With the rapid integration of sensing and communication technology, connected vehicles (CV) has been a refined data source for traffic system analysis. Penetration rate, defined by the proportion of CVs in the mixed flow, has been a fundamental obstacle to numerous applications such as OD estimation, traffic flow estimation and adaptive signal control. However, due to its stochastic nature, the accurate value of penetration rate is hard to be estimated by current methods , especially when fixed detector data is in poor quality or penetration rate is extremely small. This paper proposes a Bayesian approach for penetration rate estimation based on single-source data from CV trajectories. The proposed method is not constrained by the arrival pattern, and takes several other uncertainties of mixed traffic flow into consideration. It could also be easily extended to oversaturated condition by some simple adjustments. Markov Chain Monte Carlo (MCMC) technique is used to solve the model. Experiment is conducted using simulation data to evaluate the performance. Sensitivity analysis under varying penetration rate shows that the proposed method has a satisfying accuracy even in cases of relatively low penetration rates.

CICTP