Variable infiltration capacity model with BGSA-based wavelet neural network

In this study, we developed a hybrid form of rainfall-runoff model by integrating the variable infiltration capacity (VIC) model with a wavelet neural network (WNN) based on the binary gravitational search algorithm (BGSA). The streamflow of each subbasin in the Jinshajiang River Basin was first simulated by VIC model, then the simulated runoff of each subbasin and antecedent total basin runoff were decomposed via discrete wavelet transformation into a number of subseries components with different time scales. Finally, BGSA was employed to optimize the number of hidden layers and identify the appropriate subset of WNN inputs from a set of candidate subseries components. The proposed VIC_BGSA_WNN model was then compared to the traditional VIC model and reference methods based on correlation to determine effective wavelet components, and results indicated that our approach is feasible and effective.