Application of Hybrid Wavelet-SVM and ANN in the Prediction of Suspended Sediment Load of Aji-chay

Abstract

Simulation and evaluation of sediment and extending the relation among streamflow and sediment is an important and applicable issue in the water resource management which is useful in the organizing reservoirs, rivers and avoiding additional costs. In this paper, the efficiency of Wavelet-based Support Vector Machine (WSVM) model was examined for prediction of monthly Suspended Sediment Load (SSL) of the Aji-Chay River. Then for this purpose, at first step, SSL was predicted via ad hoc SVM and Artificial Neural Network (ANN) models. Therefore in the hybrid models, streamflow and SSL time series were decomposed into sub-signals via wavelet transform, and these decomposed subseries were imposed into SVM and ANN to simulate discharge-SSL relationship. The results showed that SVM yield to better outcomes with Determination Coefficient (DC)= 0.65 than ad hoc ANN with DC=0.61. On the other hand, WSVM showed better consequences than wavelet-based ANN (WANN) model in monthly SSL prediction, and wavelet data pre-processing could lead to catch more accurate results, e.g., DCLSSVM=0.65 was increased to the DCWLSSVM=0.82.

Keywords

• Prediction of Sediment
• Arteficial Neural network
• Support Vector Machine
• Wavelet
• Aji Chay