Sponsored by William W. Hay Railroad Engineering Seminar Railroad Grade Crossing Accident Analysis at Microscopic and Macroscopic levels Ray Benekohal Professor University of Illinois at Urbana-Champaign Date: Friday, November 14, 2014 Time: Seminar Begins 12:15 Location: Newmark Lab, Yeh Center, Room 2311 University of Illinois at Urbana-Champaign Juan Medina Postdoctoral Research Associate University of Illinois at Urbana-Champaign
40
Embed
Railroad Grade Crossing Accident Analysis at Microscopic … W. Hay... · · 2018-02-27Railroad Grade Crossing Accident Analysis . at Microscopic and Macroscopic levels. Ray Benekohal.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Sponsored by
William W. Hay Railroad Engineering Seminar
Railroad Grade Crossing Accident Analysis at Microscopic and Macroscopic levels
Ray BenekohalProfessor
University of Illinois at Urbana-Champaign
Date: Friday, November 14, 2014 Time: Seminar Begins 12:15
Location: Newmark Lab, Yeh Center, Room 2311University of Illinois at Urbana-Champaign
Juan MedinaPostdoctoral Research Associate
University of Illinoisat Urbana-Champaign
RAILROAD GRADE CROSSING ACCIDENT ANALYSIS AT
MICROSCOPIC AND MACROSCOPIC LEVELS
Rahim F. BenekohalJuan C. Medina
Nov 14, 2014 – University of Illinois at Urbana-Champaign
William W. Hay Railroad Engineering Seminar
2
1 – Introduction2 – Research Questions3 – A Micro Approach4 – State of Practice – The U.S. DOT Model5 – A Combined Model7 – Accuracy of Predictions / Rankings8 – Future Research
Outline
3
Fatalities at Public Grade Crossings
Introduction
* Source: NCHRP 755
Grade crossing accidents are more likely to be more severe, more costly, and to involve a fatality than other highway crashes (NCHRP 755, 2013)
Fatal accidents have remained at ~10% of accidents for the last 25 years
4
Accidents at Grade Crossings (2003 – 2012)
Introduction
Nationwide Data Illinois Data
Crossings remain a significant hazard. Latest trends do not show accident decline (NCHRP 755)
5
Distribution of accidents in Illinois
Introduction
Similar general patterns, but difficult to predict specific locations due to low frequencies
2007 - 20082003 - 2004 2010 - 2011
6
Distribution of accidents at public crossings in Illinois (2003 – 2012)
A Combined Model• Angle not important for gated crossings; distance to nearest highway
intersection was significant
27
• Comparison is based on two factors:
1. Absolute and relative predictions
2. Ranking of high accident locations
Accuracy of Prediction
28
Accuracy of Prediction
• (Ʃ Predicted)/(Total observed)
• True prediction of all crossings together
• Useful to check magnitude of predictions
Overall absolute predictions:
0
0.5
1
1.5
2
2.5
3
3.5
4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Σ(Pr
edic
ted)
/Tot
al O
bser
ved
Cumulative Proportion of Crossings
Active Warning Devices
Field Data
US DOT Model
ZINB Model - Average Modeland Data
29
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Σ(Pr
edic
ted)
/Tot
al O
bser
ved
Cumulative Proportion of Crossings
Crossings with Gates
Field Data
US DOT Model
ZINB Model - Average Model and Data
Accuracy of Combined Macro Model
0
0.5
1
1.5
2
2.5
3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Σ(Pr
edic
ted)
/Tot
al O
bser
ved
Cumulative Proportion of Crossings
Passive Warning Devices
Field Data
US DOT Model
ZINB Model - Average Model and Data
30
• (Ʃ Predicted)/(Total Predicted)
• Useful to check if prediction curve is similar to data
Overall relative predictions:
0
0.2
0.4
0.6
0.8
1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Σ(
Pred
icte
d)/T
otal
Pre
dict
ed
Cumulative Proportion of Crossings
Active Warning Devices
Field DataZINB ModelUS DOT Model
Accuracy of Prediction31
0
0.2
0.4
0.6
0.8
1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Σ(Pr
edic
ted)
/Tot
al P
redi
cted
Cumulative Proportion of Crossings
Crossings with Gates
Field DataUS DOT ModelZINB Model
0
0.2
0.4
0.6
0.8
1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Σ(Pr
edic
ted)
/Tot
al P
redi
cted
Cumulative Proportion of Crossings
Passive Warning Devices
Field Data
ZINB Model
US DOT Model
Accuracy of Combined Macro Model
32
Ranking of high accident locations
Top 10 Top 20Data (observed) 11 21US DOT Formula 6 10
ZINB - Average model and data 8 14Data (observed) 16 26US DOT Formula 8 13
ZINB - Average model and data 11 19Data (observed) 28 48US DOT Formula 18 31
ZINB - Average model and data 20 35
Active (Flashing Lights)
Gates
Ranking MethodWarning Device
Passive (Crossbucks)
Number of crashes predicted in top
locations
33
Summary
• Improvements for state-of-practice in accident prediction are needed:• Accuracy of predictions, understanding contributing factors, finding countermeasures
• Micro approach finds contributing factors that data aggregation may mask•
• Combination of macro and micro analysis improved accident prediction
• Better accuracy• Better ranking
34
Future (and Ongoing) Research
• Additional accident model forms, selection of best option
• Dynamic tree structure to automatically sort attributes and discover trends• Cluster crossings based on potential trends (corridors)
• Corridor Analysis
• Approaches for combining macro and micro
… More
35
A Micro Approach – Extension to Corridors?
• Can we extend the micro approach to corridors?
• Accidents along corridor can be grouped to identity possible trends
• GIS to locate crossings, add data with socio-economic and geographic info
Future (and Ongoing) Research
36
Example Corridor
• Northeast Illinois Regional CommuterRailroad
25 accidents at 8 crossings between 2003 and 2012
Future (and Ongoing) Research
37
Combining Micro and Macro• One idea is to simply include variables from micro to macro models
• We could also add categories to the macro model (dummy variables, indicators, different models) and apply factors to modify predictions of some crossings
• Micro could also lead to different classification of crossings based on risk assessments
Future (and Ongoing) Research
38
Future (and Ongoing) Research
...Continued
• Temporal analysis on the occurrence of accidents
• Verification with other datasets:• So far used Illinois data. How to generalize it