NCTCOG Regional Travel Model Improvement Experience in Travel Model Development and Data Management. Presented to TMIP VMTSC.

Transcription

1 NCTCOG Regional Travel Model Improvement Experience in 2009 and Data Management Presented to TMIP VMTSC December 7, 2009 Presenters Kathy Yu Senior Modeler Arash Mirzaei Manager Model Group Behruz Paschai Senior Modeler 1

2 Content Previous TMIP VM presentation by NCTCOG in March 2009 Data Sources for Model Improvements Improved Model Components Estimation, Calibration, and Validation Previous TMIP VM NCTCOG Presentation March 13, 2009 Topics Front end of the Regional Travel Model Travel model application software (DFWRTM, DFX) Data Collection and Maintenance Data management program examples Transit surveys Model Components Design and Improvements Described model structure Introduced improvement components in traffic assignment and feedback process We will expand major model improvements, reasons behind their selection, the building processes, and performance 2

3 Data Sources for Model Improvements TxDOT Daily Traffic Counts for 5 year program Purpose is HPMS We receive the daily count by location Data clean up is a major task Identification of wrongly coded counts is difficult Once identified, correction method is per case Systematic identification of errors is not practical Point of comparison does not exist Therefore, clean up is never over! Neither is calibration/validation Data Sources for Model Improvements Daily Counts From Other Sources + Cities, counties, airports, and toll authorities in many formats Geo coded and put in the data base Quality control issues are similar to TxDOT counts Counts and speeds from other NCTCOG sponsored projects Arterial counts and travel time studies on more than 100 corridors by time of day Toll road travel time studies on freeways All geo coded and put in the database system 3

4 Data Sources for Model Improvements Transit Data 3 transit onboard surveys in 2007 and 2008 (in previous session) Daily boarding data for all transit lines for 2007 and 2008 External Trip Data 2005 survey 2005 counts Improved Model Components Roadway Model Assignment convergence criteria Developed new volume delay function Model calibration Developed New Feedback convergence criteria Transit Model Used transit onboard surveys Developed new transit network model Developed new vehicle ownership model Developed new mode choice model 4

5 Roadway Model Assignment Convergence Criteria Assignment Convergence Issue The DFWRTM performs a UE assignment in TransCAD with a convergence criteria of 30 iterations. This process created some level of noise in the model results. Noise is the difference in link volumes between the Ideal User Equilibrium and the traffic assignment performed in DFWRTM. 5

6 Assignment Convergence Issue Study objective : are 30 iterations enough to reach stable volumes on all links? Study goals : how the convergence criteria could be changed to : reduce the RMSE from an ideal User Equilibrium; reduce the model noise level; and maintain a reasonable model run time. Assignment Convergence Study Test scenarios : 30 iterations relative gap relative gap Ideal User Equilibrium Comparison measures : RMSE Assignment Run Time Noise reduction for a controlled change 6

7 RMSE Comparison* Scenario Average 30 iterations RG RG Ideal UE * RMSE between the analyzed scenario and the Ideal UE. Assignment Run Time Comparison Scenario Iteration Run Time Iteration Run Time Iteration Run Time Iteration Run Time Iteration Run Time 30 iterations 30 0: : : : : RG 83 2: : : : : RG 126 3: : : : :47 Ideal UE : : : : :55 7

8 Comparing % Link Flow Change 30 Iterations.001 Relative Gap Controlled Change Link Volumes Reduced 2 lanes on EB IH 30 : Expect traffic on EB IH 30 to decrease Expect ramp volumes to change 30 Iterations.001 Relative Gap 8

