Study On Vibration Fault Diagnosis And Trend Prediction Of Hydroelectric Generator Units

Hydropower as a clean and renewable energy in China's entire energy structure occupies an important position.The hydropower unit,as the key core equipment for the energy conversion of hydropower stations,is actively engaged in the prediction of fault diagnosis and vibration trend of hydropower units,which can help to reduce the risk of unit accident and ensure the safe and stable operation of the unit.Due to the complicated structure of the hydropower unit,the interaction of the components and the mutual influence of the various incentives,the difficulty of predicting the fault diagnosis and vibration trend of the hydroelectric generating unit is increased to a certain extent.This paper focuses on the analysis and processing of non-stationary signals,fault feature extraction,fault feature screening,fault diagnosis model optimization and vibration trend prediction analysis.The empirical wavelet transform,maximum correlation kurtosis deconvolution,mixed gravitational search Algorithm,extreme learning machine as the theoretical basis,the nonstationary signal processing method and the filtering method are used to extract the fault characteristics of the hydropower unit effectively.The hybrid gravity search algorithm is used to improve the fault diagnosis from the two aspects of feature screening and optimization of the extreme learning machine.The vibration trend forecasting model of hydroelectric generating units combined with Gram-Schmidt orthogonal method and extreme learning machine is constructed.The main research of this paper can be as follows:Hydropower units in the course of the operation will be mechanical,hydraulic and electromagnetic incentives such as the common role,resulting in hydroelectric vibration signal with non-stationary and strong background noise.In order to effectively filter the noise reduction,the fault characteristics are extracted from the vibration signal of the hydroelectric generating set.The original vibration signal is decomposed into a series of single component modes by empirical wavelet transform.The modal model with fault characteristics is selected according to the correlation coefficient and kurtosis index Reconstruction,the maximum correlation kurtosis deconvolution is used to reconstruct the secondary filter of the signal.Through the spectral analysis and envelope analysis of the filtered signal,the fault frequency characteristics of the hydroelectric generating unit can be effectively extracted.The diagnostic accuracy of the fault diagnosis model is improved by using the hybrid gravity search algorithm from the fault feature screening and the optimization of the extreme learning machine parameters.The empirical wavelet transform decomposes the fault signal into a series of single component modalities,extracts the singular value features and energy characteristics of the modal matrix,and fuses the time domain and frequency domain statistical features extracted from the original signal into a mixed feature subset.In order to reduce the interference of the redundant features in the mixed feature subsets to the fault diagnosis,the binary gravitational search algorithm is used to filter the features.At the same time,the real weight gravitational algorithm is used to optimize the input layer weights and thresholds of the extreme learning machine.Finally,through the test data to verify the proposed fault diagnosis method has a high diagnostic accuracy.Hydroelectric group vibration signal analysis and fault diagnosis of hydropower units are post hoc decision analysis.In order to make up for the shortcomings of post hoc analysis,a method to predict the trend of the running status of the unit is proposed.According to the prediction results of the unit running trend,the unit is maintained in time to avoid further deterioration of the unit into unit failure.Based on the non-stationary vibration signal analysis and treatment of hydropower units,the trend prediction method of vibration signals of hydropower units is built by Gram-Schmidt orthogonal method and extreme learning mechine.In this method,the time series of vibration signal of hydroelectric unit is decomposed into a series of single component modal by empirical wavelet transform.For each input and output,Gram-Schmidt orthogonal method is used to filter the characteristics of the redundant features.Algorithm optimization of the extreme learning machine as a prediction model.The experimental results show that the proposed method can accurately predict the vibration trend of hydropower units by measuring the vibration data of hydropower units.