Hydroelectric generating set fault diagnosis method based on HOA-ELM

Abstract

The invention discloses a hydroelectric generating set fault diagnosis method based on HOA-ELM. The hydroelectric generating set fault diagnosis method is specifically implemented according to the following steps: step 1, collecting vibration signals of a hydroelectric generating set under different working conditions; 2, extracting a time domain feature, a frequency domain feature, an energy entropy feature, a root mean square feature and a singular value feature of the vibration signal collected in the step 1, and putting the features into a multi-dimensional feature set; 3, converting features in the multi-dimensional feature set obtained in the step 2 into low-dimensional features; 4, training an ELM model according to an HOA algorithm to obtain an optimal hidden layer weight w and deviation b, and applying the optimal hidden layer weight w and deviation b to the ELM model to obtain an HOA-ELM diagnosis model; step 5, inputting the test set in the step 3 into the HOA-ELM diagnosis model to test the HOA-ELM diagnosis model; and step 6, performing fault diagnosis by using the tested HOA-ELM diagnosis model. According to the invention, the problem of low diagnosis accuracy in the prior art is solved.

Description

Technical Field

[0001] The invention belongs to the technical field of hydropower unit fault diagnosis methods, and relates to a hydropower unit fault diagnosis method based on HOA-ELM. Background Art

[0002] At present, as an important power generation equipment, the stability of the operating state of hydropower units has a great impact on power supply. However, due to the complexity of hydropower units and the harsh operating environment, traditional fault diagnosis methods often find it difficult to accurately identify the type and extent of faults. Therefore, it is of great practical significance to develop an efficient and accurate fault diagnosis method.

[0003] The extreme learning machine model (ELM) is a model with a simple structure and fast operation speed, which can be used to realize fast and effective classification and identification of application scenarios. However, the weights and biases of the ELM model are randomly given. Therefore, the diagnostic accuracy of the extreme learning machine model applied to the hydropower unit fault diagnosis method is low. Therefore, how to introduce the extreme learning machine model into the hydropower unit fault diagnosis and ensure the diagnostic accuracy is an urgent problem to be solved. Summary of the invention

[0004] The purpose of the present invention is to provide a hydropower unit fault diagnosis method based on HOA-ELM, which solves the problem of low diagnostic accuracy in the prior art.

[0005] The technical solution adopted by the present invention is a hydropower unit fault diagnosis method based on HOA-ELM, which is specifically implemented according to the following steps:

[0006] Step 1, collecting vibration signals of the hydropower unit under different working conditions;

[0007] Step 2, extracting the time domain features, frequency domain features, energy entropy features, root mean square features and singular value features of the vibration signal collected in step 1, and putting them into a multidimensional feature set;

[0008] Step 3, converting the features in the multidimensional feature set obtained in step 2 into independent low-dimensional features by principal component analysis to form a low-dimensional feature data set and dividing it into a training set and a test set;

[0009] Step 4: train the ELM model according to the HOA algorithm to obtain the optimal hidden layer weight w and bias b, apply the obtained hidden layer weight w and bias b to the ELM model, and obtain the HOA-ELM diagnostic model;

[0010] Step 5, input the test set obtained in step 3 into the HOA-ELM diagnostic model to test the HOA-ELM diagnostic model;

[0011] Step 6: Process the vibration signal to be diagnosed according to steps 2-3, and then input it into the tested HOA-ELM diagnostic model for fault diagnosis.

[0012] The present invention is also characterized in that:

[0013] The time domain features collected in step 2 include the kurtosis, pulse factor, root mean square value, mean, standard deviation, peak value, waveform factor, skewness, margin factor, zero crossing rate, and peak-to-peak value of the vibration signal.

[0014] The frequency domain features in step 2 include the spectral density, power spectral density, root mean square frequency, frequency variance, frequency standard deviation, center of gravity frequency, mean square frequency, frequency domain amplitude mean, frequency domain amplitude skewness, frequency domain amplitude kurtosis, frequency amplitude variance, and square root ratio of the vibration signal.

[0015] The energy entropy feature, root mean square feature and singular value feature in step 2 are specifically as follows: the vibration signal is decomposed by a dynamic mode decomposition method, and the energy entropy feature, root mean square feature and singular value feature of each decomposed modal component are extracted.

