Measuring and enhancing the transferability of hidden Markov models for dynamic travel behavioral analysis

As an emerging dynamic modeling method that incorporates time-dependent heterogeneity, hidden Markov models (HMM) are receiving increased research attention with regards to travel behavior modeling and travel demand forecasting. This paper focuses on the model transferability of HMM. Based on a series of transferability and goodness-of-fit measures, it finds that HMMs have a superior performance in predicting future transportation mode choice, compared to conventional choice models. Aimed at further enhancing its transferability, this paper proposes a Bayesian conditional recalibration approach that maps the model prediction directly to the context data. Compared to traditional model transferring methods, the proposed approach does not assume fixed parameterization and recalibrates the utilities and the prediction directly. A comparison between the proposed approach and the traditional transfer-scaling favors our approach, with higher goodness-of-fit. This paper fills the gap in understanding the transferability of HMM and proposes a practical method that enables potential applications of HMM.

     

Lane changing intention recognition based on speech recognition models

Poor driving habits such as not using turn signals when changing lanes present a major challenge to advanced driver assistance systems that rely on turn signals. To address this problem, we propose a novel algorithm combining the hidden Markov model (HMM) and Bayesian filtering (BF) techniques to recognize a driver’s lane changing intention. In the HMM component, the grammar definition is inspired by speech recognition models, and the output is a preliminary behavior classification. As for the BF component, the final behavior classification is produced based on the current and preceding outputs of the HMMs. A naturalistic data set is used to train and validate the proposed algorithm. The results reveal that the proposed HMM–BF framework can achieve a recognition accuracy of 93.5% and 90.3% for right and left lane changing, respectively, which is a significant improvement compared with the HMM-only algorithm. The recognition time results show that the proposed algorithm can recognize a behavior correctly at an early stage.     

A competing Markov model for cracking prediction on civil structures

Cracks on the surface of civil structures (e.g. pavement sections, concrete structures) progress in several formations and under different deterioration mechanisms. In monitoring practice, it is often that cracking type with its worst damage level is selected as a representative condition state, while other cracking types and their damage levels are neglected in records, remaining as hidden information. Therefore, the practice in monitoring has a potential to conceal with a bias selection process, which possibly result in not optimal intervention strategies. In overcoming these problems, our paper presents a non-homogeneous Markov hazard model, with competing hazard rates. Cracking condition states are classified in three types (longitudinal crack, horizontal crack, and alligator crack), with three respective damage levels. The dynamic selection of cracking condition states are undergone a competing process of cracking types and damage levels. We apply a numerical solution using Bayesian estimation and Markov Chain Monte Carlo method to solve the problem of high-order integration of complete likelihood function. An empirical study on a data-set of Japanese pavement system is presented to demonstrate the applicability and contribution of the model.     

Exploring the influence of built environment on tour-based commuter mode choice: A cross-classified multilevel modeling approach

Understanding travel behavior and its relationship to urban form is vital for the sustainable planning strategies aimed at automobile dependency reduction. The objective of this study is twofold. First, this research provides additional insights to examine the effects of built environment factors measured at both home location and workplace on tour-based mode choice behavior. Second, a cross-classified multilevel probit model using Bayesian approach is employed to accommodate the spatial context in which individuals make travel decisions. Using Washington, D.C. as our study area, the home-based work (Home-work) tour in the AM peak hours is used as the analysis unit. The empirical data was gathered from the Washington-Baltimore Regional Household Travel Survey 2007–2008. For parameter estimation, Bayesian estimation method integrating Markov Chain Monte Carlo (MCMC) sampling is adopted. Our findings confirmed the important role that the built environment at both home location and work ends plays in affecting commuter mode choice behavior. Meanwhile, a comparison of different model results shows that the cross-classified multilevel probit model offers significant improvements over the traditional probit model. The results are expected to give a better understanding on the relationship between the built environment and commuter mode choice behavior.     

