应航天学院王振华副教授的邀请，德国卡塞尔大学Olaf Stursberg教授和Zonglin Liu（刘宗林）博士于2021年5月27日（周四）举行线上学术讲座，欢迎感兴趣的师生参加。
ZOOM会议室ID：990 9777 6082；密码：601711
Optimization-based Control of Cyber-Physical Systems
Distributed Systems, in which dynamically modeled subsystems are controlled by digital units and exchange information by communication networks, are often termed cyber-physical systems (CPS). The control design for this system class is challenging, since the task has to account for the interaction between the subsystems, for uncertainty and possibly time-varying structures, and often for large dimensions. This talk provides an overview of different techniques investigated by the Control and System Theory lab to address these challenges. Starting from a general introduction into control challenges for CPS, the talk will first describe how a hierarchy of optimization-based decision routines leads to safe cooperation of autonomous vehicles. The second part outlines an approach to use the principles of model predictive control (MPC) within a distributed control setting, in which predictions are used not only for the local control of subsystems, but also to foresee and consider delay of communication in the network connecting the controlled subsystems. The following part addresses the question of how stochastically modeled uncertainties of the behavior of interacting subsystems can be embedded into predictive control schemes. The final part reports on an approach to approximate the (often computationally demanding) step of synthesizing optimal controllers by neural networks, while guaranteeing the satisfaction of constraints.
Prof. Dr.-Ing. Olaf Stursberg is Full Professor and Head of the Control and System Theory Group at the Department of Electrical Engineering and Computer Science, University of Kassel, Germany. From 2001 to 2002 he was a post-doctoral researcher at the Carnegie Mellon University, and a senior researcher and lecturer at the University of Dortmund from 2002 to 2006. From 2006 to 2009 he was a professor for Automation Systems at the Technical University of Munich. From 2009 until now, he is Full Professor at the University of Kassel. His main research areas include methods for optimal and predictive control of networked and hierarchical systems, techniques for analysis and design of hybrid dynamic systems, applications of AI in control engineering, and the control of stochastic and uncertain systems in different domains of application.
Distributed Solution of Mixed-Integer Programs Arising for Model Predictive Control of Networked Systems
For networked systems, control schemes based on model predictive control (MPC) can provide guarantees for the satisfaction of state and input constraints. They may lead, however, also to high computational complexity, e.g. if the problem formulation involves logic conditions as in hybrid system. While the centralized solution over all subsystems may appear most promising with respect to constraint satisfaction, the high computational effort may prevent the centralized online solution. Nevertheless, recent developments of distributed MPC have shown that the overall complexity can be reduced by decomposing the global problem into a set of local problems with smaller size and solving these in parallel. However, most of these approaches are based on decomposition starting from dual formulations, but the techniques are only applicable to convex problems with only continuous variables. If integer variables are to be considered, the distributed solution using duality-based decomposition may lead to sub-optimal solutions or even infeasibility. This presentation first reviews distributed MPC for systems with only continuous variables, including the dual formulation for distributed solution and approaches to preserve recursive feasibility and stability. Then, a novel distributed solution for MPC with integer variables is introduced, which utilizes recent results on the distributed solution of mixed-integer problems (MIP). This approach decomposes the centralized MIP problem using different necessary conditions of optimality, and the distributed solution process is carried out sequentially by employing different strategies based on these conditions. Significant reductions of the computation time compared to centralized approaches are demonstrated by a larger number of numerical experiments.
Dr.-Ing. Zonglin Liu received B.Sc. from the Harbin Institute of Technology, China, in 2012, and M.Sc from the University of Kassel, Germany, in 2015. From 2015 to 2021, he was a doctoral researcher at the Control and System Theory Group of the University of Kassel, and received the best score Summa Cum Laude for his doctoral thesis with the title ''Optimizing Control of Distributed Cyber-Physical Systems''. From 2021 ongoing, he works as a Post-doctoral researcher in the same department. His research activities include developing methods for optimal and predictive control of cyberphysical systems with uncertainties, efficient distributed solutions for large-scale optimization problems, and use of system- and control-theoretic principles to help in diminishing the Corona/Covid pandemic.