Development and Evaluation of Adaptive Ramp Metering Algorithms

1
Development and Evaluation of Adaptive Ramp
Metering Algorithms

• Team Members
• Taewan Kim, Xiaojian Nie, Wenlong Jin, Yingen Ge,
  Michael Zhang
• University of California at Davis
• Lianyu Chu, Will Recker
• University of California at Irvine

2
Managing congestion by ramp metering

• Spread demand over time by holding traffic at
  ramps temporarily, and by encouraging departure
  time changes
• Redistribute demand over space to reduce demand
  pressure on bottlenecks by encouraging diversions
• Break vehicle platoons to reduce disturbances at
  merging points

3
Research effort

• Identify promising algorithms
• Evaluate their effectiveness (using PARAMICS)
• Suggest improvements

4
Classification of algorithms

• Isolated or local control is applied to an
  on-ramp independently of any other on-ramps.
• Coordinated control is applied to a group of
  on-ramps with consideration of system-wide
  traffic conditions.
• Integrated various types of control measures
  (e.g., ramp metering, signal timing, route
  guidance) are applied in concert to a traffic
  system as a whole.

5
Algorithms examined
6
Conceptual evaluationcriteria

• Good system model and sound theoretical
  foundation Accurate system model,
• reasonable assumptions and objectives,
  rigorous problem formulation, efficient
• and accurate solution methods.
• Efficiency and robustness The control actions
  should be effective to achieve
• the control objective, and degrades
  gracefully when part of the system,
• such as input links, is down.
• Flexibility and expandability Easy to modify
  and expand to account for more
• complex and perhaps more realistic
  situations encountered in the freeway system.
• Simplicity Use the simplest logic structure
  possible to reconcile demands on realism
• and theoretical elegance.

7
Selected algorithms

• Local Minnesota Zone algorithm, ALINEA
• Coordinated Bottleneck algorithm, SWARM

8
Ramp control with Paramics
9
Zone Algorithm (not strictly local)

• Minneapolis/St. Paul area along I-35 East in
  1970
• Tries to balance the volume entering leaving
  the zone
• Each zone
• 3-6 miles long
• upstream boundary free-flow area
• downstream boundary bottleneck

10
Zone Algorithm-cont.
Where F the sum of metered
freeway-freeway ramp volumes S
the space available within the zone
Maximum volume
Metering rate
11
Alinea Algorithm

• a local traffic-responsive strategy based on
  classical control methods
• several successful field applications
• Boulevard Périphérique, Paris
• A10 West Motorway, Amsterdam
• Formula

r(k) r(k-1) KROc
Oout(k) where r(k) is the metering rate in
time step k r(k) is the metering rate in time
step k-1 (previous) KR is the regulator
parameter (constant) Oout(k) is the current
occupancy measurement.
12
Bottleneck Algorithm

• Implemented on I-5 by WDOT in 1981
• Six-year evaluation study shows
• travel time dropped from 22 minutes to 11.5
• accident rate dropped about 39
• Calculate both local and bottleneck metering
  rates
• Implement the most restrictive one

13
Bottleneck Algorithm-cont.

• Bottleneck is dynamically decided
• two criteria
• surpass a pre-determined occupancy threshold
• the zone is storing vehicles
• calculate zone volume reduction if a bottleneck
  appears
• ramp metering rate reduction
• zone volume reduction ? weighting factor

14
SWARM Algorithm

• SWARM consists of swarm1 and swarm2
• swarm2 is a traditional local traffic
  responsive algorithm
• swarm2 is replaced by ALINEA
• swarm1 is a forecasting global apportioning
  algorithm
• the more restrictive of the two being
  implemented

• Developed by NET for Caltrans
• Initial field tests in Orange County and LA

15
SWARM Algorithm-cont.
16
SWARM Algorithm-cont.
Predicted density current density trend
Tcrit
Target density (current density) (1/Tcrit)
(excess density)
Volume reduction (local density target
density) (
of lanes) (distance to next Station ) Ramp
reduction volume reduction weighting factor
17
Paramics Coding

• Network

• Already coded and simulated for ATMS study
• Real time detector data available
• Congestion in peak hours
• 6 mile

18
(No Transcript)
19

• Vehicle types and characteristics

20

• Zones and demands

(demand pattern I)
21

• Demand scenarios

Three demand levels are obtained by changing
demand pattern I according to the following
proportions
()

• Estimation of critical occupancy

Fundamental diagrams obtained by simulation
Critical occupancy 0.18
22
Simulation Results

• MOE
• 1) Travel Time for a given O/D pair
• 2) TVTT Total Vehicle Travel Time

D i,j travel demand of origin i and
destination j for the simulation time T i,j k
travel time of k th vehicle between origin i and
destination j NV i,j total number of vehicles
that actually traveled between origin i and
destination j
23

• Travel Time for some O/D pairs
• (Level 2 traffic demand, target
  occupancy0.13, KR20,000)

O/D 16 -gt2
O/D 13 -gt2
O/D 11 -gt2
O/D 9 -gt2
24

• TVTT for different target occupancy values
  (0.07, 0.1, 0.13)
• (ALINEA control, Level 2 traffic demand, KR
  20,000,
• simulated 10 times for each case changing
  seed value)

(veh.hr)
t-values are 11.48 (for 0.07 and 0.1) 6.47
(for 0.07 and 0.13) 2.59 (for 0.10 and 0.13)
means are different with significance level 5
25

• TVTT for different regulator gain values
  (10000, 20000, 30000)
• (ALINEA control, Level 2 traffic demand,
  target occ. 0.13,
• simulated 10 times for each case changing
  seed value)

(veh.hr)
t-values are 1.61 (for 10000 and 20000)
0.43 (for 10000 and 30000) 0.77 (for 20000 and
30000)
means are not different with significance level 5
26

• TVTT for different control algorithms
• (target occ.0.13, KR20,000, TVTT shown is
  the average value of ten
• simulation runs, numbers in parentheses are
  percentiles for no control case)

(veh.hr,)
T-test shows that the TVTT of 3 ramp control
algorithms are less than that of no control. But
statistically there is no performance difference
among the 3 ramp control algorithms.
27

• Comparison of different demand pattern

(demand pattern II)
Demand on the mainline is the same as pattern I.
But flows at exits and entrances are different
28
lt Demand pattern I gt
(veh.hr,)
lt Demand pattern II gt
(veh.hr,)
29
Summary

• Metering reduces the total travel time up to
  7.
• No significant performance differences are
  found among ALINEA, modified BOTTLENECK, and
  ZONE algorithms under the tested scenarios.
• A remark about SWARM we found that the key to
  SWARMs performance is good traffic prediction.
  SWARM performs as good as other tested algorithms
  when traffic prediction error is small (e.g.,
  one-step prediction), but worse than other
  algorithms when prediction error is large (e.g.,
  multi-step prediction). These results are to be
  further investigated.

30
Summary-cont.

• Well tuned ramp metering parameters are critical
  for good performance. ALINEA is the easiest to
  calibrate among the tested algorithms.
• The effectiveness of the ramp control
  algorithms also depends on the level of the
  demand. As traffic demand increases, ramp
  metering tends to be more effective in reducing
  system travel time.
• Ramp metering seems to be more effective under
  certain demand patterns than others.
• Ramp metering may produce greater benefits if
  integrated with queue management, traveler
  information, and arterial street signal
  coordination