9 Traffic Assignment Conclusions Choose.001 relative gap : Reduced RMSE to 1% Significantly reduced noise from 30 iterations Is more affordable time than ideal UE since the time for an assignment could reach up to 14 hours. Feedback Test Results Total Trip Differences in Consecutive Feedbacks PADIST 60,000 50,000 MSA/.001RG.25CW/.0001RG 0CW/.001RG Naïve.05CW/.001RG.10CW/.001RG.15CW/.001RG.20CW/.001RG.25CW/.001RG.30CW/.001RG Total Trip Differences 40,000 30,000 20,000.40CW/.001RG.50CW/.001RG 10, Feedback Note : PADIST is the sum of the HBW, HNW, and NHB person trips 9

10 Feedback Tests Conclusions After implementation of Feedback loops, a relative gap of.001 was found to not have a desirable network level convergence. Therefore, a.0001 relative gap was selected in the model application. Roadway Model Volume Delay Function 10

11 Goals Provide the Capability to : reach equilibrium remove minimum speeds incorporate traffic control delay reduce model noise incorporate HCM concepts in calculating capacity and traffic control delay, if practical Approach Data availability : limited speed and travel time data no concurrent speed and volume data VDF Development : start from scratch (not from previous VDF) develop the VDF concepts formulate and implement the VDF maximize use of available data 11

12 VDF Improvements Item Expanded Travel Time Post Processing VDF Assignment VDF VDF Form Traffic Control Delay NCTCOG (w/ minimum speed) Static (not a function of v/c) CONICAL (w/ Traffic Control Delay Function) Dynamic (function of v/c) Modified VDF Delay CONGESTION DELAY Conical Delay Function (C d ) calibration parameter (function of link volume) SIGNAL DELAY Uniform Signal Delay (S d ) C : signal cycle length r: approach red time calibration parameters (function of link volume) STOP and YIELD DELAY Un Signalized Intersection Delay (U d ) calibration parameters (function of link volume) 12

13 VDF Congestion Delay Exponential Congestion Delay Function : Conical Congestion Delay Function : C d = T 0 * ( K { K d d v = 0}) c + 2 v v K = 1 A_CONICAL * (1 + dx) 2 + B_CONICAL 2 A_CONICAL * (1 + dx) B_CONICAL d c c (2 * A_CONICAL 1) B_CONICAL = (2 * A_CONICAL 2) Freeway VDF Comparisons Conical Freeway Volume Delay Function Typical Form Comparison CONICAL NCTCOG ASSGN SPD NCTCOG POST SPD Speed Limit = 60 mph, no traffic control delay 50 Speed (mph) exponential congestion delay conical congestion delay V/C 13

14 VDF Signal Delay Webster s uniform delay : PARAM_SIG, PARAM_SIG > 0, 0 v/s L 1 v MAX ( 1, L ) S S d = v 3 v 2 v CA * + CB * + CC * + CD S S S, PARAM_SIG > 0, L 1 < v/s < L 2 PARAM_SIG, PARAM_SIG > 0, v/s L 2 v MAX ( 1, L ) S 0, otherwise VDF Signal Delay Original NOT Differentiable 14 Delay Corrected Differentiable (correction range) V/S 14

15 VDF + Signal Control Arterial 45 Volume Delay Function Typical Form S/C = 1 40 Speed Limit = 40 mph, FUNCL = NCTCOG VDF Conical VDF w/out Signal Delay Speed Conical VDF w/ Signal Delay V/S Modified VDF Unsignalized Control Delay U d = MIN_DELAY + PARAM_UNSIG (v/c), PARAM_UNSIG > 0 0, otherwise V/C 5 Delay 15

16 VDF + Stop Control Arterial Volume Delay Function Typical Form Speed Limit = 40 mph, FUNCL = Conical VDF w/out Stop Delay Speed 20 NCTCOG VDF Conical VDF w/ Stop Delay V/C 2 Roadway Model Calibration Results 16

17 Location of Traffic Counts Number of locations : ~10,000 Source : TxDOT Saturation Counts 2003 (Hill County) 2005 (Hunt County) Model RMSE Comparisons FUNCL DFWRTM Expanded RMSE % F/C # OBS RMSE % F/C # OBS , , , , , , , , ALL , ,263 17

