Implementation of optimal parameter settings in GDOT's current ramp metering operation

GDOT Research Project 14-18
FINAL REPORT
Implementation of Optimal Parameter Settings in GDOT's Current Ramp Metering Operation
OFFICE OF PERFORMANCE-BASED MANAGEMENT AND RESEARCH
15 KENNEDY DRIVE FOREST PARK, GA 30297-2534

1. Report No.

2. Government Accession No. 3. Recipient's Catalog No.

FHWA-GA-19-1418

4. Title and Subtitle

5. Report Date

Implementation of Optimal Parameter Settings in GDOT's February, 2019

Current Ramp Metering Operation

6. Performing Organization Code

7. Author(s)

8. Performing Organization

Jorge A. Laval, Ph.D., Angshuman Guin, Ph.D., Bhargava Report No.

Chilukuri, Ph.D., and Hyun W. Cho

14-18

9. Performing Organization Name and Address

10. Work Unit No. (TRAIS)

School of Civil and Environmental Engineering Georgia Institute of Technology 790 Atlantic Dr.

11. Contract or Grant No. PI# 0012795

Atlanta, GA 30332-0355

12. Sponsoring Agency Name and Address

13. Type of Report and Period

Office of Performance-based Management and Research

Covered

15 Kennedy Drive

November 2014 February 2019

Forest Park, GA 30297-2534

14. Sponsoring Agency Code

15. Supplementary Notes

Prepared for the Georgia Department of Transportation

16. Abstract

The objective of this study is to optimize the parameter settings for the local ramp metering control

system currently in operation in the Metro Atlanta freeway network. The study corridor is

Eastbound/Southbound I-285 between GA 400 and I-20. Using a stochastic simulation-based

optimization framework that combines microsimulation model GTsim and a genetic algorithm-based

optimization module, we determine the optimal parameter values of the ramp-metering control system.

We generate optimal ramp metering values for 11 ramp-metering locations. The results of this project

are expected to improve ramp metering efficiency and mitigate freeway congestion.

17. Key Words

18. Distribution Statement

Ramp Metering, ALINEA, Genetic Algorithm

Optimization

19. Security Classif. (of 20. Security Classif. (of this 21. No. of pages

report)

page)

74

Unclassified

Unclassified

22. Price

Form DOT F 1700.7 (8-72) Reproduction of completed page authorized

i

Table of Contents
1 Introduction................................................................................................................. 1 2 Literature Review........................................................................................................ 3
2.1. Isolated Metering.................................................................................................. 3 2.2. Coordinated Metering .......................................................................................... 6 3. Optimization Framework ............................................................................................ 9 3.1. GTsim Application ............................................................................................. 10 3.2. Genetic Algotirhm.............................................................................................. 10 4. Methodology ............................................................................................................. 13 4.1. Study corridor..................................................................................................... 13 4.2. Traffic Data Analysis ......................................................................................... 15
4.2.1. Data ............................................................................................................. 15 4.2.2. Data Processing for Origin Destination Matrix Estimation ........................ 20 4.2.3. Calibration and Validation .......................................................................... 26 4.2.4. Critical Parameters...................................................................................... 29 4.3. Traffic Data Analysis ......................................................................................... 29 5. Results and Discussion ............................................................................................. 31
ii

5.1. Simulation Optimization Result ......................................................................... 31 5.2. Optimal Parameter Values ................................................................................. 34 5.3. MaxView implementation of Optimal Metering Rates...................................... 37
5.3.1. Peachtree Dunwoody Rd............................................................................. 38 5.3.2. Ashford Dunwoody Rd ............................................................................... 39 5.3.3. North Peachtree Rd ..................................................................................... 40 5.3.4. Peachtree Industrial Blvd ............................................................................ 41 5.3.5. Chamblee Tucker Rd .................................................................................. 42 5.3.6. Lavista Rd ................................................................................................... 43 5.3.7. Lawrenceville Hwy..................................................................................... 44 5.3.8. Church St. ................................................................................................... 45 5.3.9. Memorial Drive........................................................................................... 46 5.3.10. Covington Hwy ....................................................................................... 47 5.3.11. Glenwood Rd........................................................................................... 48 6. Conclusions............................................................................................................... 49 7. References................................................................................................................. 51 8. Appendix................................................................................................................... 55
iii

List of Figures

Figure

Page

1. Freeway-ramp configuration and parameters ............................................................... 4

2. Optimization Framework .............................................................................................. 9

3. The network of the study corridor (I-285 Eastbound/Southbound) shown in GTsim application....................................................................................................... 11

4. Study Corridor ............................................................................................................ 13

5. GDOT NaviGAtor video detection system (VDS) ..................................................... 17

6. Locations of GODT NaviGAtor's Vehicle Detection System (blue) and Varaible speed limit (red) on Google Map. ................................................................ 18

7. GDOT traffic tube counts ........................................................................................... 19

8. Flow chart of O/D matrix estimation .......................................................................... 20

9. Sample network for OD estimation ............................................................................ 21

10. O/D matrix .................................................................................................................. 25

11. (a) Time Space Speed map of NaviGAtor (upper row, field data) (b) Time Space Speed map of GTsim (lower row, simulation).. ............................................... 28

12. Speed contour map of the no control case. ................................................................. 32

13. Speed contour map of the RM only control case. . .................................................... 33

14. GA Convergence to global optimal over successive generations ............................... 34 iv

15. Typical Traffic State on I-285 Corridor on Monday PM Peak .................................. 55 16. Typical Traffic State on I-285 Corridor on Tuesday PM Peak................................... 56 17. Typical Traffic State on I-285 Corridor on Wednesday PM Peak.............................. 57 18. Typical Traffic State on I-285 Corridor on Thursday PM Peak ................................ 58 19. Typical Traffic State on I-285 Corridor on Friday PM Peak...................................... 59 20. April 2016 Monday Traffic......................................................................................... 60 21. April 2016 Tuesday Traffic.. ...................................................................................... 61 22. April 2016 Wednesday Traffic.. ................................................................................. 62 23. April 2016 Thursday Traffic.. ..................................................................................... 63 24. April 2016 Friday Traffic............................................................................................ 64 25. Flow-Occupancy (per lane) curve at D/S of Peachtree Dunwoody Road,
April 2016 ................................................................................................................... 65 26. Flow-Occupancy (per lane) curve at D/S of Ashford Dunwoody Road, April
2016............................................................................................................................. 66 27. Flow-Occupancy (per lane) curve at D/S of North Peachtree Road, April
2016............................................................................................................................. 66 28. Flow-Occupancy (per lane) curve at D/S of Peachtree Industrial Blvd, April
2016............................................................................................................................. 67
v

