An Investigation of Freeway Capacity Before and During Incidents Cuie Lu and Lily Elefteriadou Department of Civil and Coastal Engineering University of Florida March 4, 2011
Outline Database and Analysis Procedure Capacity for Normal (i.e., Non-incident) Conditions Capacity for Incident Conditions
l. Incidents on freeway may block one or more lanes and/or the adjacent shoulder lane: What is the capacity for incident conditions? What other factors affect capacity? Incident management, queue estimation Capacity-Normal
Federal Highway Administration (FHWA) : An incident blocking one lane out of three on a freeway reduces the capacity by about 50% Blocking two lanes of three reduces capacity by nearly 80% http://www.oti.dot.gov/tim/index.htm Capacity-Normal
The capacity remaining after an incident is estimated based on the number of lanes blocked by the incident. However, - doesn t consider other factors such as incident category and geometric characteristics - no models relating the capacity before an incident to the capacity after an incident Capacity-Normal
The objectives are to: Assess freeway capacity under normal (i.e., no incident) conditions Compare capacity remaining after incidents to previous research Develop models to predict incident capacity and capacity reduction Capacity-Normal
II. Database and Analysis procedure A subset of data collected for the NCHRP 3-87 project (Elefteriadou et al., 2009) Site Location Length (mi) Data Period Traffic Data * Weather Data Incident Duration Number of Lanes Closed/ Affected I-15 SB San Diego, CA 2.4 12/2006-11/2007 V, O Y Y N I-5 NB Sacramento, CA 10.4 11/2006-11/2007 V, O Y Y N QEW Toronto, Canada 6.5 01/2005-12/2005 V, O, S Y N Y I-494 SB Minneapolis, MN 3 09/2006-08/2007 V, O, S Y Y Y OR 217 SB Portland, OR 7 01/2006-07/2006 11/2007-12/2007 01/2008-12/2008 V, O, S Y Y Y * V- Volume, O- Occupancy, S-Speed
Data Analysis Procedure a. Data Screening and Incident Categorization Remove adverse weather and missing or erroneous observations Incident verification: by speed and occupancy time series plots Incidents categorization: - incidents occurring before congestion - incidents occurring during congestion - incidents occurring downstream
Data Analysis Procedure b. Breakdown Identification: speed drop of more than 10 mph, three criteria (Elefteriadou et al., 2009): S i Avg S i S i 1 0 Si 5,..., Si 1 Avg Si,..., Si 4 10mph Max S i,..., Si 9 Si 1 Defines recurrent congestion as demand-induced breakdown, congestion caused by incidents as incidentinduced breakdown
III. Capacity under Normal Conditions A. Breakdown flow; B. Maximum pre-breakdown flow C. Ave. flow for 10 minutes before breakdown; D. Average discharge flow Capacity- Normal
Capacity parameters by vs. number of lanes 2500 2000 breakdown flow 1500 1000 500 0 2 lanes (# 80 data points) 3 lanes (105 data points) 4 lanes (79 data points) 5 lanes (34 data points) max.flow 10 min before breakdown ave. flow 10 min before breakdown ave. discharge flow Capacity- Normal
As shown in the Figure: Three-lane freeways is more productive in terms of per lane capacity - might due to lane-changing behavior Breakdown flow is generally higher than the average discharge flow Breakdown flow has the largest range of values, while average discharge flow the smallest range of values. Capacity- Normal
IV. Capacity under Incident Conditions Data: three sites (Minneapolis, Portland, and Toronto), as these provide data on the number of lanes affected Factors: - Incident category (before or during congestion) - Incident location (at bottleneck or non-bottleneck) - total number of lanes - speed limit et al.
