A novel hybrid model based on STL decomposition and one-dimensional convolutional neural networks with positional encoding for significant wave height forecast

Abstract Reducing the dependence on fossil fuels and utilizing the renewable energy have become essential due to the global resource exhaustion and unfriendly environmental impact from coal, petroleum and natural gas. Therefore, the rising attention has been paid to wave energy characterized by sustainability, clean, high energy density and extensive distribution. As one of the most important parameters of wave energy, significant wave height (SWH) is difficult to forecast accurately due to the complex marine condition and ubiquitous presence of chaos in nature. In this research, a novel hybrid model called STL–CNN–PE which combines seasonal-trend decomposition procedure based on loess (STL) and one-dimensional convolutional neural networks (CNN) with positional encoding (PE) was proposed to forecast SWH efficiently and accurately. To evaluate the proposed model comprehensively, the hourly standard meteorology data at station 44007, 46087 and 51000 from NOAA’s National Data Buoy Center were selected for model training and testing. The experimental results indicated that STL–CNN–PE provided more reliable forecasting values than the single model. Meanwhile, STL–CNN–PE had enormous advantage on speed and similar precision compared with EMD-LSTM. Finally, the experimental results revealed that the models provided better forecasting metrics at deeper waters.