CN113378652A - Disturbance classification method based on EWT-MPE-PSO-BP - Google Patents
Disturbance classification method based on EWT-MPE-PSO-BP Download PDFInfo
- Publication number
- CN113378652A CN113378652A CN202110559254.7A CN202110559254A CN113378652A CN 113378652 A CN113378652 A CN 113378652A CN 202110559254 A CN202110559254 A CN 202110559254A CN 113378652 A CN113378652 A CN 113378652A
- Authority
- CN
- China
- Prior art keywords
- neural network
- pso
- value
- disturbance
- ewt
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000013528 artificial neural network Methods 0.000 claims abstract description 67
- 238000012549 training Methods 0.000 claims abstract description 36
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 16
- 230000008569 process Effects 0.000 claims abstract description 9
- 239000002245 particle Substances 0.000 claims description 46
- 230000006870 function Effects 0.000 claims description 32
- 238000004422 calculation algorithm Methods 0.000 claims description 27
- 238000000605 extraction Methods 0.000 claims description 11
- 210000002569 neuron Anatomy 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000003909 pattern recognition Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 7
- 238000001228 spectrum Methods 0.000 claims description 7
- 238000012546 transfer Methods 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 4
- 230000001174 ascending effect Effects 0.000 claims description 2
- 150000001875 compounds Chemical class 0.000 claims description 2
- 210000005036 nerve Anatomy 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000012216 screening Methods 0.000 claims description 2
- 230000011218 segmentation Effects 0.000 claims description 2
- 238000004590 computer program Methods 0.000 claims 3
- 238000012937 correction Methods 0.000 claims 1
- 238000001514 detection method Methods 0.000 abstract description 3
- 230000007547 defect Effects 0.000 abstract description 2
- 238000004458 analytical method Methods 0.000 description 13
- 238000012545 processing Methods 0.000 description 8
- 238000005070 sampling Methods 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 3
- 230000001052 transient effect Effects 0.000 description 3
- 238000003066 decision tree Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 230000010355 oscillation Effects 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
Images
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—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Public Health (AREA)
- Signal Processing (AREA)
- Water Supply & Treatment (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Complex Calculations (AREA)
Abstract
The invention relates to an electric energy quality disturbance classification method based on an EWT-MPE-PSO-BP neural network, which comprises the following steps that firstly, the EWT carries out accurate modal decomposition with anti-noise performance on different types of disturbance signals to obtain modal components with different frequencies; then, introducing a time scale concept, and optimizing the traditional permutation entropy so as 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 the PSO, the defect of low convergence speed in the BP neural network is overcome, 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 training for multiple times to obtain the final power quality disturbance classification result. The method provided by the invention effectively solves the problems of inaccurate detection of the disturbance signals and slow speed of the classification process, and has high accuracy and high working efficiency.
Description
Technical Field
The invention relates to the field of power quality analysis and control, in particular to the field of power quality disturbance classification based on feature extraction and pattern recognition.
Background
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 different operating characteristics of power equipment, time and position of occurrence of a fault, and the like, thereby generating different types of power quality disturbances. Common power quality disturbances mainly include harmonics, voltage imbalance, voltage fluctuation and flicker, transient pulses and oscillation, voltage ramp-up, voltage ramp-down, interruption, and the like. The method has the advantages of comprehensively and deeply analyzing the power quality problem, quickly and accurately identifying the type of the power quality disturbance, timely finding out the reason of the power quality disturbance, and having important significance for the management and treatment of the power quality.
The classification of the power quality disturbance mainly comprises two parts of feature extraction and pattern recognition. The parameter detection and the feature analysis of the power quality disturbance signal are the basis of disturbance classification, and the feature information with stability and difference is extracted through the analysis of an effective signal processing method, so that powerful support is provided for the power quality disturbance classification. The main existing electric energy quality disturbance detection methods are mainly divided into 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 a mainstream analysis method for electric energy quality disturbance analysis at present. The main Time-frequency analysis methods include Short-Time Fourier Transform (STFT), Wavelet Transform (WT), S-Transform, and Hilbert-Huang Transform (HHT). However, when the STFT is used for processing multi-scale signals, it is weak, the WT redundancy is large, the ST resolution is fixed, and the HHT has modal aliasing, and these methods all need to be improved to some extent.