29. Flow-Occupancy (per lane) curve at D/S of Chamblee Tucker Road, April 2016............................................................................................................................. 68
30. Flow-Occupancy (per lane) curve at D/S of Lavista Road, April 2016...................... 69 31. Flow-Occupancy (per lane) curve at D/S of Lawrenceville Hwy, April 2016 ........... 70 32. Flow-Occupancy (per lane) curve at D/S of Church Street, April 2016..................... 71 33. Flow-Occupancy (per lane) curve at D/S of Memorial Drive, April 2016. ................ 72 34. Flow-Occupancy (per lane) curve at D/S of Covington Hwy, April 2016. ................ 73 35. Flow-Occupancy (per lane) curve at D/S of Glenwood Road, April 2016................. 74
vi

List of Tables

Table

Page

1. Sample O/D calculation table ..................................................................................... 22

2. Sample calculated and observed flow......................................................................... 23

3. Flow calculation.......................................................................................................... 24

4. Calibrated Parameters ................................................................................................. 26

5. Travel time (vehicle hours) comparison of no control, the RM control ..................... 31

6. Optimal Parameter Values .......................................................................................... 36

7. Peachtree Dunwoody Rd Ramp Metering .................................................................. 38

8. Ashford Dunwoody Rd Ramp Metering..................................................................... 39

9. North Peachtree Rd Ramp Metering........................................................................... 40

10. Peachtree Industrial Blvd Ramp Metering.................................................................. 41

11. Chamblee Tucker Rd Ramp Metering ........................................................................ 42

12. Lavista Rd Ramp Metering ......................................................................................... 43

13. Lawrenceville Hwy Ramp Metering........................................................................... 44

14. Church St. Ramp Metering ......................................................................................... 45

15. Memorial Drive Ramp Metering ................................................................................ 46

16. Covington Hwy Ramp Metering................................................................................. 47

17. Glenwood Rd Ramp Metering .................................................................................... 48 vii

EXECUTIVE SUMMARY The objective of this project is to optimize the parameter settings for the ramp metering control systems in operation on the EB/SB I-285 study corridor between GA-400 and US-78. This study was accomplished with the simulation-based optimization framework GTsim, modified to account for the existing control system configuration. The project developed parameter settings for the MaxView software that GDOT acquired for the freeway and arterial signal management.
We found that each local ramp metering system has different critical density values, so it needs to be optimized carefully. This study analyzed extensive traffic data to generate origin-destination (O-D) flows of the study corridor. We used tube counters (flow) data for on- and off- ramps and NaviGAtor (flow and speed) data for the mainline freeway to estimate travel time of each O-D.
This study also developed a Genetic Algorithm-based optimization method to generate optimal parameters of the RM system. We found that optimal RM reduces almost 5 % of total travel time compared to the current control method. These savings are not trivial considering that the upstream sections of the corridor (near Peachtree Industrial Blvd and North Peachtree Rd) gets completely congested by the end of the rush hour period. Since many ramps of the study corridor operate at minimum rate after a certain time, it is safe to say that most of the benefits of parameter optimization are realized during the congestion build-up phase.
From the optimal parameter and the critical density, we generated recommended metering rates for each location, which GDOT can readily implement in MaxView.
viii

1 Introduction
The main challenge of effective ramp metering is to develop optimal strategies that not only enhance the impact of the individual ramp meters in its vicinity but maximize system benefits. Researchers have developed several ramp metering algorithms for realtime management of ramp meters (Masher et al., 1975; M Papageorgiou, Hadj-Salem, & Blosseville, 1991; E Smaragdis, Papageorgiou, & Kosmatopoulos, 2004; Emmanouil Smaragdis & Papageorgiou, 2003), but there is little guidance on critical issues such as hours of operation, determining thresholds, and efficient methodology for on-ramp queue-flush management. This problem is very challenging and location specific that even the Ramp Management and Control Handbook (Leslie Jacobson, Stribiak, Nelson, & Sallman, 2006) only makes qualitative statements in this regard.
The research team completed Georgia Department of Transportation (GDOT) Research Project "Development of Optimal Ramp Metering Strategies" (Guin & Laval, 2013) in the context of the System-Wide Adaptive Ramp-Metering (SWARM), which was acquired by GDOT and has yet to be implemented. The project produced GTsim, a ramp-metering and a simulation-based optimization platform that combines the microscopic traffic flow model, which accurately predicts traffic dynamics under rampmetering (Laval & Leclercq, 2008) with the Genetic Algorithm optimization framework. This combination allowed us to successfully identify optimal parameters for the ramp metering system within the study corridor.
1

The optimization of existing control system is very important. To identify the optimal settings of the existing ramp metering algorithms being applied in the field today by the Traffic Management Center (TMC), this research project implemented and optimized the findings and tools developed in the "Development of Optimal Ramp Metering Strategies" study (Guin & Laval, 2013).
The objective of this study is to optimize the parameter settings for the local ramp metering control system currently in operation in the Metro Atlanta freeway network. This is accomplished with the simulation-based optimization framework GTsim, developed in the GDOT Research Project 07-22, modified to account for the existing control system configuration. The study corridor is the 19.25-mile, EB/SB I-285 corridor between GA 400 and I-20.
2

2 Literature Review
Ramp meters break up the vehicle platoons entering the freeway to enable smooth merging of the on-ramp flows with the freeway flows and prevent capacity drop. They also control excessive inflows into the freeway to avoid freeway queues from blocking the off-ramps. This is important because higher outflows mean lower delay. Towards this end, this section presents a brief overview of the isolated and coordinated ramp metering methods.
2.1. Isolated Metering
Isolated metering strategies are classified in 4 categories, based on the underlying method; linear programming (Iida Y., Hasegawa T., Asakura Y., 1989; Wattleworth, 1965; Yuan L.S. and Kreer, 1968), control theory (Masher et al., 1975; M Papageorgiou et al., 1991; E Smaragdis et al., 2004; Emmanouil Smaragdis & Papageorgiou, 2003), neural networks (Zhang & Ritchie, 1997), and fuzzy-logic (Bogenberger, Vukanovic, & Keller, 2002; Taylor, Meldrum, & Jacobson, 1998). However, methods based on linear programming and control theory are popular and field implemented extensively.
Fixed-time isolated control began with the seminal work of Wattleworth (Wattleworth, 1965) who derived optimal metering rates using steady-state vehicle conservation across ramps and linear programming, using historical data. Later, Yuan and Kreer (Yuan L.S. and Kreer, 1968) and others build on the Wattleworth's methodology to maximize the flow and balance the ramp queues. Papageorgiou (Markos Papageorgiou, 1980) improved Wattleworth's method by relaxing the steady-state condition. Masher et.
3

al. (Masher et al., 1975) developed the first traffic responsive control for isolated metering; demand-capacity and percentage occupancy methods as presented below:

Consider a typical freeway-ramp section and the traffic parameters as shown in Figure 1, where q(.) and r(.) represents the average flow and o(.) represents average occupancy. The demand-capacity strategy states that the metering rate at time k is:

()

=

{

-

(

- 1), ,

<

Where Q and Ocr are the desired flow and occupancy (which are typically set to capacity and critical occupancy) and qmin is the minimum metering rate.

