behavior-oriented freight modelling

15
1 Behavior-Oriented Freight Modeling: Method and 1 Applications 2 Corresponding Author: 3 Tobias Wieczorek, PTV AG 4 Stumpfstraße 1, D-76131 Karlsruhe, Germany 5 Phone +49-721-9651-217, email [email protected] 6 Co-Authors: 7 Norbert Schick, PTV AG 8 Stumpfstraße 1, D-76131 Karlsruhe, Germany 9 Phone +49-721-9651-227, email [email protected] 10 11 Udo Heidl, PTV AG 12 Stumpfstraße 1, D-76131 Karlsruhe, Germany 13 Phone +49-721-9651-330, email [email protected] 14 15 Submission date: July 31, 2011 16 Word count: 4852 17 6602 (including Figures and Tables) 18 Figures: 8 19 Tables: 1 20 21 22 23

Upload: ronny-marcelo-aliaga-medrano

Post on 20-Jul-2016

10 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Behavior-Oriented Freight Modelling

1

Behavior-Oriented Freight Modeling: Method and 1

Applications 2

Corresponding Author: 3

Tobias Wieczorek, PTV AG 4 Stumpfstraße 1, D-76131 Karlsruhe, Germany 5 Phone +49-721-9651-217, email [email protected] 6

Co-Authors: 7

Norbert Schick, PTV AG 8 Stumpfstraße 1, D-76131 Karlsruhe, Germany 9 Phone +49-721-9651-227, email [email protected] 10 11 Udo Heidl, PTV AG 12 Stumpfstraße 1, D-76131 Karlsruhe, Germany 13 Phone +49-721-9651-330, email [email protected] 14 15 Submission date: July 31, 2011 16 Word count: 4852 17 6602 (including Figures and Tables) 18 Figures: 8 19 Tables: 1 20

21

22

23

Page 2: Behavior-Oriented Freight Modelling

2

Behavior-oriented freight modeling: Method and 1

Applications 2

3

Abstract: 4

Modeling and research in the field of freight transportation are much more complex compared to 5 passenger transport, due to specifics of the freight transportation market, the characteristics of goods and 6 vehicles and insufficient data. This paper is concerned with a behavior oriented disaggregated modeling 7 approach. Key aspects are a very detailed classification of commodities used by the impact model for trip 8 calculation and a network model containing multiple layers to account for logistics procedures within a 9 flow model. This approach has been implemented in a standardized, automated process that allowes 10 problem-specific adaptations with moderate experimental effort. So far this approach has been used to 11 develop freight models for three regions that strongly differ in terms of their geographical and economic 12 situation. 13

14

Keywords: 15

Freight traffic, modeling, behavior-oriented approach 16

17

Page 3: Behavior-Oriented Freight Modelling

3

INTRODUCTION 1

Compared to passenger transportation, freight transportation is characterized by properties that make 2 scientific research and traffic engineering models more complex (1), (2): 3

• Regarding their nature, dimensions and requirements for transportation, commodities are 4 much more heterogeneous than persons. This leads to a greater variety of transport vessels and vehicles 5 that are partly deployed for only one or few specific commodities. 6

• The market has many actors, in particular logistics companies, whose economic 7 requirements impact freight flows and thus need to be considered accordingly in freight modeling. 8

• The fact that there are special vehicles and transport companies that act separately from each 9 other lead to empty trips, which are nearly unknown in passenger transportation. In freight transportation, 10 however, they make up a significant share of the traffic volume (3). 11

• There are several capacity restraints, e.g. weight, volume, and floor space requirements (for 12 freight that cannot be stacked). 13

• Contrary to passenger transportation, freight transportation requires special technical 14 infrastructure for transshipments. The cost structure of this infrastructure, its suitability for specific 15 commodities or transport containers and its distribution across a territory have a major impact on 16 observable processes in freight transportation. 17

18 Besides freight transportation related difficulties, there are additional problems that arise from the 19

(official) practice of data collection (4). 20 • In general, freight flow data is only collected for large coverage areas (in Germany, e.g., for 21

federal states or districts). Due to the many different commodities, a later refinement down to finer spatial 22 units using general key data, e.g., number of workplaces or industrial and commercial space, normally 23 leads to unsatisfactory results. 24