The electric energy quality mode identification method mainly comprises a decision tree, a Support Vector Machine (SVM), a neural network, an expert control system and the like. And inputting the disturbance characteristic value obtained by analyzing and processing the disturbance signal into a classifier, so that the classification process of the power quality disturbance signal can be completed. The attribute weight is not credible when the decision tree is used for processing more problems, the selection problem of the SVM kernel function is 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 over-fitting problem exists.
Disclosure of Invention
The invention 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 invention is as follows: a power quality disturbance classification method based on feature extraction and pattern recognition comprises the following steps:
step 1, decomposing the power quality disturbance signal through EWT, and obtaining modal components of different frequencies in a self-adaptive manner;
step 2, calculating the EWT modal component through MPE to obtain characteristic values of different disturbance types, which can be used as a basis for distinguishing the disturbance types;
and 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 training data set and the test data set as the input of the BP neural network, and inputting a test sample to obtain a classification result after the optimized BP neural network is trained.
Compared with the prior art, the invention has the remarkable advantages that: 1) the EWT is simple and convenient to calculate, high in decomposition precision, strong in adaptability and free of mode aliasing phenomenon, and is an excellent method for signal time-frequency processing. 2) The permutation entropy algorithm is optimized through the multi-scale algorithm, and the feature extraction of complex problems is more scientific. 3) The BP neural network is optimized through the PSO, so that the neural network as a classifier has higher convergence speed when processing complex problems, 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 power quality disturbance classification method based on EWT-MPE-PSO-BP
FIG. 3 is a diagram of modal components resulting from processing of a voltage ramp signal by EWT
FIG. 4 is a diagram of the BP neural network structure of the present invention
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
The disturbance classification method based on the EWT-MPE-PSO-BP aims at solving the accuracy problem of a traditional electric energy quality disturbance signal comprehensive classification method, the EWT is introduced into a classification flow, disturbance signals are decomposed through the EWT with good performance, feature quantities are extracted by combining an MPE algorithm, PSO is introduced to optimize a BP neural network, and the feature quantities are input into the optimized neural network as input quantities to obtain a classification result. The method comprises the following steps:
step 1 decomposition of disturbance signals 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, an empirical mode decomposition method and a Fourier transformation method, can perform self-adaptive division on signals, and extracts the amplitude-frequency mode of the Fourier transformation by constructing an orthogonal filter bank so as to realize the self-adaptive decomposition of the signals. The EWT has a series of advantages of simplicity and convenience in calculation, high decomposition precision, strong adaptability, no mode aliasing phenomenon and the like, so that the EWT has a wider application range and a better application effect compared with other time-frequency analysis methods.
The decomposition of the perturbation signal using EWT comprises the following sub-steps:
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 fk(t) is expressed in the Fourier spectrum, and has the characteristic of centering on a certain frequency, so that the analysis and the extraction can be carried out by adopting a self-adaptive empirical wavelet filter bank.
Fourier transform is performed on F (t) to obtain a frequency spectrum F (omega), and the maximum value sequence of F (omega) is set asTo pairThe amplitude values are arranged in descending order, and at the same time, a reference value M is setM+α(M1-MM),α∈(0,1),The number of the maximum values which are larger than the reference value in the step (b) is N. After obtaining the number N of the modes, dividing F (omega) into N sections, wherein two adjacent frequencies omega in F (omega) are greater than the maximum value of the reference valuenAnd Ωn+1The middle point of (2), i.e. the boundary frequency ω of the segmentsnWhere N is 1, 2.., N-1, the two-terminal frequency ω of the frequency spectrum0=0,ωN=π。
Step 1.2 construction of scale function and wavelet function
Empirical wavelet scale functionAnd empirical wavelet functionAccording to the boundary frequency omeganThe construction was carried out:
wherein β (x) satisfies:
the value of γ is the ratio of the filter bandwidth to the cutoff frequency, and can be expressed as:
step 1.3 empirical wavelet transform
Performing inner product on the power quality disturbance signal f (t) and the two functions obtained in the step 1.2 to obtain an empirical wavelet approximation coefficientAnd empirical wavelet detail coefficients
In the formula (I), the compound is shown in the specification,andrespectively complex conjugates of an empirical scale function and an empirical wavelet function,andrespectively, its fourier transform;is the fourier transform of the perturbation signal f (t).
f (t) empirical mode function f obtained by decompositionk(t):
In the formula, phi1(t) and ψkAnd (t) is an empirical scale function and an empirical wavelet function respectively.
