CN112084709A - Large-scale generator insulation state evaluation method based on genetic algorithm and radial basis function neural network - Google Patents
Large-scale generator insulation state evaluation method based on genetic algorithm and radial basis function neural network Download PDFInfo
- Publication number
- CN112084709A CN112084709A CN202010922999.0A CN202010922999A CN112084709A CN 112084709 A CN112084709 A CN 112084709A CN 202010922999 A CN202010922999 A CN 202010922999A CN 112084709 A CN112084709 A CN 112084709A
- Authority
- CN
- China
- Prior art keywords
- neural network
- genetic algorithm
- basis function
- radial basis
- parameters
- 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
- 238000009413 insulation Methods 0.000 title claims abstract description 57
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 53
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 43
- 230000002068 genetic effect Effects 0.000 title claims abstract description 37
- 238000011156 evaluation Methods 0.000 title claims abstract description 13
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000000034 method Methods 0.000 claims abstract description 29
- 230000032683 aging Effects 0.000 claims abstract description 18
- 238000011478 gradient descent method Methods 0.000 claims abstract description 11
- 238000005259 measurement Methods 0.000 claims abstract description 5
- 238000012216 screening Methods 0.000 claims abstract description 4
- 210000002569 neuron Anatomy 0.000 claims description 22
- 230000015556 catabolic process Effects 0.000 claims description 14
- 239000013598 vector Substances 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 8
- 238000010521 absorption reaction Methods 0.000 claims description 4
- 238000010219 correlation analysis Methods 0.000 claims description 3
- 230000010287 polarization Effects 0.000 claims description 3
- 238000013215 result calculation Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 24
- 238000005457 optimization Methods 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 5
- 230000007547 defect Effects 0.000 description 4
- 230000035772 mutation Effects 0.000 description 4
- 210000004205 output neuron Anatomy 0.000 description 4
- 108090000623 proteins and genes Proteins 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 3
- 230000005284 excitation Effects 0.000 description 3
- 238000011423 initialization method Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000032798 delamination Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 1
- 241000772415 Neovison vison Species 0.000 description 1
- 238000010220 Pearson correlation analysis Methods 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000009529 body temperature measurement Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000008303 genetic mechanism Effects 0.000 description 1
- 230000007614 genetic variation Effects 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 210000002364 input neuron Anatomy 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 239000010445 mica Substances 0.000 description 1
- 229910052618 mica group Inorganic materials 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000005316 response function Methods 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
- 238000003878 thermal aging Methods 0.000 description 1
- 238000002411 thermogravimetry Methods 0.000 description 1
- 238000004613 tight binding model Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000004804 winding Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/04—Ageing analysis or optimisation against ageing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Software Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Physiology (AREA)
- Genetics & Genomics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a large-scale generator insulation state evaluation method based on a genetic algorithm and a radial basis function neural network, which comprises the following steps of: 1) measuring parameters related to the insulation aging state of a stator bar of a large-scale steam turbine generator; 2) classifying and screening the parameters obtained by measurement in the step 1), and then constructing a data set by using the classified and screened parameters; 3) establishing an RBF neural network; 4) optimizing the number of hidden layers in the RBF neural network, and the center and the width of a radial basis function by using a genetic algorithm; 5) training the optimized RBF neural network by using a data set, and then updating an iteration weight by using a negative gradient descent method; 6) and (5) evaluating the insulation state of the large generator by using the RBF neural network obtained in the step 5), wherein the method can accurately evaluate the insulation state of the large generator.
Description
Technical Field
The invention relates to an insulation state evaluation method for a large generator, in particular to an insulation state evaluation method for a large generator based on a genetic algorithm and a radial basis function neural network.
Background
The large-scale steam turbine generator is one of key equipment of a power system, the operation reliability of the large-scale steam turbine generator is related to the operation stability of a power grid, high attention is paid to people all the time, and one of threats of safe operation of the large-scale steam turbine generator mainly comes from an insulation system. In the running process of the motor, the stator winding is subjected to the combined action of various factors such as electricity, heat, machinery, chemistry and the like, and the insulation performance is gradually degraded. In the case of severe insulation aging, insulation failure of the generator may be caused. Since the residual breakdown voltage is affected by many factors, the accuracy of predicting the residual breakdown voltage with a small number of parameters is conventionally low. The demand for state prediction cannot be met well.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a large-scale generator insulation state evaluation method based on a genetic algorithm and a radial basis function neural network, and the method can accurately evaluate the insulation state of a large-scale generator.
In order to achieve the purpose, the method for evaluating the insulation state of the large-scale generator based on the genetic algorithm and the radial basis function neural network comprises the following steps:
1) measuring parameters related to the insulation aging state of a stator bar of a large-scale steam turbine generator;
2) classifying and screening the parameters obtained by measurement in the step 1), and then constructing a data set by using the classified and screened parameters;
3) establishing an RBF neural network;
4) optimizing the number of hidden layers in the RBF neural network, and the center and the width of a radial basis function by using a genetic algorithm;
5) training the optimized RBF neural network by using a data set, and then updating an iteration weight by using a negative gradient descent method;
6) and (5) evaluating the insulation state of the large generator by using the RBF neural network obtained in the step 5).
The parameters related to the insulation aging state of the stator bar of the large-scale steam turbine generator comprise insulation resistance R, polarization index PI, absorption ratio DAR, dielectric loss tan, dielectric loss increment delta tan, capacitance C and capacitance increment delta C.
The method for updating the iteration weight by using the negative gradient descent method comprises the following steps:
and updating the weight of an input layer, the weight of a hidden layer and the weight of an output layer in the iterative RBF neural network by using a negative gradient descent method.
And 3) constructing the RBF neural network by using the parameters classified and screened in the step 2).
The specific operation process of the step 4) is as follows:
4a) normalizing the parameters;
4b) carrying out correlation analysis on the normalized parameters to eliminate parameters with correlation lower than a preset value;
4c) generating an initial population by using a genetic algorithm;
4d) using the population parameters to make k equal to N in the Kmeans algorithm, classifying the parameters, and scoring a classification center according to the classification result, namely the center c of the hidden layer radial basis function, wherein the width sigma of the hidden layer radial basis function is according to the classification result
Calculation of where ciAnd cjCentral vectors of the ith and jth nodes respectively;
4e) constructing an RBF training model by using the number N of the neurons of the hidden layer, the center c and the width sigma of the radial basis function, training a data set, and updating the weight by using a negative gradient descent method until the precision meets the requirement or the iteration number reaches the maximum value to obtain a trained RBF neural network;
4f) predicting the residual breakdown voltage by using the trained RBF neural network, and then calculating the fitness of each population;
4g) judging whether a preset termination condition is met, and if the preset termination condition is met, turning to the step 4 h); otherwise, carrying out selective cross variation on the population according to the respective fitness, and then turning to the step 4 e);
4h) and selecting an individual with the optimal self-adaption degree in the population as the number N of hidden layers of the RBF neural network according to the training result, and then training the data set to obtain each parameter of the RBF neural network improved by the genetic algorithm.
The invention has the following beneficial effects:
the method for evaluating the insulation state of the large generator based on the genetic algorithm and the radial basis function network optimizes the number of hidden layers in the RBF neural network and the center and width of the radial basis function by using the genetic algorithm during specific operation, solves the problem that the optimal number of the hidden layers is difficult to determine in RBF neural network prediction so as to improve the accuracy of prediction, trains the optimized RBF neural network, evaluates the insulation state of the large generator by using the trained RBF neural network, has higher evaluation accuracy, and effectively solves the problem that the insulation state of the motor is inaccurate and incomplete by depending on a single variable aging factor in the traditional method.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a system diagram of the present invention.
Detailed Description
The invention is described in further detail below with reference to the accompanying drawings:
the method aims at the problem of evaluating and predicting the insulation aging state of the stator of the large-scale generator, and solves the problem that the insulation state of the motor is not accurately and comprehensively predicted by the traditional method of depending on a univariate aging factor. Meanwhile, the problem that the optimal hidden layer number is difficult to determine in RBF neural network prediction is solved by using a genetic algorithm, so that the accuracy of prediction is improved. Meanwhile, a measuring device is selected to measure the required parameters by establishing an experiment platform. Then, the RBF neural network is trained and predicted by using the measured data, and the specific process is as follows:
radial basis function neural network
The RBF is a three-layer forward network with a single hidden layer, wherein the first layer is an input layer and consists of signal source nodes; the second layer is a hidden layer, the transformation function of the neurons in the hidden layer, namely the radial basis function, is a nonnegative linear function which is radially symmetrical and attenuated to the central point, and the radial basis function is a local response function; the third layer is an output layer and is used for responding to the input mode; the input layer only plays a role of transmitting signals, the input layer and the hidden layer can be regarded as a connecting layer with a connecting weight value of 1, and tasks completed by the output layer and the hidden layer are different, so that the learning strategies are different; the output layer adjusts the linear weight and adopts a linear optimization strategy, so that the learning speed is high; the hidden layer adjusts the parameters of the activation function (green function, gaussian function, the latter is generally adopted), and a nonlinear optimization strategy is adopted, so that the learning speed is low.
The purpose of the training is to find the final weights Cj, Dj and Wj of the two layers.
The training process is divided into two steps: the first step is unsupervised learning, and weights Cj and Dj between an input layer and a hidden layer are determined by training; and the second step is supervised learning, and training determines the weight Wj between the hidden layer and the output layer.
The input vector X, the corresponding target output vector Y and the width vector Dj of the radial basis function are provided before training.
When training set data is input for training, the expression and calculation method of each parameter is as follows:
inputting a vector X: x ═ X1，x2，...，xn]TAnd n is the number of neurons of the input layer, namely the number of input training set parameter types.
And outputting a vector Y: y ═ Y1，y2,，...，yq]TWherein q is the number of neurons in the output layer, i.e., the desired value.
Initializing weights from a hidden layer to an output layer: wK＝[wk1，wk2，...，wkp]TWhere k is (1,2, … q)
Wherein p is the neuron number of the hidden layer, and q is the neuron number of the output layer.
The initialization method of the weight from the hidden layer to the output layer by the reference center initialization method comprises the following steps:
Initializing the Central parameter C of hidden layer neuronsj＝[cj1,，cj2，...cjn]TThe initialization method comprises the following steps:
where p is the total number of neurons in the hidden layer, and j is (1,2, …, p).
mini is the minimum value of all input information of the ith characteristic in the training set, and maxi is the maximum value of all input information of the ith characteristic in the training set.
Initializing width vector Dj＝[dj1，dj2，...djn]The width vector affects the range of action of the neuron on the input information: the smaller the width, the narrower the shape of the corresponding hidden layer neuron action function, the smaller the response of the information near the center of other neurons at the neuron, and the calculation method is as follows:
df is a width adjusting coefficient, the value of df is less than 1, and the df has the effect of enabling each hidden layer neuron to easily realize the sensing capability on local information and is beneficial to improving the local response capability of the RBF neural network.
Computing output z of hidden layer neuronsj，
Wherein Cj is a central vector of a jth neuron of the hidden layer, Dj is a width vector of the jth neuron of the hidden layer, and | | · | | is a euclidean norm;
computing output Y ═ Y for neurons in the output layer1，y2，...，yq]T，
Wherein, wkjThe adjustment weight between the kth neuron of the output layer and the jth neuron of the hidden layer is obtained;
wherein, wkj(t) the adjustment weight between the kth output neuron and the jth hidden layer neuron at the time of the t iteration calculation; c. Cji(t) is the central component of the jth hidden layer neuron for the ith input neuron at the time of the tth iterative computation; dji(t) is the sum of center cji(t) a corresponding width; eta is a learning factor; e is an evaluation function of the RBF neural network, wherein,
wherein, OlkIs the expected output value of the kth output neuron at the ith input sample; y islkThe net actual output value of the kth output neuron at the ith input sample is obtained.