• The classification into commodity groups used for official data collection, such as the NST 25 (Nomenclature uniforme des marchandises pour les statistiques de transport) is generally based on the 26 requirements of trade statistics, not logistics. That means although flour, e.g., is transported differently 27 from fat for animal nutrition, they are both listed under the same NST2007 group "Food, beverages and 28 tobacco". Yet ore and scrap metal can be transported using the same type of vessel and are listed in 29 different NST groups. For a transport model accounting for logistical correlations and causal 30 relationships, the nature of the goods and their logistical requirements are much more important than their 31 assignment to certain branches of industry. 32

Due to the aforementioned problems, modeling freight transportation is more complex than 33 modeling passenger transportation. For the same reasons, the data basis needed for modeling is only 34 partly available in the form required. In addition, in the past planning tasks often focused on passenger 35 transportation. The traffic models used only roughly calculated freight traffic to define a basic traffic 36 volume on the (trunk) road network. However, due to the necessity of considering emission and 37 immission and the rising importance of freight transportation, a paradigm shift has been on the horizon 38 for a few years now. With this new tendency, the requirements for transport planning and transport 39 modeling have increased. 40

As a result of the above-mentioned findings, many existing approaches of freight models reduce 41 the complexity and effort of data collection through the following ways: 42

• The origin-destination matrix is not generated synthetically via an impact model, but is 43 deviated from statistics and, if need be, synthetically refined. Due to the commodity groups used for 44 (official) data collection, essential information on the spatial and material breakdown of freight flows 45 (e.g. intermediate levels of manufacturing) is not accounted for. 46

• Particularly when freight models are merely meant to provide basic traffic volume data for a 47 passenger transportation model, the entire freight trip matrix is often simply projected using a single 48 percentage for forecast calculation. 49

Page 4: Behavior-Oriented Freight Modelling

4

• Further, these models have a low resolution regarding the zonage system and demand 1 segmentation which leads to unsatisfactory or incorrect results, especially during forecast calculation. 2 Other models are explicitly limited to a specific sub-segment of freight traffic. 3

• If at all, logistical processes such as vehicle utilization, shipment consolidation for major 4 transport runs, empty runs, intermodal transshipment, or the deployment of several vehicle types, are only 5 considered on a blanket basis. 6

All in all, there is a clear methodical shortfall in freight modeling compared to passenger 7 modeling, see (5), (6). There is not even a generally accepted, standardized procedure, such as the 4 step 8 algorithm for passenger modeling, e.g. (7). Due to the particularities of freight transport, this shortfall 9 cannot be overcome by simply applying the latest findings in passenger modeling to freight modeling. In 10 fact, freight modeling needs its own approaches. The following paper describes a behavior-oriented 11 approach for freight modeling developed by PTV AG and describes three models based thereupon in the 12 context of their application. 13

14

MODEL APPROACH 15

The objective was to develop a freight model with an improved forecast capability compared to empirical 16 traffic assessment, including small-scale changes in economic structure and the differences in 17 development over time for different branches of industry. Moreover, the model should generate a realistic 18 distribution of shipments among the modes of transportation, accounting for aspects such as transport 19 demand/volume per O-D relation and distance. 20

These requirements led to the decision firstly to apply a refined commodity classification, 21 secondly to use high resolution spatial data (in particular when the freight model has to be compatible 22 with a high resolution, passenger transportation model) and thirdly to include elements of logistical 23 processes in a basically macroscopic assignment model. Adaptations needed to be made to transport 24 supply modeling ( mainly to the network model) and transport demand (above all to the trip generation 25 process) as well as to the actual modeling process itself. These three elements are briefly described in the 26 following. 27

28 Network 29

Networks, which are used to model the transport infrastructure, also define transport supply in all freight 30 transport models. As mentioned before, here transshipments and multi-modal transportation play a much 31 bigger role than in passenger transportation. Often an A-B relation supply does not only include 32 connections with the transport mode X (e.g. road) or the transport mode Y (e.g. ship), but connections 33 where both transport modes are used. Moreover, supply is also determined by internal, operational 34 processes of logistics companies, which might result in transshipments within a single mode of transport 35 (e.g. shipment consolidation, with smaller trucks bringing the consignment to a larger truck for a direct 36 trip). So goods are not always transported directly from A to B. Their trip often includes major detours, 37 which are accordingly reflected by the traffic volume on the roads. The model should consider both of 38 these aspects, although logistics companies' internal operations could actually not be fully accounted for 39 using a macroscopic traffic model. 40

