Nonlinear autoregressive neural networks with external inputs for forecasting of typhoon inundation level

摘要: Accurate inundation level forecasting during typhoon invasion is crucial for organizing response actions such as the evacuation of people from areas that could potentially flood. This paper explores ability nonlinear autoregressive neural networks with exogenous inputs (NARX) to predict levels induced by typhoons. Two types NARX architecture were employed: series-parallel (NARX-S) and parallel (NARX-P). Based on cross-correlation analysis rainfall water-level data historical records, 10 models (five each type) constructed. The model was assessed considering coefficient efficiency (CE), relative time shift error (RTS), peak (PE). results revealed high CE performance be achieved employing more input variables. Comparisons two demonstrated NARX-S outperformed NARX-P in terms RTS, whereas both performed exceptionally PE without significant difference. highest overall identified their predictions compared those traditional ARX-based models. all three indexes, exhibited comparable superior RTS performance.