TRANSPORTATION. Ofyar Z. TAMIN

Transcription

1 TRANSPORTATION Ofyar Z. TAMIN The paper is written based on research on Dynamic Origin-Destination Matrices From Rea Time Traffic Count Information competed in March The atest deveopment in automatic traffic count data coection enabes us to obtain the traffic count information in rea time or short-timeinterva basis. For exampe, ATCS Area Traffic Contro System) aready instaed in severa arge cities in Indonesia, such as: DKI-Jaarta 1994), Bandung 1997), and Surabaya 1998) provided us the short-time-interva traffic count information for a signaised intersections. This traffic data is updated periodicay in a short-time-interva basis e.g., 5, 15, or 30 minute time interva). This information is provided at the Traffic Contro Centre TCC) of ATCS project and can be directy and easiy accessed at a very ow cost through the Internet or a teephone ine. This data is the main input for the short-time-interva OD matrix estimation. Before this traffic data is used in the OD matrix estimation process; firsty, these data have to be processed in the Data Processing Interface DPI). Having it processed; the traffic data wi then be ready for estimating the short-time-interva OD matrices. The output of short-time-interva OD matrices together with severa practica appications wi be the main input for the Rea Time Integrated Traffic Information System RITIS). This information wi be stored in a Website designed specificay and informativey for the purposes of user needs numerica and graphica). This short-time-interva traffic system information wi become the pubic-domain information which can be directy and freey accessed through the Internet by the users e.g., panning authorities, traffic authorities, Department of Pubic Wors, consutants, poice, drivers, radio stations, and TV stations, and other reated agencies, etc.). One of the most important information is the best routes from each origin zone to each destination zone which have aready considered the effect of congestion. This information wi be the main data for the deveopment of the Route Guidance System RGS) so that each driver can choose his best route through the road networ. The best route information wi be changed in a short-time-interva basis depending on the traffic condition. Moreover, this approach can aso be extended to provide the short-time-interva environmenta information. The system has been tested and vaidated in Bandung and it showed remaraby good resuts for Bandung condition. Route guidance, Inteigent transport system, Urban transport, OD matrix, Rea time traffic information As reported by Abidin 1, some papers concuded that most deveoped countries have faced osing biions of doars in transportation just due to drivers not having enough prior information of what they thought was the best routes. Besides that, traffic congestion is bamed as the main factor which resuts in the oss of wor productivity in USA around 100 biion doars per year. It is aso said that in 1991 there has been 40,000 fata injuries in traffic accidents and more than 5 miion injured. Moreover, the traffic accident has aso contributed to osses as much as 70 biion doars per year. Some other quaitative disadvantages are: deay, inefficient traffic movements, high fue consumption due to congestion, air and noise poution. Trave is an activity that has become part of our daiy ife and the demand for it aways presents probems especiay in urban areas such as congestion, deays, air poution, noise and environment. In order to aeviate these probems, it is necessary to understand the underying trave pattern. As mentioned by Tamin 2, the notion of Origin-Destination OD) matrix has been widey used and accepted by transport panners as an important too to represent the trave pattern. When an OD matrix is assigned onto the networ, a fow pattern is produced. By examining this fow pattern, one can identify the probems that exist in the networ and some ind of soution may be devised. An OD matrix gives a very good indication of trave demand, and therefore, it pays a very important roe in various types of transport studies, transport panning and management tass. Most techniques and methods for soving transportation probems urban and regiona) require OD matrix information as fundamenta information to represent the transport pattern. The conventiona method to estimate OD matrices usuay requires very arge surveys such as: home and roadside interviews; which are very expensive, engthy, abour intensive, subject to arge errors, and

2 THE DEVELOPMENT OF THE REAL TIME INTEGRATED TRAFFIC INFORMATION SYSTEM RITIS) FOR INDONESIA O. Z. TAMIN moreover, time disruptive to trip maers. As an iustration in Tamin 3, to obtain the nationa OD matrix, the Department of Transportation, Repubic of Indonesia coud ony carry out this survey three times within a 25-year period 1982, 1988, and 1996). Tamin et a. 4 mentioned that the broad outine of the 1998 Nation s Direction GBHN) of Indonesia stated that a poicies in transport deveopment shoud be directed to perform an efficient, safe, comfortabe, reiabe, and environmentay-based Nationa Transportation System Sistranas). The rapid changes in and use, popuation and empoyment, as we as vehice ownership have resuted in the conventiona methods being no onger suitabe for deveoping countries. This is due to the engthy process 2-3 years) which wi resut in the information contained in the OD matrices that do not refect the rea situation. Practicay, it is frequenty found that in soving the 2001 transportation probem, the 1996 OD matrix is sti being used due to the ac of the most recent OD matrix information. Athough the 2001 OD survey is sti undergoing, the 2001 OD matrix information wi be avaiabe perhaps in the year Furthermore, during the monetary crisis, it is amost impossibe to carry out this survey for the next 5-10 years. Moreover, for urban areas, the regiona government of Jaarta can ony afford to carry out this OD survey three times in the ast 23 years as mentioned in Tamin et a. 4, through very arge and expensive transport projects such as: Jaarta Metropoitan Area Transportation Study JMATS) in 1975, Arteria Road System Deveopment Study ARSDS) in 1987 and Transport Networ Panning Reguation Study TNPRS) in A of these require an answer. This becomes even more vauabe for probems which require quic-response treatment such as urban transport probems due to high urbanization, rapid growth of popuation, improvement of income eve, etc. Therefore, the new approach to tace a of these probems is urgenty required. The need for inexpensive methods, which require owcost data, ess time and ess manpower, generay caed an unconventiona method is therefore obvious due to time and money constraints. Traffic counts, the embodiment and the refection of the OD matrix; provide direct information about the sum of a OD pairs which use those ins. Some reasons why traffic counts are so attractive as a database are: first, they are routiney coected by many authorities due to their mutipe uses in many transport panning tass. A of these mae them easiy avaiabe. Second, they can be obtained reativey inexpensivey in terms of time and manpower, easier in terms of organization and management and aso without disrupting the trip maers. Therefore, a ey eement of the approach is a system to update the transport demand mode using ow-cost traffic count information. Previous research 2 and severa other researchers have been abe to obtain the OD matrices by ony using traffic count information. Unfortunatey, at that time, they sti used the steady-state traffic count information obtained from the traffic count survey. Nowadays, the technoogy for automatic traffic data coection is very advanced. The atest deveopment in automatic data coection for traffic count enabes us to obtain the shorttime-interva traffic count information. For exampe, as reported in AWA Pessey 5, ATCS Area Traffic Contro System) aready instaed in severa arge cities in Indonesia, such as: Jaarta 1994), Bandung 1997), and Surabaya 1998) provide us with the short-time-interva traffic count information for a signaised intersections. Furthermore, the technoogy for transferring data is aso readiy avaiabe and at a very ow cost through the use of teephone ines. By using this information, the research is directed to deveop a transport modeing system that enabes us to produce the OD matrices in a short-time-interva basis. Methods for estimating OD matrices can be cassified into two main groups as shown in Figure 1. They are as foows: conventiona and unconventiona methods 2,3. Conventiona methods rey heaviy on extensive surveys, maing them very expensive in terms of manpower and time, disruption to trip maers and most importanty the end products are sometimes short-ived and unreiabe. Another important factor is the compications that arise when foowing each stage of the modeing process. Furthermore, in many cases, particuary in sma towns and deveoping countries, panners are confronted with the tas of undertaing studies under conditions of time and money constraints, which mae the appication of the conventiona methods amost impossibe. The introduction of inexpensive techniques for the estimation of OD matrices wi overcome the probem. As a resut of dissatisfaction expressed by transport

3 TRANSPORTATION Direct Methods Roadside Interview Home Interview Fagging Method Aeria-Photography Method Car-Foowing Method Anaog Method Methods for Estimating OD Matrices Conventiona Methods Indirect Methods Unconstrained - Uniform Singy-Constrained - Production-Constrained - Attraction-Constrained Douby-Constrained - Average - Fratar - Detroit - Furness Modes Based on Trafic Counts Synthetic Method Source: Tamin 2,3 ) Unconventiona Methods Maximum-Entropy-Matrix- Estimation ME2) Transport Demand Mode Estimation TDME) Opportunity Mode Gravity Mode Gravity-Opportunity Mode panners with conventiona methods, other techniques for estimating OD matrices, which are based on traffic counts have evoved over the years; these are generay caed unconventiona methods. The aim of unconventiona methods is to provide a simper approach to sove the same probem and at a ower cost. Ideay, this simper approach woud treat the four-stage sequentia mode as a singe process. To achieve this economic goa, the data requirements for this new approach shoud be imited to simpe zona panning data and traffic counts on some ins or other ow-cost data. There are severa reasons why traffic counts are so attractive as a database to estimate the OD matrices: a. Low-cost This type of data is reativey inexpensive to obtain since they require ess manpower and automatic traffic counters can be used. They do not require preparation of questionnaires or statutory powers, and therefore they are easier in terms of organisation and management. They require simper data anaysis and output. b. Avaiabiity Traffic counts are aways avaiabe due to mutipe uses in inter-urban or urban transport studies. They are widey used for different purposes ie congestion anaysis, accident studies, maintenance panning, intersection improvement, monitoring fow eve and aso used to determine expansion factors for OD surveys and to update OD matrices. Furthermore, many oca authorities and panning bodies obtain these data reguary and hence the additiona cost of using unconventiona methods is ony margina. c. Non-disruptive Traffic counts can be obtained without generating any deay or disruption to vehices. Furthermore, the automatic coection of traffic counts is we advanced and its accuracy is very satisfactory as we as there being severa computer pacages providing efficient processing. Low 6 probaby deveoped the first mode based on traffic counts to be reported. The objective of his mode was to effectivey combine into one singe process what is usuay handed in a series of three or four sub-modes, each with its own set of errors. One of the advantages is that a the modeing errors appear in the fina output in terms of traffic voumes and can be described statisticay. The user thus has a better idea of how good his mode is something he does not now with the usua approach.

