Real-time forecasting of suspended sediment concentrations reservoirs by the optimal integration of multiple machine learning techniques

Abstract

Study region: Shihmen Reservoir is ranked the second largest designed storage capacity in Taiwan. Study focus: The accurate forecasting of suspended sediment concentrations (SSCs) during typhoons is critical for effective reservoir management. This paper proposes a two-step switched machine learning (ML)-based approach for constructing an effective model to forecast reservoir SSCs. Different ML algorithms are adopted in the first ML step to build multiple ML-based SSC forecasting models, including multilayer perceptrons, random forest, support vector machines (SVMs), deep neural networks, recurrent neural networks, long short-term memory (LSTM) networks, and gated recurrent units. To compensate for a deficiency in measured SSC data, historical typhoons are modeled using the well-validated SRH-2D numerical model. The second step develops a switched forecasting strategy to optimally integrate forecasts from multiple ML-based models to provide more accurate calculations. New hydrological insights: The SSC forecasts obtained from the SVM and LSTM are confirmed to be superior to those from other ML-based models. The proposed model (optimally integrated from multiple ML-based models) outperforms the others, particularly when forecasting 1 and 3 h ahead. The proposed model improves the accuracy of SCC forecasts and can be used for sedimentation management in reservoirs during typhoons.

Keywords

• Reservoir operation
• Suspended sediment concentration
• SRH-2D
• Machine learning
• Switched forecasting
• Optimal integration