Global Traffic State Recovery Via Local Observations With Generative Adversarial Networks

Traffic signal control for a large-scale traffic network is one challenging problem in intelligent transportation systems (ITS). High communication overheads are typically required to achieve the optimal control of the traffic signals in multiple road intersections. In this paper, in order to avoid these communication overheads among spatially distributed intersections, we propose to recover the global traffic state at each intersection in a real-time fashion by only utilizing the traffic state observed at the local intersection. Specifically, a generative adversarial network (GAN)-based traffic data recovery method is presented for each intersection controller to recover the global traffic state. We also exploit a few statistics from other intersections during the training of the proposed GAN to improve the traffic state recovery accuracy. Comprehensive numerical results demonstrate the effectiveness of the proposed scheme in recovering the global traffic state.