The philosophy of the percentage-occupancy strategy is similar to the demandcapacity strategy and only defers in its implementation. Instead of measuring upstream flow, qin(.), the percentage-occupancy strategy uses upstream occupancy to estimate the flow. Moreover, the downstream occupancy is also measured at the upstream detector. Therefore, the percentage-occupancy strategy only needs one freeway detector (upstream of the ramp location) to determine metering rate.

Figure 1 Freeway-ramp configuration and parameters 4

Papapageorgiou et. al. (M. Papageorgiou, Hadj-Salem, & Middelham, 1997; M Papageorgiou et al., 1991) developed the popular feedback controller, ALINEA, that aims to maintain critical occupancy at the merge location. According to ALINEA:
() = ( - 1) + ( - ()) where is the regulating parameter, which is in the range of 60-70 vehicle/hr. ALINEA feedback law is simple, flexible, robust, and provides smooth transitions during congestion build up and dissipation. ALINEA was field implemented extensively and found to perform better than the fixed-time plans (M. Papageorgiou et al., 1997; Markos Papageorgiou, Kosmatopoulos, Papamichail, & Wang, 2008). To overcome implementation issues, several variations of ALINEA such as AD-ALINEA (Adaptive Strategy to Dynamically Calculate Critical Occupancy), AU-ALINEA (upstreammeasurement-based version of the AD-ALINEA), FL-ALINEA (flow-based ALINEA), UP-ALINEA (upstream-occupancy-based), UF-ALINEA (upstream-flow-based), and XALINEA/Q (combination of any of the preceding strategies with queue control) were developed (Emmanouil Smaragdis & Papageorgiou, 2003; Wang, Kosmatopoulos, Papageorgiou, & Papamichail, 2014).
5

2.2. Coordinated Metering
Coordinated metering methods can be broadly divided into three categories: Multivariable control, optimal control, and rule-based control.
Multivariable Control Papageorgiou (Diakaki & Papageorgiou, 1994; Markos Papageorgiou, Blosseville, & Haj-Salem, 1990; Markos Papageorgiou, Jean-marc, & Hadj-salem, 1989) derived a linear quadratic integral control for a system of ramp meters using linear quadratic optimization theory. METALINE, as it was called, is a vectorized extension of ALIEA as shown below:
() = ( - 1) + 1((k) - ( - 1)) + 2( - ()) Where () indicates a set of controlled on-ramps = [1, 2, ... ] , () indicates a set of state measurements = [1, 2, ... ], and () indicates a subset of state measurements for which target occupancies are available = [1, 2, ... ] . Finally, and are the calibrated gain matrices. While field applications in Paris (Markos Papageorgiou et al., 1990, 1989) and Amsterdam (Diakaki & Papageorgiou, 1994) proved that METALINE is simple and robust, its performance was found to be sensitive to the gain matrices used.
Rule-Based Metering
Rule-based algorithms are popular, and field implemented extensively. Some of the popular rule-based algorithms include Zone algorithm (Stephanedes, 1994), ALINEA
6

(M Papageorgiou et al., 1991), Minnesota Zone algorithm (Liu, Wu, & Michalopoulos, 2007), Linked-ramp metering algorithm (Banks, 1993), Denver Helper ramp algorithm (Lipp, Corcoran, & Hickman, 1991), Seattle Bottleneck algorithm (L Jacobson, Henry, & Mehyar, 1989), and SWARM (Paesani, Kerr, Perovich, & Khosravi, 1997).
The Zone algorithm (Stephanedes, 1994) divides the freeway into zones with their upstream boundary in free-flow and downstream boundary a bottleneck. The algorithm uses conservation to maintain each zone at desired level. Later improvements to the algorithm balance the freeway efficiency and ramp delay to maximize freeway flow.
In the Heuristic Ramp-Metering Coordination (HERO) algorithm, each ramp is outfitted with an ALINEA algorithm. When the queue on a ramp exceeds a threshold, it becomes a "master" control and controls some upstream ramps to reduce the queues at the Master ramp. The aim of this algorithm is to efficiently use all the space available on the network but does not optimize for the freeway-ramp system.
Similar to the Zone algorithm, System Wide Area Ramp-Metering (SWARM) algorithm divides the corridor into segments. The control operates on global and local level, with the former doing the forecasting and apportioning and the latter providing local responsive metering. The most restrictive of the two metering rates is used.
One of the drawbacks of these methods is that they all generally employ ad hoc feedforward control to achieve a target flow. As they are reactive, they resort to queue flush when the queue constraints are violated.
7

This page is intentionally left blank.
8

3. Optimization Framework
The simulation-based optimization framework for determining optimal parameter values is shown in Figure 2.
Figure 2 Optimization Framework Two main components of this framework are the Generic-Algorithm-based (GAbased) optimizer and GTsim module. The optimizer will provide a set of parameter values that are utilized by the GTsim module to estimate the total travel time that will be sent back to the optimizer. The GTsim application will provide continuous state information to the ramp metering algorithm that calculates metering rates based on the state information and the parameters provided by the GA based optimizer. The sections below describe GTsim and GA based optimizer as implemented in this study.
9

3.1. GTsim Application
GTsim, which is built based on a kinematic wave model, is the first one of its kind proven to replicate traffic dynamics during congestion. GTsim implements the latest lanechanging models, which significantly improved understanding of traffic congestion. Specific explanations on GTsim modules were introduced in the final report of the "Development of Optimal Ramp Metering Strategies" study (Guin & Laval, 2013).
In GTsim, we generated the 19.25-mile corridor as in the project's objective; see Figure 3.
3.2. Genetic Algotirhm
The objective of this project is to find optimal combination of parameters of ramp metering for the study corridor. As the solution space is large, simulation-based optimization, genetic algorithm play an important role in converging to the global optimum. Parameters of the genetic algorithm were described in the final report of the "Development of Optimal Ramp Metering Strategies" study (Guin & Laval, 2013).
10

Figure 3 The network of the study corridor (I-285 Eastbound/Southbound) shown in GTsim application. 11

This page is intentionally left blank.
12

4. Methodology
4.1. Study corridor
The study involved the selection of a 19.25-mile-long I-285 East Bound/South Bound corridor between GA-400 and I-20. This corridor contains thirteen ramp meter systems and 20 variable speed limits (see Figure 4). From typical traffic congestion characteristics from Google Maps and historical data of VDS (see Appendix), this study focuses on the onset period of evening peak congestion.
Figure 4 Study Corridor 13

