Edge-AI Enabled Uncertainty-Aware Intelligent Traffic Management System Using Graph-Based Forecasting for Smart Cities

Abstract

The high rate of urbanization has brought about a lot of traffic congestion to the smart cities, requiring smart and scalable traffic control solutions. Conventional signal control systems are not responsive to changing traffic situations, whereas current prediction-based models do not take into account the uncertainty factor, and hence unreliable decisions are made. To discuss the issue of the real-time optimization of traffic, this paper offers a solution to it through the creation of an Edge-AI Enabled Uncertainty-Aware Traffic Management System (E-UTMS). The framework combines the edge-based traffic perception, graph based spatio-temporal prediction and uncertainty-conscious adaptive signal control into one architecture. At the edge the traffic descriptors extracted include the number of vehicles, the length of a queue and occupancy of a lane which are then modeled on a dynamic traffic graph to provide congestion patterns in the short term. A risk conscious control method has built in predictive uncertainty to enable sound decision making according to different traffic conditions. Our experimental analysis shows that the proposed method reduces the average waiting time by 19.2%, queue length by 21.9%, and increases the throughput by 18.9% compared to prediction-based control without uncertainty. Further, the system is real time with an average inference time of 28.5 ms. These findings demonstrate the efficiency of the combination of edge intelligence and uncertainty-aware control as a solution to the next-generation smart city traffic management systems, which is scalable and reliable.