Dynamic travel time prediction using data clustering and genetic programming

The current state-of-practice for predicting travel times assumes that the speeds along the various roadway segments remain constant over the duration of the trip. This approach produces large prediction errors, especially when the segment speeds vary temporally. In this paper, we develop a data clustering and genetic programming approach for modeling and predicting the expected, lower, and upper bounds of dynamic travel times along freeways. The models obtained from the genetic programming approach are algebraic expressions that provide insights into the spatiotemporal interactions. The use of an algebraic equation also means that the approach is computationally efficient and suitable for real-time applications. Our algorithm is tested on a 37-mile freeway section encompassing several bottlenecks. The prediction error is demonstrated to be significantly lower than that produced by the instantaneous algorithm and the historical average averaged over seven weekdays (p-value <0.0001). Specifically, the proposed algorithm achieves more than a 25% and 76% reduction in the prediction error over the instantaneous and historical average, respectively on congested days. When bagging is used in addition to the genetic programming, the results show that the mean width of the travel time interval is less than 5 min for the 60–80 min trip.     

The effects of activity-travel context and individual attitudes on car-sharing decisions under travel time uncertainty: A hybrid choice modeling approach

People’s daily decision to use car-sharing rather than other transport modes for conducting a specific activity has been investigated recently in assessing the market potential of car-sharing systems. Most studies have estimated transport mode choice models with an extended choice set using attributes such as average travel time and costs. However, car-sharing systems have some distinctive features: users have to reserve a car in advance and pay time-based costs for using the car. Therefore, the effects of activity-travel context and travel time uncertainty require further consideration in models that predict car-sharing demand. Moreover, the relationships between individual latent attitudes and the intention to use car-sharing have not yet been investigated in much detail. In contributing to the research on car-sharing, the present study is designed to examine the effects of activity-travel context and individual latent attitudes on short-term car-sharing decisions under travel time uncertainty. The effects of all these factors were simultaneously estimated using a hybrid choice modeling framework. The data used in this study was collected in the Netherlands, 2015 using a stated choice experiment. Hypothetical choice situations were designed to collect respondents’ intention to use a shared-car for their travel to work. A total of 791 respondents completed the experiment. The estimation results suggest that time constraints, lack of spontaneity and a larger variation in travel times have significant negative effects on people’s intention to use a shared-car. Furthermore, this intention is significantly associated with latent attitudes about pro-environmental preferences, the symbolic value of cars, and privacy-seeking.     

Using SenseCam to pursue “ground truth” for global positioning system travel surveys

Global Positioning System (GPS) surveys have been conducted for the past decade. Although GPS records were initially regarded as providing “ground truth” of travel, it has been found subsequently that they have some errors. SenseCam, a small passive digital camera, provides a chance to pursue ground truth by capturing images every 20 s on average. This paper discusses how SenseCam could help GPS data collection and shows potential benefits for both SenseCam and GPS research. This paper also investigates the performance of GPS devices in detail in terms of recording data by comparisons between GPS results and SenseCam images. The specific issue of missing GPS data is discussed and examined in this paper.     

The role of alternative fuel vehicles: Using behavioral and sensor data to model hierarchies in travel

Greater adoption and use of alternative fuel vehicles (AFVs) can be environmentally beneficial and reduce dependence on gasoline. The use of AFVs vis-à-vis conventional gasoline vehicles is not well understood, especially when it comes to travel choices and short-term driving decisions. Using data that contains a sufficiently large number of early AFV adopters (who have overcome obstacles to adoption), this study explores differences in use of AFVs and conventional gasoline vehicles (and hybrid vehicles). The study analyzes large-scale behavioral data integrated with sensor data from global positioning system devices, representing advances in large-scale data analytics. Specifically, it makes sense of data containing 54,043,889 s of speed observations, and 65,652 trips made by 2908 drivers in 5 regions of California. The study answers important research questions about AFV use patterns (e.g., trip frequency and daily vehicle miles traveled) and driving practices. Driving volatility, as one measure of driving practice, is used as a key metric in this study to capture acceleration, and vehicular jerk decisions that exceed certain thresholds during a trip. The results show that AFVs cannot be viewed as monolithic; there are important differences within AFV use, i.e., between plug-in hybrids, battery electric, or compressed natural gas vehicles. Multi-level models are particularly appropriate for analysis, given that the data are nested, i.e., multiple trips are made by different drivers who reside in various regions. Using such models, the study also found that driving volatility varies significantly between trips, driver groups, and regions in California. Some alternative fuel vehicles are associated with calmer driving compared with conventional vehicles. The implications of the results for safety, informed consumer choices and large-scale data analytics are discussed.     

