A neural network based predictive controller for wastewater treatment process

Abstract

Due to the high time-delays and non-linearity in the biological systems, the wastewater treatment processes based on activated sludge are often difficult to control. Furthermore, the aim of operating wastewater treatment plants as a resource recovery system increases the demand on the use of optimal control strategies. Model-based predictive control has a number of advantages compared to a conventional feedback controller due to its ability to take control action based on optimising an objective function such as reducing operational costs, better effluent quality and improved resource recovery. This work presents the idea of a using a time variant artificial neural network model as a predictive controller. The time series artificial neural networks (ANN) model has the ability to capture system dynamics without additional computational load and therefore, it can be used as an effective prediction tool for estimating plant performance. In the following work, the use of ANN based controller for the maintaining a fixed nitrate concentration in the anoxic chamber has been demonstrated. The Benchmark Simulation Model No.1 (BSM 1) is used to test the controller performance. The advantages of this predictive controller over a conventional feedback PI controller are also presented.