Step 2 feature computation using MPE
The extraction of the power quality disturbance features is the basis for classifying the disturbances, the complexity of the feature values is caused by the diversity of the power quality types, and the scientific and reasonable feature calculation of the power quality disturbance features is very critical on the basis of the modal components obtained in the step 1. Multi-scale Entropy (MPE) is a eigenvalue algorithm that combines a Multi-scale algorithm with Permutation Entropy. Different from a PE algorithm, MPE firstly performs coarse graining pretreatment on signals to form a multi-scale time sequence, and then calculates entropy values.
The disturbance signal feature extraction comprises the following substeps:
step 2.1 Modal component preprocessing
The method comprises the steps of 1, completing decomposition of an electric energy quality disturbance signal to obtain a plurality of EWT modal components, setting a reference threshold value v, removing the components with low correlation with an initial signal by comparing the correlation coefficient of each modal component with the initial signal and the size of the threshold value, only storing effective modal components, and setting the value of v to be 0.1-0.3.
Step 2.2MPE calculation
Set modal component fk(t) is of length U, and is divided into windows in sequence fk(t)={fk(1),fk(2),...fk(U), the time scale of the window is s, coarse graining treatment is carried out, and the obtained new time sequence is as follows:
when s is 1, the processed sequence becomes an original time sequence; when s is U, the original time sequence becomes U/s coarse grain sequences, and the coarse grain sequences are processedSequence reconstruction was performed as follows:
wherein l is the l reconstruction component, and l belongs to [1, N- (m-1) tau ]; τ is the delay time and m is the embedding dimension.
Reconstructing the sequenceSorting in ascending order and using1,l2,...,lm]Representing reconstructed componentsThe index of each element is as follows:
for any set of sequences, a sequence of symbols s (g) ═ l can be obtained1,l2,...,lm) Wherein G is [1, G ]]G is less than or equal to m! . For each symbol sequence, the probability of occurrence is PgPermutation entropy H of different symbol sequencesp(m) can be defined in the form of shannon entropy:
when p isg1/m! When H is presentp(m) obtaining the maximum value ln (m!), and performing normalization to obtain the permutation entropy, i.e. the characteristic value of the perturbation signal:
in the process of calculating the characteristic value, the values of the embedding dimension m and the delay time tau influence the calculation result of the characteristic value, and experiments verify that the embedding dimension of the modal component is generally set to be 3-7. Since the delay time hardly affects the result value during the calculation, 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 signal, decomposing the electric energy quality disturbance signal through EWT, obtaining modal components with different frequencies in a self-adaptive manner, and calculating the modal components through MPE to obtain characteristic values of different disturbance types, namely the characteristic values can be used as a basis for distinguishing the disturbance types.
Step 3, using PSO-BP neural network to perform pattern recognition
The mode identification is to distinguish different disturbance signals through characteristic values to obtain a classification result, and is an important step of power quality disturbance classification. The BP neural network has a simple structure, strong controllability and good fault tolerance and adaptability, is easy to fall into the problem of algorithm local optimal solution, and takes long training time, so that the BP neural network is optimized by introducing a PSO algorithm, the particles of the connection weight of each network correspond to the neural network one by particlization, and the update of the positions of the particles can lead the connection weight of the corresponding neural network to be 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 low convergence speed in the BP neural network is overcome, and the working efficiency of the BP neural network is improved.
