An Evolutionary Real-time Optimization Approach

RTO is located between the planning and control layer in a structural control system, which provides a set of set-points for local controller, keeping processes operating in optimum conditions. It is acknowledged that traditional real-time optimization approaches may suffer certain deficiencies such as tremendous waiting time for steady states and the complexity concerning model updates and optimization computations. In the face of these challenges, this paper proposes an evolutionary real-time optimization approach inspired by multi-objective evolutions. A multi-objective particle swarm optimization (PSO) algorithm is adapted to encode the model parameters and manipulated variables uniformly and thereby implement the evolutionary optimization.Initially, the basic philosophy, framework and the major weakness of traditional RTO approaches are discussed, followed by an evolutionary real-time optimization approach proposed. The period of waiting for steady-state against disturbances is divided into several pseudo-steady states, in which multi-objective evolutionary algorithms are launched to update process models and calculate optimum operating points based on real-time measurements consecutively.An in-depth investigation on multi-objective evolutionary optimization is performed, along with particle swarm based approaches to solving multi-objective optimization problems. According to requirements of evolutionary real time optimization, detailed PSO algorithms as well as the procedures towards encoding model parameters and manipulated variables simultaneously are provide in this thesis.Finally, experimental studies consisting in a classical Williams-Otto reactor are carried out, showing that the evolutionary real-time optimization method not only extends the scope of applications of real-time optimization against continuous disturbances, but also possess significant advantages in terms of optimization calculations and real-time characteristics, and demonstrating promising potential.