High-temperature power reduction state identification for wind turbines using feature correlation analysis and deep learning methods

The excessively high temperature of gearbox oil in summer leads to power reduction of wind turbines (called high-temperature power reduction state), which does harm to the performance of the power generation performance of the wind turbine. The timely and accurate identification of high-temperature power reduction state will contribute to improving the operation efficiency of the unit. This study proposes a new method based on the vine copula model and algorithm of convolutional neural network cascading to the bidirectional long short term memory network with attention mechanism (CNN-BiLSTM-attention). Firstly, the vine copula model is used to analyze the correlation of the features in supervisory control and data acquisition (SCADA) system, and the features that can reflect the high-temperature power reduction state are extracted comprehensively. Secondly, CNN is used to mine the coupling relationship between features and extract deep spatial features. Finally, BiLSTM is used to extract the time-series information in the depth spatial feature further for high-temperature power reduction state identification, and the attention mechanism is introduced to sense and identify the relevant network weights adaptively to enhance the influence of important information. The experimental results show that the method has high identification accuracy in the identification of high-temperature power reduction state. Accurate and reliable identification results can provide reference for formulating the operation and maintenance scheme of wind turbine reasonably.