Differential evolution-optimized general regression neural network and application to forecasting water demand in Yellow River Basin

Water demand of Yellow River Basin is influenced by many kinds of factors. General regression neural network (GRNN) was adopted to model the non-linear relationship between the factors and water demand. Depending on smoothing parameter, the prediction performance of GRNN can vary considerably. The most common methods for determining a suitable value of smoothing parameter are based on trial-and-error according to experience. In order to improve GRNN prediction performance, differential evolution (DE) algorithm is used to optimize GRNN and determine optimal value of smoothing parameter. For the purpose of improving the convergence and the ability of escaping from the local optimum, chaotic sequence based on logistic map was employed to self-adaptively adjust mutation factor of DE algorithm. The model of DE-optimized GRNN was employed to forecast industrial water demand, agricultural water demand and domestic water demand in Yellow River Basin. The result reveals that, compared with Back propagation based on Genetic algorithm model and GRNN based on Genetic algorithm prediction model, the new prediction model of DE-optimized GRNN is reasonable.