Abstract
Modifier Adaptation (MA) is a method of real-time optimization (RTO) which modifies a single model to match the first order properties of the plant. Known uncertainties in the parameters of this model are discarded in favor of real-time measurements, but they can be used to quantify the mismatch between the plant and model. Using multi-model methods increases the computation time, but can improve rate of convergence of the RTO scheme. This article proposes a framework, known as multiple solution modifier adaptation (MSMA), which produces several models which are all modified in the same way as standard MA, each producing a potential solution to be applied to the plant. From this framework, three recommended schemes are proposed on how to select the operating point to be applied to the plant: (1) Selecting the solution based off the modifiers; (2) Selecting the mean solution from convex models; (3) Selecting the closest solution to the current operating point. Each of these methods have different advantages, including limiting the increase in computational complexity and improving the model adequacy conditions of the scheme. These recommended schemes are shown on three different case studies of varying complexity with all three schemes showing improvements over standard MA.