Time Series Based Anomaly Data Detection For Wind Turbines

With the scarcity of fossil energy and the introduction of carbon peaks and carbon neutrality,clean energy is now the most desirable source of energy.As one of the clean energy sources,wind energy is developing rapidly,and wind power has become one of the most researched forms of power generation in Chinese new energy application technology.However,the operating environment of most wind turbines is very harsh,and a large number of abnormal data that do not correspond to the normal output of wind turbines can appear in the operating data,and the existence of these abnormal data can seriously affect the analysis of the status of the wind turbines later on.In order to make wind power a more reliable source of electrical energy,it is very important to establish an accurate wind turbine abnormality detection model.The research in this paper is as follows:Firstly,the causes of anomalous data in wind turbines and the distribution of the anomalies are analysed,extract the required data features,the data features are processed and the data is visualised to determine the distribution of the anomalous data.Secondly,according to the distribution of anomalous data,the anomaly detection algorithm is determined as the joint algorithm of KDE algorithm,linear regression algorithm,isolated forest algorithm,K-means algorithm and LSTM prediction algorithm,but due to the problem that the clustering centre of K-means algorithm may be a local optimal solution,the K-means algorithm is optimised and the K-means algorithm based on genetic algorithm is proposed,and the genetic algorithm is used to optimise each parameter of the LSTM prediction algorithm,the joint algorithm of the KDE algorithm,linear regression algorithm,isolated forest algorithm,the K-means algorithm based on genetic algorithm and the LSTM prediction algorithm based on the genetic algorithm are used to detect anomalies in wind turbine data.And according to the characteristics of the five algorithms,the joint algorithm will be used to detect the abnormality of the wind turbine.The joint algorithm uses five algorithms to detect the abnormality of the data set at the same time.When three kinds of detections are abnormal,it is judged as abnormal.Finally,the anomaly detection results of each algorithm were evaluated.Through the comparison of algorithms,the anomaly detection results were evaluated by F1 scores.The F1 scores of the joint algorithm for anomaly detection of twelve sets of wind turbine data were better than the F1 scores of the four algorithms for anomaly detection alone,with the highest F1 score of the joint algorithm being 0.9667 for 12 and the lowest F1 score being 0.9338 for wind turbine 3,therefore a joint algorithm is used to detect anomalies in the data.Through the joint algorithm to detect abnormal data in time and delete abnormal data in time,it can effectively analyze the status of wind turbines.