OPTIMAL IMPLEMENTATION OF ONLINE OPTIMIZATION

Results from a theoretical and numerical evaluation of on-line optimiz ation algorithms were used to recommend the best way to conduct on-lin e optimization. This optimal procedure conducts combined gross error d etection and data reconciliation to detect and rectify gross errors in plant data sampled from the distributed control system. The Tjoa-Bieg ler method (the contaminated Gaussian distribution) was used for gross errors in the range of 3 sigma - 30 sigma or the robust method (Loren tzian distribution) for larger gross errors. This step generates a set of measurements containing only random errors which is used for simul taneous data reconciliation and parameter estimation using the least s quares method Updated parameters are used in the plant model for econo mic optimization that generates optimal set points for the distributed control system. Applying this procedure to a Monsanto sulfuric acid c ontact plant, a 3% increase in profit and a 10% reduction in SO2 emiss ions were projected over current operating condition which is consiste nt with other reported applications. (C) 1998 Elsevier Science Ltd. Al l rights reserved.