The study corridor has the following seventeen entry locations (referred to as "origins" for the OD terminology) that feed traffic to the network:
Upstream Freeway, Peachtree Dunwoody Road, Ashford Dunwoody Road, North Peachtree Road, Peachtree Industrial Parkway, Buford Highway, the I-85 Connector, Chamblee Tucker Road, Lavista Road, Lawrenceville Highway, WB Stone Mountain Freeway (left lane merge), EB Stone Mountain Freeway, Church Street, Memorial Drive, Indian Creek Station Connector, Covington Highway, Glenwood Road.
The corridor has the following seventeen exit locations (referred to as "destinations"): Ashford Dunwoody Road, Chamblee Dunwoody Road, SB Peachtree Industrial Parkway, NB Peachtree Industrial Parkway, Buford Highway, the SB I-85 Connector, the NB I-85 Connector, Northlake Parkway, Lavista Road, Lawrenceville Highway, WB Stone Mountain Freeway, EB Stone Mountain Freeway, East Ponce de Leon Avenue, Memorial Drive, Covington Highway, Glenwood Road, Downstream Freeway.
14

4.2. Traffic Data Analysis
4.2.1. Data Within the 19.25-mile study corridor, this study used 52 GDOT NaviGAtor's Vehicle Detection System (VDS) (Figure 5, Figure 6) data that collected 20-second interval volume, speed, and occupancy (hereafter referred to as the "VDS data"). This study extracted the 52-stations VDS data during a one-month period (April 2016).
Five-minute volume data for 48 hours were measured using traffic tube counts (see Figure 7, GDOT traffic tube counts for specific locations). Tube counters could be installed in all on- and off- ramps in the corridor except for six locations: the NB GA-400 on-ramp, the SB Peachtree Industrial Blvd off-ramp, the Buford Hwy on-ramp, the I-85 connector on-ramp, the WB Stone Mountain Fwy connector on-ramp, I-20 off-ramp. Notice that most of these locations are freeway interchanges with no metering, and that this missing data was estimated as explained next).
As traffic volume data are the main input variables of this simulation case study and containing the volume data for all ramps is critical, this study focused on identifying the five-minute traffic volume of the missing locations for the same 48 hours by analyzing VDS data and the upstream and downstream ramps of the missing locations. For example, for the SB Peachtree Industrial Blvd off-ramp, we compared the VDS data of detector ID 2850034, 2850036, 2850037, which are the NB Peachtree Industrial Blvd off-ramp, and the Peachtree Industrial Blvd on-ramp, respectively. We assumed that the mainline volume on the corridor would be conserved by adding or subtracting the ramp
15

volume. However, a comparison of tube count data and the VDS data resulted in unreasonable ramp volumes (negative values) for the missing ramps.
The unrealistic ramp volumes could have resulted from the low-quality VDS data. For example, some detectors lost the data of one lane out of five or six lanes. We also tried to compensate for these missing lanes by multiplying the ratio of the missing lanes. However, we needed the lane distributions for each location to obtain the volume of correct whole lanes, which is beyond the scope of this study.
Because of these limitations, this study excluded the most upstream and downstream missing ramps, the GA-400 on-ramp, and the I-20 off-ramp. This exclusion, however, did not affect the system corridor because the GA-400 on-ramp does not contain a ramp-metering system, and the I-20 off-ramp does not affect congestion in the corridor.
16

Figure 5 GDOT NaviGAtor video detection system (VDS) 17

Figure 6 Locations of GODT NaviGAtor's Vehicle Detection System (blue) and Varaible speed limit (red) on Google Map. 18

Figure 7 GDOT traffic tube counts 19

4.2.2. Data Processing for Origin Destination Matrix Estimation
Rationally estimated origin-destination traffic volume matrix is essential in this simulation-based research. Figure 8 describes the steps of the O/D matrix estimation. We first extracted the traffic volume of the on-ramp (origin) and the off-ramp (destination) for the time periods of interest (PM peak) from the tube counts and the VDS volume data. We also calculated the travel time across each origin and destination using the spacemean speed that was converted from VDS speed data.

Tube counts and VDS data
Volume of origin and destination every 5 min.

VDS speed data
Travel time across all O/D

Time range of arrivals every 5 min. Volume
for the time range

Multiply ratio of O/D volume
Balancin g the total O/D volume sum

Non-linear optimization
Estimate d O/D matrix

Figure 8 Flow chart of O/D matrix estimation
Using these travel time data, we produced the possible time range of the arrival of the origin traffic. For example, for the I-285 downstream freeway destination, the earliest time of arrival would be the time that the first vehicle departed from the closest origin, Glenwood Road, and arrived at the destination from the beginning of the time period. Similarly, the latest time of arrival would be the time that the last vehicle departed from the farthest origin, I-285 Upstream freeway, and arrived at the destination from the end of the time periods. Using these possible time ranges of arrival, we calculated the arrival traffic volume for each destination.
20

The target time periods of our research are before the onset of the off-peak. From

the typical traffic data of Google Maps, we found that off-peak congestion on our

research corridor formed before 3:00 PM. Therefore, we chose 2:30 PM to 3:30 PM (60

minutes) as the time periods for this study.

The time periods of the origin traffic were set at 60 minutes. However, the

calculated possible time periods of destination traffic were longer than 60 minutes as they

were affected by congestion. To meet the total sum of the origin and destination traffic,

we adjusted the destination traffic volume by multiplying the ratio of the sum of the

origin volume to the sum of the possible destination volume.

After matching the sum of the origin and destination traffic volume, we estimated

the O/D matrix using a nonlinear optimization method. We explain the assumptions and

constraints of this optimization method using the simple network below (Figure 9).

1

3

5

2

4

Figure 9 Sample network for OD estimation This network consists of two origins (1, 2) and two destinations (4,5). From observed data (i.e., tube counts, VDS), the volumes of each origin and destination are generated (1-A, 2-B, 4-C, and 5-D). The volume of section 3 is calculated by adding the volumes of sections 1(A) and 2(B). To calculate the destination specific volumes, we generated the O/D matrix as Table 1. Constraints are that the sum of each row and
21

column volumes are close to the total estimated volume. For example, a volume that is generated from section 1(a) is composed of volumes heading to 4 () and 5( ). In this case, we set constraint + . Other rows and columns work in a similar manner. Subsequently, we generated a volume-calculation table, Table 2, and then we can calculate the O/D matrix using the optimization function. In the mathematical formulation, the objective function and constraints are described as follows.