Probe vehicle based real-time traffic monitoring on urban roadways

Travel time estimation and prediction on urban arterials is an important component of Active Traffic and Demand Management Systems (ATDMS). This paper aims in using the information of GPS probes to augment less dynamic but available information describing arterial travel times. The direction followed in this paper chooses a cooperative approach in travel time estimation using static information describing arterial geometry and signal timing, semi-dynamic information of historical travel time distributions per time of day, and utilizes GPS probe information to augment and improve the latter. First, arterial travel times are classified by identifying different travel time states, then link travel time distributions are approximated using mixtures of normal distributions. If prior travel time data is available, travel time distributions can be estimated empirically. Otherwise, travel time distribution can be estimated based on signal timing and arterial geometry. Real-time GPS travel time data is then used to identify the current traffic condition based on Bayes Theorem. Moreover, these GPS data can also be used to update the parameters of the travel time distributions using a Bayesian update. The iterative update process makes the posterior distributions more and more accurate. Finally, two comprehensive case studies using the NGSIM Peachtree Street dataset, and GPS data of Washington Avenue in Minneapolis, were conducted. The first case study estimated prior travel time distributions based on signal timing and arterial geometry under different traffic conditions. Travel time data were classified and corresponding distributions were updated. In addition, results from the Bayesian update and EM algorithm were compared. The second case study first tested the methodologies based on real GPS data and showed the importance of sample size. In addition, a methodology was proposed to distinguish new traffic conditions in the second case study.     

Modelling the dynamics of bus use in a changing travel environment using panel data

Panel data offers the potential to represent the influence on travel choices of changing circumstances, past history and persistent individual differences (unobserved heterogeneity). A four-wave panel survey collected data on the travel choices of residents before and after the introduction of a new bus rapid transit service. The data shows gradual changes to bus use over the four waves, implying time was required for residents to become aware of the new service and to adapt to it. Ordered response models are estimated for bus use over the survey period. The results show that the influence of level of service (LOS) is underestimated if unobserved heterogeneity is not taken into account. The delayed response to the new service is able to be well represented by including LOS as a lagged variable. Current bus use is found to be conditioned on past bus use, but with additional influence of lagged LOS and unobserved heterogeneity. It is shown how different model specifications generate different evolution patterns with the most realistic predictions arising from a model which takes into account lagged responses to change in LOS and unobserved heterogeneity. The paper demonstrates the feasibility of developing panel data models that can be applied to forecasting the effect of interventions in the travel environment. Longer panels—encompassing periods of both stability and change—are required to support future efforts at modelling travel choice dynamics.     

A latent class joint mode and departure time choice model for the Greater Toronto and Hamilton Area

This paper presents a closed-form Latent Class Model (LCM) of joint mode and departure time choices. The proposed LCM offers compound substitution patterns between the two choices. The class-specific choice models are of two opposing nesting structures, each of which provides expected maximum utility feedback to the corresponding class membership model. Such feedback allows switching class membership in response to the changes in choice contexts. The model is used for an empirical investigation of commuting mode and departure time choices in the Greater Toronto and Hamilton Area (GTHA) by using a large sample household travel survey dataset. The empirical model reveals that overall 38% of the commuters in the GTHA are more likely to switch modes than departure times and 62% of them are more likely to do the reverse. The empirical model also reveals that the average Subjective Value of Travel Time Savings (SVTTS) of the commuters in the GTHA can be as low as 3 dollars if a single choice pattern of departure time choices nested within mode choices is considered. It can also be as high as 67 dollars if the opposite nesting structure is assumed. However, the LCM estimates the average SVTTS to be around 27 dollars in the GTHA. An empirical scenario analysis by using the estimated model indicates that a 50% increase in morning peak period car travel time does not sway more than 4% of commuters from the morning peak period.

     

Bayesian inference of channelized section spillover via Markov Chain Monte Carlo sampling

Channelized section spillover (CSS) is usually referred to the phenomenon of a traffic flow being blocked upstream and not being able to enter the downstream channelized section. CSS leads to extra delays, longer queues, and a biased detection of the flow rate. An estimation of CSS, including its occurrence and duration, is helpful for analysis of the state of traffic flow, as a basis for traffic evaluation and management. This has not been studied or reported in prior literature. A Bayesian model is developed through this research to estimate CSS, with its occurrence and duration formulated as a posterior distribution of given travel time and flow rate data. Basic properties of CSS are discussed initially, followed by a macroscopic model that explicitly models the CSS and encapsulates first-in-first-out (FIFO) behavior at an upstream section, with a goal of generating the prior distribution of CSS duration. Posterior distribution is then constructed using the detected flow rate and travel time vehicles samples. The Markov Chain Monte Carlo (MCMC) sampling method is used to solve this Bayesian model. The proposed model is implemented and tested in a channelized intersection and its modeling results are compared with Vissim simulation outputs, which demonstrated satisfactory results.     

