An Improved k-Nearest Neighbor Method for Travel Time Prediction Using Dedicated Short-Range Communications Probes on Signalized Arterials
Jinhwan Jang, Korea Institute of Civil Engineering and Building Technology (KICT)Show Abstract
Real-time travel time (TT) information has become an essential component of daily life in modern society. With reliable TT information, road users can increase their productivity by choosing less congested routes or adjusting their trip schedules. Drivers normally prefer departure time-based TT, but most agencies in Korea still provide arrival time-based TT with probe data from Dedicated Short-Range Communications (DSRC) scanners, due to a lack of robust prediction techniques. Recently, interest has focused on the conventional k-nearest neighbor (k-NN) method that uses the Euclidean distance for real-time TT prediction. However, conventional k-NN still shows some deficiencies under certain conditions. This article identifies the cases where conventional k-NN has shortcomings and proposes an improved k-NN method that employs a correlation coefficient as a measure of distance and applies a regression equation to compensate for the difference between current and historical TT. The superiority of the suggested method over conventional k-NN was verified using DSRC probe data gathered on a signalized suburban arterial in Korea, resulting in a decrease in TT prediction error of 3.7 percent points on average. Performance during transition periods where TTs are falling immediately after rising exhibited statistically significant differences by paired t-tests at a significance level of 0.05, yielding p-values of 0.03 and 0.003 for two-day data. The method presented in this study can enhance the accuracy of real-time TT information and consequently improve the productivity of road users.
Minimum Sampling Size of Floating Cars for Urban Link Travel Time Distribution Estimation
Meiping Yun, Tongji UniversityShow Abstract
Wenwen Qin, Tongji University
Despite the wide application of floating car data (FCD) in urban link travel time estimation, limited efforts have been made on determining the minimum sample size of floating cars considering the requirements for travel time distribution (TTD) estimation. This study develops a framework of seeking the required minimum number of travel time observations generated from FCD for urban link TTD estimation. The basic idea is that with decreasing the number of observations, how the similarities between the distribution of estimated travel time from observations and the ground-truth one varies, which are measured by employing the Hellinger distance (HD) and Kolmogorov-Smirnov (KS) test. Finally, the minimum sample size is determined by HD value, and corresponding distribution passing the KS test. The proposed method is validated with the sources of FCD and Radio Frequency Identification Data (RFID) collected from an urban arterial in Nanjing, China. The results indicate that: (1) the average travel times derived from FCD perform good estimation accuracy for real time application; (2) the minimum required sample size range changes with the extent of time-varying fluctuations in traffic flows; (3) the minimum sample size determination is sensitive to whether observations are aggregated nearby each peak in the multistate distribution; (4) sparse and incomplete observations from FCD in most time periods cannot be used in achieving the minimum sample size. Besides, it would produce a quite deviation from the ground-truth distributions. Finally, FCD is strongly recommended for better TTD estimation incorporating both historical trends and real-time observations.
Performance Metrics for Visualizing Interdependent Regional Traffic Congestion Using Aggregated Probe Vehicle Data
Thomas Brennan, College of New JerseyShow Abstract
Ryan A Gurriell, College of New Jersey
Andrew J Bechtel, College of New Jersey
Mohan Venigalla, George Mason University
Big data from probe vehicles is an important contributor for determining the regional performance of a transportation roadway network. Recent research has applied aggregated probe vehicle speed data to quantify the changes in travel time as a result of recurring and non-recurring congestion. Through the establishment of a base travel time for all roadway segments in a region, any increased travel time characteristics can be quantified temporally and spatially. This characterization is especially important when determining the regional resiliency of a major change to the roadway network. This paper demonstrates how aggregated probe speed data can be used to characterize and visualize the interdependencies of multi-regional congestion. To demonstrate the methodologies, an analysis of the I-276 Bridge closure in Burlington County, NJ near Philadelphia, PA was conducted. The bridge was closed after a routine inspected identified a crack in one of the structural members. In total, 90 days of data, which included 90-million speed records commercially collected for 1,765 roadway segments, was analyzed. A performance metric was developed to allow an impact analysis by comparing Burlington County to two adjacent counties, Mercer and Camden. The results show that the bridge closure did have a definitive, quantifiable impact on the adjacent counties. Subsequent analysis identified specific roadways that were most impacted by the closure. Although this research explores historic speed data, the methodologies presented can be applied to real-time speed data.
