Dahomey land transport study models

:j.- This paper is prepared for staff use and is not for publication. The views expressed are those of the author and not necessarily those of the Bank. INTERNATIONAL BANK FOR RECONSTRUCTION AND DEVELOPMENT INTERNATIONAL DEVELOPMENT ASSOCIATION Economics Department Working Paper No.87 Dahomey Land Transport Study Models September 4, 1970 This is the sixth in the series of Transport Planning Models Study papers. The Study, directed by Messrs. Jan de Weille and Leon H. Miller, is a continuing investigation of mathematical models developed for transport planning. Existing transport models are being analyzed, and revised and extended where practical. New models will be developed where needed. Eventually, the Study will include cases of models' application in specific transport planning studies and a critical review of the methodology. The present paper presents the logic and concepts of the models used by N. D. Lea and Associates Ltd. and Lamarre Valois International Limitee in the Dahomey Land Transport Study, and discusses some of the problems encountered in the application. The transport network model (Trans) is essentially the same model described. in Economics Department Working Paper No. 61, Transport Network Model, while the highway model, with minor modifications, is that described in Working Paper 62, Highway Cost Performance Model. The optimization program (Opt) is a new model, designed to determine the best construction and maintenance level and to minimize the total construction, maintenance and vehicle operating costs for a transport link. The DahoXey Land Transport Study was carried out between 1967 and 1969 by the Lea and Lamarre firms, under contract to the Bank, for the UNDP and the Government of Dahomey. The Lea firm produced a supplementary study in February 1970, Trans-Opt System, presenting the methodology used for the Dahomey study. The paper here presented was adapted from the methodological study by Mr. Leon H. Miller and edited by Mrs. Judy Mijares and Mrs. Suzy Henneman. Copies of the Trans-Opt System, with detailed technical information for users and programmers, can be obtained from the Sector and Projects Studies Division. Sector and Projects Studies Division  TABLE OF CONTENTS Page No. I. CHOICE OF MODELS . . . . . . . . . . . . . . . 1 Purpose and Criteria ........ . . . . 1 Harvard Transport Model ............ .. 2 Adaptation of Harvard Transport Model to Dahomey Study The Trans-Opt System . .......... . . 6 II. THE TRANSPORT SIMULATION MODEL (TRANS) 8 . . .*.e.* . 8 General Characteristics .. ..... ... . 8 Data Inputs . . . . a . . . . . . . . e 9 General Operational Stages 11 The Operational Steps . . . . . . . . . . . . . . . . 14 III. THE LINK OPTIMIZATION PROGRAM (OPT) 22 GenerarlCharacteristics ............ 22 Data Inputs . . ........ ............. .... 23 General OperationalStages.. ........ . . . . . . 25 TheOperationalSteps ..... . . . 29 IV. EVALUATIONOFOPTOUTPUTS ............................ 38 AlternativeRoutes ........... . . . . . . . . . . . 38 NewFacilities . . ..... ............ 39 Calculation of Benefits for Individual Improvements . 40 DivertedTraffic.... . . . . . . . . . . . . . . . 42 TheOptimizationProcess. . . . . . . e . . 42 V. SIMULATIONPROB EINDANODAM OMEY.........e.. 46 Intrinsic Simulation Problems . . . . . * . . . . . . 46 DataCollection Problems.. ..... . . . . . . . 46 ProgramLimitations ................. O49 CalibrationProblem.o... .. 50 APPENDIX: GLOSARY OF TRW 1 -3 TABLE 1. Dahomey: Highway Maintenance Costs . . . . 26 Figures 1. Relationship of Harvard Transport Model and Trans .. . 3 2. Trans Flow Chart . . .................... . 13 3. Opt Flow Chart . . ... . .............. 27 4. Standard Growth Curve Procedures ..................... . 31 5. Volume Determination with Improved Transport Network . . . 33 6. Section of Transport Network . . . . . . . . . . . . . . . 39 7. Illustration of Optimization .. . .. .*..... . e 43  DAH(MEY LAND TRANSPORT STUDY MODELS I. CHOICE OF MODELS Purpose and Criteria 1. The Trans-Opt system described in this paper is partly new (the Opt model) and partly derived from models developed at Harvard Universityl/ (the Trans model). The system was applied by its authors, N.D. Lea and Associates Ltd and Lamarre Valois International Limit&e, to the transport system in Dahomey during the period 1967-1969. The work was carried out under contract to the Bank for the UNDP and the Government of Dahomey. 2. The major object of the Dahomey Land Transport Study was to deter- mine, under the existing and forecast socioeconomic conditions, the adequacy of the existing road and rail transport system, and to recommend on the basis of economic analyses a program of physical and operational improvements to the system. Because of the complex nature of the analyses and the large number of possible improvements to be considered, a computer program or package of programs was sought which could :(a) simulate the traffic which would use a given transport system under given socioeconomic conditions; (ii) analyze the costs and benefits for the country as a whole which would result from a set of given improvements; and (iii) select from many alternative potential impzovements those which would provide the greatest net economic benefits to the community. 3. A number of computer simulation programs for distributing and assigning traffic over a transport network were considered. However, most of these programs simulated urban conditions: they distributed trips split into a limited number of categories and assigned them to a trans- port link-node network, often divided into independent systems represent- ing different modes (e private transport, public transport and some- times goods transport). Most performed the distribution and assign- ment on the basis of minimum time paths, though some allowed inclusion of out-of-pocket costs with perhaps a simple relationship between vehicle operating cost and speed. They did not produce outputs suitable for 1/ Components of the Harvard transport model are described in Economics Department Working Papers 60 through 64h Regional Macroeconomic Model, Transport Network Model, Highway Cost Performance Model, Railroad Cost Performance Model, and Transfer Cost Performance Model 2/ Te transport and systems terms used in this paper are defined in the appendix, "Glossary of Terms."  - 2- economic analyses of alternative transport improvements. Most simulated one time period (ega. peak or off-peak periods, or 24 hours), and would have to be repeated to simulate more than one period. Harvard Transport Model 4. Only one computer model known at the time was specifically designed to simulate a whole country on an economic basis for each season of a year, and would handle both freight and passenger flows of many different kinds by several different types of vehicle, distribute and assign them to a single network including different categories of links (road, rail, etc.), and provide link and system outputs in terms of both freight or passenger flows and vehicles. and in terms of both financial and economic costs. This was the model developed at Harvard University and previously applied to an economic and transport study of Colombia. We refer to it as the Harvard transport model. 5. The Harvard transport model is composed of a macroeconomic model, a network model, and three modal cost models (See Fig. 1). The macro- economic model is a complex program which simulates the entire economy of a country on a regional basis for a year at a time. It requires conm)rehensive input data, including a complete nhtional input-output table relating the amount of each product produced to the amounts of other materials used in its manufacture. It provides tables of annual supply and demand of groups of commodities in each region, based on private consumption, plant and inventory investment, government expen- ditures, exports, and the input-output table; information from the previous yearts simulation is used in some of the calculations. 6. The network model takes the regional supplies and demand by industry calculated by the macroeconomic model and breaks them down into supplies and demands of individual subcommodities by node and by season of the year. It then distributes the seasonal supply and demand of each subcommodity andassigns it to the trarnport network, on the basis of minimum cost routes as perceived by the shippers. After all commodities are assigned, cost-performance measures are calculated for each link- and totaled for the whole transport system. Costs are revised according to pricing policies, supplies and demands are reaggregated, and the cost-performance measures are stored for use in simulating the next year. The highway, rail, and transfer models calculate and feed into the transport model information on link cost and performance. 7. Seven features of the network model influenced the decision to use it, subject to certain modifications, for simulation of traffic flows in the Dahomey Land Transport Study. First, up to 40 different commodities  -3- FIGURE 1: RELATlONSHI P OF HARVARD TRANSPORT MODEL AND TRANS HARVARD TRANSPORT MODE-L MACROECONC 1 MODEL (Working Paper No. 60) HIGHWAY COST __ _ _ _ MODEL (Working Paper No. 62) NETWORK-. - MODEL RAIL COST MODEI (Working Paper (Working Paper -No. 63) No. 61)I _ _ _ _ _ I TRANSFER COST MODEL (Working I Paper No. 64) Rail Cost Transfer I Look-up Cost Look-up Routine Routine '-- --2ANS MODEL -  may be handled; one or more categories of passenger trip as well as different types of freight can be represented as commodities. Distributions and assignments are performed in terms of units of each commodity rather than by vehicles. 8. Second, up to five different vehicle classes can be defined, and each commodity is assumed to travel by one of these classes. The number of vehicles of each class required per day is determined for each link by the total assigned flow of commodities which use that class. In this way, similar commodities, such as vegetables, sacks of grain, other produce and general merchandise can be simulated as sharing the same vehicle, whereas dissimilar commodities such as bulk petpoleum, bulk shipments of ore or rock, and passengers will be simulated as requiring different kinds of vehicles. 9. Third, the transport system of the region or country being studied is represented by a single network of nodes and links, which includes all available modes, i.e. types of facility, such as road, rail, air and water, and transfer facilities between modes. 10. Each link of the transport system is described in terms of a number of characteristics, from which operatiing costs and other performance measures such as travel time, delay time, probability of loss and variabi.lity of travel time can be calculated for each vehicle category. For each commodity, different values can be specified for each measure of performance: for instance, a high value of delay time can be specified for perishable produce. The quantities of the commodity requiring transportation are distributed among appropriate origins and destina.tions, and then assigned to the network, using routes along which the total perceived cost of vehicle operation, of travel time, and of the other performance measures is minimized. 11. The performance of a vehicle on a link is simulated on the basis of vehicle load and operating characteristics, link characteris- tics and the amount of other traffic on the link. As an example, the characteristics of a road link which influence vehicle performance can include such items as rate of rise and fall, design speed, type of construction aud condition of surface. Any of these characteristics on any number of links can be changed, and the effect on flows through- out the system determined by rerunning the program. In general, any commodity can travel on any link of any mode, although travel by a mode or by a group if links within a mode can be suppressed if desired. 