Particle Filter Approach for Real-Yime Freeway Traffic State Prediction

The research presented in this paper develops a multi-step particle filter traffic state prediction algorithm using spot speed measurements. The traditional Lighthill-Whitham-Richards (LWR) flow continuity equation is combined with the Van Aerde traffic stream model to generate a new speed-based partial differential equation (PDE). The numerical solution of the PDE is obtained using the Godunov discretization scheme within a particle filter to generate a time series equation that characterizes the temporal and spatial relationship of traffic speed data. This speed formulation is further enhanced by incorporating ramp flows and enhancing the boundary conditions. The numerical solution and near-term prediction accuracy (5-minute prediction) of the new speed formulation is compared with the conservative density formulation derived from LWR. Although the proposed speed formulation is non-conservative and not equivalent to the solution of LWR under the same initial and boundary conditions, it produces significant enhancements in the traffic state predictions. Specifically, the prediction error using simulated I-66 data is in the range of 3.0 to 4.5 km/h for a 5-minute prediction horizon. This error is approximately half the prediction error of the LWR formulation. Similarly, the traffic stream density prediction error is approximately half that of the LWR formulation.