An integrated dynamic model for simulating a full-scale municipal wastewater treatment plant under fluctuating conditions

In this study, an integrated dynamic model was developed through combining a mechanistic model, a neural network (NN) model and a genetic algorithm approach, in order to simulate the performance of a full-scale municipal wastewater treatment plant (WWTP) with substantial influent fluctuations. As the base of the integrated model, the mechanistic model was initially established based on the activated sludge model 3 and the EAWAG bio-P module, and was used to generate the residuals for the NN model. The NN model was employed to build the relationship between the input and output variables. The network weights of the NN model were optimized with a genetic algorithm approach. The resulting integrated model was applied to simulate the 5-month performance of a full-scale WWTP with significant influent fluctuations, and the simulation results matched the measured ones of the WWTP well even under influent disturbance conditions. Compared with the individual mechanistic model and NN model, the integrated model was able to capture sufficient residual information to compensate for the inaccuracy of the mechanistic model and improve the extrapolative capability of the NN model. This model established in our work is demonstrated to be an effective and useful tool to simulate the performance of WWTPs.