Sustainable Management Of River Water Quality Using Artificial Intelligence Optimisation Algorithms

Chia, See Leng (2021) Sustainable Management Of River Water Quality Using Artificial Intelligence Optimisation Algorithms. Final Year Project, UTAR.

PDF Restricted to Registered users only until 31 December 2023. Download (2783Kb)

Abstract

This study is aimed at proposing a superior artificial intelligence optimisation algorithm for sustainable river water quality management. The utilisation of the machine learning model can allow for predicting the water quality index (WQI) for the better management of water resources and more sustainable water supply. Least Square Support Vector Machine (LSSVM) base models with linear kernel, polynomial kernel and Radial Basis Function (RBF) kernel and its hybrid models with integration of Hybrid of Particle Swarm Optimisation and Genetic Algorithm (HPSOGA), Whale Optimisation Algorithm based on Self-adapting Parameter Adjustment and Mix Mutation Strategy (SMWOA) and Ameliorative Moth Flame Optimisation (AMFO) were developed and used to predict the WQI at stations 1K06, 1K07 and 1K08 of the Klang River in Selangor, Malaysia. River water quality data from 1999 to 2018 was utilised to generate 63 input data combinations for WQI predictions to compare the importance of water quality parameters. The performance was benchmarked using root mean squared error (RMSE), mean absolute error (MAE), Coefficient of Determination (R2 ), mean absolute percentage error (MAPE) and Global Performance Index (GPI) as well as their time cost. For the comparison of kernels, LSSVM with RBF kernel gave the best accurate WQI prediction than LSSVMs with linear kernel or polynomial kernel for all stations. LSSVM with linear kernel’s prediction resulted in negative R2 and unreliable for the predicted WQI. LSSVM with RBF kernel required more time cost while the time cost of the LSSVM with polynomial kernel was just slightly less than that of the LSSVM with RBF kernel. Among the hybrid models, in terms of accuracy, the best optimisation algorithm at station 1K06 was the AMFO while the best optimisation algorithm at station 1K07 was the HPSOGA. At station 1K08, the SMWOA was the optimisation algorithm with the most accurate prediction. For all stations, the time cost for the LSSVM-AMFO was the highest due to the kent chaotic strategy followed by the LSSVM-SMWOA and then the LSSVMHPSOGA. All optimisation algorithms in this study are quite competitive with each other in terms of prediction accuracy while RBF kernel is the best kernel type among the kernels in this study.

Actions (login required)