The aim of process optimization is to improve performance, while enforcing the satisfaction of operating, environmental, safety and quality requirements. In practice, most optimization techniques use a mathematical model of the process that is embedded in a numerical optimization solver. To determine the optimal inputs (for steady-state optimization problems) or the optimal input profiles (for dynamic optimization problems), these solvers typically maximize or minimize a cost function numerically. As the model is rarely a perfect representation of the plant, the optimal inputs obtained this way will typically not be optimal for the plant.
The research in the area of real-time optimization (RTO) has dealt with the development of adaptive optimization methods, whereby measurements are used to compensate for the effect of uncertainty on process performance. Current practice consists in successively using measurements to correct the model and the updated model to compute the optimal inputs. This approach, although appealing, can only perform well if the structure of the model is correct and thus will fail in the presence of structural plant-model-mismatch. Our research has focused on the development of methods that implement optimality via enforcing the necessary conditions of optimality (NCO) of the plant. As a result, the corresponding RTO methods are capable of dealing with the three main sources of uncertainty, namely parametric errors, structural uncertainty and process disturbances. To date, two novel RTO techniques, which deal directly with the plant NCO, have been proposed: