This document presents the summarised results of the '”Analysis of Transport for London (TfL) GPS OBU Data” mini-project commissioned by the Dutch Ministry of Transport (The Ministry) and prepared by Mapflow on original data produced during the 'Road User Charging (RUC) Mini-Trials' undertaken by TfL on selected set of vendors between November 2005 and April 2006. These trials are referred to in this report as the Stage 3 TfL trials. The results presented here have been anonymised to protect the commercial interests of the vendors taking part in the original project. The original report generated by TfL has been published and is available from the TfL website. The purpose of the Stage 3 TfL trials was to exercise and learn from the GPS-based road pricing systems available from the vendor community. The purpose of this project for the Ministry is to extend that exercise by determining the efficacy of 'simple non-map matching' methods for the calculation of distance travelled. By using a pre-defined network of road segments with different charge weightings and a number of different tariff schemes, the aim of the original TfL project was to understand the feasibility and resolution of an RUC scheme in London. The goal of this project is to understand whether road usage can be determined sufficiently without the need for detailed mapping, and all of the complexity that implies. The objectives of this study for the Ministry, is to evaluate using the TfL trial data whether non map matching algorithms can achieve performance in line with the results obtained in the “cost monitor” project for which the first phase has been completed and reported on. Three algorithms, ranging from a simple straight line distance calculation, to a simple filter and a more sophisticated algorithm were used in this study to generate distance values without the use of map matching. 4.2 million GPS points were collected from 17 device types; in some cases a vendor supplied more than one device to the TfL Stage 3 trials. Devices used in the trials ranged from road pricing devices through to telematics units. 1,265 journeys were driven in the central congestion zone in London totalling 36,617 km. The journeys were typically 24km which unfortunately does not provide a range of journey lengths for the analysis. The three algorithms used in this study demonstrated varying degrees of performance, in terms of road usage distance determined versus actual distance driven: - The simple “straight line” algorithm provided the best performance for journeys with good GPS data. For journeys with erroneous data the algorithm exhibited very poor performance. - The simple “Filter” algorithm removed the erroneous data from the calculations improving performance generally. However, this filter also adversely affected performance in cases where GPS was good. - Reviewing the results of the “Sophisticated” algorithm anecdotally suggests promise. The algorithm is able to interpolate and extrapolate out GPS error and unavailability. However, in an attempt to achieve 0% error, the algorithm was tuned to the point where it became unreliable. Taking three good vendor devices, the results from the first to methods show that it is possible to achieve 2% average distance deviation and 4.64% standard deviation, for bills of 24km journeys. The analysis shows that these devices would achieve 35% of journeys within 1% error. A margin of 10% would be required for 99% confidence. (Author/publisher)
Samenvatting