Probe vehicle data sampled by time or space: Consistent travel time allocation and estimation

A characteristic of low frequency probe vehicle data is that vehicles traverse multiple network components (e.g., links) between consecutive position samplings, creating challenges for (i) the allocation of the measured travel time to the traversed components, and (ii) the consistent estimation of component travel time distribution parameters. This paper shows that the solution to these problems depends on whether sampling is based on time (e.g., one report every minute) or space (e.g., one every 500 m). For the special case of segments with uniform space-mean speeds, explicit formulae are derived under both sampling principles for the likelihood of the measurements and the allocation of travel time. It is shown that time-based sampling is biased towards measurements where a disproportionally long time is spent on the last segment. Numerical experiments show that an incorrect likelihood formulation can lead to significantly biased parameter estimates depending on the shapes of the travel time distributions. The analysis reveals that the sampling protocol needs to be considered in travel time estimation using probe vehicle data.     

Hitting the sweet spot: Variability in commute lengths and vehicle emissions across a diverse state

In this paper we consider travel across Virginia and identify sustainability “sweet spots” where commute lengths and vehicle emissions per mile combine to maximize green travel in terms of total CO₂ emissions associated with commuting. The analysis is conducted across local voter precincts (N = 2373 in the state) because they are a useful proxy for neighborhoods and well-sized for implementing policy designed to encourage sustainable travel behavior. Virginia is especially appropriate for an examination of variability in sustainable travel behavior and technologies because the state’s transportation, demographic, and political patterns are particularly diverse and have been changing rapidly. We identify four Virginia precinct-based sustainability clusters: Sweet Spots, Emerging Sweet Spots, Neutral and Non-sustaining. A model of demographic differences among the clusters shows that sustainability outcomes, understood in terms of both local commute behavior and vehicle emissions, are significantly associated with the diverse demography and politics of the state. We also look at changes in transportation sustainability and socio-demographic trends within the clusters over the past half-decade, showing that differences in sustainability and demographic metrics are actually accelerating within the state over time. We conclude with a discussion of the implications of the differences among the clusters for developing and implementing effective transportation sustainability policies across the state.     

Help or hindrance? The travel,energy and carbon impacts of highly automated vehicles

Experts predict that new automobiles will be capable of driving themselves under limited conditions within 5–10 years, and under most conditions within 10–20 years. Automation may affect road vehicle energy consumption and greenhouse gas (GHG) emissions in a host of ways, positive and negative, by causing changes in travel demand, vehicle design, vehicle operating profiles, and choices of fuels. In this paper, we identify specific mechanisms through which automation may affect travel and energy demand and resulting GHG emissions and bring them together using a coherent energy decomposition framework. We review the literature for estimates of the energy impacts of each mechanism and, where the literature is lacking, develop our own estimates using engineering and economic analysis. We consider how widely applicable each mechanism is, and quantify the potential impact of each mechanism on a common basis: the percentage change it is expected to cause in total GHG emissions from light-duty or heavy-duty vehicles in the U.S. Our primary focus is travel related energy consumption and emissions, since potential lifecycle impacts are generally smaller in magnitude. We explore the net effects of automation on emissions through several illustrative scenarios, finding that automation might plausibly reduce road transport GHG emissions and energy use by nearly half – or nearly double them – depending on which effects come to dominate. We also find that many potential energy-reduction benefits may be realized through partial automation, while the major energy/emission downside risks appear more likely at full automation. We close by presenting some implications for policymakers and identifying priority areas for further research.     

Would personal carbon trading reduce travel emissions more effectively than a carbon tax?

The application of personal carbon trading (PCT) to transport choices has recently been considered in the literature as a means of reducing CO₂ emissions. Its potential effectiveness in changing car travel behavior is compared to the conventional carbon tax (CT) by means of a stated preferences survey conducted among French drivers (N ∼ 300). We show evidence that PCT could effectively change travel behavior and hence reduce transport emissions from personal travel. There is however a definite reluctance to reduce car travel. We were unable to demonstrate any significant difference between the effectiveness of PCT and the CT with regard to changing travel behavior. However, in the experiment, the PCT scheme provided consistent results while this was not the case for the CT scheme. Further research is needed into the “social norm” conveyed by a personal emissions allowance.     

