Parametrized Model Predictive Control in Urban Traffic Networks

Parametrized Model Predictive Control in Urban Traffic Networks: Towards real-time implementation

Jeschke, Joost (TU Delft Mechanical, Maritime and Materials Engineering)

De Schutter, B. (mentor)

Delft University of Technology

Mechanical Engineering | Systems and Control

2020-10-19

Model Predictive Control (MPC) has shown promising results in the control of urban traffic networks, but has one major drawback. The, often nonlinear, optimization that has to be performed at every control time step is computationally too complex to use MPC controllers for real-time implementations (i.e. when the online optimization is performed within the control time interval of the controlled system). This thesis proposes a parametrized MPC control approach to lower the computational complexity of the MPC controllers and to strive for real-time implementability. In parametrized MPC, the original decision variables (i.e. the inputs of the system) become a function of a parametrized control law and the parameters of this control law become the new decision variables of the optimization problem. The goal is to lower the computational complexity by reducing the number of decision variables with limited performance decrease. In this thesis, three parametrized control laws are proposed that can be used in the parametrized MPC approach for urban traffic networks. These three control laws are constructed based on the prediction model of the (parametrized) MPC controller, on ART-UTC, an existing control method, and by using supervised learning. The system performance and computational complexity of the different parametrized MPC controllers are compared to that of a conventional MPC controller by performing an extensive simulation-based case study in which different optimization algorithms, emissions, and parameter update time steps are considered. The simulation results show that the control law based on the prediction model results in a parametrized MPC controller which is real-time implementable, uses 2 parameters that are constant over the control horizon per intersection, and has a system performance decrease of less than 3%.

Model Predictive Control
Urban Traffic Control
Parametrized Controller
Parametrized Model Predictive Control

http://resolver.tudelft.nl/uuid:c06da4ef-733f-4d78-9d61-7232363af974

Student theses

master thesis

© 2020 Joost Jeschke