A NEURAL NETWORK MODEL FOR CONTROL OF WASTEWATER TREATMENT PROCESSES

Abstract

This paper discusses the development of a neural network model for the prediction of the influent disturbances, which ultimately affect the activated sludge process. Neural networks are particularly suited to problems where there is no clear understanding of the processes and the complex inter-relationship between variables. The historical data used for training and testing the neural network is actual plant data obtained from a municipal plant and weather data for the same time periods. The result of the predicted influent disturbance is used in the control of the dissolved oxygen (DO). The results are applied to a pilot wastewater treatment plant located at the Cape Peninsula University of Technology (CPUT). The number of and the type inputs are varied to find an optimal model in order to predict the Chemical Oxygen Demand (COD), Total Kjeldahl Nitrogen (TKN) and the flowrate. Three different dynamic multilayer perceptron (MLP) feed-forward neural network models are developed for the influent disturbances of COD, TKN and flowrate respectively.