Innovations in Freight Data Tuesday, July 16, 2019 2:00-3:30 PM ET TRANSPORTATION RESEARCH BOARD
Innovations in Freight Data
Tuesday, July 16, 20192:00-3:30 PM ET
TRANSPORTATION RESEARCH BOARD
The Transportation Research Board has met the standards and
requirements of the Registered Continuing Education Providers Program.
Credit earned on completion of this program will be reported to RCEP. A
certificate of completion will be issued to participants that have registered
and attended the entire session. As such, it does not include content that
may be deemed or construed to be an approval or endorsement by RCEP.
Purpose
To highlight key presentations from the April workshop entitled Innovations in Freight Data.
Learning Objectives
At the end of this webinar, you will be able to:
• Discuss new data sources and processing, analysis, or visualization methods
• Identify advances in practice for freight planning and performance measurement
• Identify immediate and long-term research needs to further advance the capabilities of freight data users and to address persistent freight data gaps and challenges
Innovations in Freight Data Workshop:Webinar Introduction
Alison ConwayAssociate Professor of Civil Engineering, City College of New York
Chair, Workshop Planning Committee
July 16, 2019
2019 TRB Innovations in Freight Data WorkshopApril 9-10, 2019, Arnold and Mabel Beckman Conference Center, Irvine, CA
Objectives:
• Bring together diverse freight data users from across the modal spectrum
• Discuss the application of emerging data sources and methods to meet local, federal, or international freight planning and performance measurement requirements
• Share freight data tools that are adaptable, flexible, user-friendly, and preserve proprietary data, or that are developed with open source technology
• Share experience and best practices for addressing new challenges associated with the use of new and emerging data sources and methods
Pooled Fund Supporters• Iowa DOT
• California DOT
• Federal Highway Administration
• Florida DOT
• Pennsylvania DOT
• Texas DOT
• Utah DOT
• Washington DOT
• Wisconsin DOT
Workshop Planning CommitteeAlison Conway, City College of New York
Avital Barnea, AASHTO
Chandra Bondzie, FHWA
Holly Cohen, Florida DOT
Chester Ford, Bureau of Transportation Statistics
Sam Hiscocks, Iowa DOT
Sherif Ishak, Old Dominion University
Nikola Ivanov, U. of Maryland CATT Lab
Yatman Kwan, CalTrans
Catherine Lawson, University at Albany, SUNY
Donald Ludlow, CPCS
Casey Wells, Texas DOT
Joel Worrell, Florida DOT
Tom Palmerlee, TRB
Scott Brotemarkle, TRB
Mai Quynh Le, TRB
Attendees
Sector Share of Attendees Federal 13%Other 13%MPO/Local 2%Private 26%State 28%University 18%
• 127 attendees
• 38 Speakers/Presenters
Program• Opening Session
• Novel Methods of Data Collection
• Developing Freight Data into Decision-Making Information
• Developing Data Systems to Manage Commercial Vehicle Behaviors
• Emerging Data Streams and Future Freight Application
• Machine Learning and Computer Vision to Collect and Improve Freight Data
• Interactive Poster Session
• Poster Session
Online Resources
Live links to PDF presentations
• Interactive Program
• E-Circular coming soon
Today’s WebinarMajor findings and key takeaways from the workshop
Presentations:• Developing Freight Data into Decision-Making Information
Joel Worrell, Transportation Data Inventory Manager, Florida DOT
• Developing Data Systems to Manage Commercial Vehicle BehaviorsDonald Ludlow, Vice President, United States, CPCS
• Machine Learning and Computer Vision to Collect and Improve Freight DataSherif Ishak, Professor and Chair, Dept. of Civil Engineering, Old Dominion University
TRB Innovations in Freight Data Workshop WebinarJuly 16th, 2019
Joel WorrellFlorida Department of Transportation
Developing Freight Data into Decision-Making
Information
Developing Freight Data into Decision-Making Information
Innovations Background:State and local transportation agencies are applying common transportation data sources such as vehicle probes, traffic sensors, and weigh-in-motion systems in innovative ways to perform freight planning and performance measurement.
Key Questions:» How can state and local transportation agencies utilize freight data for
decision-making?» What type of data is needed to support national, regional, state or local
freight issues?» What steps can state and local transportation agencies take to invest and
maintain freight datasets to answer freight issues?
