Hydroelectric generating set vibration fault diagnosis method and system

Abstract

The invention belongs to the field of vibration fault diagnosis, and discloses a hydroelectric generating set vibration fault diagnosis method and system, and the method comprises the following steps: carrying out the improvement of a particle swarm algorithm based on a Tianbei search mechanism, and obtaining a Tianbei swarm algorithm; optimizing hyper-parameters of a support vector machine based on a Tiancattle swarm algorithm, wherein the hyper-parameters of the support vector machine comprise a penalty factor C and a kernel function parameter delta; the optimized support vector machine classifies and diagnoses the vibration fault type of the hydroelectric generating set, not only improves the accuracy of fault diagnosis, but also provides important technical support for intelligent monitoring and operation and maintenance of the hydroelectric generating set, and by monitoring and analyzing the operation data of the set in real time, the quick response of the fault can be realized, the downtime can be reduced, and the operation and maintenance efficiency can be improved; through historical data analysis, data mining and other technical means, fault diagnosis is carried out on abnormal conditions in operation of the hydropower station, equipment damage and fault occurrence are effectively avoided, and the safety and reliability of the hydropower station are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of vibration fault diagnosis, and relates to a vibration fault diagnosis method and system for a hydropower unit. Background Art

[0002] Hydropower units are key equipment in a hydropower generation system, and their operating efficiency and reliability have a profound impact on the power generation efficiency and the safety of the entire life cycle of the equipment. During operation, the unit faces the complex coupling effects of multiple physical fields such as hydraulics, mechanics, and electromagnetics. Coupled with the harsh service environment, it is prone to safety hazards such as mechanical vibration exceeding the limit, multi-source coupling faults, and material fatigue deterioration. In extreme conditions, it may even lead to functional failure or structural damage of key components. If not diagnosed and processed in a timely manner, it may cause equipment damage or shutdown, resulting in economic losses and safety hazards. In the actual operation of hydropower units, the vibration signals of the unit carry approximately 80% of the fault information. Based on this characteristic, vibration signals can be utilized and, with the help of advanced fault diagnosis methods, quickly and accurately identify the fault types of the unit, providing guidance for formulating corresponding maintenance strategies, thereby ensuring the safe and stable operation of the unit.

[0003] Support Vector Machine (SVM) can effectively handle small-sample and high-dimensional data, but its performance is mainly determined by the penalty coefficient and kernel function parameters, and advanced optimization algorithms need to be used to obtain the optimal parameter combination. Summary of the Invention

[0004] Aiming at the deficiencies of the existing technology, the present invention aims to provide a vibration fault diagnosis method and system for a hydropower unit, which uses the beetle swarm algorithm to optimize the support vector machine to improve the fault recognition accuracy and classification ability.

[0005] To achieve the above object, the present invention adopts the following technical solutions: The present invention provides a vibration fault diagnosis method for a hydropower unit, including the following steps: improving the particle swarm algorithm based on the beetle antennae search mechanism to obtain the beetle swarm algorithm; optimizing the hyperparameters of the support vector machine based on the beetle swarm algorithm, where the hyperparameters of the support vector machine include the penalty factor C and the kernel function parameter δ; and classifying and diagnosing the vibration fault types of the hydropower unit using the optimized support vector machine.

[0006] Further, the beetle antennae search mechanism includes the following steps: randomly generating a direction vector and normalizing it; calculating the left and right antenna coordinates; calculating the position of the next beetle according to the fitness; and updating the step size and search distance.

[0007] Furthermore, the process of optimizing the hyperparameters of the support vector machine based on the longhorn beetle swarm algorithm includes: initializing the longhorn beetle swarm algorithm parameters; determining the initial positions and velocities of the longhorn beetle particles; evaluating the fitness of the initial longhorn beetle particles and whether the initial positions of the longhorn beetle particles are the optimal solutions. If they are the optimal solutions, the penalty factor C and the kernel function parameter δ are output; otherwise, the initial positions and velocities of the longhorn beetle particles are updated until the optimal solutions are obtained, where the optimal solutions mean that the fitness cannot be improved any further.

