Study On Key Theory And Methods For Data Cleaning Of Traffic Flow

True and accurate traffic flow data is an important guarantee for lean management and the Intelligent Transportation Systems, especially for Advanced Traffic Management Systems and Advanced Traveler Information Systems. Traffic flow condition recognition, control and optimization are the important research direction of traffic flow theory. Some quality problems including missing data, incorrect data, redundant data, etc. were existed inevitably in the dynamic data, which were detected from real road systems. Effective traffic management and control depend on high quality data. Therefore, data cleaning is essential for the followed work such as traffic flow condition recognition and data fusion, and also the key to the success of ITS.Based on review of related literatures and the demand on traffic flow data cleaning, the key theory and methods for single-source data cleaning, such as method for filling the missing data based on rough set theory or LS-SVM, method to distinguish incorrect data based on edge detection and outlier detection, method to modify incorrect data based on GM(1,1) and methods to recognize and reduce redundant data, are studied in this paper. These models or algorithms are validated with the field data from the TRG of university of Southampton of UK and the data collection from Zibo comprehensive transport planning. The results indicate that models or algorithms in the paper have better performance.