Machine learning based computational methods to study source water quality of Macau

Abstract

Main Storage Reservoir (MMR) water is one of drinking water resources for Macau residents. In order to examine the impact of natural processes on MMR water quality, this project assessed chemical oxygen demand (COD) and total dissolved solids (TDS) contents from January 2008 to April 2019. Computational efforts have been undertaken by adopting various machine learning based methods including correlation support vector machine (SVM), genetic algorithms (GA), grid search (GS) as well as sensitivity analysis. Single SVM model, hybrid GA-SVM model and hybrid GS-SVM model deals with the simulation tasks in a prediction mode by considering the current monthly water quality parameters. These said models handle computations in a forecast mode by adopting monthly water quality parameters from the last two months. The simulated results suggest that GA-SVM model and GS-SVM model demonstrated an enhanced performance than the single SVM model. Subsequently, sensitivity analysis identified the most important water quality parameters contributing to COD and TDS content as being nitrate, suspended solids and total organic carbon. The project provides a deeper insight on the interaction among various water quality parameters related to COD and TOC and offers a new computational approach in water quality monitoring.