is caused by the diversity of power quality types, and the scientific and reasonable characteristic calculation is critical based on the modal components obtained in the step 1. The Multi-scale permutation entropy (Multi-scale Permutation Entropy, MPE) is a eigenvalue algorithm that combines a Multi-scale algorithm with the permutation entropy. Different from PE algorithm, MPE coarsely processes signals to form a multi-scale time sequence, and then calculates entropy value.
The disturbance signal feature extraction comprises the following substeps:
step 2.1 Modal component pretreatment
The decomposition of the power quality disturbance signals is completed through the step 1, a plurality of EWT modal components are obtained, a reference threshold v is required to be set, the components with low correlation with the initial signals are removed by comparing the relation number of each modal component and the initial signals with the threshold, only effective modal components are saved, and the v value can be set to be 0.1-0.3.
Step 2.2MPE calculation
Set the modal component f k (t) has a length U and is divided into windows f in sequence k (t)＝{f k (1)，f k (2),...f k (U) } the time scale of the window is s, coarse-grained processing is performed, and a new time sequence is obtained:
s=1, the processed sequence becomes the original time sequence; when s=u, the original time sequence is changed into U/s coarse grain sequences, and the coarse grain sequences are compared with the original time sequenceThe sequence reconstruction is carried out:
wherein l is the first reconstruction component, l ε [1, N- (m-1) τ ]; τ is the delay time and m is the embedding dimension.
The sequence will be reconstructedAscending sort and use [ l ] 1 ，l 2 ,...,l m ]Representing reconstruction component +.>The index of each element column is: />
For any group of sequences, a symbol sequence s (g) = (l) can be obtained 1 ，l 2 ,...,l m ) Wherein g is [1, G ]]G.ltoreq.m-! . For each symbol sequence, the probability of occurrence is P g Permutation entropy H of different symbol sequences p (m) the form of shannon entropy can be defined as:
when p is g =1/m-! When H is p (m) obtaining the maximum value of ln (m|), and carrying out normalization processing to obtain the characteristic value of the permutation entropy, namely the disturbance signal, which is:
in the characteristic value calculation process, the values of the embedding dimension m and the delay time tau influence the calculation result of the characteristic value, and through experimental verification, the embedding dimension of the modal component is generally set to be between 3 and 7. Since the delay time hardly affects the result value in the calculation process, it can be set to 1.
Combining the step 1 and the step 2, namely the characteristic extraction process of the electric energy quality disturbance signals, decomposing the electric energy quality disturbance signals through EWT, obtaining modal components with different frequencies in a self-adaptive mode, and calculating the modal components through MPE to obtain characteristic values of different disturbance types, wherein the characteristic values can be used as the basis for distinguishing the disturbance types.
Step 3 Pattern recognition Using PSO-BP neural network
The pattern recognition is to distinguish different disturbance signals through characteristic values to obtain classification results, and is an important step of power quality disturbance classification. The BP neural network has simple structure, strong controllability and good fault tolerance and adaptability, but is easy to fall into the problem of local optimal solution of an algorithm, and the training time is long, so that the BP neural network is optimized by introducing a PSO algorithm, particles of the connection weight of each network are enabled to be in one-to-one correspondence with the neural network, and the connection weight of the corresponding neural network is updated due to the fact that the position of the particles is updated. The problem of searching the minimum error in the BP is converted into the problem of searching the optimal position of the PSO, the defect of slow convergence speed in the BP neural network is overcome, and the working efficiency of the BP network is improved.
Pattern recognition using PSO-BP includes the sub-steps of:
step 3.1BP neural network
The training process of the BP neural network can be described as follows:
the input layer has X sections, nodes of the same layer are not connected, and all neurons are independent from each other;
the hidden layers share Y sections, each independent hidden layer neuron is connected with all neurons of the input layer and the output layer, and the connection weight of the hidden layer and the input layer is w ij I=1, 2, X, j=1, 2, Y, and the connection weight to the output layer is w jk ,j＝1,2,...,Y,j＝1,2,...,Z；
The output layer has Z sections, nodes among the same layers are not connected, all neurons are independent, and the three relations are as follows:
the transfer function of the hidden layer adopts a Tansig type function, i.e
The transfer function of the output layer being a log-type function, i.e
Defining the input of the independent unit in the P-th training as a Pi Output is o Pi The input, output values can be expressed as:
since the initial weight settings in the different training modes are not the same, the error between the actual output and the expected value for the network can be expressed as:
wherein d Pi The output is expected for a network of P trains. In each training process, the connection weight w is needed ji The adjustment value aw along the negative gradient direction of the error can be calculated as:
and correspondingly adjusting an error formula aiming at weight errors of different layers. The error calculation process is repeated continuously until all samples are trained to obtain a global error, the global error is compared with an error value preset by a network, and if the value is smaller than a set value, a good network training result is proved; if the value is greater than the preset value, training is returned again. If the error condition is not satisfied within the predetermined number of training times, the network does not converge, and improvement is required.
Step 3.2PSO optimizing BP neural network
When the weight of the BP neural network is optimized by adopting a particle swarm optimization, the weight in the BP neural network is generally taken as a position vector of particles in a PSO algorithm, and then the updating of the BP neural network connection weight is realized by utilizing the particle swarm algorithm, so that an optimal connection weight is finally obtained. The method comprises the following steps:
1) The feature value sample obtained in the feature extraction process is used as a training sample, the structure of the BP neural network is determined, and structural parameters are set for the BP neural network;
2) Initializing particle swarm algorithm including particle swarm size S, space dimension D, and learning factor c 1 ,c 2 Inertial weight wThe method comprises the steps of carrying out a first treatment on the surface of the The particle swarm size S is determined according to the complexity of the problem, a better optimization effect can be obtained by generally taking 20-40, and the particle space dimension D corresponds to the number of the connection weights and is determined by a specific BP neural network structure.
Since the algorithm convergence speed of PSO is affected by the set parameters, the PSO algorithm is improved to a certain extent, and the learning factor c is calculated 1 ,c 2 And the inertia weight w is adjusted to a certain extent, and the adjustment formula is as follows:
wherein, c 1max 、c 1min 、c 2max 、c 2min 、w max 、w min The corresponding maximum and minimum values are respectively the maximum iteration number and the current iteration number T, t;
3) Randomly initializing particle swarm, since the range of BP neural network training function is [ -1,1]For the position interval [ x ] min ,x max ]Position x of particles in i Speed interval v min ,v max ]Velocity v of particles in i Proceeding with [ -1,1]The uniform random distribution in the interval generates r particle populations, represents neural networks with different weights, and serves as an initial solution set of a PSO algorithm, and the initialized connection weights generate an initialized BP neural network.
4) Updating the individual optimal value pbest of each particle in the particle swarm, and screening the population optimal value gbest from the individual optimal value pbest, wherein the updating formula is as follows:
wherein x is i S is the particle group size for the position of the ith particle in D-dimensional space; f (·) is an fitness function combined with BP neural network to trainMean square error accuracy is refined as a function of the fitness of the particle, and therefore, f (·) is expressed as:
wherein R is the number of training samples, Q is the number of output nerves,y jk and respectively obtaining an expected output value and an actual output value of a j-th network of a k-th sample, wherein in the BP neural network, to minimize network errors, the connection weight of an output layer is adjusted, the connection weight is adjusted through a formula (15), and when the algorithm stops iterating, the particle with the minimum training error is the optimal solution for optimization.
5) Updating the particle speed and the position, and updating the formula of the ith particle of the kth generation in the D-th dimension (1 < D < D) space is as follows:
x i (k+1)＝x i (k)+v i (k+1) (21)
wherein w is inertial weight, c 1 ,c 2 As learning factors, rand (·) ε [0,1]；x i For the position of the ith particle in D-dimensional space, v i Is the current velocity of the ith particle.
6) Judging whether an ending condition is met, stopping iteration if the iteration number reaches the set maximum number or the fitness value reaches the set precision or the minimum error requirement, wherein the position of the particle is the optimal solution obtained by optimization, namely the initial weight of the BP neural network, otherwise, returning to continuously searching the optimal value of the population;
the PSO optimization BP neural network flow is shown in figure 1.
Step 3.3 Pattern recognition Using PSO-BP
The power quality disturbance signals are selected, a certain number of samples are selected from each signal, z% of the samples are randomly selected as training data sets, the rest samples are used as test data sets, and z is generally 40-60. Inputting the feature quantity obtained in the step 1 and the step 2 into the PSO-optimized BP neural network, and inputting the rest test data set into the trained PSO-BP neural network for disturbance recognition to obtain a classification result.
