Atlanta Regional Commission (ARC) Traditional 4–step ...



University of South FloridaAtlanta Regional Commission (ARC) Traditional 4–step Travel Demand ModelTerm Project – Travel Demand Modelling (Fall ’13)Menon, Nikhil & Kamali, Mohammadreza12/9/2013MEMOTo: Dr. Abdul PinjariFrom:Nikhil Menon & Mohammedreza KamaliDate:December 9th, 2013 Term Project: Atlanta Regional Commission (ARC) Traditional 4–step Travel Demand ModelThe research paper was completed by Nikhil Menon and Mohammedreza Kamali. Nikhil Menon contributed to the first section of the paper (Introduction, Trip Generation & Trip Distribution) while Mohammedreza Kamali has contributed to the second section of the paper (Mode Choice and Network Assignment). The formatting has been done by Nikhil Menon. TABLE OF CONTENTS TOC \o "1-3" \h \z \u MEMO PAGEREF _Toc374393799 \h 11INTRODUCTION PAGEREF _Toc374393800 \h 32OVERALL MODEL STRUCTURE PAGEREF _Toc374393801 \h 43MODEL COMPONENTS PAGEREF _Toc374393802 \h 43.1TRIP GENERATION MODEL PAGEREF _Toc374393803 \h 43.1.1TRIP PRODUCTION MODEL PAGEREF _Toc374393804 \h 53.1.2TRIP ATTRACTION MODEL PAGEREF _Toc374393805 \h 73.2TRIP DISTRIBUTION PAGEREF _Toc374393806 \h 93.2.1MODEL STRUCTURE PAGEREF _Toc374393807 \h 93.2.2MODEL APPLICATION PAGEREF _Toc374393808 \h 103.2.3INFERENCES PAGEREF _Toc374393809 \h 103.3MODE CHOICE PAGEREF _Toc374393810 \h 103.4TRAFFIC ASSIGNMENT PAGEREF _Toc374393811 \h 144CONCLUSION PAGEREF _Toc374393812 \h 165BIBLIOGAPHY PAGEREF _Toc374393813 \h 16INTRODUCTIONThe Atlanta Regional Commission (ARC) travel demand model is designed to represent the state of the practice in travel demand modelling and conform to the requirements in the US EPA Transportation Conformity Rule (40 CFR Parts 51 and 93). As a result, all elements of the travel demand model are designed to support and complement all technical and policy decisions that are required towards developing a comprehensive, multimodal transportation plan, in accordance with the Intermodal Surface Transportation Equity Act of 1991 (ISTEA), the 1990 Clean Air Act Amendments (CAAA) and the Transportation Conformity Rule and the 2005, Safe, Accountable, Flexible, Efficient Transportation Equity Act: A Legacy for Users (SAFETEA-LU). This report looks into the trop based travel demand model developed for the Atlanta region. This trip based travel model encompasses several sub models which estimate the travel patterns throughout the region and then, these travel models are used to estimate the travel on the highway and transit transportation links within the Atlanta region. The model boundary originally consisted of 13 counties in order to be consistent with air quality conformity requirements, but was later extended to 20 counties in order to sustain the analysis towards conformity. Seven more additional “ring” counties along with the pre-existing 13 – county one hour ozone maintenance area received nonattainment under the fine particulate matter (PM2.5) standard. Therefore, it was felt that the number of counties undertaken for this study needed to be 20, and not 13. The model did not include the two partial counties of Heard and Putnam.The main source for the model calibration was a household travel survey of 8000 households conducted by the Atlanta Regional Commission (ARC). The time line of these surveys extended from April 2001 to April 2002. In addition, the ARC had an On-board transit survey conducted in the end of 2001 and early 2002 to aid in the development of these travel demand models. In addition to this, a survey of the air passengers at the Hartsfield International Airport was conducted in late 2001. In order to develop network speeds, extensive speed studies were conducted. This included utilizing over 5000 daily traffic counts to perform the highway assignment validation. A new commercial and truck model was developed as an added endeavor with the collection of vehicle classification counts. The air passenger survey and commercial truck model have not been included in the scope of this term project paper. The household survey was the main source of data for developing two steps of the model – namely the trip generation and the trip distribution model. The model was calibrated using a logit structure and used the ALOGIT program for this purpose. The distribution model was a gravity model which used both the transit time and the highway travel times as measures of travel impedance. The mode choice model also used the household travel survey data, but in addition, it also used the transit on – board survey. The mode choice model is a nested logit model calibrated using the program, ALOGIT. The air passenger survey was used to develop an air passenger model. This model is also a nested logit model which estimates the non – airport location of air passengers and the mode which the air passengers used to the airport. Additional modal refinements have been incorporated in due course into the ARC model – majorly in 2005 – 06, when the mode