Performance evaluation of artificial intelligence with particle swarm optimization (PSO) to predict treatment water...

The prevention of water-borne diseases requires the disinfection of water consumed. Disinfection by-products, however, are an increasing concern, and they require advanced knowledge of water treatment plants before their release for human consumption. In this study, multivariate non-linear regression (MNR) and adaptive neuro-fuzzy inference system (ANFIS: Grid partition - GP and Sub-clustering - SC) integrated with particle swarm optimization (PSO) were proposed for the prediction of haloacetic acids (HAAs) in actual distribution systems. PSO-ANFIS-GP and PSO-ANFIS-SC were trained and verified for a total of 64 sets of data with eight parameters (pH, Temperature, UVA₂₅₄, DOC, Br^-; NH₄⁺-N; NO₂^--N, residual free chlorine). With MNR, R² is 0.51842<0.8181 and average absolute error, AAE = 0.172-0.253%). Further, PSO-ANFIS-GP and PSO-ANFIS-SC for HAAs (BCAA, DCAA, TCAA, HAA5, and HAA9) prediction consistently show higher regression coefficients, with lower AAE, (0.8825< R² < 0.9363 and AAE = 0.1031-0.1499%) and (0.8826< R² < 0.9322 and AAE = 0.1108-0.1500%) respectively with). With the eight simple parameters, PSO-ANFIS-GP provides an easy way for users to monitor and predict HAAs.