[0016] Step 3 is as follows:

[0017] Step 3.1, standardize the feature data in the multidimensional feature set, obtain the standardized feature data for the standardized data, and put it into a matrix to form a standardized data matrix;

[0018] Step 3.2, calculate the covariance matrix Z based on the standardized data matrix, specifically:

[0019]

[0020] Where X is the standardized data matrix, and n is the number of multidimensional feature data sets;

[0021] Step 3.2, calculate the eigenvalue λ and corresponding eigenvector v of the covariance matrix Z; sort the eigenvalues ​​from large to small, the corresponding eigenvectors represent the projection of the characteristic data in that direction, these directions are the principal components, when the cumulative variance contribution rate of the current k eigenvalues ​​exceeds 90%, select the first k principal components;

[0022] Step 3.3, use the selected k eigenvectors as column vectors to form a projection matrix W = [v1, v2..., v k ], where v1, v2, ..., v k are the selected 1st to kth eigenvectors;

[0023] Step 3.4, the projection matrix W transforms the standardized data matrix X into a new principal component space Y = XW, where each column of Y represents a principal component, which is the transformed low-dimensional feature data set;

[0024] Step 3.5, divide the low-dimensional feature dataset into training set and test set.

[0025] Step 4 is as follows:

[0026] Step 4.1, set the population size, maximum number of iterations, population upper bound, and population lower bound, then randomly generate the initial population. Each individual in the population represents a set of possible solutions, namely, the hidden layer weights w and bias b of the ELM model;

[0027] Step 4.2, bring the hidden layer weight w and bias b of the ELM model represented by each individual in the initial population into the ELM model, and then use the data in the training set to train the ELM model. According to the training results, obtain the classification error rate corresponding to each individual in the population;

[0028] Step 4.3, record the individual with the smallest fitness value, that is, the individual with the lowest classification error rate, as the current optimal solution;

[0029] Step 4.4, update the population, and obtain the current optimal solution corresponding to the updated population in the manner of steps 4.2-4.3, until the maximum number of iterations preset for population update is reached, and the optimal parameter weight w and deviation b are output;

[0030] In step 4.5, the optimal parameter weight w and bias b are applied to the ELM model to obtain the HOA-ELM fault diagnosis model.

[0031] The population update in step 4.4 is as follows:

[0032] According to formula W i,t =W i,t-1 +γ i,t (β best -α i,t β i,t ) and formula β i,t+1 =β i,t +W i,t Update the individual speed W in the population separately i,t With position β i,t+1 , where γ i,t is a random number in [0,1]; W i,t is the speed of traveler i at iteration t; β best is the leader's position; i,t is the position of traveler i at iteration t; α i,tis the scanning factor SF of traveler i, and SF is in [1,2] until the preset maximum number of iterations is reached, and the optimal parameter weight w and bias b are output.

[0033] The training set and the test set in step 3 both include low-dimensional feature data sets corresponding to the vibration signals used in different working conditions.

[0034] The beneficial effects of the present invention are:

[0035] The present invention optimizes the parameters of the ELM through the HOA algorithm, improves the generalization ability of the model, reduces the consumption of computing resources, and thus improves the accuracy of fault diagnosis of hydropower units. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart of a hydropower unit fault diagnosis method based on HOA-ELM of the present invention;

[0037] Figure 2 It is a flow chart of the HOA algorithm used to train the ELM model in the HOA-ELM-based hydropower unit fault diagnosis method of the present invention;

[0038] Figure 3 is a convergence curve diagram of the fitness value in the HOA optimization algorithm in Example 4 of the present invention;

[0039] Figure 4 It is a time domain diagram of four operating states in Embodiment 4 of the present invention;

[0040] Figure 5 It is a frequency domain diagram of four operating states in embodiment 4 of the present invention;

[0041] Figure 6 This is a diagram of the diagnosis results of the training set samples in Example 4 of the present invention;

[0042] Figure 7a is a diagram of the diagnosis result of fault diagnosis using the HOA-ELM model in Example 4 of the present invention;

[0043] Figure 7b is a diagram of the diagnosis result of fault diagnosis using the GWO-ELM model in Example 4 of the present invention;

[0044] Figure 7c is a diagram of the diagnosis result of fault diagnosis using the SSA-ELM model in Example 4 of the present invention;

[0045] Figure 7d This is a diagram of the diagnostic results of fault diagnosis using the ELM model in Example 4 of the present invention. DETAILED DESCRIPTION

