Optimization of the Wastewater Treatment Plant Aeration Using Artificial Neural Networks Models

The present work aimed the modelling and optimization of the Wastewater Treatment Plant (WWTP) operation based on predicting its energy and quality performance indices using Artificial Neural Networks (ANNs). The best model architecture and structure were searched among three different ANN types, with different topologies. A standard dataset originating from the plant calibrated first-principle model (FPM) data was used to develop the ANN models. Their performance was evaluated by the coefficient of determination and mean squared error (MSE) values, first at testing and subsequently at the prediction performed for a new input dataset. Using the most promising identified ANN types and topologies, two ANN structures were investigated, one with three single output neural networks and another one with a single network with three outputs for predicting WWTP performance indices: aeration energy, effluent quality and pumping energy. The analytical model and the two ANN structures were used in the study of the aeration optimization of the WWTP, for finding the optimal air distribution in the aerated reactors. The obtained results were tested and compared taking into account the performance index values as well as the required computation time. The developed ANN models showed similar results to the FPM in terms of performance indices, while the required computation time was reduced by several orders of magnitude.