choice model was re – evaluated to improve the performance of the model performance on suburban intra – country trips. The main rationale behind this was to examine the existing mode choice model and improve its ability to predict travel mode of trips in the Atlanta region. Various model specifications were revised, market segments in the model were revised and an attraction shares model was incorporated into the regional model to enhance the performance of the model. This led to overall improvements in the trip generation, trip distribution and mode choice models. OVERALL MODEL STRUCTUREFigure STYLEREF 1 \s 2 SEQ Figure \* ARABIC \s 1 1 – ARC Model Structure (Rousseau, 2010)MODEL COMPONENTSBeing the traditional four step model, the four major steps involved in the ARC travel demand model are to develop sub models for trip generation, trip distribution, mode choice and assignment. TRIP GENERATION MODELThe trip generation model estimates trip ends for a typical weekday. A production trip end is where a trip begins from the home of the maker and an attraction trip end where the trip ends. This trip generation model for ARC develops trip ends by purpose and is stratified into six purposes:Home Based WorkHome Based ShoppingHome Based Grade SchoolHome Based UniversityHome Based OtherNon – Home BasedVarious sub models adorn the trip generation model – a model to estimate the trip productions by purpose and a model to estimate the trip attractions by purpose. A model to estimate the type of households in each traffic analysis zone and a model to estimate which productions are non-motorized/ motorized. TRIP PRODUCTION MODELThe trip production model was developed by using the 2000 home interview survey. This survey of approximately 8000 households consisted of identifying the travel patterns for each member of the household for two consecutive days. A substantial number of trips were recorded because of this, approximately 100,000. MODEL STRUCTUREThe trip production model was revised from a cross classification into a logit model estimating the daily trip frequency a person (not household) would make. This is a fairly unique model and the reasons for choosing this were as follows:allows more independent variable usage;allows the use of continuous independent variables, than only classification variables;allows statistical measures to be made measuring the significance of the independent variables and the entire equation.The socio – economic independent variables, specified as information for the household, are as follows:HH size (1, 2, 3, 4+)HH income group (under $20K, $20 – 50K, $50K – 100K, over $100K)Number of workers (0, 1, 2, 3+)Number of children (0, 1, 2, 3+)Number of autos (0, 1, 2, 3+)The model is a logit formulation, structured as follows: Where:Px,y(n) = probability of a person of type x making n trips by purpose y, n = 0, 1, 2, ….Ux,y(n) = utility of trips for person type x and purpose y. The utility of making no trips (U (0)) is, by convention, a zero for all models. The utility of the other numbers of trips is a linear function of various known attributes of the household – including income, size, autos, and location. Non home based trips were maintained as a single purpose in the model. A methodology used is to develop a composite time (combination of highway and transit time). The formulation of composite time combines highway and transit traffic times between traffic analysis zone pairs to reflect a single (composite) time. Although it is illogical to do this, it has been found that the ARC model incorporates this assumption into it. The accessibility measures are computed by zone as follows:Where:A(i) = accessibility for zone I; S(j) = a measure of activity in zone j; population, employment or population plus employment & Ti,j = impedance separating zones I & j; either total transit travel time or highway time.This accessibility measure incorporates urban form (number of persons or jobs “close” to a household) and the transit and highway access (more transit and / or faster highways will have more persons or jobs “close” to a household. The highway distance accessibility represents the accessibility for walking trips. The formulation, shown above, is such that the variable is continuous but produces very low values for high travel times.MODEL APPLICATIONThe ARC model was then applied to determine the percentage of automobiles in accordance to the pre – defined income groups along the summary level TAZ. This was compared to the CTPP households and the observations are as given below: Figure STYLEREF 1 \s 3 SEQ Figure \* ARABIC \s 1 1 – ARC Trip production model application – Automobile infusion into income groups. (Source: ARC 2010)INFERENCESIn conclusion, the ARC model has shown that it is possible to extend trip – based trip generation models beyond the standard income/ size based classification approach. The logit model enables to consider many relationships that would have otherwise not been possible to