Moses, R

You might also like

Download as pdf
Download as pdf
You are on page 1of 8
Evaluation of the 2010 Highway Capacity Manual Urban Street Free-Flow Speed Prediction Model Ren Moses, Enock T. Mtoi, and Eren E. Ozguven The pertrmance ofthe new 210 Highway Capacy Manual (HCM2010) procedure that predicts freefbx speed by using posted speed Hit ane ‘ght ational variables was evaluated. The eight additional variables ‘wer the proportion of segment length with restrictive median the pr portion o segment with curb onthe right-hand side, the number of aces point approaches on the eight side in the subject action of travel, he ‘number of acess point approaches om the eight side inthe opposing Aircon of travel. the segment eng, the width ofthe sigoalzed inter sections Multia pod data from 104 Florida urban street sents with speed fini rangngfrom Sto Smph were used inthe stay. Feld mean {reel speed was determined foreach analysis seament us well as the stometric an tate ateibutes required by HCM 2010 predict fee: ow sped Statistical and graphical analyses showed thatthe HCM 2010 sped prediction methodology had slow eoeticent of determination and sinderpreditd fre-ow speed who the segment posted speed limit was 5 mph or higher: The results further showed thatthe use of re-ow sped price bythe HM 2010 methodalogy in determining arterial eel of service prance lower level of service for 10 of the 20 Sees malyzed Iv adit, callbration of the HCM 2010 made coeticents| ‘hued tat speed prediction by thizmodkt cold be improved he speed lumi eneticent were doubled an that the inane of eros secthon and access density factor shanti decrease, Highway free-llow speed is considered to be the speed a driver ‘chooses under low-volume conditions when the interaction between ‘vehicles and the influence of waffie contol devices are minimal (1). ‘On inierupid-flow highways—that is, highways characterized by intersections and adsie activity suchas driveways—mean free-flow speed is influenced by several geometric, traffic, and imersection ‘control fctors. Te literature shows that the major geometric ati ing a driver's speed choice are median type (estetive ‘ctve). curb type (raised or suppressed). aceess point density, number of lanes, lane width, and segment length (2-4) ic factors infuencing a drivers speed choice are speed LU. J.compostion of trafic (6), andes folowing practices as tated by headway distbution, The intersection factors that affect the choice of speed along sgnalized coidorinelude inal spacing, Dopertmert fC Enger, Frida ABM Univers Paria State Univers loge of Engnearng, Rao 129°, 2525 Peeacrer Sires. alana, FL32010, responding auor-€.T MR, my fs ac, Trosperttin Reseach Rect Join a he Transparent, e205, anspor Rear Berd of oe ena Rear, Wagan, 3. 2084, 1-8 or 404a4/2464-01 progress such a effective green time and cycle length 7. 8) Free-fow speed must be determined correctly because it pays an important role in planning, operational analysis, and performance evaluation of transportation systems (). In systems planning and highway performance evaluation, free-flow speed is sed as an inp in ravel forecasting tor assessment ofsiquaity and in determining congestion ngics. Traffic engineers se fee low speed for highway capacity and level-of service (LOS) evaluation, traffic modeling and simulation, and fuel consumption and emission statis Because ofthe level of importance of free-flow speed in waic analysis, accurate field determination of free-flow speed is eucial However, field determination of fce-low speed for use in analysis isresource intensive because it requires the aequsition of sutcient sample size of vehicle speeds athe appropiate time of low vehicle imteroction and at appropriate locations with homoge ‘uate, andsigalizaton features, To overcome challenges field data ype, and signal timing and coordination parameters collection, most highway agencies use modeling techniques to predict free-flow speed ‘caus free-flow speed is affected by many factors, an elie ‘mode! for predicting fee-low speed mas ake ino acount numerous independent variables related o geometics, traf, and sigznalization Such models have boon propased over the yeas, 4,10), the most recent being the one incorporated into the 2010 Highssay Capacity