. . ( - )2 + ( - )2 + ( + - - )2 + ( - )2 + ( - )2 Subject to + , + , + , +

O\ D 1 2
TARGET SUM

Table 1 Sample O/D calculation table

4

5

TARGET





A





B

C

D

c

d

SUM a b

22

Table 2 Sample calculated and observed flow

Calculated

Observed

Square of Differences

1

a

2

b

3

a+b

4

c

5

d

A B A+B C D

( - ) ( - ) ( + - - ) ( - ) ( - )

We used the computer program Generalized Reduced Gradient Algorithm (Lasdon, Fox, & Ratner, 1974), which is useful for solving the nonlinear optimization problem. The objective function of our problem is to minimize the total sum of the squared differences of the estimated volume (last column of Table 3), and the constraints are the sums of each cell of rows and columns (see Figure 10 O/D matrix). In Table 3, the green cells represent origin traffic, and the pink cells indicate destination traffic. In Figure 10, the gray cells must be zero because these destinations are upstream of the origins. With the algorithm, we found that the objective value decreased to a two-digit value.

23

Table 3 Flow calculation

Link 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67

Ramps

Calculated Flow Observed FlowDifference Diff. sqrd.

U/S Fwy

7914

7914

0

0

Peachtree Dunwoody

1030

1030

0

0

8944

8944

0

0

Ashford Dunwoody

1317

1317

0

0

7627

7627

0

0

Ashford Dunwoody

1279

1279

0

0

8906

8906

0

0

Chamblee Dunwoody

1099

1099

0

0

7807

7807

0

0

North Peachtree

978

978

0

0

8785

8785

0

0

SB P'tree Ind.

189

189

0

0

8596

8596

0

0

NB P'tree Ind.

1492

1492

0

0

7104

7104

0

0

P'tree Ind.

1963

1963

0

0

9067

9067

0

0

Buford Hwy

577

577

0

0

8490

8490

0

0

SB I-85

1440

1440

0

0

7050

7050

0

0

NB I-85

3181

3181

0

0

3869

3869

0

0

Buford Hwy

411

411

0

0

4280

4280

0

0

I-85

3941

3941

0

0

8221

8221

0

0

Chamblee Tucker

428

428

0

0

8649

8649

0

0

Northlake Pkwy

1055

1055

0

0

7594

7594

0

0

Lavista

996

996

0

0

6598

6598

0

0

Lavista

1063

1063

0

0

7661

7661

0

0

Lawrenceville Hwy

647

647

0

0

7014

7014

0

0

Lawrenceville Hwy

502

502

0

0

7516

7516

0

0

Stone Mt.

910

910

0

0

6606

6606

0

0

Stone Mt. EB

1276

1276

0

0

5330

5330

0

0

Stone Mt. Left merge

1017

1017

0

0

6347

6347

0

0

Stone Mt.

587

587

0

0

6934

6934

0

0

E Ponce De Leon

664

664

0

0

6270

6270

0

0

Church St.

382

382

0

0

6652

6652

0

0

Memorial Dr.

1008

1008

0

0

5644

5644

0

0

Memorial Dr.

891

891

0

0

6535

6535

0

0

Indian Creek

21

22

1

1

6556

6557

1

0

Covington

666

666

0

0

5891

5891

0

0

Covington

593

593

0

0

6484

6484

0

0

Glenwood

484

484

0

0

5999

6000

1

1

Glenwood

507

508

1

1

6506

6508

2

4

I-20

3203

3200

-3

9

D/S Fwy

3303

3300

-3

9

24

Origin/Destination
u/s Fwy P'tree Dunwoody Ashford Dunwoody
N. P'tree Rd P'tree Industrial
Buford Hwy I-85 C/D SYS
CHAMBLEE TUCKER RD
Lavista Rd Lawrenceville Hwy Stone Mt. FWY Left
Stone Mt. FWY Church St. Memorial Dr. Indian Creek Covington Glenwood

1 3 5 7 11 17 19
21
25 27 31 33 35 37 39 41 43 Target Sum

Ashford Chamblee SB P'tree NB P'tree Buford Dunwoody Dunwoody Industrial Industrial Hwy

4

6

8

10

12

1316 984

189

1258

1

1

68

0

136

7

47

0

67

36

0

30

22

512

SB I-85
14 1104 101 98 72 65

NB I-85
16 1196 523 786 654 22

Northlake pkwy
22 114 0 0 0 174 24 722

Lavista
24 99 2 1 1 152 15 714

LAWRENCE STONE VILLE HWY MOUNTAIN FWY

STONE MOUNTAI N FWY EB

EAST PONCE DE LEON AVE

Memorial

Covington

Glenwood

26

28

30

34

36

40

42

533

130

120

232

121

0

0

8

2

1

6

0

1

0

4

0

26

3

0

1

0

4

0

25

3

0

2

1

47

201

246

0

212

21

310

0

21

44

2

13

0

1

12

464

615

295

386

11

0

22

12

2

9

43

1

14

0

0

36

57

103

36

73

8

2

24

52

0

23

0

0

68

92

0

0

18

41

10

67

32

11

78

601

1

0

0

25

1317 1099

189 1492

577 1440 3181 1055

996

647

1317 1099

189 1492

577 1440 3181 1055

996

647

1317 1099

189 1492

577 1340 3081

955

896

547

910 1276

664 1008

666

484

910 1276

664 1008

666

484

810 1176

564

908

566

384

I20 d/s Fwy Target

44

46

298

219

7914

88

86

1030

109

99

1279

85

79

978

2

0

1963

159

132

411

270

453

3941

160

165

428

365

381

1063

222

180

502

361

496

1017

199

252

587

117

145

382

140

149

891

21

0

22

290

278

593

319

188

508

3200 3300 8

3203 3303

2929 3183

Sum
7914 1030 1279 978 1963 411 3941
428 1063 502 1017 587 382 891
21 593 507
0

Figure 10 O/D matrix

25

4.2.3. Calibration and Validation

GTsim has several parameters that must be calibrated (Chilukuri, Laval, & Guin, 2014). The parameters are categorized into capacity parameters (i.e., free-flow speed, jam density, and wave speed), lane-changing parameters (i.e., longitudinal distance between a vehicle and an exit ramp, tau (i.e., time to execute a lane-changing maneuver), epsilon (i.e., relaxation speed gap), and driver behavior parameters (friction speed). These calibrated parameters are summarized in Table 4. We used all parameter values in Table 4 for the entire corridor, except the value for the parameter of the longitudinal distance between a vehicle and an exit ramp. For some sections of the study corridor, the higher value of this parameter was needed to replicate feasible congestion propagation.

