Research On Heihe Groundwater Level Forecast Based On Grey Theory And Machine Learning

Due to the impact of climate changes and human activities,the increasing use of groundwater resources has led to serious groundwater overexploitation and serious environmental problems.Protecting the environment and the groundwater has become serious problems that we have to face with nowadays in our society.Accurate prediction of groundwater level change can provide better decision-making basis for groundwater resource management,and is conducive to the sustainable utilization of groundwater resources.However,the single model is difficult to meet the requirements of the water level prediction,hence the groundwater level prediction based on the hybrid model emerged,which can take into account the merits,drawbacks and working conditions of single model and has the ability to improve the prediction effectively.In this paper,the groundwater level prediction model based on Grey Model and Machine Learning method in the middle reaches of the Heihe River Basin is established,and it is applied to the prediction of groundwater level time series and the space-time restoration of missing data.The results show that the hybrid model based on Grey Model and Machine Learning method(BP neural network,RBF neural network,SVM,GVM)is good at predicting the variation trend and volatility of groundwater level in the middle reaches of Heihe River.Taking the Xingou observation site as an example,all the hybrid models are better than the single model.The IGM-GVM hybrid model has the best prediction results,in which the root mean square error is 0.3161 m and the mean absolute error is 0.2557 m.For the missing space-time data of groundwater level in the middle reaches of the Heihe River,the combined model has completed the data restoration.Taking the Yanuanzhangwan observation site as an example,the hybrid model has better missing data restoration performance than the traditional Kriging interpolation method.The GRA-GVM hybrid model has the best restoration performance,in which the root mean squared error is 0.2850 m and the mean absolute error is 0.2607 m.