Who likes being stuck in traffic?
Data-driven control helps to reduce traffic congestion - by Philippe Brigger
In urban areas, traffic congestion poses a significant challenge, often without the possibility of expanding road infrastructure.
To address this, smart ways of managing our existing road infrastructure are necessary. Control can help here, for example by adapting the green-red cycle of our traffic lights or by adjusting speed limits based on the current traffic conditions.
We use Data-EnablEd Predictive Control (DeePC), a data-driven control approach developed by NCCR Automation researchers, to dynamically control traffic lights. Rooted in behavioural systems theory, DeePC utilises the system’s trajectories to describe its behaviour. Hence, DeePC does not require a parametric model of the system as Model Predictive Control (MPC) would. Similar to MPC, DeePC determines optimal control inputs by minimising a cost function and adhering to system constraints.
Due to the nonlinear and complex nature of traffic networks, large amounts of data would be required to describe the full behaviour of the system. However, solving the DeePC optimization problem becomes computationally intractable if too much data is used. Yet, it is not trivial to know which data and how much data is relevant depending on the current traffic conditions.
To bridge this gap, we are researching how to implement a data selection method for DeePC in urban traffic control. Ultimately, this effort aims to enable the deployment in a time-efficient and reliable manner, contributing to the alleviation of traffic congestion issues.
Simulation of congestion in the city of Zurich with heavy traffic (left) and moderate traffic (right).
Text by Philippe Brigger, simulation by Alessio Rimoldi, picture by Kaique Rocha from www.pexels.com
Data-driven control helps to reduce traffic congestion - by Philippe Brigger