Traffic Control Optimization for Multi-Modal Operations in a Large-Scale Urban Network

Transcription

1 Traffic Control Optimization for Multi-Modal Operations in a Large-Scale Urban Network Cameron Kergaye, PhD, PMP, PE UDOT Director of Research 13th Annual NJDOT Research Showcase October 27 th, 2011

2 Improve Pedestrian Operations Reduce Vehicular Emissions Provide Priority for Transit Improve Traffic Flow

3 Source: Signal Timing Manual, FHWA 2008 Do we have right objectives and policies? No, sustainable policies to consider transit, emissions, person-based costs, etc. Do we (always) know what performance measures to use? No, we use surrogate performance measures (in lieu of emission-related metrics) Can current models support our goals? No, we need better models and better interfaces between models and field operations

4 Minimizing Delay of: Signal Timings to Optimize: Private Cars Person* Transit Vehicles Basic Very common Uncommon Common TSP Uncommon** Uncommon Very common** Basic + TSP Uncommon Uncommon Uncommon * Delay per person in the system regardless of travel mode used. ** TSP parameters not optimized (but manually adjusted) since commercial optimization tools do not adjust these parameters. 4

5 Unsuitable software (modeling tools) Signal optimization tools No transit operations No TSP optimization Simulation tools No optimization of signal timings Current practice Optimization of basic signal timings TSP settings adjusted based on engineering judgment Simulation used for fine-tuning and quality control but not in the optimization process 5

6 Basic Signal Timing Concepts Cycle length Split (A phase) Offset 6

7 Passive Priority Active Priority early green (red truncation) green extension actuated transit phase phase insertion phase rotation Adaptive/Real-Time Control 7

8 Distance Early start time Normal start time Transit vehicle trajectory Time 8

9 Source: Transit Signal Priority Evaluation Results Woodward Avenue. PTV America 9

10 Distance Normal end time End time with extension Transit vehicle trajectory Time 10

11 Gives the TSP phases extended green while shortening the next cycle s TSP phase Recovery at the end of the next cycle 11

12 Interface a signal optimization tool with a microsimulation tool Use GA as an optimization routine Use microsimulation (VISSIM) to model transit operations Use signal control emulator (within VISSIM) to truthfully model TSP operations 12

13 Better than searching techniques used in deterministic tools Optimization in complex search space (relationships between signal timing parameters - complex) Running many simulations is a slow process GA achieves good solution after a relatively small number of generations 13

14 Technique for solving search and optimization problems Solutions are evolved Stochastic search process based on survival of the fittest Mimics natural evolution GA in general independent from the specific problem 14

15 Encode Timing Plans Initialize First Population Run VISSIM & Evaluate Population End Criteria Satisfied? no Create Next Generation of Population yes Return Best Timing Plan 15

16 GA Optimization Inputs 16

17 Woodward Ave, Detroit, Michigan 13 miles long 70 signalized intersections 3 bus routes 12 buses per hour TSP Evaluated by PTV America (2006) Source: Transit Signal Priority Evaluation Results Woodward Avenue. PTV America 17

18 Vehicle Delay (s) Initial Vehicle Delay Optimization of Basic Signal Timings Optimization of TSP Settings Optimization of All Signal Timings Number of Generations 157 a. 18

19 Transit Vehicle Delay (s) Number of Generations b c. Initial Transit Vehicle Delay Optimization of Basic Signal Timings Optimization of TSP Settings Optimization of All Signal Timings Number of Generations 19

20 Person Delay (s) Number of Generations a. Initial Person Delay Optimization of Basic Signal Timings Optimization of TSP Settings Optimization of All Signal Timings Number of Generations b. 20

21 Optimization of basic signal timings most important 2nd best are optimizations of TSP settings Non-traditional performance measures (e.g. person delay) are successfully used as optimization objective functions 21

22 Signal Timings Optimized to Minimize Delay of Transit Vehicles y = x R² = 0.86 y = x Signal Timings Optimized to Minimize Delay of Private Vehicles a. 22

23 Signal Timings Optimized to Minimize Multi-Modal Delay per Person Signal Timings Optimized to Minimize Delay of Private Vehicles a. y = x R² = y = x Signal Timings Optimized to Minimize Delay of Transit Vehicles b. 23

24 Initial Signal Timings y = x R² = y = x Signal Timings Optimized to Minimize Delay of Private Vehicles a. 24

25 Initial Signal Timings Signal Timings Optimized to Minimize Delay of Private Vehicles y = x R² = a. y = x Signal Timings Optimized to Minimize Delay of Transit Vehicles b. 25

26 Linearity of the investigated network Simplified phase designs of signal controllers (reducing the chance for benefits from TSP) Total number of parameters to be optimized Disparity between number of private drivers and transit passengers Good coordination provided for buses which join vehicular platoons 26

27 Repeat the study on a network where major transit and traffic flows do not share ROW but compete for priority from different approaches. In this research: 1 sec of delay of transit pax = 1 sec of driver s delay Instead, social costs of passenger s VMT vs driver s VMT can be used for fuller cost accounting These social costs may be further contrasted if sustainability costs (long-term impact on human environment) are involved in calculations 27

28 Questions & Comments