Multi-state supernetwork framework for the two-person joint travel problem

Most travel behavior studies on route and mode choice focus only on an individual level. This paper adopts the concept of multi-state supernetworks to model the two-person joint travel problem (JTP). Travel is differentiated in terms of activity-vehicle-joint states, i.e. travel separately or jointly with which transport mode and with which activities conducted. In each state, route choice can be addressed given the state information and travel preference parameters. The joint travel pattern space is represented as a multi-state supernetwork, which is constructed by assigning the individual and joint networks to all possible states and connecting them via transfer links at joints where individuals can meet or depart. Besides route choice, the choices of where and when to meet, and which transport mode(s) to use can all be explicitly represented in a consistent fashion. A joint path through the supernetwork corresponds to a specific joint travel pattern. Then, JTP is reduced to an optimization problem to find the joint path with the minimum disutility. Three standard shortest path algorithm variants are proposed to find the optimal under different scenarios. The proposed framework further indicates the feasibility of multi-state supernetworks for addressing high dimensional problems and contributes to the design of a next generation of joint routing systems.     

Modeling commuting mode choice jointly with work start time and work duration

This paper presents a joint trivariate discrete-continuous-continuous model for commuters’ mode choice, work start time and work duration. The model is designed to capture correlations among random components influencing these decisions. For empirical investigation, the model is estimated using a data set collected in the Greater Toronto Area (GTA) in 2001. Considering the fact that work duration involves medium- to long-term decision making compared to short-term activity scheduling decisions, work duration is considered endogenous to work start time decisions. The empirical model reveals many behavioral details of commuters’ mode choice, work start time and duration decisions. The primary objective of the model is to predict workers’ work schedules according to mode choice, which is considered a skeletal activity schedule in activity-based travel demand models. However, the empirical model reveals many behavioral details of workers’ mode choices and work scheduling. Independent application of the model for travel demand management policy evaluations is also promising, as it provides better value in terms of travel time estimates.     

Activity imputation for trip-chains elicited from smart-card data using a continuous hidden Markov model

Although smart-card data were expected to substitute for conventional travel surveys, the reality is that only a few automatic fare collection (AFC) systems can recognize an individual passenger's origin, transfer, and destination stops (or stations). The Seoul metropolitan area is equipped with a system wherein a passenger's entire trajectory can be tracked. Despite this great advantage, the use of smart-card data has a critical limitation wherein the purpose behind a trip is unknown. The present study proposed a rigorous methodology to impute the sequence of activities for each trip chain using a continuous hidden Markov model (CHMM), which belongs to the category of unsupervised machine-learning technologies. Coupled with the spatial and temporal information on trip chains from smart-card data, land-use characteristics were used to train a CHMM. Unlike supervised models that have been mobilized to impute the trip purpose to GPS data, A CHMM does not require an extra survey, such as the prompted-recall survey, in order to obtain labeled data for training. The estimated result of the proposed model yielded plausible activity patterns that are intuitively accountable and consistent with observed activity patterns.     

A copula-based approach for estimating the travel time reliability of urban arterial

Estimating the travel time reliability (TTR) of urban arterial is critical for real-time and reliable route guidance and provides theoretical bases and technical support for sophisticated traffic management and control. The state-of-art procedures for arterial TTR estimation usually assume that path travel time follows a certain distribution, with less consideration about segment correlations. However, the conventional approach is usually unrealistic because an important feature of urban arterial is the dependent structure of travel times on continuous segments. In this study, a copula-based approach that incorporates the stochastic characteristics of segments travel time is proposed to model arterial travel time distribution (TTD), which serves as a basis for TTR quantification. First, segments correlation is empirically analyzed and different types of copula models are examined. Then, fitting marginal distributions for segment TTD is conducted by parametric and non-parametric regression analysis, respectively. Based on the estimated parameters of the models, the best-fitting copula is determined in terms of the goodness-of-fit tests. Last, the model is examined at two study sites with AVI data and NGSIM trajectory data, respectively. The results of path TTD estimation demonstrate the advantage of the proposed copula-based approach, compared with the convolution model without capturing segments correlation and the empirical distribution fitting methods. Furthermore, when considering the segments correlation effect, it was found that the estimated path TTR is more accurate than that by the convolution model.     

