Assessment of Effective Monitoring Sites in a Reservoir Watershed by Support Vector Machine Coupled with Multi-Objective Genetic Algorithm for Sediment Flux Prediction during Typhoons

摘要

Effectively assessing crucial monitoring sites with suspended sediment concentration (SSC) is a vital challenge for achieving accurate prediction of sediment flux on sluice gates at a dam in a reservoir watershed. To address this issue, an assessment framework based on a core concept of Data-Information-Knowledge-Wisdom (DIKW) hierarchy is proposed in this study. First, for the reasonable training of the coupled method, a two-dimentional layer-averaged density current model, SRH2D, is applied to simulate reasonable SSC data. The limited SSC data at monitoring sites collected from the field and at dam face, inflow, and outflow discharges are collected for validation of a calibrated numerical model. Second, a well-known data-driven method, Support Vector Machine (SVM), is coupled with Multi-Objective Genetic Algorithm (MOGA) as a sediment-flux-prediction (SFP) model in the proposed framework to evaluate effective monitoring sites with SSC. An application in the Shih-Men Reservoir is implemented to demonstrate the contribution of the proposed investigation framework. The results indicate that the spatial turbidity current movement is reasonably simulated by the numerical model and appropriate as reliable data for the SFP model. The SSCs at measured points located on the lower level at dam face are significantly higher. Moreover, the results also show that the simulated SSC at the monitoring sites located near the inflow point and dam face are relatively useful for SFP. The analyzed results are concluded that the well-established observation equipment at the inflow point and near the dam is necessary for obtaining high-quality measured data, which has become a significant key issue on reservoir operation management (ROM). Also, the proposed framework is expected to be helpful to improve the benefit of ROM as reference for decision makers.