Developing Freight Data into Decision-Making Information
Key Presentations:
» Truck Empty Backhaul - A Florida Freight StoryJoel Worrell, Florida Department of Transportation
» Using Freight Data to Inform the 2018 Texas 100 Most Congested Road SectionsDavid Schrank, Bill Eisele, Texas A&M Transportation Institute
» Trucks and the Port of Virginia: Understanding Freight Patterns with Big DataRobert Case, Hampton Roads Transportation Planning Organization;
Catherine Manzo, StreetLight Data
Truck Empty Backhaul - A Florida Freight Story
Courtesy of FDOT – TRB Innovations in Freight Data Workshop – Final Program
Innovation: Utilizing data programs such as the weigh-in-motion sensor program can increase the effectiveness of organization’s freight planning and operational programs.
Contribution: A new methodology to develop a granular analysis of cubed-out and partially empty commercial vehicles based on overall axle weight distribution for each vehicle can support analysis for the truck empty backhaul phenomenon.
Truck Empty Backhaul - A Florida Freight Story
Courtesy of FDOT – TRB Innovations in Freight Data Workshop – Final Program
Truck Empty Backhaul - A Florida Freight Story
Courtesy of FDOT – TRB Innovations in Freight Data Workshop – Final Program
Truck Empty Backhaul - A Florida Freight Story
Courtesy of FDOT – TRB Innovations in Freight Data Workshop – Final Program
Using Freight Data to Inform the 2018 Texas 100 Most Congested Road Sections
Courtesy of TTI and TXDOT – TRB Innovations in Freight Data Workshop – Final Program
Innovation: Development of visualization tools for congestion and delay statistics support communication of patterns and illustrations to decision makers.
Contribution: Utilization of travel time data for all traffic and trucks only support freight data-decision making processes for various geographic reporting needs.
Using Freight Data to Inform the 2018 Texas 100 Most Congested Road Sections
Courtesy of TTI and TXDOT – TRB Innovations in Freight Data Workshop – Final Program
Using Freight Data to Inform the 2018 Texas 100 Most Congested Road Sections
Courtesy of TTI and TXDOT – TRB Innovations in Freight Data Workshop – Final Program
Trucks and the Port of Virginia: Understanding Freight Patterns with Big DataInnovation: Leveraging probe datasets support the analysis of commercial truck behaviors around ports, gateways, and other origins and destinations.
Contribution: The research and analysis of large probe datasets support the field of understanding big-data processing techniques and needs.
Courtesy of Hampton Roads TPO – TRB Innovations in Freight Data Workshop – Final Program
Trucks and the Port of Virginia: Understanding Freight Patterns with Big Data
Courtesy of Hampton Roads TPO – TRB Innovations in Freight Data Workshop – Final Program
Trucks and the Port of Virginia: Understanding Freight Patterns with Big Data
Courtesy of Hampton Roads TPO – TRB Innovations in Freight Data Workshop – Final Program
Developing Freight Data into Decision-Making Information
Summary:
» Dataset Resources – Freight analysis can be supported through state agency dataset analysis or data program improvements.
» Dataset Governance – Understanding how datasets are created will support an understanding of capabilities and opportunities for freight analysis applications.
» Scale of Data – The area of interest may determine the most suitable freight data to utilize for decision making.
» Data Processing – Identifying how large freight datasets are processed can identify the best fit scope of work required.
Next Steps:
» Continue Research for Multimodal Freight Issues
» Focusing on Resiliency
» Establishing Data Governance
» Developing Freight Data for Real-Time Information
» Advancing Data Processing
Thank You
Developing Data Systems to Manage Commercial Vehicle Behaviors Innovations in Freight DataTRB Webinar | July 16, 2019
Donald LudlowVice President, CPCS
Challenges• Congestion (urban)• “Last Mile” Access• “Last 50-Feet”
Access• Land Use• Truck Parking
2
What Commercial Vehicle Challenges are We Trying to Solve?
3
Key Questions
Basic Premise: Understanding truck activity patterns and supply chain behaviors is critical for planning for and managing goods movements and prioritizing infrastructure investments.
• What did we learn about data systems that can improve the ability to manage commercial vehicle behaviors?