12. Fourth, the program operates on modal supplies and demands of each commodity for a particular year. The distribution and assigmment routines operate in terms of daily flows in each season. As well as  operating for a past year on historic data, the program can be run for a future year using forecasts of supplis and demards and of transport system characteristics. 13. Fifth, the available output includes tables of origin-destination (O-D) movements in each season to calibrate against O-D surveys; daily commodity and vehicle flows by class and season on each link for calib- ration against vehicle counts; and aggregate system performance measures for each season, from which can be determined such items as total number of vehicles required, vehicle operator costs and profits, and government tax revenues. 14. Sixth, it is possible to study the consequences of various policy changes, such as revision of rate structures on any mode; programs of improvements to existing links or construction of new links; changes in vehicle size or weight restrictions; changes in taxation, etc. The effects of such changes on transport movements as a whole are reflected in the aggregate system performance measures. 15. Seventh, two distribution routines are available: a gravity model formulation which is suitable for passenger trip distributions, in which all supply nodes ship to all demand nodes, and a linear programming routine which is suitable for distribution of individual commodities from their production points to locations where they are consumed or reprocessed: each supply node of al commodity supplies a limited number of demand nodes, so that the total perceived cost of distribution is minimized. The program also provides for insertion of additional distribution routines if desired. Adaptation of Harvard Transport Model to Dahomey Study 16. Consideration was given to using the entire Harvard package in the Dahomey Land Transport Study in order to select worthwhile packages of improvements to the transport system. However, the amount of exist- ing Dahomey information of the type required by the entire Harvard model was very small, and collection of the necessary data would have been expensive and time-consuming. Detailed studies would have had to be ,d.one on all aspects of the economy, and input-output factors would have had to be determined for each industry using information on raw material consumption and commodity production. 17. The macroeconomic and network models are run in combination for each year of the study period. The 24-year period covered by the Dahomey Study (1967 to 1990 inclusive) would have required running both models 24 times for any particular set of conditions; for each alternative set of transport improvements being evaluated the models would have had to be rerun for all years following the year of the first improvement. This procedure would have required an inordinate amount of computer time and expenditure, as both models are quite large and complex, and there were many alternative improvements to be considered. In addition,  -6- the macroeconomic model had not yet been successfully calibrated in the Colombian Study, so there was no guarantee of the accuracy of its simulation procedures. 18. Because of limitations of time, cost and data availability, the macroeconomic model of the Harvard package was not chosen. But the desirable features of the network model led to the decision to use it. The package consisting ofe the network model, a modified highway cost model, and two look-up routines indicating unit costs for rail and transfer operations was given the name Trans (Transport Simulation). Inistead of running the model for each of 24 years, it was decided to run it for two points in time only: a base year, the most recent year for which data were obtainable (1967); and a future year for which the required inputs would be predicted (1975). 19. An entirely new link optimization program., Opt, was devised to interpolate and extrapolate costs and volumes for other years and to perform link-by-link economic analyses of the alternative transport improvements, because there was no known existing program which could perform this type of operation. The Trans-Opt System 20. The Trans-Opt system, then, is a package of two self-contained computer programs: a transport simulation program adapted from the Harvard package and a highway link optimization program. They are designed for use together, to analyze various alternative courses of action in developing an existing transport network, and to select a combination of capital and maintenance improvements that will give rise to a minimum present value of all incurred costs over a given period. The programs may also be used independently. In particular, the Trans program can be used to examine the transport system in any particular year and test the effects of various taxation, pricing and regulatory pol.icies. The Trans program has the same concept and design as the transport network model of Working Paper 61 plus the highway submodel described in Working Paper 62 and the two look-up routines; the Opt program is unique. 21. The inlput information required for a Trans run consists of a link-by-link description of a link-node transport network representing the physical transport system, with information for calculating transport costs and performance, the taxation, pricing and regulatory policies, plus a list of seasonal supply and demand quantities by node for each commodity, i.e. freight categories and passengers. The  -7- major output item is a table of assigned daily commodity flows by each link of the network. 