Wind Turbine Condition Monitoring Based on Autoencoder and K-means

Wind turbine condition monitoring (WTCM) is important to reduce the operation and maintenance cost of wind turbine. This paper proposes a WTCM approach based on autoencoder (AE) and K-means cluster. After data preprocessing, we firstly build an AE model with long short-term memory layer, and the construction of the model is determined by cross-validation experiment. Use the bottleneck layer of the AE model as the feature vector and establish the feature vector space of normal data. Secondly, the K-means cluster is employed. We gather the features of normal data into a cluster, then the cluster center and Euclidian distance are used to set the threshold. Thirdly, obtain the feature of testing data and calculate the Euclidian distance between the feature and the cluster center of normal data. The calculated Euclidian distance is used as the evaluation basis. Two cases with typical faults have been studied to demonstrate the feasibility of the proposed method.