A Dynamic Bayesian Network Model for Predicting Congestion ...
A Smart Application for Predicting Network-wide Congestion ... · A Smart Application for...
Transcript of A Smart Application for Predicting Network-wide Congestion ... · A Smart Application for...
http://www.abdullahshabarek.com
A Smart Application for Predicting
Network-wide Congestion Hot Spots under
Adverse Weather Conditions
Presenter: Abdullah Shabarek
New Jersey Institute of Technology
NJDOT Annual Research Showcase
October 23rd, 2019
Agenda
Introduction1
Objective2
Methodology3
Application4
Conclusions5
2
INTRODUCTION
• Most of the studies are concerned about monitoring traffic conditions
under adverse weather conditions
• Previous studies predict traffic speed under normal conditions using
deep learning using:
- Deep Neural Networks
- Recursive Neural Networks
• There are some research papers that use deterministic models to
predict traffic speed under weather conditions, but they lack the
ability to consider various weather variables
3
Why to predict traffic congestion due
to adverse weather conditions?
• Predict congestion hot spots due to adverse weather conditions for
congestion mitigation plans
• Allocate larger resources to higher congestion hot spot segments at
certain times depending on the output of the application
4
BACKGROUND
• Adverse weather conditions that affect traffic speed can be categorized into:
- Rain Conditions
- Fog Conditions
- Snow Conditions
Weather Conditions Freeway Average Speed Reduction*
Light Rain/Snow 3% - 13%
Heavy Rain 3% - 16%
Heavy Snow 5% - 40%
Fog 10% - 12%
* FHWA. (2018, September 17). How Do Weather Events Impact Roads? Retrieved October 8, 2019, from https://ops.fhwa.dot.gov/weather/q1_roadimpact.htm.
5
BACKGROUND (CONT.)
Increases vehicles headways and decreases
traffic speed.
Fog Conditions
Can cause Capacity reduction (10% - 30%)
depending on the rain intensity.
Rain Conditions
Snow accumulation impedes the traffic reducing
the traffic speed.
Snow Conditions
Wind conditions can reduce drivers’ visibility
when combined with rain or snow conditions.
Wind Conditions
Weather Conditions
6
OBJECTIVE
• The objective of this study is to propose a smart application that predicts
freeway congestion hot spots due to adverse weather conditions.
• This study provides a system that covers the New Jersey freeway
network and can capture the effect of three different weather conditions:
- Rain Conditions
- Snow Conditions
- Fog Conditions
7
SYSTEM FRAMEWORK
Database DevelopmentBig data analysis is
conducted on traffic speed
and weather conditions data
Predictions of traffic speed under adverse weather conditions
Based on the output from traffic speed prediction under normal
conditions and weather data
Predictions of traffic speed under normal conditions
Based on the traffic speed from previous time stamps
Smart ApplicationBased on real-time feed from the
databases. The application show a
network-wide prediction of hot spot
congestions
Database
Development
Prediction of traffic
speed under normal
conditions
Prediction of traffic
speed under adverse
weather conditions
Smart
Application
8
Probe Vehicle
Data
Captures Traffic
Speed
New Jersey
Congestion
Management Systems
Estimates Traffic
Volume
New Jersey
Straight Line
DiagramRelates All the
Databases in terms
of Geographical
Locations
NOAA
Provides Weather
Information
DATABASE DEVELOPMENT
9
DISTRIBUTION OF THE ADVERSE WEATHER CONDITIONS IN TERMS OF (MILES-HOURS)
Rain 57%Fog 32%
Snow 11%
*The data is based on weather conditions from 2014 until 2019 in all New Jersey Freeway Network
10
RECURRENT NEURAL NETWORKS
11
𝑥1 𝑥2 𝑥3 𝑥𝑛
𝑦1 𝑦2 𝑦3 𝑦𝑚
Model Inputs
Hidden Layers
Model Outputs
PREDICTING TRAFFIC SPEED UNDER NORMAL CONDITIONS
Prediction of Traffic Speed
under Normal Conditions
for the next 24 hours
Traffic Speed of previous 2 days Recurrent Deep Learning model
through two hidden layers
12
Model Inputs Model Outputs
PREDICTING TRAFFIC SPEED UNDER ADVERSE WEATHER CONDITIONS
Volume/Capacity
VisibilityPredicted Traffic Speed
under Normal Conditions
Wind Speed
Traffic Speed
under Adverse
Weather
Conditions
Wind Direction
Rain/Snow Intensity 13
The Model uses recurrent neural networks
through four hidden layers to account for
queuing congestions over time
Weather Condition Type
APPLICATION
• Heavy Rain occurred in July 22nd , 2019
• The rain starts around 4:00 PM at some New Jersey areas and extends
until 10:00 PM
• The analysis is conducted July 22nd , 2019 at 1:00 PM (3 hours prior
to the prediction starting time)
• Interstate-78 (Eastbound direction) is selected for further illustrations
• Hot spot congestion is considered when traffic speed is below 25 mph
14
© 2019 by Abdullah Shabarek from: http://abdullahshabarek.com/res/NJ-DOT-2019.mp4
15
15
NORMAL CONDITIONS VS. PREDICTED CONDITIONS
0
10
20
30
40
50
60
16:00:00 17:00:00 18:00:00 19:00:00 20:00:00 21:00:00 22:00:00
Mil
es o
f C
onges
ted H
ot
Spots
Time
I-78
Garden State Parkway
I-95
I-80
I-287
0
10
20
30
40
50
60
16:00:00 17:00:00 18:00:00 19:00:00 20:00:00 21:00:00 22:00:00
Mil
es o
f C
onges
ted H
ot
Spots
Time
I-78
Garden State Parkway
I-95
I-80
I-287
16
RAIN INTENSITY AND PREDICTED NORMAL CONDITIONS(I-78 EASTBOUND)
5:30 7:00 8:30 10:00 Time
20
40
60
Milepost
0.00 0.15 >0.30
Precipitation Intensity (in/hour)
5:30 7:00 8:30 10:00
20
40
60
Milepost
0.00 30.0 >65.0
Traffic Speed (mph)
ACTUAL TRAFFIC SPEED VS. PREDICTED TRAFFIC SPEED(I-78 EASTBOUND)
Time
Milepost
5:30 7:00 8:30 10:00 Time
20
40
60
Milepost
0.00 30.0 >65.0
Traffic Speed (mph)18
5:30 7:00 8:30 10:00
20
40
60
MAPE VS. RMSE(I-78 EASTBOUND)
5:30 7:00 8:30 10:00 Time
20
40
60
Milepost
7.0 >14.0
RMSE (mph)
0.00
19
5:30 7:00 8:30 10:00
20
40
60
Milepost
100
MAPE (%)
0
CONCLUSIONS
• This model provides a smart application to predict hot spot
congestion on a network level due to adverse weather conditions
• Transportation agencies can use this application for congestion
mitigation plans when adverse weather conditions are forecasted.
• The application can be used to optimize the resources when assigned
to a network-wide locations depending on the predicted level of
congestion.
20
RESEARCH OPPORTUNITIES
Opportunity #1
Opportunity #2
Database Coverage to
Enhance the Application
Performance
Incorporate the Application
with New Jersey Best
Practices
Include Mobility as a Service
aspects to improve traveler
decision choice
Opportunity #3
21
Thank YouAbdullah Shabarek
New Jersey Institute of Technology
NJDOT Annual Research Showcase
October 23rd, 2019
http://www.abdullahshabarek.com
22
23