Author(s): B. Raduly; K. V. Gernaey; A. G. Capodaglio; P. S. Mikkelsen; M. Henze
Linked Author(s): Peter Steen Mikkelsen
Keywords: NN; Modelling; Performance evaluation; Time series; WWTPKeywords
Abstract: Wastewater treatment plants (WWTPs) are an essential component of the integrated urban water system. The simulation of the plant behaviour is increasingly becoming an essential tool for design, evaluation and everyday operation of the plants. Rather long influent time series containing a wide range of influent disturbances are needed to allow a simulation-based WWTP performance evaluation of sufficient quality, but this requires long simulation times. The approach proposed in this paper combines an influent disturbance generator with a deterministic WWTP model for generating a limited sequence of training data (6 months of dynamic data). Artificial neural networks (ANNs) are then trained on the available WWTP input-output data, and are subsequently used to simulate the remainder of the influent time series (20 years of dynamic data) generated with the influent disturbance generator. It is demonstrated that the ANN reduces simulation time by a factor of 35, even when including the time needed for ANN training. ANN prediction of effluent ammonium, BOD5 and total suspended solids was good when compared to deterministic WWTP model predictions (correlation coefficient > 0. 95), whereas prediction of effluent COD and total nitrogen concentrations was less satisfactory (correlation coefficient > 0. 80) but still within acceptable limits.
Year: 2005