Hydrological modeling using a dynamic neuro-fuzzy system with on-line and local learning algorithm

Abstract

This paper introduces the dynamic neuro-fuzzy local modeling system (DNFLMS) that is based on a dynamic Takagi–Sugeno (TS) type fuzzy inference system with on-line and local learning algorithm for complex dynamic hydrological modeling tasks. Our DNFLMS is aimed to implement a fast training speed with the capability of on-line simulation where model adaptation occurs at the arrival of each new item of hydrological data. The DNFLMS applies an on-line, one-pass, training procedure to create and update fuzzy local models dynamically. The extended Kalman filtering algorithm is then implemented to optimize the parameters of the consequence part of each fuzzy model during the training phase. Local generalization in the DNFLMS is employed to optimize the parameters of each fuzzy model separately, region-by-region, using subsets of training data rather than all training data. The proposed DNFLMS is applied to develop a model to forecast the flow of Waikoropupu Springs, located in the Takaka Valley, South Island, New Zealand, and the influence of the operation of the 32 Megawatt Cobb hydropower station on spring flow. It is demonstrated that the proposed DNFLMS is superior in terms of model complexity and computational efficiency when compared with models that adopt global generalization such as a multi-layer perceptron (MLP) trained with the back propagation learning algorithm and the well-known adaptive neural-fuzzy system (ANFIS).