Development of vehicle trajectory forecasting framework allowing capacity fluctuation for multi-lane condition

Abstract

This study presents a framework for short-term trajectory forecast based on variational theory of kinematic waves. The framework is tested for non-lane based heterogeneous traffic condition where different characteristics of the roadway (e.g. non-motorized vehicles, bus parking, vendors, and pedestrian movement) generate side friction, which in turn causes capacity fluctuation. Considering these unique aspects of non-lane based system the framework is developed in a way that considers multiple lanes, and capacity fluctuation. The study is carried out by following a step-wise approach starting from the explanation of the variational theory, then trajectory reconstruction and evaluation, and finally, the short-term trajectory forecast framework. Trajectory reconstruction is carried out considering two boundary options consisting of cumulative count and probe data. The dataset including speed, flow, and density is used to develop a fundamental diagram for the study section and to determine parameters (e.g. jam density, forward wave speed, and backward wave speed) that are necessary to develop the time-space solution domain (e.g. lopsided network) using variational theory. The probe data is then placed into the mesh to set up the probe trajectory in the newly formed lopsided network. Finally, the trajectories are reconstructed using the probes as a reference and the effect of capacity fluctuation or varying capacity are investigated. The accuracy of the estimated trajectories is evaluated through error analysis using different reference probes. The analysis shows that the probes are able to estimate adjacent trajectories significantly well compared to distant vehicles. The application of the trajectory forecast framework is tested on a corridor of 1 km in length and consisting of two intersections. The boundary conditions are set along these two intersections. The proposed framework provides an opportunity to combine historical and real-time dataset in an effort to predict trajectories within the small forecast window. The temporal distribution of the forecast window (1500 second for this study) depends on the distance between the two boundaries and the available real-time dataset. Note that the prediction dataset is prepared considering the signal cycle, turning movement, and passing vehicles; whereas the trajectory estimation process is carried out following the variational formulation of kinematic waves. The application of the framework involves a modular concept that is built into a Graphical User Interface for automation. The modular concept consists of four modules as follows: (1) Data Module; (2) Master Control; (3) Trajectory Reconstruction Module; and (4) Trajectory Forecast Module. Master control facilitates access to all modules and maintains the interconnectivity between modules during any particular operation, e.g. trajectory estimation or forecast. The accuracy of the predicted trajectory is evaluated through travel time analysis using different probes as a reference within the forecast window.