Flood Flow Modeling in a River System Using Adaptive Neuro-Fuzzy Inference System

Abstract

This paper presents the application of a data-driven model, Adaptive Neuro-Fuzzy Inference System (ANFIS) in forecasting flood flow in a river system. ANFIS uses neural network algorithms and fuzzy reasoning to map an input space to an output space. In the present study, ANFIS models are used to forecast common downstream flow rates and flow depths in a river system having multiple inflows. Three different ANFIS model forms: (i) depth-depth (H-H) model, (ii) depth-discharge (H-Q) model and (iii) discharge-discharge (Q-Q) models are considered in this study. The models are used for forecasting one-hour ahead common downstream flow rates and flow depths in a river system based on past upstream flows. The flow and flow depths data are divided arbitrarily into different categories (2, 3, 4, 6) and different number of membership functions (Triangular, Gaussian, Trapezoidal and Bell) selecting two categories with Gaussian input and constant output membership functions based on trial and error. Performances of the ANFIS model with selected categories and membership functions are tested and verified by applying a time-series model, Autoregressive Integrated Moving Average (ARIMA) to the same river system. ARIMA has been successfully used in time-series forecasting leading to satisfactory performances. A further validation of the ANFIS model has been done by applying it to another river basin, Tar River Basin in USA. The results evaluated on the basis of standard statistical criteria showed improved performances by the ANFIS depth-depth forecasting models. The results also indicate that performances of the ANFIS models with multiple inflows are more satisfactory and closely follow performances of the ARIMA models. The study demonstrates applications of the multiple inflows ANFIS models in forecasting downstream flood flow and flow depth in a river system.