Development and optimization of a neural network model using genetic algorithm to predict the performance of a packed bed reactor treating sulphate-rich wastewater

Abstract

The performance of a packed bed reactor (PBR) containing immobilized sulphate reducing bacteria for sulphate removal from wastewater, utilizing carbon monoxide (CO) as the sole electron donor, is demonstrated. The performance of the PBR system in terms of CO and sulphate removal efficiencies (%RE_CO and %RE_sulphate, respectively) was predicted using three parameters, i.e. the hydraulic retention time (HRT, h), inlet concentrations of CO (IC_CO, mg/L) and sulphate (IC_sulphate, mg/L). An artificial neural network (ANN) model with 3-14-2 topology was developed by training the experimental data through the Levenberg Marquardt (LM) algorithm. Using genetic algorithm (GA) with appropriate objective functions, optimal sets of inputs were obtained to ensure maximum RE at a minimum HRT. The ANN had an overall accuracy above 98%, with a correlation coefficient of 0.99 and a root mean square error of 1.66%, suggesting its good performance. The automation of sulphate-rich wastewater industry through GA identified solutions might be leveraged for efficient operation in terms of saving time and resources.