Evaluation of daily rainfall-runoff model using multilayer perceptron and particle swarm optimization feed forward neural networks

Abstract

In recent years, Artificial Neural Networks (ANNs) have been successfully used as a tool to model various nonlinear relations, and the method is appropriate for modeling the complex nature of hydrological systems. They are relatively fast and flexible, and are able to extract the relation between the inputs and outputs of a process without knowledge of the underlying physics. ANNs with sufficient hidden units are able to approximate any continuous function to any degree of accuracy by performing efficient training. In this study, two types of ANNs, namely, the multilayer perceptron neural network (MLP) and the newly developed particle swarm optimization feedforward neural network (PSONN) are applied to model the daily rainfall-runoff relationship for the Bedup Basin, Sarawak, Malaysia. Various models are investigated in searching for the optimal configuration of ANNs. Results are evaluated using the coefficient of correlation (R) and the nash-sutcliffe coefficient (E2). With the input data of current rainfall, antecedent rainfall and antecedent runoff, MLP simulated the current runoff perfectly for training with R=1.000 and E2=1.000, and R=0.911 and E2=0.8155 for testing data set. Meanwhile, PSONN also simulated the current runoff accurately with R=0.872 and E2=0.7754 for training data set, and R=0.900 and E2=0.8067 for testing data set. Thus, it can be concluded that ANNs are able to model the rainfall-runoff relationship accurately. The performance of the newly developed PSONN is comparable with the well-known MLP network, which had been successfully used to model rainfall-runoff for the Bedup Basin.