The pattern recognition using the PSO-BP comprises the following substeps:
step 3.1BP neural network
The training process of the BP neural network can be introduced as follows:
the input layer has X nodes, the nodes on the same layer are not connected, and the neurons are independent;
the hidden layer has 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 wijI 1,2,.. times, X, j 1,2,. times, Y, and the connection weight w to the output layerjk,j=1,2,...,Y,j=1,2,...,Z;
The output layer has Z nodes, the nodes on the same layer are not connected, and the neurons are independent of each other, and the relationship among the three is as follows:
Define the input of the independent unit in the P training as aPiThe output is oPiThen the input and output values can be expressed as:
since the initial weight settings in different training modes are not the same, the error between the actual output and the expected value for the network can be expressed as:
in the formula (d)PiAn output is expected for the P trained networks. In each training process, the connection weight w needs to be adjustedjiWith continuous adjustment, the adjustment value Δ w in the negative gradient direction of the error can be calculated as:
and correspondingly adjusting an error formula aiming at the weight errors of different layers. Continuously repeating the error calculation process until all samples are trained to obtain a global error, comparing the value with an error value preset by the network, and if the value is smaller than a set value, proving that the network training result is good; if the value is larger than the preset value, the training is returned again. If the error condition is not satisfied within the predetermined number of training times, the network does not converge and needs to be improved.
Step 3.2PSO optimizing BP neural network
When the weight of the BP neural network is optimized by adopting the particle swarm algorithm, the weight in the BP neural network is generally taken as a position vector of a particle in the PSO algorithm, then the update of the BP neural network connection weight is realized by utilizing the particle swarm algorithm, and finally an optimal connection weight is obtained. The method comprises the following steps:
1) determining the structure of the BP neural network by using a characteristic value sample obtained in the characteristic extraction process as a training sample, and setting the structural parameters of the BP neural network;
2) initializing and setting a particle swarm algorithm, wherein the initialization comprises a particle swarm scale S, a space dimension D and a learning factor c1,c2The inertial weight w; the particle swarm size S is determined according to the complexity of the problem, a good optimization effect can be achieved by generally taking 20-40 as the value, and the particle space dimension D is determined by a specific BP neural network structure corresponding to the number of the connection weights.
Because the PSO algorithm convergence speed is influenced by the set parameters, the learning factor c is improved to a certain extent aiming at the PSO algorithm1,c2And the inertia weight w is adjusted to a certain degree, and the adjustment formula is as follows:
in the formula, c1max、c1min、c2max、c2min、wmax、wminRespectively, its corresponding maximum and minimum values, T, t respectively being the maximum number of iterations and the current number of iterationsCounting;
3) randomly initializing the particle swarm since the range of the BP neural network training function is [ -1,1]For the position interval [ xmin,xmax]Inner particle position xiSpeed interval [ v ]min,vmax]Inner particle velocity viIs carried out in the range of [ -1,1]And (3) uniformly and randomly distributing in the interval to generate r particle populations which represent neural networks with different weights and serve as an initial solution set of the PSO algorithm, and generating an initialized BP neural network by the initialized connection weights.
4) Updating the individual optimum value pbest of each particle in the particle swarm, and screening out the population optimum value gbest from the individual optimum value pbest, wherein the updating formulas are respectively as follows:
wherein x isiThe position of the ith particle in the D-dimensional space is shown, and S is the particle swarm size; f (-) is a fitness function, and in combination with a BP neural network, the training mean square error precision is taken as the fitness function of the particle, so f (-) is expressed as:
wherein R is the number of training samples, Q is the number of output nerves,yjkand (3) respectively setting an expected output value and an actual output value of the jth network of the kth sample, in the BP neural network, in order to minimize the network error, adjusting the connection weight of an output layer, adjusting the connection weight by an equation (15), and when the algorithm stops iteration, training the particle with the minimum error, namely, an optimal solution of the optimization.
5) Updating the speed and the position of the particle, wherein the updating formula of the ith particle of the kth generation in the space of the D dimension (1 < D < D) is as follows:
xi(k+1)=xi(k)+vi(k+1) (21)
wherein w is the inertial weight, c1,c2For learning factor, rand (. epsilon.) [0,1 ]];xiIs the position of the ith particle in D-dimensional space, viIs the current velocity of the ith particle.
6) Judging whether an end condition is met, if the iteration frequency reaches a set maximum frequency or the fitness value reaches a set precision or a minimum error requirement, stopping iteration, and if 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 the attached figure 1.
Step 3.3 Pattern recognition Using PSO-BP
Selecting power quality disturbance signals, selecting a certain number of samples for each signal, randomly selecting z% of the samples as a training data set, and selecting the rest samples as a test data set, wherein z can be generally 40-60. Inputting the characteristic quantities obtained in the training data set through the steps 1 and 2 into the PSO-optimized BP neural network, and inputting the rest test data sets into the trained PSO-BP neural network for disturbance identification to finally obtain a classification result.
