An Abnormal Data Processing Method Based on An Ensemble Algorithm for Early Warning of Wind Turbine Failure

In the modeling process of fault warning models, modeling data plays an important role, the quality of which affects the performance of the model. The manual selection of modeling data according to fault records is time-consuming and makes it difficult to guarantee the high quality of the data because of inconsistencies, errors, and losses of records in the fault log file. For this reason, the present study proposes a framework of abnormal data processing based on an unsupervised serialization ensemble algorithm, which considers the high dimensional characteristics of operational data and the relationship between known low dimensional variables. Meanwhile, the influence of modeling data using different data processing methods on fault prediction performance is studied. The improved stacked autoencoder (I-SAE) based on the idea of partial data reconstruction is proposed to learn the high dimensional characteristics of operational data, which can enhance the separability of normal and abnormal operational data. The improved density-based spatial clustering of applications with noise (DBSCAN) clustering algorithm based on the density ratio is developed to handle sparse abnormal data with data imbalance characteristics. Finally, the case analysis results demonstrate that the abnormal data processing model proposed in this article has better performance than other methods, and the performance of the fault prediction model can be effectively enhanced by improving the quality of modeling data.