“A Decision Support System For Intermodal Transportation Networks Management”

This paper specifies a Decision Support System (DSS) devoted to manage Intermodal Transportation Networks (ITN). With the aim to support decision makers in the management of the complex processes in the ITN, we describe the architecture of a DSS and

Please download to get full document.

View again

of 6
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.


Publish on:

Views: 5 | Pages: 6

Extension: PDF | Download: 0

  A DECISION SUPPORT SYSTEM FOR INTERMODALTRANSPORTATIONNETWORKS MANAGEMENT Maria Pia Fanti (a) ,Giorgio Iacobellis (b) ,George Georgoulas (c) ,Chrysostomos Stylios (d) ,Walter Ukovich (e) , (a) Dept. of Electrical and Electronic Engineering -Polytechnic ofBari (b,d,) Dept. of Informatics and Telecommunications Technology, Technological Educational Institute of Epirus (c) Teorema EngineeringS.r.l., Area Science Park Basovizza, Trieste, Italy (e) Dept. of Engineering and Architecture-University of Trieste (a) fanti@deemail.poliba.it, (b) iacobellis@deemail.poliba.it, (c) georgoul@teleinfom.teiep.gr , (d) stylios@teiep.gr, (e) ukovich@units.it, ABSTRACT This paper specifiesaDecision Support System(DSS) devoted to manage IntermodalTransportation Networks (ITN). With the aim to support decisionmakersin the managementof thecomplexprocessesinthe ITN,we describe the architecture of a DSS and itsmain components.In order to obtain a generic DSS, weemploy the Unified Modeling Language (UML) todescribe the components and the architecture of theDSS and we apply the solutions based on ServiceOriented Architecture,the simulation drivers and window services.Keywords: Decision Support System, IntermodalTransport Network, Unified Modeling Language,Simulation. 1.INTRODUCTION An Intermodal Transportation Network (ITN) is defined as a logistically linked system integrating differenttransportation modes (rail, ocean vessel, truck etc.) tomove freight or people from one place to another in atimely manner (Boschian et al. 2011;Crainic and Kim2007;Macharis and Bontekoning 2004). There is agreat availability of various alternative means and routes and so the complexity of ITN is continuouslyincreased. On the other hand the availability of the newInformation and Communication Technologies (ICT)that could be usedto support the Decision Makers(DM) and the huge amount of acquired information,require the development of new models and methods for decision support(Coronado et al. 2009,Giannopoulos2004). It is widely accepted that ITN decision makingis a very complex process, due to the dynamical and large-scale nature of the intermodal transportationchain, the hierarchical structure of decisions, as well asthe randomness of various inputs and operations.Researchers have followed similar approaches fromother application areas (e.g., production processes), and they have identifieddifferent hierarchical/functionallevelsfor transportation systems(Dotoli et al. 2009,Caris, Macharis, and Janssens 2008).More precisely thelevels are the following: a) the strategic level,related tothe long-term definition of the transportation network,to the selection of the different transportation modesand to the evaluation of the feasible flows (capacities of the nodes and of the arcs); b) the tactical level,related (on a middle-short term) to the management of logisticflows connected to the information flow and to thetransportation network that is topologically and dimensionally defined at the higher (strategic) level; c)the operational level,includingreal-time decisional processes thatconcern the resource assignment, thevehicle routing definition, and so on. In particular,assuming a real time availability of the informationregarding the conditions of the network (likeunexpected requests of transportation, variations in theavailability of the transportation system, road conditionsand traffic flows), operational decisions should be takenin a dynamic context.Since intermodal transportation is more data-intensive than conventional transportationmeans,themodern ICT tools help to produce, manipulate, store,communicate, and/or disseminate information and  provide useful information about the state of the systemin real time and therefore manage and change on-line paths, vehicle flows, orders and deliveries. Dotoli et al.