1、Application of a Multi-Level Logit Function HOV/SOV Forecasting Procedure in theGreater Vancouver Transportation ModelChris HoffMinistry of Transportation GCSOV = Generalized cost from previous auto assignment;PKSOV = Parking Cost for SOV;VOT = Assumed value of time for trip purpose.HOV IMP WT* GC +
2、PN K *60VOTSH +DF1 AT+ BiasHOV Where:HOV IMP = HOV Impedance for specific to vehicle classification;WT = Factor representing reduced Ride-Share opportunity as impedance increases;GCHOV = Generalized cost from previous model run;PN = Represents inconvenience and additional delay of making a rideshare
3、 trip;PKHOV = Parking Cost for HOV;VOT = Assumed value of time for trip purpose;SH = Number of occupants sharing the parking charge;DF = Density factor to represent impact of trip density on Ride-Share opportunity;AT = Number of trip attractions per trip purpose in destination zone; HA = Destination
4、 zone size in hectares; and,Bias = Sub-modal bias reflecting the desire for privacy.Ride-Share Impedance was calculated as a weighted mean of 2HOV impedance and 3+HOV impedance, namelyA Multi-Level Logit Function HOV/SOV Model for Greater Vancouver Page 7EMME/2 Users Group Conference, Houston Octobe
5、r, 1998Edwin Hull Associates LTD Consulting Transportation EngineersIMPrs = ln (exp (- * IMP2hov) + exp (- * IMP3+hov) / -CalibrationTwo sets of data were available for model calibration: The 1992 Home Interview Survey (2% sample) The 1996 screen-line traffic and vehicle occupancy counts from roadsi
6、de observationsThe Home Interview Survey provided data on vehicle occupancy by trip purpose. Unfortunately, with a 2% sample rate, the number of reported HOV trips, particularly 3+ HOV trips, was quite small. Even with reported trips aggregated to nine regional superzones, the reported values for AM
7、 peak hour work trips were not statistically significant for more than 50% of the 81 cells in the superzone matrix. For the other modelled purposes Post-Secondary, Grade School, Business and Other the reported values were statistically significant in less than 10% of the superzone matrix cells.Despi
8、te these concerns, assigning the synthetic matrices based on calibrating the model to the Interview Survey data produced a reasonable fit to total HOV use at regional screen-lines. Unfortunately, the 2HOV/3+HOV split showed considerable deviation from the screen-line occupancy counts. The 3+HOV synt
9、hetic volumes were typically between two and four times the observed volumes. Discussion with the Ministrys traffic data manager indicated that a low level of confidence in the roadside observations of occupancy. In particular, it was considered likely that many 3+HOVscould have been recorded as 2HO
10、Vsbecause of the difficulty in seeing back seat passengers in high-speed vehicles on multi-lane highways and bridge approaches.Consequently, it was decided to calibrate the parameters to generate screen-line volumes consistent with the total HOVs reported by the roadside counts. However, the breakdo
11、wn of HOVs into 2-person and 3-plus-person was calibrated to 3+HOV volumes approximately midway between the observed counts and the values derived from the Home Interview Survey.It was also observed that there was relatively little variation in the percentage of vehicles in each occupancy category,
12、either from screen-line to screen-line or among the cells in the superzone matrix derived from the Home Interview Survey. We recognize the difficulties inherent in attempting to calibrate so many parameters from a limited range of observations. Consequently, although the model based on the “calibrat
13、ed” parameters provides a good fit to the target screen-line volumes, the values assigned to each co-efficient owe as much to professional judgement and intuition as to any rigorous evaluation of the sensitivity of the forecasts to changes in the relative values of the coefficients.The narrow range
14、of observed values also raises concerns about application of the model to future conditions. The network of HOV facilities planned by 2006, in combination with other TDM measures intended to increase out-of-pocket costs for SOV use, will radically A Multi-Level Logit Function HOV/SOV Model for Great
15、er Vancouver Page 8EMME/2 Users Group Conference, Houston October, 1998Edwin Hull Associates LTD Consulting Transportation Engineerschange the relationship between SOV and HOV impedances. This means that the model will be applied to conditions well outside the range of observations available for cal
16、ibration. Clearly, there is a need for a comprehensive data collection and monitoring program to complement the implementation of the HOV lane program in the region. The data collected in this program will allow the coefficients used in the HOV/SOV model to be refined.Table 1 summarizes the paramete
17、rs used for the HOV impedance calculations for four of the five trip purposes. Note that the HOV/SOV split for grade school trips is determined based on fixed percentages for all origins and destinations.Table 1: AM HOV Impedance Calculation ParametersWork Post Secondary Other Business2 person weigh
18、t 1.1 1.05 1.1 12 person penalty 5 8 10 52 person density factor 32 26 7 152 person HOV bias 11 5 10 163+ person weight 1.2 1.1 1.15 1.13+ person penalty 12 15 15 103+ person density factor 36 30 8 253+ person HOV bias 17 8 12 20SOV/Rideshare 0.075 0.07 0.08 0.052HOV/3+HOV 0.085 0.08 0.085 0.06Sensi
19、tivity of ForecastsApplication of the model to 2006 scenarios indicated a significant increase in HOV use, particularly 3+HOV use. Examination of the SOV and HOV generalized costs and impedances indicated that the relative difference between SOV and 3+HOV impedance had changed by up to 30 impedance
20、units in favour of HOV in some cases. The “typical” change in relative impedance was in the order of ten impedance units.For some cells, the reduction was entirely due to a relative reduction in the HOV generalized cost generated by the auto assignment. For other cells, with significant increases in
21、 employment, the reduction was due in part to an increase in the trip end density at the destination zone.We examined the sensitivity of the percentage of work auto person trips forecast to be made by 3+SOV to changes in generalized cost for a typical trip from an origin in an inner A Multi-Level Lo
22、git Function HOV/SOV Model for Greater Vancouver Page 9EMME/2 Users Group Conference, Houston October, 1998Edwin Hull Associates LTD Consulting Transportation Engineerssuburb to downtown Vancouver. When the generalized cost for 3+HOV, for 2HOV and for SOV are all equal, the effects of the various we
23、ights, coefficients and biases is such that the percentage of auto person trips forecast to be 3+HOV is approximately 6% (equivalent to approximately 2% of vehicle trips).The forecast percentages of auto trips by 3+HOV for generalized cost differences ranging from +10 impedance units to 10 impedance
24、 units are shown in Figure 3.Figure 3It can be seen that a 10-impedance unit reduction in the relative generalized cost of 3+HOV compared with the respective generalized costs of SOV and 2HOV would increase the forecast 3+HOV from 6% of auto person trips to 14% of auto person trips. The generalized
25、cost of travel for each sub-mode is the sum of the travel time by that mode and the share of vehicle operating costs and other out-of-pocket costs incurred by each vehicle occupant.A 10-minute reduction in 3+SOV travel time compared with SOV time would be sufficient to effect a 10-impedance unit red
26、uction in comparative generalized cost. To achieve the same effect through increased vehicle operating cost would require an increase of $1.90 per vehicle trip. The observed mean trip length for AM peak hour work trips is 13.7 km. To achieve the required cost increase per trip would require an increase in vehicle operating cost of approximately 14 cents per km almost 2 times current costs. This would Sensitivity of P(3HV) to Changes in Generalized Cost0.000.050.100.150.20-10 -8 -6 -4 -2 0 2 4 6 8 10Change in Generalized Cost UnitsP(3HV)
