Integration of gradient adaptation and quadratic approximation in real-time optimization

This paper is concerned with combining the use of models and of measurement to drive a plant to its optimal operation. The iterative gradient-modification optimization scheme [1] is integrated with the quadratic approximation approach [2] to enforce the global convergence and to improve the rate of convergence. The key elements of the integration, i.e. model-quality monitoring, optimization switching and empirical gradients estimation, are analyzed theoretically. Simulation studies for the optimization of chromatographic separation with considerable plant-model mismatch and process-dependent constraints are used to illustrate the potential of the proposed scheme.