Observer design for the activated sludge process

Övrigt konferensbidrag, 2010

The activated sludge process for degradation of organic matter is one of the main processes commonly used in biological wastewater treatment, and aeration in that process stands for a large part of the energy consumed in a plant. Hence, there has been many attempts to optimize the operation of the activated sludge process, which requires a model of the process. The advanced models used has in general their origin in IWA (former IAWQ) activated sludge model no 1 (ASM1). Unfortunately, feasible optimization is limited because several of the most important variables; bacterial biomass, readily biodegradable soluble substrate, and slowly biodegradable particulate substrate cannot be reliably measured online because of their complexity hiding behind their notation. One way to resolve this problem is to estimate these concentrations using an observer and other online measurements at hand. One of the developed methods is an Extended Kalman Filter (EKF) that estimates the relevant concentrations in the ASM1 based on oxygen measurements, and supplied air. For faster convergence, measurements of totally suspended solids in the influent flows are included in the algorithm. It is concluded that estimation does not work for one stirred tank alone, but when the activated sludge process is described by several tanks in series with oxygen measurements in each of them, the estimates converge. The filter has more interesting convergence properties, and to explain these observability properties are investigated. For an implementation of the observer it is necessary to estimate the oxygen transfer function and methods for this are evaluated and further developed. One of these and the EKF were evaluated for a real wastewater plant. The observer is not convergent for this, which among others can be explained by the many uncertainties regarding the model. The optimization problem is considered briefly and is solved for one control variable, and results based on real data are presented.