Geneticizing input selection based advanced neural network model for sediment prediction in different climate zone

Abstract

The study focuses on developing an accurate prediction model for suspended sediment load (SSL) based on antecedent SSL and water discharge values. Two Artificial Intelligence (AI) models, Hybrid and Parallel, were employed to test on the Kelantan and Mississippi Rivers in different climate zones and river sizes. The parallel model showed better performance than the hybrid in most cases, with the best results based on Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) (432.06 and 782.15 respectively) for Kelantan and (31672.25 and 62356.60 respectively) for Mississippi. The multifunctional GA neural-network model results have proven its ability to predict SSL in tropical and semi-arid zones. In the Kelantan River, the 8-input combination set was the best prediction model, showing an improvement of more than 38% compared to traditional models. The proposed method has proven to be more accurate than traditional models, ensuring better water resource planning, agricultural management and reservoir operation.