Hybrid unsupervised-supervised artificial neural networks for modelling activated sludge wastewater treatment plants

Mathematical modelling of wastewater treatment plant process is important for improving its treatment efficiency and thus the quality of the effluent released into the receiving water body. However, due to the highly complex and non-linear characteristics of this biological system, traditional mathematical modelling of the treatment process has remained a challenge. This work presents a hybrid modelling strategy based on the Kohonen Self Organising Map (KSOM) and feed-forward, back-propagation artificial neural networks (BP-ANN) for modelling the activated sludge wastewater treatment plant. The hybrid approach involved a 2-stage process: firstly, the KSOM was used for data preparation, visualisation of high dimensional data and features extraction; and secondly, these features were then used for the training and validation of the BP-ANN. Comparison of this hybrid modelling approach against the straight modelling of the original raw data using BP-ANN showed that the hybrid approach resulted in much more improved model performance.
The study demonstrated that the hybrid modelling strategy offers viable, flexible and robust modelling methodology for effectively handling noisy data for environmental systems modelling.