Speed up grid-search for Kernels selection of support vector regression

Abstract

The aerobic granular sludge (AGS) is a one of the promising technologies for wastewater treatment. In this paper, several modelling strategies are developed to predict the behaviour of AGS. The modelling approaches are cautiously chosen to address the complex dynamic of AGS due internal interactions between the sludge characteristics and variables. Since only a small dataset is available, the support vector regression (SVR) method is employed. Instead of using the time-consuming and trial-and-error or grid search methods to determine the pair of kernels, the particle swarm optimization (PSO) and genetic algorithm (GA) techniques are proposed. Using a dataset generated from an AGS process in sequential batch reactor at a working temperature 30 ˚C, the SVR-PSO, SVR-GA and SVR-Grid Search predict models are developed and compared. The results show that the proposed SVR-PSO and SVR-GA models improve the prediction accuracy of chemical oxygen demand (COD) by 10% as compared to the conventional SVR-Grid Search model. The computational time also was reduced up to 86% and 79% respectively.