Development of a Traveler-Oriented Traffic Performance Metric in a Highly-Congested CBD Using Travel Time Data
Zeng Xu, New York UniversityShow Abstract
Elena Prassas, Associate Professor Department of Civil Engineering Indian Institute of Technology Guwahati
John Falcocchio, Professor Department of Civil and Urban Engineering NYU Tandon School of Engineering
In the highly congested street network of the Manhattan central business district (CBD), where average peak period traffic speed is in the range of 5 to 8 mph, a conventional arterial performance measure, such as average traffic speed, is an inadequate metric to describe drivers’ perception of changes in traffic conditions resulting from changes in traffic control strategies. This is because an improvement in average speed in such a highly-congested network is typically too small (usually 1 or 2 mph) to be perceptible to travelers. The objective of this research was to develop and evaluate a methodology that more appropriately represents driver experience in urban arterials, where high traffic density and frequent signal controls coexist. When the distribution of travel time shows multi-modality, conventional travel time estimations that use indexes of traffic performance from a unimodal distribution as estimators do not accurately reflect traffic conditions as perceived by motor vehicle users. This study proposes a stop-based performance measure of traffic flow that can be derived from the multi-modality analysis of travel time data. The stop-based approach of measuring traffic performance makes it possible to explicitly reflect the effect on stop frequency of traffic signal progression on road segments with densely signalized intersections, and to measure the benefits (e.g., reduction in stop frequency) of implementing advanced traffic control systems. The methodology produced a relationship between average speed and the proportion of drivers stopping at targeted frequency levels.
Individual Truck Speed Estimation from Advanced Single Inductive Loops
Yiqiao Li, University of California, IrvineShow Abstract
Yeow Chern Tok, University of California, Irvine
Stephen Ritchie, University of California, Irvine
Trucks are known to be an essential element in freight movements. They transport 73 percent of freight tonnage among all modes. However, they have been attributed with severe adverse impacts on roadway congestion, safety and air pollution. Truck speeds by truck body types have been considered as an indicator of traffic conditions and roadway emissions. Even though vehicle speed estimation has been researched for decades, there exists a gap in focus on truck speeds particularly at the individual vehicle level. A wide diversity of vehicle lengths associated with trucks makes it especially challenging to estimate truck speeds from conventional inductive loop detector data. This paper presents a new speed estimation model that uses detailed vehicle signature data from single inductive loop sensors equipped with advanced detectors to provide very accurate truck speed estimates. This model uses a new inductive signature features that shows a stronger correlation with truck speeds. A modified feature weighting K-means algorithm was used to cluster vehicle length related features to 16 specific groups. Then, individual vehicle speeds regression models were developed within each cluster. Finally, a multi-layer perceptron neural network model was used to assign single loop signatures to the pre-determined speed related clusters. The new model delivers a promising estimation results on both a truck-focused dataset and general traffic dataset.
A Statistical Approach for Estimating Speed Threshold for Traffic Breakdown Event Identification: A Model Accounting for Data Variations
Emmanuel Kidando, Florida State UniversityShow Abstract
Ren Moses, Florida State University
Thobias Sando, University of North Florida
This study aims at developing a robust Bayesian statistical approach to determine the speed threshold (ST) for detecting a traffic breakdown event using traffic flow parameters. Data collected from a freeway in Jacksonville, Florida was used as a case study segment. The approach particularly is based on the change-point regression, in which two models – the Student-t and Gaussian residual distributed regressions – were developed and compared. The study found promising results in detecting the ST value when verified using the hypothesis test and simulated data. Moreover, it was found that the Student-t regression outperformed the Gaussian residual distributed regression in fitting the speed-occupancy relationship. The methodology described in the current study can be used in the procedures of analyzing the breakdown process, stochastic roadway capacity analysis, congestion duration analysis, assessing recurring traffic conditions, and clustering different traffic conditions. The results from these analyses provide useful information required in developing advanced traffic management strategies for highway operations.
