In this work, we propose a chance constrained optimization framework which integrates base layer control and model-based optimization. Here, unlike the definition where controls are decision variables, in the proposed closed framework the set-points of the measurable outputs are defined as decision variables. Moreover, the controller performance based on the minimum variance control is regarded as a random input, and thus is also included in the chance constrained formulation of the model-based stochastic optimization problem. The result is a cyclic adjustment of the operating point which guarantees the compliance with the product quality restrictions and assures the controller performance under parametric uncertainty, uncertain boundary conditions, and the random offset.
To show the efficiency of the integrated framework, the proposed approach has been applied to a pilot plant(a high-pressure column for separating an azeotropic mixture). In order to satisfy the operational constraints the nominal optimal decisions are adjusted cyclical. Fast disturbances are treated in the basic regulatory control layer whereas in the optimization layer optimal set-points are computed. Explicit inclusion of closed-loop deviations and model uncertainty in the problem formulation guarantee feasible and optimal operation. The application of the developed methodology will also be presented by implementation of a soft-sensor in order to compute optimal set-points e.g. for temperature control loops. Furthermore, solving a reduced model for the stripping section, robust set-points are evaluated to meet the bottom product specification. In general, the robust and optimal operation of the high-pressure column and its experimental verification will be presented.