When E is less than the expected error, the training ends.
Genetic algorithm
Genetic Algorithm (Genetic Algorithm) is a kind of randomized search method which is evolved by the evolution law (survival of the fittest, and superior-inferior Genetic mechanism) of the biology world for reference, and is firstly proposed by professor J.Holland in the United states in 1975, and the method is mainly characterized in that the method directly operates the structural object and does not have the limitations of derivation and function continuity; the method has the advantages of inherent hidden parallelism and better global optimization capability; by adopting a probabilistic optimization method, the optimized search space can be automatically acquired and guided, the search direction can be adaptively adjusted, and no determined rule is needed. The characteristics of the genetic algorithm can well solve the problems of combination optimization and machine learning, and the genetic algorithm is a key technology in modern related intelligent computing.
The basic operation process of the genetic algorithm is as follows:
a) initialization: setting an evolution algebra counter T to be 0, setting a maximum evolution algebra T, and randomly generating M individuals as an initial population P (0);
b) individual evaluation: calculating the fitness of each individual in the population P (t);
c) selecting and operating: acting a selection operator on the population, wherein the selection is to directly inherit the optimized individuals to the next generation or generate new individuals through pairing and crossing and then inherit the new individuals to the next generation, and the selection operation is based on the fitness evaluation of the individuals in the population;
d) and (3) cross operation: acting a crossover operator on the population, wherein the crossover operator plays a core role in the genetic algorithm;
e) and (3) mutation operation: acting mutation operators on the population, namely changing the gene values of certain loci of the individual strings in the population, and obtaining a next generation population P (t +1) after selection, intersection and mutation operations of the population P (t);
f) and (4) judging termination conditions: and if T is equal to T, outputting the individual with the maximum fitness obtained in the evolution process as the optimal solution, and stopping the calculation.
Genetic algorithm is a search algorithm for solving optimization in computational mathematics, and is one of evolutionary algorithms, which is formed based on the idea of genetic variation of biological genes. The genetic algorithm is a global optimization algorithm, can avoid falling into local optimal solution and overfitting phenomenon, and can greatly improve the convergence rate of optimization through a cross variation method.
Genetic algorithms first encode and then further complete the selection, crossover and mutation of genes.
The method for optimizing the RBF neural network by the genetic algorithm comprises the following steps:
1) in order to facilitate the cross variation of genes, binary coding is adopted in the invention to generate an initial population and generate any N groups of central vectors ciAnd width parameteriThe network parameter of (2);
2) applying gradient descent method to network parameter ciAnditraining is carried out to obtain the weight w of the networkiAnd the fitness f of the population individualsComprises the following steps:
wherein E is an error function, Cmax＝max{E}；
3) Performing cross variation operation on the individuals with high fitness in the step 2);
for example, an individual a1 is randomly selected, the corresponding fitness of the individual a1 is f1, the individual a2 is randomly selected, the fitness of the individual a2 is selected, and the fitness of the individual an is fn; selecting part of individuals with higher fitness in the cross variation to perform exchange or negation on part of bits of the binary code; further generate a new binary code, i.e. a new ciAndi；
4) subjecting the new population c obtained in step 3)iAndicontinuing to train by using the step 2).
5) When the error function meets the error precision requirement, the training is stopped, and the c with the highest fitness at the moment is takeniAndias the optimal parameter value of the RBF neural network.