• What are some of the specific innovations?• What are the next steps and remaining gaps?
• The challenge: GPS data on its own does not link to industries.
• The innovation: Advanced machine-learning and spatial analyses to classify trucks by industry based on activity patterns derived from anonymous GPS
• Accomplished by linking industry classification and land use data to truck GPS activity patterns
4
Discerning Truck Industry Activity Patterns with GPS and other Data
Akter and Hernandez, 2019
5
Making the Linkage (simplified)
Akter and Hernandez, 2019
• Represents an improvement over using modeling alone to estimate industry-level O-Ds, routes, and other truck activitydata
• Truck activity patterns by industry type using GPS data to improve modeling, forecasting, operations
• Future research could include different extraction methods, more ground truthing, and more integration of satellite imagery
6
Outcomes
Akter and Hernandez, 2019
• The Challenge: How to develop a forecasting tool that accounts for both short and longer term truck behaviorand to link long-hauland short-haul activity
• The Innovation: Advanced uses of truck GPS data for truck tour typologies, connecting long-haul and short-haul trucks, and longitudinal analysis. Uses multiple years ofdata.
7
The Challenge: Modeling Long and Short-Haul Truck Trips
Shabani, Smith, Bernadin, 2019
8
Washington, DC Regional Results
Heavy Duty Medium Duty Light Duty
Shabani, Smith, Bernadin, 2019
• Examined major depots (e.g. UPS and Costco DCs) to observe interchange between long and short-haul
• Linkages between long haul and short haul movements can be observed region-wide and modeled
• “Commercial” vehicle noise eliminated by examining tours (e.g. excluding superfluous ‘taxis’, ‘buses’)
• Gap in understanding detail about supply chains—the paths along which freight shipments move—and end-to-end trip performance.
• Not just commercial vehicle movements—but also ail, air, water with full supply-chains (end-to-end across modes) and component segments
• What is the innovation: fusing public and private data sources—including supply chain descriptions from private companies” to provide a national snapshot of 24 industries on major corridors.
• This approach takes real freight movement and fluidity data, and monitors it over time.
9
The Challenge: Understanding Freight Fluidity on a Supply Chain Level
10
Truck Routes in the National Platform24 National-Level Industry Sectors
• Makes supply chain performance visible on the infrastructure that public agencies manage (in the case of highways) using existing technology (GIS and Tableau)
Truck Network Change in PTI Q4 2018 to Q1 2019
I-95 Corridor Coalition, Parker and Bryan, 2019
• Truck probe data provides insight on truck behavior—where the trucks are parked, how long, and whether its ‘informal’ parking
• Gap in understanding variation in demand for truck parking spaces (how many spaces are open, where, and when)?
• What is the innovation: Crowdsourced truck parking updates provide hundreds of thousands of observations by drivers
11
The Challenge: Understanding Which Truck Parking Spaces are Full
12
Truck Parking: Crowdsourced Updates Enables ADOT to Make Rest Area and other Investment Decisions
• Data Fusion. Most “new” data sources must be “fused” with other sources—especially location-based data—to provide value.
• Connections. Making progress on establishing connections between vehicles, operations, and industries.
• Analytical Frameworks. The analytical procedures are as important as the data sources.
13
Summary: Key Themes in Commercial Vehicle Data Innovations
Using cost data to understand supply chains
Using machine learning and business data to link trucks to industries
Crowdsourcing truck space availability
Connecting short and long-haul trips through longitudinal analysis
• Improved industry/commodity connections• Success in tying trucks to tours and industries but not
specific commodities
• Difficult industry sectors (e.g. retail petroleum, farm-to
market, aggregates, drayage)
• Leveraging additional image and satellite data• Leveraging ELD data to improve connections
14
Future Steps: Additional Connections
15
Other Resources
NCFRP 49 Web Guide SHRP2 C20 Freight Data
Florida DOT Going the Last Mile NCFRP Project Library
Online Guidance and Case Studies
Thank You!
