Quantitative Data Cleaning and Analysis for Large Databases of Passenger Car Fuel Consumptions

This paper presents the work a development of data cleaning and analysis for on-board diagnostic system (OBD II) data. The OBD device can collect speed and mass air flow via ECU, from which the instantaneous fuel consumptions are also calculated, along with the vehicle trajectories collected by an integrated GNSS module. Therefore, a large vehicle trips survey collects trajectories and instantaneous fuel consumption of over 8000 vehicles. The large trajectory database consists a number of trips whose positioning failures are significantly higher than the average. The datasets are proceeded with quantitative analysis to identify the faulty trip data by fitting with the power law distribution. After the cleanup, the analysis results demonstrate that the datasets can deliver a reliable estimation of the average daily passenger car fuel consumption across Beijing.