Study On Groundwater Level Prediction Method Of Heihe River Basin Based On Multi Model Fusion

Groundwater resources are widely used in residential life,industry,farmland irrigation and other fields,and play an important role in production,life and economic development.Effective prediction of groundwater level can provide strong support for water resource management.In this paper,the task of groundwater level prediction is divided into short-term prediction and medium and long-term prediction.The short-term prediction is helpful to understand the short-term dynamic change of water level in time,and the prediction accuracy is higher.The medium and long-term prediction provides the decision-makers with the trend of water level change in the future,which is relatively more challenging.In order to solve the problems of data volatility and spatiotemporal dynamics in groundwater level prediction,a depth neural network water level prediction model based on multi model fusion is proposed in this thesis.Taking the groundwater level data in the middle reaches of Heihe River as an example,a better prediction effect is achieved.In this thesis,a time series water level prediction model based on attention mechanism is proposed,and the experimental results show that the model can get higher prediction accuracy in short-term prediction and medium-term and long-term prediction than auto-regressive model,traditional machine learning method and single neural network model.In order to make more effective use of the spatiotemporal data of groundwater level,this thesis further proposes a prediction model of groundwater level based on spatiotemporal attention mechanism,which uses multi-level spatiotemporal attention mechanism to model the spatiotemporal dynamics and external characteristics of groundwater level.Through a large number of experiments and effect evaluation,our spatiotemporal attention model has greatly improved in the three evaluation indexes of RMSE,MAE and R2.The three indexes of short-term prediction are 0.0754 m,0.0952 m and 0.4053,which is 28.65% higher than the average prediction accuracy of regression support vector machine.In terms of medium and long-term expectation,the three indexes reached 0.0916 m,0.1104 m and 0.3051,which improved the average prediction accuracy by 41.95% compared with the difference moving average auto-regressive model.The neural network model,which integrates spatiotemporal attention mechanism,can effectively realize the prediction of groundwater level,and the prediction accuracy is greatly improved compared with the traditional method,which is of great significance to the protection and management of groundwater resources.