Parameter and state estimation of the activated sludge process: On-line algorithm

The Linearized Maximum Likelihood (LML) method for the simultaneous estimation of activated sludge states and parameters from noisy process measurements (Kabouris and Georgakakos, 1996a, Wat. Res., 30, 2853–2865) is simplified, in terms of its memory storage and computational requirements, for efficient on-line implementation. This is achieved by processing only the four most recent sets of 5-min on-line measurements at each estimation instance, along with the utilization of simplified estimation equations for tracking state and parameter variations, following the initial convergence period. The algorithm is tested in a computational case study involving a nitrifying activated sludge process, modelled by the IAWQ Activated Sludge Model 1 and incorporating a dynamic settling and clarification model. The on-line LML algorithm is capable of tracking the process states and parameters under dynamic conditions of process inputs and model parameters.