Wind Turbine Performance Assessment Modeling Using Machine Learning Method for Condition Based Maintenance

Abstract: This thesis work proposes a performance assessment framework to estimateoperation states of wind turbines for the sake of condition monitoring. Theframework uses the data in the supervisory control and data acquisition systemsas original input and some machine learning methods including K-NearestNeighbour, K-Means Clustering, Support Vector Machine and Artificial NeuralNetwork are implemented to analyze the data. The framework mainly consistsof three stages: power curve prediction, real time power tracking and turbineperformance assessment. At the first stage, two main methods including quartilemethod and k-means clustering and density-based clustering are implementedseparately for the elimination of bad measurements. Then di↵erent methods,including both parametric and non-parametric methods, are applied to estimatethe ideal wind turbine power curve, which is used as a reference value to assessthe real one. At the second stage, a sliding window method is introduced toanalyze the real time performance of wind turbines. The di↵erence between theexpected power output and real measurements are computed and used as theanomaly. At the third stage, performance zone is defined to evaluate the overallhealth condition of the turbines. The proposed approach has been applied withthe experience data of six onshore wind turbines in a single wind farm which islocated in southern Europe. The results show that the method in the frameworkcan monitor the wind turbine operation condition and evaluate the performancefor a wind turbine in this study case.