Hidden Markov Model-based population synthesis

Micro-simulation travel demand and land use models require a synthetic population, which consists of a set of agents characterized by demographic and socio-economic attributes. Two main families of population synthesis techniques can be distinguished: (a) fitting methods (iterative proportional fitting, updating) and (b) combinatorial optimization methods. During the last few years, a third outperforming family of population synthesis procedures has emerged, i.e., Markov process-based methods such as Monte Carlo Markov Chain (MCMC) simulations. In this paper, an extended Hidden Markov Model (HMM)-based approach is presented, which can serve as a better alternative than the existing methods. The approach is characterized by a great flexibility and efficiency in terms of data preparation and model training. The HMM is able to reproduce the structural configuration of a given population from an unlimited number of micro-samples and a marginal distribution. Only one marginal distribution of the considered population can be used as a boundary condition to “guide” the synthesis of the whole population. Model training and testing are performed using the Survey on the Workforce of 2013 and the Belgian National Household Travel Survey of 2010. Results indicate that the HMM method captures the complete heterogeneity of the micro-data contrary to standard fitting approaches. The method provides accurate results as it is able to reproduce the marginal distributions and their corresponding multivariate joint distributions with an acceptable error rate (i.e., SRSME=0.54 for 6 synthesized attributes). Furthermore, the HMM outperforms IPF for small sample sizes, even though the amount of input data is less than that for IPF. Finally, simulations show that the HMM can merge information provided by multiple data sources to allow good population estimates.     

An adaptive hawkes process formulation for estimating time-of-day zonal trip arrivals with location-based social networking check-in data

Location-Based Social Networking (LBSN) services, such as Foursquare, Facebook check-ins, and Geo-tagged Twitter tweets, have emerged as new secondary data sources for studying individual travel mobility patterns at a fine-grained level. However, the differences between human social behavioral and travel patterns can cause significant sampling bias for travel demand estimation. This paper presents a dynamic model to estimate time-of-day zonal trip arrival patterns. In the proposed model, the state propagation is formulated by the Hawkes process; the observation model implements LBSN sampling. The proposed model is applied to Foursquare check-in data collected from Austin, Texas in 2010 and calibrated with Origin-Destination (OD) data and time of day factor from Capital Area Metropolitan Planning Organization (CAMPO). The proposed model is compared with a simple aggregation model of trip purposes and time of day based on a prior daily OD estimation model using the LBSN data. The results illustrate the promising benefits of applying stochastic point process models and state-space modeling in time-of-day zonal arrival pattern estimation with the LBSN data. The proposed model can significantly reduce the number of parameters to calibrate in order to reduce the sampling bias of LBSN data for trip arrival estimation.     

A travel time reliability model of urban expressways with varying levels of service

Urban expressways usually experience several levels of service (LOS) because of the stop-and-go traffic flow caused by congestion. Moreover, multiple shock waves generate at different LOS interfaces. The dynamic of shock waves strongly influences the travel time reliability (TTR) of urban expressways. This study proposes a path TTR model that considers the dynamic of shock waves by using probability-based method to characterize the TTR of urban expressways with shock waves. Two model parameters are estimated, namely distribution of travel time (TT) per unit distance and travel distances in different LOS segments. Generalized extreme value distribution and generalized Pareto distribution are derived as distributions of TT per unit distance for six different LOS. Distribution parameters are estimated by using historical floating car data. Travel distances in different LOS segments are calculated based on shock wave theory. The range of TT along the path, which can help drivers arrange their trips, can be obtained from the TTR model. Finally, comparison is made among the proposed TTR model, generalized Pareto contrast model, which does not consider different LOS or existence of shock waves, and normal contrast model, which assumes TT per unit distance as normal distribution without considering shock wave. Results show that the proposed model achieves higher prediction accuracy and reduces the prediction range of TT. The conclusions can be further extended to TT prediction and assessment of measures to improve reliability of TT in a network.     

Bayesian estimation of multinomial probit models of work trip choice

In this paper, we estimate a multinomial probit model of work trip mode choice in Seoul, Korea, using the Bayesian approach with Gibbs sampling. This method constructs a Markov chain Gibbs sampler that can be used to draw directly from the exact posterior distribution and perform finite sample likelihood inference. We estimate direct and cross-elasticities with respect to travel cost and the value of time. Our results show that travel demands are more sensitive to travel time than travel cost. The cross-elasticity results show that the bus has a greater substitute relation to the subway than the auto (and vice versa) and that an increase in the cost of an auto will increase the demand for bus transport more so than that of the subway.     

