Towards essentially decentralized interior point methods for distributed non-convex optimization

Title:
Towards essentially decentralized interior point methods for distributed non-convex optimization

Speaker:
Alexander Engelmann

Abstract:
Distributed and decentralized optimization methods are key in distributed model predictive control, in distributed sensing, and estimation. Non-linear models, however, lead to problems with non-convex constraints for which established distributed and decentralized algorithms often lack convergence guarantees. Moreover, decentralized algorithms frequently exhibit rather slow linear convergence rates. In this talk we propose an essentially decentralized primal-dual interior point method with local convergence guarantees for non-convex problems at a superlinear rate. We draw upon different examples from power systems and control illustrating its performance. The numerical results indicate that the proposed method is able to outperform ADMM in terms of computation time and it has the potential to overcome difficulties associated with active-set detection in the context of distributed optimization.

Bio:
Alexander Engelmann (GSM'18) received the B.Sc. and M.Sc. degrees in electrical engineering and computer science (with distinction) from the Karlsruhe Institute of Technology, Karlsruhe, Germany, in 2014 and 2016, respectively, where since 2017, he has been working toward the Ph.D. degree with the Optimization and Control group, Institute for Automation and Applied Informatics, where he is focusing on distributed optimization and optimal control for power and multi-energy systems.