Multiple-step accurate prediction of wave energy: A hybrid model based on quadratic decomposition, SSA and LSTM

Abstract

ABSTRACT Accurate and efficient prediction of wave energy flux (WEF) holds significant engineering value. However, the inherently non-stationary and nonlinear characteristics of WEF pose a formidable challenge to forecasting efforts. A novel hybrid model for WEF prediction was proposed, which integrated efficient modal decomposition and fusion strategies, employs long short-term memory neural networks (LSTM) for forecasting, and utilizes the sparrow search algorithm (SSA) for overall optimization. First, the WEF sequences were pre-processed using complete ensemble empirical modal decomposition with adaptive noise (CEEMDAN) to reduce their complexity. Second, the mid-frequency and low-frequency sequences after sample entropy (SE) reconstruction were used for prediction, while the high-frequency sequences needed to undergo quadratic decomposition before being applied to the prediction. Third, SSA was employed to optimize the prediction model for higher performance. At last, the three predictions obtained are merged into the final prediction of the WEF. Experiments were carried out for one-step, three-step, and six-step predictions using the two buoy station data in the eastern North Pacific Ocean. The results indicated that the proposed model has higher prediction accuracy, with the lowest MAE of 0.6348, 0.8694, 2.4579 and the highest R 2 of 0.9985, 0.9953, 0.9770, respectively. Therefore, this study provides an effective prediction tool for engineering applications of wave energy.