OPTIMISING THE DESIGN OF MULTIMODAL FREIGHT TRANSPORT NETWORK IN INDONESIA

Transcription

1 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , 2005 OPTIMISING THE DESIGN OF MULTIMODAL FREIGHT TRANSPORT NETWORK IN INDONESIA Bon Frzl RUSS Tdsh YAMADA Grdute Student Assocte Professor Grdute School of Engneerng Grdute School of Engneerng Hroshm Unversty Kyoto Unversty Kgmym, Hgsh-Hroshm, Skyo-ku, Kyoto, Jpn Jpn Fx: Fx: E-ml: E-ml: Jun T. CASTRO Hronor YASUKAWA Reserch Assocte Professor Grdute School of Engneerng Grdute School of Engneerng Hroshm Unversty Hroshm Unversty Kgmym, Hgsh-Hroshm, Kgmym, Hgsh-Hroshm, Jpn Jpn Fx: Fx: E-ml: E-ml: Abstrct: Ths pper copes wth freght trnsport network desgn, prtculrly n the selecton of the most fesble set of nfrstructure projects. The problem nvolves nvestment plnnng, whch cn be trnsformed to combntorl optmston problem of dentfyng nd selectng the best set of fesble ctons from number of possble ctons. A mthemtcl model s developed wthn the frmework of b-level progrmmng problem, where multmodl mult-user ssgnment technque s ncorported wthn the lower level problem nd the combnton of ctons for cpcty expnson s optmsed usng genetc lgorthm-bsed procedures n the upper level problem. Model pplcton to the ctul freght trnsport network n Indones revels tht the model cn dequtely select the best combnton of ctons for desgnng the optml multmodl freght network. Key Words: Inter-regonl freght trnsport, Dscrete network desgn, B-level progrmmng, Genetc locl serch 1. INTRODUCTION Indones s vst rchpelgc country composed of more thn 17,500 slnds. Freght trnsport usng severl modes s therefore ndspensble. However, reltvely poor ttenton s gven on the plnnng of multmodl trnsport systems, prtculrly n the ntonl level. Most regons n Indones hve ndequte trnsport nfrstructure nd ther qulty hs much to be desred. In ddton, the emphss of trnsport plnnng hs been on the rod-mode, further worsenng the trnsport system nd resultng n severe socl nd envronmentl mpcts. Bsed on the 2001 Ntonl Orgn-Destnton Survey, the shre of other-thn-rod-mode n movng regonl freght s less thn 5% n the mjor slnd of Jv where the rod densty s reltvely hgh. Even n regons wth low rod densty such s Sulwes Islnd, Mluku nd Irn Jy Provnces, the shre of other-thn-rod-mode s less thn 50%. Ths pper dels wth the desgn of freght trnsport network, hghlghtng the selecton of the 2894

2 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , 2005 most fesble set of nfrstructure projects. The problem nvolves nvestment plnnng, whch cn be trnsformed to problem of dentfyng nd selectng set of fesble ctons from number of possble ctons to optmse some objectve functon. Ths problem cn therefore be consdered s combntorl optmston problem. Ths pper lso puts forwrd method to cope wth such problems wthn the frmework of b-level progrmmng, where multmodl mult-user equlbrum trffc flow s descrbed n the lower level problem nd the combnton of ctons s optmsed usng GA (Genetc Algorthm)-bsed procedures n the upper level problem. The dvntge of doptng such heurstc technques s tht they cn fcltte the desgn of b-level progrmmng optmston problems f ppled s combntorl optmston technques n the upper level problem, s well s they cn provde better solutons wthn resonble computton tmes. Ths type of problem lso nvolves mthemtcl problem wth equlbrum constrnts. The model s tested on n nvestment problem for freght trnsport network n Indones, bsed on set of possble ctons tht ncludes the mprovement or upgrdng of exstng nfrstructure nd constructon of new rods, rlwys nd freght termnls. 2. GENERAL FRAMEWORK A b-level progrmmng pproch s employed for formultng the problems, wth the soluton process shown n Fgure 1. The lower level cn be formulted s vrtonl nequlty problem, whle the upper level cn be formulted s combntorl optmston problem. A smplfcton s ppled n modellng the network by omttng the nfluence of shpper-crrer behvour nd ther ntercton n the freght trnsport decson. Ths s resonble ssumpton becuse n mny cses, especlly n Indones, the mcro level (.e. mult-commodty, shpper-crrer compny level) dt re unvlble. The vlble dt, whch were prmrly bsed on the ntonl orgn-destnton nd trnsport fclty surveys, re more vble for n ggregte-bsed model (Lubs et l., 2002). Optml Type nd Locton of Actons Ftness of Combnton Upper Level Objectve Functon Network Performnces Modl Splt nd Assgnment Generton of Combntons (Sets of Actons) Lower Level Alterntves of Actons (Type nd Locton) Fgure 1. Modellng nd Computtonl Frmework 2895

3 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , LOWER LEVEL PROBLEM 3.1 Network Representton We consder n bstrct mode network η(n,a), where N s the set of nodes nd A s the set of lnks. Nodes represent ctes or junctons where there re no ssocted delys or costs. Lnks, whch re the conduts for flow between two nodes, represent not only physcl nfrstructures (.e. rods, rlwys), but lso ctvtes tht cuse dely/cost to the flow (.e. lnks representng the lodng process n the termnl). A number of lnk types re used, such s centrod connectors tht connect the orgn/destnton pont (centrod) to the network, lnk wys tht vry mong modes nd termnl lnks. Termnls re very mportnt for freght movements such tht t ttrcts specl ttenton n freght network modellng. Tvsszy (1996) proposed smple trnshpment lnk wth fxed vlues of cost nd dely tme, whle Guelt et l. (1990) used more specfc trnsfer lnk tht cretes more lnks n the multmodl termnl. A more explct representton of termnls n the network proposed by Southworth et l. (2000) seprtes trnshpment lnk nto termnl ccess lnk nd trnsfer lnk nsde the termnl, lthough the pplcton s for dtbse nd routng purposes only (Fgure 2). To determne the explct effect of termnls, t s necessry to dd more lnks representng the processes n the termnl. For three-modl (multmodl) termnl, there would be lodng-unlodng ctvtes, trn spottng nd swtchng, dryge, wtng for vehcles or storge ncludng nspectons nd other dmnstrtve processes. However, t s necessry to mke smple representton of the termnl to vod complcted clculton nd nputs. Therefore, the termnl representton shown n Fgure 3 cn be used. Unlodng lnks represent unlodng/dschrgng ctvty nd storge, whle lodng lnks represent lodng nd other ctvtes n the termnl such s dryge, nspectons nd other dmnstrtve processes. Lnk wys re ctegorsed by mode. For the se nd rl mode, the ctvty of wtng for the vehcle s ncluded snce they re ssumed to hve lmted vehcle vlblty. Thus, the lnks cn be ctegorsed nto centrod connectors, rod lnks, rl lnks, se lnks, lodng lnks nd unlodng lnks for ech user type. Ths confgurton s used due to the chrcterstcs of the proposed cost nd dely functons descrbed n the followng secton. ) Tvsszy b) Guelt et l. c) Southworth et l. Note: Lnk of Mode 1 Lnk of Mode 2 Lnk of Mode 3 Trnsfer Lnk Node Fgure 2. Exmples of Termnl Representtons l1 u3 l3 u2 l2 u1 w1 w2 w3 Note: l1 Lodng lnk for Mode 1 u1 Unlodng lnk for Mode 1 l2 Lodng lnk for Mode 2 u2 Unlodng lnk for Mode 2 l3 Lodng lnk for Mode 3 u3 Unlodng lnk for Mode 3 w1 Lnk used by Mode 1 (ncludng wtng for vehcle) w2 Lnk used by Mode 2 (ncludng wtng for vehcle) w3 Lnk used by Mode 3 Node Fgure 3. Multmodl Termnl Representton 2896

4 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , Formultons Denotng k s the pth nd K ω s the set of ll pths n the network connectng the orgn-destnton (OD) pr ω, where ll OD prs belong to the set Ω, nd ssumng user type belongs to the set I wth p types of users, then f kω cn be defned s the flow of user type on pth k connectng ω, nd x s the flow of user type on lnk. The followng lnk flow conservton should then hold: where δ k ω = x = ω Ω k Kω f kω δ kω, A, I 1, f pth k connectng ω for user uses lnk 0, otherwse (1) Representng q ω s the demnd ssocted wth OD pr ω, then the followng OD flow conservton nd non-negtve pth flow should lso hold: q ω = f k Kω kω, ω Ω, I (2) f kω 0 (3) Wrdrop s user optml prncples stte tht the flow s dstrbuted on the network such tht the trvel costs on ll used routes between orgn nd destnton re equl, whle ll unused routes hve equl or greter trvel costs. Dfermos (1980) recognzed these user equlbrum (UE) condtons to follow vrtonl nequlty problem. In the cse when the Jcobn mtrx of the lnk cost functon s symmetrc, the UE flow my be obtned s the soluton of n equvlent convex cost mnmston problem. In ths pper, freght nd pssengers re treted s mult-clss users, wth modl splt nd route choce beng crred out smultneously by convertng the multmodl network nto unmodl bstrct mode network. Therefore, the UE problem to be delt wth s non-seprble nd symmetrc Jcobn mtrx cost functon mong user types. Ths cn be stted s vrtonl nequlty problem s follows: Fnd x * κ such tht: p = 1 A c ( x~ ) ( x Here, x * represents the lnk flow of ech user type tht s the soluton for the UE problem. x ~ s pn-dmensonl column vector wth the components {x 1,...,x n 1,..., x p,...,x n p } where n represents the number of lnks. κ cn be defned s κ { x ~ stsfyng Equtons (1), (2) nd (3)}. 3.3 Lnk Cost Functons * x * ) 0 x~ κ Cost on lnk for user type (except for the centrod connectors where the cost s neglected) s expressed s generlsed cost composed of fre component nd tme cost component. The tme cost component conssts of the product of the dely tme nd tme vlue for ech user type. (4) 2897

5 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , 2005 c ( x ) = ρ + α d ( x where c : generl cost on lnk (Rp) ρ : fre on lnk for user type (Rp) α : tme vlue for user type (Rp/hr) d (x ) : tme spent on lnk for user type (hr) x : flow on lnk for user type ) (5) The fre component s fxed vlue nd does not depend on volume, whle the tme cost component, prtculrly the tme spent on the lnk, s functon of volume nd dffers by lnk type. For smplcty, t s ssumed tht the termnl s seres of (M/M/1) queue system. The dely functon (.e. the functon for the tme spent on the lnk) for lodng nd unlodng s derved from resdence tme (men tme n tem spends n the system), whle dmnstrton nd other processes re consdered to be fxed vlues. In order to keep the lnk cost functon monotonclly ncresng, the tme spent for the dmnstrton process s ttched to the lodng process, whle the functon for the wtng process s ttched to the lnk wys of se nd rl modes. However, these functons hve symptotc behvours tht requre specl procedures prtculrly when the mgntude of flow s bout the totl cpcty, whch cn led to complex objectve functon. Therefore, the dely functons re requred to be converted to contnuous functon. Ths pper ncorportes the followng polynoml pproxmton method proposed by Crnc et l. (1990): T T T x d ( x ) = t0 1 + φ1x + φ2 T r where x T : totl flow on lnk (veh) t 0 : free runnng tme on lnk r : overll cpcty of lnk (veh) φ 1, φ 2,γ : prmeters to be clbrted γ (6) By pplyng ths equton to ll the lnks, the multmodl nd mult-type lnks network s then converted to sngle bstrct mode network. 3.4 Soluton Technques A method wdely used for solvng ths cse of mult-clss user ssgnment problem s dgonlston (Florn et l., 1982). Essentlly, ths method keeps ntercton effects constnt whle solvng the ssgnment problem by descent drecton lgorthm. When updtng the flow of one user type n the next terton, the other user type s consdered constnt. Ths cn smultneously be undertken untl no sgnfcnt chnges on the flows re obtned. Bsed on Sheff (1985) nd other dgonlston results n Thoms (1991), the bsc condton for rechng convergence s tht the lnk cost s only domnted by the flow on t. Even f the condton s volted, stsfctory result cn stll be obtned s long s the lnk 2898

6 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , 2005 cost s Jcobn postve defnte. The lgorthm used for the computton s bsed on the Frnk-Wolfe Method ccordng to Thoms (1991). 4. UPPER LEVEL PROBLEM 4.1 Problem Formulton We use the subset of A wth A=A 1 A 2 A 3, where A 1 ={:=1,2,...,n} cn be defned s the set of exstng lnks tht wll not be modfed, A 2 ={:=n+1,n+2,...,n+m} s the set of exstng lnks wth possble ctons to be mplemented, nd A 3 ={:=n+m+1,...,n+2m} s the updted verson of set A 2 (fter the cton s mplemented). Lnks n A 2 nd A 3 re numbered such tht f A 2 s selected, tht s, the cton ssocted wth t s beng mplemented, lnk +m n A 3 wll replce, otherwse +m n A 3 wll be dscrded. We then defne the set of possble combntons of cton Y ssocted wth A 3 (or A 2 ), tht s, Y = {y =n+m+1,...,n+2m} (or Y={y =n+1,...,n+m}), where Y s the set of combnton of ctons, nd y s the cton mplementton ndctor whch hs bnry vlue of 1 f the cton relted to lnk on the set of A 2 nd A 3 s mplemented, nd 0 f t s otherwse. Ths yeld Equton (5) nd (6) to the followng equtons: c ( x, y d ( x, y ) = t0 ) = ρ + α d ( x, y ) 1 + φ1x + φ 2 r 0 x + y r γ (7) (8) where c (x,y ) r 0 r : generlsed cost on lnk by user type wth the ctons beng mplemented or not (cton mplementton ndctor y ) (Rp) : exstng lnk cpcty on lnk : dded lnk cpcty on lnk f the cton s mplemented The objectve functon of the upper level problem cn be bsed on the rto of reduced totl generlsed cost nd nvestment cost requred by the combnton of ctons. Ths s the smplfcton of economc fesblty term beneft-cost rto (BCR), whch ndctes the economc effectveness of the cton. The dvntge of usng BCR vlue, nsted of just totl network generlsed cost, s tht ths prmeter cn ssess not only the reltve mprovement (compred to the ntl condtons) of combnton of ctons, but lso how effectve the combnton s. Furthermore, t cn lso nspect the occurrence of the Bress prdox (.e. see Bell et l., 1997),.e. those combntons of cpcty expnson tht ncrese the totl user cost. Denotng x * o s the equlbrum lnk flow for user type n ll the ntl lnks, c s the generlsed lnk cost functon on lnk, nd F s the set of user types for freght trnsport, then G o, the totl generlsed cost of the ntl network wthout ny cton mplemented cn be formulted s follows: G * * o = xoc ( xo ) F A (9) 2899

7 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , 2005 The objectve functon for selectng the best combnton of ctons wll be to mxmse the BCR vlue of freght. The followng objectve functon z(y) cn therefore be formulted: mxmse y Y z( y ) = where * x : lnk flows of ech user type tht re the soluton for the UE problem wth combnton of ctons mplemented (veh/dy) c (x *,y ) : generlsed cost on lnk by user type tht depends on the equlbrum flow nd whether the ctons re mplemented or not (cton mplementton ndctor y ) (Rp) b : nvestment cost of lnk tht reltes to freght nfrstructure development f correspondng cton s mplemented (Rp) 4.2 Soluton Technques G o F A1 x c ( x * * b A2 ) + Genetc Algorthm (GA) s used to solve the upper level problem. GA s metheurstc method ntroduced by Hollnd (1979) bsed on the mechnsm of nturl selecton nd genetcs. Lke other metheurstc lgorthms, GA cnnot ensure obtnng exct optml solutons but cn provde resonble nd prctcl solutons, especlly n solvng combntorl optmston problems (Goldberg, 1989). In GA, the possble soluton s constructed nto ndvduls known s chromosomes or strngs. Every poston n the chromosome s clled gene nd the vlue of gene s clled the llelc vlue. The most common llelc vlues re the bnry vlues {0,1}. In ths pper, gene represents the cton number nd the llelc vlue represents y whch s the cton mplementton ndctor. A number of ndvduls comprse populton. The process strts wth populton of rndom strngs representng possble solutons to the problem, nd the populton of strng members re evluted usng defned ftness functon nd the new set of populton s generted by genetc opertors. Genetc Locl Serch (GLS) s one of the GA-bsed procedures tht perform recombnton opertors wthn the locl genes, or n generl performs locl mprovement. In ths pper, the GLS procedures nclude reproducton, unform crossover nd creep mutton opertors. Lner ftness sclng s lso employed n the reproducton process. Furthermore, eltst model s used to preserve some of the best ndvduls for further generton (e.g. Ymd et l., 1999). The followng detls the lgorthm: Step 1: (Intlston) Generte the ntl populton (set of rndom strngs) nd set the number of generton (g) = 0. Step 2: (Ftness computton) Determne the ftness vlue of ech ndvdul n the populton. Step 3: (Elte preservton) Identfy number of ndvduls (number of eltes to be preserved) wth hgh ftness vlues. y A2 x c ( x * *, y ) (10) 2900

8 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , 2005 Step 4: (Evoluton) - Perform unform crossover. Ths s process of genertng new strngs by combnng the prent strngs from the prevous populton. - Perform creep mutton by determnng the locton (gene) of mutton rndomly nd the vlue t tht gene s swpped. Creep mutton llows multple muttons perform n one ndvdul. - Perform locl serch by selectng gene rndomly nd ech of the neghbourng gene s swpped for ts llele. Ths cn produce three ndvduls ncludng the orgnl one. The best ndvdul mong these three s dentfed nd preserved for the next step. Step 5: (Reproducton) - Determne the ftness vlue of ech ndvdul n the populton resulted from the prevous step. - Perform reproducton ccordngly to develop new populton. - Set g = g + 1 Step 6: (Elte nserton nd repetton) - Insert the eltes from Step 3 to the new populton. - If the termnton condton (.e. g = mxmum number of generton) s stsfed, determne the ftness vlue of the lst generton nd stop. Otherwse, go bck to Step APPLICATION 5.1 Test Condtons Indones hs n re of bout 7.8 Mllon sq. km., wth 75% comprsng the se re. The current trnsport system s composed of 330,495 km of rod, 515 km of toll rod nd 4,553 km of rlwy trck. There re 668 rl termnls, 31 of them mn termnls, whle the totl number of exstng se ports s 1,881 consstng of 725 publc ports nd 1,156 specl ports. Lck of mntennce hs resulted n ncresed dmge to the trnsport nfrstructure. Hevly dmged rods n the yer 2002 comprse 15% of the totl length, whle lght dmged rods comprse 23.6%. Totl rlwy length hs contnuously decresed by 10% untl the yer Bsed on the ntonl orgn-destnton survey n 2001, t could be notced tht most of the regonl freght movements re concentrted n Jv nd Sumtr Islnd, except for the movements of ol, tmber nd mnng products tht lso exst n other slnds (see Fgure 4). Almost 83% of the totl regonl freght movements were from or to res n Jv Islnd, nd the other 10% re from or to Sumtr Islnd. Surby nd Jkrt, the mn cty n Jv Islnd were observed to hve the hghest flow of freght enterng the re. 5.2 Network nd Actons The trnsport network s modelled nto 51 zones ggregted from regences nd ctes, 761 nodes nd 3,466 lnks comprsng the ntonl, provncl nd toll rods, rlwys, port-to-port connectons rl termnls, 25 seports (strtegc ports) nd 19 mn ferry ports. 2901

9 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , 2005 Generton Attrcton 400,000 ton/dy Fgure 4. Totl Generton nd Attrcton of Freght Movements n Indones (from Ntonl OD Survey 2001) Prmeter vlues for the lnks re obtned from sttstcl dt on rods, rls, ports nd termnls s culled from vrous sources. Rod cpctes nd speed dt re cqured from the dtbse of the Indonesn Inter-urbn Rod Mngement System (IRMS) nd the Indonesn toll rod opertor PT. Js Mrg. Rlwy dt nd relted nformton re obtned from the Deprtment of Communctons nd the sem-prvte rlwy compny PT. KAI. Port nformton nd other se network dt re collected from the Drectorte Generl of Se Communcton under the Deprtment of Communctons. The verge speed for the rl mode s set t 60 km/hour nd t 12 km/hour for the se mode, nd frequency nd vehcle cpcty vlues re verged from the vlble yerly trp sttstcs. Cpctes for lodng/unlodng lnks re derved from the shp hndlng cpctes of severl ports/termnls. Other dely tmes such s tme of nspecton, nventory, dmnstrtons on termnls re ssumed, rngng from 6 to 48 hours dependng on termnl type. The number of berths s derved from the port s berth length nd the verge shp length for lodng/unlodng t se termnls, whle for rl termnls t s equl to the number of yrds. The prmeters for Equton (6) nd (8) re then clbrted usng γ = 5 tht gves the best ft polynoml result. There re 16 lterntve ctons, whch nclude upgrdng or mprovement of exstng nfrstructure nd development of new ones for ll modes (see Fgure 5). Improvements, ncludng rod wdenng, re set on the ssumpton tht the overll cpcty s mproved 1.5 tmes the ntl cpcty. In generl, the development of new expresswys or toll rods s reltvely more expensve compred wth the other types, whle rl termnl mprovement s mong the chepest. 5.3 Results Before the full computtonl process s performed, the modl splt-ssgnment model (lower level problem) ws vldted to confrm the ftness of the flows resultng from the model to the rel flow dt, nd to mke djustments to the model prmeters. The rel dt used for ths vldton process s derved from vrous sources s follows: 2902

10 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , Fgure 5. Test Network nd Alterntve Actons No. Type of Acton Locton 1 Se Port Improvement Jkrt (Tnjung Prok) 2 Se Port Improvement Bnten (Cwndn) 3 Se Port Improvement South Sulwes (Mksr) 4 Se Port Improvement Irn (Sorong) 5 Se Port Improvement North Sulwes (Btung) 6 New Se Port Jmb (Mur Sbk) 7 Rl Termnl Improvement West Jv (Bndung) 8 Rl Termnl Improvement Centrl Jv (Clcp) 9 Rl Termnl Improvement Lmpung (Bndr Lmpung) 10 New Brdge Jv-Sumter Brdge 11 New Brdge Jv-Mdur Brdge 12 Rod Wdenng Lmpung-South Sumter 13 Rod Wdenng Centrl-South Klmntn 14 New Express Wy Semrng-Surby 15 New Express Wy Crebon-Semrng 16 New Express Wy Bndung-Crebon - rod mode dt s composed of trffc count dt. In the IRMS dtbse, the dt s vlble n the form of dly trffc volumes by vehcle type nd drecton. The dly trffc volumes re converted nto tonnge unt for freght vehcles nd number of pssengers for pssenger vehcles. - rlwy mode dt s composed of termnl-to-termnl flows relted to the Ntonl OD Survey dt - se mode dt s composed of port-to-port flows relted to the Ntonl OD Survey dt It cn be seen from Fgure 6 tht the ssgnment model performed well nd cn be used for the next step of computton. Though t s not perfect ft, estmted vlues from the model hve strong correlton wth the ctul vlues from the vlble dt set s ndcted by the correlton coeffcents rngng from Fgure 7 shows the best-of-generton nd generton verge result. The frst populton (generton 0) strts out wth n extremely low verge objectve functon vlue. Ths s becuse mny ndvduls n ths generton hve negtve BCR vlues (freght trnsport cost by mplementng the combnton of cton s hgher thn the ntl condton). Model Flow (Thousnd Ton/Dy) Model Flow (Thousnd Ton/Dy) Model Flow (Thousnd Ton/Dy) Dt Flow (Thousnd Ton/Dy) Dt Flow (Thousnd Ton/Dy) Dt Flow (Thousnd Ton/Dy) Rod mode, R 2 =0.70 Rl mode, R 2 =0.72 Se mode, R 2 =0.85 Fgure 6. Flow Dstrbuton Comprsons 2903

11 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , 2005 The best BCR vlue of ths populton s Better ndvduls ppered n generton 3 nd the best ndvdul ppered n generton 6. The results cn be consdered to rech convergence fter 10 genertons snce the verge vlue n ths generton (1.09) hs been close to the best vlue of At the ntl generton, the ccumulted number of evluted ndvduls ncresed ccordngly wth the number of ndvduls n ech generton. However, t grdully ncreses s t pproches convergence (see Fgure 8). In the 30th generton, the ccumulted number of evluted ndvduls reches 1222, whch s 1.86% of the totl number of possble combntons of 16 lterntves,.e. (2 16-1) = 65,535 combntons. The low rto obtned between the smpled ndvduls nd the possble combntons cn be consdered s one of the dvntges of usng ths pproch. The best ndvdul (optml soluton) contns the cton number 6, 14 nd 16 (see Fgure 5 for ther loctons) tht ncludes new seport development n Sumtr Islnd nd new expresswy development n Jv Islnd. Ths combnton of ctons would led to reducton n totl freght trnsport costs of 79.8% nd reducton n totl trvel tme of 59.5% s compred to the ntl condton before mplementton. Beneft-Cost Rto Averge Hghest Generton No. Fgure 7. BCR Vlues n Ech Generton No. of Indvdul Generton No. Fgure 8. Accumulted Number of Evluted Indvduls 2904

12 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , 2005 Although no ctons re nvolved concernng rl network or termnl mprovements, ths combnton of cton cn ncrese the mode shre of rl for freght trnsport by 16.2% f compred to tht n the ntl condtons n Jv Islnd. There re lso 2.6% of pssenger movements shftng to the use of trn fter the mplementton of these ctons. Fgures 9 nd 10 dsply the freght flow dstrbuton before nd fter the optml combnton of ctons re mplemented. It cn be found tht there re some lnks n Jv Islnd tht become very hevly loded fter the mplementton of the optml ctons. These lnks re the rlwy lnks, whch re reltvely under-utlsed n the ntl condtons. Furthermore, ths combnton of ctons reduces some overloded rod lnks tht occur n the ntl condtons whle reducng the se mode shre by 14.2%, prtculrly on the se lnks round Jv Islnd. In Sumtr Islnd, the effects of mplementng ths combnton of ctons would be better utlston of rods connectng new port nd the western prt of the Islnd. For the other res, there re no sgnfcnt mpcts before nd fter mplementton of the optml ctons. Moreover, the combnton of ctons hvng hgher BCR vlues typclly ncludes ctons undertken n Jv nd Sumtr Islnd snce the genertons nd ttrctons of freght movements re mnly concentrted on these slnds. Optml solutons obtned ndcte tht t s necessry to develop new expresswys n Jv Islnd to provde better ccess to the mn ctes nd to mprove other-thn-rod mode for freght movements. On the other hnd, the exstng freght se network s found to be less ttrctve for the user, snce t produces hgher costs thn other modes. Hgher mode shre for se mode could be relsed by developng the se trnsport system tht cn ncrese the frequency nd the fleet cpcty of se lnks s well s mprove port performnces. Note: Low Loded Lnk Moderte Loded Lnk Hevy Loded Lnk Very Hevy Loded Lnk Fgure 9. Freght Flow Dstrbuton (Intl Condtons) 2905

13 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , 2005 Note: Low Loded Lnk Moderte Loded Lnk Hevy Loded Lnk Very Hevy Loded Lnk Fgure 10. Freght Flow Dstrbuton (After Implementng the Optml Combnton of Actons) 6. CONCLUSIONS Ths pper proposed model tht cn be used s tool for strtegc level of plnnng, prtculrly n the development of freght trnsport network. It presented method for solvng dscrete network desgn problem, n ths cse selectng the best combnton of ctons for nvestment plnnng bsed on ther economc effcency. An optmston model ws developed, where multmodl mult-clss user ssgnment technque s ncorported wthn the lower level problem nd the optml combnton of ctons for cpcty expnson s determned usng GLS procedures n the upper level problem. A vrton on the GLS pproch s ppled on the ctul freght trnsport network n Indones. Results reveled tht the model usng GLS procedures dequtely found the best soluton, wth low rto between smpled ndvduls nd totl possble combntons. The best combnton of ctons ncludes new port nd expresswy development, whch could led to sgnfcnt reductons n freght trnsport costs nd ncrese rl mode shre. However, these results depend hevly on the set of possble ctons, nd therefore severl dfferent lterntves wll be ssessed n future reserches. More ccurte dt re lso necessry for the ctul mplementton of selected ctons. There s crtcl mblnce n trnsport demnd n Indones. Some prts of Indones hve very low demnd, whle the other prts hve very hgh demnd. Network desgn models bsed on economc effcency such s tht proposed n ths pper typclly only promote ctons to be mplemented n res hvng hgher demnd. Therefore, s for the nton-wde trnsport plnnng n Indones, t mght lso be effectve to nvestgte how to estblsh freght trnsport systems tht cn offer blnced development ll over the country. 2906

14 Journl of the Estern As Socety for Trnsportton Studes, Vol. 6, pp , 2005 ACKNOWLEDGMENT Ths reserch s prtly supported by the Jpn Socety for the Promoton of Scence (JSPS) s prt of the Group Actvtes on Mrne Trnsport Engneerng under the JSPS Mrne Trnsportton Progrm. REFERENCES Bell, M. G. H., Id Y. (1997) Trnsport Network Anlyss. John Wley & Sons, Englnd. Crnc T. G., Florn M., Lel J. (1990) A Model for The Strtegc Plnnng of Ntonl Freght Trnsportton by Rl, Trnsportton Scence, Vol. 24, No. 1, Dfermos S. C. (1980) Trffc Equlbrum nd Vrtonl Inequlty, Trnsportton Scence, Vol. 14, Drectorte Generl of Hghwy, Deprtment of Publc Works, Republc of Indones (1997) Indonesn Hghwy Cpcty Mnul. Florn, M., Spess, H. (1982) The Convergence of Dgonlzton Algorthms for Asymmetrc Network Equlbrum Problems, Trnsportton Reserch, Vol. 16B, No. 6, Goldberg, D. E. (1989) Genetc Algorthms n Serch, Optmzton, nd Mchne Lernng. Addson-Wesley. Guelt J., Florn M. nd Crnc, T. G. (1990) A Multmode Multproduct Network Assgnment Model for Strtegc Plnnng of Freght Flows, Trnsportton Scence, Vol. 24, No. 1, Lubs H. A., Sjfruddn A. nd Frzl, R. B. (2002) Port Plnnng wth Multmodl Anlyss, Report of Group Actvtes on Mrne Trnsportton Engneerng, JSPS Mrne Trnsportton Progrm, Hroshm Unversty, Jpn, Sheff, Y. (1985) Urbn Trnsportton Networks: Equlbrum Anlyss wth Mthemtcl Progrmmng Methods, Prentce Hll, Englewood Clffs, New Jersey. Southworth F. nd Peterson, B. E. (2000) Intermodl nd Interntonl Freght Modelng, Trnsportton Reserch, Prt C, Vol. 8, Tvsszy L. A. (1996) Modelng Europen Freght Trnsport Flows, PhD Thess, T. U. Delft. Thoms R. (1991) Trffc Assgnment Technques, Avebury Techncl, Englnd. Ymd, T., Tnguch, E. nd Nortke, M. (1999) Optml Locton Plnnng of Logstcs Termnls Bsed on Multobjectve Progrmmng Method, Urbn Trnsport V, L. J. Suchrov (Ed.), WIT Press,