Reservoir turbidity forecasting coupled adaptive network-based fuzzy inference system

Abstract

The purpose of this study is to apply ANFIS (Adaptive Network-based Fuzzy Inference System) and BPNN (Back-Propagation Neural Network) with a coupled and non-coupled structure to construct suitable reservoir turbidity forecast models for short and long lead times. The proposed models can be used by reservoir operators to predict reservoir turbidity while releasing water during typhoon periods. The study site was at Shih-Men reservoir. The Lung-Chu-Wan and the second pumping stations were selected as the prediction locations. Results showed that the coupled structures in the ANFIS and BPNN models demonstrated superior tolerance and ability to handle predictive errors in the stable flow regime compared to those in the turbulence flow regime when forecasting reservoir turbidity. Furthermore, the precision of predicting turbidity and stability was better with BPNN than with ANFIS. However, the training CPU time needed in constructing BPNN was sixty times greater than that of ANFIS.