Travel Time Distribution Estimation for Urban Arterial Using Pair-Copula Construction
Wenwen Qin, Tongji UniversityShow Abstract
Xiaofeng JI, Kunming University of Science and Technology
Feiwen Liang, Guangxi University of Science and Technology
Urban arterial travel times are inherently uncertain due to the volatile traffic flows, signal controls, bus stops and roadside parking, etc. An effective way of characterizing such uncertainty is the estimation of travel time distribution (TTD), which provides some features of distributions to assist with looking at travel time variability. However, a common method both in literature and practice prefer to estimate average arterial travel times rather than TTDs, even when multistate travel times and link correlations may exist. In this context, a pair-copula construction approach is proposed to estimate the probability distribution of urban arterial travel times from link TTDs. First, link correlations are investigated and empirical evidence reveals that adjacent link travel times present multiple relationships. Then, in each pair-copula associated with two connected links, a copula mixture model is developed for capturing multiple relationships between link travel times. Finally, the arterial TTD is derived as the product of the pair-copulas and marginal probability distributions of link travel times. The proposed approach is validated with Radio Frequency Identification Data collected from an urban arterial including four links in Nanjing, China. The results indicate that the approach can dynamically capture the positively correlated, negatively correlated, and uncorrelated relationships between link travel times. The convolution model of individual link TTDs and recently developed bivariate copulas-based method are also used to benchmark the proposed approach, and comparison results confirm the effectiveness and accuracy of the approach, especially when multistate distributions and correlations are encountered.
Dynamic Prediction of Arterial Travel Time Mean and Variability
Zifeng Wu, Kittelson & Associates, Inc. (KAI)Show Abstract
Laurence Rilett, University of Nebraska, Lincoln
Weijun Ren, Chang'an University
Accurate travel time prediction is very important for real-time traveler information systems. Existing traveler information systems can provide mean travel times, but no indicators exist regarding the potential arrival time variability. Also, the mean corridor travel time is usually estimated as a direct summation of the travel times on the consisting links, which neglects the correlation between link travel times and may lead to inaccurate estimations. This paper improves the simple addition method and explores the potential of the nonlinear autoregressive with exogenous inputs (NARX) model to forecast corridor travel time mean and variability. The results verified that the proposed NARX model outperforms the other models in terms of predicting the mean corridor travel time and is a promising model to predict arrival time variability as well.
Evaluating the Reproducible Relationship Between Travel Times and Vehicle Accumulation Over Extended Freeway Segments: Detecting Onset of Traffic Conditions Leading to a Marked Increase in Travel Times
Yoonseok Oh, Korea UniversityShow Abstract
Koohong Chung, California Department of Transportation (CALTRANS)
Seungmo Kang, Korea University
Kitae Jang, The Cho Chun Shik Graduate School for Green Transportation Korea Advanced Institute of Science and Technology 291, Daehak-ro, Yuseong-gu, Daejeon, Rep. of KOREA 34141
The precursor states signaling the onset of traffic congestion were investigated in an effort to proactively detour traffic to detouring routes to reduce system-wide delay. The precursor states were identified by monitoring travel times (TT) and estimated vehicle accumulation (VA) over an extended freeway segments. The bivariate relationship between TT and VA displayed a slanted elliptical shape under recurrent congestion condition. Investigation of the evolution of traffic states together with the reproducible elliptical bivariate relationship revealed the following: (i) traffic states evolved in a counterclockwise direction; (ii) departures of the traffic states from the elliptical shapes were observed prior to increases in travel times; and (iii) the rate at which traffic states evolved along the elliptical circle could also be used to predict marked increases in segment travel times in advance. These findings show great promise for the proposed method to help government agencies in detecting the onset of traffic conditions leading to marked increases in travel times in advance and detouring traffic at opportune times.