Thus, in the network model, the networks of the individual transport system (road, rail, ship, 41 pipeline) are depicted as layers, connected at the points of transshipment. In addition, in order to 42 differentiate between the first run, main run and follow-up run of a transport, the road network had to be 43 modeled in three layers. Figure 1 shows this schematically. 44

Page 5: Behavior-Oriented Freight Modelling

5

Figure 1: Schematic diagram of the 5-layer network model: For example, the route of a 1

consignment might start on the lower layer (senders side), source 1 (O1 in Figure 2 below-left), then continue to transfer point 1, through transshipment change to layer 3 2; continue to transfer point 2, then through transshipment reach the receivers side 4 network level and from there finally reaches destination D3. 5

The Figure shows that connections with and without transshipment can be depicted. The model 6 architecture is flexible enough to show the most relevant transport operations. For instance, assuming a 7 pertinent low cost direct link connection of a sender or recipient to a railway network , the model can 8 account for a direct rail service. When one picture the middle layers in the overview separately from each 9 other, it becomes clear that one can also model several transshipments, e.g. seagoing vessel - inland 10 waterway vessel - truck. 11

If the individual layers and the arcs connecting them (which represent the transshipments) are 12 assigned generalized costs, you can perform an inter-modal route search. 13

14 Demand 15

The use of disaggregated commodities (models based on the new approach distinguish between more than 16 100 commodities) and a refined spatial resolution, allowed to synthetically create the demand model, 17 instead of deriving it from existing empirical freight flow matrices. This approach is not entirely new, but 18 there are few models kwon to the authors with more than 20 different commodities, e.g. (5),(8). 19 Thereby the following aspects came into play: 20

• Freight transport volume in smaller traffic cells is often dominated by a single or few 21 factories, with a specific transport characteristic. This information should be maintained wherever 22 possible. 23

• Regional economy data is partly based on freight traffic, listed with a minor or even negative 24 sum (e.g. due to excavation construction, sludge, refuse), so that deriving data from monetary flows (e.g. 25 as is possible from export or import data) does not seem advisable. 26

Page 6: Behavior-Oriented Freight Modelling

6

• The information available on multi-stage production processes within the same commodity 1 group is blurred by official statistics and cannot be reconstructed by means of the coded refinements of an 2 empirically formed matrix. 3

• Major producers of a traffic cell, such as mills, sawmills, steelworks, glassworks, etc. can be 4 accounted for relatively easily by calibrating the model for the respective cell, since the economic 5 processes and traffic processes resulting thereof are actually traceable. 6

7 When classifying the commodities into groups for the models developed, the following data was 8

decisive: quantity of the goods in the area investigated, the logistical procedures, the assignment to a 9 multi-stage production process (raw material, semi-finished product, finished product), the value of the 10 goods and - for pragmatics reasons - the availability of data to assess the marginal totals of freight 11 transport volume. 12

Figure 2: Selected commodity groups and their quantities in an analysis and forecast of the 13

freight model for the United Arab Emirates 14

The freight transport demand and O-D pairs were estimated for individual commodities, using an 15 interface to structural data (when precise data was available) and specific production and consumption 16 rates. For freight transport demand data that is closely linked to the population size (such consumption of 17 foodstuff, consumer goods, gas and fuels, waste generation, etc.) this worked well. For many 18 commodities either the production or attraction side is characterized by a few major producers/consumers. 19 In these cases, it is unavoidable to research locations and production figures. The information available 20 mostly refers to the main product (e.g. steel for a steelworks). A basic knowledge about the respective 21 production process allows to estimate the production and attraction of raw materials (charcoal and iron 22 ore/junk iron), and by-products. 23

It turned out to be necessary for most commodities to develop an individual (in general, however, 24 not complex) solution in order to calculate or estimate the production and attraction totals. Moreover, it 25 was helpful to keep mass balances of each production level (output = input, with the exception of adding 26 or removing water and gases) or across several production levels (input = output of previous level). For 27 calibration the following data, mostly aggregated, from several statistics were used: marginal totals 28 (production and attraction), import and export data, transshipment data of harbors, etc. 29

