Rainfall-Runoff Modeling Using a Self-reforming Neuro-Fuzzy System

Rainfall-runoff modelling is one of the fundamental hydrological problems which is crucial for many applications including water resources management and flood forecasting. To date, many Artificial Intelligence (AI) techniques have been proposed in modelling the dynamic hydrological processes such as the rainfall-runoff process, from which Neuro-Fuzzy Systems (NFS) have gained popularity due to their capability in capturing complex associations between inputs and outputs. However, conventional NFS models suffer from poor adaptability, underestimation of peak events, extensive rule base and inapplicability for real-time application. As such, this research proposes a novel self-reforming Neuro- Fuzzy System coupled with fuzzy rule interpolation/extrapolation (FRIE) and episodic memory mechanism. The proposed model aims to resolve the shortcoming of conventional NFS models in rainfall- runoff modeling and subsequently improve the model’s adaptability and performance in peak events estimation.