Left-right brain interaction inspired bionic deep network for forecasting significant wave height

As a promising source of clean energy in carbon neutrality, ocean wave energy generation depends heavily on forecasting significant wave height (SWH), whose evolution is too complex to accurately model due to multifactor mixed effects. Additionally, most existing deep models present intelligent fitting via making some tricks and further have mining-extraction capabilities of hidden features, while they ignore the support of biologicallyinspired ideas. An important and interesting open issue is how to utilize SWH data characteristics with advanced brain structures and functions to construct its high-performance forecasting network. Specifically, from SWH data analysis, the overall framework of the proposed network separately extracts autocorrelation and causality via two brain-interaction-inspired (BII) modules at first, and then integrates them via the attention fusion module, which coincides with the idea of "divide and conquer". From a micro view, 1) through imitating both structures and functions in left-right brain interaction, the designed BII module stacks the one-dimensional convolutions and gate mechanisms to implement the gate, collaboration, and inhibition functions for capturing long short-term dependencies. 2) The attention mechanism with dynamic weights is designed to integrate captured information and real-timely grasp the main features for making high-accuracy forecasts. The proposed network not only has some interpretability in the design process but also effectively enhances the feature completeness. In six experiments of two real-world datasets, the proposed network improves the root mean squared error by averages of 26.3% and 23.7% compared with 11 baselines, respectively.