The state evaluation step of the stator bar by the novel RBF neural network designed after the genetic algorithm is fused is as follows:
1) and selecting an electric parameter and a non-electric parameter related to the insulation aging state of the stator bar of the large-scale steam turbine generator as the parameter type to be measured. Insulation resistance R, polarization index PI and absorption ratio DAR can well reflect insulation moisture and dirt accumulation conditions, are sensitive parameters for stator insulation moisture absorption and dirt characteristic investigation, dielectric loss tan, dielectric loss increment delta tan, capacitance C, capacitance increment delta C and the like can reflect air gap and layering size and insulation integral defects in insulation, generally, the aging degree of insulation is evaluated according to the variation trend of tan and C along with voltage, and if delta tan or delta C is close to 0, the insulation is free of defects. When the insulation contains a large number of air gaps, as the alternating voltage applied to the insulation increases, the leakage current flowing through the insulation changes abruptly with the increase of the air gap discharge, and a first current excitation point and a second current excitation point appear, so that the alternating current increase rate Delta I and the internal air gap distributionThe condition and the quantity are related, and the overall aging state of the insulation can be reflected. The insulation is damaged most seriously by partial discharge, the discharge mainly comes from air gap breakdown inside the insulation, and the maximum partial discharge quantity QmaxThe maximum local defect in the insulation is reflected, and the local aging condition of the insulation can be effectively judged. The above characteristic quantities are all found to have a correlation with the residual breakdown voltage. According to IEC standard, the insulation residual breakdown voltage U of the stator barBDCan be used to reflect the degree of insulation aging when U isBDWhen the breakdown voltage drops to half the initial breakdown voltage, the insulation is considered to have reached the end of life.
2) Measuring a bar of the large-scale turbo generator accelerated aging to obtain required parameters;
a thermal aging test is carried out on the stator bar in an oven or in a heating plate mode, a vibration exciter is applied to apply vibration excitation to the stator bar to simulate the condition of vibration of the stator bar, and voltage is applied to two ends of the bar to simulate the influence of electrical aging on the bar. Insulation resistance measurements were made using an IR tester in measuring the parameters. The IR tester includes an ohmmeter, a built-in generator for generating a high dc voltage, applying the voltage to the surface of the stator insulation and flowing current around the insulation surface, giving IR readings in ohms, then calculating PI with insulation resistance values at 1min and 10min and DAR with insulation resistance values at 30s and 60s, and a generator. The dielectric loss tan and the dielectric loss increment delta tan were measured by a dielectric loss tester. And calculating the capacitance C and the capacitance increment delta C by measuring the capacitance voltage and the capacitance current. The value of the leakage current is measured by a current transformer. Measuring the maximum partial discharge Q by using a partial discharge measuring instrumentmax. The residual breakdown voltage was measured using a voltmeter. In order to better control the temperature condition in the aging test, the invention adopts an infrared dot matrix temperature measurement method to detect the temperature. The measurement of the deformation amount may be accomplished by a displacement sensor.