18 Link RMSE Comparisons Daily Traffic Count (vpd) %RMSE Expanded > 50, ,000 X < 50, ,000 X < 25, ,000 X < 10, X < 5, ALL Model Volumes All Volumes Model Volume y = x R 2 = Counts 18

19 Freeway Volumes Freeway Volumes Model Volume y = x R 2 = Counts Freeway Corridors RMSE Comparisons Freeway %RMSE I30 WB 9.22 I30 EB 9.30 I20 WB I20 EB I35W SB I35W NB I35E SB 8.84 I35E NB US 75 SB US75 NB I635 WB I635 EB I635 SB 5.73 I635 NB 8.96 SH360 SB SH360 NB I45 SB I45 NB % RMSE Expanded

20 Model Cut Lines Model Cut Lines RMSE Comparisons 6% 3% 7% +7% 4% 20

21 Model Cut Lines RMSE Comparisons 1% +2% 7% 11% 8% +4% Model Cut Lines RMSE Comparisons 18% +3% +1% 8% 8% +3% +19% 21

22 Model Screen Lines Model Screen Lines %Error Screen Line Absolute % Error % 90.00% 80.00% Expanded Cumulative Percentage 70.00% 60.00% 50.00% 40.00% 30.00% ~41% ABS(%ERROR) < 10% ~69% ABS(%ERROR) < 20% ~82% ABS(%ERROR) < 30% 20.00% 10.00% 0.00% Absolute %Error 22

23 Freeway Travel Time Database Route Lengths : miles Freeway Speed Comparisons AM Freeway Speed Comparisons (Scenario 29) Average Speed (mph) AMMSPD AMFSPD _TT 056_TT 055_TT 014_TT 003_TT 021_TT 023_TT 063_TT 036_TT 054_TT 004_TT 012_TT 035_TT 008_TT 029_TT 025_TT 020_TT 013_TT 030_TT 062_TT 007_TT 026_TT 022_TT 002_TT 039_TT 001_TT 011_TT 040_TT 23

24 Freeway Speed Comparisons PM Freeway Speed Comparisons (Scenario 29) Average Speed (mph) PMMSPD PMFSPD _TT 056_TT 055_TT 014_TT 003_TT 021_TT 023_TT 063_TT 036_TT 054_TT 004_TT 012_TT 035_TT 008_TT 029_TT 025_TT 020_TT 013_TT 030_TT 062_TT 007_TT 026_TT 022_TT 002_TT 039_TT 001_TT 011_TT 040_TT SkyComp Freeway AM Speeds LEGEND 0 < x < 30 mph 30 < x < 50 mph Model Data SkyComp Data 24

25 SkyComp Freeway PM Speeds LEGEND 0 < x < 30 mph 30 < x < 50 mph SkyComp Data Model Data Arterial Travel Time Database Number of routes : 131 Total length : 303 center line miles 25

26 Arterial Speed Comparisons AM Average Speed Comparisons (Scenario 29) AMMSPD AMFSPD Average Speed (mph) _TIPTT 107_TIPTT 120_TIPTT 127_TIPTT 133_TIPTT 141_TIPTT 153_TIPTT 163_TIPTT 171_TIPTT 186_TIPTT 205_TIPTT 224_TIPTT 238_TIPTT 255_TIPTT 265_TIPTT 277_TIPTT 288_TIPTT 333_TIPTT 343_TIPTT 354_TIPTT 366_TIPTT 378_TIPTT 393_TIPTT 403_TIPTT 412_TIPTT 424_TIPTT Arterial Speed Comparisons PM Average Speed Comparisons (Scenario 29) 60 PMMSPD PMFSPD 50 Average Speed (mph) _TIPTT 107_TIPTT 120_TIPTT 127_TIPTT 133_TIPTT 141_TIPTT 153_TIPTT 163_TIPTT 171_TIPTT 186_TIPTT 205_TIPTT 224_TIPTT 238_TIPTT 255_TIPTT 265_TIPTT 277_TIPTT 288_TIPTT 333_TIPTT 343_TIPTT 354_TIPTT 366_TIPTT 378_TIPTT 393_TIPTT 403_TIPTT 412_TIPTT 424_TIPTT 26

