Developing an Adaptive Traffic Signal Control System

1
Developing an Adaptive Traffic Signal Control
System

• Michael Shenoda, P.E.

2
Purpose of Research

• Defining traffic control generally
• Three types of traffic control
• Pretimed, Actuated and Adaptive
• Goals of traffic control research
• Why focus on adaptive control?

3
Purpose of Research

• Looking at previous implementations of adaptive
  control what was found
• Setting of signal timings based on predicted
  volume of arrivals (over various interval
  lengths)
• Set length of horizon over which to analyze
  traffic patterns
• fixed or fixed-variability (i.e. variable by
  fixed intervals) cycle lengths
• Closed and/or convoluted structure

4
Aims of Research

• Development of new algorithm centered around four
  concepts
• easy to understand and implement
• flexible (i.e. useful in as many configurations
  as possible)
• improve upon previous implementations
• allow continual improvement
• Incorporation would allow the best opportunity to
  advance the concept

5
Progress of Research

• Steps involved
• review of previously implemented traffic control
  techniques
• developing optimization technique
• developing an algorithm framework
• coding of the algorithm (Excel, then C)
• testing of the algorithm
• application to a real-world setting

6
Aims of Adaptive Control

• How do we expect it to perform in relation to
  other control methodologies?
• Traffic demand regimes
• Low traffic / capacity less than 50
• Medium traffic / capacity between 50 and 75
• High traffic / capacity between 75 and
  100
• Oversaturated traffic / capacity greater than
  100

7
Aims of Adaptive Control
Comparative Benefit of Adaptive Control vs.
other control methods based on demand
8
Aims of Adaptive Control
Improvement of measures of effectiveness of
Adaptive Control vs. Pretimed Control
9
Structure of Adaptive Control
10
Structure of Adaptive Control

• Detection
• The process by which vehicles entering the system
  are recognized and processed
• Prediction
• The process by which detected patterns of
  arriving vehicles are used to determine patterns
  of future vehicles
• Optimization
• The process by which detected and/or predicted
  vehicle arrivals are used to distribute green
  time

11
Optimization Highlights

• Formulation of optimization function
• Based on simple two-approach intersection

12
Optimization Highlights

• Phases i 1,,N vehicle ID j
• Horizon length, H
• l prop. over which the phase remains green
• Phase start (ps), phase length (pl ps lH),
  phase end (pe ps H)

13
Optimization Highlights

• The optimization function
• Subject to
• Where xij 1 if , 0 otherwise
• yij 1 if , 0 otherwise
• i approach i j vehicle j k phase k
• ps start of phase
• lk proportion of horizon elapsed at phase k
• H length of horizon
• tij arrival time of vehicle j at approach i
  

14
Optimization Highlights

• Preliminary two-phase experimentation
• Single- and dual-intersection networks
• Changes in demand over simulation length
• Arrivals by truncated Poisson for low to med.
  vol.

? Proposed algorithm adjusts green time
proportions to changes in demand
Gap in MOE ? performance widens with deviation
from initial demand
15
Optimization Highlights

• Solving the optimization function
• Exhaustive enumeration (originally used in
  optimization at 0.0001 resolution)
• Numerical search procedures (e.g. steepest
  descent, golden section, etc.)
• Heuristic/meta-heuristic search procedures
• Need to solve for larger networks within a
  reasonable time frame

16
Optimization Highlights

• How do we accomplish this?
• Consider that the signal timing can only be
  optimized by coinciding phase changes within some
  e of arrival time points
• For each phase, select lk corresponding to the
  arrival time of a vehicle utilizing the phase
• Each lk becomes a point to be visited in the
  solution space

17
Optimization Highlights

• What are the advantages of a heuristic/
  meta-heuristic approach?
• reduces the number of possible l solutions
• reduces the time to reach an optimal solution
• provides a logic to the solution space search
• Exhaustive search last-to-first
• Heuristic proportional
• Meta-heuristic tabu search

18
Research Highlights

• Other considerations
• Addition of steps to overall algorithm to allow
  recursive operation
• Detection considerations advanced use of single
  detectors video, other detection means for
  vehicle recognition to improve MOE measurement
• Prediction considerations very short-term
  volume prediction time series w/feedback,
  upstream feeding

19
Conclusions/Summary

• What can the algorithm do?
• phase-by-phase operation
• updates on a user-defined interval
• based on actual arrival times
• user-defined or optimized horizon lengths
• perform meta-heuristic search on solution space
• improvement upon pretimed signal control
• compare favorably to or outperform other
  methodologies
• open structure with unique features and easy
  understanding

20
Conclusions/Summary

• What are the final goals?
• Ability to mirror actual field conditions
• Optimization of any desired MOE on any
  real-world network
• Consideration of adaptive control as a basic
  theoretical traffic control method

Qs? As?