TRALICO

Multi-Input Deep Learning for Congestion Prediction and Traffic Light Control

© EIG Concert Japan
  • Dr. Vilmos Simon - Budapest University of Technology and Ecnomics - Hungary
  • Dr. Ismail Arai - Nara Institute of Science and Technology - Japan
  • Dr. Muhammed Emın Abbaszade - ISBAK Istanbul IT and Smart City Technologies Inc. - Türkiye

Congestion in crowded metropolises poses significant challenges to city residents, leading to wasted time and energy, increased air pollution, and health problems. Effective management of intersections in the road network plays an important role in increasing traffic efficiency and ensuring smooth traffic flow on urban roads. However, traditional traffic management solutions used in metropolises no longer adapt to the dynamic nature of today’s traffic conditions.

The current approaches to traffic management primarily involve intervening in traffic congestions, incidents after they have already occurred, resulting in prolonged inconvenience and inefficiency. By leveraging historical and real-time data in an AI-based prediction model, proactive intervention in traffic control becomes feasible, empowering more efficient and responsive traffic management in urban transportation systems. Thus, we can proactively optimize traffic flow by fine-tuning traffic light settings based on predicted future traffic conditions, effectively preventing congestion and vehicle queues.

Our project proposal is a transformative research initiative aimed at revolutionising urban traffic management to create sustainable and carbon-neutral cities. The proposal introduces a multidisciplinary approach, combining advanced deep learning techniques with innovative traffic control strategies, to predict congestion patterns and optimise traffic light control. Relevant traffic-related data, including traffic flow, historical congestion patterns, weather data, and traffic light operation logs, are collected and preprocessed to ensure accurate training and evaluation of the deep learning models. One of the novelties of our solution lies in developing an advanced multi-input deep learning model that uses various inputs such as traffic data and weather to accurately predict future traffic and congestion states. For this purpose we plan to utilise transformer models, known for their self-attention mechanism, expecting to make significant improvements.

This predictive model acts as a valuable tool for city planners and traffic operators, allowing them to proactively identify traffic bottlenecks for effective traffic management. The adaptive traffic light control strategy allows for dynamic adjustment of traffic light timing based on traffic congestion forecasts, to minimise traffic metrics such as travel time, the average delay per vehicle, average queue length, CO2 emissions. Following the integration of the traffic light control strategy with the city’s infrastructure, it will be tested first in simulated conditions and after under realistic conditions. This advanced traffic management system will allow city authorities to control traffic lights more precisely, influencing traffic flow within their cities. It is a valuable solution sought by many cities worldwide, offering great business potential.

Call related news and events