The flow of classifying the power quality disturbance through the EWT-MPE-PSO-BP neural network is shown in figure 2.
Examples
In order to verify the effectiveness of the inventive protocol, the following simulation experiments were performed.
1) based on the basic concept of the power quality disturbance and the disturbance frequently occurring in the actual power quality problem, eight power quality disturbances of normal voltage (C0), voltage sag (C1), voltage dip (C2), voltage interruption (C3), harmonic wave (C4), transient oscillation (C5), transient pulse (C6) and voltage fluctuation (C7) are selected as examples for analysis. Setting the fundamental frequency f of disturbance signals 0 ＝50Hz，ω＝2πf 0 Sampling frequency f s =5khz, the sampling duration is 0.4s (20 power frequency cycles), and the voltage amplitude is normalized. And obtaining signal waveforms of different disturbance through a mathematical model, and carrying out EWT decomposition treatment to obtain modal components after overlapping 30dB Gaussian white noise. Taking a voltage transient rise signal as an example, the mathematical model is:
f(t)＝{1+α[u(t 2 )-u(t 1 )]}sin(ωt)0.1≤α≤0.8
let α=0.2, t 1 ＝0.14s，t 2 =0.2 s, and EWT decomposition is performed after noise is superimposed, resulting in the modal components shown in fig. 3. And carrying out EWT decomposition processing on different disturbance signals to obtain modal components.
2) The reference threshold v=0.2 is set, components with low correlation with the initial signal are removed, and only valid mode components are saved.
The MPE value of the effective modal component is calculated and the average value is taken as the eigenvector. The MPE algorithm is set, and when the time scale factor is taken to be 10, the subsequent entropy change amplitude is smaller, so that the time scale is selected to be 10. Referring to the parameter selection in the usual method, the value of the embedding dimension m is taken as 6. The eigenvalues calculated for the different disturbance signals using MPE are shown in table 1.
Table 1 eigenvector values of power quality disturbance signals
3) Setting a BP neural network:
the BP neural network is set to be a three-layer structure of an input layer, an output layer and an hidden layer. The characteristic quantity obtained after the disturbance signal is processed by EWT-MPE is taken as input, the input layer is 10 neurons, different disturbance types are taken as output, the output layer is 8 neurons, the number of neurons of the hidden layer can be set to be 6-16 according to the formula (12), and the value is set to be 15 through simulation calculation.
Setting an activation function as a Sigmoid type, setting a transfer function as a Log-Sigmoid type, and setting a training function tranlm type.
Improving the parameters of the PSO algorithm by referring to the formula (16), and learning the improved factor c 1 ,c 2 And inertial weights w are 0.5, 2.5 and 0.4, respectively. The particle swarm size S is generally 20-40, so that a better optimization effect can be achieved, and S is 30. The particle dimension D corresponds to the connection weight, and the BP neural network is of a 3-layer structure type 10-15-8, and has 270 connection weights, so that the D value is 270. And (3) improving the connection weight of the BP neural network by using a PSO algorithm, wherein the improvement process refers to formulas (17) to (21).
The BP neural network structure of the application is shown in figure 4 after design.
4) The relevant parameters of the disturbance signal are set as follows: fundamental frequency f 0 ＝50Hz，ω＝2πf 0 Sampling frequency f s =5khz, the sampling duration is 0.4s (20 power frequency cycles), and the voltage amplitude is normalized. 8 disturbance signals are selected, disturbanceThe coefficient value of the signal is generated by adopting random numbers, each signal generates 100 groups of samples, and Gaussian white noise of 20dB and 30dB is superimposed on the signal.
Firstly, 60 groups of disturbance signals are taken as training data sets, the characteristic quantity of the disturbance signals is input into a BP neural network after PSO optimization, then the rest 40 groups of samples are taken as test data sets, the rest 40 groups of samples are input into the PSO-BP neural network after training for disturbance recognition, and finally the obtained classification result is shown in a table 2.
Table 2 results of power quality disturbance classification
The classification accuracy of the application can be judged through the final classification result. The classification accuracy of the application in the example operation is over 95 percent, and the application has good classification performance.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as the combinations of the technical features do not have a spear shield, they should be considered as the scope of the description.
The above examples illustrate only a few embodiments of the application, which are described in detail and are not to be construed as limiting the scope of the application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of protection of the present application is to be determined by the appended claims.