A tensor-based Bayesian probabilistic model for citywide personalized travel time estimation

Urban travel time information is of great importance for many levels of traffic management and operation. This paper develops a tensor-based Bayesian probabilistic model for citywide and personalized travel time estimation, using the large-scale and sparse GPS trajectories generated by taxicabs. Combined with the knowledge learned from historical trajectories, travel times of different drivers on all road segments in some time slots are modeled with a 3-order tensor. This tensor-based modeling approach incorporates both the spatial correlation between different road segments and the person-specific variation between different drivers, as well as the coarse-grain temporal correlation between recent and historical traffic conditions and the fine-grain temporal correlation between different time slots. To account for the variability caused by the intrinsic uncertainties in urban road network, each travel time entry in the built tensor is treated as a variable following a log-normal distribution. With the help of the fully Bayesian treatment, the model achieves automatic hyper-parameter tuning and model complexity controlling, and therefore the problem of over-fitting is prevented even when the used data is large-scale and sparse. The proposed model is applied to a real case study on the citywide road network of Beijing, China, using the large-scale and sparse GPS trajectories collected from over 32,670 taxicabs for a period of two months. Empirical results of extensive experiments demonstrate that the proposed model provides an effective and robust approach for urban travel time estimation and outperforms the considered competing methods.     

A practical approach to identifying the market potential for high speed rail: A case study in the Sydney-Canberra corridor

This paper presents the methodology and selective empirical results from a study of the demand for a high speed rail system serving the Sydney-Canberra corridor currently dominated by air travel for business trips and car travel for non-business trips. We outline the steps involved in the study from problem specification, data needs, development of base year trip tables, model specification and estimation to establish switching behaviour in the presence of a new mode and calculation of induced demand for current travellers. A stated choice heteroskedastic extreme value switching model is used to evaluate the choice of fare type for business and non-business travel given the current mode used in the corridor for each sampled traveller conventional train, charter coach, scheduled coach, plane or car. Starting with the current travel profile, patronage can be predicted under alternative fare regimes, taking into account diverted traffic, induced traffic and growth. Treating fare class as endogenous enhances the real choice context facing potential patrons.     

Estimating a model of dynamic activity generation based on one-day observations: Method and results

In this paper we develop and explore an approach to estimate dynamic models of activity generation on one-day travel-diary data. Dynamic models predict multi-day activity patterns of individuals taking into account dynamic needs as well as day-varying preferences and time-budgets. We formulate an ordered-logit model of dynamic activity-agenda-formation decisions and show how one-day observation probabilities can be derived from the model as a function of the model’s parameters and, with that, how parameters can be estimated using standard loglikelihood estimation. A scale parameter cannot be identified because information on within-person variability is lacking in one-day data. An application of the method to data from a national travel survey illustrates the method. A test on simulated data indicates that, given a pre-set scale, the parameters can be identified and that estimates are robust for a source of heterogeneity not captured in the model. This result indicates that dynamic activity-based models of the kind considered here can be estimated from data that are less costly to collect and that support the large sample sizes typically required for travel-demand modeling. We conclude therefore that the proposed approach opens up a way to develop large-scale dynamic activity-based models of travel demand.     

Tour-based mode choice of joint household travel patterns on weekend and weekday

Joint household travel, with or without joint participation in an activity, constitutes a fundamental aspect in modelling activity-based travel behaviour. This paper examines joint household travel arrangements and mode choices using a utility maximising approach. An individual tour-based mode choice model is formulated contingent on the choice of joint tour patterns where joint household activities and shared ride arrangements are recognised as part of the joint household decision-making that influences the travel modes of each household member. Two models, one for weekend and one for weekday, are estimated using empirical data from the Sydney Household Travel Survey. The results show that weekend travel is characterised by a high joint household activity participation rate while weekday travel is distinguished by more intra-household shared ride arrangements. The arrangements of joint household travel are highly associated with travel purpose, social and mobility constraints and household resources. On weekends, public transport is mainly used by captive users (i.e., no-car households and students) and its share is about half of that on weekdays. Also, the value of travel time savings (VOTs) are found to be higher on weekends than on weekdays, running entirely counter to the common belief that weekend VOTs are lower than weekday VOTs. This paper highlights the importance of studying joint household travel and using different transport management measures for alleviating traffic congestion on weekdays and weekends.