Groundwater Level Prediction Based on BP and RBF Neural Network

Groundwater level is an important indicator to measure groundwater resources and their exploitation amount. The accurate prediction of groundwater level is important for efficient use and management of groundwater resources. Because of mainly affected by natural and human factors, groundwater level has evident randomness. So, building stochastic model for prediction of groundwater level is of great significance in the evaluation of groundwater resources. In the paper, BP and RBF neural network models are built and they are applied in Yichang Irrigation District of Hetao Irrigation District in Inner Mongolia. Forecasting the groundwater level fluctuations in the irrigation district can provide references for many aspects, such as saving groundwater resources, restoring groundwater homeostasis in the region, establishing the optimum irrigation system of well irrigation, developing water-saving irrigation and promoting the sustainable development of agriculture and water resources. Overall, simulation results of the neural network models suggest that predictions of two models are reasonably accurate. The average absolute value of relative error of BP neural network is 5.28% and RBF neural network is 4.84%. Comparative analysis shows that RBF neural network is simpler, converges faster and has more stable prediction results.