Distributed model predictive control of nonlinear systems with asynchronous communication

点击翻译按钮获取中文摘要

Networked and complex systems are of growing importance in the field of automation. Practically relevant examples are smart grids and water supply networks that can be described by modular, distributed systems with a large number of dynamically coupled subsystems. Accordingly, modern control methods for this kind of systems must be able to cope with modularity and scalability. In particular, distributed model predictive control (DMPC) splits the central problem in subproblems for the individual components of the network that are solved in parallel by local agents. In a previous DFG project, a modular and computationally efficient DMPC framework for nonlinear networked systems was developed and implemented in the open-source toolbox GRAMPC-D.State-of-the-art optimization methods for DMPC such as ADMM (Alternating Direction Method of Multipliers) perform the local optimizations and communication steps synchronously over the network. This synchronization has the major drawback that the network has to wait for the slowest agent of the network. First results in the literature and own preliminary work show that an asynchronous handling of the subsystems can massively improve the efficiency and at the same time is able to significantly reduce the communication over the network. Motivated by these results, this project investigates the asynchronous coordination of DMPC for nonlinear systems and the corresponding communication aspects and eventually extends the existing open-source toolbox GRAMPC-D to handle nonlinear networked systems in an asynchronous and modular fashion.