IDENTIFYING HIGH-RISK ROADWAYS THROUGH JERK-CLUSTER ANALYSIS Seyedeh-Maryam Mousavi, Louisiana State University Scott Parr, California State University - Fullerton Anurag Pande, California Polytechic State University Brian Wolshon, Louisiana State University 41st Annual International Forum on Traffic Records and Highway Information SystemsCosta Mesa, CA - November 2015 1
28
Embed
IDENTIFYING HIGH-RISK ROADWAYS THROUGH JERK-CLUSTER ANALYSIS Seyedeh-Maryam Mousavi, Louisiana State University Scott Parr, California State University.
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
IDENTIFYING HIGH-RISK ROADWAYS THROUGH JERK-CLUSTER ANALYSIS
Seyedeh-Maryam Mousavi, Louisiana State University
Scott Parr, California State University - Fullerton
Anurag Pande, California Polytechic State University
Brian Wolshon, Louisiana State University
41st Annual International Forum on Traffic Records and Highway Information SystemsCosta Mesa, CA - November 2015
1
BACKGROUND
Research was initiated as part of a National Science Foundation study to examine stop-and-go, speed, and travel time parameters associated with the Two Fluid Model of traffic flow
GPS data collected to analyze those movements also showed evidence of “abrupt” and “abnormal” driving maneuvers
Based on these observations, it was suggested that there could be a correlation between the areas of “abrupt, atypical” movement and locations of crashes over time
In effect, the idea of this research was is that if we see a fatality in one of every 300 crashes and 1 crash occurs for every 10,000 abrupt maneuvers, we may be able to identify dangerous locations by looking first at locations with very high abrupt maneuvers These locations would be much easier to identify because abrupt maneuvers occur much more
frequently
This could suggest problem areas that could be corrected before crashes, injuries, and fatalities ever occur.
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
2
LITERATURE REVIEW
State of practice for identifying high-risk road segments: using long-term historic traffic crash data
However, crashes
Would it be more effective to develop and apply surrogate measures of safety to identify high-risk roadway segments?
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
1. Are a reactive measure: They can only be used after damage, injury, and loss of life have occurred
2. Might occur due to several reasons
3. May not be recorded uniformly
3
WHAT IS A SURROGATE MEASURE OF SAFETY?
An observable, non-crash event that has a relationship with crashes
Can be observed long before crashes
have occurred
Occur far more frequently than actual crashes; in some cases, 10 and 15 times more
frequently
enabling more powerful statistical analyses to be
applied
Advantages:
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
4
RESEARCH HYPOTHESIS AND OBJECTIVE
Hypothesis
Locating high concentrations of abnormal negative jerk values
(jerk-clusters: rate of change of acceleration_ ft./s³) would
enable high-risk locations to be identified in advance and
potentially with greater accuracy.
Objective:
Develop measures to proactively identify high-risk roadway
segments.Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
5
METHODOLOGY
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
6
1. Data Collection
GPS data collection and processing, road selection, ADT, curve location
2. Data Processing
Data errors, linear referencing, jerk analysis, crash analysis
3. Sensitivity Analysis
“Appropriate” jerk threshold and segment length
4. Crash Frequency Modeling and Log Likelihood ratio test
Crash intensity v. jerk-clusters v. horizontal curvature, model fitness
1. DATA COLLECTION :GPS DATA LOGGERS
GPS Data Loggers:•Rechargeable battery;
•Recorded data in a comma-separated format;
•Collected: latitude, longitude, altitude, heading, speed, number of satellites utilized, universal time code (UTC) and date, and etc.
•Placed in the center console or glove box of the tested vehicles.
Readings were recorded at a rate of three
hertz .
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
7
1. DATA COLLECTION :PARTICIPANTS
1) To ensure the consistency of travel over the same routes for developing jerk-clusters: a questionnaire based on personal characteristics and driving patterns was developed;
2) Participants: 31 staff members and their household at Louisiana State University;
3) Participants were asked to refrain from allowing anyone
else to use their vehicle;
4) To maintain confidentiality: a unique random identifier
was assigned to each participant;
5) Data collection period: spanned from July 2012 to January
2013- With each driver contributing around 10 days.
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
8
20-29 30-45 46-65 650.0
5.0
10.0
15.0
20.0
25.0
30.0
35.0
Male FemaleAge
Per
cent
age
Interrupted state highways in Baton Rouge, LA were selected in this
study, based on the data frequency:
1) A 7.5 mile long stretch of LA 42,
a four-lane divided highway with
posted speeds varying from 45 to 55 mph;
2) A 5.15 mile long stretch of LA 1248,
a four-lane divided highway with
posted speeds varying from 30 to 45
mph.
1. DATA COLLECTION :ROADWAY SELECTION
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
9
1) Average Daily Traffic (ADT) count data: accessed through
LADOTD’s estimated annual average daily traffic counts
2) Roadway geometric features: drawn from Google maps.
1. DATA COLLECTION :ADT AND ROADWAY CURVATURE
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
10
2. DATA PROCESSING:GPS DATA ERRORS
GPS “noise” at intersections GPS position wandering GPS reading gaps
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
11
2. DATA PROCESSING:GIS LINEAR REFERENCING
Linear referencing geographic locations (x, y) to a measured linear feature (LA 42 and LA 1248) within a radius of 300 ft. [14 and 16].
This was applied to the collected GPS data and crash data.
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
12
2. DATA PROCESSING:JERK ANALYSIS
Each participants’ acceleration and jerk between consecutive data points calculated from the
driving data by:
Where:
: Acceleration (ft/s²)
: Change in velocity between successive observations (ft/s)
: Jerk (ft/s³)
: Change in acceleration between successive observations (ft/s²)
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
13
2. DATA PROCESSING:SEGMENTING THE ROADWAYS
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
The roads were divided into three different segment lengths in ArcGIS:
1. Short: 1/8 mile segments
2. Medium: ¼ mile segments
3. Long: ½ mile segments
14
2. DATA PROCESSING:CRASH RATE ANALYSIS
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement
Crash rate was calculated for each segment based on [18]:
Where:
: Road segment crash rate expressed for 100 million vehicle-miles traveled
: Total number of crash on the roadway segment
: Traffic volume (ADT)
: Number of years of crash data
: Length of the road segment
15
3. SENSITIVITY ANALYSIS
1. 21 jerk thresholds were evaluated, started at -0.5 ft. /sec3 and
decreasing by 0.5 ft. /sec3 until reaching -10.5 ft. /sec3
2. Calculating “crash rate” and “jerk ratio” per each segment of 1/8, ¼,
and ½ mile.
3. Implementing a sensitivity analysis for LA42 and LA1248 separately.
Introduction and Objective Background Information Methodology Crash Modeling and LL Test Conclusion Acknowledgement