Hydroelectric generating set fault identification method based on support vector machine

Abstract

The invention provides a hydroelectric generating set fault identification method based on a support vector machine, and relates to the technical field of hydropower station equipment. The method specifically comprises the following steps: acquiring vibration signal data of different parts in the hydroelectric generating set in real time by using a vibration sensor, constructing a hydroelectric generating set vibration signal data set, and preprocessing the hydroelectric generating set vibration signal data set to obtain a time sequence data set; performing dimension reduction processing on the time sequence data set by adopting a principal component analysis method to obtain a dimension-reduced time sequence data set, inputting the dimension-reduced time sequence data set into a support vector machine, and training the support vector machine by adopting a multi-strategy improved sparrow search algorithm to obtain an optimized hydroelectric generating set fault recognition model; vibration signal data of different parts in the hydroelectric generating set are collected again and preprocessed, a new time sequence data set is constructed and input into the optimized hydroelectric generating set fault recognition model for fault recognition, and a fault recognition result of the hydroelectric generating set is obtained. The accuracy and the real-time performance of fault diagnosis of the hydroelectric generating set are effectively improved.

Description

Technical Field

[0001] The invention relates to the technical field of hydropower station equipment, and in particular to a hydropower unit fault identification method based on a support vector machine. Background Art

[0002] As an important energy production equipment, the operational stability and reliability of hydropower units directly affect the safety and sustainability of energy supply. However, due to the complexity of the equipment and natural wear and tear during long-term operation, equipment failures often occur, which may lead to equipment damage, shutdown and production losses.

[0003] At present, traditional fault diagnosis methods mainly rely on experience judgment and regular inspection and maintenance, but there are several major problems: First, fault diagnosis methods that rely on manual experience and periodic inspections cannot monitor and diagnose equipment status in real time and comprehensively; second, in view of the changing working environment and equipment operating conditions, traditional fault diagnosis methods are difficult to accurately predict and identify potential fault risks; third, facing the challenges of the big data era, traditional fault diagnosis methods also have shortcomings in data processing and analysis efficiency.

[0004] To solve the above problems, intelligent fault diagnosis methods based on machine learning and data mining technology have gradually attracted attention in recent years. As a powerful supervised learning algorithm, support vector machine performs well in processing low-dimensional data and classification problems, and is widely used in the field of industrial equipment fault prediction and diagnosis. However, the application effect of traditional support vector machine methods under complex data conditions still needs to be further optimized, especially in terms of real-time performance and accuracy. Summary of the invention

[0005] In view of the deficiencies of the above-mentioned prior art, the present invention is based on support vector machine algorithm, data preprocessing technology and intelligent optimization technology, through in-depth analysis of the vibration signal data of the hydropower unit, and combined with wavelet denoising, principal component analysis and sparrow search algorithm, a hydropower unit fault identification method based on support vector machine is proposed, which aims to better discover and identify the fault status of the hydropower unit, thereby effectively improving the accuracy and real-time performance of the fault diagnosis of the hydropower unit.

[0006] The present invention proposes a method for identifying faults of hydropower units based on a support vector machine, the method comprising the following steps:

[0007] Step 1: collect vibration signal data of different parts of the hydropower unit in real time and construct a vibration signal data set of the hydropower unit. Preprocess the vibration signal data set of the hydropower unit to obtain a time series data set;

[0008] Step 2: Use the principal component analysis method to reduce the dimension of the time series data set to obtain the reduced-dimensional time series data set;

[0009] Step 3: Input the reduced-dimensional time series data set into the support vector machine, and use the multi-strategy improved sparrow search algorithm to train the support vector machine to obtain the optimized hydropower unit fault identification model;

[0010] Step 4: Recollect the vibration signal data of different parts of the hydropower unit and perform preprocessing and dimensionality reduction processing, construct a new time series data set and input it into the optimized hydropower unit fault identification model for fault identification, and obtain the fault identification result of the hydropower unit;

[0011] The vibration signal data of different parts of the hydropower unit in step 1 include: vibration signal data of the motor end bearing seat, vibration signal data of the turbine end bearing seat, vibration signal data of the generator casing, vibration signal data of the turbine casing, vibration signal data of the coupler and vibration signal data of the base;

[0012] The pretreatment method in step 1 comprises:

[0013] Step A1: using a wavelet denoising method to perform denoising on the vibration signal data in the hydropower unit vibration signal data set to obtain denoised vibration signal data;

[0014] Step A2: Use the neighboring mean method to supplement the missing values ​​of the vibration signal data after noise reduction to obtain complete vibration signal data;

[0015] Step A3: normalize the complete vibration signal data using the min-max normalization method, and construct a time series data set using the normalized vibration signal data;

[0016] The step 2 further comprises:

[0017] Step 2.1: In the time series data set, the vibration signal data of each part of the hydropower unit is taken as a group of sample data, the covariance between every two groups of sample data is calculated respectively, and the covariance matrix of the time series data set is constructed;

[0018] Step 2.2: Perform eigenvalue decomposition on the covariance matrix to obtain all eigenvalues ​​and eigenvectors of the covariance matrix, arrange all eigenvalues ​​in descending order, and select the eigenvectors corresponding to the first w eigenvalues ​​in the arrangement result as w principal components;

[0019] Step 2.3: Use the selected w principal components to project the time series data set from the original high-dimensional space to the new low-dimensional space to obtain the reduced-dimensional time series data set;

