Wind wheel imbalance fault diagnosis method based on multi-kernel learning

Abstract

The invention mainly relates to the technical field of wind power generation, and discloses a wind wheel imbalance fault diagnosis method based on multi-kernel learning. The time domain characteristic parameters of the wind turbine generator in normal and fault states are extracted, the classification precision of the support vector machine is improved by introducing the multi-kernel learning algorithm, and the characteristic parameters of the support vector machine are automatically optimized in combination with the intelligent optimization algorithm, so that the accuracy of wind wheel imbalance fault diagnosis is remarkably improved, and the fault diagnosis efficiency is improved. The specific technical scheme is as follows: step 1, data acquisition and preprocessing; step 2, feature extraction and normalization processing; 3, designing and optimizing a multi-kernel learning algorithm; 4, establishing an intelligent optimization support vector machine model; and step 5, wind wheel imbalance fault diagnosis is carried out, and accurate identification and classification of wind wheel imbalance faults are realized. The method has the beneficial effect that the operation reliability and the maintenance efficiency of the wind generating set are improved through the non-sensing technology.

Description

Technical Field

[0001] The present invention relates to the field of wind power technology, and in particular to a wind wheel imbalance fault diagnosis method based on multi-core learning. Background Art

[0002] The reliable operation of wind turbines is crucial to modern energy supply, especially in large-scale wind farms, where rotor imbalance failure is one of the key factors affecting wind turbine performance and life. Rotor imbalance may cause vibration, noise, excessive equipment wear, and even serious damage, which brings huge challenges to the maintenance and management of wind turbines. Therefore, timely and effective diagnosis of rotor imbalance failure can significantly improve the operational stability and maintenance efficiency of wind turbines.

[0003] Some detection methods have been proposed for wind rotor imbalance faults. The literature [“Imbalance Fault Detection of Direct-Drive Wind Turbines Using Generator Current Signals,” (IEEE Transactions on Energy Conversion, vol. 27, no. 2, pp. 468-476, June 2012.)] mainly relies on the power spectrum of the stator current signal of the generator to identify the characteristics of wind rotor imbalance faults. The literature [“Wavelet Package Energy Transmissibility Function and Its Application to Wind Turbine Blade Fault Detection,” (IEEE Transactions on Industrial Electronics, vol. 69, no. 12, pp. 13597-13606, Dec. 2022.)] proposes a new wavelet packet energy method, which improves the resolution in the high-frequency range and can be further applied to wind rotor imbalance fault detection.

[0004] The above-mentioned literature provides valuable research ideas for the research on wind rotor imbalance fault detection, but the former is easily affected by the stator current under complex working conditions and cannot accurately distinguish the type of imbalance. The latter may not be able to quickly and effectively diagnose and shut down a wind turbine in real-time operation, which may easily cause greater damage to the wind turbine. The present invention proposes a wind rotor imbalance fault diagnosis method based on multi-core learning, which uses a multi-core learning method to optimize the feature selection process of a support vector machine, and combines an intelligent optimization algorithm to automatically optimize feature parameters to achieve efficient real-time diagnosis of wind rotor imbalance faults and accurately distinguish fault types. Through this method, the status of the wind rotor of a wind turbine can be monitored in real time without the need for additional sensors, which significantly improves the accuracy and response speed of fault diagnosis, and further improves the operating reliability and maintenance efficiency of wind turbines. Summary of the invention

[0005] The technical problem to be solved by the present invention is to propose a multi-core learning-based method for wind rotor imbalance faults, and finally establish a multi-core learning support vector machine model based on intelligent optimization to accurately realize wind rotor imbalance fault diagnosis.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for rapidly diagnosing wind rotor imbalance faults based on multi-core learning comprises the following steps:

[0008] Step 1: Data collection and preprocessing: collect raw vibration data and perform signal processing;

[0009] Step 2: Feature extraction and normalization: extract the required time domain feature parameters and perform normalization;

[0010] Step 3: Design and optimize multi-core learning algorithm: Establish a multi-core learning algorithm and obtain the optimal feature parameters of the support vector machine;

[0011] Step 4: Establishment of intelligent optimization support vector machine model: Based on intelligent optimization technology, an optimized multi-core learning support vector machine model is designed;

[0012] Step 5: Wind rotor imbalance fault diagnosis: Based on the model output, accurate identification and classification of wind rotor imbalance faults are achieved.

[0013] In step 1, the original vibration data comes from the tower vibration sensors installed on the wind turbine, which are installed in the left and right directions of the tower, the front and back directions of the tower, etc., and the data types include offline and online data;

[0014] In step 1, in order to improve the data analysis effect, the original vibration data is preprocessed by an efficient signal decomposition algorithm. The algorithm first determines the optimal decomposition layer number, and then decomposes the original vibration signal according to the number of layers to obtain n natural modal components and their residual components, and realizes signal denoising at the same time;

[0015] The step 2 adopts a feature extraction method combined with fault features to extract time domain feature parameters that can effectively characterize the fault features, and divides the feature parameters of the n inherent modal components and their residual components obtained by decomposition into n+1 feature groups;

[0016] In step 2, a normalization method is used to normalize the characteristic parameters of each characteristic group to a standard interval for subsequent analysis, and the normalized characteristic group is used as input data for the next step of analysis;

[0017] In step 3, a multi-kernel learning method integrating multi-kernel functions is used, and the kernel functions used include but are not limited to linear kernels, Gaussian kernels, etc., and the combination of these different kernel functions is used to capture the fault characteristics of different time domain feature parameters. The multi-kernel learning method can be further combined with other optimization methods to obtain the optimal feature parameters of the support vector machine;

[0018] In step 4, an intelligent optimization algorithm is used to adaptively optimize the multi-objective feature parameters, and a multi-core learning support vector machine model based on intelligent optimization is constructed by combining the multi-core learning method of the aforementioned fusion multi-core function. This optimization design effectively improves the diagnostic efficiency and accuracy of the model;

[0019] In step 5, the wind wheel imbalance fault is divided into mass imbalance fault and aerodynamic imbalance fault, and common causes include blade icing, blade sweep tower impact, and blade skin tearing, etc. By constructing a multi-core learning support vector machine model based on intelligent optimization, automatic classification and accurate diagnosis of different types of faults can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Attached Figure 1 is a system block diagram of the present invention; DETAILED DESCRIPTION

[0021] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are used to explain the present invention, and should not be interpreted as limiting the present invention.

[0022] Example

[0023] Combination Figure 1 A wind wheel imbalance fault diagnosis method based on multi-core learning of the present invention comprises the following steps:

[0024] Step 1: Data collection and preprocessing: collect raw vibration data and perform signal processing;

[0025] The original tower vibration signal x(t) from the offline data of the wind turbine is processed. First, the advantage of empirical mode decomposition is used to determine the decomposition level n. Although mode aliasing will occur in the process of empirical mode decomposition, it can adaptively obtain the decomposition level n without any prior knowledge.

[0026] As an adaptive decomposition algorithm, variational mode decomposition can decompose the signal into a set of finite intrinsic mode components u, combined with the decomposition layer number n. k , the frequency domain characteristics of each mode function are around their respective center frequency w k , get each mode u k The analytical signal of , and construct the variational constraint problem:

[0027]

[0028] in, represents the regularization parameter, α is the penalty factor for controlling the bandwidth, λ(t) is the Lagrange multiplication operator, * represents convolution, and δ(t) is the unit pulse function.

[0029] By solving the variational constraint problem, n intrinsic modal components and their residual components are obtained;

[0030] Step 2: Feature extraction and normalization: extract the required time domain feature parameters and perform normalization;

[0031] Firstly, a fuzzy entropy function is constructed, and the time-domain fault characteristic parameters of the data are introduced, including standard deviation, kurtosis factor, kurtosis factor, margin factor, clearance factor and pulse factor. The fuzzy entropy function is used to combine these fault characteristics to obtain the fuzzy entropy value matching the fault characteristics as the time-domain characteristic parameters to be used. Then, the time-domain characteristic parameters of n intrinsic modal components and their residual components are divided into n+1 characteristic groups.

[0032] Use the Z-Score normalization method to normalize each feature group to a uniform interval for subsequent analysis. The normalized feature groups are used as inputs for subsequent steps.

[0033] Step 3: Design and optimize multi-core learning algorithm: Establish a multi-core learning algorithm and obtain the optimal feature parameters of the support vector machine;

[0034] First, the fault characteristics of different time domain feature parameters are captured by combining different kernel functions. The kernel functions used include linear kernel, Gaussian kernel, Sigmoid kernel, polynomial kernel, Morlet wavelet kernel, and Marr wavelet kernel, which are expressed by formula (2):

[0035]

[0036] Among them, K1(x i ,x j ) represents the linear kernel, K2(x i, x j ) represents the Gaussian kernel, K3(x i, x j ) represents the Sigmoid kernel, K4(x i, x j ) represents the polynomial kernel, K5(x i, x j ) and K6(x i, x j ) represents the Morlet wavelet kernel and the Marr wavelet kernel, x i ,x j Represents two input vectors, σ represents the Gaussian kernel adjustment parameter, β and θ represent the Sigmoid kernel adjustment parameters, c represents the polynomial kernel constant term, d represents the number of scaling, a represents the wavelet kernel adjustment parameter, and n represents the wavelet number;

[0037] Secondly, K-means clustering is combined with multi-kernel function learning to randomly initialize K cluster centers, calculate the distance from each point to the K cluster centers, and then assign each data point to the cluster center closest to the cluster center. The cluster center is recalculated for each category, and the above steps are repeated to perform clustering in multiple feature spaces until the optimal weight of each kernel function is obtained. The optimal feature parameters of the support vector machine are obtained through continuous data iteration.

[0038] Step 4: Establishment of intelligent optimization support vector machine model: Based on intelligent optimization technology, an optimized multi-core learning support vector machine model is designed;

[0039] The performance of support vector machines depends on the selection of kernel functions, kernel parameters and the size of penalty coefficients. The above multi-kernel learning combined with the K-means clustering method can well select and combine kernel functions. In order to further improve the performance of support vector machines and select better kernel parameters and penalty coefficients, an intelligent optimization algorithm is needed.

[0040] A multi-objective particle swarm optimization algorithm is adopted. First, the particle swarm and its related parameters are randomly initialized, all non-dominated particles in the particle swarm are found and inserted into the non-dominated set, and then the non-dominated particles obtained in each generation are inserted into the storage set. Finally, the global extreme value is selected in the storage set. The particle swarm continuously searches for a better solution under the guidance of the extreme value.

[0041] The mathematical model is established as formula (3):

[0042]

[0043] Among them, x is the input vector, err represents the number of misclassifications, nsv represents the total number of support vectors, and N p Represents the prediction sample, N t represents the test sample, y = (f1(x), f2(x)) represents the fitness value corresponding to the multi-objective particle optimization algorithm, f1(x) represents the misclassification rate, and f2(x) represents the support vector occupancy rate;

[0044] Through formula (3), the optimal feature parameters and the support vector machine model are continuously updated using the optimization steps of the multi-objective particle swarm optimization algorithm, and finally the support vector machine model based on the multi-objective particle swarm optimization of the fusion multi-kernel function and K-means clustering is obtained;

[0045] Step 5: Wind rotor imbalance fault diagnosis: Based on the model output, accurate identification and classification of wind rotor imbalance faults are achieved;

[0046] Wind rotor imbalance faults are divided into mass imbalance faults and aerodynamic imbalance faults. Mass imbalance faults are usually caused by blade icing, blade breakage, and blade wear and corrosion. Aerodynamic imbalance faults are usually caused by installation problems, uneven wind speed, tower shadow effect, and inconsistent blade angles.

[0047] By establishing a multi-core learning support vector machine model based on intelligent optimization, various faults are automatically classified, thereby achieving efficient real-time diagnosis of wind wheel imbalance faults and accurately distinguishing fault types.

[0048] The present invention proposes a wind rotor imbalance fault diagnosis method based on multi-core learning, which uses the multi-core learning method to optimize the feature selection process of the support vector machine, combines the intelligent optimization algorithm to automatically optimize the feature parameters, and establishes a multi-core learning support vector machine model based on intelligent optimization. Through this model, the wind rotor status of the wind turbine can be monitored in real time without the need for additional sensors, which significantly improves the accuracy and response speed of fault diagnosis, and further improves the operating reliability and maintenance efficiency of the wind turbine.

[0049] The above description of the present invention is only an example for clearly illustrating the present invention, and is not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A wind rotor imbalance fault diagnosis method based on multi-core learning, which systematically realizes accurate diagnosis of wind rotor imbalance fault by combining multi-core learning algorithm with intelligent optimization technology; it is characterized by: The method comprises the following steps: Step 1: Data collection and preprocessing: collect raw vibration data and perform signal processing; Step 2: Feature extraction and normalization: extract the required time domain feature parameters and perform normalization; Step 3: Design and optimize multi-core learning algorithm: Establish a multi-core learning algorithm and obtain the optimal feature parameters of the support vector machine; Step 4: Establishment of intelligent optimization support vector machine model: Based on intelligent optimization technology, an optimized multi-core learning support vector machine model is designed; Step 5: Wind rotor imbalance fault diagnosis: Based on the model output, accurate identification and classification of wind rotor imbalance faults are achieved.

2. The wind wheel imbalance fault diagnosis method based on multi-core learning according to claim 1 is characterized in that: In the step 1, the selection of the preprocessing method of the original vibration signal is the key to this step, including wavelet decomposition and reconstruction, empirical mode decomposition, and order resampling method.

3. The wind wheel imbalance fault diagnosis method based on multi-core learning according to claim 1 is characterized in that: In the step 2, the time domain feature parameter extraction method includes fuzzy entropy function method, principal component analysis method, and fractal dimension method.

4. The wind wheel imbalance fault diagnosis method based on multi-core learning according to claim 1 is characterized in that: In step 2, the normalization processing methods include Z-Score standardization method, Box-Cox transformation method, and rank transformation method.

5. The wind wheel imbalance fault diagnosis method based on multi-core learning according to claim 1, characterized in that: In step 3, the multi-core learning algorithm includes constrained optimization multi-core learning, semi-positive definite programming multi-core learning, and linear combination kernel learning.

6. The wind wheel imbalance fault diagnosis method based on multi-core learning according to claim 1, characterized in that: In step 4, the intelligent optimization algorithm includes genetic algorithm, particle swarm optimization algorithm, ant colony optimization algorithm, and simulated annealing algorithm.

7. The wind wheel imbalance fault diagnosis method based on multi-core learning according to claim 1, characterized in that: In step 5, the types of wind rotor imbalance faults include mass imbalance faults and aerodynamic imbalance faults caused by long-term blade wear, blade icing, and blade breakage. The diagnosis and classification of wind rotor imbalance faults are achieved through the classification results output by the model.

Images