The process of classifying the power quality disturbance through the EWT-MPE-PSO-BP neural network is shown in the attached figure 2.
Examples
To verify the validity of the inventive scheme, the following simulation experiment was performed.
1) Based on the basic concept of power quality disturbance and the disturbance frequently occurring in the actual power quality problem, normal voltage (C0), voltage temporary rise (C1), voltage temporary drop (C2), voltage interruption (C3) and normal voltage (C0) are selected,Eight types of power quality disturbances, harmonic (C4), transient oscillation (C5), transient pulse (C6), and voltage fluctuation (C7), were analyzed as examples. Setting the fundamental frequency f of the disturbance signal0=50Hz,ω=2πf0Sampling frequency fs5kHz, the sampling time length is 0.4s (20 power frequency periods), and the voltage amplitude is normalized. And obtaining signal waveforms of different disturbances through a mathematical model, superposing 30dB of Gaussian white noise on the signal waveforms, and performing EWT decomposition processing to obtain modal components. Taking the voltage ramp signal as an example, the mathematical model is:
f(t)={1+α[u(t2)-u(t1)]}sin(ωt)0.1≤α≤0.8
take alpha as 0.2, t1=0.14s,t2Noise was superimposed and EWT decomposition was performed for 0.2s, resulting in the modal component shown in fig. 3. And performing EWT decomposition processing on different disturbance signals to obtain modal components.
2) And setting the reference threshold value v to be 0.2, removing the components with low correlation with the initial signal, and only saving the effective modal components.
MPE values of the effective modal components are calculated, and the average value of the MPE values is taken as a feature vector. Setting the MPE algorithm, and considering that when the time scale factor is 10, the subsequent entropy change range is small, so that the time scale is selected to be 10. With reference to the parameter selection in the usual method, the value of the embedding dimension m is taken to be 6. The characteristic vector values calculated for different perturbation signals using MPE are shown in table 1.
TABLE 1 electric energy quality disturbance signal eigenvector value
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 a hidden layer. The characteristic quantity obtained after the disturbance signal is processed by EWT-MPE is used as input, an input layer is 10 neurons, different disturbance types are used as output, an output layer is 8 neurons, the number of the 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.
The activation function is set to be a Sigmoid type, the transfer function is set to be a Log-Sigmoid type, and the training function is set to be a train type.
The parameter of PSO algorithm is improved by referring to formula (16), and the improved learning factor c1,c2And inertial weights w are 0.5, 2.5 and 0.4, respectively. The particle swarm size S is generally 20-40, and S is 30. The particle dimension D corresponds to the connection weight, and since the BP neural network is a 3-layer structure 10-15-8 type, the total number of the connection weights is 270, and therefore the D value is 270. And improving the connection weight of the BP neural network by using a PSO algorithm, and referring to equations (17) to (21) in the improvement process.
The structure of the designed BP neural network of the invention is shown in figure 4.
4) Setting relevant parameters of the disturbance signal as follows: fundamental frequency f0=50Hz,ω=2πf0Sampling frequency fs5kHz, the sampling time length is 0.4s (20 power frequency periods), and the voltage amplitude is normalized. 8 kinds of disturbance signals are selected, the coefficient value of the disturbance signals is generated by random numbers, 100 groups of samples are generated for each kind of signals, and Gaussian white noise of 20dB and 30dB is superposed on the signals.
Firstly, 60 groups of each disturbance signal are taken as a training data set, the characteristic quantity of each disturbance signal is input into a BP neural network after PSO optimization, then the rest 40 groups of samples are taken as a test data set, the training PSO-BP neural network is input for disturbance identification, and finally the obtained classification result is shown in Table 2.
Table 2 electric energy quality disturbance classification result
The classification accuracy of the invention can be judged by the final classification result. In the example operation, the classification accuracy of the invention is more than 95%, and the classification performance is good.
