Study on Optimal Allocation of Water Resources Based on Surrogate Model of Groundwater Numerical Simulation

Abstract

The characteristics of groundwater systems are highly complex. It will take substantial computational resources and running time to optimize a groundwater numerical simulation model. In this study, in order to realize the coupling of simulation and optimization models, the improved backpropagation (BP) neural network was used as a surrogate model of a groundwater numerical simulation; the improved BP neural network was trained with the groundwater level drawdown–pumping volume data output of the simulation model. The method was applied to the water resource optimal allocation in the near future of Wenshang County, Shandong Provence of China. The results show that the water level drawdown output of the improved BP neural network model fits the results of the simulation model well, showing that the improved BP neural network can effectively be the surrogate of a groundwater numerical simulation to be embedded in an optimization model. The improved simulation and optimization technique can make full use of water resources in the whole area. Under an assurance rate of 50%, both water shortage and water shortage rate reduced to zero in the whole area. Under an assurance rate of 75%, water shortage and water shortage rate reduced to about 10% of the conventional scheme, which dramatically improves the comprehensive benefit of the whole area.