be tracked. This is seen as a suitable mid-point between trip based classification models and the newly developed tour – based micro simulation approach. It is also found to be less time consuming. The results have not shown many surprises – Income is an extremely important determinant in household trip making. All the trip purposes also show a decreasing influence on higher income on non – motorized trips.However, it has shown that the number of children in a household is not a strong determinant – it doesn’t matter if the children are one or four. This, although fundamentally incorrect has come up in this manner primarily due to the fact that the total size among adult trip makers had been replaced by “number of other adults” and “presence of any children” – which negated the effect. Car ownership once again proved to be the deciding factor in bringing out the effect of “induced travel”. Induced travel effects come directly from households located in areas where walking and transit are not found to be viable options and therefore, leading to increased car ownership & greater number of trips. The accessibility variables were found to be too statistically weak to show any effects on trip making. Propensity to travel was found to be more in high density areas than the low density areas. When it came to non – motorized travel, the ARC model showed that it is possible to estimate the proportion of non – motorized trips using information on the wealth of the traveler, the urban form in the terms of density and the availability of opportunities in terms of the number of jobs and people within a reasonable distance. Since the provision of school buses is a local policy decision, no mathematical model was applied to determine its usage by the ARC. TRIP ATTRACTION MODEL The trip production model used the home interview survey performed in 2000. ARC staff conducted an Establishment survey where both employees and visitors were surveyed. This data enabled the creation of cross classification attractions models instead of the regression models normally developed using the home interview data. All the four models developed have a basic structural 2 – way cross classification matrix of trip rates per employee, per person or per household. The four trip purposes for which the cross classification matrix has been implemented are as follows:Home – based Work (HBW)Home – based Shop (HBS)Home – based Other (HBO)Non – home – based (NHB)As per revisions in 2005, an attractions share model was incorporated into the existing attraction model. The purpose of this model was to determine zonal attractions by the socio – economic groups based on their highway and transit accessibility. Trips are attracted not only to non – residential establishments, but also to residences within the study area. Thus a correct balance can only be achieved by combining the information from the Establishment survey with that of the home interview survey. MODEL STRUCTUREBy looking at the SMARTRAQ household data, the following four strata were implemented:Households without cars. (Strata 1)Households with the number of cars < the number of workers (Strata 2)Low income households with # cars >= # workers (Strata 3)High income households with # cars >= # workers (Strata 4)The trip attraction share model used the SMARTRAQ data for estimation. This home interview based survey was performed in 2001. This model implemented a multinomial logit model to split total attractions into attractions for each stratum. A separate model is applied for each purposeFigure STYLEREF 1 \s 3 SEQ Figure \* ARABIC \s 1 2 – Utility variables and their influences on trip attraction (Source: ARC 2010)Coefficients for the attraction shares logit equations were derived by taking the natural logarithm of the relative shares instead of going through the logit estimation process. The total employment independent variable is calculated as the minimum of 10,000 or the total employment divided by 10,000.INFERENCESThe attraction share model utility coefficients have been obtained and are as shown below:Figure STYLEREF 1 \s 3 SEQ Figure \* ARABIC \s 1 3 – Attraction share model utility coefficients (Source: ARC 2010)The attractions at the socio – economic level are further classified into motorized and non – motorized trips. If non-motorized production trips is less than or equal to total attraction trips in the zone, then non-motorized attraction are set to be the same as non-motorized production and vice versa.A comparison of the number of the zero household auto trips between the CTPP 2000 and the model revealed that the model underestimated these trips. An adjustment factor was then computed to resolve this anomaly. TRIP DISTRIBUTIONThe initial challenge in the development of the trip distribution model was to determine the need or not for separate distribution models for individual income groups within each of the home based trip purposes. Another challenge was the development of a new spatial separation variable, like the composite time variable, which was done during the trip generation stages. The same procedure has been adopted here as well. The stratification of market groups were done majorly on the basis of two criteria – sufficiency of the samples and the average trip length. The formulation of the composite time includes combining of the highway and transit times in such a way that they receive equal weightage. MODEL STRUCTUREThe trip distribution model utilizes two factors as the primary reasons which travelers use in making a decision concerning a trip from a particular origin to a particular destination: The relative number of destinations at a particular destination traffic analysis zone;The spatial separation between the traffic analysis zones.The trip generation model has already made the decision regarding the number of destinations in a particular destination TAZ, leaving only the spatial separation parameter to be unique to the distribution model. The composite time was developed using the following relationship:Where:CT = Composite Time (mins)HT = Highway Travel Time (mins)TT = Transit Travel Time (mins)X = Weighing factor varying by income group (Group 1: 0.5; G2: 0.12; G3: 0.06; G4: 0.03)The gravity model used the 2001 – 2002 Atlanta Regional Commission’s (ARC) household travel survey data. The home – based work trips were stratified and reviewed. Highway and transit times were combined using the composite time files where different factors were used to weigh the contribution of transit travel time in the calculation of composite time. The calibration process was for a standard gravity model based on the 2001 – 2002 survey trip tables. This involved adjusting the friction factors obtained in the gravity model until the computed trip lengths were inside 3% of the average trip length observed in the survey data. The ARC model addresses the issue of area bias through a “topo” penalty. “Area” bias is more significant when natural features are combined with political and/ or service delivery boundaries. In this case, area bias arises is formed by the Chattahoochee River. The “topo” penalty is a lump sum of travel time (in mins) added to the composite time for all zonal pairs on either of the Chattahoochee. MODEL APPLICATIONThe ability of the calibrated gravity model to replicate observed travel patterns was checked by a comparison of the county-to-county work trips between the gravity model, the 2001-2002 Household Travel Survey (O-D), and the 2000 Census Journey-to-Work data.INFERENCESWhen the “topo” penalties were added to the composite times, the results thrown by the gravity model seem to have greatly improved.When compared to the OD survey and 2000 Census, it was shown that the gravity model provided a reasonable estimate of trips for the individual counties. Where there were differences, it was realized that slight adjustments towards intra – zonal travel times within the counties would largely reduce it.MODE CHOICEIn this step the ARC estimates the probability that users will choose different modes of travel in the Atlanta region. To achieve this goal, a nested logit structure was employed. All the variables and their coefficients were estimated with the software named ALOGIT using input data from the household travel survey. Later on, a regional travel on-board survey was conducted by the ARC to understand better the travel patterns. Some of the inputs from this later survey were used for u[dating and calibrating this model. Below is the nested logit model structure:Figure 3-4 – ARC Model mode choice nesting structureAs mentioned before, an expansion of the preliminary survey was conducted and initial data from the expanded survey were used for some adjustments. These adjustments were implemented in transit targets by trip purpose, mode of access and socioeconomic classes. Below are the results of the survey expansion and its downward effect on the model:Figure STYLEREF 1 \s 35 – Adjustments for Year 2005.It is worth explaining the terms “Kiss and ride” and “Park and ride”. Kiss and ride mode of access is when somebody drops off the passenger at the transit station that can be a subway station or a bus stop. Park and ride is the case when the passenger drives to the station, parks her car and then takes the train or bus.Furthermore, it was concluded that the minimum walk distance to a transit station was 1 mile and this distance was assumed to be a straight line from the zone centroid to the transit stop. The figure below shows the cumulative walk access distance and their share of the transit market:Figure 3-6 – Cumulative Walk Access DistanceThe next step in constructing the mode choice model was doing some calibrations including adjusting some bias coefficients. An iterative process was employed by comparing the estimated shares with the observed shares and adjusting the bias accordingly. These adjustments include Transfer Penalty, Pedestrian Environment Factor, CBD Constant