Manual (HEM 2010) (8). Tae regression model consined in HEM 2010 for predicting free-flow speed is S50 h o free-flow speed, flow speed, and f= adjustment factor for segment However, Equation 1 is not as simple as it looks, Nested Within the two variables Sand fare other variables that must be recalculated (8) Sio=Ss+ fa fe 2 5,=256+0.475, ) 15Piq~O47P~ 3. Pos Pos “ ore, pee o 2 5,280( Nay + Nao) io (=) finro2-47—Sa— 8 o ay! speed constant, adjustment factor for erss section, “fa adjusiment factor for access density, speed Fimit (mph), proportion of segment length with restrictive median, = proportion of segment with curb on right-hand side, [gu = numberof acess point approaches on right side in subject, direction of travel, ‘Ngo = number of access point approaches on right siden opposing direction of travel, segment length, ‘width ofthe signalized imersection, numberof through lanes, distance between intersections, and 002s point density ‘The total count of independent variables thar need to be quantified for use ofthe HCM 2010 free-flow speed equation i nine. The HCM 2010 free-flow speed prediction methodology is easy (0 apply ifan agency or an analyst already has a database ofthese ine variables for alloadway segments within its jurisdiction. Without such acomplete ‘data set, the elficacy of the HCM 2010 free-flow speed prediction ‘mode! is questionable. So, the challenge facing modelers i either to develop comprehensive database of ll nine modeling variables oF'0 evaluate other altematives of modeling free-flow speed, particularly forthe purposes of systems planning and the LOS analysis. This research study was motivated by the need of the Feria Department (of Transportation to evaluate the efficacy ofthe HCM 2010 free-flow speed prediction model and to assess simpler alternative models that can serve the same purpose TABLE 1 Descriptive Statistic of Modeling Vere Variable Min, ‘Speedlimit (mph 5 ‘Number of anes 2 Length with curb (@) Proportion wih cit ° Acces pots in sbjet diction o Access pots in oppsite direction ° Access pint density (pe ile) 47 ‘Legh of segment) Lots Proportion wih restive median os ‘Wisin of downstream inersecions (@) 40, Median widen) 0 Lae wid (8) ° Feld re-Row speed ph) 2 ‘Trensportavon Researeh Record 2461 METHODOLOGY “The research (eam took advantage of archived data collected by the Florida Depastment of Transportation's Office of Statistic, ‘which operates temporary and permanent count stations strategically placed at various locations on the state highway system. This study focuses onthe evaluation of the HICM 2010 urban street fce-ow speed estimation methodology: therefor, only count stations located con urban arterials were selected. t was important thatthe selected sites represented diferent area types, failiy types. and facility sizes. ‘The area types were categorized as central business distri, suburban, and residential: facility types included divided arerils, undivided aerials, collector trees, and centroid connectors. Facility size refers to the numberof lanes inane direction of travel ‘Analysis ofthe distribution ofthe monitoring sites showed that their installation is skewed toward collection of trafMe data on ‘major arterials that have high speed limits. To fil this data gap, ‘which resulted fom an insufficient numberof roadways wit lower classes and speed limits, speed data were collected on selected oad sway segments Because ofthe lack of resources, only 20 sites were selected all in the city of Tallahassee, Site selection was based on speed limit and trafic volume to ensure that roadways of lower class were included in the overall analysis. Table I shows descriptive Statistics of 104 sites selected for analysis. DETERMINATION OF FIELD FREE-FLOW SPEED “The archived speed data acquired from the Florida Department ‘of Transporation hd already been aggregated in hourly speed bins. "There was one speed bin for vehicles waveling at 20 mph or slower. there were 13 speed bins for vehicles traveling at speeds between 20 and 85 mph a S-mph intervals, and there was abn for vehicles traveling faster than 85 mph Because the speed bins inthe raw data acquired from the Florida Department of Transportation were aggregated on an hourly basis, the fist step toward determining free-flow speed was to remove data recorded on weekends and holidays and during special events. Then hours with low traffic ow 104 St © Confenve Mem Max, 8D. Lower Upper “eS 68 ss 25 4 06 2 ase 1S MG aw s7 8 5 a 3200 2 : ms Pa ae 7a | sa as 1 to erase m7 os «ons sia ee Moses, Mic and Caguven were segregated. HCM 2010 defines free-flow speed as te average running speed under conditions of very low volume. In this study, low volume was defined as 200 or fewer vehicles per hour per lane, representing an average headway of 18 s. Ater low-volume hours ‘ovcurrng at night were removed, the hourly harmonic mean speeds were calculated and ranked in ascending order: these included only ‘ours that had a volume of 200 or fewer vehicles per hour per lane. “The formula used forthe calculation ofthe hourly harmonic mean speeds is count,) harmonic mean speed where b= speed bin index (1 0 15), count, = numberof vehicles in speed bin band speed = midpoint ofthe speed range in bin & Forte 20 sites from which speed data were manually collected, individual vehicle records wore acquited. Three-day speed data on 20 roadway segments in Tallahassee wore collected with pneumatic tube counters, providing an opportunity for determining free-flow speed in accordance with the procedure specified in Chapter 30 of CM 2010. The HM 2010 free-flow speed determination procedure has three main stops: ‘Step Conducta spt speed study ata midsegment locaton during losvolume conditions, Record the speed of 100 or mere free-flowing passenger cars. Acar is free-flowing when it has headway of 8s oF ‘more to the vehicle ahead and 5 s or more tothe vehicle behind in the same trafic lane. Step 2. Compute the average of the spot speeds (Sa) and their standard deviation (6p ‘Step 3. Compute the segment free-flow speed (Sasa space mean ‘peed withthe following equation Spa Se 9) eS o ‘These thre steps were applied to the speed data collected by the pneumatic tube counters. A computer program was written {0 ‘extract vehicles whose leading and following gaps were more than 8 sand more than 5 s, respectively. The speeds of these vehicles were then summed and averaged, The final step was the calculation ‘of the space mean speed with Equation 9. Only daytime speeds of| passenger cars (Le, twovaxle vehicles) were used inthe analysis STATISTICAL ANALYSIS OF HOM 2010 FREEFLOW SPEED PREDICTION MODEL. ‘The HCM 2010 model was coded into a spreadshoet containing both dependent and independent variables. The HCM 2010 free-flow speed prediction model already has a deterministic mathematical form, as shown in Equation 1. Therefore, the geometric and traffic tributes of each of the 104 analysis segments were plugged into ‘the model equation to determine the expected free-low speed, The ‘outputs from this model were compared with fcld data to assess the model's goodness offi. SAS statistical software was used in the statistical analysis (1) “The customary accepted statistical procedures for a model per- formance check were used inthe analysis. The statistical analysis involved examining scatter plots of dependent variable versus inde~ pendent variables, checking measures of variation, esting iniieance ‘of parameters, and testing for molticollinearty. Analysis of Scatter Plots Ficid-measured free-flow speed (as response variable) was plot- ted against al nine descriptor variables required bythe HCM 2010 speed prediction model. Te plot of posted speed Fimit(S,) versus Feld-measured free-flow speed showed that there was positive linea comelation between speed limit and free-flow speed. The plot showed that the slope of the line was slightly below the 1:1 line ‘Although the plot of the free-flow speed versus the proportion of restrictive median did not show a clear-cut tren, it appeared that {he analysis sopments had a slight increase in free-flow speed with increasing proportion of restrictive median. Forte free-flow speed versus proportion of segment with curbs, the plot showed a slight . be Loan Field i ow seed, § : noc osat o4se7 © 0 oss Oe Dove m0 ‘moot “ooo 9722 ose 002 Speedlimit, ? 9a Loom 00873. 21s oi? onsen 0.3946 rvs ‘<0001 ae Fo ones re ew Namber of rouh lanes, Ne ° 00m 00173 1000094001 0s89 0.0920 oom 40s pralee S21 68a mm 38S TET 28S 38197 Prop. of egment With eu, Pag a 04567 03215, 0100) 1.000023 uso ons rae ‘ooo 003805 mde TR S69 R006 ” og oa aso 02.0000 0109 0.895 pralve ‘ooo "non 127 a7 ts mR “1012 Access pia density Dy 7 0008 o4es 00920-00483 4992-906 pratue ica | rere, ase eee ‘<0001 #1 Segment eth Log 7 pom ones ots og oo ous76 pale TR SRD 3H S60 < m8 Late With Coen . 04198 0396-01495 ons £ nat OUTS 1.0000 Pale ‘002 ooo t97k 806 ee) m Now ono peal po. = ean ‘Moses, Mio, nd Ozguven free-flow speed and LOS determined with fcld-measured feee- flow speed. The automobile LOS criteria in HCM 2010 are based ‘on performance measures from the feld and as perceived by the traveler The criteria use travel speed of through vehicles to describe Aalomabile LOS. The segment travel speed reflects the factors th influence running time along the segment andthe control delay at intersections within te segment “The segment LOS i expressed as travel speed divided by the base free-flow speed. Unlike HCM 2000, HCM 2010's base free-flow speed sa function of the segment posted speed, the cross-section adjustment factor, and the access point adjustment factor. The free-flow speed is the product ofthe base fee-flow speed and the signal spacing adjust- ‘ment fator as shown in Equation 1, For calculating the signal spacing adjustment factor, signal timing daa forthe 20 analysis segments were acquired from the cty of Tallahasec's traffic signal systems office. TABLE G Analysis of LOS ‘Table3 showsthe resus ofthe analysis. Inall eases, ll other variables ‘oma segment ae the sume excep for free-flow sped, ‘The results in Table 3 show that use of HCM 2010 predicted speed results in one LOS lowee than the LOS obtained bythe field measured value for 10 ofthe 20 segments analyzed, a 50% error rate, These results mainly reflect thatthe arterial automobile LOS. procedure is too sensitive to input free-flow speed. Since the HCM. 2010 model underpreiets free-flow speed when field free flow is about 40 mph or higher, these results ate not surprising. A recent study conducted by Ozkul etal. tha evaluated the HCM 2010 urban streets automobile LOS methodology found that compared with HEM 2000, the current methodology produces poorer LOS. par- ticularly for short arterisl segments wit lower posted speed limits ‘such as those found in central business dstrits (72). These results ‘are somewhat consistent with the findings of Ozkul etl. ‘Speed Limit Prediction FreFlow Cycle Control Segment Segment enh) Method Speed (nh) Length(s) IC Ratio _Delay(s)__ LOS ‘Apalachee Pukovay 45 HM Pm 150 040 ass Feld measured si 150 040 6B South Adan Suest 4s HM “ 150 oss ese Feld measured ea 150 as uC [Noth Macomb Sst Ho 26 130 04s a5 € Field messed x 130 neers 55) I Lake Bedford Steet, Hom 0 10 oss 0 Feld measured 8 10 os = 0 Thomasville Road t 3s eM 38 10 os 3550 ld measured 2 160 0% = 3550 ‘Tomaville Read 2 38 eM 40 10 0360 eld measured 40 160 eee soe! CI Tennessee Steet, 3s eM 36 160 035 3e10 Field measred 8 160 bas 310 Bla Stone Road 1 30 hem a us as 3610 Feld measured 7 ist as solo D Bhai Stone Road 2 3s eM 3 rn ne Field measred a ue 0 eo ‘Onage Avenue 3s im 3 180 035 3110 Fld measured 2 bo 033 u € ‘Tarpe Stet 0 Hem a7 150. 04s 90D Feld measured BM 150 043 sas ‘Tennesse Stet x0 HM a 10 oat 3500 c Feld measured x 160 oat 3300 North Mone Stost 2s Hem 36 130 o4s 315 ¢ Feld measured 3 10 oa 535 Paul Russel Road x0 HOM u 150 o4 = to > Feld measored % 150 ou = IoD (Cait Cle NE 1 6 Hew “4 160 oat 350 Cc Feld measured 2 10 oat 35 8 Capita Circe NE 4s eM 3 160 oat 340 Field eased 2 16 oat beac Miccosukee Road | 3s eM 8 10 oss asc Field measured a 10. 04s BoC Micosiee Road 2 30 Hes 36 130 048 3i30 > Feld measred 36 160. oa ais Wes Pensacola Hest a 120 050 saa > Feld mesared eo 140 oc) 0) Seuth Monroe Stet as He 28 150 oo ail led measured a 130 oD [Nort i geen ie cel lng; NE w noes IMPROVING HCM 2010 FREE-FLOW ‘SPEED PREDICTION MODEL Giventhe abundance ofthe data acquired fr this study, the researchers explored several techniques with which to improve the HCM 2010 {hee-flow speed prediction model. Two methods were considered ‘Thefts method used a backward elimination technique to find the best-fit model. Te second method tweaked the current HCM 2010 ‘model coeficiens to ineease the model's predictive power “The bestfit