Convergence Analysis of Bias-Update Process Optimization Applied to a Chemical Reactor

A. G. MARCHETTI

Buenos Aires

Congreso; 24º Congreso Argentino de Control Automático, AADECA'14; 2014

AADECA

In the framework of process optimization, bias-update schemes correct the model predictions using additive plant-model bias terms instead of estimating the model parameters. In constraint adaptation algorithms these bias terms are used to correct the predicted values of the process-dependent constraints in the optimization problem. Based on previous global convergence results for run-to-run control algorithms, this paper analyzes the global convergence of the constraint-adaptation algorithm for the case of a single active process-dependent constraint. The convergence condition is illustrated by means of a continuous stirred tank reactor for which constraint adaptation does not converge without an appropriate filtering of the constraint bias.