Estimation of Traffic Flow Rate With Data From Connected-Automated Vehicles Using Bayesian Inference and Deep Learning

Connected automated vehicles (CAVs) hold promise to replace current traffic detection systems in the near future. However, traffic state estimation, particularly flow rate, poses a major challenge at low CAV penetration rates without other supporting infrastructure of sensors. This paper proposes flow rate estimation methods using headway data from CAVs. Specifically, Bayesian inference and deep learning based methods are developed and compared with a naïve method based on a simple arithmetic mean of observed headways. The proposed methods are investigated via numerical experiments to evaluate their performance with respect to the CAV penetration rate, traffic demand, and availability of historical data. The methods are further validated with real data. The results show that the Bayesian inference based method, which estimates the flow rate distribution by integrating current (real-time) data and previous knowledge, can perform well even at low penetration rates with good prior information. However, in high CAV penetration, its relative advantage to the other methods diminishes because the prior information always influences the flow rate estimation. The deep learning based method can be effective with a large amount of data to train the model; however, in low CAV penetration, it tends to converge to the mean of target output values regardless of the observed data. At last, in relatively high CAV penetration, the relative advantage of the advanced methods is negligible and in fact, the naïve method is preferred in terms of accuracy as well as efficiency.