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

Area of Science:

• Environmental Science
• Analytical Chemistry
• Data Science

Background:

• Accurate water quality monitoring is crucial for digital city water management.
• Existing methods for measuring chemical oxygen demand (COD) face accuracy challenges under diverse environmental conditions.

Purpose of the Study:

• To develop a robust COD measurement method using ultraviolet-visible (UV-Vis) spectroscopy and machine learning (ML).
• To account for and remove interferences from temperature, pH, turbidity, and common ions.
• To assess the method's economic viability and environmental impact.

Main Methods:

• Utilized UV-Vis spectrum analysis combined with ML algorithms.
• Applied principal component analysis (PCA) for data processing.
• Employed Random Forest (RF) as the primary ML model for COD prediction.

Main Results:

• Achieved a low mean absolute percentage error (MAPE) of 6.73% for total, dissolved, and particulate COD.
• Demonstrated excellent transferability to new water samples with an average MAPE of 8.17%.
• Techno-economic assessment showed significant cost reductions compared to traditional methods (60.9% of lab, 49.3% of automatic stations).
• Life cycle assessment indicated a 31.32% reduction in environmental impact with ML integration.

Conclusions:

• The proposed UV-Vis and ML method offers accurate and cost-effective COD measurement for urban water systems.
• The approach demonstrates strong performance and transferability across different water environments.
• This method presents a feasible and sustainable solution for future digital water management.