[0008] Furthermore, the support vector machine adopts a kernel function, and the kernel function is:

[0009] where, is the data point in the input space, is the feature vector, is the kernel function reference.

[0010] Furthermore, the optimized support vector machine is the support vector machine optimized by the longhorn beetle swarm algorithm; the classification accuracy of the support vector machine optimized by the longhorn beetle swarm algorithm is higher than that of a single support vector machine, the support vector machine optimized by the genetic algorithm, and the support vector machine model optimized by the particle swarm algorithm.

[0011] Furthermore, the longhorn beetle swarm algorithm is:

[0012] where, is the historical position of the longhorn beetle particle, is the updated velocity of the longhorn beetle particle.

[0013] Furthermore, the updated velocity of the longhorn beetle particle is:

[0014]

[0015] where, is the inertia factor, is the historical velocity of the longhorn beetle particle, is the self-cognition term, is the swarm-cognition term, is a constant set artificially, is a random number, is the velocity update rate, is the step size at the t-th time, is the direction vector, is the fitness of the right antenna, is the fitness of the left antenna.

[0016] Further, the function of the longicorn beetle swarm algorithm for optimizing the support vector machine is as follows:

[0017] wherein, is the number of training samples, is the Lagrange multiplier, is the class label, is the kernel function, is the threshold determined by the training samples.

[0018] Further, the fitness of the initial longicorn beetle particle

[0019]

[0020] wherein, is the error rate.

[0021] The present invention also provides an abnormal data cleaning system based on multi-model fusion, including: an improvement module: for improving the particle swarm algorithm based on the longicorn beetle antenna search mechanism to obtain a longicorn beetle swarm algorithm; an optimization module: for optimizing the hyperparameters of the support vector machine based on the longicorn beetle swarm algorithm, where the hyperparameters of the support vector machine include the penalty factor C and the kernel function parameter δ; a diagnosis module: classifying and diagnosing the vibration fault types of the hydro-generator set by the optimized support vector machine.

[0022] Compared with the prior art, the present invention has the following beneficial technical effects: A vibration fault diagnosis method for a hydro-generator set according to the present invention not only improves the accuracy of fault diagnosis, but also provides an important technical support for the intelligent monitoring and operation and maintenance of the hydropower station. By real-time monitoring and analyzing the operation data of the unit, early diagnosis and rapid response to faults can be achieved, and the operation and maintenance efficiency can be improved.

[0023] A vibration fault diagnosis method for a hydro-generator set according to the present invention can timely detect and handle potential problems of the hydro-generator set through accurate fault diagnosis, avoid the expansion of faults, and thus enhance the operation reliability and safety of the unit.

[0024] A vibration fault diagnosis method for a hydro-generator set according to the present invention automatically optimizes the kernel function and penalty factor in the support vector machine SVM through the global search ability of the longicorn beetle swarm algorithm, significantly improving the generalization ability and classification accuracy of the support vector machine SVM.

[0025] A vibration fault diagnosis method for a hydro-generator set according to the present invention, the support vector machine SVM can effectively classify and predict unseen fault samples by finding the optimal classification hyperplane in the high-dimensional feature space.