Speed Estimation Using Smartphone Accelerometer Data
ILYAS USTUN, Wayne State UniversityShow Abstract
Mecit Cetin, Old Dominion University
This paper is focused on developing an algorithm to estimate vehicle speed from accelerometer data generated by an onboard smartphone. The kinetic theory tells that the integration of acceleration gives the speed of a vehicle. Thus, the integration of the acceleration values collected with the smartphone in the direction of motion would theoretically yield the speed. However, speed estimation by the integration of accelerometer data will not yield accurate results, since the accelerometer data in the direction of motion is not pure acceleration, but involves white noise, phone sensor bias, vibration, gravity component, and other effects. To account for these sources of noise and error, a calibration method that can adjust the speed at certain times or points is needed. The exact times when the vehicle stops and starts are identified and used to calibrate the estimated speed. Based on the collected sample data, the proposed method yields that the estimated speed is on average within 10 mph of the actual speed with a lower margin at the street-level driving. This suggests that with more information to calibrate the speed, the model accuracy can be improved further.
The Estimation of Queue Length, Probe Vehicle Penetration Rate, and Traffic Volume Using Probe Vehicle Trajectories
Yan Zhao, University of Michigan, Ann ArborShow Abstract
Jianfeng Zheng, Didi Chuxing LLC
Wai Wong, University of Michigan, Ann Arbor
Xingmin Wang, University of Michigan, Ann Arbor
Yuan Meng, Didi Chuxing LLC
Henry Liu, University of Michigan, Ann Arbor
With the development of connected vehicle technology and the emergence of e-hailing services, a huge amount of vehicle trajectory data are being collected everyday. Connected vehicles and vehicles with e-hailing services, like mobile sensors, can provide rich information of traffic conditions. The huge amount of trajectory data could provide a new perspective for the sensing, diagnosis, and optimization of transportation networks. There has been some literature estimating traffic volume and queue length at intersections using data collected from these probe vehicles. However, some of the existing models can only work when the penetration rate of probe vehicles is high enough. Some others require two critical inputs, the distribution of queue lengths and the penetration rate of probe vehicles, which might vary a lot both spatially and temporally and usually are not known in real life. To fill the gap, this paper proposes a novel method for the estimation of queue length, penetration rate, and traffic volume at signalized intersections. The proposed method is validated by both simulation data and real-field data. The testing results show that the method is ready for large-scale real-field applications.
An Entropy-Based Model to Evaluate Stochastic Travel Time Pattern Prediction Uncertainty
Xuechi Zhang, Research Institute of Highway, Ministry of TransportShow Abstract
Saini Yang, Beijing Normal University
Ali Haghani, University of Maryland, College Park
Zhentian Sun, Ministry of Transport Research Institute of Highway
This paper develops a general entropy-based model to evaluate parameter prediction uncertainty in stochastic system. Comparing against the existing uncertainty evaluation models developed in the field of traffic parameter prediction, the proposed model is not strictly dependent on the format of prediction model. Moreover, it can be utilized to evaluate the goodness of a particular system measurement method with the measurement error inexplicitly incorporated. With a newly defined traffic state depiction variable named as travel time pattern, we apply the proposed model to test the prediction uncertainty on corridor travel time patterns with spatiotemporal real-time measurements. Prediction uncertainties upon real-time temporal measurements and spatial measurements for real-world corridor segments are compared against the prediction uncertainties solely based on historical inference. Based on the empirical results, we conclude both real-time temporal and spatial travel time measurements can improve the short-term travel time pattern predictability, especially for the predictions during AM and PM peak hours. Moreover, comparing the overall uncertainty reductions induced by temporal measurement and spatial measurement, the temporal one is more effective to reduce the prediction uncertainty with respect to travel time pattern.
