Improved differential evolution based BP neural network for prediction of groundwater table

Abstract

Groundwater table often shows complex nonlinear characteristic. Back Propagation (BP) neural network is increasingly used to predict groundwater table. However man-made selecting the structure of BP neural network has blindness and expends much time, so differential evolution (DE) algorithm was adopted to automatically search BP neural network weight matrix and threshold matrix. In order to improve the convergence of DE algorithm, a chaotic sequence based on logistic map was introduced to self-adaptively adjust mutation factor. Furthermore, a self-adapting crossover probability factor was presented to improve the population's diversity and the ability of escaping from the local optimum. Study case shows that, compared with groundwater level prediction model based on traditional BP neural network, the new prediction model based on DE and BP neural network can greatly improve the convergence speed and prediction precision.