3) Classifying and screening the measured parameters by using Pearson correlation analysis, removing factors which are not very large in correlation degree with the residual breakdown voltage, measuring the correlation (linear correlation) between two variables X and Y by a Pearson algorithm, wherein the value of the correlation is between-1 and 1, and the calculation formula is as follows:
where ρ isx，yFor the correlation coefficient, cov (X, Y) is the covariance of two variables,x、yrespectively the standard deviation of the two parameters, when the correlation coefficient is more than 0.7, the correlation of the two parameters is considered to be very high, otherwise, the two parameters are considered to be redundant parameters and eliminated, and in order to improve the convergence speed and accuracy of the model, the parameters selected above are normalized, namely
Wherein x, y ∈ Rn；xmin＝min(x)；xmax＝max(x)；
4) RBF neural network modeling improved by using genetic algorithm
4a) Initializing a data set
Normalizing the dielectric parameter data measured by the experiment to reduce the influence of data with different orders of magnitude on prediction and reduce the calculated amount;
4b) carrying out correlation analysis on the dielectric parameter data obtained in the step 4a) to remove parameters with low correlation;
4c) generating an initial population by using a genetic algorithm;
4d) classifying the dielectric parameter data by using the population parameter and making k equal to N in the Kmeans algorithm, and obtaining a classification center according to a classification result, namely the center c of the radial basis function of the hidden layer, wherein the width sigma can be determined according to the classification result
Calculation of where ciAnd cjCentral vectors of the ith and jth nodes, respectively;
4e) constructing an RBF training model by using the number N of the neurons of the hidden layer, the center c and the width sigma of the radial basis function, training a data set, and updating the weight by using a negative gradient descent method until the precision meets the requirement or the iteration number reaches the maximum value to obtain a trained RBF neural network;
4f) predicting the residual breakdown voltage by using the trained RBF neural network, and then calculating the fitness of each population, namely the average absolute percentage error MAPE;
4g) judging whether the conditions stop, if not, carrying out selective cross variation on the population according to the respective fitness, and then turning to the step 4e), and if so, turning to the step 4 h);
4h) selecting an individual with the optimal fitness in the population as the number N of hidden layers in the RBF neural network according to the training result, then training the data of the whole training set to obtain each parameter of the RBF neural network improved by a genetic algorithm;
4i) and predicting the residual breakdown voltage to be predicted by using a RBF neural network improved by a genetic algorithm to obtain a predicted value of the residual breakdown voltage.
The delamination phenomenon and air gap distribution of the aged mica are observed by a scanning electron microscope method and a thermogravimetric analysis method, the aging conditions of delamination and air gaps are graded and are graded into 4 grades, and the grades are shown in table 1:
TABLE 1
4g) The analytic hierarchy process is based on the problem situation and the strategy target, and summarizes all the factors according to different levels through subjective judgment and analysis of interaction and mutual inclusion relation among the problem factors to obtain a problem hierarchical structure chart. Generally speaking, a multi-level analysis structure is composed of a target layer, a standard layer and a decision scheme layer in sequence from high to low. Therefore, the problem of complexity can be converted into a multi-level problem with clear order and strong logic. After a hierarchical structure chart of the problem is established, the hierarchical analysis method should sort the relative importance among the factors in the same hierarchy, and determine the weight of the factors according to the sequence result, so as to provide a quantification standard for the final decision.
Claims (5)
1. A large generator insulation state evaluation method based on a genetic algorithm and a radial basis function neural network is characterized by comprising the following steps:
1) measuring parameters related to the insulation aging state of a stator bar of a large-scale steam turbine generator;
2) classifying and screening the parameters obtained by measurement in the step 1), and then constructing a data set by using the classified and screened parameters;
3) establishing an RBF neural network;
4) optimizing the number of hidden layers in the RBF neural network, and the center and the width of a radial basis function by using a genetic algorithm;
5) training the optimized RBF neural network by using a data set, and then updating an iteration weight by using a negative gradient descent method;
6) and (5) evaluating the insulation state of the large generator by using the RBF neural network obtained in the step 5).
2. The method for evaluating the insulation state of the large-scale generator based on the genetic algorithm and the radial basis function neural network as claimed in claim 1, wherein the parameters related to the insulation aging state of the stator bar of the large-scale steam turbine generator comprise insulation resistance R, polarization index PI, absorption ratio DAR, dielectric loss tan, dielectric loss increment delta tan, capacitance C and capacitance increment delta C.
3. The large generator insulation state evaluation method based on the genetic algorithm and the radial basis function neural network according to claim 1, wherein updating the iteration weight by using a negative gradient descent method comprises the following steps:
and updating the weight of an input layer, the weight of a hidden layer and the weight of an output layer in the iterative RBF neural network by using a negative gradient descent method.
4. The method for evaluating the insulation state of the large-scale generator based on the genetic algorithm and the radial basis function neural network as claimed in claim 1, wherein the RBF neural network is constructed in step 3) by using the parameters classified and screened in step 2).
5. The large-scale generator insulation state assessment method based on genetic algorithm and radial basis function neural network as claimed in claim 1, characterized in that, the specific operation process of step 4) is as follows:
4a) normalizing the parameters;
4b) carrying out correlation analysis on the normalized parameters to eliminate parameters with correlation lower than a preset value;
4c) generating an initial population by using a genetic algorithm;
4d) using the population parameters to make k equal to N in the Kmeans algorithm, classifying the parameters, and scoring a classification center according to the classification result, namely the center c of the hidden layer radial basis function, wherein the width sigma of the hidden layer radial basis function is according to the classification result
Calculation of where ciAnd cjCentral vectors of the ith and jth nodes respectively;
4e) constructing an RBF training model by using the number N of the neurons of the hidden layer, the center c and the width sigma of the radial basis function, training a data set, and updating the weight by using a negative gradient descent method until the precision meets the requirement or the iteration number reaches the maximum value to obtain a trained RBF neural network;
4f) predicting the residual breakdown voltage by using the trained RBF neural network, and then calculating the fitness of each population;
4g) judging whether a preset termination condition is met, and if the preset termination condition is met, turning to the step 4 h); otherwise, carrying out selective cross variation on the population according to the respective fitness, and then turning to the step 4 e);
4h) and selecting an individual with the optimal self-adaption degree in the population as the number N of hidden layers of the RBF neural network according to the training result, and then training the data set to obtain each parameter of the RBF neural network improved by the genetic algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010922999.0A CN112084709B (en) | 2020-09-04 | 2020-09-04 | Large-scale generator insulation state evaluation method based on genetic algorithm and radial basis function neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010922999.0A CN112084709B (en) | 2020-09-04 | 2020-09-04 | Large-scale generator insulation state evaluation method based on genetic algorithm and radial basis function neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112084709A true CN112084709A (en) | 2020-12-15 |
| CN112084709B CN112084709B (en) | 2024-04-05 |
Family
ID=73733127
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010922999.0A Active CN112084709B (en) | 2020-09-04 | 2020-09-04 | Large-scale generator insulation state evaluation method based on genetic algorithm and radial basis function neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112084709B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112651456A (en) * | 2020-12-31 | 2021-04-13 | 遵义师范学院 | Unmanned vehicle control method based on RBF neural network |
| CN112884234A (en) * | 2021-03-04 | 2021-06-01 | 电子科技大学 | Method for searching optimal working parameters of power module of high-power millimeter wave gyrotron traveling wave tube |
| CN115469259A (en) * | 2022-09-28 | 2022-12-13 | 武汉格蓝若智能技术有限公司 | RBF neural network-based CT error state online quantitative evaluation method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2011006344A1 (en) * | 2009-07-15 | 2011-01-20 | 北京航空航天大学 | Temperature regulating device and intelligent temperature control method for sand dust environment test system |
| CN109634121A (en) * | 2018-12-28 | 2019-04-16 | 浙江工业大学 | More parent genetic algorithm air source heat pump multiobjective optimization control methods based on radial basis function neural network |
| US20190242936A1 (en) * | 2018-02-05 | 2019-08-08 | Wuhan University | Fault diagnosis method for series hybrid electric vehicle ac/dc converter |
-
2020
- 2020-09-04 CN CN202010922999.0A patent/CN112084709B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2011006344A1 (en) * | 2009-07-15 | 2011-01-20 | 北京航空航天大学 | Temperature regulating device and intelligent temperature control method for sand dust environment test system |
| US20190242936A1 (en) * | 2018-02-05 | 2019-08-08 | Wuhan University | Fault diagnosis method for series hybrid electric vehicle ac/dc converter |
| CN109634121A (en) * | 2018-12-28 | 2019-04-16 | 浙江工业大学 | More parent genetic algorithm air source heat pump multiobjective optimization control methods based on radial basis function neural network |
Non-Patent Citations (2)
| Title |
|---|
| 朱沛恒;: "基于果蝇算法优化的概率神经网络在变压器故障诊断中的应用", 电力大数据, no. 06 * |
| 赵磊;成永红;陈小林;郭亮;: "用RBFNN评估发电机主绝缘剩余击穿电压", 高电压技术, no. 08 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112651456A (en) * | 2020-12-31 | 2021-04-13 | 遵义师范学院 | Unmanned vehicle control method based on RBF neural network |
| CN112651456B (en) * | 2020-12-31 | 2023-08-08 | 遵义师范学院 | Unmanned vehicle control method based on RBF neural network |
| CN112884234A (en) * | 2021-03-04 | 2021-06-01 | 电子科技大学 | Method for searching optimal working parameters of power module of high-power millimeter wave gyrotron traveling wave tube |
| CN115469259A (en) * | 2022-09-28 | 2022-12-13 | 武汉格蓝若智能技术有限公司 | RBF neural network-based CT error state online quantitative evaluation method and device |
| CN115469259B (en) * | 2022-09-28 | 2024-05-24 | 武汉格蓝若智能技术股份有限公司 | CT error state online quantitative evaluation method and device based on RBF neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112084709B (en) | 2024-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ma et al. | Discriminative deep belief networks with ant colony optimization for health status assessment of machine | |
| CN112084709B (en) | Large-scale generator insulation state evaluation method based on genetic algorithm and radial basis function neural network | |
| CN105930901B (en) | A kind of Diagnosis Method of Transformer Faults based on RBPNN | |
| CN112101220A (en) | Rolling bearing service life prediction method based on unsupervised model parameter migration | |
| CN113094980B (en) | Solder paste printing quality prediction method and system based on IGA-DNN | |
| CN110929763A (en) | Multi-source data fusion-based mechanical fault diagnosis method for medium-voltage vacuum circuit breaker | |
| KR102275126B1 (en) | Electrical facility safety evaluation system using artificial intelligence technique | |
| CN109030790A (en) | A kind of method for diagnosing fault of power transformer and device | |
| CN110542819A (en) | transformer fault type diagnosis method based on semi-supervised DBNC | |
| CN112326276B (en) | High-speed rail steering system fault detection LSTM method based on generation countermeasure network | |
| CN111104972A (en) | Method for identifying low-voltage risk of distribution room based on genetic algorithm optimization support vector machine multi-classifier | |
| CN112926620B (en) | Defect type identification method and system for partial discharge of transformer bushing | |
| CN111310722A (en) | Power equipment image fault identification method based on improved neural network | |
| CN111695452A (en) | Parallel reactor internal aging degree evaluation method based on RBF neural network | |
| CN116662925A (en) | Industrial process soft measurement method based on weighted sparse neural network | |
| Ferreira et al. | Frequency response analysis interpretation using numerical indices and machine learning: A case study based on a laboratory model | |
| Kuo | Artificial identification system for transformer insulation aging | |
| CN115712871A (en) | Power electronic system fault diagnosis method combining resampling and integrated learning | |
| CN118155662A (en) | Transformer voiceprint fault identification method based on artificial intelligence | |
| CN113591402A (en) | Digital power transformer health state fuzzy comprehensive evaluation method | |
| CN113642822A (en) | VAE-based sample set directional expansion method for evaluating building group structure safety | |
| CN114565051B (en) | Method for testing product classification model based on influence degree of neurons | |
| CN109490793A (en) | A kind of X-ray high voltage power supply failure prediction method based on wavelet decomposition and LSTM | |
| CN115165164A (en) | Concrete compressive stress identification method and system based on deep learning admittance characteristics | |
| Guo et al. | Application of support vector regression to predict the remaining useful life of polymerized styrene butadiene rubber of cable insulation |
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 |