A case study of industrial data analysis: Gearbox temperature prediction of wind turbines using ensemble deep learning regression

The damage and effectiveness of gearboxes may directly lead to the shutdown of wind turbines (WTs) and increase the unplanned maintenance cost. The variation and distribution of gearbox temperature can reflect its working condition. The predicting gearbox temperature has emerged as an essential method for guaranteeing the safety, availability, and efficiency of WTs. A method called ensemble deep learning regression (EDLR) is proposed for gearbox temperature prediction. Using random forest method, the collected features are then selected as main impacting features. Long short-term memory is sequentially employed for the temperature prediction. The proposed approach is investigated with real data from a wind farm. The experimental results reveal that EDLR has excellent regression performance in terms of the adopted criteria and that it can accurately predict the gearbox temperature.