Water turbine main shaft working condition classification method based on data analysis

Abstract

The invention relates to the technical field of main shaft working condition classification, and provides a water turbine main shaft working condition classification method based on data analysis, comprising the following steps: collecting main shaft stress data, calculating main shaft power based on the main shaft stress data, and pre-classifying main shaft working conditions according to the main shaft power; constructing a main shaft power data set through the main shaft power, calculating an overall contour coefficient of the main shaft power data set, and respectively constructing a first objective function of a K-Means algorithm and a second objective function of a main shaft working condition classification model based on the overall contour coefficient; selecting a first optimal clustering number, an optimal Euclidean distance and an optimal Manhattan distance through a second objective function; setting the number of individuals and the total number of iterations, and carrying out iterative training on the second target function to obtain optimized individual positions; and inputting the optimized individual positions into a first objective function, performing clustering iteration on the first objective function based on the first optimal clustering number, the optimal Euclidean distance and the optimal Manhattan distance to obtain a second final clustering number, and performing optimal classification on the spindle working conditions based on the second final clustering number. According to the method, the convergence precision and the iteration speed of main shaft working condition classification are improved, the local optimum dilemma is effectively avoided, and accurate classification of the main shaft working conditions is realized.

Description

Technical Field

[0001] The invention relates to the technical field of main shaft working condition classification, and in particular to a method for classifying main shaft working conditions of a turbine based on data analysis. Background Art

[0002] At present, for the classification of spindle working conditions, traditional empirical classification methods are often too simplified and cannot fully capture the complexity of the working conditions. This single classification method affects the reliability of the classification results, thereby limiting the in-depth analysis of the spindle state and fatigue life. Although existing technologies such as the K-Means clustering algorithm have been applied to working condition classification, the algorithm is greatly affected by the initial clustering center in terms of clustering effect and efficiency, which easily leads to the generation of local optimal solutions, limiting its applicability under complex working conditions. Therefore, it is urgent to explore better clustering methods to achieve more accurate working condition classification. Summary of the invention

[0003] In view of this, the present invention proposes a method for classifying turbine main shaft operating conditions based on data analysis, which solves the problem that the prior art is only sensitive to the initial cluster center and easily produces local optimality.

[0004] The technical solution of the present invention is implemented as follows: The present invention provides a method for classifying the working conditions of a main shaft of a hydraulic turbine based on data analysis, comprising the following steps:

[0005] S1, collecting spindle stress data, calculating spindle power based on the spindle stress data, and pre-classifying the spindle working conditions according to the spindle power;

[0006] S2, constructing a spindle power data set through the spindle power, calculating the overall contour coefficient of the spindle power data set, and constructing a first objective function of a K-Means algorithm and a second objective function of a spindle operating condition classification model based on the overall contour coefficient;

[0007] S3, selecting the first optimal number of clusters, the optimal Euclidean distance, and the optimal Manhattan distance through the second objective function;

[0008] S4, setting the number of individuals and the total number of iterations, iteratively training the second objective function to obtain the optimized individual position;

[0009] S5, inputting the optimized individual position into the first objective function, performing clustering iteration on the first objective function based on the first optimal number of clusters, the optimal Euclidean distance and the optimal Manhattan distance to obtain a second final number of clusters, and performing optimal classification of the spindle operating conditions based on the second final number of clusters.

[0010] On the basis of the above technical solution, preferably, step S1 includes:

[0011] S11, installing a main shaft stress sensor on the main shaft of the turbine, and collecting original main shaft stress data through the main shaft stress sensor;

[0012] S12, preprocessing the original spindle stress data to obtain spindle stress data, wherein the preprocessing includes data filtering and normalization processing;

[0013] S13, calculating the spindle power by combining the spindle stress data with the torque angular velocity method;

[0014] S14, pre-classifying the main shaft operating condition of the turbine based on the main shaft power, wherein the pre-classification includes dividing the main shaft power into a plurality of preset power intervals, and dividing different preset power intervals into different pre-classified operating conditions.

[0015] Based on the above technical solution, preferably, step S2 includes:

[0016] S21, constructing a spindle power data set according to the spindle power data, wherein the spindle power data set includes a plurality of spindle power data points;

[0017] S22, calculating a silhouette coefficient of each spindle power data point, where the silhouette coefficient is used to evaluate the relative position of the spindle power data point in the cluster;

[0018] S23, calculating the overall silhouette coefficient of the spindle power data set based on the silhouette coefficient of each spindle power data point;

[0019] S24, constructing a first objective function of the K-Means algorithm based on the overall silhouette coefficient, wherein the first objective function is used to optimize the clustering effect;

[0020] S25, constructing a second objective function of the spindle operating condition classification model based on the overall silhouette coefficient, wherein the second objective function is used to optimize clustering parameters.

[0021] On the basis of the above technical solution, preferably, step S3 includes:

[0022] S31, optimizing clustering parameters by using the second objective function, calculating the overall silhouette coefficient under different numbers of clusters, and selecting the optimal number of clusters based on the clustering effect;

[0023] S32, calculate the best Euclidean distance and the best Manhattan distance for the optimal number of clusters as the distance metric for the K-Means algorithm;

[0024] S33, inputting the first optimal number of clusters, the optimal Euclidean distance, and the optimal Manhattan distance into the K-Means algorithm.

[0025] Based on the above technical solution, preferably, step S4 includes:

[0026] S41, setting the number of individuals and the total number of iterations, where the number of individuals is the number of individuals of the candidate solutions used in the optimization process;

[0027] S42, initializing the position of the population, using a population initialization strategy, wherein the population initialization strategy includes random initialization, chaotic mapping, and random reverse learning;

[0028] S43, calculating the fitness value of each individual under the second objective function to evaluate its clustering effect;

[0029] S44, sorting according to fitness values, and selecting the individual with the highest fitness value as the current optimal solution;

[0030] S45, executing an optimization iteration process, wherein the optimization iteration process includes an exploration phase, an exploration phase, and an exploitation phase, and updating the position of the individual to find a better solution of the current optimal solution;

[0031] S46, increasing population diversity through individual mutation strategy and updating individual positions;

[0032] S47, after each iteration, determine whether the set number of iterations has been reached. If the set number of iterations has not been reached, return to step S43 to continue iterating. If the set number of iterations has been reached, output the current optimal optimized individual position.

[0033] On the basis of the above technical solution, preferably, step S4 further includes:

[0034] The calculation formula for the random initialization is:

[0035] χ i :x ij =ll j +r·(ul j -ll j ),i=1,2,...,N,j=1,2,...,M;

[0036] Among them, χ i is the position of the i-th candidate solution, x ij is the position coordinate of the i-th candidate solution, ll j and j denote the lower and upper bounds of the j-th decision variable, r is a random value between 0 and 1, N is the number of candidate solutions, and M is the number of decision variables;

[0037]

[0038] The calculation formula of the chaotic map is:

[0039]

[0040] Among them, X k+1 is the number of individuals after the kth chaotic mapping, X k is the number of individuals before the kth chaotic mapping, mod(·) is the modulus function;

[0041] The calculation formula of the random reverse learning is:

[0042]

[0043] in, For X k The reverse solution, U b and L b are the upper and lower boundaries of the problem to be optimized, and J is a random value between 0 and 1.

[0044] On the basis of the above technical solution, preferably, the calculation formula of the individual variation strategy is:

[0045] G hippo =D hippo [1+aC(0,σ 2 )+bG(0,σ 2 )∑;

[0046]

[0047] Among them, G hippo is the position of the individual with the highest fitness after mutation, D hippo is the position of the individual with the highest fitness before mutation, C(0,σ 2 ) is a random quantity of Cauchy distribution, G(0,σ 2 ) is the Gaussian distribution random quantity, a and b are the Cauchy distribution dynamic parameter and Gaussian distribution dynamic parameter respectively, is the current individual, f(·) is the fitness function, exp(·) is the exponential function, and σ is the standard deviation.

[0048] Based on the above technical solution, preferably, step S5 includes:

[0049] S51, inputting the first optimal number of clusters, the optimal Euclidean distance and the optimal Manhattan distance into a K-Means algorithm;

[0050] S52, performing K-Means clustering iterative calculation on the spindle power data set based on the input parameters;

[0051] S53, in the clustering iteration process, dynamically updating the cluster center of each data point until the clustering result converges to obtain the second best number of clusters;

[0052] S54: finally classify the spindle operating conditions according to the second optimal number of clusters, and output the classification results.

[0053] Based on the above technical solution, preferably, step S52 includes:

[0054] At the beginning of each iteration, the distance between the current cluster center and each data point is calculated, and the data point is assigned to the nearest cluster center; at the end of each iteration, the mean of each cluster center is calculated to update the position of the cluster center, and the clustering results of each iteration and the position changes of the cluster center are recorded.

[0055] Based on the above technical solution, preferably, step S54 includes:

[0056] According to the second best number of clusters, the silhouette coefficient of each cluster is calculated to evaluate the clustering effect; the clustering results are compared with the preset working condition standards, and a classification report is generated to point out the characteristics of each working condition category; when outputting the classification results, the sample number and main statistical characteristics of each working condition category are provided, and the working condition classification is performed based on the sample number and the main statistical characteristics.

[0057] The method for classifying the working conditions of a hydraulic turbine main shaft based on data analysis of the present invention has the following beneficial effects compared with the prior art:

[0058] (1) By collecting spindle stress data and calculating spindle power, the spindle working conditions are pre-classified, the spindle power data set is constructed, and the overall silhouette coefficient is calculated to optimize the objective function and classification model of the K-Means algorithm, thereby improving the clustering effect and parameter accuracy. By optimizing the objective function, the optimal number of clusters and distance metric are selected, and iterative training is performed in combination with the population optimization algorithm. The individual mutation strategy is introduced to increase the diversity of the population, thereby improving the convergence accuracy and iteration speed, which can effectively avoid falling into the local optimal dilemma and ultimately achieve accurate classification of the spindle working conditions.

[0059] (2) By setting the number of individuals and the total number of iterations, random initialization, chaotic mapping and random reverse learning are used to initialize the population position. By calculating the fitness value of each individual under the second objective function, the clustering effect is evaluated and the individual with the highest fitness value is selected as the current optimal solution. The exploration, exploration and exploitation stages in the optimization iteration process are performed, which helps to continuously update the individual position and find a better solution. At the same time, the individual mutation strategy is used to increase the population diversity and improve the clustering optimization effect.

[0060] (3) By inputting the optimal clustering parameters into the K-Means algorithm, clustering is performed under the optimal conditions to ensure the accuracy of the clustering results. The cluster center of each data point is dynamically updated during the clustering iteration until the clustering results converge, and a more stable and reliable second optimal number of clusters is obtained, laying the foundation for the final working condition classification. The clustering results of each iteration and the position changes of the cluster center are recorded to provide a basis for the subsequent clustering result analysis, which is helpful to evaluate the clustering effect;

[0061] (4) The generated classification report points out the characteristics of each operating condition category, provides the sample number and main statistical characteristics, and enhances the interpretability of the classification results. The operating condition classification is performed based on the sample number and statistical characteristics, and the operating condition information contained in the data is fully mined, providing a reliable basis for the intelligent management of the turbine. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0063] Figure 1 The present invention is a flow chart of a method for classifying turbine main shaft operating conditions based on data analysis. DETAILED DESCRIPTION

[0064] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] See also Figure 1 , a method for classifying operating conditions of a turbine main shaft based on data analysis is provided, comprising the following steps:

[0066] S1, collecting spindle stress data, calculating spindle power based on the spindle stress data, and pre-classifying the spindle working conditions according to the spindle power;

[0067] S2, constructing a spindle power data set through the spindle power, calculating the overall contour coefficient of the spindle power data set, and constructing a first objective function of a K-Means algorithm and a second objective function of a spindle operating condition classification model based on the overall contour coefficient;

[0068] S3, selecting the first optimal number of clusters, the optimal Euclidean distance, and the optimal Manhattan distance through the second objective function;

[0069] S4, setting the number of individuals and the total number of iterations, iteratively training the second objective function to obtain the optimized individual position;

[0070] S5, inputting the optimized individual position into the first objective function, performing clustering iteration on the first objective function based on the first optimal number of clusters, the optimal Euclidean distance and the optimal Manhattan distance to obtain a second final number of clusters, and performing optimal classification of the spindle operating conditions based on the second final number of clusters.

[0071] Specifically, this embodiment collects spindle stress data and calculates spindle power to achieve pre-classification of spindle operating conditions, constructs a spindle power data set, and calculates the overall silhouette coefficient, so as to optimize the objective function and classification model of the K-Means algorithm, thereby improving the clustering effect and parameter accuracy. By optimizing the objective function, selecting the optimal number of clusters and distance metric, combining the population optimization algorithm for iterative training, and introducing an individual mutation strategy to increase the diversity of the population, the convergence accuracy and iteration speed are improved, which can effectively avoid falling into the local optimal dilemma and ultimately achieve accurate classification of the spindle operating conditions.

[0072] Step S1 includes:

[0073] S11, installing a main shaft stress sensor on the main shaft of the turbine, and collecting original main shaft stress data through the main shaft stress sensor;

[0074] S12, preprocessing the original spindle stress data to obtain spindle stress data, wherein the preprocessing includes data filtering and normalization processing;

[0075] S13, calculating the spindle power by combining the spindle stress data with the torque angular velocity method;

[0076] S14, pre-classifying the main shaft operating condition of the turbine based on the main shaft power, wherein the pre-classification includes dividing the main shaft power into a plurality of preset power intervals, and dividing different preset power intervals into different pre-classified operating conditions.

[0077] In a specific embodiment, the torque angular velocity method calculates power by the formula P=T·ω, where P is power, T is torque, and ω is angular velocity. According to the calculated spindle power, it is divided into multiple preset power intervals (such as low power, medium power, and high power), and different power intervals correspond to different pre-classified working conditions (such as normal working conditions, slightly abnormal working conditions, and severely abnormal working conditions).

[0078] Specifically, this embodiment improves the accuracy and reliability of collected data by installing a stress sensor on the main shaft of the turbine and performing data filtering and normalization processing. The torque angular velocity method is used to calculate the main shaft power, which can more accurately reflect the actual operating status of the turbine under different working conditions and ensure the authenticity and validity of the power data. By dividing the main shaft power into multiple preset power intervals, the operating status of the turbine can be quickly identified, laying the foundation for achieving accurate working condition classification.

[0079] Step S2 includes:

[0080] S21, constructing a spindle power data set according to the spindle power data, wherein the spindle power data set includes a plurality of spindle power data points;

[0081] S22, calculating a silhouette coefficient of each spindle power data point, where the silhouette coefficient is used to evaluate the relative position of the spindle power data point in the cluster;

[0082] S23, calculating the overall silhouette coefficient of the spindle power data set based on the silhouette coefficient of each spindle power data point;

[0083] S24, constructing a first objective function of the K-Means algorithm based on the overall silhouette coefficient, wherein the first objective function is used to optimize the clustering effect;

[0084] S25, constructing a second objective function of the spindle operating condition classification model based on the overall silhouette coefficient, wherein the second objective function is used to optimize clustering parameters.

[0085] In a specific embodiment, a spindle power data set is constructed based on the collected spindle power data, and the data set includes a plurality of spindle power data points, each data point representing a spindle power value in a specific time period.

[0086] Calculate the silhouette coefficient for each spindle power data point. The calculation formula for the silhouette coefficient is:

[0087]

[0088] Among them, s(m) is the silhouette coefficient of the mth data point, B(m) is the average distance from the mth data point to the nearest other cluster, and A(m) is the average distance from the mth data point to other points in the same cluster. The silhouette coefficient is used to evaluate the relative position of each data point in the cluster. The closer the value is to 1, the better the clustering effect.

[0089] Based on the silhouette coefficient of each spindle power data point, the overall silhouette coefficient of the spindle power data set is calculated. The overall silhouette coefficient is the average of the silhouette coefficients of all data points, reflecting the clustering effect of the entire data set.

[0090] The first objective function of the K-Means algorithm is constructed based on the overall silhouette coefficient. The first objective function is in the form of:

[0091]

[0092] Among them, O 1 is the first objective function, Q is the number of data points, and the first objective function aims to maximize the overall silhouette coefficient to optimize the clustering effect.

[0093] The second objective function of the spindle working condition classification model is constructed based on the overall contour coefficient. The second objective function is in the form of:

[0094]

[0095] Among them, O 2 is the second objective function, C l is the lth cluster, μ l is the cluster center, d(x,μ l ) is the distance from the data point to the cluster center, and the second objective function is used to optimize the clustering parameters.

[0096] Specifically, this embodiment systematically organizes the collected power data by constructing a spindle power data set, and can effectively evaluate the relative position of the data point in the cluster by calculating the silhouette coefficient of each data point, thereby providing a basis for selecting the optimal number of clusters and optimizing the clustering effect. By calculating the overall silhouette coefficient and constructing the objective function, it is possible to quantify the clustering effect and ensure that parameter optimization in the clustering process is based on actual data performance, thereby improving the accuracy and reliability of clustering.

[0097] Step S3 includes:

[0098] S31, optimizing clustering parameters by using the second objective function, calculating the overall silhouette coefficient under different numbers of clusters, and selecting the best number of clusters based on the clustering effect,

[0099] S32, calculate the best Euclidean distance and the best Manhattan distance for the optimal number of clusters as the distance metric for the K-Means algorithm;

[0100] S33, inputting the first optimal number of clusters, the optimal Euclidean distance, and the optimal Manhattan distance into the K-Means algorithm.

[0101] In a specific embodiment, clustering parameters are optimized by a second objective function. First, a range of cluster numbers is set (e.g., from 2 to 10), and then K-Means clustering is performed on each cluster number to calculate the corresponding overall silhouette coefficient. By comparing the overall silhouette coefficients under different cluster numbers, the cluster number with the highest silhouette coefficient is selected as the first optimal cluster number.

[0102] After determining the optimal number of clusters, the optimal Euclidean distance and the optimal Manhattan distance under the optimal number of clusters are further calculated. This can be done by statistically analyzing the distance between the sample points of each cluster and the cluster center, and selecting the distance metric that optimizes the clustering effect.

[0103] The first optimal number of clusters, optimal Euclidean distance, and optimal Manhattan distance are input as parameters into the K-Means algorithm.

[0104] Specifically, this embodiment can effectively identify the optimal number of clusters by calculating the overall silhouette coefficient under different numbers of clusters, ensuring that the clustering results can truly reflect the intrinsic structure of the data. By calculating the optimal Euclidean distance and the optimal Manhattan distance, it can provide a suitable data metric for the K-Means algorithm to ensure the adaptability of the clustering algorithm under different data distributions. The optimized clustering parameters are input into the K-Means algorithm to ensure the systematicness and effectiveness of the entire classification process.

[0105] Step S4 includes:

[0106] S41, setting the number of individuals and the total number of iterations, where the number of individuals is the number of individuals of the candidate solutions used in the optimization process;

[0107] S42, initializing the position of the population, using a population initialization strategy, wherein the population initialization strategy includes random initialization, chaotic mapping, and random reverse learning;

[0108] S43, calculating the fitness value of each individual under the second objective function to evaluate its clustering effect;

[0109] S44, sorting according to fitness values, and selecting the individual with the highest fitness value as the current optimal solution;

[0110] S45, executing an optimization iteration process, wherein the optimization iteration process includes an exploration phase, an exploration phase, and an exploitation phase, and updating the position of the individual to find a better solution of the current optimal solution;

[0111] S46, increasing population diversity through individual mutation strategy and updating individual positions;

[0112] S47, after each iteration, determine whether the set number of iterations has been reached. If the set number of iterations has not been reached, return to step S43 to continue iterating. If the set number of iterations has been reached, output the current optimal optimized individual position.

[0113] The calculation formula for the random initialization is:

[0114] χ i :x ij=ll j +r·(ul j -ll j ),i=1,2,...,N,j=1,2,...,M;

[0115] Among them, χ i is the position of the i-th candidate solution, x ij is the position coordinate of the i-th candidate solution, ll j and j denote the lower and upper bounds of the j-th decision variable, r is a random value between 0 and 1, N is the number of candidate solutions, and M is the number of decision variables;

[0116]

[0117] The calculation formula of the chaotic map is:

[0118]

[0119] Among them, X k+1 is the number of individuals after the kth chaotic mapping, X k is the number of individuals before the kth chaotic mapping, mod(·) is the modulus function;

[0120] The calculation formula of the random reverse learning is:

[0121]

[0122] in, For X k The reverse solution, U b and L b are the upper and lower boundaries of the problem to be optimized, and J is a random value between 0 and 1.

[0123] The calculation formula for the exploration stage is:

[0124]

[0125] in, For the current individual, is the position coordinate of the current individual, D hippo is the position of the individual with the highest fitness before mutation, χ i is the current individual position, is the exploration stage i Location Update, I 1 is an integer between 1 and 2, y 1 is a random number between 0 and 1, F i is the objective function value.

[0126] The calculation formula for the exploration phase is:

[0127]

[0128] Among them, Predator j is the location of the predator, is a random vector with a modulus between 0 and 1, is the position of the individual when facing the predator, is a random vector following Levy distribution, β is a uniform random number between 2 and 4, c is a uniform random number between 1 and 1.5, d is a uniform random number between 2 and 3, and g is a uniform random number between -1 and 1. and is a random vector of dimension 1*m, For the exploration phase i location updates.

[0129] The calculation formula for the exploitation stage is:

[0130]

[0131]

[0132] Among them, t is the current iteration number, T is the maximum iteration number, is the new upper bound for the variable decision, is the new lower bound for the variable decision, To find the individual location of the nearest safe place, r 3 is a random number between 0 and 1, ζ 1 is a selected random vector or random number, is a random vector with a modulus between 0 and 1, r 12 is a normally distributed random number, r 13 is a random number between 0 and 1, Exploitation stage i location updates.

[0133] In a specific embodiment, the number of individuals and the total number of iterations are set. Assuming the number of individuals is 50, it means that 50 candidate solutions are used in the optimization process; the total number of iterations is set to 100, which means that the algorithm will find the optimal solution in 100 iterations.

[0134] Initialize the position of the population. Use a random initialization strategy to generate 50 candidate solution positions, and the decision variable range of each candidate solution is between the preset lower and upper bounds.

[0135] Calculate the fitness value of each individual under the second objective function. The fitness value is used to evaluate the performance of each candidate solution in clustering effect. The higher the fitness value, the better the clustering effect.

[0136] Sort by fitness value and select the individual with the highest fitness value as the current optimal solution. Record the position and fitness value of the current optimal solution.

[0137] The optimization iteration process is performed. The process includes the exploration phase, the exploration phase and the exploitation phase, updating the position of the individual to find a better solution for the current optimal solution.

[0138] The individual mutation strategy is used to increase population diversity and update the position of individuals. The mutation strategy can include random mutation, crossover and other methods to prevent the algorithm from falling into the local optimum.

[0139] After each iteration, it is determined whether the set number of iterations has been reached. If the set number of iterations has not been reached, the process returns to step S43 to continue the iteration; if the set number of iterations has been reached, the currently optimal optimized individual position is output.

[0140] Specifically, this embodiment ensures that the optimization process has sufficient search space and time to increase the possibility of finding the global optimal solution by reasonably setting the number of individuals and the number of iterations. The random initialization strategy can effectively increase the diversity of the population and avoid the centralization of the initial solution, thereby improving the exploration ability of the algorithm. By calculating the fitness value, the clustering effect of each candidate solution can be accurately evaluated. Through the exploration, exploration and exploitation stages, the algorithm can effectively balance between the global and local. Through the individual mutation strategy, the diversity of the population can be maintained, the algorithm can be prevented from falling into the local optimal solution, and the accuracy and reliability of the final clustering result can be improved.

[0141] The calculation formula of the individual variation strategy is:

[0142] G hippo =D hippo [1+aC(0,σ 2 )+bG(0,σ 2 )];

[0143]

[0144] Among them, G hippo is the position of the individual with the highest fitness after mutation, D hippo is the position of the individual with the highest fitness before mutation, C(0,σ 2 ) is a random quantity of Cauchy distribution, G(0,σ 2 ) is the Gaussian distribution random quantity, a and b are the Cauchy distribution dynamic parameter and Gaussian distribution dynamic parameter respectively, is the current individual, f(·) is the fitness function, exp(·) is the exponential function, and σ is the standard deviation.

[0145] In a specific embodiment, during the optimization process, in order to increase the diversity of the population and prevent the algorithm from falling into a local optimal solution, an individual mutation strategy is used to mutate the individual with the highest fitness.

[0146] Cauchy distribution random quantity C(0,σ 2 ) and Gaussian distribution random quantity G(0,σ 2 ) can be generated by a random number generator to ensure the randomness and diversity of each mutation.

[0147] The individual position G after mutation hippo Used for fitness evaluation and optimization iterations.

[0148] After mutation, the fitness value of the mutated individual is recalculated to evaluate its performance in clustering effect. If the fitness value of the mutated individual is higher than the current optimal solution, the current optimal solution is updated.

[0149] Specifically, this embodiment can effectively increase the diversity of the population by mutating the individuals with the highest fitness, avoid the algorithm from falling into the local optimal solution, and thus improve the global search capability. The mutation strategy enables the algorithm to explore new solution spaces and find better solutions. By using random quantities of Cauchy distribution and Gaussian distribution, the mutation strategy has dynamic adaptability and can adjust the mutation amplitude according to the current state of the population. Through appropriate mutation, the convergence process of the algorithm can be accelerated, making the final clustering result more stable and reliable.

[0150] Step S5 includes:

[0151] S51, inputting the first optimal number of clusters, the optimal Euclidean distance and the optimal Manhattan distance into a K-Means algorithm;

[0152] S52, performing K-Means clustering iterative calculation on the spindle power data set based on the input parameters;

[0153] S53, in the clustering iteration process, dynamically updating the cluster center of each data point until the clustering result converges to obtain the second best number of clusters;

[0154] S54: finally classify the spindle operating conditions according to the second optimal number of clusters, and output the classification results.

[0155] Step S52 includes:

[0156] At the beginning of each iteration, the distance between the current cluster center and each data point is calculated, and the data point is assigned to the nearest cluster center; at the end of each iteration, the mean of each cluster center is calculated to update the position of the cluster center, and the clustering results of each iteration and the position changes of the cluster center are recorded.

[0157] Step S54 includes:

[0158] According to the second best number of clusters, the silhouette coefficient of each cluster is calculated to evaluate the clustering effect; the clustering results are compared with the preset working condition standards, and a classification report is generated to point out the characteristics of each working condition category; when outputting the classification results, the sample number and main statistical characteristics of each working condition category are provided, and the working condition classification is performed based on the sample number and the main statistical characteristics.

[0159] In a specific embodiment, the first optimal number of clusters, the optimal Euclidean distance and the optimal Manhattan distance are input into the K-Means algorithm. These parameters are obtained by step S3 to ensure that the K-Means algorithm can run under optimal conditions.

[0160] Based on the input parameters, K-Means clustering iterative calculation is performed on the spindle power data set. The specific steps are as follows:

[0161] Initialize cluster centers: Randomly select the first best number of clustering data points as the initial cluster centers.

[0162] The iterative process includes:

[0163] At the beginning of each iteration, the distance between the current cluster center and each data point is calculated (using the best Euclidean distance and the best Manhattan distance).

[0164] Assign each data point to the nearest cluster center.

[0165] At the end of each iteration, the mean of each cluster is calculated to update the location of the cluster center.

[0166] Record the clustering results of each iteration and the position changes of the cluster centers.

[0167] In the clustering iteration process, the cluster center of each data point is dynamically updated until the clustering result converges. The convergence criterion can be that the change in the location of the cluster center is less than a preset threshold, or the maximum number of iterations is reached.

[0168] According to the second best number of clusters, the spindle working conditions are finally classified and the classification results are output. The output classification results include:

[0169] The silhouette coefficient of each cluster is used to evaluate the clustering effect.

[0170] The clustering results are compared with the preset working condition standards to generate a classification report, pointing out the characteristics of each working condition category and providing the sample quantity and main statistical characteristics of each working condition category.

[0171] Specifically, the clustering results are compared with the preset working condition standards, a classification report is generated, the characteristics of each working condition category are pointed out, and the sample quantity and main statistical characteristics of each working condition category are provided, which includes the following steps:

[0172] The clustering results are compared with the preset working condition standards to generate a classification report. The classification report points out the characteristics of each working condition category, such as the statistical characteristics of the spindle power, the working condition duration, etc., providing detailed information for the working condition classification.

[0173] When outputting the classification results, the sample quantity and main statistical features of each working condition category are provided, such as mean value, standard deviation, peak value, etc. These statistical features can be used as the basis for working condition classification.

[0174] The spindle working conditions are finally classified according to the number of samples and main statistical characteristics of each working condition category. For example, the power threshold can be set to divide the power data into different working condition categories; or the working conditions can be divided into stable, fluctuating, etc. according to the power fluctuation characteristics.

[0175] The final classification results include the characteristic description of each operating condition category, sample quantity statistics, and corresponding main shaft power data. This information can provide a basis for turbine operation monitoring, fault diagnosis, and maintenance decision-making.

[0176] Specifically, this embodiment can perform clustering under optimal conditions by inputting the first optimal number of clusters, the optimal Euclidean distance and the optimal Manhattan distance into the K-Means algorithm, thereby ensuring the accuracy and effectiveness of the clustering results. During the clustering iteration process, the cluster center of each data point is dynamically updated to better adapt to data changes. By obtaining the second optimal number of clusters, the spindle operating conditions can be optimally classified.

[0177] At the beginning of each iteration, the distance between the current cluster center and each data point is calculated to ensure that the data points are accurately assigned to the nearest cluster center. At the end of each iteration, the position of the cluster center is updated by calculating the mean of each cluster center, which can make the cluster center gradually move to the concentrated area of ​​​​the data points and enhance the convergence of clustering. Recording the clustering results of each iteration and the position changes of the cluster center is helpful to analyze the evolution of the clustering process.

[0178] By calculating the silhouette coefficient of each cluster, evaluating the clustering effect, and comparing the clustering results with the preset working condition standards, a detailed classification report can be generated, which can clearly point out the characteristics of each working condition category. When outputting the classification results, the sample quantity and main statistical characteristics of each working condition category are provided, which can provide richer information support for working condition classification. Working condition classification based on sample quantity and main statistical characteristics can fully mine the working condition information contained in the data, and provide reliable data support for intelligent management and maintenance decisions of turbines.

[0179] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for classifying turbine main shaft working conditions based on data analysis, characterized in that: The following steps are involved: S1, collecting spindle stress data, calculating spindle power based on the spindle stress data, and pre-classifying the spindle working conditions according to the spindle power; S2, constructing a spindle power data set through the spindle power, calculating the overall contour coefficient of the spindle power data set, and constructing a first objective function of a K-Means algorithm and a second objective function of a spindle operating condition classification model based on the overall contour coefficient; S3, selecting the first optimal number of clusters, the optimal Euclidean distance, and the optimal Manhattan distance through the second objective function; S4, setting the number of individuals and the total number of iterations, iteratively training the second objective function to obtain the optimized individual position; S5, inputting the optimized individual position into the first objective function, performing clustering iteration on the first objective function based on the first optimal number of clusters, the optimal Euclidean distance and the optimal Manhattan distance to obtain a second final number of clusters, and performing optimal classification of the spindle operating conditions based on the second final number of clusters.

2. A method for classifying turbine main shaft operating conditions based on data analysis as claimed in claim 1, characterized in that: Step S1 includes: S11, installing a main shaft stress sensor on the main shaft of the turbine, and collecting original main shaft stress data through the main shaft stress sensor; S12, preprocessing the original spindle stress data to obtain spindle stress data, wherein the preprocessing includes data filtering and normalization processing; S13, calculating the spindle power by combining the spindle stress data with the torque angular velocity method; S14, pre-classifying the main shaft operating condition of the turbine based on the main shaft power, wherein the pre-classification includes dividing the main shaft power into a plurality of preset power intervals, and dividing different preset power intervals into different pre-classified operating conditions.

3. A method for classifying turbine main shaft operating conditions based on data analysis as claimed in claim 2, characterized in that: Step S2 includes: S21, constructing a spindle power data set according to the spindle power data, wherein the spindle power data set includes a plurality of spindle power data points; S22, calculating a silhouette coefficient of each spindle power data point, where the silhouette coefficient is used to evaluate the relative position of the spindle power data point in the cluster; S23, calculating the overall silhouette coefficient of the spindle power data set based on the silhouette coefficient of each spindle power data point; S24, constructing a first objective function of the K-Means algorithm based on the overall silhouette coefficient, wherein the first objective function is used to optimize the clustering effect; S25, constructing a second objective function of the spindle operating condition classification model based on the overall silhouette coefficient, wherein the second objective function is used to optimize clustering parameters.

