Multi-level predictive traffic control for large-scale urban networks

Project members:prof.dr.ir. B. De Schutter (Bart), A. Jamshidnejad (Ana)
Keywords:Distributed and large-scale systems, Hybrid systems, Model predictive control, Multi-level and multi-agent control
Sponsored by:NWO, NSFC

The aim of the project is to use Model-based Predictive Control (MPC) to develop innovative algorithms and methods for fast and efficient traffic management of large-scale urban networks, with a main emphasis on congestion prevention and avoidance, as well as reduction of emissions. We propose the use of MPC, which allows us to anticipate the future events and to integrate and coordinate various traffic control measures that are spatially dispersed over a large area. Our aim is to make MPC tractable in practice for large-scale urban traffic networks.

In this project we will consider a multi-level approach with three control levels. The top layer should control the future traffic flows among subnetworks in order to achieve a balance between subnetworks and to avoid traffic congestion from the macroscopic point-of-view. Here, we will define an optimization problem that will be solved in the receding horizon fashion.
For the middle level, we will turn the MPC optimization problem into a constraint satisfaction problem, in which the conditions for preventing and removing traffic congestion are translated into hard constraints. From the point of view of emission control, the task of the middle layer is additionally to protect sensitive areas.
For the bottom layer, an important issue is to develop a distributed control approach that could deal with nonlinear and hybrid dynamics and guarantee fast convergence to the optimal solution.

At the top and middle level we use MPC, with different types of models due to the different temporal and spatial scale involved. At the bottom level we use data-driven, rule-based, or fast MPC controllers embedded in a distributed control framework in order to obtain coordination. A fast MPC will be developed through the use of parametrized control laws, approximations, and exploitation of the specific structure of the MPC optimization problem inside the optimization solvers, in combination with distributed optimization and control. The project is done in collaboration with the Department of Automation, Shanghai Jiao Tong University (China).

© Copyright Delft Center for Systems and Control, Delft University of Technology, 2017.