Resumen
Inundation forecast models with non-sequential regressors are advantageous in efficiency due to their rather fewer input variables required to be processed. This type of model is nevertheless rare mainly because of the difficulty in finding the proper combination of regressors for the model to perform accurate prediction. A novel methodology is proposed in this study to tackle the problem. The approach involves integrating a Multi-Objective Genetic Algorithm (MOGA) with forecast models based on ARX (Auto-Regressive model with eXogenous inputs) to transfer the search for the optimal combination of non-sequential regressors into an optimization problem. An innovative approach to codifying any combinations of model regressors into binary strings is developed and employed in MOGA. The Pareto optimal sets of three types of models including linear ARX (LARX), nonlinear ARX with Wavelet function (NLARX-W), and nonlinear ARX with Sigmoid function (NLARX-S) are searched for by the proposed methodology. The results show that the optimal models acquired through this approach have good inundation forecasting capabilities in every aspect in terms of accuracy, time shift error, and error distribution.