Identification of groundwater contaminant source information based on ensemble surrogate model and genetic algorithm

Journal Information

J. Hefei Univ. Tech. (Nat. Sci.), Jun.2024, Vol.47 No.6: 731-738

DOI: 10.3969/j.issn.1003-5060.2024.06.002

Authors

LIU Meng $ ^{1} $, LUO Qiankun $ ^{1} $, AN Jimin $ ^{2} $, ZHAO Meng $ ^{1} $, QIAN Jiazhong $ ^{1} $

(1. School of Resources and Environmental Engineering, Hefei University of Technology, Hefei 230009, China; 2. Unit 93263 of PLA, Jinzhou 121000, China)

Abstract and Keywords

Abstract: Identifying contaminant source information accurately is the premise for efficient remediation of groundwater pollution. In order to solve the problem of computational burden in the process of identifying contaminant source information using traditional simulation-optimization methods, this paper firstly establishes back propagation neural network (BPNN) model, supports vector regression (SVR) model and kernel extreme learning machine (KELM) model to replace the groundwater flow and solute transport model in traditional simulation-optimization methods. Then, the simple average method and genetic algorithm (GA) are used to calculate the weight values, and the ensemble surrogate models are established to further improve the accuracy of the model. Finally, the best-performing ensemble surrogate model based on GA is embedded in the optimization model for identifying contaminant source information. The calculation results show that compared with the traditional simulation-optimization model, the calculation time of the optimization model embedded with the ensemble surrogate model based on GA is shortened from 11 d to 39 min and the contaminant source information parameters obtained by the solution are close to the true value, which can be used to solve the contaminant source information identification problem.

Keywords: groundwater contamination; contaminant source identification; surrogate model; genetic algorithm(GA)

作者信息

刘蒙 $ ^{1} $，骆乾坤 $ ^{1} $，安济民 $ ^{2} $，赵梦 $ ^{1} $，钱家忠 $ ^{1} $

(1. 合肥工业大学资源与环境工程学院，安徽合肥 230009；2. 中国人民解放军 93263 部队，辽宁锦州 121000)

摘要和关键词

摘要: 准确识别污染源信息是高效治理和修复地下水污染的前提。为解决使用传统模拟优化方法识别污染源信息过程中产生的严重计算负担问题，文章首先建立 BP 神经网络（back propagation neural network，BPNN）模型、支持向量回归（support vector regression，SVR）模型和核极限学习机（kernel extreme learning machine，KELM）模型代替传统模拟优化方法中的地下水水流和溶质运移模拟模型。然后，使用简单平均法和遗传算法（genetic algorithm，GA）计算权重值并建立集成替代模型进一步提高模型精度。最后，将表现最优的基于遗传算法建立的集成替代模型嵌入识别污染源信息的优化模型中。算例结果分析表明，嵌入基于遗传算法建立的集成替代模型的优化模型相较于传统模拟优化模型计算时间由 11 d 大幅度缩短至 39 min，且求解所得的污染源信息参数接近真实值，可用于解决污染源信息识别问题。

关键词: 地下水污染；污染源识别；替代模型；遗传算法（GA）

Funding

国家自然科学基金资助项目(41831289); 安徽省自然科学基金资助项目(1708085QD82)和中央高校基本科研业务费专项资金资助项目(JZ2018HGTB0251)

合肥工业大学学报(自然科学版)

Navigation