[0020] The step 3 further comprises:

[0021] Step 3.1: Randomly divide the training set from the time series data set after dimensionality reduction according to the preset ratio;

[0022] Step 3.2: Set the value range of the support vector machine penalty parameter C, the value range of the kernel function parameter g, the population size and the maximum number of iterations of the sparrow search algorithm;

[0023] Step 3.3: Use Latin hypercube sampling to generate the initial sparrow population, and each sparrow individual is an array consisting of a penalty factor C and a kernel function parameter g;

[0024] Step 3.4: Construct a fitness function based on the k-fold cross validation method;

[0025] Step 3.5: Input the initial sparrow population and the training set into the support vector machine SVM, and use the fitness function to calculate the fitness value of each individual sparrow in the initial sparrow population;

[0026] Step 3.6: Determine the explorers, followers and vigilants in the initial sparrow population according to the fitness value of each individual sparrow in the initial sparrow population, and use the initial sparrow population as the sparrow population in the current iteration round;

[0027] Step 3.7: Use the multi-strategy improved sparrow search algorithm to update the position of each individual sparrow in the sparrow population in the current iteration round, and update the ratio of explorers to followers in the sparrow population in the current iteration round to obtain an updated sparrow population;

[0028] Step 3.8: Use the fitness function to calculate the fitness value of each individual sparrow in the updated sparrow population;

[0029] Step 3.9: Determine whether the current iteration round has reached the maximum number of iterations. If so, use the globally optimal sparrow individual in the updated sparrow population as the optimal parameter of the support vector machine; if not, return to step 3.7;

[0030] Step 3.10: Construct an optimized hydropower unit fault identification model based on the optimal parameters of the support vector machine;

[0031] The calculation formula of the fitness function described in step 3.4 is:

[0032]

[0033] Where F(k,a,c,r) represents the fitness function; k is the number of validation folds; a is the a-th fold validation set; c represents correctly classified samples; e is incorrectly classified samples; s is the total number of samples; a c is the number of samples that are correctly classified in the a-fold validation set;

[0034] The process of updating the position of each individual sparrow in the sparrow population in the current iteration round described in step 3.7 includes: updating the position of the explorer, updating the position of the follower, and updating the position of the vigilant;

[0035] The method for updating the explorer's position is:

[0036]

[0037] in is the position of the i-th sparrow individual in the j-th dimension at the t-th iteration, and j = 1, 2, 3, ..., D; t is the current iteration number; D is the number of dimensions; is the position of the i-th sparrow in the j-th dimension at the t+1-th iteration; α is a random number with an amplitude range of (0,1]; R 2 is the warning value, and R 2 ∈[0,1]; ST is a safe value, and the value range of ST is [0.5,1]; Q is a random number that obeys the normal distribution; L is a 1×D matrix whose matrix elements are all 1; iter max Indicates the maximum number of iterations;

[0038] The method for updating the position of the explorer is: using the Levy flight strategy to update the position of the explorer, expressed as:

[0039]

[0040] Where X worst is the global worst position of the explorer in the current iteration round; represents the global optimal position of the explorer at the t+1th iteration; A represents a 1×D matrix, and the elements of the matrix A are {-1,1}, and satisfy A + =A T (AA T ) -1 ; n is the number of sparrows; Levy (α) is the flight step length;

[0041] The method for updating the position of the vigilant is:

[0042]

[0043] in represents the global optimal position of the sparrow population at the tth iteration; β is a random number that follows a normal distribution; f i represents the fitness value of the i-th sparrow individual; f g is the global optimal fitness value of the sparrow population; K is a random number, and K∈[-1,1]; represents the global worst position at the tth iteration; fw is the global worst comfort value of the sparrow population; ε is a constant;

[0044] The method for updating the ratio of explorers to followers in the sparrow population in the current iteration round described in step 3.7 is: using the strategy of adaptively updating the ratio of explorers to followers to update the ratio of explorers to followers in the sparrow population in the current iteration round, expressed as:

[0045]

[0046] Where r is the proportion of explorers in the sparrow population; m is the proportion coefficient; l is the deviation disturbance factor; FD is the number of explorers; SD is the number of discoverers; N p is the population size;

[0047] The step 4 further comprises:

[0048] Step 4.1: Use vibration sensors to re-collect vibration signal data from different parts of the hydropower unit and perform preprocessing and dimensionality reduction to obtain a new time series data set;

[0049] Step 4.2: Input the new time series data set into the optimized hydropower unit fault identification model for fault identification, and obtain the fault identification result of the hydropower unit;

[0050] The fault identification process is as follows: inputting a new time series data set into the optimized hydropower unit fault identification model, extracting features of any vibration signal data in the new time series data set, and comparing the extracted features with the fault feature library stored in the optimized hydropower unit fault identification model, and then determining, based on the comparison result, that a potential unit fault exists at a location corresponding to the vibration signal data in the hydropower unit or that the location corresponding to the vibration signal data in the hydropower unit is in a normal state;

[0051] Step 4.3: Perform fault diagnosis and alarm processing based on the fault identification results of the hydropower unit.

[0052] The beneficial effects of adopting the above technical solution are:

[0053] The method of the present invention collects vibration signal data of key parts of the hydropower unit, and obtains a time series data set by performing wavelet denoising, missing value supplementation and min-max normalization on the collected vibration signal data to ensure data quality.

