Modelling of Activated Sludge Treatment Process in a Pulp Mill Using Neural Networks

The control of activated sludge treatment plants depends on on-line and laboratory measurements. Even though the behavior of activated sludge is basically known, long lags and the complex characteristic of biological activity set a challenge for controlling the process. In this paper MultiLayer Perceptron (MLP) neural network is applied for modelling Chemical Oxygen Demand (COD) concentration, which is one of the most important parameter of the activated sludge process. A data set of the process, extracted from databases of the pulp mill, was used for training neural networks, which could be used for forecasting the COD concentration of effluent. The results of validation show that the forecasting of COD concentration using the MLP models is good one day and at least moderate two days ahead. Therefore the results indicate that the MLP method can also be efficiently applied to the analysis of the activated sludge treatment process and so towards better process control.