Probabilistic modelling of surface drift prediction in marine disasters based on the NN–GA and ARMA model

Abstract

This paper studies the issue of target drift prediction based on ocean current from high-frequency (HF) ground-wave radar. A large number of unpowered drift observation experiments were carried out with a typical South China Sea offshore fishing vessel in the range of radar radiation. A target drift prediction model was proposed based on the neural network-genetic algorithm (NN–GA) and Auto-regressive Moving Average (ARMA) model. Replacing the traditional AP98 model, the NN–GA was used to establish the target wind-induced drift characteristic model, while the ARMA model replaced the traditional sub-grid model to refine the time series characteristics of sub-grid velocity. Firstly, trajectory simulation based on the Runge-Kutta method was used to preliminarily verify the improvement of target drift prediction accuracy by the ARMA model. In addition, the MonteCarlo method was used to calculate and compare the prediction range of different sub-grid velocity models. The study also investigated the NN–GA and ARMA model with different continuous prediction time steps. Compared with the traditional AP98 and sub-grid velocity model, the proposed method does not need to pay attention to the detailed force analysis of the drifting target and can significantly improve the performance of the target drift prediction model in terms of mean prediction error and separation. However, the ARMA model is also limited by its continuous prediction step size to a large extent. This study provides a new approach for trajectory analysis of drifting targets and verifies that the NN–GA and ARMA model with short-term continuous prediction can improve the performance of the target drift prediction model.