Model predictive traffic control: Efficiency versus accuracy

Staff Mentor: B. De Schutter (Bart)

Other Mentor(s):

Mohammad Hajiahmadi


Optimal and model predictive control; Automotive and intelligent transportation systems; Distributed and large-scale systems



With the increasing number of vehicles, the freeways are becoming more and more congested. This
along with increasingly stringent traffic requirements necessitates use of efficient large-scale traffic
management and control algorithms. One particular solution to this problem is based on Model
Predictive Control (MPC), where a finite-horizon constrained optimal control problem is solved
in a receding horizon fashion. However, in the MPC framework a model of the process is required
to predict the behavior of it in a prediction horizon. For traffic networks, a wide range of traffic
flow models have been developed. Among these, a model that can provide accurate predictions
of the traffic states while it has low computational complexity is needed. The METANET model
is a second-order model that is able to model the traffic network with good accuracy. However,
for the simple case study shown in Figure, it has been shown that finding optimal control inputs
in the MPC framework based on the METANET, takes considerable time. One way to overcome
this problem especially for large-scale traffic networks is to use first-order models like the Cell
Transmission Model and the Link Transmission Model. These models are mostly used for dynamic
traffic assignment problems. But in this project we consider using them for control purposes.


In this project we consider modeling a case study with these different macroscopic traffic flow
models: METANET, CTM, and LTM. In the first step, these models should be carefully studied
and next be implemented in MATLAB for a traffic network case study. The evolution of traffic
states (flow, density, velocity) should be determined and compared for the 3 models. Computation
time should be calculated and compared. After this step, a model predictive controller for
minimizing the total time that vehicles spend in the traffic network will be implemented. As a
prediction model in the MPC framework, any of the afforementioned models can be used. The aim
is to design an MPC controller for the case study modeled by each of the 3 models (METANET,
CTM, and LTM) and next, to compare the performance of the MPC controllers in terms of computational
efficiency and total cost. As an extension to this work, a real traffic network can be
modeled by the LTM and the model can be calibrated using real data.

Traffic network case study

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