Application of Dual-Response Surface Methodology and Radial Basis Function Artificial Neural Network on Surrogate Model of the Groundwater Flow Numerical Simulation
Application of Dual-Response Surface Methodology and Radial Basis Function Artificial Neural Network on Surrogate Model of the Groundwater Flow Numerical Simulation
复制标题
DOI：
10.15244/pjoes/68854
复制
发表时间：
2017-07
影响因子：
1.8
通讯作者：
Yanping Yi;Wenxi Lu;Defa Hong;Hongchao Liu;Lei Zhang
中科院分区：
文献类型：
--
作者：
Yanping Yi;Wenxi Lu;Defa Hong;Hongchao Liu;Lei Zhang
替代模型是地下水流数值模拟中连接模拟和优化模型的有效途径；克服了传统方法在优化模型中嵌入和调用仿真模型的局限性，大大减少了优化模型求解过程中直接调用仿真模型带来的计算量。本文应用双响应面法和径向基函数人工神经网络法建立了内蒙古金泉工业园区地下水流数值模拟的代理模型。采用拉丁超立方采样法确定5口抽水井的随机抽水负荷，并以此为输入数据地下水流数值模拟模型计算10个观测井的降水数据集（输出数据集）。基于输入和输出数据集，采用双响应面法和径向基函数人工神经网络法建立地下水模拟模型的替代模型，并对比检验替代模型的有效性。结果表明，两个替代模型的结果与模拟模型的结果吻合较好，说明两个替代模型能够逼近地下水流数值模拟模型；与双响应面模型相比，RBF神经网络模型在样本量要求、拟合模拟结果精度方面更具优势。
The surrogate model is an effective way to connect the simulation and optimization models in groundwater flow numerical modeling; it could overcome the limitations of embedding and calling simulation models in the optimization model by conventional methods, which greatly reduces the computational load caused by directly calling the simulation model in the solving process of the optimization model. In this paper, the dual-response surface method and radial basis function artificial neural network method were applied to establish the surrogate model of groundwater flow numerical simulation in Jinquan Industrial Park, Inner Mongolia, China. The Latin hypercube sampling method was used to determine random pumping load of the five pumping wells, which were taken as the input data groundwater flow numerical simulation model for calculating 10 observation wells drawdown data sets (output data sets). Based on the input and output data sets, the dual-response surface method and radial basis function artificial neural network method were used to establish the surrogate model of groundwater simulation model, and the validity of surrogate models were comparatively tested. The results showed that both the results of two surrogate models fit well with the results of the simulation model, which indicates that two surrogate models were capable of approaching the groundwater flow numerical simulation model; compared with the dual response surface model, the RBF neural network model had more advantages in terms of sample size requirements, fitting the accuracy of simulation results.