Transfer learning for robust urban network-wide traffic volume estimation with uncertain detector deployment scheme

Real-time and accurate network-wide traffic volume estimation/detection is an essential part of urban transport system planning and management. As it is impractical to install detectors on every road segment of the city network, methods on the network-wide flow estimation based on limited detector data are of considerable significance. However, when the plan of detector deployment is uncertain, existing methods are unsuitable to be directly used. In this study, a transfer component analysis (TCA)-based network-wide volume estimation model, considering the different traffic volume distributions of road segments and transforming traffic features into common data space, is proposed. Moreover, this study applied taxi GPS (global positioning system) data and cellular signaling data with the same spatio-temporal coverage to improve feature extraction. In numerical experiments, the robustness and stability of the proposed network-wide estimation method outperformed other baselines in the two subnetworks selected from the urban centers and suburbs.

Automated summary

In this study, a transfer component analysis (TCA)-based network-wide volume estimation model is proposed to be used when the plan of detector deployment is uncertain. The method improves feature extraction by using taxi GPS data and cellular signaling data with the same spatio-temporal coverage. Numerical experiments show that the proposed method outperforms other baselines in terms of robustness and stability in two selected subnetworks from urban centers and suburbs.