Inferring patterns in the multi-week activity sequences of public transport users

The public transport networks of dense cities such as London serve passengers with widely different travel patterns. In line with the diverse lives of urban dwellers, activities and journeys are combined within days and across days in diverse sequences. From personalized customer information, to improved travel demand models, understanding this type of heterogeneity among transit users is relevant to a number of applications core to public transport agencies’ function. In this study, passenger heterogeneity is investigated based on a longitudinal representation of each user’s multi-week activity sequence derived from smart card data. We propose a methodology leveraging this representation to identify clusters of users with similar activity sequence structure. The methodology is applied to a large sample (n = 33,026) from London’s public transport network, in which each passenger is represented by a continuous 4-week activity sequence. The application reveals 11 clusters, each characterized by a distinct sequence structure. Socio-demographic information available for a small sample of users (n = 1973) is combined to smart card transactions to analyze associations between the identified patterns and demographic attributes including passenger age, occupation, household composition and income, and vehicle ownership. The analysis reveals that significant connections exist between the demographic attributes of users and activity patterns identified exclusively from fare transactions.     

Using the built environment to oversample walk,transit, and bicycle travel

Characteristics of the built environment (BE) have been associated with walk, transit, and bicycle travel. These BE characteristics can be used by transportation researchers to oversample households from areas where walk, transit, or bicycle travel is more likely, resulting in more observations of these uncommon travel behaviors. Little guidance, however, is available on the effectiveness of such built environment oversampling strategies. This article presents measures that can be used to assess the effectiveness of BE oversampling strategies and inform future efforts to oversample households with uncommon travel behaviors. The measures are sensitivity and specificity, positive likelihood ratio (LR+), and positive predictive value (PPV). To illustrate these measures, they were calculated for 10 BE-defined oversampling strata applied post-hoc to a Seattle area household travel survey. Strata with an average block size of <10 acres within a ¼ mile of household residences held the single greatest potential for oversampling households that walk, use transit, and/or bicycle.     

Multi-step prediction of experienced travel times using agent-based modeling

This paper develops an agent-based modeling approach to predict multi-step ahead experienced travel times using real-time and historical spatiotemporal traffic data. At the microscopic level, each agent represents an expert in a decision-making system. Each expert predicts the travel time for each time interval according to experiences from a historical dataset. A set of agent interactions is developed to preserve agents that correspond to traffic patterns similar to the real-time measurements and replace invalid agents or agents associated with negligible weights with new agents. Consequently, the aggregation of each agent’s recommendation (predicted travel time with associated weight) provides a macroscopic level of output, namely the predicted travel time distribution. Probe vehicle data from a 95-mile freeway stretch along I-64 and I-264 are used to test different predictors. The results show that the agent-based modeling approach produces the least prediction error compared to other state-of-the-practice and state-of-the-art methods (instantaneous travel time, historical average and k-nearest neighbor), and maintains less than a 9% prediction error for trip departures up to 60 min into the future for a two-hour trip. Moreover, the confidence boundaries of the predicted travel times demonstrate that the proposed approach also provides high accuracy in predicting travel time confidence intervals. Finally, the proposed approach does not require offline training thus making it easily transferable to other locations and the fast algorithm computation allows the proposed approach to be implemented in real-time applications in Traffic Management Centers.     

Analysis of plug-in hybrid electric vehicles’ utility factors using GPS-based longitudinal travel data

The benefit of using a PHEV comes from its ability to substitute gasoline with electricity in operation. Defined as the proportion of distance traveled in the electric mode, the utility factor (UF) depends mostly on the battery capacity, but also on many other factors, such as travel pattern and recharging pattern. Conventionally, the UFs are calculated based on the daily vehicle miles traveled (DVMT) by assuming motorists leave home in the morning with a full battery, and no charge occurs before returning home in the evening. Such an assumption, however, ignores the impact of the heterogeneity in both travel and charging behavior, such as going back home more than once in a day, the impact of available charging time, and the price of gasoline and electricity. Moreover, the conventional UFs are based on the National Household Travel Survey (NHTS) data, which are one-day travel data of each sample vehicle. A motorist’s daily travel distance variation is ignored. This paper employs the GPS-based longitudinal travel data (covering 3–18 months) collected from 403 vehicles in the Seattle metropolitan area to investigate how such travel and charging behavior affects UFs. To do this, for each vehicle, we organized trips to a series of home and work related tours. The UFs based on the DVMT are found close to those based on home-to-home tours. On the other hand, it is seen that the workplace charge opportunities significantly increase UFs if the CD range is no more than 40 miles.     

