Hierarchical federated learning model for traffic light management in future smart

The present era is marked by rapid improvement and advances in technology. Nowadays inefficient traffic light management systems can make long delays and waste energy improving the efficiency of such complex systems to save energy and reduce air pollution in future smart cities. In this paper, we propose to take real-time traffic information from the surrounding environment. Such a process, which is called profilization constantly gathers and analyses information for vehicles and pedestrians throughout smart cities in order to fairly predict their actions and behaviours. We develop an efficient multi-level traffic light control system to schedule traffic signals’ duration based on a distributed profile database, which is generated by embedding sensors in streets, Vehicles and everywhere. We deploy pervasive deep learning models from the cloud to users (vehicles, bikes and pedestrians) to learn and control the traffic lights. In the cloud-level learning model, the maximum waiting time of different vehicles and pedestrians is calculated based on their profiles. The profilization process is a constant learning process throughout the whole city at the user level. Each vehicle deploys a separate learning model (decision-making) based on its average and maximum speed in a different area, waiting times at the intersections and possible trips and destinations. Such a multi-level deep learning model in the level of intersection and cloud aims to locally schedule the traffic with deadlines toward their destinations within a certain period. The results show that the proposed multi-level traffic light system can significantly improve the efficiency of the traffic system in future smart cities.