Bayesian predictive travel time methodology for advanced traveller information system

Abstract

SUMMARYTravellers can benefit from the availability of point-to-point driving time estimates on a real time basis for making travel decisions such as route choice at strategic locations (e.g. junctions of major routes). This paper reports a predictive travel time methodology that features a Bayesian approach to fusing and updating information for use in advanced traveller information system. The methodology addresses the issue that data captured in real time on travel conditions becomes obsolete and has archival value only unless it is used as an input to a predictive travel time method for updating the information. The need for fusing real time data with other factors that influence travel time is defined and the concept of predictive travel time is discussed. The methodological framework and its components are advanced and an example application is provided for illustrating the fusion of data captured by infrastructure-based and mobile technology with model-based predictions in order to produce expected travel times.

An improved incremental assignment model for parking variable message sign location problem

“…Drivers choose parking lots based on the information of space availability provided by parking VMS. However, drivers expect to receive the information estimated reasonably, rather than rough real-time information [26]. The arrival parking flow volume of parking lot k can be calculated as follows:…”

Comparative analysis of implicit models for real‐time short‐term traffic predictions

“…Van Hinsbergen et al [21] introduced the concept of a Bayesian committee, which combines the predictions of multiple NNs with different structures and different weight distributions. Khan [22] proposed a Bayesian approach to combine estimates obtained by naïve estimation from real-time detected data and estimates provided by even different models. Wang et al [23] modified the Bayesian combination method by introducing a dynamic computation of credits and incorporated three single predictors.…”

How to Provide Accurate and Robust Traffic Forecasts Practically?

“…Besides the above neural networks models, computational intelligence (CI) techniques that encompass fuzzy systems, machine learning and evolutionary computation have been successfully developed in the field of traffic forecasting. For instance, some literature applies Bayesian networks (Zhang et al , 2004;Castillo et al, 2008) and Bayesian inference based regression techniques (Khan, 2011;Tebaldi et al, 2002;Sun et al, 2005Sun et al, , 2006Zheng et al, 2006;Ghosh et al, 2007), some literature uses fuzzy systems or fuzzy NNs to predict the traffic states (Dimitriou et al, 2008;Quek et al, 2009). While others start to explore support vector regression (SVR) to model traffic characteristics and produce prediction of traffic states (Castro-Neto, 2009;Ding et al, 2002;Hong, 2011;Hong et al, 2011;Wu et al, 2004;Vanajakshi & Rilett, 2004).…”