4 THE DEVELOPMENT OF THE REAL TIME INTEGRATED TRAFFIC INFORMATION SYSTEM RITIS) FOR INDONESIA O. Z. TAMIN 3.1 Genera Nguyen 7, Wiumsen 8, and Tamin 3 provide a very good and comprehensive overview on the state of art in this research domain reated to the OD matrix estimation based on traffic counts. They state the genera probem in the foowing way. Let P denote the set of origins, Q denote the set of destinations and I = PxQ denote the set of origin-destination OD) pairs. Most of the existing modes to estimate an OD matrix [T id ] from traffic counts may be written in the form: minimum or maximum S = fˆv, V )... 1) subject to i d T id p id = ˆV for L...2) T id ) where: T id = number of trips traveing from origin i to destination d; pid = proportion of trips traveing from each origin i to each destination d that use in, 0 pid 1; ˆV, V = observed and estimated voume on in. It can be seen that the vaue of p id is defined by the route chosen by each user within the study area which can be estimated by appying suitabe route choice technique. There are now avaiabe severa route choice techniques ranging from the simpest one a-or-nothing) to the most sophisticated one equiibrium) 9. To obtain the vaues of p id, Tamin 2,3,10,13 has deveoped the procedure to estimate the vaues of p id based on the predetermined trip assignment technique. Thus, theoreticay, by nowing the information on ˆV and p id, the vaue of T id can be estimated through the mechanism of optimization equations 1) 3). 3.2 Transport demand estimation approach The centra idea is to deveop estimation methods that can be used not ony for estimating currenty prevaiing OD matrix in a short-time-interva basis and hence the OD fows but aso for forecasting OD matrices and OD fows which wi prevai in the future. One possibe way to deveop methods for estimating OD matrices from traffic counts is by modeing the trip maing behaviour. The transport demand mode estimation approach assumes that the trave pattern behaviour is we represented by a certain genera transport mode, e.g., a gravity mode. The main idea is to appy a system of transport modes to represent the trave pattern 2,3. It shoud be noted here that the transport demand modes are described as functions of some panning variabes ie popuation or empoyment and one or more parameters. Whatever the specification and the hypotheses underying the modes adopted, the main tas is to estimate their parameters on the basis of traffic counts. Once, the parameters of the postuated transport demand modes have been caibrated, they may be used not ony for the estimation of the current OD matrix, but aso for predictive purposes. The atter requires the use of future vaues for the panning variabes. Consider a study area that is divided into N zones, each of which is represented by a centroid. A of these zones are inter-connected by a road networ that consists of a series of ins and nodes. Furthermore, the OD matrix for this study area consists of N 2 trip ces. N 2 N) trip ces if intrazona trips can be disregarded. The most important stage is to identify the paths foowed by the trips from each origin to each destination. The variabe pid is used to define the proportion of trips by mode traveing from zone i to zone d through in. Thus, the fow on each in is a resut of: trip interchanges from zone i to zone d or combination of severa types of movement traveing between zones within a study area = T id ); and the proportion of trips by mode traveing from zone i to zone d whose trips use in, which is defined by pid 0 pid 1). The tota voume of fow ˆV ) in a particuar in is the summation of the contributions of a trip interchanges by mode between zones within the study area to that in. Mathematicay, it can be expressed as foows: V = i d T id p id... 4) Given a the p id and a the observed traffic counts ˆV ), then there wi be N2 unnown T id s to be estimated from a set of L simutaneous inear equations 1) where L is the tota number of traffic passenger) counts. In principe, N 2 independent and consistent traffic counts are required in order to determine uniquey the OD matrix [T id], N 2 N) if intrazona trips can be disregarded. In practice, the number of observed traffic counts is much ess than the number of unnowns T id s.

5 TRANSPORTATION 3.3 Fundamenta basis Using unconventiona methods, it is assumed that a certain type of transport demand mode such as the gravity mode can represent the trip-maing behavior. The in fows can then be represented as a function of a mode form and its reevant parameters. The parameters of the postuated mode are then estimated so that the errors between the estimated and observed traffic counts are minimized. Consider now that there are K trip purposes or commodities traveing between zones within the study area. Assume aso that a certain transport demand mode such as gravity GR) mode can represent the interzona movement within the study area. Hence, the tota number of trips T id with origin in i and destination d for a trip purposes or commodities can be expressed as: T id = T id... 5) T id is the number of trips for each trip purpose or commodity traveing from zone i to zone d as expressed by equation 6) generay nown as a douby constrained gravity mode DCGR). T id = O i D d A i B d f id... 6) A i and B d = baancing factors expressed as: A i = [ d B d D d f id)] 1 and B d = [ i A i O i f id)] ) f id = the deterrence function = exp β C id)... 8) The readers who want to now more about the Gravity-Opportunity GO) mode are suggested to read Tamin 12 and Tamin and Soegondo 13. By substituting equation 6) to 4), the fundamenta equation for estimating the transport demand mode based on traffic counts is: V = i d O i D d A i B d f id pid)... 9) The fundamenta equation 9) has been used by many papers not ony to estimate the OD matrices but aso to caibrate the transport demand modes from traffic count information 2,3,10,11,12,13. Theoreticay, having nown the vaues of ˆV and pid, T id can be estimated by foowing the optimization mechanism of equations 1) 3). Equation 9) is a system of L simutaneous equations with ony one unnown parameter β that needs to be estimated. The probem now is how to estimate the unnown parameter β so that the mode reproduces the estimated traffic fows as cose as possibe to the observed traffic counts. 3.4 Estimation methods Tamin 3 expains severa types of estimation methods that have been deveoped so far by many researchers as: Least-Squares estimation method LLS or NLLS) Maximum-Lieihood estimation method ML) Bayes-Inference estimation method BI) Maximum-Entropy estimation method ME) Least-Squares estimation method LS) Tamin 2, 3 have deveoped severa Least-Squares LS) estimation methods of which its mathematica probem can be represented as equation 10). to minimize S = [ ˆV V )2 ]... 10) ˆV V = observed traffic fows for mode = estimated traffic fows for mode The main idea behind this estimation method is that we try to caibrate the unnown parameters of the postuated mode so as to minimize the deviations or differences between the traffic fows estimated by the caibrated mode and the observed fows. Having substituted equation 9) to 10), the foowing equation is required in order to find an unnown parameter β which minimizes equation 10): δs = [2 T p Vˆ ) T id p ))] = 0 id id δβ i d i d β id... 11) Equation 11) is an equation which has ony one unnown parameter β that needs to be estimated. Then it is possibe to determine uniquey a the parameters, provided that L > 1. Newton-Raphson s method combined with the Gauss-Jordan Matrix Eimination technique can then be used to sove equation 11) 14,15. The LS estimation method can be cassified into two: Linear-Least-Squares LLS) and Non-Linear-Least- Squares NLLS) estimation methods. Tamin 2 has concuded that the NLLS estimation method requires a onger processing time for the same amount of parameters. This may be due to the NLLS estimation method containing a more compicated agebra compared to the LLS so that it requires a onger time to process. However, the NLLS estimation method aows us to use the more reaistic transport demand mode in representing the trip-maing behavior. Therefore, in genera, the NLLS provides better resuts compared to the LLS.

