Numerical Simulation Of Ground Water Problem Using A Deep Learning Approach

Regional groundwater problem is one of the important issues in the field of groundwater resources management and protection.The unreasonable exploitation of groundwater has become nowadays the main reason causes geologic hazard such as sink of ground and subsidence.Understanding the regional movement of groundwater has always played an important role in predicting the flow of groundwater and controlling the exploitation of groundwater,which often involves the numerical simulation of regional groundwater model.As we all know,an important issue in regional groundwater simulation is scale disparity.It often brings many difficulties and challenges to the numerical simulation.For example,in an aquifer system where the hydraulic conductivity changes by an order of magnitude,when a standard numerical method is used to solve the equations,the degrees of freedom of the resulting discrete system can be extremely large,due to the necessary resolution for obtaining meaningful solutions,which will consume tremendous amount of memory storage and computing time.In recent years,many researchers have done a lot of fruitful work and have also developed some mature simulation software.In these works,some pay attention to the grid-based numerical model,but this kind of method often has some limitations.For example,when dealing with the problem of pumping wells,it is not easy to design an appropriate numerical algorithm to match the grid;some focus on meshfree numerical methods,such as the radial basis function method(RBF)and the Monte Carlo method(MCM).The drawbacks of these methods compared to traditional mesh based methods are for example numerical stability for RBF and inefficiency for MCM.In this thesis,a new deep learning method called “GWPINN” is designed to simulate the ground water problem and different sampling methods and training strategies are proposed,which can make up for the shortcomings of traditional numerical simulation methods to a certain extent.This thesis also designed different test cases to train GWPINN,and compared it with the well-known groundwater numerical simulation software MODFLOW.The results demonstrate that GWPINN has strong ability in resolving the issue of scale disparity in ground water problem.