GPS-Based Highway Performance Monitoring: Characterization of Travel Speeds on any Roadway Segment Alain L. Kornhauser Professor, Operations Research & Financial Engineering Director, Program in Transportation Founder, ALK Technologies, Inc. and Vice Chairman, New Jersey Commission on Science and Technology
Typical Recurring Congestion
Objective To readily Monitor Speed/TravelTime/Congestion/Delay Anywhere Using Vehicle Probe as the data Source
The Measurement Problem How to collect the Speed/TravelTime Data? Incremental Infrastructure In pavement loop detectors (single point) (single loop) radar/laser/video signpost systems (single point) EZ Pass readers (2 point span measurement, Excellent) Bluetooth Readers CrowdSourced Data Location services data: NYT article Wireless Location Technology (Cellular Probes, see Fontaine, et al) Cell-tower trilateration» Yet to demonstrate sufficient accuracy Cell-handoff processing» maybe OK for simple networks Floating Car (Vehicle Probe) data processing (see Demers et al) Week 8
Location Data (time sequenced) GPS Tracks (Location Bread crumbs, long sequences by unique traveling entity) ID (can be anonymous, but must be unique) Position (lat, lon) Date UTC (time) Instantaneous velocity (speed and heading) Other attributes Major source class: Qualcomm (QASPAR) @Road (GPS Commercial Tracking) CoPilot (GPS Consumer Crowd Sourced)
Position QASPR (Qualcomm Automated Satellite Position Reporting) Part of Qualcomm s OmniTACS No instantaneous velocity Data rate ~ typically every 45 minutes Positional accuracy ~ ¼ mile Standard of the long-haul trucking industry for more than 10 years; installed in over 300,000 trucks
@Road GPS Commercial Vehicle Data GPS-Quality Position resolution (~+15 m) Time Resolution ~ every minute Focused on medium to long-haul commercial operations Speed and heading not generally recorded Reconstruct Speed and Heading from adjacent poitions
GPS Commercial Vehicle Data, every 2 minutes # unique IDs Total GPS data points Total travel hours Total distance traveled (km) Average distance traveled (km/ ID) Average travel time (hr./ ID) 4,950 60,659,746 1,345,475 118,357,762 23,910 271.8
View of a couple of the 4,950 IDs
Distribution of data rate through corridor
CoPilot (GPS Consumer Crowd Sourced) Accuracy ~ 10 meters under most conditions Can be as frequent as every second Has instantaneous velocity (heading unreliable at low speeds)
FedEx Contract Carrier (for one week) breadcrumb every 3 seconds
Typical view of GPS Tracks
Scatter due to temporary obstructed line-of-sight
Map-matching {Link#, t, dir}
Urban Canyon Scatter
Observed Travel Time
Simple Approaches Managing the data
concept: Monuments concept: OneMon (Critical Position-Speed-Time Stamp)
concept: Monuments A readily identifiable location along a road segment. Could be anything Digital Map is an ensemble of links connected at end nodes. Chose: Mid-point of link i as monument i
ALK digital map database showing Monuments (blue squares) north of Toronto.
concept: Monuments A readily identifiable location along a road segment. Could be anything Digital Map is an ensemble of links connected at end nodes. Chose: Mid-point of link i as monument i concept: OneMon (Critical Position-Speed-Time Stamp) Given set of Position-Speed-Time Stamps ( GPS Track) for the k th vehicleid, map-matched to i th Link: The one nearest monument i is OneMon k,i Travel time is observed between monument i and monument j by differencing the data contained in OneMon k,i and OneMon k,j
Monument Link
OneMon
OneMon
OneMon
OneMon
OneMon
OneMon
Pair them up to get segment performance Monument2Monument
One of the segments: {851,850} M2M Pair
0 Speed 120 kph M2M Weekly Performance Days of Week Monday Tuesday Wednesday Thursday Friday Saturday Sunday 0 Cumulative Probability 1
M2M Weekly Performance
M2M Weekly Performance
M2M Weekly Performance
M2M Weekly Performance
Putting them together for the Windsor to Montréal Corridor http://orfe.princeton.edu/~alaink/trancanada/transportcanadafinalreport_v7.pdf
Doing it for all of North America Adding speeds to all 31x10 6 arcs of ALK s digital map of NA Forms the basis of MinETA Stochastic Route Optimization Beginning with Median of observed in closest neighborhood with sufficient data
Interactive Functionalities
Interactive Functionalities Overall Bandwidth display of Median Speed by ToD ampeak, midday, pmpeak, Overnight, Weekend, Nominal
AM Peak Median Speeds
Functionalities Overall Bandwidth display of Median Speed by ToD ampeak, midday, pmpeak, Overnight, Weekend, Nominal mouseover Speeds by ToD Display Distance v Time for all ToD on picked link
Functionalities Overall Bandwidth display of Median Speed by ToD ampeak, midday, pmpeak, Overnight, Weekend, Nominal mouseover Speeds by ToD Display Distance v Time (DvT) for all ToD on picked link Dragged-route specific Bandwidth display of Median Speed by ToD Dragged-route specific DvT display of Median Speed for all ToD
estimated be to parameters are C K and Where C C C K t f TT Time Weekday Travel i i i t e,,, 2 1 ), ( : ), ( ), ( ), ( ) ( 2 2 / 2 ) ( 2 3 3 3 2 2 2 1 1 1 Downtown Zoo Interchange
Thank You