Tracking congestion throughout the network road is a critical component ofIntelligent transportation network management systems. Understanding how thetraffic flows and short-term prediction of congestion occurrence due torush-hour or incidents can be beneficial to such systems to effectively manageand direct the traffic to the most appropriate detours. Many of the currenttraffic flow prediction systems are designed by utilizing a central processingcomponent where the prediction is carried out through aggregation of theinformation gathered from all measuring stations. However, centralized systemsare not scalable and fail provide real-time feedback to the system whereas in adecentralized scheme, each node is responsible to predict its own short-termcongestion based on the local current measurements in neighboring nodes. We propose a decentralized deep learning-based method where each nodeaccurately predicts its own congestion state in real-time based on thecongestion state of the neighboring stations. Moreover, historical data fromthe deployment site is not required, which makes the proposed method moresuitable for newly installed stations. In order to achieve higher performance,we introduce a regularized Euclidean loss function that favors high congestionsamples over low congestion samples to avoid the impact of the unbalancedtraining dataset. A novel dataset for this purpose is designed based on thetraffic data obtained from traffic control stations in northern California.Extensive experiments conducted on the designed benchmark reflect a successfulcongestion prediction.
展开▼
