Simulation assisted performance optimization of large-scale multiparameter technical systems

During the past two decades the role of dynamic process simulation within the research and development work of process and control solutions has grown tremendously. As the simulation assisted working practices have become more and more popular, also the accuracy requirements concerning the simulation results have tightened. The accuracy improvement of complex, plant-wide models via parameter tuning necessitates implementing practical, scalable methods and tools operating on the correct level of abstraction. In modern integrated process plants, it is not only the performance of individual controllers but also their interactions that determine the overall performance of the large-scale control systems. However, in practice it has become customary to split large-scale problems into smaller pieces and to use traditional analytical control engineering approaches, which inevitably end in suboptimal solutions. The performance optimization problems related to large control systems and to plant-wide process models are essentially connected in the context of new simulation assisted process and control design practices. The accuracy of the model that is obtained with data-based parameter tuning determines the quality of the simulation assisted controller tuning results. In this doctoral thesis both problems are formulated in the same framework depicted in the title of the thesis. To solve the optimization problem, a novel method called Iterative Regression Tuning (IRT) applying numerical optimization and multivariate regression is presented. IRT method has been designed especially for large-scale systems and it allows the incorporation of domain area expertise into the optimization goals. The thesis introduces different variations on the IRT method, technical details related to their application and various use cases of the algorithm. The simulation assisted use case is presented through a number of application examples of control performance and model accuracy optimization