for Walk Access and Drive Access Modes, and Alternative Specific Constants.In the next step, the model was constructed. First of all, we have to explain some of the mathematics behind a nested logit model. The following scenario depicts a nested structure containing three modes of transportation: car, bus, and metro. It is assumed that bus and metro are correlated so they will be considered at the lower nest.Figure 3-7 – Nested Logit Model in a Trinomial Model Choice SituationThe lower public transport nest would be modelled by a simple binary logit model:PMPT=expWMexpWM+expWBAndPBPT=1-PMPTThe utilities W include only those elements that are not common to both bus and metro. Another binary logit model at the higher nest between car and the composite alternative PT is modelled as: PC=exp?(VC)expVC+exp?(VPT)AndPPT=1-PCIn this step VC includes all the attributes of the car choice. The public transport utility will be:VPT=?EMU+kαkZkAndEMU=ln?(expWB+expWM)Finally the probabilities are:PC=PCPB=PBPTPPTPM=PMPTPPTThis was a brief and small example of a nested logit model which is different from the model at hand in ARC Travel Demand Model. However, they both follow the same pattern. The following table shows the nested logit model developed by ARC. Figure STYLEREF 1 \s 38 ARC Nested Logit ModelAfter developing the model, it was applied to the data base to estimate the mode choice. To compare between the model results and actual observations, MARTA rail system was chosen and daily boarding on this system was considered for comparison. The following table and figure shows this comparison.Figure STYLEREF 1 \s 39 – Observed and Estimated Daily Weekday Boardings on Marta Station.Figure STYLEREF 1 \s 310 – MARTA Rail Station Entries Observed vs. Estimated Trend LineTRAFFIC ASSIGNMENTAfter taking the first three steps, the assignment model and highway assignment validation had to be done. To this end, daily trips were split into four time periods – a morning peak (6am-10am), a mid-day period (10am-3pm), an evening peak (3pm-7pm) and an evening/night period (7pm-6am). Separate highway assignments are made for each time period. The highway assignment procedure employs a standard equilibrium technique to assign the trips throughout the network. This equilibrium assignment includes iterative series of all-or-nothing traffic assignment technique with some adjustments. Five different vehicle categories were considered for the traffic assignment: heavy duty truck, medium duty truck, commercial vehicles, single occupancy vehicles (SOVs), and high occupancy vehicles (HOVs). Moreover, some modifications have been done to the model like passenger car equivalency for the influence of trucks.To proceed with constructing the model several models had to be calibrated. First of all, Time-of-Day model was calibrated to show the percentage of travel made in the four times of the day. Moreover, it was necessary to understand how travel times increase as roadway traffic reaches and goes beyond its capacity. A feature in the ARC Travel Demand Model is that a feedback component was added to the model to make sure that the travel times resulting from assigning the traffic were similar to those used for trip generation. To achieve this, the travel time resulted from assigning the volumes to the links were fed back to the first step, which is trip generation, and the whole model was run again, this process continued until the travel times converged.Continuing with the model development, highway trips were assigned to the network using initially estimated volume delay functions. The following table shows one of the validation checks for the model.Figure STYLEREF 1 \s 311 – Percent Root Mean Square Error Comparison by Area Type for the 2000 Atlanta Regional ModelThe above table shows the error estimates for ARC are comparable with those for models used in urban areas. The following graph shows a link specific check for the model. The coefficient of determination is shown on the graph and its value is 0.97 which is well above the target value 0.88 meaning that the model is estimating with satisfying precision. Figure STYLEREF 1 \s 312 – Atlanta 2000 Travel Demand Model Scatter PlotCONCLUSIONThe ARC Traditional 4 step travel demand model has been studied and the term paper summarizes the procedure and findings in very brief detail. It has been observed that the model developed for the Atlanta region showed reliable estimates in comparison with the observed data viz. OD surveys, other comparative metropolitan travel models (like BMC 2000) during the various stages of the traditional 4 steps. BIBLIOGAPHYAtlanta Regional Commission (2010), The Travel Forecasting Model Set for the Atlanta Region 2010 Documentation ; Rousseau, G (2010), Atlanta Travel Forecasting Methods – Traditional Trip based & Activity based models; of Civil Engineering, Indian Institute of Technology, Mumbai. Nested Logit Model; ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download