Human-cognition-inspired deep model with its application to ocean wave height forecasting

Abstract

Ocean wave height (OWH) forecasting is indispensable but challenging task since that the series evolution involves mixed effects of numerous factors. However, most deep models only focus on nonlinear fitting in the data layer, are hard to accurately learn its evolution. By the fact that experienced fishermen achieve cognition for complex marine phenomena, this paper develops a human-cognition-inspired deep model for forecasting OWH including the diverse sense, brain analysis, and anticipation module. Firstly, through imitating the function of extracting diverse features based on multi-senses, the first module converts the original series into multiple simple modes via the multivariate variational mode decomposition (MVMD). Secondly, through imitating the gate and collaboration functions in the brain, the second module performs the capture of internal relevance and long short-term dependencies from each mode. Thirdly, through imitating the function of achieving reactions to complex environments, the third module sums forecasts of each mode and reconstructs final forecasts. Deep simulations of the handling flowchart and functions ensure effective forecasts. Five experiments and six discussions under two real-world OWH show that the proposed model is superior to 12 baselines, improves the mean absolute percent error of 64.6% and 63.9% on average, and provides reliable evidences for ocean wave management.