Claims (4)

1. The disturbance classification method based on EWT-MPE-PSO-BP is characterized by comprising the following steps:
step 1 decomposition of disturbance Signal Using EWT
Decomposing the power quality disturbance signal through EWT, and obtaining modal components with different frequencies in a self-adaptive mode;
step 2 feature calculation Using MPE
Calculating the modal components through MPE to obtain characteristic values of different disturbance types, wherein the characteristic values are used as the basis for distinguishing the disturbance types;
step 3 Pattern recognition Using PSO-BP neural network
Optimizing the BP neural network by introducing a PSO algorithm, enabling particles of the connection weight of each network to correspond to the neural network one by one, converting the problem of searching the minimum error in BP into the problem of searching the optimal position of PSO, training the optimized BP neural network, and inputting a test sample to obtain a classification result;
in the step 1, the disturbance signal is decomposed by using the EWT, and the specific method is as follows:
step 1.1 Spectrum segmentation and boundary determination
Performing Fourier transform on the power quality disturbance signal F (t) to obtain a frequency spectrum F (omega), and setting a maximum value sequence of F (omega) asFor->The amplitude is arranged in descending order, and meanwhile, a reference value M is set M +α(M 1 -M M ),α∈(0,1)，/>The number of maxima larger than the reference value is N, after the mode number N is obtained, F (omega) is divided into N segments, the maxima of two adjacent characteristic values larger than the reference value in F (omega) are taken, and the corresponding frequency omega is obtained n And omega n+1 The midpoint of (a) is the boundary frequency omega of each segment n Where n=1, 2,..n-1, the frequency ω at both ends of the spectrum 0 ＝0，ω n ＝π；
Step 1.2 construction of a Scale function and a wavelet function
Empirical wavelet scale functionAnd empirical wavelet function->According to boundary frequency omega n The construction is carried out:
wherein, beta (x) satisfies:
gamma is the ratio of the filter bandwidth to the cut-off frequency, expressed as:
step 1.3 empirical wavelet transform
The power quality disturbance signal f (t) is subjected to inner product with the two functions obtained in the step 1.2 to obtain an empirical wavelet approximation coefficientAnd empirical wavelet detail coefficients->
In the method, in the process of the application,and->Complex conjugate of empirical scale function and empirical wavelet function, respectively, +.>And->Respectively fourier transforms thereof; />Is the fourier transform of the disturbance signal f (t);
f (t) empirical mode function f obtained by decomposition k (t) is:
in phi 1 (t) and ψ k (t) an empirical scale function and an empirical wavelet function, respectively;
in the step 2, MPE is used for carrying out characteristic calculation, and the specific method comprises the following steps:
step 2.1 Modal component pretreatment
The decomposition of the power quality disturbance signals is completed through the step 1, a plurality of EWT modal components are obtained, a reference threshold v is required to be set, the components with low correlation with the initial signals are removed by comparing the relation number of each modal component and the initial signals with the threshold, and only effective modal components are saved;
step 2.2MPE calculation
Set the modal component f k (t) has a length U and is divided into windows f in sequence k (t)＝{f k (1),f k (2),...f k (U) } with a window time scale of s, coarse-granulating to obtain new timeThe sequence is as follows:
s=1, the processed sequence becomes the original time sequence; when s=u, the original time sequence is changed into U/s coarse grain sequences, and the coarse grain sequences are compared with the original time sequenceThe sequence reconstruction is carried out:
wherein, l is the first reconstruction component, l epsilon [1, U- (m-1) tau' ]; τ' is the delay time and m is the embedding dimension;
the sequence will be reconstructedAscending sort and use [ l ] 1 ,l 2 ,...,l m ]Representing reconstruction component +.>The index of the column in which each element is located is: />
For any group of sequences, a symbol sequence s (g) = (l) is obtained 1 ,l 2 ,...,l m ) Wherein g is [1, G ]]G.ltoreq.m-! The method comprises the steps of carrying out a first treatment on the surface of the For each symbol sequence, the probability of occurrence is P g Permutation entropy H of different symbol sequences p (m) is defined in terms of shannon entropy:
when p is g =1/m-! When H is p (m) obtaining a maximum value of ln (mThe characteristic value of the permutation entropy, namely the disturbance signal, is obtained through the line normalization processing:
in the step 3, a PSO-BP neural network is used for pattern recognition, and the specific method comprises the following steps:
step 3.1BP neural network
The input layer has X sections, nodes of the same layer are not connected, and all neurons are independent from each other;
the connection weight of the hidden layer and the input layer is w ij I=1, 2., X, j=1, 2., Y, the connection weight of the hidden layer and the output layer is w jh ,j＝1,2,...,Y，h＝1,2,...,Z；
The output layer has Z sections, nodes among the same layers are not connected, all neurons are independent, and the three relations are as follows:
the transfer function of the hidden layer adopts a Tansig type function, i.e
The transfer function of the output layer being a log-type function, i.e
Step 3.2PSO optimizing BP neural network
1) The feature value sample obtained in the feature extraction process is used as a training sample, the structure of the BP neural network is determined, and structural parameters are set for the BP neural network;
2) Initializing particle swarm algorithm including particle swarm size S, space dimension D, and learning factor c 1 ,c 2 Inertial weight w; wherein, the particle swarm scale S is determined according to the complexity of the problem, 20-40 is taken, and the space dimension D is correspondingly connectedThe weight number is determined by the BP neural network structure;
since the algorithm convergence speed of PSO is affected by the set parameters, the PSO algorithm is improved to a certain extent, and the learning factor c is calculated 1 ,c 2 And the inertia weight w is adjusted to a certain extent, and the adjustment formula is as follows:
wherein, c 1max 、c 1min 、c 2max 、c 2min 、w max 、w min Respectively C 1 、C 2 Maximum and minimum values corresponding to w, T and T are respectively the maximum iteration number and the current iteration number;
3) Randomly initializing particle swarm, since the range of BP neural network training function is [ -1,1]For the position interval [ x ] min ,x max ]Position x of particles in i' Speed interval v min ,v max ]Velocity v of particles in i' Proceeding with [ -1,1]Uniformly and randomly distributing in the interval to generate r particle populations, representing neural networks with different weights, and generating an initialized BP neural network by using the initialized connection weights as an initial solution set of a PSO algorithm;
4) Updating the individual optimal value pbest of each particle in the particle swarm, screening the population optimal value gbest from the individual optimal value pbest, and respectively updating the following formulas:
wherein x is i' The position of the ith particle in the D-dimensional space, and S is the particle group size; f (·) is an fitness function, combined with the BP neural network, to train the mean square error accuracy as the fitness function of the particles, therefore, f (·) is expressed as:
wherein R is the number of training samples, Q is the number of output nerves, y j'h' 、d j'h' The expected output value and the actual output value of the j 'th network of the h' th sample respectively;
5) Updating the particle speed and the position, and updating the formula of the ith particle of the kth generation in the d-th dimensional space is as follows:
x i' (k'+1)＝x i' (k')+v i' (k') (21)
wherein w is inertial weight, c 1 ,c 2 As learning factors, rand (·) ε [0,1]；x i' For the position of the ith particle in D-dimensional space, v i' A current velocity for the i' th particle;
6) Judging whether an ending condition is met, stopping iteration if the iteration number reaches the set maximum number or the fitness value reaches the set precision or the minimum error requirement, wherein the position of the particle is the optimal solution obtained by optimization, namely the initial weight of the BP neural network, otherwise, returning to continuously searching the optimal value of the population;
step 3.3 Pattern recognition Using PSO-BP
Inputting the feature quantity obtained in the step 1 and the step 2 into the PSO-optimized BP neural network for training, inputting the test data set into the trained PSO-BP neural network for disturbance recognition, and finally obtaining a classification result.
2. Disturbance classification system based on EWT-MPE-PSO-BP, characterized in that a disturbance classification based on EWT-MPE-PSO-BP is achieved based on the method of claim 1.
3. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing an EWT-MPE-PSO-BP based disturbance classification based on the method of claim 1 when the computer program is executed by the processor.
4. A computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements EWT-MPE-PSO-BP based disturbance classification based on the method of claim 1.
CN202110559254.7A 2021-05-21 2021-05-21 Disturbance classification method based on EWT-MPE-PSO-BP Active CN113378652B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110559254.7A CN113378652B (en) 2021-05-21 2021-05-21 Disturbance classification method based on EWT-MPE-PSO-BP

