A Novel Framework to Estimate Missing Flow at Urban Junctions

Data acquisition is the base for traffic management and control and the availability of data also influences the operation and analysis of the intelligent transportation system. In some traffic systems, traffic flow data is missing from the detectors. To complete the missing data, the paper presents a novel framework to estimate missing flow at urban junctions by considering approaches from four aspects: (1) historical lane flow pattern (2) distributed lanes in the same signal plan, (3) synced lane flow with timing plans, (4) sources from other sensors such as Floating car data. First three approaches consider direct observations in the same system from time and space dimension. The last approach considers the traffic flow theory between flows and traffic states from another sensor. A combined method is carried out by training data iteratively from first two approaches by applying their updated flow relations. The result of the cases turns out to be positive, with half of the individual short-term missing flows cases of mean absolute percentage error lower than 10%, and stable performance of flow estimation over a longer period of missing time. The fourth approach is tested individually, and it shows the possibility of making estimation from a data fusion aspect; though the current performance does not show enough accuracy, Floating car data from the fourth approach is also expected to determine the weighting factors of iteration between previous two approaches. A combination of all the approaches is expected from the future research.