Processing SHRP2 Time Series Data to Facilitate Analysis of Relationships Between Speed and Roadway Characteristics
Jayson Stibbe, Texas A&M Transportation InstituteShow Abstract
Marcus Brewer, Texas A&M Transportation Institute
The Strategic Highway Research Program 2 Naturalistic Driving Study (SHRP2 NDS) dataset is a source of “big data” that provides researchers with opportunities to analyze more than just crashes and near crashes. In fact, the vast majority of the available data comes from trips where no such event takes place. By studying driving under normal conditions, it is possible to see what sort of effects the geometrics of a roadway can have on the operating speed of a vehicle. A current project tasked researchers with using the SHRP2 NDS dataset to evaluate how geometric characteristics affect the operating speed on freeway ramps. Through achieving this goal, the researchers encountered and solved unique problems that arose due to either the data-collection method or the sheer volume of data. One such problem was the task of locating a vehicle at a specific distance along a ramp, where a 50-foot error could be the difference between being in the middle of a sharp-radius curve and being stopped at an intersection. The researchers explored various ways of processing the time series data, and found that it was possible to locate vehicles using measurements obtained from aerial imagery.
Do Not Use Harmonic Mean for Probe Vehicle Average Speed Calculations: Methodology and Demonstrations with NGSIM Data
Yuandong Liu, University of Tennessee, KnoxvilleShow Abstract
Yang Zhang, Apple, Inc.
Lee Han, University of Tennessee, Knoxville
Traffic engineers are often interested in measuring speed along a stretch of roadway for a given period of time. Typically in the past, speed values are measured at a given location over some duration. After some initial confusions, traffic engineers correctly determined that harmonic means, instead of arithmetic means, should be used to calculate the average speed of the traffic stream. In the modern age of ubiquitous devices of mobile phones and GPS, vehicle speed data can be collected along a stretch of roadway frequently. The speed calculations, however, have not always been performed correctly with these data. Many users are under the impression that as long as individual speed data were aggregated using harmonic mean method, the result would be correct, or at least “close enough.” This, as is shown in this paper, is far from the truth. This paper examines calculation methods for the mean speed of probe vehicle data using different sampling strategies. It is demonstrated that average speed can be accurately obtained by taking the arithmetic mean of vehicle spot speeds if the sampling is done by time. Real-world vehicle trajectory data from the NGSIM database were used to verify and demonstrate that the traditional harmonic mean based calculation can be quite erroneous and the average speed should be computed using simple arithmetic means if time-based sampling strategy is used. Aggregating the vehicle spot speeds by taking harmonic means usually leads to an underestimate of the mean speed, compared to the arithmetic mean approach.
Predicting Link Travel Speed in Urban Road Networks Using Variational Mode Decomposition
Eui-Jin Kim, Seoul National UniversityShow Abstract
Ho-Chul Park, Seoul National University
Seung-Young Kho, Seoul National University
Dong-Kyu Kim, Seoul National University
Predicting travel speeds in the urban road networks is a challenging subject due to its uncertainty stemming from travel demand, geometric condition, traffic signals, and other exogenous factors. This uncertainty appears in the data as non-linearity, non-stationarity, and volatility, and it also creates a spatiotemporal heterogeneity of the link travel speed by combining with the correlation of the neighbor links. In this study, we propose a hybrid model using variational mode decomposition (VMD) to investigate and mitigate the uncertainty of urban travel speeds. VMD allows the travel speed data to be divided into orthogonal and oscillatory sub-signals, called modes. The regular components are extracted as the low-frequency modes, and the irregular components presenting uncertainty are transformed into a combination of modes that is more predictable than the original uncertainty. For the prediction, the VMD decompose the travel speeds data into modes, and they are predicted respectively and summed to represent the predicted travel speed. The evaluation results in the urban road networks show that a hybrid model outperformed the benchmark models in congested area and in general. The improvement in performance increases significantly over specific link-days which generally are hard to predict. To explain the significant variance of the prediction performance according to each link and each day, correlation analysis between the properties of modes and the performance of the model is conducted. Based on the results, discussion on the interpretation of the correlation analysis and future research are addressed.