Hierarchical travel demand estimation using multiple data sources: A forward and backward propagation algorithmic framework on a layered computational graph

Aiming to develop a theoretically consistent framework to estimate travel demand using multiple data sources, this paper first proposes a multi-layered Hierarchical Flow Network (HFN) representation to structurally model different levels of travel demand variables including trip generation, origin/destination matrices, path/link flows, and individual behavior parameters. Different data channels from household travel surveys, smartphone type devices, global position systems, and sensors can be mapped to different layers of the proposed network structure. We introduce Big data-driven Transportation Computational Graph (BTCG), alternatively Beijing Transportation Computational Graph, as the underlying mathematical modeling tool to perform automatic differentiation on layers of composition functions. A feedforward passing on the HFN sequentially implements 3 steps of the traditional 4-step process: trip generation, spatial distribution estimation, and path flow-based traffic assignment, respectively. BTCG can aggregate different layers of partial first-order gradients and use the back-propagation of “loss errors” to update estimated demand variables. A comparative analysis indicates that the proposed methods can effectively integrate different data sources and offer a consistent representation of demand. The proposed methodology is also evaluated under a demonstration network in a Beijing subnetwork.     

Reconstructing maximum likelihood trajectory of probe vehicles between sparse updates

Data from connected probe vehicles can be critical in estimating road traffic conditions. Unfortunately, current available data is usually sparse due to the low reporting frequency and the low penetration rate of probe vehicles. To help fill the gaps in data, this paper presents an approach for estimating the maximum likelihood trajectory (MLT) of a probe vehicle in between two data updates on arterial roads. A public data feed from transit buses in the city of San Francisco is used as an example data source. Low frequency updates (at every 200 m or 90 s) leaves much to be inferred. We first estimate travel time statistics along the road and queue patterns at intersections from historical probe data. The path is divided into short segments, and an Expectation Maximization (EM) algorithm is proposed for allocating travel time statistics to each segment. Then the trajectory with the maximum likelihood is generated based on segment travel time statistics. The results are compared with high frequency ground truth data in multiple scenarios, which demonstrate the effectiveness of the proposed approach, in estimating both the trajectory while moving and the stop positions and durations at intersections.     

Application of Lagrangian relaxation approach to α-reliable path finding in stochastic networks with correlated link travel times

This study investigates the important problem of determining a reliable path in a stochastic network with correlated link travel times. First, the distribution of path travel time is quantified by using trip records from GPS probe vehicles. Second, the spatial correlation of link travel time is explicitly considered by using a correlation coefficient matrix, which is incorporated into the α-reliable path problem by Cholesky decomposition. Third, the Lagrangian relaxation based framework is used to handle the α-reliable path problem, by which the intractable problem with a non-linear and non-additive structure can be decomposed into several easy-to-solve problems. Finally, the path-finding performance of this approach is tested on a real-world network. The results show that 15 iterations of calculation can yield a small relative gap between upper and lower bounds of the optimal solution and the average running time is about 5 s for most OD settings. The applicability of α-reliable path finding is validated by a case study.     

Local online kernel ridge regression for forecasting of urban travel times

Accurate and reliable forecasting of traffic variables is one of the primary functions of Intelligent Transportation Systems. Reliable systems that are able to forecast traffic conditions accurately, multiple time steps into the future, are required for advanced traveller information systems. However, traffic forecasting is a difficult task because of the nonlinear and nonstationary properties of traffic series. Traditional linear models are incapable of modelling such properties, and typically perform poorly, particularly when conditions differ from the norm. Machine learning approaches such as artificial neural networks, nonparametric regression and kernel methods (KMs) have often been shown to outperform linear models in the literature. A bottleneck of the latter approach is that the information pertaining to all previous traffic states must be contained within the kernel, but the computational complexity of KMs usually scales cubically with the number of data points in the kernel. In this paper, a novel kernel-based machine learning (ML) algorithm is developed, namely the local online kernel ridge regression (LOKRR) model. Exploiting the observation that traffic data exhibits strong cyclic patterns characterised by rush hour traffic, LOKRR makes use of local kernels with varying parameters that are defined around each time point. This approach has 3 advantages over the standard single kernel approach: (1) It allows parameters to vary by time of day, capturing the time varying distribution of traffic data; (2) It allows smaller kernels to be defined that contain only the relevant traffic patterns, and; (3) It is online, allowing new traffic data to be incorporated as it arrives. The model is applied to the forecasting of travel times on London’s road network, and is found to outperform three benchmark models in forecasting up to 1 h ahead.     

