The iTSS Lab will be presenting at the 98th TRB Annual Meeting from January 13-17, 2019 in Washington, DC.
Keep an eye out for our poster and lectern sessions. Our research will be featured in the following areas:
Poster Session: 1431, Current Issues in Transportation and Air Quality
Tuesday, Jan 15, 2019 8:00AM – 9:45AM
Topic: Traffic Emission Estimation Using High-Resolution Traffic and Signal Control Data: A Vehicle Trajectory-Based Approach
Authors: Anjie Liu; Liping Fu
Abstract
This paper presents a new approach for estimating real-time traffic emissions in a signalized traffic corridor using high-resolution detector and traffic control data. The research is motivated by the growing need to manage traffic congestion and vehicle emissions by taking advantage of the increasing availability of new big data sources. The proposed approach consists of a vehicle trajectory reconstruction algorithm using Newell’s car-following model and a set of empirical vehicle motion functions assuming that some traces of individual vehicles’ trajectories can be detected when traversing an intersection using a video camera or other tracking technologies, and that signal indication history of all lights are known. A simulation environment, coupling two microscopic traffic and emission simulators (VISSIM and MOVES), is used to evaluate the proposed approach. The VISSIM traffic simulator is used to generate detailed ground-truth trajectory data and the assumed available traffic data while the MOVES emission model generates the emission estimates based on the original trajectories from the traffic simulator as well as from the reconstruction algorithm. The emission estimates are compared to evaluate the quality of the trajectory reconstruction-based method. A case study based on a real traffic corridor is conducted to show the performance and limitations of the proposed method and identify the factors that influence the performance, such as traffic conditions.
Lectern Session: 1473, The Science of Snowfighting
Tuesday, Jan 15, 2019 10:15AM – 12:00PM
Topic: Field Evaluation of Different Pre-wetting Ratios for Sustainable Salting
Authors: Jaspreet Kaur; Taimur Usman; Liping Fu
Abstract
This research presents the findings from a field study aimed at comparing the performance of different pre-wet ratios of salt for their impacts on snow melting performance/friction of road surfaces. The research was motivated by the question whether or not better snow melting performance can be achieved by using higher pre wetting ratios. Field tests were conducted on three sections of a provincial highway, located in Western Ontario in winter season 2016-2017 under three pre-wet ratios, i.e., 5% (current practice), 10% and 20%. Based on a comprehensive statistical analysis of the field testing data, it was found that salt pre-wetted at 20% improved friction levels by approximately 12% while reducing the salt usage by 19% and sand by 35% when compared to those at a pre-wet ratio of 5%. Examination of images collected during snow storms showed that sections treated with salts with higher pre-wet ratios generally had lower snow coverage.
Poster Session: 1564, Highway-Rail Grade Crossing Research
Tuesday, Jan 15, 2019 1:30PM – 3:15PM
Topic: A Resource Optimization Model for Improving Railway-Highway Grade Crossing Safety in Canada
Authors: Lalita Thakali; Liping Fu; Jay Rieger; Shuangshuo Wang
Abstract
This paper presents a new approach to the problem of allocating federal resources and identifying upgrading projects for improving safety at grade crossings in Canada. The proposed approach is unique in two key aspects. First, a risk-based network screening process is adopted to identify the priority sites for providing a justifiable basis for distributing the total budget at a regional level as well as narrowing the search space in the subject optimization step. Secondly, a mathematical programming approach is applied to formalize the resource allocation process with an explicit consideration of the expected benefits – risk reduction and the costs of implementing the projects. This approach is expected to improve the process of identifying the optimal set of upgrading projects within each region, thus maximizing the return of investment. A full-scale case study using real data from the Canadian crossing network is conducted to demonstrate the applications of the proposed model and its potential in generating solutions that balance the need for achieving fair allocation of funds among the regions and retaining the optimality of the identified improvement projects.
Poster Session: 1565, Application of Machine Learning Methods for Operation and Maintenance of Transportation Systems (Part 1)
Tuesday, Jan 15, 2019 1:30PM – 3:15PM
Topic: Evaluation of Alternative Pre-trained Convolutional Neural Networks for Winter Road Surface Condition Monitoring
Authors: Guangyuan Pan; Liping Fu; Ruifan Yu; Matthew Muresan; Tae Kwon
Abstract
Real-time winter road surface condition (RSC) monitoring is of critical importance for both winter road maintenance operators and the travelling public. Accurate and timely RSC information during snow events can help maintenance operators to deliver better maintenance services, such as plowing and salting, for reduced costs and salt usage and improved level of service. With this information, the traveling public can make more informed decisions on whether or not to travel, where to go, and which highways to drive on. In our previous effort we have shown the potential of applying a pre-trained convolutional neural network (CNN) for automatically detecting winter road surface conditions based on images from fixed traffic/weather cameras or in-vehicle devices. This paper focuses on comparing the performance of four most successful CNN models available from the leaders of this technology, namely, VGG16 (Oxford University), ResNet50 (Microsoft), Inception-V3 (Google) and Xception (Google), for solving the RSC classification problem. The models were first customized with additional fully-connected layers of neurons for learning the specific features of the RSC images. The extended models were then trained with a low learning rate for fine-tuning by using a small set of RSC images. The models were tested using a hold-out set of images from cameras installed at different locations, showing highly encouraging results.
