基于自组织映射(SOM)算法的地下水污染源反演

Groundwater Contaminant Source Inversion Based on Self-Organizing Map Algorithm

根据污染物浓度监测数据进行地下水污染源反演是一类典型的地下水逆问题,通过对地下水污染源位置与强度的确定有助于提高地下水污染治理与修复的效果。提出一种基于自组织映射(SOM)算法的地下水污染溯源模型的替代模型的构建方法,并通过数值算例研究了SOM替代模型中训练数据样本数目和神经元数目的最优组合,以得到能够准确地表征地下水污染源强信息与观测数据之间联系的最优替代模型。算例研究结果表明:基于SOM算法的替代模型作为一种快速反演算法,在应用于地下水污染源反演问题时能有效减少计算量,并能以较高的精度对地下水污染源进行快速识别,具有良好的稳定性和鲁棒性。

Groundwater pollution source identification based on concentration monitoring data is a typical groundwater inversion problem.The determination of the location and intensity of groundwater pollution sources can help improve the effectiveness of groundwater pollution control and restoration.This study puts forward a method to construct surrogate model of the groundwater contaminant source identification model using a self-organizing map(SOM)algorithm,and obtains the optimal combination of the number of training data and the number of neurons in the SOM-based surrogate model through numerical examples.The optimal surrogate model can accurately represent the relationship between source intensity information and observed data of the groundwater pollution.The results show that the proposed SOM-based surrogate model is a fast inversion algorithm,and can identify pollution sources with high accuracy and has good stability and robustness.