Improved Traffic State Estimation by Bayesian Network Data Fusion of V2X and Vehicular Bluetooth Data
Kim Klüber, Deutsches Zentrum für Luft- und RaumfahrtShow Abstract
Marek Junghans, German Aerospace Center (DLR)
Konstantin Fackeldey, Technical University Berlin
Robert Kaul, German Aerospace Center (DLR)
A recently published probabilistic method for vehicle based traffic state estimation on the basis of fusing two wireless communication based technologies, i.e. Bluetooth and Vehicle-to-X communication data, is analyzed in three different scenarios ranging from “academic” to “realistic”. On the one hand there is accurate, but extremely rare V2X speed data and on the other, there is frequent, but inaccurate speed data based on a one unit Bluetooth reader approach. Therefore, this analysis takes into account specific traffic related variables, such as traffic flow and traffic density as well as the consideration of traffic light control (TLC), which affect the Bluetooth based detection results. The findings are used to improve the method. A Bayesian Network (BN) is developed that merges Bluetooth and V2X based speed detection results to provide an improved speed estimation. The novel BN is compared to the previous one for different V2X penetration ratios using the open source microscopic traffic simulation package SUMO (Simulation of UrbanMObility). The results show that the novel method can improve the vehicular speed estimation in the academic as well as in the realistic scenarios.
Gradient Boosting Regression Tree for Urban Link Travel Speed Prediction
Shuaichao Zhang, Zhejiang UniversityShow Abstract
Xiqun Chen, Zhejiang University
Short-term traffic flow prediction is the premise of intelligent traffic management and control, traffic status identification, and real-time traffic guidance. Due to stochastic perturbations, it is more difficult to predict urban link travel speed. In this paper, we employ the gradient boosting regression tree (GBRT) for urban link travel speed prediction considering temporal and spatial correlations of adjacent intersections. The core idea of the ensemble learning is that each new regression tree is established to reduce the residual error on the gradient direction of the previous regression trees, by adjusting the weights of basic learners to formulate a strong learner. Parameter tuning of GBRT is presented to show the influence of a variety of parameter combinations on the prediction performance. To verify the effectiveness of the method, license plate recognition data are collected from the Traffic Management Platform of Hangzhou, China, in November 2016. The first 2-week data are utilized as the training set, the following 1-week data are used as the validation set, and the remaining 1-week data are used as the test set. Several experiments are conducted and compared with the autoregressive integrated moving average (ARIMA) and random forests (RF) to test the performance of GBRT in short-term speed prediction. The results show the GBRT algorithm outperforms the other two approaches.
The Effect of Inherent Variation and Spatio-Temporal Dependency in Predicting Travel Speed in Urban Networks
Ho-Chul Park, Seoul National UniversityShow Abstract
Seungmo Kang, Korea University
Seung-Young Kho, Seoul National University
Dong-Kyu Kim, Seoul National University
Urban traffic prediction is a challenging task due to the complexity of urban network. Many studies have been conducted to improve the accuracy of the urban traffic prediction, but the limitation still remains that their accuracy varies with location and time. To overcome this limitation, it is necessary to investigate in depth the various phenomena that change the traffic flow patterns. Among the phenomena, this study aims to analyze the effect of inherent variation in a link and spatiotemporal dependency between links in predicting travel speed in urban networks and to identify the factors that influence the two phenomena. We present three measures to quantify them, i.e., coefficient of variation, forecastable component analysis, and cross-correlation function. The results show that the variation and dependency have significant differences according to locations. The results also indicate that the effects of the dependency of the upstream and the downstream are different each other and the effects of the two phenomena vary depending on the prediction horizon of the prediction model. In the longer prediction horizon, in particular, the effect of the variation is increased, but the effect of the dependency between adjacent links is decreased gradually. We also identify the factors that affect the two phenomena and recommend guidelines for urban traffic prediction.
