Estimation of Missing Flow at Junctions Using Control Plan and Floating Car Data

This paper proposes a new and consistent approach for estimating missing flow by analyzing data from SCATS (containing both flow and timing plan at junctions) and FCD (Floating Car Data). SCATS system provides flow data and timing plan at a 5-minute interval, while FCD contains information of taxi trajectories with speed and position for each vehicle at a 30-second interval.Two objectives are defined in this paper: 1) to summarize major methods of flow estimation and create a generalized framework in flow estimation, 2) to research for the possibility of improving utilization of traffic flow data, by comparing methods from multiple aspects, to provide accurate and reliable source for traffic research and application.The paper devises three consistent methods to estimate missing flow at junctions. Firstly, historical flow data of a specific lane is used to make an initial estimation. Flow estimation values are estimated from each single lane and normalized to further complement Secondly, flow values from adjacent lanes with similarities are processed to compare with the estimated lane, for which timing plan is applied to identify relevant control group (same turning lanes) with their relative phase time proportions. Proportions and flow rate from observed lanes on the same control group are normalized to make an estimation at missing lanes. Thirdly, the information of FCD (such as speed) is used to estimate corresponding flow value. Typically detected flow and FCD speed relationship is established from junction streams. This relation is then applied back to the stream to calculate traffic flow. Each of the methods is expected to perform under varying situations. If all the results are proved to be reliable to a certain degree, they can be iterated for mutual verification and consistency.This methodology has been applied to Changsha municipality in China. Initial results indicate that suggested methods give promising indication that almost all missing lane flow could be recovered using these three methods. Further research is ongoing to investigate specific data fusion mechanism or interchangeable data source for traffic state estimation and its quality. Further research will also consider adjacent junctions within a given area to understand how data and flow relationship works at the network level.