Hybrid Wavelet–Genetic Programming Approach to Optimize ANN Modeling of Rainfall–Runoff Process

Abstract

In this paper, the wavelet analysis was linked to the genetic programming (GP) concept for constructing a hybrid model to detect the seasonality patterns in the rainfall–runoff time process. This approach was used to determine the dominant input variables of an artificial neural network (ANN) rainfall–runoff model via a sensitivity analysis. In this way, the main time series of two variables, rainfall and runoff, were decomposed into some multi frequency time series by the wavelet transform. Then, these decomposed time series were imposed as input data to the GP to optimize the input structure of ANN model. This methodology was utilized in daily and monthly timescale modeling for two watersheds with distinct climatologic regimes. The obtained results were compared favorably to ANN and GP models. The obtained results showed that the proposed model can monitor both short and long term patterns due to the use of multiscale time series of rainfall and runoff data as the GP inputs. Moreover, using the proposed sensitivity analysis, the number of input variables in the ANN modeling was decreased.