6 THE DEVELOPMENT OF THE REAL TIME INTEGRATED TRAFFIC INFORMATION SYSTEM RITIS) FOR INDONESIA O. Z. TAMIN Maximum-Lieihood estimation method ML) Tamin 2,3 has aso deveoped an estimation method that tries to maximise the probabiity as expressed in equation 12). The framewor of the ML estimation method is that the choice of the hypothesis H maximising equation 12) subject to a particuar constraint, wi yied a distribution of V giving the best possibe fit to the survey data ˆV ). The objective function for this framewor is expressed as: V to maximize L = c p ˆ... 12) subject to: V ˆV T = ) where: ˆV T =tota observed traffic fows c = constant p V = ˆV T By substituting equation 9) to 12), finay, the objective function of ML estimation method can then be expressed as equation 14) with respect to unnown parameters β and θ. max. L 1 = [ ˆV og e i d T id p id) θ i d T id p id] + θ ˆV T ˆV T og e ˆV T + og e c...14) The purpose of an additiona parameter θ, which appears in equation 14), is that to ensure the constraint equation 13) shoud aways be satisfied. In order to determine uniquey parameter β of the GR mode together with an additiona parameter θ, which maximizes equation 14), the foowing set of equations are then required. They are as foows: δtid δl = V ˆ id 1 i d δβ p id δt id δβ T i d δβ p id p id θ = 0 i d... 15a) δl 1 = θ [ T p id V ˆ id ] = 0 δθ T...15b) i d Equation 15ab) is in effect a system of two simutaneous equations which has two unnown parameters β and θ that need to be estimated. Again, the Newton- Raphson s method combined with the Gauss-Jordan Matrix Eimination technique can then be used to sove equation 15ab) Bayes-Inference estimation method BI) Tamin 3 has deveoped the Bayes-Inference BI) estimation method in which the main idea is to combine the prior beiefs and observations to produce posterior beiefs. If one has 100% confidence in one s prior beief then no random observations, however remarabe, wi change one s opinions and the posterior wi be identica to the prior beiefs. If, on the other hand, one has itte confidence in the prior beiefs, the observations wi then pay the dominant roe in determining the posterior beiefs. In other words, prior beiefs are modified by observations to produce posterior beiefs; the stronger the prior beiefs, the ess infuence the observations wi have to produce the posterior beiefs. The objective function of the Bayes-Inference BI) estimation method can be expressed as: to maximize BI T V ) = ˆV og e V )...16) By substituting equation 9) to 16), the objective function can then be rewritten as: to maximize BI = [ ˆV og e T i d id Pid)]...17) In order to determine the unique parameter β of the GR mode, which maximizes equation 17), the foowing equation is then required: BI β = ˆV T id p id) i d Tid id β p = ) Equation 18) is an equation which has one unnown parameter β that needs to be estimated. Again, the Newton-Raphson s method combined with the Gauss-Jordan Matrix Eimination technique can then be used to sove equation 18) Maximum-Entropy estimation method ME) Tamin 5 has deveoped the maximum-entropy approach to caibrate the unnown parameters of the gravity mode. Now, this approach is used to deveop procedures to caibrate the unnown parameters of the transport demand mode based on traffic count information. The basis of the method is to accept that a micro states consistent with our information about macro states are equay iey to occur. Wison 16 expains that the number of micro states W{V } associated with the meso state V is given by: W [V ] = V!... 19) V! T i d

7 TRANSPORTATION As it is assumed that a micro states are equay iey, the most probabe meso state woud be the one that can be generated in a greater number of ways. Therefore, what is needed is a technique to identify the vaues [V ] which maximize W in equation 19). For convenience, we see to maximize a monotonic function of W, namey og e W, as both probems have the same maximum. Therefore: T og e W = og V! e = og e V T! og e V!..20) V! Using Stiring s approximation for og e X! Xog e X X, equation 20) can be simpified as: og e W = og e V! V og e V V )...21) T Using the term og e V T! is a constant; therefore it can be omitted from the optimization probem. The rest of the equation is often referred to as the entropy function. og e W' = V og e V V )... 22) By maximising equation 22), subject to constraints corresponding to our nowedge about the macro states, enabes us to generate modes to estimate the most iey meso states in this case the most iey V ). The ey to this mode generation method is, therefore, the identification of suitabe micro-, meso- and macro-state descriptions, together with the macro-eve constraints that must be met by the soution to the optimisation probem. In some cases, there may be additiona information in the form of prior or od vaues of the meso states, for exampe observed traffic counts ˆV ). The revised objective function becomes: og e W'' = V ˆ og e ) V + V... 23) Vˆ Equation 23) is an interesting function in which each eement in the summation taes the vaue zero if V = ˆV and otherwise is a positive vaue which increases with the difference between V and ˆV. The greater the differences, the smaer the vaue of og e W". Therefore, og e W" is a good measure of the difference between V and ˆV. Mathematicay, the objective function of the ME estimation method can be expressed as: V to maximise V E 1 = og e W'' = V ˆ og e ) V + V... 24) Vˆ In order to determine the unique parameter β of the GR mode that maximizes the equation 24), the foowing equation is then required: T i d id id i d β p T id p id δe 1 = og e = ) ˆV Equation 25) is an equation which has ony one unnown parameter β that needs to be estimated. Again, the Newton-Raphson s method combined with the Gauss- Jordan Matrix Eimination technique can then be used to sove equation 25) Test case with steady state traffic count data The rea data set of urban traffic movement in Bandung in terms of steady state traffic count information was used to vaidate the proposed estimation methods. Bandung is the capita of West Java Province and its popuation is around 6.4 miion in 1998 and expected to increase to 13.8 miion by The tota area of Bandung is around 325,096 hectares and is divided into 66 ecamatans and 590 eurahans. The study area was divided into 146 zones of which 140 are interna zones and 6 are externa. The road networ of the study area consisted of 653 nodes and 1,811 road ins. There are 95 observed steady-state traffic counts ˆV ), traffic generation and attraction O i and D d ) for each zone, and observed OD matrix for comparison purpose. The units used in equation 9) are as foows: ˆV = traffic counts in vehices/hour O i, D d = trip generation/attraction for each zone in vehices/hour The most important thing in transport demand mode estimation from traffic counts is to now how good the caibrated transport modes are in reproducing the observed OD matrix. There are two ways of doing this tas: a. the accuracy of the estimated OD matrices compared to the observed one; b. if the estimated OD matrix is assigned onto the networ then the corresponding traffic fows in each in shoud be as cose as possibe with the observed in fow obtained from ATCS contro center. In order to estabish the strategy for vaidity and sensitivity tests, it is necessary to introduce at this stage the main issues affecting the accuracy of the estimated OD matrix produced by the caibrated modes. These are as foows: the choice of the transport demand mode itsef to be used in representing the trip maing behaviour within the study area or, perhaps, a system of the rea word;

8 THE DEVELOPMENT OF THE REAL TIME INTEGRATED TRAFFIC INFORMATION SYSTEM RITIS) FOR INDONESIA O. Z. TAMIN the estimation method used to caibrate the parameters of the transport mode from traffic count information; ocation and number of traffic count information; the eve of errors in traffic counts; and the eve of resoution of the zoning system and the networ definition. The vaidity and sensitivity tests can then be estabished from these five main issues. Two transport demand modes, namey gravity GR) and gravity-opportunity GO) modes, and four estimation methods NLLS, ML, BI, ME) have been used in the vaidity tests. The four estimation methods mentioned above have been discussed in detai in Section 3.4. The vaue of R 2 statistics as expressed in equation 26) 27) is used to compare the observed and estimated OD matrices to ascertain how cose they are 2,3. Tˆ id T id ) 2 R 2 = 1 i d Tˆ id T 1 ) ) i d 1 T 1 = T id N N 1) i d for i d...27) The best ocation of traffic counts It is mentioned that the unconventiona method uses traffic count information as the main input for estimating the OD matrices. Because of that, any process regarding the traffic counts shoud be ceary and deepy understood in order to obtain the best estimates of OD matrices; especiay those which are reated to data coection process, e.g., number of traffic counts and their best ocations. The data coection process is very important since it is the first action in the whoe process of OD matrix estimation. Some basic anayses used in finding the best ocation are as foows: a. Proportion of trip interchanges on a particuar in The tota voume of fow in a particuar in ˆV ) is the summation of the contributions of a trip interchanges between zones within the study area to that in. Mathematicay, it can be expressed as equation 4). Tamin 3 stated that the most important stage for the estimation of OD matrix from traffic counts is to identify the paths foowed by the trips from each origin i to each destination d. In other words, the proportion of trip interchanges between zone i and zone d have to be uniquey identified for a those ins invoved. In this case, the variabe p id is used to define the proportion of trip interchanges from origin i to destination d traveing through in. Therefore, in finding the best ocation, the traffic counts having much information of the trip interchanges shoud be chosen. This information can be identified by anaysing the tota number and vaue of p id in each in. This information wi then be taen as the main criteria to determine the choice of the best ocation for traffic counts. b. Inter-in reationship Inter-dependence Figure 2 shows that fows on in 5-6 are the summation of fows on in 1-5 and on in 2-5, then there is no additiona information that can be extracted by counting fows on in 5-6 because of the fow continuity condition, V 56 = V 15 + V V15 V25 Source: Tamin 3 ) V V63 V64 In principe, we need counts for ony three independent counts in order to find the fows of a ins in Figure 2. Therefore, from an economic point of view, some efforts are needed in choosing the appropriate ins to be counted 3. Inconsistency In practice, the probem of inconsistency in in counts may arise when the fow continuity conditions are not satisfied by the observed voumes. In the case of Figure 2, it may we happen that the observed fows are such that: V 56 V 15 + V ) or V 15 + V 25 V 63 + V ) This inconsistency in counts may arise due to human or counting errors and counting at different times or dates. As a resut of a this, no soution for the OD matrix can be estimated that reproduces a these inconsistent traffic counts. One possibe way to remove this probem is by choosing ony independent ins to be counted

9 TRANSPORTATION c. Optimum number of traffic counts Equation 4) is the fundamenta equation deveoped for estimating the OD matrices from traffic counts information. In this mode, the parameters p id are estimated using traffic assignment technique. Given a the p id and a the observed traffic counts ˆV ), then there wi be N 2 T id s to be estimated from a set of L simutaneous inear equations 1) where L is the tota number of traffic counts. In principe, N 2 independent and consistent traffic counts are required in order to determine uniquey the OD matrix [T id ]; [N 2 N] if intrazona trips can be disregarded. In practice, the number of observed traffic counts is much ess than the number of unnowns T id s. Therefore, it is impossibe to determine uniquey the soution 2. d. The determination of the optimum number of traffic counts As mentioned above, the determination of the optimum number of traffic counts wi be conducted under one condition representing the sensitivity between the number of traffic counts and the in ran to the accuracy of the estimated OD matrices, namey: random condition. In this condition, severa combinations of traffic counts wi be created based on random seection. Each combination of traffic counts wi then be used to estimate the OD matrices. In this research, the initia OD matrix was created by caibrating the Gravity-Opportunity GO) mode from traffic counts by using a seected ins 646 ins). Tamin et a. 18 reports that the best of vaues of parameters ε and µ) are ε = 0,4 and µ = 1,0 for the GO mode. The other unnown parameters α and β) of the GO mode were then caibrated using 646 seected traffic counts by using Non- Linear-Least-Squares NLLS) estimation method. The initia OD matrix to be used for comparison purposes wi then be created using the GO mode together with the vaues of its caibrated parameters. Figure 3 shows the reationship R Number of Traffic Counts Condition II Random) between the eve of accuracy of the estimated matrices compared to the initia one and the number of seected traffic counts under random conditions. It can be seen from Figure 3 that the use of 90 ins has reproduced the reativey high accuracy of estimated OD matrices compared to the initia one in terms of R 2 ). The use of 90 ins has reativey the same accuracy with the use of 646 ins. It can be concuded that the optimum number of traffic counts is 90 ins 14% of 646 seected ins or 3.6% of 2,485 avaiabe ins) Important findings Tabe 1 shows the vaues of R 2 statistics of the observed OD matrix compared with the estimated OD matrices obtained from traffic counts. Mode Estimation Methods NLLS ML BI ME GR/GO 1 GR GO Note: 1) obtained using the observed OD matrix information Some fina concusions can then be drawn from Tabe 1. They are as foows: In terms of OD matrix eve, it was found that the GO mode aways produced the best estimated matrices. However, these are ony marginay better than those obtained by the GR mode. Taing into account the resuts of using other criteria, it can be concuded that the best overa estimation methods are the combination of GO mode with NLLS estimation method and the worst is GR mode estimated by BI estimation method. With the evidence so far, it was found that the estimated modes and therefore OD matrices are ony sighty ess accurate than those obtained directy from the fu OD surveys. This finding concudes that the transport demand mode estimation approach is found encouraging in terms of data coection and transport mode estimation costs. Severa important findings can be concuded as given in Tabe 2, which shows the performance raning of mode s estimation method according to specified criteria. The purpose of this tabe is to provide guidance to choose the best overa mode s estimation method regarding its behaviour to severa criteria such as: accuracy, computer time, sensitivity to errors in traffic counts, sensi-

10 THE DEVELOPMENT OF THE REAL TIME INTEGRATED TRAFFIC INFORMATION SYSTEM RITIS) FOR INDONESIA O. Z. TAMIN tivity to zoning eve and networ soution, and sensitivity to number of traffic counts. Sma differences of R 2 on Tabe 1 wi then be regarded as reasonabe differences to determine the superiority/inferiority among the estimation methods. The vaues of R 2 are transferred to the raning scae ranging from 1 to 8 to see the performance of estimation methods based on the above criteria. This approach is used to homogenise severa types of quantitative scaing systems between each criteria into a 1 8 scaing system. Scae 1 shows the worst performance, whie scae 8, shows the best performance. It can be seen from Tabe 2 that in terms of accuracy and sensitivity to number of traffic counts criteria, the GO mode together with NLLS estimation performs the best. Whie, in terms of computer time, sensitivity to errors in traffic counts, sensitivity to zoning eve and networ resoution, the GR mode with NLLS estimation performs the best. In genera, it can be concuded that the NLLS estimation method shows the best raning performance based on severa types of criteria. 4.1 Introduction The Area Traffic Contro System ATCS) which has been instaed in three arge cities Jaarta, Bandung, and Surabaya) enabed us to obtain the rea-time or short-timeinterva traffic count information automaticay for a signaized intersections. DLGT 17 and AWA Pessey 5 report that the ATCS has been fuy operationa in Bandung since The technoogies for transferring data via the Internet and teephone ines are aso avaiabe and at very ow cost that enabes us to obtain the traffic count information in a short-time-interva basis. Basicay, the objective of ATCS is to achieve the optimum traffic performance through minimization of intersection deay and creating continuous traffic fow caed a green wave aong the coordinated intersections. To achieve the above condition, the oop detectors record the traffic fow passing through the approaches. Then, the traffic data wi be used for traffic signa arrangement interactivey. The traffic data woud be saved in the database system at the ATCS contro center through teecommunication networ. This traffic data is updated periodicay in a shorttime-interva basis. The database system can be accessed very easiy at a very ow cost through the Internet and teephone ine faciities. This data woud be the main input data for short-time-interva OD matrix estimation. As an iustration, as reported in AWA Pessey 5, Bandung has 117 intersections under ATCS and divided into two areas: the northern area consists of 59 intersections and the southern area consists of 58 intersections. The traffic data obtained from ATCS is traffic data in the approach of intersection. It is required to convert the data into in traffic data as required by the estimation process. This can be done through the conversion factors. 4.2 The deveopment of Data Processing Center DPC) As we now that the short-time-interva data traffic count wi be used as the main data. This system has some sensors detectors) in every direction point and in genera terms it can be seen in Figure 4. For the fina process, the traffic count fies are required to estimate the OD matrices. These fies are taen Mode and estimation methods Accuracy *) Computer time GR GO Criteria Sensitivity to Sensitivity to zoning eve Sensitivity to number errors in traffic counts and networ resoution of traffic counts NLLS ML BI ME NLLS 8 4 NA NA 8 ML 5 2 NA NA 7 BI 4 3 NA NA 5 ME 7 2 NA NA 6 Note : *) concuded from Tabe 1 Source : Anaysis

11 TRANSPORTATION periodicay from ATCS. A the traffic data of every sensor in each intersection are sent to the ATCS server and saved into a fie which have to be ordered first. In ordering the fie, some variabes shoud be specified such as: ocation of intersection, and the saving period date, hour, minute). Another probem is the data communication. Some steps have to be foowed: deveoping data communication from ATCS server to aboratory The data communication required is something reiabe and not too expensive. After some anaysis, the teephone ine medium is chosen which is quite cheap especiay in operationa costs. deveoping communication to the ATCS server The operating system in ATCS server is quite different with other we-nown operating systems. It uses the OpenVMS of Digita/DEC. For those two steps above, an attached computer faciity is required. The main purpose of this system is to provide data in short-time-interva basis; therefore, a time-basis process is required. The automation process incudes: process in ordering the data, taing data from the ATCS server, sending data using a teephone ine, converting data from VS formatted into text formatted, processing data into MOTORS pacage program, and, saving it into a documentation fie. In this step, the textformatted fies need to be converted into a fie that can be used by MOTORS pacage program Basic configuration In genera, the basic configuration of the system is summarized in Figure 5. Coecting automaticay the traffic count information obtained from the Traffic Contro Center TCC) of ATCS in a short-time-interva basis deveops this information system. However, some important processes are required such as data transferring, data processing, and output processing, as foows: Data transferring: Traffic counts at signaised intersections were detected via wire detectors and automaticay connected to the Traffic Contro Center TCC) of ATCS. This traffic data is updated periodicay in a short-time-interva basis. The traffic data information is then provided at the TCC and can be directy and easiy accessed at a very ow cost via a teephone ine. Data processing: This data is the main input for shorttime-interva OD matrix estimation. Before the traffic data is used in the estimation process; firsty, those data have to be processed in the Data Processing Center DPC). The process incudes: error treatment due to transfer process, data formatting, database preparation of zoning and networ system, etc. Moreover, the traffic data obtained from ATCS is traffic at the approach of intersection. It is demanded to convert the data into in traffic data as required by the estimation process. This can be done through conversion factor. Having it processed, the traffic data wi then be ready to be used for estimating the short-time-interva OD matrices. Output processing: The estimated short-time-interva OD matrices and their practica appications are stored in a Rea Time Integrated Traffic Information System RITIS) so that the users can directy and easiy access the information through the Internet faciity. The RITIS is designed specificay and informativey for the purposes of user needs. Route information processing: The driver can contact the Data Processing Center DPC) to inform his ocation and his destination zone in which his ocation wi be determined by GPS. The DPC wi then process the best route by considering the atest traffic conditions and send the best route information bac to the vehice. A of these processes wi be designed in the forms of Route Guidance System RGS) and the Rea Time Integrated Transportation Information System RITIS).

12 THE DEVELOPMENT OF THE REAL TIME INTEGRATED TRAFFIC INFORMATION SYSTEM RITIS) FOR INDONESIA O. Z. TAMIN ATCS Traffic Data Contro in Bandung Route Guidance System Transfer Information via Teephone Line Data Processing Interface DPI) Rea Time Traffic Count Data Coection GPS Dispay Transfer Information via Internet Output Data Processing Numerica/Graphica) Rea Time OD Matrices and Their Practica Appications Rea Time OD Matrix Estimation From Rea Time Traffic Count Data USERS Government, Traffic Authority, Panning Authority, Department of Pubic Wors, Poice, Consutants, etc.) As mentioned above, some techniques and methods have been deveoped in very recent years which enabe us to obtain the OD matrices by using ony easiy avaiabe and ow-cost traffic count information. Unfortunatey, at that time, the modes sti used the steady-state traffic count information obtained from the traffic count survey. The atest deveopment in automatic data coection for traffic counts enabes us to obtain the short-time-interva traffic count information. For exampe, ATCS Area Traffic Contro System) aready instaed in Bandung since 1997 provides us with rea-time or short-time-interva traffic count information for a signaized intersections. Furthermore, the technoogy for transferring data is aso readiy avaiabe and at a very ow cost through the use of the Internet or a teephone ine faciity. The use of short-time-interva traffic count information enabes us to anayze the dynamic phenomena of OD matrices in a short-time-interva basis. The deveoped mode wi give high added vaue through high efficiencies in terms of time and cost especiay to be used to sove dynamic urban transportation probems. In other words, we can obtain the accurate and ow-cost OD matrix information reguary within a very short period such as every 30 minutes. Severa things that have to be studied more carefuy in order to increase the accuracy of the estimated OD matrices are as foows: a. deveopment of the Data Processing Interface DPI) and to study the best procedure for coecting shorttime-interva traffic count data from ATCS Contro Center; b. the conversion factor to convert the intersection-based traffic data into in-based traffic data; c. better nowedge and obtaining more advanced trans-

13 TRANSPORTATION port demand modes that wi represent more accuratey some specific trave demand patterns; d. the optimum time-sice of OD matrices; e. the optimum ocation and number of traffic count data and its impact on the accuracy of the estimated OD matrices; f. expanation on some unanswered questions reating to the impact of eve of detai of zoning system and networ definition on the OD matrices accuracy; g. more advanced route choice techniques capacity-restrained or equiibrium) to tae into account the effect of congestion especiay in urban areas in reation to dynamic OD matrix estimation from traffic counts; h. the impact of the intersection deay to route choice and its effect to the accuracy of the estimated OD matrices; i. the evoution of short-time-interva OD matrices due to traffic fow fuctuation. The output of short-time-interva OD matrices together with their practica appications wi be stored in a Rea Time Integrated Traffic Information System RITIS) designed specificay for the purposes of user needs numerica and graphica). A users panning authorities, traffic authorities, Department of Pubic Wors, consutants, poice, and other reated agencies) can directy and easiy access this information at a very ow cost through Internet faciities. Severa transport anayses can be conducted and severa appications can be carried out by using the RITIS, some of them are: to predict short-time-interva 30 minute) OD matrices based on fuctuating traffic, hence to provide the evoution of in fows as sources in identifying an appropriate road management scheme; to provide short-time-interva information on the performance of the networ, both numerica and graphic, e.g., in fows, in speeds, VCR vaues for a ins, route guidance, ocations of bottenecs, and many other short-time-interva practica information; to assess merits of the introduction of new transport poicy on the road networ performance before it is impemented; to anayze the effect of ATCS impementation on road traffic circuation; severa important appications which wi sove dynamic urban transport probems. This short-time-interva traffic information wi become the pubic-domain information that can be directy and freey accessed via the Internet by users e.g., road users, traffic poice, traffic panners, traffic authority, radio stations and TV stations, etc.). Moreover, this approach can aso be extended to provide the short-time-interva environmenta information. This method has been tested and vaidated in Bandung and it shows remaraby good resuts for Bandung conditions. Severa further appications can aso be deveoped such as: the route guidance system for private vehice and taxi, bus operating system, feet management, etc. The paper expains briefy the concept of deveoping the Rea Time Integrated Traffic Information System RITIS) and the Route Guidance System RGS) by utiizing the short-time-interva traffic count information. A nove unifying approach to describe the estimation of OD matrices from traffic count information has been given. The significance of the mode is, both theoretica and practica vaues; by understanding thoroughy the use of short-time-interva traffic count information in obtaining the short-time-interva OD matrices is a breathrough giving high added vaue for appications in deveoping countries due to its effectiveness and efficient uses in transport panning, engineering, management and poicy tass. The resut of previous research, which utiised the steady-state traffic count information, was found very usefu in deveoping the modified mode based on shorttime-interva traffic count information. By using this traffic count information, the short-time-interva OD matrix can aso be estimated. The OD matrices together with their appications wi be stored and provided in the RITIS designed specificay for the purposes of users numericay and graphicay) so that it can be directy accessed and used via the Internet at a very ow cost. are One of the most important pieces of information is the best routes from each origin zone to each destination zone which has aready considered the effect of congestion. This information wi be the main data for the deveopment of the Route Guidance System RGS) so that each driver can choose his best route through the road networ. The best route information wi be changed in a short-time-interva basis depending on the traffic condition. Moreover, this approach can aso be extended to provide short-time-interva environmenta information.

14 THE DEVELOPMENT OF THE REAL TIME INTEGRATED TRAFFIC INFORMATION SYSTEM RITIS) FOR INDONESIA O. Z. TAMIN 1. Abidin. The potentia and prospect of GPS for transportation sector in Indonesia. Proceedings of 1 st Symposium of Indonesian Inter-University Transport Studies Forum IIUTSF), 3 December 1998, Bandung in Indonesian). 1998). 2. Tamin, O.Z.The estimation of transport demand modes from traffic counts. Ph.D. Dissertation, University of London. 1988). 3. Tamin, O.Z. Transport Panning and Modeing, 2 nd Edition. ITB Press, Bandung. 2000). 4. Tamin, O.Z., Hidayat, H., Sjafruddin, A. The concept of deveoping the rea time OD matrices based on ATCS data. Proceedings of 1 st Symposium of Indonesian Inter-University Transport Studies Forum IIUTSF), December 1998, Bandung in Indonesian). 1998). 5. AWA Pessey. Bandung Area Traffic Contro System: After Traffic Study, Bandung. 1997). 6. Low, D.E. A new approach to transportation systems modeing. Traffic Quartery 263): pp ). 7. Nguyen, S. An agorithm for the traffic assignment probem. Transportation Science 83): pp ). 8. Wiumsen, L.G. An Entropy Maximising Mode for estimating trip matrices from traffic counts. Ph.D. Dissertation, University of Leeds. 1981) 9. Van Viet, D. and Dow, P.D. Capacity restrained road assignment. Traffic Engineering and Contro 206): pp ). 10. Tamin, O.Z. and Wiumsen, L.G. Freight Demand Mode estimation from traffic counts. Proceedings of the 16 th PTRC Summer Annua Conference, Bath UK). 1988). 11. Tamin, O.Z. Pubic transport demand estimation by caibrating a trip distribution-moda choice from traffic counts. Proceedings of the 5 th CODATU Conference, Sao Pauo Brazi). 1990). 12. Tamin, O.Z. Appication of transport demand modes for inter-regiona vehice movements in East Java. Proceedings of the 6th Word Conference on Transport Research, Lyon France). 1992). 13. Tamin, O.Z. and Soegondo, T. Freight demand modeing from traffic counts: A case study in Bai. Proceedings of the 5th Word Conference on Transport Research, Yoohama Japan). 1989). 14. Batty, M. Urban Modeing: Agorithm, Caibrations and Predictions. Cambridge University Press, Cambridge. 1976). 15. Wison, A.G. and Bennet, R.J. Mathematica Methods in Human Geography and Panning. John Wiey and Sons Ltd. 1985) 16. Wison, A.G. Entropy in Urban and Region Modeing. Pion Ltd, London. 1970). 17. DLGT. Bandung Area Traffic Contro System, Fina System Design, AWA Traffic System. 1996). 18.Tamin, O.Z., Soedirdjo, T.L., and Suyono, R.S. The impact of ocation and number of traffic counts in the accuracy of OD matrices estimated from traffic counts: A case study in Bandung Indonesia), to be pubished in the 4 th Journa of the Eastern Asia Society for Transportation Studies EASTS). 2001). The authors woud ie to than the University Research for Graduate Education URGE) Project, Directorate Genera of Higher Education, Department of Nationa Education for financiay supporting this research Graduate Team Research Grant, Batch IV, 1998/1999 with the tite: Dynamic Origin-Destination OD) Matrices Estimation From Rea Time Traffic Count Information under contract No. 029/HTTP-IV/URGE/1999.