Genetic algorithm and deep learning models compared for swell wave height prediction

Abstract

A comparative study has been conducted between genetic algorithm (GA) and deep learning models to predict swell wave heights in the Bay of Bengal (BOB) region. To simulate the required parameter SWAN (Simulating Waves Nearshore) model is integrated with daily 25 km wind from 2009 to 2018 for July and December separately representing the southwest and northeast monsoons respectively. For the BOB region empirical orthogonal function (EOF) analysis is applied on the swell parameter to study the spatial and temporal patterns. GA is applied on the principal component of swell wave heights to generate a forecast explicit equation and thus a basin scale EOF-GA model is established. Next a grid (20⁰ N, 90⁰E) is chosen in the head bay region and the outcomes of the standalone GA model and the deep learning models are compared to predict the time series data of swell wave heights (SWS). It is observed that the performances of the deep learning model is better during the calm conditions in December than the rough seas in July. Another grid (15⁰ N, 82⁰E) is chosen along the east coast through which the severe cyclonic storm PHETHAI (13–18 December 2018) passed and the model accuracies are tested. The EOF-GA model serves as an effective computationally cheap basin scale forecast model. Thus, both the genetic algorithm and deep learning models can be developed and utilized for normal and extreme wave prediction having wide application in the ocean engineering domains.