Table 4 Calibrated Parameters

Calibrated Parameter

Parameter Value

Free-flow speed Jam density Wave speed
Longitudinal distance between a vehicle and an exit ramp
Tau (time to execute a lane-changing maneuver)
Epsilon (relaxation speed gap) Friction speed

100 km/hr 150 veh/km 20 km/hr
2 (4) km
4 s
2 20 km/hr

26

We validated the model by comparing the speed contour maps of (a) NaviGAtor's VDS data (2016/04/12) and (b) GTsim (see Figure 11), which used the estimated O/D flow of the same day data. In Figure 11, the color legend shows the speed scale (unit: km/hr). Note that in the speed plot of NaviGAtor (Figure 11 (a)), vehicle speeds over 100 (km/hr) are capped at 100 (km/hr) to meet the free-flow speed of GTsim. We found that in the real-world corridor on the date (Figure 11 (a)), congestion formed around the 34mile post area at about 2:45 PM and around the 38-mile post area at about 3:30 PM. We confirmed similar patterns in the GTsim plots (Figure 11 (b)).
27

Figure 11 (a) Time Space Speed map of NaviGAtor (upper row, field data) (b) Time Space Speed map of GTsim (lower row, simulation). In the map, x-axis is the time, y-axis is the location. Blue represents high speed while brown represents low speed.
28

4.2.4. Critical Parameters In this project, we aim to produce optimal metering rates for each on-ramp meter system. GDOT implemented ALINEA algorithm, which are noted as below
() = ( - 1) + ( - ()). In this equation, is a critical occupancy (density) that is characterized by location. To obtain this value for each location, we plotted flow-density (q-k) curves using NaviGAtor data (in Appendix). As NaviGAtor data produce flow, speed, and occupancy for 20-second, we combined those data to 5-minute data. We also converted time-mean speed to space-mean speed. After that, we produced density by dividing flow to spacemean speed.
4.3. Traffic Data Analysis
The strategy development and evaluation is an iterative process. The number of allowable values of each of the eleven parameters evaluated in this study which corresponds to a total of 85,899,300,000,000,000,000 combinations of parameter values. Determining the optimal set of parameter values from this huge set is extremely time consuming even with any sophisticated search algorithm. Using the genetic algorithm, we selected optimal solution (combination of the eleven parameters). With the same method, the research team also investigated optimal queue flush parameters, which are maximum and minimum critical density of designated location.
29

This page is intentionally left blank.
30

5. Results and Discussion

5.1. Simulation Optimization Result

The results of the simulation-optimization for three cases (no control, the RM control only, the VSL-RM control) are summarized in Table 5. We found that the VSL-RM with optimized parameters outperforms the RM-control-only model with its optimized parameters in terms of reducing total travel time.

Table 5 Travel time (vehicle hours) comparison of no control, the RM control

Case

System

Ramp

Freeway

No control

6561

175

6386

RM control

6254 (4.7%)

194 (-10.9%)

6061 (5.1%)

Figure 12 show the speed contour maps of each control case. In the figure, we found that most congestion arises upstream of the 32-milepost. The VSL in this study corridor is located at the 34-milepost, and the target bottleneck location is downstream of the 35milepost, highlighted by the red oval line in the figures. Both controls reduce congestion in the target area. The main benefits of the controls are that they delay the bottleneck formation time and lessen the severity of the bottleneck (passing speed).

31

Figure 12 Speed contour map of the no control case. Red circles mean low speed areas to be compared with other figures.
32

Figure 13 Speed contour map of the RM only control case. Red circles mean low speed areas to be compared with other figures. The congestion is reduced compared
to Figure 12.
33

5.2. Optimal Parameter Values
Based on all possible values for parameters and queue flush density parameters, the solution space was still very large. Therefore, the optimal values of parameters were determined using a different set of GA parameters. A population of 30 and a mutation rate of 0.4 were used and the GA was run for 300 generations. The 15th percentile values of all the generations are plotted in Figure 14 that shows the convergence of GA over generations.
Figure 14: GA Convergence to global optimal over successive generations Figure 14 shows that after 208 generations, the algorithm converged to a minimum. To confirm that it is the global minima, tens of thousands combination of allowable
34

values of impact parameters were simulated. It is found that the minimum obtained from the GA after 200 generations was indeed the global minimum. Thus, it was confirmed that the GA parameters used above would converge to the global minimum.
After obtaining optimal parameters, we performed same GA test for queue flush density parameters. Table 4 describes these optimal parameter values.
35

Table 6 Optimal Parameter Values

Location



Critical

Density

Queue Flush Queue Flush Max. Density Min. Density

(veh/km/lane) (veh/km/lane) (veh/km/lane)

Peachtree Dunwoody 139

20

76

28

Rd

Ashford Dunwoody Rd 137

20

75

30

North Peachtree Rd 144

20

74

18

Peachtree Industrial 87

18

82

15

Blvd

Chamblee Tucker Rd 59

20

91

9

Lavista Rd

152

18

76

30

Lawrenceville Hwy

147

20

87

5

Church St

86

25

63

12

Memorial Dr

97

25

89

7

Covington Hwy

155

25

84

23

Glenwood Rd

94

23

66

20

36

5.3. MaxView implementation of Optimal Metering Rates
In this section we provide the necessary input information for GDOT to implement the optimal settings found in our study within the current version of MaxView. Notice that the current version of MaxView does not have the capabilities of implementing real-time control such as ALINEA, and therefore the results in the previous section cannot be directly implemented in the current system. However, in the new version of MaxView that is about to be implemented by GDOT this will be possible. In the meantime, the following tables are approximations of our results that can be implemented in the current system.
37

5.3.1. Peachtree Dunwoody Rd

=139, Critical Density = 20 (veh/km/lane), Maximum rate =1800 veh/hr, Minimum rate = 400 veh/hr

Table 7 Peachtree Dunwoody Rd Ramp Metering

Rate (veh/hr) 1800 1800 1800 1800 1800 1800 1800 1600 1450 1300 1150 1000 850 700 550 400

Density (veh/km/ln) 5 or less 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 or more

38

5.3.2. Ashford Dunwoody Rd

=137, Critical Density = 20 (veh/km/lane), Maximum rate =1800 veh/hr, Minimum rate = 400 veh/hr

Table 8 Ashford Dunwoody Rd Ramp Metering

Rate (veh/hr) 1800 1800 1800 1800 1800 1800 1800 1600 1450 1300 1150 1000 850 700 550 400

Density (veh/km/ln) 5 or less 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 or more

39

5.3.3. North Peachtree Rd

=144, Critical Density = 20 (veh/km/lane), Maximum rate =1800 veh/hr, Minimum rate = 400 veh/hr

Table 9 North Peachtree Rd Ramp Metering

Rate (veh/hr) 1800 1800 1800 1800 1800 1800 1800 1600 1450 1300 1150 1000 850 700 550 400

Density (veh/km/ln) 5 or less 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 or more

40

5.3.4. Peachtree Industrial Blvd

=87, Critical Density = 18 (veh/km/lane), Maximum rate =1700 veh/hr, Minimum rate = 400 veh/hr

Table 10 Peachtree Industrial Blvd Ramp Metering

Rate (veh/hr) 1700 1600 1500 1400 1300 1200 1100 1000 900 800 700 600 500 400

Density (veh/km/ln) 5 or less 6 7 8 9 10 11 12 13 14 15 16 17 18 or more

41

5.3.5. Chamblee Tucker Rd

=59, Critical Density = 20 (veh/km/lane), Maximum rate =1700 veh/hr, Minimum rate = 800 veh/hr

