Real-time vehicle location estimation in signalized networks using partial connected vehicle trajectory data

Real-time vehicle location estimation is essential for diverse transportation applications, such as travel time estimation, arrival pattern estimation, and adaptive signal control. Existing connected vehicle-based studies rely on either black-box neural networks requiring large training datasets or computationally intensive time-continuous movement simulations grounded in car-following models. However, they often overlook the distinct vehicle location patterns in source lanes, which define network boundaries and experience random arrivals, and intermediate lanes, situated between intersections and receiving traffic discharged from upstream. These patterns are critical for accurate vehicle location estimation. To address these limitations, this study proposes a generic and fully analytical CV-based vehicle location (CVVL) model for estimating vehicle locations within a signalized lane in a network using readily available partial CV trajectory data. The proposed model is applicable to any signal timing, traffic demand, and CV penetration rate and consists of two sub-models: CVVL-S and CVVL-I. The CVVL-S sub-model estimates vehicle locations in source lanes, where vehicle distribution tends to be relatively homogeneous owing to random arrivals. In contrast, the CVVL-I sub-model focuses on estimating vehicle locations in intermediate lanes, where sequential discharges from different upstream lanes can lead to the formation of multiple platoons, adding complexity to vehicle location estimation. The proposed model decomposes the complex task into three sequential sub-problems: identifying candidate platoons (CPs), estimating the number of vehicles in each CP, and determining the spatial distribution of vehicles within each CP. Extensive numerical experiments were conducted under various traffic conditions, CV penetration rates, and times of interest using the VISSIM platform and the real-world Next Generation Simulation dataset. The results demonstrate that the proposed CVVL model achieved improvements of 0–45 %, 0–37 %, and 4–34 % in precision, recall, and F1 score, respectively, compared with the competing method. These results highlight the model’s potential to enhance the accuracy and reliability of various downstream applications.