Expert tuning of multivariable model-based controllers

摘要

This thesis focuses on the use of an expert system to tune a GPC algorithm for industrial process applications.;An evaluation of an expert systems based tuning strategy for Long Range Predictive Control (LRPC) algorithms is performed.;The Adaptive Long range predictive control Performance Supervisor or ALPS (Walther, 1992) is used to tune the parameters of the Generalized Predictive Control (GPC) algorithm when applied to a heavy oil fractionater. Gensym's G2 was used for developing ALPS. The tuning strategy is found to work well in improving the closed loop performance of the heavy oil fractionater.;Experience gained from simulations of the heavy oil fractionater has been translated into rules applicable to most multi-input multi-output (MIMO) systems.;A non-exhaustive approach is taken to write these rules, by using a few procedures which simplify the logic.;Simulations of the heavy oil fractionater with model plant mismatch show that increasing the input weighting in proportion to the sensitivity of the outputs to inputs can eliminate the 'noisy' interaction in the outputs while maintaining good tracking of the set points. (Abstract shortened by UMI.).