[0026] A vibration fault diagnosis method for a hydropower unit according to the present invention can extract features and perform mapping on the vibration signals of the hydropower unit, enabling diagnosis at the initial stage of a fault, providing a valuable time window for maintenance decision-making, and reducing downtime and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flowchart of a vibration fault diagnosis method for a hydropower unit according to the present invention; Figure 2 It is a comparison graph for verifying the advantages of the longhorn beetle swarm algorithm with a non-linear function as an example in an embodiment of the present invention; Figure 3 It is an iteration curve of the longhorn beetle swarm algorithm, particle swarm algorithm, and genetic algorithm in an embodiment of the present invention; Figure 4 It is an optimization effect graph of hyperparameters for comparing three algorithms with the classification function of SVM as an example in an embodiment of the present invention; Figure 5 It is the fault diagnosis result of a support vector machine in an embodiment of the present invention; Figure 6 It is the fault diagnosis result of a support vector machine optimized by a genetic algorithm in an embodiment of the present invention; Figure 7 It is the fault diagnosis result of a support vector machine optimized by a particle swarm algorithm in an embodiment of the present invention; Figure 8 It is the fault diagnosis result of a support vector machine optimized by the longhorn beetle swarm algorithm in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] In order to enable those skilled in the art of the present technology to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0029] Currently, the main methods for vibration fault diagnosis of hydropower units are empirical mode decomposition (EMD), artificial neural network (ANN), and support vector machine (SVM). Empirical mode decomposition is an adaptive decomposition method for analyzing non-linear and non-stationary signals. By decomposing the vibration signals, its intrinsic mode functions are extracted for fault diagnosis. An artificial neural network is a computational model that simulates the working mechanism of a biological neural network, processes complex vibration signals, and can automatically discover and identify fault features through its powerful learning ability. After appropriate training and testing, it can effectively improve the accuracy and efficiency of fault diagnosis. As a supervised learning model, the support vector machine has wide applications in classification and regression analysis and is suitable for vibration fault diagnosis.

[0030] Fault diagnosis of hydroelectric generating units usually faces the challenge of limited sample size, belonging to the small-sample classification problem. Support vector machines have excellent performance in solving small-sample classification problems, so they have become a commonly used method in the field of fault diagnosis.

[0031] Example 1 The present invention provides an abnormal data cleaning method based on multi-model fusion, as Figure 1 shown, including the following steps: improving the particle swarm algorithm based on the beetle antennae search mechanism to obtain the beetle swarm algorithm; optimizing the hyperparameters of the support vector machine based on the beetle swarm algorithm, where the hyperparameters of the support vector machine include the penalty factor C and the kernel function parameter δ; classifying and diagnosing the vibration fault types of hydroelectric generating units using the optimized support vector machine.

[0032] Improving the particle swarm algorithm based on the beetle antennae search mechanism, specifically: adding the beetle antennae search mechanism to the particle swarm algorithm.

[0033] The particle swarm optimization algorithm is a swarm intelligence algorithm that simulates the foraging behavior of birds. By continuously updating its own position and speed, and communicating and comparing with other individuals, the position update mainly depends on the individual's current position and speed information.

[0034] The particle position is updated as follows:

[0035] Among them, is the position of the th iteration of the th particle, is the .

[0036] The particle speed is updated as follows:

[0037] Among them, the th iteration of , is the inertia weight, used to control the tendency of the particle to maintain its original motion state, the th iteration of the th particle's speed, is the component for the particle to learn from its individual optimal position, is the component for the particle to learn from the global optimal position. It should be noted that the particle speed update here is the particle update speed of the particle swarm algorithm.

[0038]

[0039] Among them, is the individual learning factor, is a random number, is the th iteration of the th particle's individual optimal position, is the th iteration of the th particle's position.

[0040]

[0041] Among them, is the global learning factor, is a random number, is the th iteration of the global optimal position, is the th iteration of the th particle's position.

[0042] The particle swarm optimization algorithm overly relies on the group cooperation mechanism to exert an influence on individual particles, and it is easy to ignore the judgment ability of the particles themselves, resulting in certain limitations in its global search ability and being prone to falling into the dilemma of local optimal solutions. To overcome this defect, the beetle antennae search mechanism is introduced to optimize the update rule of the particle swarm optimization algorithm.

[0043] The process of the beetle antennae search mechanism includes calculating the coordinates of the left and right antennae; calculating the position of the next beetle according to the fitness; updating the step size and the search distance. Before calculating the coordinates of the left and right antennae, a random direction vector is first generated and normalized Randomly generating and normalizing the direction vector is:

[0044] Among them, is a function that generates a random number, is the dimension of generating a random number, and 1 indicates generating a vector. is to calculate of the modulus or norm, that is, the length of the vector.

[0045] The calculation of the left and right antennae coordinates is as follows:

[0046]

[0047] Among them, is the coordinate of the right antenna of the beetle in the th iteration, is the position of the beetle in the th iteration, is the The step size of the longhorn beetle in the i-th iteration is a standardized direction vector with a modulus of 1 and a random direction is the coordinate of the left antenna of the longhorn beetle in the i-th iteration

[0048] Calculate the position of the next longhorn beetle according to the fitness, that is, by comparing the fitness values of the left and right antennae, the longhorn beetle will move in the direction with higher fitness. Specifically:

[0049] where is the position of the updated longhorn beetle particle is the step size at the k-th time is the direction vector is the fitness of the right antenna is the fitness of the left antenna

[0050] The longhorn beetle explores the surrounding environment through its left and right antennae and calculates the fitness values of these two points respectively. By comparing the fitness values of the left and right antennae, the longhorn beetle can sense which direction is more conducive to finding a better solution. Specifically, if the fitness value of the left antenna is higher than that of the right antenna, the longhorn beetle will tend to move to the left; otherwise, it will move to the right. This movement mechanism based on fitness feedback ensures that the longhorn beetle can gradually approach the global optimal solution, thus reflecting the efficiency and intelligence of the longhorn beetle swarm algorithm in optimization problems

[0051] Update step size and search distance

[0052]

[0053] where is the step size at the (k + 1)-th time is an artificially set attenuation coefficient is the step size at the k-th time is the distance from the centroid to the antenna at the (k + 1)-th time, and C3 is an artificially set constant

[0054] The longhorn beetle swarm algorithm is an improved algorithm developed on the basis of the particle swarm optimization algorithm. It obtains the longhorn beetle swarm algorithm by introducing the longhorn beetle antenna search mechanism, and optimizes the position update rule of the particles. In the optimized rule, the position update of the particles is no longer limited to the historical optimal solution and the global optimal solution, but also incorporates the key factor of the particle velocity update rate. The improved algorithm solves the local optimal solution problem by comparing the positions of the left and right antennae and then determining the update strategy of the population according to the comparison result

[0055] To sum up, the longhorn beetle swarm algorithm is as follows

[0056]

[0057]

[0058] Among them, is the inertia factor, is the historical longhorn beetle particle velocity, is the self-cognition term, which is a vector pointing from the current point to the best point of the particle itself, is the swarm cognition term, which is a vector pointing from the current point to the best point of the population, is a constant set artificially, is a random number, is the velocity update rate, is the step size at the t-th time, is the direction vector, is the right antenna fitness, is the left antenna fitness.

[0059] To verify that the longhorn beetle swarm algorithm has an advantage in global search, the longhorn beetle swarm algorithm is compared with the particle swarm algorithm and the genetic algorithm.

[0060] The longhorn beetle swarm algorithm, the particle swarm algorithm, and the genetic algorithm are used to solve the maximum value of the non-linear function, and search is carried out in the interval -2 ≤ x ≤ 2, -2 ≤ y ≤ 2. Taking the calculation result as the fitness function value, look at the output optimal solution and the corresponding value. The results are as Figure 2 shown. Figure 2 Taking the non-linear function as an example, verify the search results of the longhorn beetle swarm algorithm, the particle swarm algorithm, and the genetic algorithm. The results show that the longhorn beetle swarm algorithm has an advantage.

[0061] On the left is the solution result of the longhorn beetle swarm algorithm, in the middle is the solution result of the particle swarm algorithm, and on the right is the solution result of the genetic algorithm. As described in Table 1, the absolute value of the relative error of the longhorn beetle swarm algorithm is 0.021%, which is the smallest among the three algorithms, so the calculation accuracy is the highest. The abbreviation of the longhorn beetle swarm algorithm is BSO, the abbreviation of the particle swarm algorithm is PSO, and the abbreviation of the genetic algorithm is GA.

[0062] Table 1

[0063] As Figure 3As shown, premature convergence occurred in the genetic algorithm and the particle swarm algorithm, and they fell into local optimal solutions. The optimal solution obtained by the longhorn beetle swarm algorithm is the largest among the three algorithms, indicating that the longhorn beetle swarm algorithm has a higher accuracy rate compared to the genetic algorithm and the particle swarm algorithm, and can better solve the problem that the optimization process is prone to falling into local optima.

[0064] Optimize the hyperparameters of the support vector machine based on the longhorn beetle swarm algorithm. The hyperparameters of the support vector machine include the penalty factor C and the kernel function parameter δ.

[0065] Specifically: Initialize the parameters of the longhorn beetle swarm algorithm, including the population size, the upper limit of the number of iterations, the value range of the penalty factor C, and the kernel function parameter δ. Randomly generate the initial positions and velocities of the longhorn beetle particles, and use the non - linear function to calculate the fitness value of each longhorn beetle particle, and determine whether it is the optimal solution. If it is the optimal solution, output the penalty factor C and the kernel function parameter δ. Otherwise, update the positions and velocities of the longhorn beetle particles, and continue to iterate until the optimal solution is obtained. Input the optimized penalty factor C and kernel function parameter δ into the support vector machine for model training. Use the trained support vector machine model to classify and diagnose the vibration signal data of the hydropower unit. The vibration signal data of the hydropower unit includes seven types, namely normal, dynamic - static rubbing, rotor imbalance, hydraulic imbalance, excessive bearing clearance, rotor misalignment, and eccentric vortex in the draft tube.

[0066] Example 2 Collect 1820 groups of experimental samples and use 1680 groups of data with fault labels to train the SVM model.

[0067] Use 140 samples without fault labels to verify the SVM model. That is, under the assumption that the fault type is unknown, input 140 samples into the algorithm model and classify them into seven state types through model classification.

[0068] Initialize the parameters of the longhorn beetle swarm algorithm. Set the population size to 20; set the maximum number of population iterations Kmax to 200; set the search dimension to 2; set the weight to 0.7; set the maximum change of the penalty factor C to 100 and the minimum to 0.1; set the maximum of the kernel function parameter δ to 1000 and the minimum to 0.01. C1 = 0.4, C2 = 2, C3 = 3, as shown in Table 2.

[0069] Table 2

[0070] To evaluate the classification effect of the model, based on the same dataset, the penalty factor C and kernel function parameter δ of the support vector machine are optimized by the genetic algorithm, particle swarm algorithm, and beetle swarm algorithm respectively, and the optimized parameter combinations are applied to the support vector machine to construct a support vector machine optimized by the genetic algorithm, a support vector machine optimized by the particle swarm algorithm, and a support vector machine optimized by the beetle swarm algorithm. The parameter optimization results are shown in Table 3, and the iteration curves are as Figure 4 shown. The abbreviation of the beetle swarm algorithm is BSO, the abbreviation of the particle swarm algorithm is PSO, and the abbreviation of the genetic algorithm is GA. Figure 4 Taking the classification function of SVM as an example, the optimization effect diagrams of the hyperparameters by the beetle swarm algorithm, particle swarm algorithm, and genetic algorithm are compared.

[0071] Table 3

[0072] The support vector machine, the support vector machine optimized by the genetic algorithm, the support vector machine optimized by the particle swarm algorithm, and the support vector machine optimized by the beetle swarm algorithm are respectively used to detect the samples of the 7-hydroelectric generator unit vibration fault test set, and the classification results are as Figure 5 、as Figure 6 、as Figure 7 、as Figure 8 shown, and the classification accuracy of the model is shown in Table 4.

[0073] Table 4

[0074] The experimental results show the advantages of the beetle swarm algorithm compared with the particle swarm algorithm, especially its remarkable ability to avoid falling into local optimal solutions. By introducing the beetle swarm algorithm to optimize the parameter settings of the support vector machine, the accuracy has been effectively improved. It not only enhances the generalization ability of the model but also provides new ideas for dealing with complex classification tasks. The application of the beetle swarm algorithm not only solves the problem that traditional optimization methods are prone to falling into local optima but also further improves the performance of the support vector machine in various application scenarios.

[0075] Example 3 An abnormal data cleaning system based on multi-model fusion of the present invention includes a preprocessing module, a first separation module, a second separation module, a building module, a combination module, and a cleaning module.

[0076] Preprocessing module: used to preprocess the data that needs to be cleaned; First separation module: used to separate discrete abnormal data; Second separation module: used to separate piled-up abnormal data; Establishment module: used to establish a random forest model and a long short-term neural network model; Combination module: used to combine the random forest model and the long short-term neural network model based on the artificial fish swarm algorithm to obtain an RF-LSTM prediction model; Cleaning module: used to input the data with missing values into the RF-LSTM prediction model, output the prediction results and fill them into the data set to complete the cleaning of abnormal data.

[0077] The abnormal data cleaning system based on multi-model fusion provided by the present invention can implement the method steps consistent with the above method implementation, so it will not be elaborated here.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: still can modify the specific implementation manners of the present invention or make equivalent replacements, and any modification or equivalent replacement without departing from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A method for diagnosing vibration faults of a hydropower unit, characterized in that: The following steps are involved: The particle swarm algorithm is improved based on the beetle whisker search mechanism to obtain the beetle swarm algorithm. Optimizing the hyperparameters of a support vector machine based on a beetle swarm algorithm, wherein the hyperparameters of the support vector machine include a penalty factor C and a kernel function parameter δ; The optimized support vector machine is used to classify and diagnose the vibration fault types of hydropower units.

2. The method for diagnosing vibration faults of a hydroelectric generator set according to claim 1, characterized in that: The longicorn beard search mechanism comprises the following steps: Randomly generate direction vectors and normalize them; Calculate left and right whisker coordinates; Calculate the next beetle's position based on fitness; Update the step size and search distance.

3. The method for diagnosing vibration faults of a hydroelectric generator set according to claim 1, characterized in that: The process of optimizing the hyperparameters of the support vector machine based on the beetle swarm algorithm includes: Initialize the parameters of the beetle swarm algorithm; Determine the initial position and velocity of the longicorn particle; Evaluate whether the initial beetle particle fitness and initial beetle particle position are the optimal solution. If they are the optimal solution, output the penalty factor C and kernel function parameter δ; Otherwise, the initial position and speed of the longicorn particle are updated until an optimal solution is obtained, wherein the optimal solution is a solution that cannot further improve the fitness.

4. The method for diagnosing vibration faults of a hydroelectric unit according to claim 1, characterized in that: The support vector machine adopts a kernel function. for: in, is a data point in the input space, is the feature vector, For kernel function reference.

5. The method for diagnosing vibration faults of a hydroelectric generator set according to claim 4, characterized in that: The optimized support vector machine is a support vector machine optimized by the beetle swarm algorithm; The classification accuracy of the support vector machine optimized by the beetle swarm algorithm is higher than that of a single support vector machine, a genetic algorithm optimized support vector machine, and a particle swarm algorithm optimized support vector machine model.

6. The method for diagnosing vibration faults of a hydroelectric generator set according to claim 5, characterized in that: The beetle swarm algorithm for: in, Historical positions of longicorn particles, is the updated speed of the longicorn particles.

7. The method for diagnosing vibration faults of a hydroelectric generator set according to claim 6, characterized in that: The updated longicorn particle velocity for: in, is the inertia factor, is the historical longicorn particle velocity, For self-awareness, is the group cognition item, is an artificially set constant. is a random number, is the speed update rate, is the step size at time t, is the direction vector, is the fitness of the right whisker, is the left whisker fitness.

8. The method for diagnosing vibration faults of a hydroelectric generator set according to claim 6, characterized in that: The function of the beetle swarm algorithm to optimize the support vector machine for: in, is the number of training samples, is the Lagrange multiplier, is the category label, is the kernel function, The threshold determined for the training samples.

9. The method for diagnosing vibration faults of a hydroelectric generator set according to claim 3, characterized in that: The initial longicorn particle fitness in, is the error rate.

10. An abnormal data cleaning system based on multi-model fusion, comprising: Improvement module: used to improve the particle swarm algorithm based on the beetle whisker search mechanism to obtain the beetle swarm algorithm; Optimization module: used for optimizing the hyperparameters of the support vector machine based on the beetle swarm algorithm, wherein the hyperparameters of the support vector machine include a penalty factor C and a kernel function parameter δ; Diagnosis module: The optimized support vector machine classifies and diagnoses the vibration fault types of hydropower units.

Images