Trip-activity Chain Pattern Recognition Based On LBS Trajectory

It is the revolution of Internet, the application of mobile communication, and the maturity of intelligent transportation technology, which offers a new thinking on traditional trip survey, travel behavior research and travel demand prediction. The aim of this paper is to introduce an approach to extract travel mode and activity type information from the travel trajectory data collected by the location based services offered by smart-phone, which will be helpful to increase the efficiency of resident trip survey, decrease the subjectivity during the survey, retrench the cost and duration of the data collection, and present the data and decision support for urban transportation planning and management.This research combines the theories and techniques of trip chain, pattern recognition, passive trip data collection, and location based services. Firstly, the paper establishes a trip-activity chain pattern through the structure of activity chain and trip chain, divides the pattern into trip sub-pattern and activity sub-pattern, analyzes the characteristics of the sub-patterns, and studies the relationship between travel trajectories and trip-activity chain pattern. Then, the location and sensor modules of smart-phone are involved into the method of travel trajectory data collection by the thinking of LBS location information. The interpolation is applied to complement the missing points of the trajectories, the Kalman Filter is adopted to denoise the trajectory data, and the sliding threshold method is constructed to segment the trajectories into trip parts and activity parts. Moreover, the paper concludes the feature vectors of trip sub-pattern and activity sub-pattern, gives a method to convert the trajectory parameter vectors into the pattern feature vectors, uses frequency distribution histogram and F-score method to illustrate the divisibility between two classes through the qualitative and quantitative perspectives, and trains the decision tree, BP Neural Networks, RBF Neural Networks, and support vector machines as the classifiers to proceed the recognition of the sub-pattern samples.The trajectory data collected in Dalian is involved and practiced by the approach mentioned above in the empirical study at last. The case study evaluates the effect of the approach including data complement, smoothing, segmentation, and recognition. The results show that the approach presented by this paper is effective to recognize the trip-activity chain pattern information from the LBS trajectories.