Spatio-temporal modeling and prediction on travel behaviour analysis

Abstract

Travel behaviour analysis has been an essential topic since 1970s. Research shows that there is a strong relationship between citizens’ daily movements with activities and travel demand, so analysing human behaviours in different time and locations is quite significant to understand travel demand, and improve the transportation design. Primarily, traditional data collection methods are questionnaires and interviews. However, previous research have clear gaps: first, trip patterns extracted from the small set of data are limited and not convincing enough. This makes traditional human behaviour models inaccurate; second, the travel patterns and activity patterns discovered from previous research are limited. Moreover, previous research often ignored the impact from human activity patterns on their travel behaviours. Nowadays, it becomes feasible to obtain large trajectory data with development in sensing technologies, such as GPS trajectory data. How to use large trajectory data to extract trip patterns, forecast travel behaviours in different dimensions (spatially and temporally), and estimate dynamic travel demands using citizens’ travel behaviours have become urgent research questions to be answered. Focusing on the research questions, my PhD project analyses people’s travel behaviours from three aspects: (1) discovers spatial-temporal patterns of common movements; (2) estimate activity patterns that are not explained in trip records; (3) predict travel demand and city movement status during different periods and locations. In order to carry out the experiments, I have collected a large amount of relevant data, which are taxi trajectory data, trip diaries, house price data, bike sharing data, employment data etc. Three types of methods are employed to conduct my experiments: (i) geography and data analysis methods (Huff model, Geographical Temporal Weighted Regression, Bayesian probabilities, Distance decay function, Monte Carlo simulation, ARIMA etc.); (ii) machine learning methods (K-means Clustering, ANN, SVR, etc.); and (iii) Simulation method (Agent-based modeling and simulation). After a mass of experiments and explorations, my PhD project hastravel behaviours. The main conclusions of my PhD studies are as follows: (1) passengers have very different travel behaviours in different time and locations, which are largely related to the amount of their spare time and their income; (2) passengers’ activity can be highly inferred from their trip information (travel time, drop-off time, and drop-off location); (3) whether a trip has return trip is closely related to passengers’ travel time; (4) citizens’ daily routine, activity schedule, and trip numbers have strong regularities. In addition to the academic contributions, my research results have potential managerial contributions. For example, the model from Chapter 4 could be used to determine the locations of new infrastructure within specified conditions. The model from Chapter 5 could be used for targeted advertising. The forecasting model from Chapter 6 can be used to forecast travel demands, and then assist the department of municipal transport administration plan the scale of road construction, and the number of traffic lights.

Date of Award	8 Nov 2020
Original language	English
Awarding Institution	Univerisity of Nottingham
Supervisor	Ruibin Bai (Supervisor), John Cartlidge (Supervisor) & Guoping Qiu (Supervisor)