Donald Ludlow, MCP, AICPVice President, United [email protected] | +1 202 791 9055
Artificial Intelligence and Machine Learning in TransportationOverview of basic knowledge, challenges, and opportunities
Innovations in Freight Data WebinarJuly 2019
Sherif Ishak, PhD, PEOld Dominion University
Outline
• Overview of AI, ML, and DL
• ABJ70 Scope and Mission
• Tools, Applications, Technologies, and Data
• AI and Critical Issues in Transportation
Introduction
• AI is big right now with large investments made to tackle a variety of problems
• A few basic questions:
• What is AI?
• What is the public perception of AI?
• How different is it from Machine Learning and Deep Learning?
• What problems can these paradigms best tackle?
What is AI?
• AI vs Automation
• Automation: what we can do with computers
• AI: what we wish we could do to create intelligence by machines
• Perception of AI
• Extraordinary AI: Skynet, general reasoning with human-like intelligence or surpassing human capabilities
• Ordinary AI: specialized algorithms to answer specific questions in well defined domains (e.g. object recognition, pattern recognition, detection, prediction, etc.)
Big Data
• Massive amounts of data – thanks to technological and computational advances!
• Troves of data from smartphones and low-cost sensors
• Key to addressing resilience, sustainability, and safety issues of infrastructure
• Insight into how we move and how the transportation system works
• Challenging questions:
• How to tap into AI and ML to handle the sheer volume of information in the data we have and how to achieve that in a timely manner!
• How to capture subtle patterns and make short and long term predictions!
What is Machine Learning
(ML)?
• A way to create AI
• Computational methods to learn from data w/o predetermined equations
• When: rules are not clear or cannot be defined
• How: ML develops a set of rules on its own through the learning process – better learning with more data!
• Why now: we generate, store, and manage large amounts of data
• Challenge: which model to use? And how to strike the balance between simplicity (easy to train and work with) and sophistication (capture complicated relationships)
• Caution: big data does not necessarily mean better models!
Types of ML
• Supervised and Unsupervised
• Supervised: learn from input and output, predict future output
• Classification – prediction of discrete responses by classifying data into categories
• Regression – prediction of continuous responses
• Unsupervised: find hidden patterns and structures in input data w/o output
• Clustering – data exploration
• Which algorithm do we use? size and type of data, what insight we hope to gain – a trial and error approach!
Supervised Learning
• Binary vs. Multiclass Classification
• Classification Methods:
• Logistic Regression
• K Nearest Neighbor (KNN)
• Support Vector Machines (SVM)
• Neural Networks
• Discriminant Analysis
• Decision Trees
• Bagged and Boosted Decision Trees
• Regression Methods:
• Gaussian Process Regression
• Support Vector Machine
• Generalized Linear Model
• Regression Tree
Unsupervised Learning
• Cluster analysis for data exploration
• Partition of data into groups with similar characteristics
• Hard Clustering – a data point belongs to one cluster (K-means, hierarchical clustering, self-organizing maps)
• Soft Clustering – a data point may belong to more than one cluster with some probability (Fuzzy c-means, Gaussian Mixture Model
How to Improve Models
• Better accuracy and predictive power without overfitting
• Identification of the most relevant features forbetter prediction
• Transforming features to reduce dimensionality and simplify training
• Tuning hyperparameters to provide the bestmodel
What is Deep Learning (DL)?
• A subset of ML models with a network structure of multiple layers (DNN)
• Extract features and learn complexities (e.g. Google, Amazon)
• Fueled by Big Data and parallel computing for faster training
• Efficient in computer vision and object recognition applications
• Example: Convolutional Neural Networks (CNN)
ML Development
Process
• Access and load data
• Preprocess the data
• Derive features
• Train models
• Iterate to find best model
• Integrate trained model into application system
Questions and Challenges
• Questions:
• What kind of data do we have?
• What insights do we hope to gain?
• How and where will our model be applied?
• Challenges:
• Data – messy, incomplete, incompatible, redundant
• Data preprocessing – selection of features and information extraction, reducing dimensionality (PCA, factor analysis, etc.)
• Finding best model – overfitting vs overgeneralization!
