| The research aim was to develop a mathematical model for predicting the behaviour of an activated sludge plant treating pulp and paper wastewater. A one-dimensional mechanistic model for the primary and secondary clarifiers was selected. The Activated Sludge Model No. 1, selected for the bioreactor, was modified to include components and processes related to activated sludge treatment of pulp and paper wastewaters. The mechanistic models were calibrated against data originating from full-scale facilities using genetic algorithms. The mechanistic models were connected to neural networks to form hybrid models.; The response of the primary clarifier mechanistic model was fair for both overflow and underflow suspended solids. The hybrid model did not improve the mechanistic model response.; The secondary clarifier mechanistic model response was very good for the underflow suspended solids and poor for the overflow suspended solids. A hybrid model improved the secondary clarifier mechanistic model in predicting overflow suspended solids concentrations. The neural network model introduced pH and BODS as variables related to clarification.; The activated sludge mechanistic model predictions for effluent COD, mixed liquor suspended solids, phosphorus and oxygen uptake rate were very good. The model response was acceptable for nitrite plus nitrate, but inadequate for ammonia.; The steady-state neural network model did not improve the activated sludge mechanistic model predictions. Temperature- and pH-dependent growth and decay rates were introduced in the activated sludge mechanistic model to gain an insight into the impact of variable model parameters on the overall model predictions. A conclusion was that a reason for using the pH- and temperature-dependent parameters existed only if the model responses for readily biodegradable COD and oxygen consumption were sought.; The introduction of different coefficients representing the COD content of active biomass and a remaining portion of the mixed liquor suspended solids provided better model predictions for effluent COD and mixed liquor suspended solids.; The calibration results indicated the importance of having long term, dynamic data for both influent and effluent COD fractions to improve model accuracy. Thus, a need for extending full scale plant measurements to include influent and effluent COD fraction measurements was recognized. |