[0046] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Example 1

[0048] The present invention is based on the HOA-ELM hydropower unit fault diagnosis method, the process is as follows Figure 1 As shown, the specific steps are as follows:

[0049] Step 1, collecting vibration signals of the hydropower unit under different working conditions;

[0050] Step 2, extracting the time domain features, frequency domain features, energy entropy features, root mean square features and singular value features of the vibration signal collected in step 1, and putting them into a multidimensional feature set;

[0051] Step 3, converting the features in the multidimensional feature set obtained in step 2 into independent low-dimensional features by principal component analysis to form a low-dimensional feature data set and dividing it into a training set and a test set;

[0052] Step 4: train the ELM model according to the HOA algorithm to obtain the optimal hidden layer weight w and bias b, apply the obtained hidden layer weight w and bias b to the ELM model, and obtain the HOA-ELM diagnostic model;

[0053] Step 5, input the test set obtained in step 3 into the HOA-ELM diagnostic model to test the HOA-ELM diagnostic model;

[0054] Step 6: Process the vibration signal to be diagnosed according to steps 2-3, and then input it into the tested HOA-ELM diagnostic model for fault diagnosis.

[0055] The Hiking Optimization Algorithm (HOA) is a metaheuristic optimization algorithm inspired by hiking experience. It was proposed by Sunday O.Oladejo et al. in 2024. The algorithm solves the optimization problem by simulating the process of hikers trying to climb mountains, hills or rocks.

[0056] The present invention uses the HOA algorithm to optimize the two parameters of the weight w and the bias b of the ELM hidden layer, thereby improving the model training effect and thus improving the fault diagnosis accuracy.

[0057] Example 2

[0058] Based on Example 1, the time domain features collected in step 2 include the kurtosis, pulse factor, root mean square value, mean, standard deviation, peak value, waveform factor, skewness, margin factor, zero crossing rate, and peak-to-peak value of the vibration signal.

[0059] The frequency domain features in step 2 include the spectral density, power spectral density, root mean square frequency, frequency variance, frequency standard deviation, center of gravity frequency, mean square frequency, frequency domain amplitude mean, frequency domain amplitude skewness, frequency domain amplitude kurtosis, frequency amplitude variance, and square root ratio of the vibration signal.

[0060] The energy entropy feature, root mean square feature and singular value feature in step 2 are specifically as follows: the vibration signal is decomposed by a dynamic mode decomposition method, and the energy entropy feature, root mean square feature and singular value feature of each decomposed modal component are extracted.

[0061] Example 3

[0062] Based on Example 2, step 3 is specifically as follows:

[0063] Step 3.1, standardize the feature data in the multidimensional feature set, obtain the labeled feature data for the standardized data, and put it into a matrix to form a standardized data matrix;

[0064] Step 3.2, calculate the covariance matrix Z based on the standardized data matrix, specifically:

[0065]

[0066] Where X is the standardized data matrix, and n is the number of multidimensional feature data sets;

[0067] Step 3.2, calculate the eigenvalue λ and corresponding eigenvector v of the covariance matrix Z; sort the eigenvalues ​​from large to small, the corresponding eigenvectors represent the projection of the characteristic data in that direction, these directions are the principal components, when the cumulative variance contribution rate of the current k eigenvalues ​​exceeds 90%, select the first k principal components;

[0068] Step 3.3, use the selected k eigenvectors as column vectors to form a projection matrix W = [v1, v2..., v k ], where v1, v2, ..., v k are the selected 1st to kth eigenvectors;

[0069] Step 3.4, the projection matrix W transforms the standardized data matrix X into a new principal component space Y = XW, where each column of Y represents a principal component, which is the transformed low-dimensional feature data set;

[0070] Step 3.5, divide the low-dimensional feature dataset into training set and test set.

