Estimation of scaling factors for traffic counts based on stationary and mobile sources of data

Abstract

OBJECTIVES: Mobile sources and stationary sources are two major types of data sources, from which traffic information can be acquired and analyzed. However, either approach is subject to certain limitations. Mobile sources, such as global positioning system (GPS), may only be available from a subset of probe vehicles, whereas stationary sources, such as fixed detectors, are only confined to some given detector locations. Thus, combination of the two sources is worth doing, in which linear data projection is considered to be a straightforward method. Whereas the bias of linear projection is found to increase with the variability of the scaling factors. Therefore, a modeling approach that is able to quantify the variability of linear projection function using a non-linear regression method will be established in this study, followed by a case study of volume prediction in Hong Kong, 2011. DATA AND METHODOLOGY: A modeling framework that is able to track the variability of scaling factors is proposed using a non-linear regression method. In order to estimate a scaling factor of a certain location, different weights that vary spatial-temporally are assigned to other surrounding scaling factors, followed by a normalized weighted average function to calculate the objective scaling factor. The framework was applied to a case study in Hong Kong. GPS data was obtained from 460 probe taxis, which recorded information including time, coordinates and speed every 30 seconds. Trip purpose database was acquired from the Traffic Characteristic Survey (TCS) conducted in 2011, which was applied to represent zonal land use patterns by cluster analysis. Annual Traffic Census (ATC) database provided traffic information that was collected from 85 fixed stations. Based on the proposed framework, the scaling factors were derived to project zonal taxi volume to all the 406 zones in Hong Kong. EXPECTED RESULTS: According to the GPS and ATC databases, the scaling factors for each of the 85 ATC stations were obtained as the ratio of total daily taxi flow to the daily probe taxis. Based on these factors, several projection functions are calibrated by minimizing the difference between observed and predicated scaling factors. Metrics, such as rRMSE, MAPE and AIC, will be used to evaluate the performance of the estimation. The final projection function will make it possible to infer the total zonal taxi volume throughout Hong Kong, which is able to be applied to future accident analyses and network modeling, etc.