[0054] The proposed method uses principal component analysis to reduce the dimensionality of time series data sets, that is, by calculating the covariance matrix, eigenvalue decomposition and eigenvector sorting, the data is projected into a new low-dimensional space, which achieves the extraction of the most significant features of the original data in the time series data set while reducing the data dimension, which facilitates the subsequent training and analysis of the fault identification model.

[0055] The method of the present invention optimizes the penalty factor C and the kernel parameter g of the support vector machine through a multi-strategy improved sparrow search algorithm to maximize the fault judgment accuracy of the hydropower unit fault identification model on the test set. In addition, the method of the present invention uses cross-validation to evaluate the performance of the hydropower unit fault identification model on the training set and the test set, and further improves the generalization ability and accuracy of the model by adjusting the model parameters, thereby optimizing the hydropower unit fault identification model.

[0056] The method of the present invention applies the optimized hydropower unit fault identification model to perform fault identification on the re-collected time series data set, and timely issues a fault alarm and notifies maintenance personnel to perform necessary maintenance and repair according to the fault identification result, so as to ensure the safe operation of the hydropower unit and extend the life of the equipment, provide efficient and reliable technical support for the operation and maintenance management of the hydropower station, and reduce the impact and loss of equipment failure on production.

[0057] In summary, the hydropower unit fault identification method based on support vector machine proposed in the present invention can play an important role in practical applications, improve the stability and safety of equipment operation, while reducing maintenance costs and downtime, and provide strong support for the operation and management of hydropower stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a flow chart of a method for identifying vibration faults of hydropower units based on a support vector machine in this embodiment;

[0059] Figure 2 This is a flow chart of using a multi-strategy improved sparrow search algorithm to assist support vector machine training in this implementation. DETAILED DESCRIPTION

[0060] In order to facilitate the understanding of the present application, the specific embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and embodiments. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thoroughly understood.

[0061] A method for identifying faults of hydropower units based on support vector machines in this embodiment is as follows: Figure 1 As shown, the method comprises the following steps:

[0062] Step 1: Collect vibration signal data from different parts of the hydropower unit in real time and construct a vibration signal dataset of the hydropower unit. Preprocess the vibration signal dataset of the hydropower unit to obtain a time series dataset.

[0063] The vibration signal data of different parts of the hydropower unit include: vibration signal data of the motor end bearing seat, vibration signal data of the turbine end bearing seat, vibration signal data of the generator casing, vibration signal data of the turbine casing, vibration signal data of the coupler and vibration signal data of the base.

[0064] In this embodiment, high-precision vibration sensors, such as accelerometers or vibration measuring instruments, are precisely installed at key locations such as the generator-end bearing seat, turbine-end bearing seat, generator casing, turbine casing, coupler and base of the hydropower unit to collect vibration signal data in real time and ensure data quality and reliability. By setting a suitable sampling frequency and duration, it is ensured that the acquired vibration signal data has sufficient time resolution and spectral resolution, and can accurately reflect the vibration characteristics of the equipment's operating status, thereby ensuring that the vibration conditions of the hydropower unit equipment at different locations during operation are captured.

[0065] The pretreatment method comprises:

[0066] Step A1: using a wavelet denoising method to perform denoising on the vibration signal data in the hydropower unit vibration signal data set to obtain denoised vibration signal data.

[0067] In this embodiment, the collected vibration signal data is imported into the data acquisition system for real-time monitoring and recording to obtain a vibration signal data set of the hydropower unit. In order to reduce the interference introduced by equipment operation or environmental noise, the vibration signal data is processed by wavelet transform using a wavelet noise reduction method. The application of wavelet transform can effectively separate the frequency domain components of the signal. By selecting an appropriate wavelet basis function, such as Daubechies wavelet, unnecessary high-frequency noise and low-frequency drift are removed, and the main spectral components of the vibration signal data are retained, thereby highlighting the main spectral characteristics of the vibration signal data and improving the accuracy and reliability of subsequent analysis.

[0068] In this embodiment, the wavelet denoising method is used to perform denoising on the vibration signal data of the hydropower unit. The specific process is as follows: all the collected vibration signal data are subjected to wavelet transform according to the selected wavelet basis function, and the detail coefficients and approximate coefficients of different scales are obtained by decomposing the vibration signal data into sub-bands of different frequencies. Then, for the part containing noise in the high-frequency sub-band, the soft threshold method is used to suppress the noise, that is, the high-frequency sub-band is subjected to threshold processing using the soft threshold method. Next, the processed signal is subjected to inverse wavelet transform, and the denoised vibration signal data is reconstructed by recombining the signals at multiple scales. Finally, the denoising process effectively removes the high-frequency noise in the vibration signal data of the hydropower unit and retains the main features of the vibration signal data of the hydropower unit.

[0069] Step A2: Use the nearest mean method to supplement the missing values ​​of the vibration signal data after noise reduction to obtain complete vibration signal data.

[0070] In this embodiment, if there is data missing, such as data loss due to sensor failure or communication interruption, the adjacent mean method is used to supplement the missing vibration signal data. Since the vibration signal data has been arranged in chronological order, interpolation is performed based on the known vibration signal data values ​​at adjacent time points to fill the data missing caused by sensor failure or data transmission interruption, so as to restore the complete vibration signal data and ensure that subsequent analysis and model training are not affected by data missing.

