基于机器学习MLR模型的地下水循环井优化设计

Optimization Design of Groundwater Circulation Well Based on Machine Learning with Multiple Linear Regression Model

针对污染场地地下水循环井(groundwater circulating well, GCW)的优化设计问题,提出一种基于机器学习多元线性回归(multiple linear regression, MLR)模型的优化设计方法。该方法首先利用有限差分法建立不同条件下单个GCW运行的数值模型,通过运行数值模型,得到不同条件下GCW的运行效果,从而构建数据集;然后利用MLR算法对模型进行训练,构建计算多种GCW运行效果刻画指标的数学模型,并比较各个数学模型的拟合精度,结果显示纵向影响半径(RL)、横向影响半径(RT)模型的拟合程度较好,具有一定的泛化能力;最后根据机器学习所得的数学模型,对某试验场地GCW进行优化设计,得到最终优化设计方案,通过优化前的设计方案相比,RL、RT指标有了一定的提升,验证了方法的有效性。该研究结果可为GCW前期结构的快速设计提供参考,具有一定的实际意义。

For optimizing the structure of groundwater circulation well(GCW) in remediation site, the machine leaning with multiple linear regression(MLR) method is proposed in this study.Firstly, the groundwater numerical models in different conditions are developed by the finite difference method.The data set with GCW operation effect in different conditions is constructed by runninggroundwater numerical models.the numerical models for single GCW operation induced by in different conditions are developed by finite difference method.The data set with GCW operation effect in different conditions are constructed by running a large number of groundwater numerical models.Then MLR algorithm in machine learning is used to train to get the mathematical model constructed to calculate various indices for the operation efficacy of GCW.Compared with the results confirmed by numerical simulation, the mathematical model trained by MLR algorithm shows good consistent result, especially the longitudinal influence radius(RL) and transverse influence radius(RT).Finally, the optimal design of the GCW in the test site is carried out according to the mathematical model obtained by machine learning.The final optimal design scheme is confirmed by applying the model.Compared with the design scheme before optimization, the optimized RLand RTindicators are significantly improved.The study provides a reference for rapid design of the GCW structure.