(2009) proposedan integrated system that isbased on areferencemodel and a simulation moduleto support thedecision making process. The integrated systemtracksthe state changes from the real ITN and evaluates performance indices typical of the tactical and real timemanagement, such as utilization, traffic indices and delivery delays (Viswanadham,1999).This paper takes into accountthe integrated system proposed in (Dotoli et al. 2009), and it specifies aDecision Support System (DSS) devoted to manageITNs fortakingtactical decisions, i.e., in an off-linemode, and operational decisions, i.e., in real time.Indeed, due to the complexity of the system, the DMneeds support during thedecision making process.There are proposed typical methodologies for decisionmaking, which depend on the type of ITN problemsthat are addressed. Moreover, suitable optimization Proceedings of the European Modeling and Simulation Symposium, 2012978-88-97999-09-6; Breitenecker, Bruzzone, Jimenez, Longo, Merkuryev, Sokolov Eds.150  algorithms can be used for the definition of new longterm policies, and employ the treatment of massivevolumes of data to address multicriteria problems. Inany case, the utilization of computer-based approachesto support the decision making procedures is anecessity. Different type of computer-based systemshave been developed that are used widely according tothe activity and the type of problem that they support.This paper proposesagenericDSSthat is devoted to manage complex ITN systems,based on theUnified Modeling Language (UML) (Miles and Hamilton2006),a visual modeling language that is suitable todescribe and specify software engineering.Moreover,we specify the components and the architecture of theDSSand the corresponding software tools. In addition,allsolutions included in the proposed DSS arebased onaService Oriented Architecture(SOA)approachthatguaranteesthe interoperability of the simulationand optimizationdrivers. 2.THE DSS STRUCTURE In this section,we describe the main componentsof the proposed DSS.UsuallyDSSs are categorizedon the basis of different characteristics of the systems,e.g. whether they are for personal or group oriented decision making(Gorry and Scott Morton 1971; Keen and Scott Morton1978; Delen et al. 2010; Power 2002). Based on thetype of the application,DSSs are divided in desktop and web based applications. Despite the different categoriesof DSS, all of themshare common characteristics; i.e., atypical DSS should include four main components: thedata component, the model component, the decisioncomponent, and the interface component. Here, aUMLclass diagramis used to describe the main structureof the DSS.In particular,UML has standard notation and syntax and is composed of thirteen main types of diagrams, each one serving a different purpose and describing the systemfrom different points of view. Weconsider twoviewsof the DSS: the structural view isdescribedby the package and class diagrams thatillustrate the different types of objects which the systemconsists ofand their relationships; the behavioral view isdescribed by activities diagrams that describetherules that the system follows to operate in a completeand correct manner, to avoid misunderstanding on boththe user side and the developer side. 2.1.The DSS Components Figure 1 showsthe UML class diagram of the proposed DSS architecture. More precisely, each class isrepresented by a rectangular box divided intocompartments. The first compartment holds the classname, the second holds attributes and the last holdsoperations. More precisely, attributes are qualities and named property values that describe the characteristicsof a class. In addition, operations are features thatspecify the class behavior. Moreover, classes canexhibit relationships that are represented by differentgraphic connections: association (solid line),aggregation (solid line with a clear diamond at one end),composition (solid line with a filled diamond at oneend), inheritance or generalization (solid line with aclear triangle at one end), realization (dashed line with aclear triangle at one end) and dependency (dashed linewith an arrow at one end). By class diagrams eachcomponent can be modeled as a different classillustrating the different types of objects that the systemcan have and their relationships.Figure 1: The DSS StructureIn the following we briefly specify the DSS componentsshown in Fig. 1. Data component. This moduleis a database thatcan be denoted as internal and external data base. Thedata are internalif they come from organiza tion’s internal procedures and sources such as products and services prices, recourse and budget allocation data, payroll cost, cost-per-product etc.Moreover, externaldata are related to thecompetition market share,government regulations etc. and may come from variousresources such as market research firm, governmentagencies, etc. In some cases the DSS can have its owndatabase or it may use other organizational databasesthat can be connected directly with them. Model component. Thiscomponent mainlyincludes a simulation model, a mathematical model, and a set of optimization algorithms suitable to analyzeeffects of choices on the system performances. Themodels describe the operations at different managementlevels and the type of functions varies withtheoperation that theysupport. Interface component. This module isthe part of the DSS that is responsible of the communication and interaction of the system with the DM.Such acomponent is very important because regardless of thequality and quantity of the available data;theaccuracyof the model is based on this interface. Indeed, thiscomponent includes an Information CommunicationSystem (ICS) that is able to interact withthereal systemand maintainsthe consistency between the stored dataand the real system. Decision component .Thiscomponent consists of two second level classes:the operationaldecisionclass Proceedings of the European Modeling and Simulation Symposium, 2012978-88-97999-09-6; Breitenecker, Bruzzone, Jimenez, Longo, Merkuryev, Sokolov Eds.151  and the tactical decision class. Moreover, such classesincludethe performance indices that have to beconsidered in order to take the decisions. In addition, inrelation with the performance indices and the object of the decisions, the DSS hasto collect the decision rulesand the optimization procedures that are used by themodel and the simulation component. 