Predicting travel time reliability using mobile phone GPS data

Estimates of road speeds have become commonplace and central to route planning, but few systems in production provide information about the reliability of the prediction. Probabilistic forecasts of travel time capture reliability and can be used for risk-averse routing, for reporting travel time reliability to a user, or as a component of fleet vehicle decision-support systems. Many of these uses (such as those for mapping services like Bing or Google Maps) require predictions for routes in the road network, at arbitrary times; the highest-volume source of data for this purpose is GPS data from mobile phones. We introduce a method (TRIP) to predict the probability distribution of travel time on an arbitrary route in a road network at an arbitrary time, using GPS data from mobile phones or other probe vehicles. TRIP captures weekly cycles in congestion levels, gives informed predictions for parts of the road network with little data, and is computationally efficient, even for very large road networks and datasets. We apply TRIP to predict travel time on the road network of the Seattle metropolitan region, based on large volumes of GPS data from Windows phones. TRIP provides improved interval predictions (forecast ranges for travel time) relative to Microsoft’s engine for travel time prediction as used in Bing Maps. It also provides deterministic predictions that are as accurate as Bing Maps predictions, despite using fewer explanatory variables, and differing from the observed travel times by only 10.1% on average over 35,190 test trips. To our knowledge TRIP is the first method to provide accurate predictions of travel time reliability for complete, large-scale road networks.     

Walking,cycling and the urban form: A Heckman selection model of active travel mode and distance by young adolescents

Physical inactivity of children and adolescents is a major public health challenge of the modern era but, when adequately promoted and nurtured, active travel offers immediate health benefits and forms future sustainable and healthy travel habits. This study explores jointly the choice and the extent of active travel of young adolescents while considering walking and cycling as distinct travel forms, controlling for objective urban form measures, and taking both a “street-buffer” looking at the immediate home surroundings and a “transport-zone” looking at wider neighborhoods. A Heckman selection model represents the distance covered while cycling (walking) given the mode choice being bicycle (walk) for a representative sample of 10–15 year-olds from the Capital Region of Denmark extracted from the Danish national travel survey. Results illustrate the necessity of different urban environments for walking and cycling, as the former relates to “street-buffer” urban form measures and the latter also to “transport-zone ” ones. Results also show that lessening the amount and the density of car traffic, diminishing the movement of heavy vehicles in local streets, reducing the conflict points with the density of intersections, and intervening on crash frequency and severity, would increase the probability and the amount of active travel by young adolescents. Last, results indicate that zones in rural areas and at a higher percentage of immigrants are likely to have lower probability and amount of active travel by young adolescents.     

Volume and GHG emissions of long-distance travelling by Western Europeans

It is generally recognised that long distance travelling accounts for a significant part of the mileage of person travel. However, estimates have been hardly made. The paper estimates volume and GHG emissions of long-distance travel by Western Europeans. The analysis is predominantly based on data of the DATELINE project, the only EU-wide survey on long-distance travelling, conducted in 2001 and 2002. Some studies demonstrate that DATELINE suffers from serious underreporting of journeys. We analysed the causes for underreporting and developed expansion factors that correct for that. These gave the opportunity to estimate long-distance travel volumes and related GHG emissions in 2001/2002. Next an update to 2013 is made using statistics on the development of tourist travel and patronage of long-distance modes. Defining long distance ⩾100 km crow-fly, the estimates per capita in the Western European countries in 2013 are 7.5 journeys (defined as round-trips), 8600 km, and 1300 kg greenhouse gasses. The estimated total GHG emissions of long-distance travelling is 520 megaton. In the Netherlands and Flanders, countries where data on short-distance travelling were available, long-distance travelling accounts for 45% of the mileage and nearly 50% of the GHG emissions of all person transport. Long-distance travelling is growing and is expected to continue to grow, particularly by air. The GHG emissions are expected to grow as well, though to a smaller extent. Because short-distance travelling is stagnating, the shares of long distance travelling in both mileage and GHG emissions are likely to increase.     

A study of feasibility and potential benefits of organised car sharing in Ireland

In this study, the market potential of car sharing has been evaluated using multiple alternative scenarios which examine the geographic, financial and environmental factors influencing car sharing adoption. The scenarios are applied to the available and collected travel information of the Irish population to estimate the potential impact of introducing car sharing in Ireland. The analysis identified that car owners who travel predominantly on alternative modes, could make significant cost and CO₂ savings through car sharing. A reduction of yearly CO₂ emissions of 86 kt is readily achievable through car sharing, with reductions up to 895 kt possible with appropriate policy and financial support. These figures are comparable to other measures proposed under the Irish National Climate Change Strategy.