Neural-Adaptive Control of a Biological Wastewater Treatment Process

Abstract

In a mixed liquor biological wastewater treatment process, it is very important to control the dissolved oxygen (DO) concentration. In this research work a robust control strategy is developed to maintain a DO set point in aerated bioreactors. The nonlinearities in the system are considered unknown and are modeled by an adaptive neural network. The neural network utilized is the Cerebellar Model Arithmetic Computer (CMAC). The weight update method tested in this thesis prevents weight drift and associated bursting, without sacrificing performance. The controller is tested on two models of an activated sludge process. The first one is a simple model with four state variables. The second simulation model is based on the benchmark simulation model number 1 (BMS1) for wastewater treatment plants, in which the activated sludge model no. 1 (ASM1) is considered to model the biological process. It uses real influent data files. The proposed controller outperforms the default PI controller that is tuned for dissolved oxygen control in the BSM1.