A New Method for Travel Time Prediction
This research adopts a functional data analysis method that is mainly based on a mixture prediction method to analyze and predict travel times; such analysis and prediction constitute an essential component in Intelligent Transportation Systems applications. The mixture prediction method is developed through three major modules, i.e., functional clustering for historical functional travel time patterns, probabilistic functional classification for newly observed travel time trajectories, and linear regression model fitting for travel time prediction.
The research framework was demonstrated with data on 80 days of Electronic Toll Collection (ETC) travel times retrieved from the website under the database TDCS_M04A constructed between interchanges 01F0155S (Donghu) and 01F0880S (Chupei) on Taiwan Area National Freeway Bureau of Republic of China’s Ministry of Transportation and Communications. The demonstration encompassed 57 weekdays and 23 holidays from 2016/09/01 to 2016/11/30.
The preliminary result shows the best combination of observed time (ω) and unobserved (ν) time occurred at (ω=3, ν=2) with mean absolute percentage error (MAPE) equal to 7.26 and the usefulness of functional data analysis in analyzing and predicting the travel time trajectories on freeways is supported, similar to results for the traffic flow trajectories (Chiou, 2012). However, intensive research on different combination of (ω,ν) under various traffic conditions must be performed before a firm conclusion can be reached. Moreover, the merit of the functional data analysis, particularly the functional clustering method, can be readily employed by other “decomposition” type methods, such as Hilbert-Hwang Transform (HHT), to enhance their accuracy in prediction of travel times.
Keywords - Functional Data Analysis; Functional Clustering; Posterior Cluster Membership Probability; Functional Mixture Prediction Model; Electronic Toll Collection Travel Times.