[0071] In this embodiment, the specific process of using the neighboring mean method to supplement missing values ​​in the vibration signal data after noise reduction is as follows: missing vibration signal data points are detected in the vibration signal data after noise reduction, and for each missing vibration signal data point, known vibration signal data adjacent to the vibration signal data point before and after are selected, and the average value of all selected known vibration signal data is calculated, and the calculated average value is used as the vibration signal data of the vibration signal data point, thereby restoring the continuity of the vibration signal data.

[0072] Step A3: The complete vibration signal data is normalized using the min-max normalization method, and a time series data set is constructed using the normalized vibration signal data.

[0073]

[0074] Where X norm represents the normalized vibration signal data value; X represents the original vibration signal data value; X max and X min Respectively represent the maximum and minimum values ​​in the original vibration signal data.

[0075] In this embodiment, the min-max standardization method is used to normalize the vibration signal data and map the data to a range between 0 and 1, thereby ensuring that the vibration data collected by different sensors are in the same numerical range and eliminating the data deviation caused by different dimensions of vibration signal data at different positions, so that all data can be compared and comprehensively analyzed at the same scale, so as to facilitate subsequent data analysis and model training comparison.

[0076] Step 2: Use the principal component analysis method to reduce the dimension of the time series data set to obtain the reduced-dimensional time series data set.

[0077] Step 2.1: In the time series data set, the vibration signal data of each part of the hydropower unit is taken as a group of sample data, the covariance between every two groups of sample data is calculated respectively, and the covariance matrix of the time series data set is constructed.

[0078] In this embodiment, the covariance matrix describes the linear correlation between different data features and is one of the mathematical foundations of the principal component analysis method. By calculating the covariance matrix, the strength and direction of the correlation between the data features can be understood.

[0079] Step 2.2: Perform eigenvalue decomposition on the covariance matrix to obtain all eigenvalues ​​and eigenvectors of the covariance matrix, arrange all eigenvalues ​​in descending order, and select the eigenvectors corresponding to the first w eigenvalues ​​in the arrangement result as w principal components.

[0080] In this embodiment, eigenvalue decomposition is a basic concept in linear algebra, also known as spectral decomposition, which refers to the process of decomposing a matrix into the eigenvalues ​​and eigenvectors of the matrix. Specifically, for an h×h square matrix H, where the number of rows and columns of the square matrix H are both h, if there is a reversible matrix P and a diagonal matrix G, so that the square matrix H can be expressed as P times G times the inverse of P, denoted as H=PGP-1, then this decomposition is called the eigenvalue decomposition of H.

[0081] The w principal components are expressed as:

[0082]

[0083] where Y p A represents the pth principal component, and p=1,2,…,w; qp is the eigenvector corresponding to the i-th eigenvalue in the permutation result, and q = 1, 2, …, w; Z xq is the value of the j-th dimension of the sample data x in the time series dataset.

[0084] In this embodiment, the eigenvalue represents the importance of the covariance matrix in the direction of each eigenvector, that is, the variance of each principal component in the data set, and the eigenvector describes the linear combination of these directions, that is, the direction of the data in the new coordinate system. The obtained eigenvalues ​​are arranged in descending order. Usually, the eigenvectors corresponding to the eigenvalues ​​in the front are selected as the principal components, because the eigenvalues ​​in the front can retain the variance information in the original data set to the greatest extent, that is, the main features of the data set are retained. According to the requirement of retaining the total variance, the eigenvectors corresponding to the first few principal components are selected as the new feature space. This embodiment selects to retain 80% to 95% of the total variance to ensure that the main data information is retained.

[0085] Step 2.3: Use the selected w principal components to project the time series data set from the original high-dimensional space to the new low-dimensional space to obtain the reduced-dimensional time series data set.

[0086] In this embodiment, the time series data set is projected using the selected principal component, and the original high-dimensional data set is mapped to a new low-dimensional space through a linear transformation, so as to realize the projection of the high-dimensional data into the new low-dimensional space and obtain a time series data set after dimensionality reduction. The data set after dimensionality reduction retains the main information of the original data, but reduces the dimension of the data, which helps to simplify the training and computational complexity of the subsequent model, and improve the efficiency and accuracy of the subsequent model training and analysis.

[0087] In this embodiment, the process of projecting the time series data set from the original high-dimensional space to the new low-dimensional space is: constructing a new coordinate system using the selected w principal components, i.e., w eigenvectors, and projecting the sample data in the time series data set to the new coordinate system. Specifically, the coordinates of each sample data in the original coordinate system are dot-producted with each eigenvector to obtain the coordinates of the sample data in the new coordinate system, and these coordinates are the sample data after dimensionality reduction.

[0088] Step 3: Input the reduced-dimensional time series data set into the support vector machine, and use the multi-strategy improved sparrow search algorithm to train the support vector machine to obtain the optimized hydropower unit fault identification model.

[0089] In this embodiment, if Figure 2As shown in the figure, the sparrow search algorithm is an emerging meta-heuristic optimization algorithm, which is inspired by the foraging and escaping predator behaviors of sparrows in nature. It has the advantages of simple structure, easy implementation, and few adjustment parameters, but it still has problems such as dependence on the initial population distribution, weak ability to jump out of the local optimal solution, reduced population diversity in the later iteration, and weakened global search ability. In this implementation method, Latin hypercube initialization, Levy flight strategy, and adaptive update strategy of explorer and follower ratio are introduced to the traditional sparrow search algorithm. By introducing multiple strategy improvements, the optimization performance of the sparrow search algorithm is improved and the training effect of the support vector machine is further improved.

[0090] Step 3.1: Randomly divide the training set from the reduced-dimensional time series dataset according to the preset ratio.

[0091] In this embodiment, the time series data set after dimension reduction is divided into training set D and training set D according to the ratio of 8:2. 1 and the test set D 2 , where the training set is used to train the support vector machine model, and the test set is used to evaluate the generalization ability and prediction accuracy of the model.

[0092] Step 3.2: Set the value range of the support vector machine penalty parameter C, the value range of the kernel function parameter g, the population size of the sparrow search algorithm, and the maximum number of iterations.

[0093] In this embodiment, in order to optimize the performance of the support vector machine, the penalty factor C and the kernel function parameter g of the support vector machine are used as optimization variables, and the value range of the penalty factor C and the kernel function parameter g of the support vector machine is set to (0,100], where the penalty factor C determines the degree of penalty of the model for misclassified samples, and the kernel function parameter g controls the model's adaptability to data distribution during training. The maximum number of optimization iterations is used as the optimization termination condition, and the population size of the sparrow search algorithm is set to 100 and the number of iterations is set to 100.

[0094] Step 3.3: Use Latin hypercube sampling method to generate the initial sparrow population, and each sparrow individual is an array consisting of a penalty factor C and a kernel function parameter g.

[0095] In this embodiment, the optimization ranges of the hyperparameters C and g of the support vector machine are set to (0, 100], and the population size is 100. In order to ensure the uniform distribution of the initial population in the parameter space, the Latin hypercube sampling method is used to initialize the sparrow population: first, the value range of C and g (0, 100] is equally divided into 100 sub-intervals; in each sub-interval, a value is randomly selected for C and g respectively; the sample values ​​of C and g are randomly arranged and combined to generate 100 different arrays as different sparrow individuals, forming a uniformly distributed initial sparrow population, so that the population has good diversity at the beginning of the search, which is helpful for global search.

[0096] Step 3.4: Construct a fitness function based on the k-fold cross-validation method.

[0097] In this embodiment, the performance of the model on the training set and the test set is evaluated by cross-validation technology. Cross-validation is a commonly used evaluation method, which is particularly suitable for situations where the data set is small or the generalization ability of the model is to be accurately evaluated. The basic idea of ​​cross-validation is to divide the data set into multiple subsets, such as k subsets in k-fold cross-validation, and use one of the subsets as the validation set in turn, and the rest as the training set, and repeat the training and validation process multiple times. The average accuracy of the validation is used as the fitness value of the sparrow individual.

[0098] The calculation formula of the fitness function is:

[0099]

[0100] Where F(k,a,c,r) represents the fitness function; k is the number of validation folds; a is the a-th fold validation set; c represents correctly classified samples; e is incorrectly classified samples; s is the total number of samples; a c is the number of samples that are correctly classified in the a-fold validation set.

[0101] Step 3.5: Input the initial sparrow population and the training set into the support vector machine SVM, and use the fitness function to calculate the fitness value of each individual sparrow in the initial sparrow population.

[0102] Step 3.6: Determine the explorers, followers and vigilants in the initial sparrow population according to the fitness value of each individual sparrow in the initial sparrow population, and use the initial sparrow population as the sparrow population in the current iteration round.

[0103] The method for determining the explorer is: sorting the individual sparrows in the initial sparrow population in order of fitness values ​​from high to low, and selecting a number of individual sparrows as explorers from the sorting results according to a preset ratio.

[0104] In this embodiment, the explorers are composed of sparrows with high fitness in the sparrow population. Generally, the top 20% of sparrow individuals are selected from the fitness ranking results as explorers. The explorers are responsible for extensive exploration of the search space, randomly searching for food and optimal solutions, and have a high degree of freedom, ensuring the global search capability of the algorithm and preventing it from falling into the local optimum. While looking for food sources, the explorers also provide foraging directions for the followers, so they have a wider search range.

[0105] The method for determining the vigilant ones is: randomly selecting a number of sparrow individuals from the initial sparrow population as vigilant ones according to a preset ratio.

[0106] In this embodiment, the sparrow individuals who can be aware of danger in time are called vigilant. A small proportion of sparrow individuals are randomly selected from the population as vigilant individuals, generally about 10% of the population, to monitor changes in the surrounding environment, prevent and avoid potential threats, simulate the self-protection mechanism of the group, and prevent individuals from falling into an extremely unfavorable search state.

[0107] The method for determining followers is: taking the sparrow individuals except explorers and vigilant ones in the initial sparrow population as followers.

[0108] In this embodiment, followers account for the majority of the population, about 80%. Followers are responsible for following the actions of explorers and adjusting their positions according to the choices of explorers, mainly performing local searches to help the algorithm refine the accuracy of the solution.

[0109] Step 3.7: Use the multi-strategy improved sparrow search algorithm to update the position of each individual sparrow in the sparrow population under the current iteration round, and update the ratio of explorers to followers in the sparrow population under the current iteration round to obtain the updated sparrow population.

[0110] In this embodiment, a multi-strategy improved sparrow search algorithm is used to search and optimize the hyperparameters of the support vector machine to obtain the optimal hydropower unit fault identification model. The sparrow search algorithm is a heuristic optimization algorithm that aims to help find the optimal support vector machine parameter combination within a given parameter range. In order to improve the performance of the algorithm, this embodiment introduces two improvements on the basis of the original sparrow search algorithm: the Levy flight strategy and the strategy of adaptively updating the ratio of explorers and followers. These two improvements further improve the local optimization efficiency of the algorithm while ensuring the global search capability, so that the hyperparameters C and g of the support vector machine are optimized more accurately, and finally the accuracy and robustness of the fault identification model are improved.

[0111] The process of updating the position of each individual sparrow in the sparrow population in the current iteration round includes: updating the position of the explorer, updating the position of the follower and updating the position of the vigilant.

[0112] In this embodiment, the sparrow individual refers to an array composed of the penalty factor C and the kernel parameter g of the SVM. The accuracy of SVM fault identification is closely related to the penalty factor C and the kernel parameter g. These two parameters are generally set manually based on experience, but they cannot achieve the best fault identification effect. Therefore, in this embodiment, a multi-strategy improved sparrow search algorithm is used to search for the combination of penalty factor C and kernel parameter g that can achieve the best SVM effect.

[0113] The method for updating the explorer's position is:

[0114]

[0115] in is the position of the i-th sparrow individual in the j-th dimension at the t-th iteration, and j = 1, 2, 3, ..., D; t is the current iteration number; D is the number of dimensions; is the position of the i-th sparrow in the j-th dimension at the t+1-th iteration; α is a random number with an amplitude range of (0,1]; R 2 is the warning value, and R 2 ∈[0,1]; ST is a safe value, and the value range of ST is [0.5,1]; Q is a random number that obeys the normal distribution; L is a 1×D matrix whose matrix elements are all 1; iter max Indicates the maximum number of iterations.

[0116] In this embodiment, followers can always find explorers with food sources in the entire sparrow population, and the identities of explorers and followers change dynamically; when a follower joins an explorer, a corresponding explorer becomes a follower.

[0117] The Levy flight strategy simulates the random walk process of animals foraging, which can expand the search space, increase its diversity, improve the situation of falling into local optimality, and enhance the global exploration ability.

[0118] The method for updating the position of the explorer is: using the Levy flight strategy to update the position of the explorer, expressed as:

[0119]

[0120] Where X worst is the global worst position of the explorer in the current iteration round; represents the global optimal position of the explorer at the t+1th iteration; A represents a 1×D matrix, and the elements of the matrix A are {-1,1}, and satisfy A + =A T (AA T ) -1; n is the number of sparrows; Levy (α) is the flight step length, and:

[0121]

[0122] Where α is 1.5; μ and ν follow a normal distribution.

[0123] The method for updating the ratio of explorers to followers in the sparrow population in the current iteration round is: using a strategy of adaptively updating the ratio of explorers to followers to update the ratio of explorers to followers in the sparrow population in the current iteration round, expressed as:

[0124]

[0125] Where r is the proportion of explorers in the sparrow population; m is the proportional coefficient, which is 0.8 in this implementation; l is the deviation disturbance factor; FD is the number of explorers; SD is the number of discoverers; N p is the population size, which is set to 100 in this embodiment.

[0126] In this embodiment, the ratio of explorers to followers in the population is adaptively updated according to the optimization process of the algorithm to improve the performance of the sparrow search algorithm. The ratio of explorers to followers is closely related to the performance of the sparrow search algorithm. In the initial stage, explorers dominate, focus on improving the global performance of the algorithm, and conduct extensive searches. As the iteration proceeds, the number of explorers decreases and the number of followers increases, gradually switching from global search to local search to improve convergence accuracy.

[0127] The method for updating the position of the vigilant is:

[0128]

[0129] in represents the global optimal position of the sparrow population at the tth iteration; β is a random number that follows a normal distribution; f i represents the fitness value of the i-th sparrow individual; f g is the global optimal fitness value of the sparrow population; K is a random number, and K∈[-1,1]; represents the global worst position at the tth iteration; f w is the global worst comfort value of the sparrow population; ε is a constant.

[0130] Step 3.8: Use the fitness function to calculate the fitness value of each individual sparrow in the updated sparrow population.

[0131] Step 3.9: Determine whether the current iteration round has reached the maximum number of iterations. If so, use the globally optimal sparrow individual in the updated sparrow population as the optimal parameter of the support vector machine; if not, return to step 3.7.

[0132] In this embodiment, it is determined whether the current number of iterations has reached the maximum number of iterations. If not, the updated sparrow population is updated to the sparrow population under the current iteration round and returns to step 3.7 to continue model training. Otherwise, the position information of the globally optimal sparrow individual is the optimal solution, that is, the optimal (C, g) combination.

[0133] Step 3.10: Construct an optimized hydropower unit fault identification model based on the optimal parameters of the support vector machine.

[0134] Step 4: Re-collect the vibration signal data of different parts of the hydropower unit and perform preprocessing and dimensionality reduction processing, build a new time series data set and input it into the optimized hydropower unit fault identification model for fault identification, and obtain the fault identification result of the hydropower unit.

[0135] Step 4.1: Use vibration sensors to re-collect vibration signal data from different parts of the hydropower unit and perform preprocessing and dimensionality reduction to obtain a new time series data set.

[0136] In this embodiment, the new time series data set that has been preprocessed and reduced in dimension is input into the optimized hydropower unit fault identification model. After the processing in the previous steps, these data have included the vibration signal data collected from the key parts of the hydropower unit, and have been preprocessed by wavelet denoising, missing value supplementation and normalization, as well as the principal component analysis method of dimensionality reduction, so that the data has good feature expression and a form suitable for model input. To ensure data quality and consistency.

[0137] Step 4.2: Input the new time series data set into the optimized hydropower unit fault identification model for fault identification, and obtain the fault identification result of the hydropower unit.

[0138] The fault identification process is as follows: a new time series data set is input into an optimized hydropower unit fault identification model, and for any vibration signal data in the new time series data set, feature extraction is performed on the vibration signal data, and the extracted features are compared with a fault feature library stored in the optimized hydropower unit fault identification model, and then, based on the comparison result, it is determined whether a potential unit fault exists at a location corresponding to the vibration signal data in the hydropower unit or whether the location corresponding to the vibration signal data in the hydropower unit is in a normal state.

[0139] In this embodiment, the fault identification model of the hydropower unit will perform fault identification on each sample in the new data set, and obtain the operating status of the hydropower unit by extracting features from the vibration signal data. The extracted features will be compared with the fault feature library stored in the optimized fault identification model of the hydropower unit. The fault feature library is formed during the training process and contains typical features under various fault conditions. By comparison, the fault identification model can determine whether the features of the current vibration signal data are similar to one or some features in the fault feature library. If the features of the current sample have a high degree of similarity to the features in the fault feature library, the model will determine that the sample exhibits potential unit fault features. On the contrary, if the similarity is low, the model will consider that the sample is in a normal state.

[0140] Step 4.3: Perform fault diagnosis and alarm processing based on the fault identification results of the hydropower unit.

[0141] In this embodiment, fault diagnosis and alarm processing are performed according to the output results of the fault identification model of the hydropower unit. If there is a potential unit fault in the hydropower unit, it is necessary to trigger the corresponding alarm system or notify the relevant maintenance personnel in time, and provide specific fault diagnosis information, including: fault location, fault type and fault degree. The fault location is the generator end bearing seat, turbine end bearing seat, generator casing, turbine casing, coupler, or base of the hydropower unit; the fault type includes: bearing wear, rotor imbalance, turbine blade damage, resonance, and excessive axial vibration; fault degree: the severity of the fault is evaluated according to the amplitude and frequency of the amplitude data, such as: slight, medium, and severe. Fault diagnosis information helps the operation and maintenance team to quickly locate and confirm the fault location, and to maintain and repair the fault location in time, which can effectively prevent the expansion and damage of equipment failures, thereby ensuring the safe operation of the hydropower unit and the extension of equipment life to the greatest extent. If the hydropower unit is in a normal state, there is no need to trigger the corresponding alarm system or notify the relevant maintenance personnel.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. A method for identifying faults of hydropower units based on support vector machines, characterized in that: The method comprises the following steps: Step 1: collect vibration signal data of different parts of the hydropower unit in real time and construct a vibration signal data set of the hydropower unit. Preprocess the vibration signal data set of the hydropower unit to obtain a time series data set; Step 2: Use the principal component analysis method to reduce the dimension of the time series data set to obtain the reduced-dimensional time series data set; Step 3: Input the reduced-dimensional time series data set into the support vector machine, and use the multi-strategy improved sparrow search algorithm to train the support vector machine to obtain the optimized hydropower unit fault identification model; Step 4: Re-collect the vibration signal data of different parts of the hydropower unit and perform preprocessing and dimensionality reduction processing, build a new time series data set and input it into the optimized hydropower unit fault identification model for fault identification, and obtain the fault identification result of the hydropower unit.

2. According to the method for identifying faults of hydropower units based on support vector machines according to claim 1, it is characterized in that: The vibration signal data of different parts of the hydropower unit described in step 1 include: vibration signal data of the motor end bearing seat, vibration signal data of the turbine end bearing seat, vibration signal data of the generator casing, vibration signal data of the turbine casing, vibration signal data of the coupler and vibration signal data of the base.

3. According to claim 2, a method for identifying faults of hydropower units based on support vector machines is characterized in that: The pretreatment method in step 1 comprises: Step A1: using a wavelet denoising method to perform denoising on the vibration signal data in the hydropower unit vibration signal data set to obtain denoised vibration signal data; Step A2: Use the neighboring mean method to supplement the missing values ​​of the vibration signal data after noise reduction to obtain complete vibration signal data; Step A3: The complete vibration signal data is normalized using the min-max normalization method, and a time series data set is constructed using the normalized vibration signal data.

4. The method for identifying faults of hydropower units based on support vector machines according to claim 3 is characterized in that: The step 2 further comprises: Step 2.1: In the time series data set, the vibration signal data of each part of the hydropower unit is taken as a group of sample data, the covariance between every two groups of sample data is calculated respectively, and the covariance matrix of the time series data set is constructed; Step 2.2: Perform eigenvalue decomposition on the covariance matrix to obtain all eigenvalues ​​and eigenvectors of the covariance matrix, arrange all eigenvalues ​​in descending order, and select the eigenvectors corresponding to the first w eigenvalues ​​in the arrangement result as w principal components; Step 2.3: Use the selected w principal components to project the time series data set from the original high-dimensional space to the new low-dimensional space to obtain the reduced-dimensional time series data set.

