基于替代模型和流向算法的地下水污染源反演识别  

Inversion and identification of groundwater pollution sources based on surrogate model and flow direction algorithm

作　　者：罗成明 卢文喜 潘紫东 王梓博 徐亚宁 白玉堃 LUO Cheng-ming;LU Wen-xi;PAN Zi-dong;WANg Zi-bo;XU Ya-ning;BAI Yu-kun(Key Laboratory of Groundwater Resources and Environmental Ministry of Education,College of New Energy and Environment,Jilin University,Changchun 130012,China)

机构地区：[1]吉林大学新能源与环境学院,地下水与资源环境教育部重点实验室,吉林长春130012

出　　处：《中国环境科学》2023年第11期5823-5832,共10页China Environmental Science

基　　金：国家自然科学基金资助项目(42272283,41972252);吉林大学研究生创新基金资助项目(2022186)。

摘　　要：应用模拟-优化的理论和方法,对地下水污染源的相关信息、模拟模型的渗透系数以及抽水井的抽水量进行同步识别.首先,根据假想例子构建地下水污染数值模拟模型.然后,分别运用BP神经网络(BPNN)方法和核极限学习机(KELM)方法构建模拟模型的替代模型,优选出拟合精度更高的替代模型嵌入到后续的优化模型中,以此减少计算负荷并提升替代模型对模拟模型的逼近精度.最后,采用流向算法(FDA)对优化模型进行求解,得到反演结果,同时,将其分别与麻雀搜索算法(SSA)和粒子群优化算法(PSO)得到的反演结果进行对比.结果表明:相比于KELM替代模型,BPNN替代模型的拟合精度较高,确定性系数、平均相对误差和均方根误差分别为0.9999、0.1723%和0.5625;与PSO和SSA相比,FDA的收敛速度更快,对优化模型的求解精度更高,其识别结果的平均相对误差小于7%,提升了地下水污染源反演识别的精度和效率,能够为地下水污染修复、风险评定和责任认定提供可靠的依据.In this paper,the theory and method of simulation-optimization was applied to identify the relevant information of groundwater pollution sources,the hydraulic conductivities of the simulation model,and the pumping capacity of pumping wells simultaneously.First,a numerical simulation model of groundwater contamination was constructed based on a hypothetical example.Then,the BP neural network(BPNN)and kernel extreme learning machine(KELM)methods were applied to construct surrogate models of the simulation model,and the surrogate model with better fitting accuracy was selected and embedded in the subsequent optimization model to reduce the computational load and improve the approximation accuracy of the surrogate model to the simulation model.Finally,the inversion results were obtained by solving the optimized model with flow direction algorithm(FDA)and comparing them with those obtained by sparrow search algorithm(SSA)and particle swarm optimization(PSO)respectively.The results showed that compared with the KELM surrogate model,the BPNN surrogate model had higher fitting accuracy,with the coefficient of determination,the average relative error and the root mean square error of 0.9999,0.1723%and 0.5625,respectively;compared with PSO and SSA,FDA had faster convergence speed and higher accuracy in solving optimization model.The average relative error of its identification results was less than 7%,which improved the accuracy and efficiency of groundwater pollution source inversion identification and provided a reliable basis for groundwater pollution remediation,risk assessment and liability determination.

关 键 词：同步识别 模拟-优化方法 替代模型 流向算法 BP神经网络 

分 类 号：X523[环境科学与工程—环境工程]