3.THE DSS ARCHITECTURE This section describes in detail the DSS components presented inSection 2. Moreover, Fig.2shows thearchitecture of the realized systemby enlighteningtheconnections among the modules.Figure 2: The DSS architecture. 3.1.The Data Component We distinguish three different kindsof data. Thefirstdata are managed by the Data Base ManagementSystem (DBMS)that stores theinternal data used by thedecisionand the simulation components. More precisely, the DBMS stores therequests of a newsimulation with its input data, thedata related tointernal variablesof the simulation model, and theoutputs produced by simulations.Furthermore, theDBMS contains the queue of the requests to be sent tothe simulation server, the state of each requestand theresults ofthe simulation. Moreover, it stores thedescription of all the simulation models that areavailable for the simulation runs.The second and third kind of data are stored intheData System:the internal data and the external data.The internal data represent all data necessary to describethe internal procedures, e.g. the time required for eachactivity, the number of available resources, the capacityof parking areas and safety levels, etc.On the other hand, the external data areinformation coming in realtime from the system: the current number of vehicles,theinformation about the conditions of the roads, theaccidents, the road maintenance works, the weather conditions,etc. 3.2.The Model Component The model component is the core of the DSS: it consistsof the simulation modeland the optimizationcomponent. 3.2.1.The simulation model The simulation modelmimes the system, applies theoptimization strategies proposed by the optimizationmodule and provides the performance measures.Inthe proposedsolution, the simulation model isimplemented by the Arena servers. In particular, eachArena server consists of the followingmain elements: x TheARENA software. In such a server thesimulation models are implemented and executedin the Arena Rockwell environment(Kelton 2009). x The Windows service. The service isautomatically started, it monitors the DBMS,as soon as a new simulation request is loaded,the service executesthe simulation by thearena driver and loadsresults on the DB. x ARENA driver. This component is the driver that connects the system to the softwareARENA:it is in charge of startingand stoppingsimulations.Bythe proposed architectural solution, we can provide a set of Arena serversand each server isindependent from the others. Then,the numberof operative Arena servers candynamicallychange on the basis of the computational effort that isrequired in realtime. Moreover, such anapproach allowsa parallelexecution of the simulation requestsby reducing thetotal execution time. 3.2.2.The Optimization Component The second basic module of the model component is theoptimization module (see Fig. 1)that combines avariant of Particle Swarm Optimization (PSO)(Kennedy 1995) with an Optimal Computing BudgetAllocation(OCBA) scheme (Chen 2000). In particular,the srcinal PSO algorithm was introduced in 1995 byEberhart and Kennedy (Kennedy 1995). Its mainconcept includes a population, called a swarm, of  potential solutions of the problem at hand, called the particles, probing the search space. The particlesiteratively move in the search space with an adaptablevelocity, retaining in a memory the best positions theyhave ever visited, i.e., the positions with the lowestfunction values (consideringonly minimization problems). The exploration capability of PSO is promoted by information exchange among particles.More specifically, each particle is assignedto aneighborhood. In the global PSO variant, also known as gbest  , the neighborhood of each particle is the wholeswarm and the overall best position is the maininformation provider for all particles. On the other hand,in the local PSO variant, also known as lbest  , theneighborhoods are strictly smaller, usually consisting of a few particles. In such cases, each particle may have itsown leader that influences its velocity update.In real MBMSDATA SYSTEMDGMS -Windows service-ArenaDriver-ArenaRockwell DBMS -Requests-Simulationresults-Settings Data    ARENA servers PollingOutputPushnewrequestGetresults-WebSerivce-TCP socket-WatcherCSV/XML files   -Wa   t   c erCSV/XML es ICS REAL SYSTEM -DecisionComponent-OptimizationComponent   -- USER   USER Simulation Module DSS Proceedings of the European Modeling and Simulation Symposium, 