Machine Learning-Driven Dynamic Measurement of Environmental Indicators in Multiple Scenes and Multiple Disturbances.

Journal: Environmental science & technology
Published Date:

Abstract

Digital city water management systems require extensive data sensing for various environmental indicators, yet measurement accuracy often falls short under diverse extreme conditions. This study proposes a chemical oxygen demand (COD) measurement method based on ultraviolet-visible spectrum analysis and machine learning (ML), taking into account the removal of interferences, including temperature, pH, turbidity, common anions and cations, as well as COD composition and different water environments. The data collected from the river and wastewater were processed through principal component analysis, and random forest (RF) performed the best among the multiclass models with a mean absolute percentage error (MAPE) of only 6.73% for total COD (TCOD), dissolved COD (SCOD), and particulate COD (PCOD). RF has excellent transferability with an average MAPE of 8.17% for TCOD, PCOD, and COD in another real wastewater and river. Interpretability analysis elucidates the mechanism of PCA downscaling on the model. Techno-economic assessment revealed that this method incurs only 60.9% of the costs of laboratory monitoring and 49.3% of the costs of conventional automatic monitoring stations. Life cycle assessment showed that the introduction of ML can reduce environmental impacts by 31.32%. The study concludes with a discussion of the dynamic feasibility of this approach in future urban water systems.

Authors

  • Yu-Qi Wang
    State Key Laboratory of Urban Water Resource and Environment, School of Eco-Environment, Harbin Institute of Technology, Shenzhen 518055, China.
  • Han-Bo Zhou
    State Key Laboratory of Urban-Rural Water Resource and Environment, School of Eco-Environment, Harbin Institute of Technology, Shenzhen 518055, China.
  • Xiao-Qin Luo
    Department of Geriatrics, The Second Xiangya Hospital of Central South University, Changsha, China.
  • Shang-Wen Deng
    State Key Laboratory of Urban-Rural Water Resource and Environment, School of Eco-Environment, Harbin Institute of Technology, Shenzhen 518055, China.
  • Hao-Ran Xu
    State Key Laboratory of Urban-Rural Water Resource and Environment, School of Eco-Environment, Harbin Institute of Technology, Shenzhen 518055, China.
  • Yun-Peng Song
    State Key Laboratory of Urban Water Resource and Environment, School of Eco-Environment, Harbin Institute of Technology, Shenzhen 518055, China.
  • Jia-Ji Chen
    CAS Key Laboratory of Environmental Biotechnology, Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences, Beijing 100085, PR China.
  • Wan-Xin Yin
    College of the Environment, Liaoning University, Shenyang 110036, PR China; State Key Laboratory of Urban Water Resource and Environment, School of Civil and Environmental Engineering, Harbin Institute of Technology, Shenzhen 518055, PR China.
  • Hao-Yi Cheng
    State Key Laboratory of Urban-Rural Water Resource and Environment, School of Eco-Environment, Harbin Institute of Technology, Shenzhen 518055, China.
  • Ai-Jie Wang
    State Key Laboratory of Urban Water Resource and Environment, School of Civil and Environmental Engineering, Harbin Institute of Technology, Shenzhen 518055, PR China.
  • Hong-Cheng Wang
    State Key Laboratory of Urban Water Resource and Environment, School of Civil and Environmental Engineering, Harbin Institute of Technology, Shenzhen 518055, PR China. Electronic address: wanghongcheng@hit.edu.cn.