Prediction of Groundwater Quality Indexes Using the Linear and Non-linear Model

As an important part of the ecological civilization construction, the protection of the groundwater environment should take prevention as the main and remediation as the secondary. To prevent the groundwater pollution crisis in advance, short-term prediction models are established to grasp the groundwater quality changes. The purpose of this study is to investigate the suitability of linear and non-linear prediction models for groundwater quality indexes in Pinggu Plain, Beijing. The water quality indexes in monitoring wells W1 and W2 are selected as research objects, autoregressive integrated moving average (ARIMA) model and backpropagation neural network (BP) model are constructed to predict Cl−, SO42− and TDS concentrations in these two wells, respectively. Furthermore, the ARIMA and BP models of monitoring well W2 are combined with equal weighting method and optimal weighting method. The results show that the linear and nonlinear prediction effects of W1 are unsatisfactory, while both ARIMA and BP models extract effective water quality change information to a certain extent in W2. The prediction accuracy of the ARIMA-BP model based on the combination of different methods is higher than that of the single model, which verified the superiority of the combined model prediction effect. The relevant conclusions can provide important theoretical and methodological support for groundwater pollution prevention and control in Pinggu District and have important practical significance for promoting sustainable utilization of groundwater resources.