regression model related the fleld-estimated free-flow speed (FFS) with the segment posted speed limit (S,), the numberof ‘through lanes (Na), the proportion ofa segment witha raised cut (Pap) the proportion of a segment with a restrictive median (Pa), the access point density (D,) the segment length Ly) and the Tane width Ly) The model was specifi with the following regression equation: N(0,0") 9) FS =f 358.X,+ INTER +, here By = intercept ofthe model, efficients of predictor variables X, ror terms, and INTER « interaction terms between the variables, Statistical results ofthe fitted model are shown in Table 4. With a p-value of 05 asa cutoff point, Table 4 shows that al independent varables were significant except numberof lanes and lane width ‘The generalized equation forthe resulting model is 6.8008 +0.96615,~ 5.6238Py + 0.0003Py + 0.0144, + 0.0001La, +6, ay Compared withthe existing HCM 2010 speed prediction model, the most notable change isthe speed limit coefficient, which changes rom 0.47 inthe current modelo 0,9661 in the best-fit model show {in Equation 11 In comparison with the HCM 2010 model, the best-fit ‘model suggests that speed limit has almost double the influence on free-flow speed Transportation Research Record 2461 The best-fit model was checked for multcollinearty through assessment ofthe variance inflation factor (VIF) for each variable “The VIF is a widely used measure ofthe degre of multicollinearity ‘ofthe ith independent variable with the other independent variables ina regression model. The most common rule of thumb i that if ‘the VIF is more than 10, this shouldbe regarded asa sign of severe ‘ultcolineaity among the variables (13). The results in Table 4 show that the VIF values ae well below 10. Hence. the use of ridge regression (14, 15) o futher elimination of variables is unnecessary Case diagnostics and analysis of residuals were also carried out ta check model validity and to check cases that would need further analysis, The results showed tha the residuals sats the assumpxions ‘of normality and there was no indication of outers in the data used. The performance of the bes-fi model (Equation 11)in predicting {rec-low speed was graphically compared tothe performance ofthe HCM 2010 speed prediction model. The field free flow speed was plotted agains the predicted free-flow spood and aginst the predictive power ofthese two models, as shown in Figure 2 ‘The figure shows that the best-fit regression model traces the field ata wel across the whole range of measured field ree-low speeds. Incontrast, the HICM 2010 madel appears to underpredit free-flow speeds when the feld-estimated free-flow speed is approximately above 40 mph ‘Tweaking HCM 2010 Model Coefficients “To increase the predictive power of the HCM 2010 model, the mode's base free-flow speed equation was calibrated with feld data, Asindicated earlier, the HCM 2010 base free-flow speed equation is Syp=256+047S + fat Su 2) Calibration involved determining optimum model coefficients that would improve the predictive power ofthe model with the data sequired for his study. The model tobe fited took the following for: Si0=Bo+ BiSu + Bafa Bada as) “The field base free-flow speed was estimated as the upper bound ofthe 95th confidence interval ofthe average free-flow speed. The choice ofthe upper bound of the 95th confidence interval was based ‘on the assumption tha base free-flow speed is higher than the aver: age free-flow speed and should not include afew observed extreme ‘values. All 104 segments were used inthe calibration of the base TABLE 4 Bost-Fit Regression Mods! Parameter Standard Variable) Esuimate(B) Ener Pr>iel__VIF Inercep 68008 1784200 Speedlimit (mph) 096610032 <000 [Namber of anes oom = 037 Proportion with ob see 149680196, Proportion wthresictve median «0.000300 ‘Access point density (per mile) oor ons? sas Length of segment) 0.0001 on Lane width (6) 0.394 ns 138 [Nove VIF = variance nfton fat aed Mawes, Mic ara Caquven Predicted Free-Flow Speed (mph) 30 350s ——+Feld Data —enem eesti - so $5 60 6570 Field-Measured Free-Flow Speed (mph) FIGURE 2 Soattar plot of fed vers TABLE 5 Calibration of HOM 2010 Parameter Standard arable (X)_Estimate(B) Emer Niue Pr>|e|_VIF Consant(B) 1089 —«4.208 «9080

You might also like