27 Arterial Speed Difference Distribution Non Freeway Speed Difference Distribution % 90.00% Cumulative Percentage (%) 80.00% 70.00% 60.00% 50.00% 40.00% 30.00% 20.00% 10.00% AM Speeds PM Speeds 58% of speed differences are less than 5.0 mph 97% of speed differences are less than 10.0 mph 0.00% Absolute Value of Speed Difference (mph) Roadway Model Feedback Convergence Criteria 27

28 Goals Provide the Capability to : perform consistent model runs reduce the model noise report model volume accuracy statements Define the Assignment Convergence Criteria as : a multi dimensional criteria that is a function of the individual link volume and travel time in addition to the relative gap as defined in TransCAD DFWRTM Model Process Trip Generation Trip Distribution Mode Choice Roadway Assignment Convergence Criteria Satisfied? YES Transit Assignment NO Zone Layer Demographics Roadway Skims 2 Feedbacks 30 iterations 0 relative gap Roadway Network Transit Network Transit Skims FEEDBACK LOOP 28

29 Current DFX Model Process Trip Generation Trip Distribution Mode Choice Roadway Assignment Convergence Criteria Satisfied? YES Transit Assignment NO Zone Layer Demographics Roadway Skims 5 Feedbacks Unlimited iterations relative gap Roadway Network Transit Network Transit Skims FEEDBACK LOOP Proposed DFX Model Process Zone Layer Trip Generation Demographics Trip Distribution Mode Choice Roadway Assignment Skim Averaging (Constant Weight/.25/.75) [D i ] = F([Di 1], [Ai]) [D i 1 ] = Average Roadway Skims (i 1) [A i ] = Roadway Skims (i) FEEDBACK LOOP [A i ] Roadway Skims (i) Convergence NO Criteria Satisfied? Unconstrained Feedbacks YES Unlimited iterations Pre Defined Criteria Transit relative gap Assignment 29

30 Proposed DFX Model Application TransCAD 1. Number of iterations : Unlimited 2. Relative gap : Number of feedbacks : min 3, max 10 (18 45 hours) DFX 2. Skim Matrices RMSE : 1% 3. Maximum change in Skim Matrix cells : 10% 4. Link Volume RMSE : 2% 5. Link Volume Change : ΔV F1 15% one lane capacity ΔV F2 20% one lane capacity ΔV F3 25% one lane capacity ΔV F4 25% one lane capacity ΔV F6 25% one lane capacity ΔV F7 50% one lane capacity Feedback Results 5.00% Skim RMSE Feedback i vs i 1 (PKNOHOV) 4.50% 4.00% 3.50% 1 SKIM RMSE 3.00% % RMSE 2.50% 2.00% 1.50% 1.00% 0.50% 0.00% Feedback Number (i) 30

31 Feedback Results Skim Cell by Cell Percent Difference Feedback i vs i 1 (PKNOHOV) % % 2 SKIM DIFF % % % % % % % % Feedback Results 4.000% Link Volume RMSE Feedback i 2 vs i 1 (AMVOL) 3.500% 3.000% 3 AMVOL RMSE 2.500% % RMSE 2.000% 1.500% 1.000% 0.500% 0.000% Feedback Number (i) 31

32 Feedback Results Ratio of Maximum Link Volume Difference over One Lane Capacity 50.00% 45.00% FUNCL 1 FUNCL 2 FUNCL 3 FUNCL 4 FUNCL 6 FUNCL 7 FUNCL % 35.00% 30.00% Ratio (%) 25.00% 20.00% 15.00% 10.00% 5.00% 0.00% Feedback Number (i) Feedback Results Run Time (hours) Time Time (hours) Feedback 32