干旱区生态环境知识资源中心

This paper deals with an application of the Neuro-Wavelet (WNN) and Neuro-Fuzzy (ANFIS) systems for modeling the flow of a catchment located in Algeria. This catchment is prone to a semi-arid climate and a strong variability in runoff. The time series of its daily rainfall-runoff are split into two subsets: a training one and a testing one. In the first model, the series of rainfall and flows are decomposed into a succession of approximation and details using the discrete wavelet transform and used as inputs in a model of artificial neural networks. The second model corresponds to Adaptive Network-based Fuzzy Inference System (ANFIS), which generates an input-output based on both fuzzy rules and stipulated rainfall-runoff data pairs. The obtained results show that the performances of WNN and neuro-fuzzy models exceed those of neural network model.