Applying Gradient Boosted Regression Trees to Predict Multi-Route Bus Headway
Da Lei, Southeast UniversityShow Abstract
Siteng Wang, Southeast University
Ronggen Luo, Southeast University
Long Cheng, Ghent University
An Augmented Bayesian Tensor Factorization Model for Missing Traffic Speed Data Imputation
Yixian ChenShow Abstract
Zhaocheng He, Sun Yat-Sen University
Xinyu Chen, Sun Yat-Sen University
Yaxiong Han, Sun Yat-Sen University
In data-driven intelligent transportation systems, advanced sensor technologies have broadened our ways to collect a large quantity of urban traffic data. However, due to the sparsity of some kinds of data and the uncertainty of sensors in data collection, we frequently suffer from the problem of incomplete data. Therefore, it still remains a challenge on improving the accuracy of missing traffic data imputation. In this study, we propose an augmented Bayesian CP factorization (AugBCPF) model based on the standard Bayesian CP factorization one (BCPF) to estimate the missing traffic data accurately. With exception of the CP decomposition structure, which allows us to capture the interactions between different dimensions, we further add global parameter and bias terms to the mean parameter of the Gaussian assumption on tensor entries. Therefore, our model is capable of modeling the variation effect that only associated with every specific object itself. Afterwards, we present a fully Bayesian treatment of AugBCPF by placing conjugate priors over all model parameters and using Markov chain Mento Carlo (MCMC) methods to perform approximation inference. Empirically, relying on the urban traffic speed data set collected from Guangzhou, China, we evaluate the performance of data completion task with the proposed and competing methods by varying the missing rate from 10\% to 80\%. We find that the proposed model outperforms the state-of-the-art methods -- BCPF and SVD-combined Tensor Decomposition (STD). Moreover, the additive global parameter and bias terms of AugBCPF have profound implications in effect that can capture the real traffic pattern.
Modeling the Provisioning of Individual Travel Time for Advanced Traveler Information Systems
Qing Tang, Missouri University of Science and TechnologyShow Abstract
Xianbiao (XB) Hu, Missouri University of Science and Technology
The purpose of the Advanced Traveler Information System (ATIS) is to collect, process and provide travel time information to assist users’ travel from origin to destination. The heterogeneous driving behaviors from different travelers are not considered in most current systems such as Google Maps and 511 systems, which leads the systems to generate the same travel time for everyone who inputs the same origin and destination, although the actual travel times for each individual may turn out to be different. This paper explores the modeling of “individualized” travel time based on the individual behavior of each driver as opposed to “average” traffic information, with the ultimate goal of enabling individualized traffic information provision for the ATIS and subsequently reducing travel time prediction errors. A back propagation neural network (BPNN) model was built to quantitatively estimate the driving behavior differences (i.e., the “delta”) between individual drivers and the surrounding traffic. A travel time estimation algorithm is then proposed to derive link-level traffic information that considers individual behavioral difference. Finally, individualized route travel time is computed for each traveler based on the derived link-level traffic information and individual behavioral difference. The proposed model is implemented and tested on an open-source NGSIM dataset, which demonstrated the feasibility and effectiveness of the proposed model.
An Integrated Model for Transportation Networks and Travel Time Reliability
Xilei Zhao, Georgia Institute of Technology (Georgia Tech)Show Abstract
James C. Spall, Johns Hopkins University
Real-time navigation services, such as Google Maps and Waze, are widely used in daily life. These services provide rich data resources in real-time traffic conditions and travel time predictions; however, these resources have not been fully recognized and used in transportation modeling. This paper aims at taking advantage of the traffic data from Google Maps and applying cutting-edge technologies in maximum likelihood to model transportation networks and travel time reliability. This paper integrates travel time data for routes and traffic condition data for links to model the complexities of transportation networks. Based on our network model, we formulate the Fisher information matrix and use asymptotic normality to obtain the probability distribution of the travel time estimates for a random route within the network of interest. We propose a novel method to compute travel time reliability, which takes into account two levels of uncertainties, i.e., the uncertainty of the route's travel time and the uncertainty of its travel time estimates. The proposed method can provide a more realistic and robust travel time reliability estimate. The methodology is applied to a small network in the downtown Baltimore area, where we propose a link data collection strategy and provide empirical evidence to show data independence by following this strategy. We also present results for maximum likelihood estimates and travel time reliability measures for different routes within the network.
