A new hybrid model based on secondary decomposition, reinforcement learning and SRU network for wind turbine gearbox oil temperature forecasting. (June 2021)

• Record Type: Journal Article
• Authors: Liu, Hui; Yu, Chengqing; Yu, Chengming
• Is Part Of: Measurement. Volume 178(2021)
• Publication Date: 2021-06
• Subjects: Oil temperature forecasting -- Simple Recurrent Unit -- Reinforcement Learning -- Secondary decomposition method; VMD variational mode decomposition -- WPD wavelet packet decomposition -- EMD empirical mode decomposition -- SSA singular spectrum analysis -- SE sample entropy -- GA genetic algorithm -- RL reinforcement learning -- ARIMA auto-regressive integrated moving average -- NARX nonlinear autoregressive with external input -- SVM support vector machine -- LSSVM least squares support vector machine -- MLP multi-layer perception -- ENN elman neural network -- GMDH group method of data handling -- ESN echo state network -- DBN deep belief network -- RNN recurrent neural network -- LSTM long short-term memory -- BiLSTM bi-directional long short-term memory -- GRU gated recurrent unit -- SRU simple recurrent unit -- MAPE mean absolute percentage error -- MAE mean absolute error -- RMSE root mean square error; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2021.109347
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: The oil temperature series is predicted by time series method. The Sarsa method is applied as the feature selection method. The SRU is used as the main predictor to predict the oil temperature. The proposed model is compared with twenty-three mainstream forecasting models. Abstract: Oil temperature forecasting technology can realize real-time detection of the gearbox status of wind turbines. To make the oil temperature forecasting more accurate, a new hybrid model is presented in this study. The main modeling process of the presented method consists of three main steps. In step I, the proposed secondary decomposition method is utilized to preprocess the raw oil temperature data. In step II, the feature selection algorithm based on reinforcement learning selects the features of each sub-series. In step III, the simple recurrent unit network establishes forecasting models for each sub-series after feature selection and obtains the final forecasting results. By analyzing the forecasting results of multiple experiments, it can be concluded that: (1) the presented hybrid model can obtain satisfying forecasting results. Its RMSE values are 0.1101 °C, 0.1683 °C, and 0.1784 °C in three cases. (2) The presented hybrid model can get higher forecasting accuracy than the seventeen alternative models and six existing models in all cases. It improves the performance of traditional neural networks by over 90 percent.