Research On Spatial And Temporal Prediction Method Of Groundwater Level In Changwu Area Based On KNN-LSTM Model

Groundwater is an important source of water supply in most areas of China.In recent years,with the rapid development of economy in our country,the overexploitation of groundwater has caused a series of environmental geological problems,such as water resources attenuation,land subsidence and ground fissures.Moreover,groundwater dynamics can reveal the state of water level,water quantity and water quality in aquifer changing with time,and it can reflect the development and utilization of groundwater resources.The dynamic monitoring data of groundwater level is a typical nonlinear and non-stationary geographic space-time data.Predicting the dynamic trend of groundwater level is an important basis for optimizing the allocation of water resources.While most of the current groundwater level prediction models are obviously insufficient in considering the time,space and space-time correlation.Therefore,it is necessary to construct a spatial-temporal prediction model of groundwater level taking into account the spatial-temporal factors to further improve the prediction accuracy of dynamic changes of groundwater level.Most importantly,the rational and scientific development and utilization of groundwater resources has important theoretical and practical significance for the sustainable development of groundwater resources utilization.Based on the consideration of spatial-temporal correlation of groundwater level data,this paper presents a spatial-temporal prediction model of groundwater level considering spatial-temporal factors,and realizes the prediction of groundwater regime in the study area.The main contents and conclusions are as follows:(1)The temporal and spatial characteristics of groundwater level monitoring values are analyzed and revealed.Based on temporal and spatial series analysis and geostatistical methods,combined with the monitoring data of water level of the second confined aquifer in Changwu area.By analyzing the discreteness,spatio-temporal correlation and heterogeneity of data,the nonlinearity and non-linearity of the groundwater level monitoring value is determined,which provides data characteristics for the selection and construction of the spatio-temporal prediction hybrid model of groundwater level.(2)The mixed model of spatio-temporal prediction of groundwater level considering spatio-temporal factors are constructed.Based on the spatio-temporal characteristics of groundwater level monitoring data and the characteristics of deep learning algorithms,the Wavelet Transform algorithm is used to remove noise in the original data,and the K-Nearest Neighbor(KNN)is to screen the monitoring wells in the study area for spatial correlation.The spatial correlation screening results are used to reconstruct the spatio-temporal data set and the Long Short-Term Memory(LSTM)algorithm,thereby constructing a groundwater level spatiotemporal prediction model KNN-LSTM that takes into account the temporal and spatial factors.(3)Prediction of spatio-temporal variation of groundwater level based on KNNLSTM model.The KNN-LSTM model is used to predict the groundwater level of the second confined aquifer in Changwu area.The spatial and temporal expression of groundwater level data is realized,and the dynamic change trend is revealed.At the same time,cross-validation method is used to verify the reliability and accuracy of the model.Compared with LSTM,support vector regression(SVR)and differential integrated moving average auto-regressive model,the prediction accuracy of KNNLSTM is improved by 20.68%,46.54% and 55.34% respectively,which proves that the prediction accuracy of KNN-LSTM is higher.