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110559254.7A CN113378652B (en) 2021-05-21 2021-05-21 Disturbance classification method based on EWT-MPE-PSO-BP

Publications (2)

Publication Number Publication Date
CN113378652A CN113378652A (en) 2021-09-10
CN113378652B true CN113378652B (en) 2023-11-24

Family

ID=77571709

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110559254.7A Active CN113378652B (en) 2021-05-21 2021-05-21 Disturbance classification method based on EWT-MPE-PSO-BP

Country Status (1)

Country Link
CN (1) CN113378652B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115144647B (en) * 2022-08-30 2022-12-30 国网江西省电力有限公司电力科学研究院 Intelligent overvoltage identification method and system
CN117060405B (en) * 2023-10-11 2024-02-02 广东鑫钻节能科技股份有限公司 Electric energy quality control method, system, equipment and medium for digital energy oxygen station

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110111015A (en) * 2019-05-13 2019-08-09 中南大学 A kind of power quality analysis method based on the multiple dimensioned arrangement entropy of variation mode decomposition

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110111015A (en) * 2019-05-13 2019-08-09 中南大学 A kind of power quality analysis method based on the multiple dimensioned arrangement entropy of variation mode decomposition

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
基于ＰＳＯ－ＢＰ神经网络的双机架炉卷轧机轧制力预测;王智;《中国优秀硕士学位论文全文数据库工程科技Ⅰ辑》;20180115(第1期);第26-33、56-59页 *
转子振动信号的故障特征提取方法研究;李霁蒲;《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》;20180915(第9期);第23-27页 *
针对群桥水域航路规划问题的ＰＳＯ 算法选比研究;关宏旭等;《武汉理工大学学报》;20150228;第39卷(第1期);第220-223页 *

Also Published As

Publication number Publication date
CN113378652A (en) 2021-09-10

Similar Documents

Publication Publication Date Title
CN109800700B (en) Underwater acoustic signal target classification and identification method based on deep learning
CN113378652B (en) Disturbance classification method based on EWT-MPE-PSO-BP
CN109033719B (en) Wind turbine bearing fault diagnosis method
CN109948597B (en) High-voltage circuit breaker mechanical fault diagnosis method
CN109034054B (en) Harmonic multi-label classification method based on LSTM
CN110728360A (en) Micro-energy device energy identification method based on BP neural network
CN113571067B (en) Voiceprint recognition countermeasure sample generation method based on boundary attack
CN106529428A (en) Underwater target recognition method based on deep learning
CN111723701B (en) Underwater target identification method
CN103675616A (en) Online partial discharge detection signal recognition method of cable
CN109344751B (en) Reconstruction method of noise signal in vehicle
CN111488801A (en) Ship classification method based on vibration noise identification
CN112580588A (en) Intelligent flutter signal identification method based on empirical mode decomposition
CN114897023B (en) Underwater sound target identification method based on extraction of sensitive difference features of underwater sound targets
Zhang et al. Temporal Transformer Networks for Acoustic Scene Classification.
CN113111786B (en) Underwater target identification method based on small sample training diagram convolutional network
Ding et al. Mine microseismic time series data integrated classification based on improved wavelet decomposition and ELM
CN110610203A (en) Electric energy quality disturbance classification method based on DWT and extreme learning machine
CN112434716B (en) Underwater target data amplification method and system based on condition countermeasure neural network
Shi et al. Underwater target recognition based on wavelet packet entropy and probabilistic neural network
Gao et al. Target signal extraction method based on enhanced ICA with reference
CN113269261B (en) Intelligent classification method for distribution network waveforms
Lei et al. Power Quality Disturbance Identification Using Improved Particle Swarm Optimizer and Support Vector Machine
Ullah et al. Development of a novel system for speaker verification
Xin et al. Classification for multiple power quality disturbances based on deep forest

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant