Research On Wind Turbine Fault Early Warning Technology Based On Operating Data

In recent years,wind energy has played an increasingly important role in China’s energy structure as a clean energy source.With the large-scale operation of wind turbines,the imbalance of wind energy distribution and irregular alternating loads cause frequent failures of wind turbines,which Affects the stable operation of wind farms and brings serious challenges to the development of wind power technology.Therefore,the research on fault warning technology for wind turbines is gradually becoming a focus for scholars.Machine learning as a method of implementing artificial intelligence has been successfully applied in various fields,which has greatly improved work efficiency.In this paper the machine learning algorithms is applied to the fault warning technology for wind turbines.In order to realize early warning of wind turbine faults,a comprehensive fault warning method for wind turbines based on feature extraction is proposed.The method includes two parts: prediction of state parameters of wind turbines and assessment of health status.The specific work is as follows.(1)A state parameter prediction algorithm for wind turbines(LPP-LightGBM)based on the Bureau’s projection and machine learning algorithm was proposed.The premise of realizing wind turbine fault early warning is to establish a state parameter prediction model with high stability and high accuracy.Therefore,according to the high-dimensional and disordered characteristics of SCADA data of wind turbines,the feature vector extraction of wind turbines is adopted to extract the characteristic vectors of wind turbines and then reuse XGBoost establishes a prediction model for the parameter feature vectors extracted by the insurance projection algorithm,but the pre-sorted algorithm used by XGBoost needs to pre-sort the features and store the sorted index values in order to find the data separation points more accurately.Consumption of memory,thus introducing LightGBM based on histogram algorithm to model and predict the characteristic vector of the main bearing parameters,and finally obtain the state prediction parameters of the wind turbine.(2)A fuzzy comprehensive evaluation method for assessing the health status of wind turbines has been designed.This method can transform qualitative evaluation to quantitative evaluation for objects affected by multiple factors,and comprehensiv ely analyze the health status assessment of the parameter residual change trend.Due to the characteristics of the simple evaluation process and high accuracy of the fuzzy comprehensive evaluation method.After the state parameters of the wind turbine are obtained through the proposed LPP-LightGBM method,the health status assessment of the wind turbine can be achieved through the fuzzy comprehensive evaluation.Eventually early warning of wind turbine failure was realized.The main bearing temperature related parameter data is selected as an example to verify the effectiveness of the proposed method in wind turbine fault early warning.Simulation results show that the proposed LPP-LightGBM algorithm can better mine low-dimensional information in high-dimensional data and improve the model’s performance.The prediction accuracy,combined with the fuzzy comprehensive evaluation method for health assessment,through comparison with the selected algorithm modeling,the proposed fault early warning mod el can better realize early warning of wind turbine failure and improve the safety and reliability of wind turbine operation.