Incident capacity (two parameters): the average discharge flow per open lane when both incident and congestion are present (in red) the minimum 10- minute flow rate when both incident and congestion are present (in blue) Example: March 9, 2005 (Toronto)
Relationship between incident capacity and # lanes affected # Lanes open # Lanes affected # Data points Average flow per total lanes (veh/h/ln) Average flow per open lanes (veh/h/open lane) Minimum 10-min flow rate (veh/h/ln) 0 2 11 418 836* 237 shoulder +2 6 763 1289* 410 1 1 28 1004 2031 858 shoulder +1 2 798 1595 443 2 8 756 2267 541 2 Shoulder 14 1383 1383 1268 1 28 1390 1390 792 3 0 1 1388 1388 1180 Note: *assumes there is a passage for the vehicles on the shoulder or between the lanes, or that all lanes are closed only for a brief amount of time.
Capacity remaining after incidents (calculated in two ways): as the ratio of the minimum 10-min flow rate to the average discharge flow for normal conditions (averaged for each site) as the ratio of the minimum 10-min flow rate to the 10-min flow before breakdown for normal conditions (averaged for each site)
Compare capacity remaining after incidents to previous research: Author Number of lanes Lanes blocked shoulder 1 lane 2 lanes 3 lanes Goolsby, 1971 3 (27 data points) 0.67 0.50 0.21 0.00 2 0.81 0.35 0.00 N/A HCM 2000 3 0.83 0.49 0.17 0.00 4 0.85 0.58 0.25 0.13 5 0.87 0.65 0.40 0.20 Smith et al., 2003 3 (27 data points) N/A 0.37 0.23 N/A 2 0.75 0.32 0.00 N/A Chin et al., 2004 3 0.84 0.53 0.22 0.00 4 0.89 0.56 0.34 0.15* 5 0.93* 0.75 0.50 0.20* Lu and Elefteriadou, 2011 (average discharge flow) Lanes affected 2 (60 data points) 0.77 0.50 0.14 N/A 3 (30 data points) N/A 0.43 0.32 c N/A 2 (60 data points) 0.68 0.46 0.13 N/A Lu and Elefteriadou, 2011 (ave. flow 10-min before 3 (30 data points) breakdown) N/A 0.40 0.29 N/A
Reasons of differences: Normal capacity: - Previous research: the peak/maximum of the flow-density curve - This research: the ave. discharge flow or ave. flow 10-min before breakdown the number of lanes affected vs the number of lanes blocked by incidents
Estimate of the minimum 10-min flow rate: Parameter Description Estimate Error t Value Pr> t Incident category: 1-during congestion -165.8 67.9-2.44 0.0166 0-before congestion One lane affected 885.5 46.2 19.17 <.0001 Two lanes affected 408.5 69.6 5.87 <.0001 Shoulder affected 1291.2 79.0 16.35 <.0001 R-square 0.49 Root MSE 293.20
Estimates of total capacity reduction: uses average discharge flow as capacity for normal conditions and the minimum 10-min flow as capacity for incident conditions Parameter Description Estimate Error t Value Pr> t Incident category: 1-during congestion 0-before congestion 386.5 167.4 2.31 0.0234 Total number of lanes 1361.1 166.6 8.17 <.0001 One lane affected -1171.6 442.7-2.65 0.0097 Two lanes affected -76.7 442.6-0.17 0.8629 Shoulder affected -2008.3 387.7-5.18 <.0001 R-square 0.67 Root MSE 718.40
Total capacity reduction results in a better model: capacity.. reduction 386 inccate 1361 nlane 2008,..shoulder - affected -1172,..1- lane..affected - 77,..2- lanes..affected
V. Three-lane freeways seem to be the most efficient in terms of per lane capacity for normal conditions Two incident parameters: the minimum 10-min flow rate provides a better fit with the data reported Three factors significant in incident capacity analysis: - number of lanes affected - incident category - total number of lanes No linear relationship between incident capacity and number of lanes affected
Recommendations: Record incident data in more detail on the number of lanes closed More data on: shoulder plus one lane affected shoulder plus two lanes affected
My contact information: Cuie Lu clu614lce@ufl.edu Questions and Comments? Thank you!