Case Studies Data Thinking

Developing an intelligent control system for large traffic interchange using reinforcement learning.

Initial situation and problem

  • Prioritization of mobile units passing through a traffic intersection has been done manually until now. As a result, the average speed is not optimized, which usually leads to delays.

  • Due to the high complexity and dynamic nature, an analytical solution/optimization is not possible.

  • Therefore, the goal of the project is to minimize delays throughout the system by supporting dispatchers with artificial intelligence. This should be trained based on simulations and historical data to achieve an optimal prioritization of mobile units.

     

 

mm1 approach and solution

  • Engineering data pipelines from various IT systems to enable simulations for training an ML model.
  • Developing and training an intelligent agent using reinforcement learning. The agent learns to make the optimal mobile unit prioritization decisions from the available data to minimize delays in the system.
  • Roll out a Minimum Viable Product (MVP) in the IT systems to help dispatchers determine the order of mobile units in real time.