5. The method for identifying faults of hydropower units based on support vector machines according to claim 4 is characterized in that: The step 3 further comprises: Step 3.1: Randomly divide the training set from the time series data set after dimensionality reduction according to the preset ratio; Step 3.2: Set the value range of the support vector machine penalty parameter C, the value range of the kernel function parameter g, the population size and the maximum number of iterations of the sparrow search algorithm; Step 3.3: Use Latin hypercube sampling to generate the initial sparrow population, and each sparrow individual is an array consisting of a penalty factor C and a kernel function parameter g; Step 3.4: Construct a fitness function based on the k-fold cross validation method; Step 3.5: Input the initial sparrow population and the training set into the support vector machine SVM, and use the fitness function to calculate the fitness value of each individual sparrow in the initial sparrow population; Step 3.6: Determine the explorers, followers and vigilants in the initial sparrow population according to the fitness value of each individual sparrow in the initial sparrow population, and use the initial sparrow population as the sparrow population in the current iteration round; Step 3.7: Use the multi-strategy improved sparrow search algorithm to update the position of each individual sparrow in the sparrow population in the current iteration round, and update the ratio of explorers to followers in the sparrow population in the current iteration round to obtain an updated sparrow population; Step 3.8: Use the fitness function to calculate the fitness value of each individual sparrow in the updated sparrow population; Step 3.9: Determine whether the current iteration round has reached the maximum number of iterations. If so, use the globally optimal sparrow individual in the updated sparrow population as the optimal parameter of the support vector machine; if not, return to step 3.7; Step 3.10: Construct an optimized hydropower unit fault identification model based on the optimal parameters of the support vector machine.

6. The method for identifying faults of hydropower units based on support vector machines according to claim 5 is characterized in that: The calculation formula of the fitness function described in step 3.4 is: Where F(k,a,c,r) represents the fitness function; k is the number of validation folds; a is the a-th fold validation set; c represents correctly classified samples; e is incorrectly classified samples; s is the total number of samples; a c is the number of samples that are correctly classified in the a-fold validation set.

7. The method for identifying faults of hydropower units based on support vector machines according to claim 6 is characterized in that: The process of updating the position of each individual sparrow in the sparrow population in the current iteration round described in step 3.7 includes: updating the position of the explorer, updating the position of the follower, and updating the position of the vigilant; The method for updating the explorer's position is: in is the position of the i-th sparrow individual in the j-th dimension at the t-th iteration, and j = 1, 2, 3, ..., D; t is the current iteration number; D is the number of dimensions; is the position of the i-th sparrow in the j-th dimension at the t+1-th iteration; α is a random number with an amplitude range of (0,1]; R2 is the warning value, and R2∈[0,1]; ST is the safety value, and the value range of ST is [0.5,1]; Q is a random number that obeys the normal distribution; L is a 1×D matrix with all matrix elements being 1; iter max Indicates the maximum number of iterations; The method for updating the position of the explorer is: using the Levy flight strategy to update the position of the explorer, expressed as: Where X worst is the global worst position of the explorer in the current iteration round; represents the global optimal position of the explorer at the t+1th iteration; A represents a 1×D matrix, and the elements of the matrix A are {-1,1}, and satisfy A + =A T (AA T ) -1 ; n is the number of sparrows; Levy (α) is the flight step length; The method for updating the position of the vigilant is: in represents the global optimal position of the sparrow population at the tth iteration; β is a random number that follows a normal distribution; f i represents the fitness value of the i-th sparrow individual; f g is the global optimal fitness value of the sparrow population; K is a random number, and K∈[-1,1]; represents the global worst position at the tth iteration; f w is the global worst comfort value of the sparrow population; ε is a constant.

8. A method for identifying faults of hydropower units based on support vector machines according to claim 7, characterized in that: The method for updating the ratio of explorers to followers in the sparrow population in the current iteration round described in step 3.7 is: using the strategy of adaptively updating the ratio of explorers to followers to update the ratio of explorers to followers in the sparrow population in the current iteration round, expressed as: Where r is the proportion of explorers in the sparrow population; m is the proportion coefficient; l is the deviation disturbance factor; FD is the number of explorers; SD is the number of discoverers; N p For population size.

9. A method for identifying faults of hydropower units based on support vector machines according to claim 8, characterized in that: The step 4 further comprises: Step 4.1: Use vibration sensors to re-collect vibration signal data from different parts of the hydropower unit and perform preprocessing and dimensionality reduction to obtain a new time series data set; Step 4.2: Input the new time series data set into the optimized hydropower unit fault identification model for fault identification, and obtain the fault identification result of the hydropower unit; The fault identification process is as follows: inputting a new time series data set into the optimized hydropower unit fault identification model, extracting features of any vibration signal data in the new time series data set, and comparing the extracted features with the fault feature library stored in the optimized hydropower unit fault identification model, and then determining, based on the comparison result, that a potential unit fault exists at a location corresponding to the vibration signal data in the hydropower unit or that the location corresponding to the vibration signal data in the hydropower unit is in a normal state; Step 4.3: Perform fault diagnosis and alarm processing based on the fault identification results of the hydropower unit.

Images