[0071] like Figure 2 As shown, step 4 is specifically as follows:

[0072] Step 4.1, set the population size, maximum number of iterations, population upper bound, and population lower bound, then randomly generate the initial population. Each individual in the population represents a set of possible solutions, namely, the hidden layer weights w and bias b of the ELM model;

[0073] Step 4.2, bring the hidden layer weight w and bias b of the ELM model represented by each individual in the initial population into the ELM model, and then use the data in the training set to train the ELM model. According to the training results, obtain the classification error rate corresponding to each individual in the population;

[0074] Step 4.3, record the individual with the smallest fitness value, that is, the individual with the lowest classification error rate, as the current optimal solution;

[0075] Step 4.4, update the population, and obtain the current optimal solution corresponding to the updated population in the manner of steps 4.2-4.3, until the maximum number of iterations preset for population update is reached, and the optimal parameter weight w and deviation b are output;

[0076] In step 4.5, the optimal parameter weight w and bias b are applied to the ELM model to obtain the HOA-ELM fault diagnosis model.

[0077] The population update in step 4.4 is as follows:

[0078] According to formula W i,t =W i,t-1 +γ i,t (β best -α i,t β i,t ) and formula β i,t+1 =β i,t +W i,t Update the individual speed W in the population separately i,t With position β i,t+1 , where γ i,t is a random number in [0,1]; W i,t is the speed of traveler i at iteration t; β best is the leader's position; i,t is the position of traveler i at iteration t; α i,t is the scanning factor SF of traveler i, and SF is in [1,2] until the preset maximum number of iterations is reached, and the optimal parameter weight w and bias b are output.

[0079] The training set and the test set in step 3 both include low-dimensional feature data sets corresponding to the vibration signals used in different working conditions.

[0080] Example 4

[0081] Based on Example 3, this example uses the vibration data of a domestic hydropower unit under different working conditions to identify fault states to verify the model effect. The collected data is divided into four operating states: normal state, dynamic and static friction, thrust head looseness, and rotor misalignment. The time domain and frequency domain diagrams of each operating state are shown in Figure 1. Figure 4 , Figure 5 The sampling frequency is 1000 Hz, the sample length is 1000 points, and 100 samples are collected for each operating state. This paper divides the training set and the test set in a ratio of 7:3. The fault labels and sample data division are shown in Table 1.

[0082] Table 1

[0083] Running status Label Training set / piece Test set / piece normal 1 70 30 Dynamic and static friction 2 70 30 Thrust head loose 3 70 30 Rotor misalignment 4 70 30

[0084] In this embodiment, the population size N is set to 30, the maximum number of iterations T is set to 50, and the classification error rate of the training set is used as the fitness function. Figure 3 The figure shows the convergence process of the HOA optimization algorithm. After 13 iterations, the fitness value converges and the diagnosis results of the training set samples are as follows: Figure 6 As shown in the figure, it can be seen that the diagnostic accuracy of the training set reached 97.86%, and each operating state achieved good diagnostic accuracy, indicating that the HOA algorithm found better parameter weights w and deviations b.

[0085] In order to verify the effectiveness and superiority of the diagnostic model in this paper, the diagnostic model composed of ELM optimized by Sparrow Search Algorithm (SSA) and Grey Wolf Optimizer (GWO) is compared with the diagnostic model without optimization algorithm. The diagnostic results of the test set samples are shown in the figure. Figure 7a-7d shown.

[0086] As can be seen from the figure, the HOA-ELM diagnostic model of the present invention misdiagnoses 5 samples with an accuracy rate of 95.83%. The accuracy rates of SSA-ELM, GWO-ELM and traditional ELM diagnostic models are 93.33%, 92.50% and 90% respectively, which shows that the accuracy rates of the HOA-ELM model of the present invention exceed these models by 2.68%, 3.60% and 6.48% respectively, indicating that HOA has stronger optimization ability and faster convergence speed.

Claims

1. The fault diagnosis method of hydropower unit based on HOA-ELM is characterized by: The specific steps are as follows: Step 1, collecting vibration signals of the hydropower unit under different working conditions; Step 2, extracting the time domain features, frequency domain features, energy entropy features, root mean square features and singular value features of the vibration signal collected in step 1, and putting them into a multidimensional feature set; Step 3, converting the features in the multidimensional feature set obtained in step 2 into independent low-dimensional features by principal component analysis to form a low-dimensional feature data set and dividing it into a training set and a test set; Step 4: train the ELM model according to the HOA algorithm to obtain the optimal hidden layer weight w and bias b, apply the obtained hidden layer weight w and bias b to the ELM model, and obtain the HOA-ELM diagnostic model; Step 5, input the test set obtained in step 3 into the HOA-ELM diagnostic model to test the HOA-ELM diagnostic model; Step 6: Process the vibration signal to be diagnosed according to steps 2-3, and then input it into the tested HOA-ELM diagnostic model for fault diagnosis.