Table 11 Chamblee Tucker Rd Ramp Metering

Rate (veh/hr) 1700 1640 1580 1520 1460 1400 1340 1280 1220 1160 1100 1040 980 920 860 800

Density (veh/km/ln) 5 or less 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 or more

42

5.3.6. Lavista Rd

=152, Critical Density = 18 (veh/km/lane), Maximum rate =1800 veh/hr, Minimum rate = 400 veh/hr

Table 12 Lavista Rd Ramp Metering

Rate (veh/hr) 1800 1800 1800 1800 1800 1600 1450 1300 1150 1000 850 700 550 400

Density (veh/km/ln) 5 or less 6 7 8 9 10 11 12 13 14 15 16 17 18 or more

43

5.3.7. Lawrenceville Hwy

=147, Critical Density = 20 (veh/km/lane), Maximum rate =1800 veh/hr, Minimum rate = 400 veh/hr

Table 13 Lawrenceville Hwy Ramp Metering

Rate (veh/hr) 1800 1800 1800 1800 1800 1800 1800 1600 1450 1300 1150 1000 850 700 550 400

Density (veh/km/ln) 5 or less 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 or more

44

5.3.8. Church St.

=86, Critical Density = 25 (veh/km/lane), Maximum rate =1800 veh/hr, Minimum rate = 400 veh/hr

Table 14 Church St. Ramp Metering

Rate (veh/hr) 1800 1800 1700 1600 1500 1400 1300 1200 1100 1000 900 800 700 600 500 400

Density (veh/km/ln) 10 or less 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 or more

45

5.3.9. Memorial Drive

=97, Critical Density = 25 (veh/km/lane), Maximum rate =1800 veh/hr, Minimum rate = 400 veh/hr

Table 15 Memorial Drive Ramp Metering

Rate (veh/hr) 1800 1800 1700 1600 1500 1400 1300 1200 1100 1000 900 800 700 600 500 400

Density (veh/km/ln) 10 or less 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 or more

46

5.3.10. Covington Hwy

=155, Critical Density = 25 (veh/km/lane), Maximum rate =1800 veh/hr, Minimum rate = 400 veh/hr

Table 16 Covington Hwy Ramp Metering

Rate (veh/hr) 1800 1800 1800 1800 1800 1800 1750 1600 1450 1300 1150 1000 850 700 550 400

Density (veh/km/ln) 10 or less 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 or more

47

5.3.11. Glenwood Rd

=94, Critical Density = 23 (veh/km/lane), Maximum rate =1800 veh/hr, Minimum rate = 400 veh/hr

Table 17 Glenwood Rd Ramp Metering

Rate (veh/hr) 1800 1600 1500 1100 1300 1200 1100 1000 900 800 700 600 500 400

Density (veh/km/ln) 10 or less 11 12 13 14 15 16 17 18 19 20 21 22 23 or more

48

6. Conclusions
This study used a simulation based optimization framework to determine optimal parameter values of GDOT's MaxView ramp-metering system to minimize the total travel time in the study corridor. As a part of this framework, the microsimulation model (GTsim) and a GA based optimization module were integrated. The optimal values derived from this study were found to reduce travel times by more than 4.5% compared to no-metering scenario. These savings are not trivial considering that the upstream sections of corridor (near Peachtree Industrial Blvd and North Peachtree Rd) gets completely congested by the end of the congestion period. Considering that many ramps of the study corridor operate at minimum rate after certain time, it can be stated that most of the benefits of parameter optimization is realized during the congestion build-up phase.
From the optimal parameter values and the critical density downstream of the ramp metering that are derived in this study, we generated optimal metering rates for each location of the study corridor, which can be readily implemented in GDOT's MaxView ramp metering system. Notice that these metering rates are approximations of our results that can be implemented in the current system, which does not have the capabilities of implementing real-time control such as ALINEA. However, the new version of MaxView that is about to be implemented by GDOT, will have the ability to directly implement the results of our study.
49

This page is intentionally left blank.
50

7. References
Banks, J. H. (1993). Effect of response limitations on traffic-responsive ramp metering. Transportation Research Record, (1394).
Bogenberger, K., Vukanovic, S., & Keller, H. (2002). ACCEZZ--Adaptive Fuzzy Algorithms for Traffic Responsive and Coordinated Ramp Metering. In Journal of Transportation Engineering (Vol. 49, pp. 744753). ASCE.
Chilukuri, B. R., Laval, J. A., & Guin, A. (2014). MicroSimulation-Based Framework for Freeway Travel Time Forecasting. Transportation Research Record: Journal of the Transportation Research Board, 2470, 3445.
Diakaki, C., & Papageorgiou, M. (1994). Design and Simulation Test of Coordinated Ramp Metering Control (METALINE) for A10-West in Amsterdam. Internal Report No. 1994-2. ATT-Project Eurocor V2017.
Guin, A., & Laval, J. (2013). Development of Optimal Ramp Metering Strategies Final Report (GDOT Research Project TO 02-98; RSCH PROJ 11-18 Georgia Tech Project No. 2006S18).
Iida Y., Hasegawa T., Asakura Y., S. C. F. (1989). A formulation of on-ramp traffic control system with route guidance for urban expressway. In IFAC/IFFIP/IFORS/Sixth International Conference on Control in Transportation Systems (pp. 229236).
Jacobson, L., Henry, K., & Mehyar, O. (1989). Real-time metering algorithm for centralized control. Transportation Research Record, (1732), 2032.
Jacobson, L., Stribiak, J., Nelson, L., & Sallman, D. (2006). Ramp Management and 51

Control Handbook. Lasdon, L. S., Fox, R. L., & Ratner, M. W. (1974). Nonlinear optimization using the
generalized reduced gradient method. RAIRO Operations Research, 8, 73103. Lipp, L. E., Corcoran, L. J., & Hickman, G. A. (1991). Benefits of central computer
control for Denver ramp-metering system. Transportation Research Record, (1320). Liu, H. X., Wu, X., & Michalopoulos, P. G. (2007). Improving Queue Size Estimation
for Minnesota's Stratified Zone Metering Strategy. Transportation Research Record, 2012(1), 3846. https://doi.org/10.3141/2012-05 Masher, D. P., Ross, D. W., Wong, P. J., Tuan, P. L., Zeidler, H. M., & Peracek, S. (1975). Guidelines for design and operating of ramp control systems. SRI, Menid Park, CA, Stanford Res. Inst. Rep. NCHRP 3-22, SRI Project 3340. Menlo Park, California. Paesani, G., Kerr, J., Perovich, P., & Khosravi, F. E. (1997). System wide adaptive ramp metering (SWARM). Merging the Transportation and Communications Revolutions. Abstracts for ITS America Seventh Annual Meeting and Exposition. Papageorgiou, M. (1980). A new approach to time-of-day control based on a dynamic freeway traffic model. Transportation Research Part B: Methodological, 14(4), 349360. https://doi.org/10.1016/0191-2615(80)90015-6 Papageorgiou, M., Blosseville, J.-M., & Haj-Salem, H. (1990). Modelling and real-time control of traffic flow on the southern part of Boulevard Peripherique in Paris: Part II: Coordinated on-ramp metering. Transportation Research Part A: General, 24(5),
52

