Evaluation of machine learning models for predicting suspended sediment inflows in artificial open channels

초록

Accurate prediction of suspended sediment load (SSL) in waterways is essential for ensuring the long-term sustainability of hydraulic systems. Machine learning (ML) models have been widely adopted in recent years for SSL prediction in natural rivers, but their application to artificial open channels (AOCs) remains limited. Various ML models, including multilayer perceptron (MLP), radial basis function neural network (RBFNN), and long short-term memory (LSTM), were evaluated to investigate the governing temporal characteristics of SSL dynamics in an AOC. The best-performing optimization algorithm for each ML model was selected, and the performance of the models was assessed using performance metrics. The results showed that the MLP model with the SGDM optimizer performed best for SSL inflow prediction with R2 = 0.805, PBIAS = +5.894, WI = 0.988, NSE = 0.892, and RSR = 0.329. Notably, the feedforward MLP provided robust predictions relative to the recurrent models, suggesting that SSL inflows in an AOC are strongly influenced by gate operation. The performance hierarchy based on Friedman ranking and Wilcoxon pairwise comparisons was MLP ' Bi-LSTM ' Uni-LSTM ' RBFNN. The results highlight the potential of ML models to outperform conventional SSL prediction methods by offering higher accuracy and greater adaptability. © 2026 The Authors.

키워드