Wind Turbine Dynamic State Prediction Method Based on the Similarity of Historical Data

This paper proposes a wind turbine dynamic state prediction method using similarity matching in historical data. Because similarity matching is inefficient for large-scale multivariate time series (MTS) data, we use feature selection method based on random forest to determine main attributes and remove irrelevant attributes. Then, by using the principal component analysis, the dimension of MTS data can be reduced into one dimension. After that, we search the most similar historical data for a given current synthetic attribute time series, and then use the consequent data of the chosen historical data as the predicting state of the wind turbine. Theoretical analysis and experimental results show that the proposed method is robust and effective.