ROBUST PARAMETER-ESTIMATION IN ONLINE OPTIMIZATION .1. METHODOLOGY AND SIMULATED CASE-STUDY

This paper describes a two-step method for the design of a robust parameter estimation scheme that may be used to update the process model employed in on-line optimization. The first step involves determining the "key" model parameters (i.e. those having a significant effect on the calculated optimum). The second step involves finding the "best" set of measurements to estimate these key parameters. This is done by passing all possible measurements through a series of "sieves" (comprising structural analysis, singular value analysis and calculation of parameter confidence regions), with only the more promising measurements passing on to the next stage. The Williams-Otto (simulated) plant was used to assess the proposed methodology. For this purpose, it was found that the parameter estimation scheme was robust to both the presence of measurement noise and uncertainties in the non-key parameters.