Automatic Detection of Major Freeway Congestion Events Using Wireless Traffic Sensor Data: A Machine Learning Approach
Sanaz Aliari, University of Maryland, College ParkShow Abstract
Kaveh Farokhi Sadabadi, University of Maryland, College Park
Monitoring the dynamics of traffic in major corridors can provide invaluable insight for traffic planning purposes. An important requirement for this monitoring is the availability of methods to automatically detect major traffic events and to annotate the abundance of travel data. This paper introduces a machine learning based approach for reliable detection of highway traffic congestion events from hundreds of hours of traffic speed data. Indeed, the proposed approach is a generic approach for detection of changes in any given time series, which is the wireless traffic sensor data in the present study. The speed data is initially time-windowed by a ten-hour long sliding window and fed into three Neural Networks that are used to detect the existence and duration of congestion events (slowdowns) in each window. The sliding window captures each slowdown event multiple times and results in increased confidence in congestion detection. The training and parameter tuning are performed on 17,483 hours of data that includes 168 slowdown events. This data is collected and labeled as part of the ongoing probe data validation studies at the Center for Advanced Transportation Technologies (CATT) at the University of Maryland. The Neural networks are carefully trained to reduce the chances of over-fitting to the training data. The experimental results show that this approach is able to successfully detect most of the congestion events, while significantly outperforming a heuristic rule-based approach. Moreover, the proposed approach is shown to be more accurate in estimation of the start-time and end-time of the congestion events.
Evaluating Quality and Penetration Rate Relation for Commercially Available Floating Car Data
oruc altintasi, Middle East Technical UniversityShow Abstract
Hediye Tuydes-Yaman, Middle East Technical University
kagan tuncay, Middle East Technical University
Urban traffic monitoring requires reliable traffic data which is traditionally obtained from various sources such as loop detectors, point sensors, video cameras etc. Due to its lower cost and high coverage area, Floating Car Data (FCD) is being increasingly used as an alternative traffic data source. However, its quality depends significantly on the penetration rate of the GPS-equipped vehicle providing individual track data processed anonymously for commercially available FCD. While it is possible to get the total number of registered vehicles feeding to FCD, spatio-temporal distribution of these vehicles in traffic is the real input affecting the quality FCD essentially, which must be evaluated. This paper presents a framework for estimating the current penetration rate of the FCD as well as investigating the impact of FCD penetration rate on its quality using Monte Carlo simulations. For a numerical case, commercially available FCD speed data for Ankara, Turkey was obtained from Be-mobile and compared with video-based ground truth (GT) data. FCD speed quality was measured with Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE) and correlation coefficient (R2). The results of the Monte Carlo simulations using GT speeds revealed that FCD penetration rate of 15% or more would increase quality significantly, while the current penetration rate in Turkey was estimated at between 1.02%-5.06%. Evaluation of FCD penetration rate for urban arterial Level of Service (LOS), a qualitative measure derived based on speed, showed that current FCD estimated the LOS with MAPE of 21.89%.
Application of Clustering and Association Rule Mining in Large-Scale Loop Detector Troubleshooting
Amin Ariannezhad, University of ArizonaShow Abstract
Yao-Jan Wu, University of Arizona
Traffic sensors are becoming increasingly common on roadways across the United States. The archived data from these sensors is used in a wide range of traffic management applications, including transportation planning, incident and congestion monitoring, performance measurement, and travel time analysis. However, missing or invalid data is becoming an important concern. This study proposes a systematic approach to identify and characterize data error patterns to facilitate large-scale loop detector troubleshooting. Data was collected from 102 dual loop detector stations located on freeways in Phoenix, Arizona for May 2016. A set of quality control criteria selected and applied on daily 20-second data to find the error percentage for each loop detector. A Fuzzy C-means clustering methods was implemented on the data quality check results to reveal six patterns in the data errors. Then, an association rule mining method was applied on data subsets found by the clustering method to discover the most frequent rules. Three loop detector stations with different error patterns were visited in the field to verify the clustering and association rule mining results, potential causes and recommend appropriate solutions. The analysis identified four key patterns, indicating that the proposed approach successfully found the relationships in the data errors. The findings of this study will enable agencies to focus on a relatively small number of patterns and help traffic engineers to more easily diagnose and troubleshoot large-scale loop detector errors.