The Study On The Surrogate Model Of The Numerical Simulation Model Of Groundwater

Through theoretical study and engineering practice proved that simulation models can only solve the problem of prediction, and simulation and optimization can solve the groundwater remediation process to decision problem. Therefore, the optimization model that bases on the numerical simulation increasingly widespread attention by people. However, in order to ensure the optimization process is flowing the inherent principle and the discipline of groundwater. the optimization mode must be based on the simulate model. Hence the need for a way and means, the simulation model being embedded into the optimization model, making it an integral part of the optimization model, achieve the connection of simulation model and'optimization model. While the technical methods such as embedding method, response matrix, the state transition equation, and so on, that couple the simulation model and optimization model, that have their own scope of application and limitations in solving the problem that embed and call simulation model in the optimization.Surrogate model is an effective way can connect simulation model and optimization model. The groundwater surrogate model can approximate numerical simulation model in function, it can be directly called in the iterative process of optimization model. Not only overcome the limitations of coupling methods, but also significantly reduce the computational load in the process of optimization model calls simulation model, and save a lot of time. Therefore, the method of surrogate model has the potentiality and practical value to effective solve the problem way. more and more people will recognize and pay attention to its research and application.Surrogate model's quality depends on two key aspects:sampling methods and types of surrogate model. Choose an appropriate sampling method combining simulation model to obtain the input (pumping capacity)-Output (drawdown) sample data sets, is the basic premise to establish surrogate model. Therefore, Jinquan water for industrial park as the study area of this paper, based on the establishment of groundwater numerical simulation model. Monte Carlo method and Latin Hypercube method are applied to sample the pumping data of the region's five pumping wells, to determine the pumping test program, this will not only meet the practical needs and save costs, but also improve efficiency. In cases of selecting the same number of samples, comparing the two random sampling methods, the results showed that data obtained of Latin Hypercube Sampling is more representative. Therefore, use the results of the Latin hypercube sampling as the pumping test data program, and input the data to the groundwater numerical simulation model to run, obtaining the drawdown data sets of the ten observation wells.On the basis of the input and output data sets, this paper use regression analysis and artificial neural network, respectively establish Dual-response Surface Model and Radial basis function (RBF) Neural Network Model, those are two surrogate models of groundwater numerical simulation model, and test the effectiveness of the model. The results show that the obtained data from two surrogate models with the goal of data fitting results are well, to illustrating the two surrogate models that can approximate the function of numerical simulation model of groundwater.Finally, analyze the application process and results, those of Dual-response Surface Model and RBF Neural Network Model. Comparing with the Dual-response Surface Model, the results show that RBF Neural Network Model is more advantage and better prospects in the requirements of the calculation variables, with the goal of data fitting results, expressions and so on. Indicate that RBF Neural Network Model is more suitable for surrogating the numerical simulation model of groundwater.