2012978-88-97999-09-6; Breitenecker, Bruzzone, Jimenez, Longo, Merkuryev, Sokolov Eds.152  life problems we usually accept good enough solutionsinstead of the globally optimal solution. Therefore, in case of “noisy”functions, “following” the “best”  particle becomes a bit involved since the actual value of a particle is obscured by noise and repeated functionevaluations are required in order to more accurateestimate the true value. Especially in situations where afunction evaluation is a costly process,a compromiseshould be reached between the need for an accurateestimate of the true value and the need to have as few as possible unnecessary function evaluations.The OCBA was proposed by Chen (2000) as a procedure to optimal allocate a predefined number of trials/replications in order to maximize the probabilityof selecting the best system/design: allocate replicationsnot only based on the variance of the different designs but also taking into account the respective means. Morespecifically, the noisier the simulation output (larger variance), the more replications are allocated whilemore replications are also given to the design that itsmean is closer to that of the best design.The global version of the PSO uses the meanvalues for each performance measure that are evaluated  byalownumber of replications. Then using the OCBA procedure more replications are allocated in order toincrease the probability of correctly identifythe best particle,whose value is used to guide the search of theswarm. 3.3.Interface Component In the presented architecture we implement twointerfaces:the first interface connectsthe simulationmodule withthe Model Base Management Server (MBMS)by means ofthe DialogGeneration/Management Server (DGMS)module;thesecond one connectsthe DSS and the real systemthrough the MBMS.In particular, the ICS module (see figure 2)represents the information system of the wholeinfrastructure and is the interface between the realsystem and the information system. It updates thesystem status stored in the Data System in real time.The DGMS allowsthe communication between thedecision component and the simulation component. Itreceivesrequests from the decision component such asrunning a new simulation or retrievingthe results of afinished simulation. All requests coming from theMBMS are loaded into the DBMS by the DGMS. Inorder to guarantee the modularity of the architecture and the possibility to execute the DSS components under different platforms, we implement the following threedifferent waystocommunicate with the DGMS: x Web Service(WS). The WScan trigger a newsimulation, provides information about thestate of the request and gives the simulationresults. A web service is a processrunning on aserver andallowinga client to execute a process on the server. This kind of applicationcommunicates by the Simple Object AccessProtocol (SOAP). x TCP Socket. This applicationis a processrunning on the server that createsa listener ona preconfigured communication gate. TheMBMSsendsmessages to DGMS by the TCPchannel. These messages are strings codified  by a prefixed communication protocol. By thisapplication the MBMS can request a newsimulation or access to stored data. x File Watcher .This process monitors the fileson a shared folder. As soon as a new file isloaded by anFTP client,it analyses thecontents of the file and executes thecorresponding operations. When a simulationgoes to an end,the file watchercreates a reportfile inthe same folder so that all data areavailable to the client.Furthermore,the MBMS shows data to the DM: if the current performance of the system is notsatisfactory,then the DM can decide to evaluate the impact of some decision using a “what if” approa ch.The MBMS allows the DM to run asimulation directlywithout the decision module control. This proposed solution is very useful for the DM and contributes togeneralize all the features offered by the DSS. 3.4.The DecisionComponent The MBMS server implementsthe decision component.Inparticular,it monitors the system statestored in theData System component and decides if it is necessary torun a new simulationor not.Indeed,ifthe performancesof the systemdecrease, thenthe decision componentstartsthe optimization and thedecision procedure.Figure 3 shows the used decision approach that is based on the optimization module and the validation by thesimulation. More precisely,the optimization algorithms propose some solutions that are sentto the simulationmodule. The ICS provides the current state of thesystem by returning the values of the variables thatcannot be determined by the DSS.Then the DSSinvokes the simulation module:thesimulation starts and applies the proposed managementstrategies. Theobtained performance indices allow evaluating theimpact of the proposed solution on thesystem.Then anew set of candidate variables are passed to thesimulation model and the process continuestill thealgorithm leadsto a satisfactory decision described by aset of candidate variables. In this model,the candidatevariables arethe controllable inputs thatmay change inthetactical or operational decision