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, however, as long as there is no such combination, the scope of the present description should be considered as being described in the present specification.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (7)
1. A disturbance classification method based on EWT-MPE-PSO-BP is characterized by comprising the following steps:
step 1 decomposition of disturbance signals using EWT
Decomposing the power quality disturbance signal through the EWT, and obtaining modal components of different frequencies in a self-adaptive manner;
step 2 feature computation using MPE
Calculating the modal components through MPE to obtain characteristic values of different disturbance types as a basis for distinguishing the disturbance types;
step 3, using PSO-BP neural network to perform pattern recognition
And optimizing the BP neural network by introducing a PSO algorithm, so that particles of the connection weight of each network correspond to the neural network one by one, converting the problem of searching the minimum error in the BP into the problem of searching the optimal position of the PSO, training the optimized BP neural network, and inputting a test sample to obtain a classification result.
2. The method for classifying disturbance based on EWT-MPE-PSO-BP as claimed in claim 1, wherein in step 1, EWT is used to decompose disturbance signal, the specific method is:
step 1.1 Spectrum segmentation and boundary determination
Fourier transform is carried out on the power quality disturbance signal F (t) to obtain a frequency spectrum F (omega), and the maximum value sequence of the F (omega) is set asTo pairThe amplitude values are arranged in descending order, and at the same time, a reference value M is setM+α(M1-MM),α∈(0,1),The number of the maximum values which are larger than the reference value in the F (omega) is N, after the modal number N is obtained, the F (omega) is divided into N sections, wherein the frequency omega of two adjacent maximum values which are larger than the reference value in the F (omega) is omeganAnd Ωn+1The middle point of (2), i.e. the boundary frequency ω of the segmentsnWhere N is 1, 2.., N-1, the two-terminal frequency ω of the frequency spectrum0=0,ωN=π;
Step 1.2 construction of scale function and wavelet function
Empirical wavelet scale functionAnd empirical wavelet functionAccording to the boundary frequency omeganThe construction was carried out:
wherein β (x) satisfies:
the value of γ is the ratio of the filter bandwidth to the cutoff frequency, expressed as:
step 1.3 empirical wavelet transform
Performing inner product on the power quality disturbance signal f (t) and the two functions obtained in the step 1.2 to obtain an empirical wavelet approximation coefficientAnd empirical wavelet detail coefficients
In the formula (I), the compound is shown in the specification,andrespectively complex conjugates of an empirical scale function and an empirical wavelet function,andrespectively, its fourier transform;is the fourier transform of the perturbation signal f (t).
f (t) empirical mode function f obtained by decompositionk(t) is:
in the formula, phi1(t) and ψkAnd (t) is an empirical scale function and an empirical wavelet function respectively.
3. The method for classifying disturbance based on EWT-MPE-PSO-BP as claimed in claim 1, wherein MPE is used for feature calculation in step 2, and the specific method is as follows:
step 2.1 Modal component preprocessing
The method comprises the following steps that 1, decomposition of an electric energy quality disturbance signal is completed to obtain a plurality of EWT modal components, a reference threshold value v needs to be set, the components with low correlation with an initial signal are removed by comparing the correlation coefficient of each modal component with the initial signal and the size of the threshold value, and only effective modal components are stored;
step 2.2MPE calculation
Set modal component fk(t) is of length U, and is divided into windows in sequence fk(t)={fk(1),fk(2),...fk(U), the time scale of the window is s, coarse graining treatment is carried out, and the obtained new time sequence is as follows:
when s is 1, the processed sequence becomes an original time sequence; when s is U, the original time sequence becomes U/s coarse grain sequences, and the coarse grain sequences are processedSequence reconstruction was performed as follows:
wherein l is the l reconstruction component, and l belongs to [1, N- (m-1) tau ]; τ is the delay time, m is the embedding dimension;
reconstructing the sequence Yl (s)Sorting in ascending order and using1,l2,...,lm]Representing the reconstructed component Yl (s)The index of the column in which each element is located is:
for any set of sequences, a sequence of symbols s (g) ═ l can be obtained1,l2,...,lm) Wherein G is [1, G ]]G is less than or equal to m! (ii) a For each symbol sequence, the probability of occurrence is PgPermutation entropy H of different symbol sequencesp(m) is defined in the form of shannon entropy as:
when p isg1/m! When H is presentp(m) obtaining the maximum value ln (m!), and performing normalization to obtain the permutation entropy, i.e. the characteristic value of the disturbance signal:
4. the method for classifying disturbance based on EWT-MPE-PSO-BP as claimed in claim 1, wherein in step 3, PSO-BP neural network is used for pattern recognition, and the specific method is as follows:
step 3.1BP neural network
The input layer has X nodes, the nodes on the same layer are not connected, and the neurons are independent;
the hidden layer has Y sections, each independent hidden layer neuron is connected with all neurons of the input layer and the output layer, and the connection weight value of the hidden layer and the input layer is wijI 1,2,.. times, X, j 1,2,. times, Y, and the connection weight w to the output layerjk,j=1,2,...,Y,j=1,2,...,Z;
The output layer has Z nodes, the nodes on the same layer are not connected, and the neurons are independent of each other, and the relationship among the three is as follows:
Define the input of the independent unit in the P training as aPiThe output is oPiThen the input and output values are expressed as:
since the initial weight settings in different training modes are not the same, the error between the actual output and the expected value for the network is expressed as:
in the formula (d)PiFor the expected output of the network with P training times, the connection weight w needs to be adjusted in each training processjiContinuously adjust and recordRepresenting the unit error correction, the adjustment value Δ w in the negative gradient direction of the error is calculated as:
aiming at weight errors of different layers, correspondingly adjusting an error formula, continuously repeating the error calculation process until all samples are trained to obtain a global error, comparing the value with an error value preset by a network, and if the value is smaller than a set value, proving that the network training result is good; if the value is larger than the preset value, returning to training again, and if the error condition is not met within the specified training times, the network is not converged and needs to be improved;
step 3.2PSO optimizing BP neural network
1) Determining the structure of the BP neural network by using a characteristic value sample obtained in the characteristic extraction process as a training sample, and setting the structural parameters of the BP neural network;
2) initializing and setting a particle swarm algorithm, wherein the initialization comprises a particle swarm scale S, a space dimension D and a learning factor c1,c2The inertial weight w; the particle swarm size S is determined according to the complexity of the problem, generally 20-40, the number of the connection weights corresponding to the particle space dimension D is determined by a specific BP neural network structure.
Because the PSO algorithm convergence speed is influenced by the set parameters, the learning factor c is improved to a certain extent aiming at the PSO algorithm1,c2And the inertia weight w is adjusted to a certain degree, and the adjustment formula is as follows:
in the formula, c1max、c1min、c2max、c2min、wmax、wminThe maximum value and the minimum value are respectively corresponding to the current iteration times, and T, t is the maximum iteration time and the current iteration time respectively;
3) randomly initializing the particle swarm since the range of the BP neural network training function is [ -1,1]For the position interval [ xmin,xmax]Inner particle position xiSpeed interval [ v ]min,vmax]Inner particle velocity viIs carried out in the range of [ -1,1]Uniformly and randomly distributing in the interval to generate r particle populations which represent neural networks with different weights and serve as an initial solution set of a PSO algorithm, and generating an initialized BP neural network by the initialized connection weights;
4) updating the individual optimum value pbest of each particle in the particle swarm, screening out the population optimum value gbest from the individual optimum value pbest, wherein the updating formulas are respectively as follows:
wherein x isiThe position of the ith particle in the D-dimensional space is shown, and S is the particle swarm size; f (-) is a fitness function, and in combination with a BP neural network, the training mean square error precision is taken as the fitness function of the particle, so f (-) is expressed as:
wherein R is the number of training samples, Q is the number of output nerves,yjkin the BP neural network, in order to minimize the network error, the connection weight of an output layer is adjusted, the connection weight is adjusted through a formula (15), and when the algorithm stops iteration, the particle with the minimum training error is the optimal solution of optimization;
5) updating the particle speed and the position, wherein the updating formula of the ith particle of the kth generation in the space of the D dimension (1 < D < D) is as follows:
xi(k+1)=xi(k)+vi(k+1) (21)
wherein w is the inertial weight, c1,c2For learning factor, rand (. epsilon.) [0,1 ]];xiIs the position of the ith particle in D-dimensional space, viIs the current velocity of the ith particle;
6) judging whether an end condition is met, if the iteration frequency reaches a set maximum frequency or the fitness value reaches a set precision or a minimum error requirement, stopping iteration, and if 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 characteristic quantities 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 identification, and finally obtaining a classification result.
5. A disturbance classification system based on EWT-MPE-PSO-BP, characterized in that the disturbance classification based on EWT-MPE-PSO-BP is implemented based on the method of any of claims 1-4.
6. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing EWT-MPE-PSO-BP based perturbation classification based on the method of any of claims 1-4 when executing the computer program.
7. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out EWT-MPE-PSO-BP based classification of disturbances based on the method of any one of claims 1-4.
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 true CN113378652A (en) | 2021-09-10 |
CN113378652B 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) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114878900A (en) * | 2022-06-20 | 2022-08-09 | 宁夏隆基宁光仪表股份有限公司 | Adaline neural network and FFT compensated flicker value measuring method |
CN115144647A (en) * | 2022-08-30 | 2022-10-04 | 国网江西省电力有限公司电力科学研究院 | Intelligent overvoltage identification method and system |
CN117060405A (en) * | 2023-10-11 | 2023-11-14 | 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)
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 |
-
2021
- 2021-05-21 CN CN202110559254.7A patent/CN113378652B/en active Active
Patent Citations (1)
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)
Title |
---|
关宏旭等: "针对群桥水域航路规划问题的PSO 算法选比研究", 《武汉理工大学学报》 * |
李霁蒲: "转子振动信号的故障特征提取方法研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》 * |
王智: "基于PSO-BP神经网络的双机架炉卷轧机轧制力预测", 《中国优秀硕士学位论文全文数据库工程科技Ⅰ辑》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114878900A (en) * | 2022-06-20 | 2022-08-09 | 宁夏隆基宁光仪表股份有限公司 | Adaline neural network and FFT compensated flicker value measuring method |
CN115144647A (en) * | 2022-08-30 | 2022-10-04 | 国网江西省电力有限公司电力科学研究院 | Intelligent overvoltage identification method and system |
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 |
CN117060405A (en) * | 2023-10-11 | 2023-11-14 | 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 |
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 |
Also Published As
Publication number | Publication date |
---|---|
CN113378652B (en) | 2023-11-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113378652A (en) | Disturbance classification method based on EWT-MPE-PSO-BP | |
CN113834656B (en) | Bearing fault diagnosis method, system, equipment and terminal | |
CN109948597B (en) | High-voltage circuit breaker mechanical fault diagnosis method | |
CN109033719B (en) | Wind turbine bearing fault diagnosis method | |
Han et al. | A recursive sparse representation strategy for bearing fault diagnosis | |
CN111832353B (en) | Steam turbine rotor fault diagnosis method based on EMD and BA optimization SVM | |
CN109034054B (en) | Harmonic multi-label classification method based on LSTM | |
CN109271975A (en) | A kind of electrical energy power quality disturbance recognition methods based on big data multi-feature extraction synergetic classification | |
CN107506330A (en) | A kind of variation mode decomposition algorithm parameter optimization method based on particle cluster algorithm | |
CN108761287B (en) | Transformer partial discharge type identification method | |
CN111723701B (en) | Underwater target identification method | |
Caicedo et al. | A systematic review of real-time detection and classification of power quality disturbances | |
Dash et al. | Mining for similarities in time series data using wavelet-based feature vectors and neural networks | |
CN112580588B (en) | Intelligent flutter signal identification method based on empirical mode decomposition | |
CN113642484B (en) | Magnetotelluric signal noise suppression method and system based on BP neural network | |
CN114897023B (en) | Underwater sound target identification method based on extraction of sensitive difference features of underwater sound targets | |
CN115238759A (en) | Electric power data compression and electric energy quality disturbance identification method based on compressed sensing | |
CN113971416A (en) | Cable early fault identification method | |
Cai et al. | EMD and GNN-AdaBoost fault diagnosis for urban rail train rolling bearings | |
CN114036977A (en) | Fan power quality disturbance monitoring method based on two-dimensional image coding features | |
CN117077025A (en) | Power quality disturbance identification method based on time domain characteristic kurtosis and neural network | |
CN116559728A (en) | Power transmission line fault diagnosis method based on wavelet transformation-multilayer perceptron | |
Abubakar et al. | Identification of multiple power quality disturbances problems in wind-grid integration system | |
CN116665698A (en) | Pulse sound identification method based on Hilbert-Huang transform and Mel spectrum transform | |
CN115270857A (en) | Transformer substation fault diagnosis method based on data fusion |
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 |