Hierarchical neurofuzzy model for real-time flood forecasting

Abstract

The current study employs a hierarchical adaptive network-based fuzzy inference system for flood forecasting by developing a rainfall–runoff model for the Narmada basin in India. A hybrid learning algorithm, which combines the least-square method and a back propagation algorithm, is used to identify the parameters of the network. A subtractive clustering algorithm is used for input space partitioning in the fuzzy and neurofuzzy models. The model architectures are trained incrementally each time step and different models are developed to predict one-step and multi-step ahead forecasts. The number of input variables is determined using a standard statistical method. An artificial neural network (ANN) model which uses an Levenberg–Marquardt (LM) backpropagation training algorithm has been developed for the same basin. The results of this study indicate that the hierarchical neurofuzzy model performs better compared to an ANN and the standard fuzzy model in estimating hydrograph characteristics, especially at longer forecast time horizons.

Keywords:

• hierarchical neurofuzzy model
• Takagi–Sugeno fuzzy model
• subtractive clustering
• flood forecasting