22. For a run of Opt, the input required includes the assigned link flows produced by Trans, and descriptions and costs of the alter- native improvements to the links under consideration. Opt analyzes each link pair in the network and selects the investment alternative which will minimize the present value of total cost for the link pair. The selection is made on the basis of traffic volumes derived from the input. Although Opt can calculate costs for ewxisting standards on link pairs of road, rail and transfer modes, it can analyze alternative improved standards only for road link pairs. It can also select an optimum standard for proposed new road links. 23. The Trans and Opt models are described separately in detail in the next chapters.  -8- II. THE TRAISPORT SIMIUTION MODEL (TRANS) General Characteristics 1. Trans is a mathematical model which can simulate the movement of goods and/or passengers on a multimodal transport system for each season of a given year. It is in the form of a computer program which can presently be run on either an IB 7094, a UNIVAC ll08 or a Burroughs 5500 computer. 2. Each commodity is assigned to on-a of the five vehicle classes. For each season, each commodity, through its vehicle class, is assigned to the network along paths of minimum cost, as perceived by the shippers, by one of two distribution routines. When all the commodities have been assigned to the network, the costs, times, and other transportation performance measures are calculated for each vehicle class on each link in each season. The available output includes origin-destination tables of costs, flows, etc,, by commodity in each season.; costs, flows, etc. in each season on each link of the system, and aggregate systems per- formance measures by season. 3. The Trans program as used in the Dahomey Study has certain capacity limitations. It can handle up to: 300 one-way links 149 nodes 40 commodities 20 supply nodes per commodity 40 demand nodes per commodity 10 mode-submode combinations 5 vehicle classes 4 seasons Some of these limits could be extended by means of programming changes. 4. Before the Trans model can be used it has to be calibrated to ensure that it ca-n in fact accurately simulate the transportation system. This is done by iteratively running the model using data for current conditions, and comparing the resulting flows and performance data with those actually measured in the field. After suitable programming and data modifications have been made, the model is ready to forecast future traffic on the basis of projected changes in supply and demand patterns and in the transport system characteristics. 5 The Trans model can be used in one of two ways. First, it can be used on its own to analyze the transport situation in one specific year or to give a better understanding of the interaction between the various parts of the system and to point out areas for improvement.  -9- Second, Trans can be run as part of the Trans-Opt System, the output from Trans for different years and networks being used as input to the Opt program, which determines an optimum network. Data Inputs 6. The information required to run the model for a particular country or region consists in a description of the transportation network, a description of the vehicles operating on the network, a description of the commodities to be moved on the network,and a series of equations and parameters relating the performance of the vehicles to the characteristics of the network and to the amount and type of traffic using the network. 7. The network is described in terms of links and nodes; the links represent transportation routes - road, rail, air, etc., and intermodal transfer facilities - and the nodes represent points where two or more routes join and points where traffic is generated or attracted. The links are described by their characteristics -mode, length, rise and dall, etc. The vehicles are divided into five classes which are described by characteristics such as weight, capacity, power, etc., as they relate to each mode considered. The commodities are describ.ed by their transportation preference characteristics, i.e., the relative importance of travel cost, travel time, probability of loss, etc., and by the nodes where they are supplied and the nodes where they are demanded. Passengers are considered as one or more separate commodities. 8. The data required to ruin the Trans program are divided into ten categories or data types; for the first run of a study all data types must be supplied but in subsequent runs only the input data types which differ from the previous run need to be specified. The highway mode is the only mode for which a separate cost model is used (data type (4)); rail and transfer link costs are determined in a simpler way, described further below. 9. The ten data types are: (1) General Data. Information regarding the input of other data types and the optional outputs is required; also basic dimensions for the program, including number of commodities, number of nodes and links in the transport network, number and lengths of seasons and number of vehicle categories. (2) Node Names. Each node of the transport system may be given a code name if desired.  - 10 - (3) Preliminary Link Performance Measures. In the first of a series of iterations, approximate unit values of travel cost and up to four other performance measures (travel time, waiting time, variability of travel time and probability of loss) are required; in subsequent iterations, performance measures calculated on the basis of traffic assigned to the network in the previous iteration are used. (4) Highway Cost Model Data. Items required include characteristics of up to five classes of highway vehicle, such as the weight, horsepower, carrying capacity, vehicle cost, unit costs of fuel, oil, maintenance, tires, etc.; also parameters for use in equations relating vehicle speed and fuel consumption to rate of rise and fall., speed to volume and capacity, and operating costs to condition of road surface. (5) Commodit