ABJ70: TRB Committee on Artificial Intelligence and Advanced Computing Applications
TRB Committee Structure
7/11/2019 15
Policy & Organization
Data and Information Systems ABJ00
ABJ70: Artificial Intelligence andAdvanced Computing
ResearchEducation and outreachCommunication
ABJ10 National Transportation Data Requirements and ProgramsABJ15T Task Force for the Using Census Data for Transportation Applications ConferenceABJ20 Statewide Transportation Data and Information SystemsABJ25T Task Force on the Traffic Monitoring ConferencesABJ30 Urban Transportation Data and Information SystemsABJ35 Highway Traffic MonitoringABJ40 Travel Survey MethodsABJ45T Task Force on Understanding New Directions for the National Household Travel SurveyABJ50 Information Systems and TechnologyABJ60 Geographic Information Science and ApplicationsABJ70 Artificial Intelligence and Advanced Computing ApplicationsABJ80 Statistical MethodsABJ90 Freight Transportation DataABJ92T Task Force on Understanding Big Data in Freight TransportationABJ95 Visualization in Transportation
ABJ70 Mission and Scope
• Promote and advance the applications of AI and ML tools through strategic research, education, and outreach initiatives
• Bridge the Computer Science and Information Technology community with the transportation community to leverage more effective and efficient methods for solving challenging problems in all areas of transportation
• Focus research and education activities on the deployment of AI and ML tools to address critical issues and challenges at the intersection areas of emerging technologies and transportation applications
AITools
TransportationApplications
Technology & Data
ABJ70 Strategic
Goals
• Identify the critical transportation issues and challenges that are best tackled by AI and Machine Learning tools
• Take the lead on educating the transportation community about the critical role of AI and Machine Learning tools in Big Data Analytics
• Leverage collaboration with TRB committees and professional societies on research and education activities
•Crash Prediction
•Distracted Driving
•Collision Detection
•Traffic Accidents
Safety
•Imputation
•Noise Reduction
Data
•Behavior
•Characteristics
•Response Pattern
•Driving Performance
Driver
•Object Recognition
•Vehicle & Pedestrian Detection
•Decision Making Process
Automated Vehicles
•Delivery Scheduling
•Truck Parking
•Truck Classification
•Commodity Classification
•Other Computer Vision Apps
Freight
•Incident Detection/Duration
•Congestion Detection
•Signal Coordination
•Traffic Flow Estimation/Prediction
•Ramp Metering
•Travel Time Estimation
•Variable Speed Limit
Traffic Management
•Situational Awareness
Maritime
•Ridership Prediction
•Mode Classification
•Taxi Ridership
•Travel Time Estimation
Transit
•Route Choice
•Travel Demand Forecasting
Planning
•Ridesharing
•Ridesourcing
Shared Mobility
AI/ML Applications
Critical Issues in
Transportation (TRB 2019)
• Transformational tech and services
• Energy and sustainability
• Serving a growing and shifting populations
• Resilience and security
• Safety and public health
• Equity
• Governance
• System performance and management
• Funding and financing
• Goods movement
• Institutional and workforce capacity, and
• Research and innovation
Machine Learning and
Computer Vision to
Collect and Improve
Freight Data
• New Data Sources• Video Imaging for truck classification
• Integrated information and data fusion
• New Methods• ML and DL techniques for commodity code entries, transforming
video imaging into truck taxonomy using vehicle features (log, company name) to classify trucks and infer commodities
• New Challenges• Provide more details and more data for better CFS estimates
• Collect CFS data more frequently and reduce respondent burden
• Address data management and governance aspects of data from video imaging
• Expand overall understanding of the role of AI and ML in freight transportation planning and modeling
Thank You!
Today’s Speakers
• Alison Conway, City College of New York, [email protected]
• Joel Worrell, Florida Department of Transportation, [email protected]
• Donald Ludlow, CPCS, [email protected]
• Sherif Ishak, Old Dominion University, [email protected]
Get Involved with TRB• Getting involved is free!• Join a Standing Committee (http://bit.ly/2jYRrF6)• Become a Friend of a Committee
(http://bit.ly/TRBcommittees)– Networking opportunities– May provide a path to become a Standing Committee
member• Sponsoring Committee: ABJ90• Innovations in Freight Data Workshop website• For more information: www.mytrb.org
– Create your account– Update your profile
Receiving PDH credits
• Must register as an individual to receive credits (no group credits)
• Must attend entire webinar (including Q&A) to receive credits
• Credits will be reported two to three business days after the webinar
• You will be able to retrieve your certificate from RCEP within one week of the webinar