Multi-machine learning methods for rapid and synergistic inversion of groundwater contamination source, hydrogeologic parameter and boundary condition.

Journal: Journal of contaminant hydrology
Published Date:

Abstract

The application of machine learning methods to the groundwater pollution inversion problem has become a hot research topic in recent years. However, applying machine learning methods to achieve synergistic and rapid identification of pollution source information, hydrogeological parameter, and boundary condition is much limited. This study proposed to use multi-machine learning methods, including: multilayer perceptron (MLP), kernel extremum learning machine, support vector machine (SVR), and back-propagation neural network, to directly establish the inverse mapping relationship between the outputs of the simulation model and the inputs, and to realize the synergistic identification of multiple variables to be identified. The recognition accuracies of different machine learning methods for different types of variables to be recognized were compared, and the methods with good inversion performance were combined. The results showed that the SVR method had excellent accuracy in identifying the hydraulic conductivity and specific head boundary. The MLP method had good accuracy in identifying the release intensity of the pollutant sources. Therefore, by combining SVR and MLP (SVR-MLP), SVR was used to construct an inverse mapping relationship identifying hydraulic conductivity coefficients and head-specific boundary values, and MLP was used to identify pollutant release intensities, thus having the synergistic identification of all three realized. Overall, SVR-MLP improved the overall inversion accuracy. In order to verify the reliability of the method, several sets of reference values were selected to assess the inversion performance of the method, and the average absolute percentage error of the identification results of the multiple sets was less than 4 %, which emphasized the stability and reliability of the inversion method. It can provide a reliable basis for groundwater pollution remediation and treatment.

Authors

  • Chengming Luo
    Guangdong Provincial Key Laboratory of Chip and Integration Technology, School of Electronic Science and Engineering (School of Microelectronics), South China Normal University, Foshan 528225, People's Republic of China.
  • Xihua Wang
    College of Civil Engineering, Tongji University, 1239 Siping Road, Shanghai 200092, China; Department of Earth and Environmental Sciences, University of Waterloo, ON N2L 3G1, Canada. Electronic address: 21531@tongji.edu.cn.
  • Y Jun Xu
    School of Renewable Natural Resources, Louisiana State University Agricultural Center, Baton Rouge, Louisiana, USA.
  • Qinya Lv
    College of Civil Engineering, Tongji University, 1239 Siping Road, Shanghai 200092, China.
  • Xuming Ji
    College of Civil Engineering, Tongji University, 1239 Siping Road, Shanghai 200092, China.
  • Boyang Mao
    College of Civil Engineering, Tongji University, 1239 Siping Road, Shanghai 200092, China.
  • Shunqing Jia
    College of Civil Engineering, Tongji University, 1239 Siping Road, Shanghai 200092, China.
  • Zejun Liu
    College of Civil Engineering, Tongji University, 1239 Siping Road, Shanghai 200092, China.
  • Yanxin Rong
    College of Civil Engineering, Tongji University, 1239 Siping Road, Shanghai 200092, China.
  • Yan Dai
    Laboratory of Veterinary Drug Development and Evaluation, College of Animal Science and Technology, Henan University of Science and Technology, Luoyang 471023, China.