4. A method for classifying turbine main shaft operating conditions based on data analysis as claimed in claim 3, characterized in that: Step S3 includes: S31, optimizing clustering parameters by using the second objective function, calculating the overall silhouette coefficient under different numbers of clusters, and selecting the optimal number of clusters based on the clustering effect; S32, calculate the best Euclidean distance and the best Manhattan distance for the optimal number of clusters as the distance metric for the K-Means algorithm; S33, inputting the first optimal number of clusters, the optimal Euclidean distance, and the optimal Manhattan distance into the K-Means algorithm.

5. A method for classifying turbine main shaft working conditions based on data analysis as claimed in claim 4, characterized in that: Step S4 includes: S41, setting the number of individuals and the total number of iterations, where the number of individuals is the number of individuals of the candidate solutions used in the optimization process; S42, initializing the position of the population, using a population initialization strategy, wherein the population initialization strategy includes random initialization, chaotic mapping, and random reverse learning; S43, calculating the fitness value of each individual under the second objective function to evaluate its clustering effect; S44, sorting according to fitness values, and selecting the individual with the highest fitness value as the current optimal solution; S45, executing an optimization iteration process, wherein the optimization iteration process includes an exploration phase, an exploration phase, and an exploitation phase, and updating the position of the individual to find a better solution of the current optimal solution; S46, increasing population diversity through individual mutation strategy and updating individual positions; S47, after each iteration, determine whether the set number of iterations has been reached. If the set number of iterations has not been reached, return to step S43 to continue iterating. If the set number of iterations has been reached, output the current optimal optimized individual position.

6. A method for classifying turbine main shaft operating conditions based on data analysis as claimed in claim 5, characterized in that: Step S4 also includes: The calculation formula for the random initialization is: χ i :x ij =ll j +r·(ul j -ll j ),i=1,2,...,N,j=1,2,...,M; Among them, χ i is the position of the i-th candidate solution, x ij is the position coordinate of the i-th candidate solution, ll j and j denote the lower and upper bounds of the j-th decision variable, r is a random value between 0 and 1, N is the number of candidate solutions, and M is the number of decision variables; The calculation formula of the chaotic map is: Among them, X k+1 is the number of individuals after the kth chaotic mapping, X k is the number of individuals before the kth chaotic mapping, mod(·) is the modulus function; The calculation formula of the random reverse learning is: in, For X k The reverse solution, U b and L b are the upper and lower boundaries of the problem to be optimized, and J is a random value between 0 and 1.

7. A method for classifying turbine main shaft operating conditions based on data analysis as claimed in claim 6, characterized in that: The calculation formula of the individual variation strategy is: G hippo =D hippo [1+aC(0,σ 2 )+bG(0,σ 2 )]; Among them, G hippo is the position of the individual with the highest fitness after mutation, D hippo is the position of the individual with the highest fitness before mutation, C(0,σ 2 ) is a random quantity of Cauchy distribution, G(0,σ 2 ) is the Gaussian distribution random quantity, a and b are the Cauchy distribution dynamic parameter and Gaussian distribution dynamic parameter respectively, is the current individual, f(·) is the fitness function, exp(·) is the exponential function, and σ is the standard deviation.

8. A method for classifying turbine main shaft operating conditions based on data analysis as claimed in claim 7, characterized in that: Step S5 includes: S51, inputting the first optimal number of clusters, the optimal Euclidean distance and the optimal Manhattan distance into a K-Means algorithm; S52, performing K-Means clustering iterative calculation on the spindle power data set based on the input parameters; S53, in the clustering iteration process, dynamically updating the cluster center of each data point until the clustering result converges to obtain the second best number of clusters; S54, performing final classification on the spindle operating conditions according to the second optimal number of clusters, and outputting the classification results.

9. A method for classifying turbine main shaft working conditions based on data analysis as claimed in claim 8, characterized in that: Step S52 includes: At the beginning of each iteration, the distance between the current cluster center and each data point is calculated, and the data point is assigned to the nearest cluster center; at the end of each iteration, the mean of each cluster center is calculated to update the position of the cluster center, and the clustering results of each iteration and the position changes of the cluster center are recorded.

10. A method for classifying turbine main shaft working conditions based on data analysis as claimed in claim 9, characterized in that: Step S54 includes: According to the second best number of clusters, the silhouette coefficient of each cluster is calculated to evaluate the clustering effect; the clustering results are compared with the preset working condition standards, and a classification report is generated to point out the characteristics of each working condition category; when outputting the classification results, the sample number and main statistical characteristics of each working condition category are provided, and the working condition classification is performed based on the sample number and the main statistical characteristics.

Images