Poster Session: 1565, Application of Machine Learning Methods for Operation and Maintenance of Transportation Systems (Part 1)
Tuesday, Jan 15, 2019 1:30PM – 3:15PM
Topic: Deep Learning and Traffic Signal Control: A Comparative Analysis of Current State-of-the-Art Methods
Authors: Matthew Muresan; Guangyuan Pan; Liping Fu
Abstract
The past decade has seen significant advancement in the field of machine learning and artificial intelligence. In particular, recent advancements in the area of deep learning has resulted in new models and techniques that are able to solve problems with high accuracy that previously only humans were able to. One particular technique, deep reinforcement learning, has been applied to solve problems where complex strategy is required to achieve an objective. One prominent example of this is the recent development of AlphaGO, a game-playing AI that has bested the world’s best players of the ancient abstract board game of Go. In a previous work, we applied these techniques and extended them to develop an adaptive traffic signal control method. The approach used a Deep Q Learning approach in combination with a novel representation of the traffic system to search for an optimal traffic control policy. This paper extends these results by comparing the performance of some additional recently developed state-of-the-art deep reinforcement learning models, A3C and UNREAL, to our originally proposed structure, and to existing traditional control methods. Comparisons are made on the basis of the training time and the overall performance of each approach. The results can be used as a foundation for the development of the next generation of adaptive traffic algorithms, as presently, many current traffic control strategies rely on traditional techniques and methods.
Poster Session: 1639, Intercity Passenger Rail Research
Tuesday, Jan 15, 2019 3:45PM – 5:30PM
Topic: Modeling the Influence of Disturbances in a High-Speed Rail System
Authors: Ping Huang; Qiyuan Peng; Liping Fu; Chaozhe Jiang; Yuxiang Yang
Abstract
Accurately predicting the influence of disturbances in High-Speed Railways (HSR) has great significance for improving real-time train dispatching and operation management. In this paper, we established distribution models to estimate the inferences of high-speed train disturbances (HSTD), including the number of affected trains (NAT) and total delayed time (TDT). We extracted data about the disturbances and their affected train groups from historical train operation records of Wuhan-Guangzhou (W-G) HSR in China and used a K-Means clustering algorithm to classify them into four categories. Five common distribution models were applied to fit the distributions of NAT and TDT of each clustered category and the models with best goodness-of-fit were determined according to the Kolmogorov-Smirnov (K-S) test. The validation results show that the models accurately revealed the characteristics of HSTD and that these models can be used in real-time dispatch to predict the NAT and TDT once the type of delays and its basic characters are known.
Poster Session: 1655, Vehicle Trajectory Data-Enabled Traffic Signal Control
Tuesday, Jan 15, 2019 6:00PM – 7:30PM
Topic: Continuous Updating of Traffic Signal Timing Plans Using Bluetooth and Wifi Data
Authors: Matthew Muresan; Liping Fu; Ming Zhong
Abstract
This paper discusses a method to adjust traffic signal timings in real-time using travel time data obtained from Bluetooth and WiFi sensors. The proposed method attempts to leverage the opportunity of increasing availability of “Big Data” for improving the prediction of traffic characteristics. This type of data is widely available but has not seen use beyond traffic monitoring. A peudo-adaptive signal updating is proposed on the basis of an analytical approach from Highway Capacity Manual (HCM), enabling the best use of the available data while keeping the computational time low. A simulation platform (VISSIM) with scenarios covering a wide variety of penetration rates and traffic configurations is used for testing and development. Bluetooth and WiFi detections based on typical technical parameters are simulated in an external module developed to connect to the simulator. A sensitivity analysis is conducted to assess the effect of some key parameters such as the frequency of signal adjustment, the size of the adjustment, and the market penetration. Performance data is also extracted from multiple simulation runs to contrast the benefits of the proposed system compared to the traditional methods. The results of the sensitivity analysis show that the system can perform well in situations with penetration rates as low as 10% and adjustment intervals as short as 5 minutes. Travel time delays are also reduced when compared to semi-actuated control.
We look forward to seeing you there!
Leave a Reply
You must be logged in to post a comment.