making process.In the proposed DSS,the hybrid optimizationmodule (PSO+OCBA) is connected by a web service tothe simulation module. The optimization module sendsinputsand thenumber of replications for each candidatesolution/design and itrecordsthe outputs, again throughthe web service.The average values of the outputs are sent to theOCBA procedure that decidesif anumber of extrareplications are required.The average outputs are used to guide the PSO procedure:the process is repeated tilla goodnumber of available replications is obtained. Proceedings of the European Modeling and Simulation Symposium, 2012978-88-97999-09-6; Breitenecker, Bruzzone, Jimenez, Longo, Merkuryev, Sokolov Eds.153  Figure 3: The Decision approachOn the basis of the current state of the system and the design variables, the DSScan be used to takedecision both at the tactical andoperational levels. In particular,at the tactical level the input variables aregenerated stochastically, whileat the operational levelthe inputs come from the ICT tools of the real systems. 4.THE PILOT CASE The proposed DSS has been realized to manage attactical level anITNcomposed ofthe port of Trieste(Italy),the dry port of Fernetti,and the railway stationof the Roll on-Roll off freight trains. Mainly, the DSSdeals with the decisions aboutthe import and exportflow of goodsandthe containers management and transportation. The DSSis designed in the framework of the SAIL Project, sponsored by the EuropeanCommission under the 7th Framework Program,Specific ProgramPEOPLE-Marie Curie Actions.Figure 4: ARENA MODELIn order to specifythe simulation model inBoschian et al. (2011),the authors identified the mainevents occurring in a transportation systemand describedthe main classesof the system: the port area,the inland port area, the road connection system and therail network system. Furthermore, someinterviews are previously performed totheactors involved in the process and thenecessary data and the relations arecollected.Theinvolved resourcesand their availabilityare identifiedand the followingperformanceindices areconsidered: x throughput,i.e.,the number of units boarded ina time units; x average lead time,i.e.,the time elapsed fromthe arrivalof a new unitstill its system output; x the average resources utilization; x the average queue lengths of the users thathave to acquire resources.Figure 4 shows a sketch of the simulation modelrealized in the ARENA environment. The DSS allowsusto identifyall the weakness and thebottlenecksof thesystem.Moreover, someoptimization strategies are proposed and analyzed at tactical level. 5.CONCLUSIONS This paper specifies a Decision Support System devoted to manage Intermodal Transportation Networks and totake tactical and operational decisions. We describe theDSS architecture and, in particular, the decision processthat is based on two main modules: the optimization and the simulation module. Moreover, weshow how theDSS is designed by applying SOA and web server approaches.Hence, the obtained DSS can be realized ina distributed framework in order to face the systemcomplexity.Future work will describe in more detail thedecision procedures of the DSS. ACKNOWLEDGMENT This work is supported by the E.U FP7-PEOPLE-IAPP-2009, Grant Agreement No 251589, Acronym:SAIL. REFERENCES Boschian, V., Dotoli, M., Fanti, M.P., Iacobellis, G. and Ukovich, W., 2011. A metamodeling approach tothe management of intermodal transportationnetworks.  IEEE Transactions on AutomationScience and Engineering , vol. 8: pp. 457-469.Caris, A., Macharis, C., and Janssens, G.K., 2008.Planning Problems in Intermodal FreightTransport: Accomplishments and Prospects. Transportation and Planning Technology , vol. 31: pp. 277-302.Chen,C.H., Lin, J., Yücesan, E. and Chick, S.E., 2000.Simulation Budget Allocation for Further Enhancing the Efficiency of Ordinal Optimization.  Journal of Discrete Event Dynamic Systems:Theory and Applications , Vol. 10: pp. 251-270.Coronado, A.E., Coronado, E., and Lalwani, C.S., 2009.Wireless Vehicular Networks to Support Road Haulage and Port Operations ina MultimodalLogistics Environment. Proceedings of  IEEE/INFORMS International Conference onService Operations, Logistics and Informatics , pp.62-67, July 22-24, Chicago (IL, USA).Crainic, T.G. and Kim, K.M., 2007. Intermodaltransportation. In: C. Barnhart and G. Laporte, ed. (Transportation)Handbooks in Operations Research and Management Science , vol. 12.Amsterdam: North-Holland/Elsevier, 467  –  537.Delen, D., Sharda, R., and Turban, E., 2010.  DecisionSupport and Business Intelligence Systems. 9th Edition . New Jersey, USA: Prentice Hall.Dotoli, M., Fanti, M.P., Iacobellis, G., and Mangini,A.M., 2009. A First Order Hybrid Petri Net Model Arena simulation Module Optimization ModuleCurrentState of the systemPerformance Indicators Candidate variables Proceedings of the European Modeling and Simulation Symposium, 2012978-88-97999-09-6; Breitenecker, Bruzzone, Jimenez, Longo, Merkuryev, Sokolov Eds.154
Related Search
Similar documents
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks