Anomaly Detection and Classification Method for Wind Speed Data of Wind Turbines Using Spatiotemporal Dependency Structure

Detection of anomalous wind speed time series in measurements is considered crucial for the operation and maintenance of wind speed sensors on wind turbines, as they generally reveal some inherent defects of sensors or extreme environmental conditions in wind farms. In this article, an anomaly detection and classification method termed detectable spatial and temporal dependency structure (DSTDS) is developed for wind speed data. The main idea is to transform wind speed data into dependency matrix, and then, the problems of anomaly detection are turned into dependency modelling and cross-verification. First, the attributes of raw wind speed data are analyzed and the abnormal data are classified into three categories from statistical perspective. Next, a temporal segmentation scheme for wind speed data using change-points algorithm is proposed to improve both the accuracy and sensitivity of anomaly detection. Then, the spatial dependency structure of wind speed series, which is captured by t-copula-based method, is employed to detect the fault data and weak fault data. The temporal context relation is used to formulate the velocity constraints to detect the noisy data. Finally, experiments for anomaly detection are conducted on simulation data and real-world data to validate effectiveness and universality of the proposed method.