A Data-Driven Multi-Fidelity Approach for Traffic State Estimation Using Data From Multiple Sources

Accurate estimation of traffic state is crucial in the successful implementation of advanced traveler information systems and advanced traffic management systems. Currently, traffic data from various sources such as GPS, cell phones, and connected vehicles are increasingly available. Consequently, efforts have moved from estimating traffic using a single source of data to incorporating heterogeneous data, i.e. data from multiple sources, for traffic estimation. In these works, data from different sources are mostly considered to have the same level of fidelity or accuracy. This is while, in reality, different data sources may contain different levels of noise. In this work, we propose using multi-fidelity Gaussian processes to exploit data with different fidelities for traffic estimation. We demonstrate the performance of the proposed approach using a simulation study, where the results show that accounting for multi-fidelity nature of data can significantly improve the accuracy when multiple data sources are used for traffic state estimation.