2. The fault diagnosis method of a hydropower unit based on HOA-ELM according to claim 1 is characterized in that: The time domain features collected in step 2 include the kurtosis, pulse factor, root mean square value, mean, standard deviation, peak value, waveform factor, skewness, margin factor, zero crossing rate, and peak-to-peak value of the vibration signal.

3. The fault diagnosis method of a hydropower unit based on HOA-ELM according to claim 2 is characterized in that: The frequency domain features in step 2 include the spectral density, power spectral density, root mean square frequency, frequency variance, frequency standard deviation, center of gravity frequency, mean square frequency, frequency domain amplitude average, frequency domain amplitude skewness, frequency domain amplitude kurtosis, frequency amplitude variance, and square root ratio of the vibration signal.

4. The fault diagnosis method for hydropower units based on HOA-ELM according to claim 3 is characterized in that: The energy entropy feature, root mean square feature and singular value feature in step 2 are specifically: the vibration signal is decomposed by a dynamic mode decomposition method, and the energy entropy feature, root mean square feature and singular value feature of each decomposed modal component are extracted.

5. The fault diagnosis method for hydropower units based on HOA-ELM according to claim 4 is characterized in that: The step 3 is specifically as follows: Step 3.1, standardize the feature data in the multidimensional feature set, obtain the standardized feature data for the standardized data, and put it into a matrix to form a standardized data matrix; Step 3.2, calculate the covariance matrix Z based on the standardized data matrix, specifically: Where X is the standardized data matrix, and n is the number of multidimensional feature data sets; Step 3.2, calculate the eigenvalue λ and corresponding eigenvector v of the covariance matrix Z; sort the eigenvalues ​​from large to small, the corresponding eigenvectors represent the projection of the characteristic data in that direction, these directions are the principal components, when the cumulative variance contribution rate of the current k eigenvalues ​​exceeds 90%, select the first k principal components; Step 3.3, use the selected k eigenvectors as column vectors to form a projection matrix W = [v1, v2..., v k ], where v1, v2, ..., v k are the selected 1st to kth eigenvectors; Step 3.4, the projection matrix W transforms the standardized data matrix X into a new principal component space Y = XW, where each column of Y represents a principal component, which is the transformed low-dimensional feature data set; Step 3.5, divide the low-dimensional feature dataset into training set and test set.

6. The fault diagnosis method of a hydropower unit based on HOA-ELM according to claim 5 is characterized in that: The step 4 is specifically as follows: Step 4.1, set the population size, maximum number of iterations, population upper bound, and population lower bound, then randomly generate the initial population. Each individual in the population represents a set of possible solutions, namely, the hidden layer weights w and bias b of the ELM model; Step 4.2, bring the hidden layer weight w and bias b of the ELM model represented by each individual in the initial population into the ELM model, and then use the data in the training set to train the ELM model. According to the training results, obtain the classification error rate corresponding to each individual in the population; Step 4.3, record the individual with the smallest fitness value, that is, the individual with the lowest classification error rate, as the current optimal solution; Step 4.4, update the population, and obtain the current optimal solution corresponding to the updated population in the manner of steps 4.2-4.3, until the maximum number of iterations preset for population update is reached, and the optimal parameter weight w and deviation b are output; In step 4.5, the optimal parameter weight w and bias b are applied to the ELM model to obtain the HOA-ELM fault diagnosis model.

7. The fault diagnosis method of a hydropower unit based on HOA-ELM according to claim 6 is characterized in that: The population update in step 4.4 is specifically as follows: According to formula W i,t =W i,t-1 +γ i,t (β best -α i,t β i,t ) and formula β i,t+1 =β i,t +W i,t Update the individual speed W in the population separately i,t With position β i,t+1 , where γ i,t is a random number in [0,1]; W i,t is the speed of traveler i at iteration t; β best is the leader's position; i,t is the position of traveler i at iteration t; α i,t is the scanning factor SF of traveler i, and SF is in [1,2] until the preset maximum number of iterations is reached, and the optimal parameter weight w and bias b are output.

8. The fault diagnosis method of a hydropower unit based on HOA-ELM according to claim 1 is characterized in that: The training set and the test set in step 3 both include low-dimensional feature data sets corresponding to vibration signals used in different working conditions.

Images