361370. https://doi.org/10.1016/0191-2607(90)90048-B Papageorgiou, M., Hadj-Salem, H., & Blosseville, J. M. (1991). Alinea. A local feedback
control law for on-ramp metering. Transportation Research Record: Journal of the Transportation Research Board, 1320, 5864. https://doi.org/10.1002/cncr.11316 Papageorgiou, M., Hadj-Salem, H., & Middelham, F. (1997). ALINEA Local Ramp Metering: Summary of Field Results. Transportation Research Record, 1603(1), 90 98. https://doi.org/10.3141/1603-12 Papageorgiou, M., Jean-marc, B., & Hadj-salem, H. (1989). Macroscopic Modelling of traffic flow on the Boulevard Peripherique in Paris. Trafnsportation Research PartB:Methodological, 236(ii), 2947. https://doi.org/10.1016/01912615(89)90021-0 Papageorgiou, M., Kosmatopoulos, E., Papamichail, I., & Wang, Y. (2008). A Misapplication of the Local Ramp Metering Strategy ALINEA. IEEE Transactions on Intelligent Transportation Systems, 9(2), 360364. Smaragdis, E., & Papageorgiou, M. (2003). Series of New Local Ramp Metering Strategies. Transportation Research Record: Journal of the Transportation Research Board, 1856, 7486. https://doi.org/10.3141/1856-08 Smaragdis, E., Papageorgiou, M., & Kosmatopoulos, E. (2004). A flow-maximizing adaptive local ramp metering strategy. Transportation Research Part B, 38(3), 251 270. Stephanedes, Y. J. (1994). Implementation of on-line Zone Control Strategies for optimal
53

ramp metering in the Minneapolis Ring Road. In Road Traffic Monitoring and Control, 1994., Seventh International Conference on (pp. 181184). Taylor, C., Meldrum, D., & Jacobson, L. (1998). Fuzzy ramp metering: design overview and simulation results. Transportation Research Record: Journal of the Transportation Research Board, 1634(1), 1018. Wang, Y., Kosmatopoulos, E. B., Papageorgiou, M., & Papamichail, I. (2014). Local Ramp Metering in the Presence of a Distant Downstream Bottleneck: Theoretical Analysis and Simulation Study. IEEE Transactions on Intelligent Transportation Systems, 15(5), 20242039. https://doi.org/10.1109/TITS.2014.2307884 Wattleworth, J. a. (1965). Peak-period analysis and control of a freeway system. Highway Research Record, (89), 125. Yuan L.S. and Kreer, J. B. (1968). An optimal control algorithm for ramp metering of urban freeways. In 6th Annual Allerton Conference on Circuit and System Theory. Zhang, H. M., & Ritchie, S. (1997). Freeway ramp metering using artificial neural networks. Transportation Research Part C: Emerging ..., 5(5), 273286.
54

8. Appendix
Figure 15 Typical Traffic State on I-285 Corridor on Monday PM Peak 55

Figure 16 Typical Traffic State on I-285 Corridor on Tuesday PM Peak 56

Figure 17 Typical Traffic State on I-285 Corridor on Wednesday PM Peak 57

Figure 18 Typical Traffic State on I-285 Corridor on Thursday PM Peak 58

Figure 19 Typical Traffic State on I-285 Corridor on Friday PM Peak 59

Figure 20 April 2016 Monday Traffic. Low-speed regions are in brown and highspeed regions are in yellow.
60

Figure 21 April 2016 Tuesday Traffic. Low-speed regions are in brown and highspeed regions are in yellow.
61

Figure 22 April 2016 Wednesday Traffic. Low-speed regions are in brown and high-speed regions are in yellow.
62

Figure 23 April 2016 Thursday Traffic. Low-speed regions are in brown and highspeed regions are in yellow.
63

Figure 24 April 2016 Friday Traffic. Low-speed regions are in brown and highspeed regions are in yellow.
64

Figure 25 Flow-Occupancy (per lane) curve at D/S of Peachtree Dunwoody Road, April 2016. Flow is in vehicle/hr and occupancy is defined as the percentage of time
that the sensor is occupied by a vehicle.
65

Figure 26 Flow-Occupancy (per lane) curve at D/S of Ashford Dunwoody Road, April 2016. Flow is in vehicle/hr and occupancy is defined as the percentage of time
that the sensor is occupied by a vehicle.
Figure 27 Flow-Occupancy (per lane) curve at D/S of North Peachtree Road, April 2016. Flow is in vehicle/hr and occupancy is defined as the percentage of time that
the sensor is occupied by a vehicle.
66

Figure 28 Flow-Occupancy (per lane) curve at D/S of Peachtree Industrial Blvd, April 2016. Flow is in vehicle/hr and occupancy is defined as the percentage of time
that the sensor is occupied by a vehicle.
67

Figure 29 Flow-Occupancy (per lane) curve at D/S of Chamblee Tucker Road, April 2016. Flow is in vehicle/hr and occupancy is defined as the percentage of time that
the sensor is occupied by a vehicle.
68

Figure 30 Flow-Occupancy (per lane) curve at D/S of Lavista Road, April 2016. Flow is in vehicle/hr and occupancy is defined as the percentage of time that the
sensor is occupied by a vehicle.
69

Figure 31 Flow-Occupancy (per lane) curve at D/S of Lawrenceville Hwy, April 2016. Flow is in vehicle/hr and occupancy is defined as the percentage of time that
the sensor is occupied by a vehicle.
70

Figure 32 Flow-Occupancy (per lane) curve at D/S of Church Street, April 2016. Flow is in vehicle/hr and occupancy is defined as the percentage of time that the
sensor is occupied by a vehicle.
71

Figure 33 Flow-Occupancy (per lane) curve at D/S of Memorial Drive, April 2016. Flow is in vehicle/hr and occupancy is defined as the percentage of time that the
sensor is occupied by a vehicle.
72

Figure 34 Flow-Occupancy (per lane) curve at D/S of Covington Hwy, April 2016. Flow is in vehicle/hr and occupancy is defined as the percentage of time that the
sensor is occupied by a vehicle.
73

Figure 35 Flow-Occupancy (per lane) curve at D/S of Glenwood Road, April 2016. Flow is in vehicle/hr and occupancy is defined as the percentage of time that the
sensor is occupied by a vehicle.
74