Research On Health Status Assessment And Prediction Of Wind Turbines Based On Data Minin

In recent years,with the increasing demand for renewable energy,the installed capacity of wind turbines has been growing,and the scale of wind farms is expanding.As wind turbines are often installed in remote areas and work in harsh natural environments,their operational performance can decrease continuously due to aging and the effects of harsh operating conditions.This leads to high failure rates,short lifespan,and high maintenance costs,which have become increasingly prominent issues.To ensure the long-term stable operation of wind turbines and reduce the huge losses caused by downtime due to failures,it is important to carry out research on monitoring and evaluating the operational and health status of wind turbines.The Supervisory Control And Data Acquisition(SCADA)system for wind turbines can obtain a large amount of operational status information for the turbines.This thesis,based on the study of the working principle and operating characteristics of wind turbines,fully utilizes the operational data of wind turbines to conduct research on data mining-based monitoring and health status evaluation and prediction of wind turbines.The main research content is as follows:(1)In view of the problem that SCADA data of wind turbines are greatly affected by the natural environment and operating conditions,making it difficult to directly use the raw data,this study has investigated the method for processing abnormal data of wind turbines.By analyzing the operational characteristics and raw data of the SCADA system of wind turbines,the distribution characteristics of abnormal and fault data are determined.Using the "wind speed-power" curve of wind turbines as a starting point and combining the local anomaly factor algorithm to clean the abnormal operational data of turbines,a data set of normal operating states for wind turbines is established,providing a data basis for subsequent research.(2)In view of the subjectivity of traditional feature selection methods and the problem of insufficient mining of hidden spatio-temporal feature information in SCADA data,a spatio-temporal fusion model based on Self-Attention(SA)feature selection combined with Convolutional Neural Networks(CNN)and Bidirectional Gating Recurrent Unit(BiGRU)is established to predict multiple operating parameters of wind turbines.By calculating the mean square error between the model’s predicted values and actual values,and using the exponential weighted moving average method to monitor the fluctuation trend of errors,the operational status of wind turbines is monitored.Case studies show that the model has high accuracy and can effectively monitor early abnormal states of turbines.(3)In view of the problem that traditional health status evaluation methods are subject to subjective factors in weight allocation,resulting in low reliability and poor stability of evaluation results,a health status evaluation and prediction model based on the Sparrow Search Algorithm(SSA)optimized Kernel Extreme Learning Machine(KELM)is constructed.Firstly,using the SA-CNN-BiGRU model to obtain the standard residual set under healthy conditions and predict the operating parameters of the turbines to obtain the predicted residual relative to the healthy state.Then,the health deterioration index of each subsystem is calculated using the Mahalanobis distance and the standard residual set,and KELM is used to predict the overall health status of the turbines.To further improve the accuracy and stability of the wind turbine health assessment model,the parameters of KELM are optimized by combining SSA.Finally,the effectiveness of the model is verified through case studies.