Page 7: Behavior-Oriented Freight Modelling

7

Results were initially available in the form of production and attraction totals for individual 1 traffic zones in the unit tons/year. Freight distribution was calculated under consideration of a simple 2 impedance matrix. What at first might appear inaccurate compared to the rather detailed the trip 3 generation, is actually one of the advantages of the high-resolution modeling approach: Because of the 4 detailed classification of commodities, the number of possible productions and attractions mostly remains 5 low, at least for commodities with a unilateral major producer/consumer. Ideally, distance distribution can 6 then mostly be derived from network modeling details and trip generation instead of - as is common 7 practice - accounting for it in freight distribution calculation as an external variable in complexly 8 calculated impedance matrices. This is illustrated in Figure 3 in a schematic example with two major 9 producers (e.g. a mill and a meat processing company) supplying goods to receivers distributed across the 10 area (e.g. warehouses of food store chains). If the model does not distinguish between the commodities of 11 the two producers and depending on the impedance function used, this might result in completely 12 different freight flows in freight distribution calculation and accordingly lead to different distance 13 distributions. The upper half of the diagram shows a simple case in which producers deliver to the 14 receivers they are closest to. If the model, however, distinguishes between the commodities and knows 15 approximately how much of each the receivers needs, a specific interrelation is produced, independently 16 from the impedance matrix. This is depicted in the lower half of the diagram. 17

Figure 3: Example of freight distribution calculation for two major producers without 18

distinction between commodities (top) and with distinction between commodities 19 (bottom). 20

Although one cannot record and model all economic freight interrelations by applying even the most 21 detailed analysis, the approach chosen here still allows for a more traceable, transparent and realistic 22 recording of freight traffic volume than the highly aggregated models. 23

Procedure 24

In a conventional 4-step algorithm model for passenger transportation, the first steps are trip generation 25 and distribution, followed by mode choice, (where applicable) conversion of person movements in 26 vehicle movements and route choice. A freight model basically has to accomplish the same tasks. 27 However, to a certain extent, these latter three tasks depend on transport modalities, such as direct 28 transportation/consolidated transportation including transshipments, the use of special transshipment 29

Page 8: Behavior-Oriented Freight Modelling

8

facilities, size of containers, etc. All this is summarized in the described model under "logistics system". 1 Logistics systems are subsystems of freight transport supply and have their own characteristic processes. 2 To distinguish between these systems, different types of storage vessels as a distinctive feature were used 3 and additionally a difference between food and non-food products was made. The sole exception is the 4 transport container. The most important logistics systems are: tank transportation, bulk material, non-food 5 products, pallet transportation, fresh pallet transportation, containers, packages/bagged goods, special 6 purpose trucks. For each logistics system attributes in the network model had to be specified, controlling 7 which links (or layers, see Figure 1) and transshipment locations it could use. Once selected, the logistic 8 system then has an influence on mode choice and route choice. The selection of a logistics system itself is 9 influenced by the commodity transported. Many commodities can be unambiguously allocated to a 10 logistics system (e.g. fresh concrete is only transported by truck). For some commodities an allocation to 11 several logistics systems is possible (e.g. fuel, transported by truck as well as rail or ship). In such cases, 12 the criteria used for allocation can be transport distance, demand per O-D pair, or border crossing. 13

To account for this fact, an additional aggregation level for freight transportation was introduced, 14 the shipment, see (9), (10) for a similar approach. It was added between the unit (metric) ton (used to 15 calculate the demand for individual commodities) and the unit vehicle (for trip assignment per mode of 16 transport in the network). In a logistics system, shipments typically represent the consolidated quantity of 17 a commodity transported. Accordingly, typical weight and volume limits for all shipment units in the 18 model were defined. Examples of shipments that do not come in tons or vehicle loads are pallets, 19 containers, and packages. 20

These shipments are assigned in a multi-layer model where mode choice and route choice take 21 place simultaneously. For route choice impedance, costs consisting of two major factors are relevant: 22

• transport costs incur by transport distance and transport time, plus loading and unloading at 23 the beginning and end of route as well as transshipments along the route. The cost structures of the 24 network layers differ. Normally rail and inland waterway transportation are less expensive than truck 25 transportation in terms of distance and time related costs of the transport run. However, they are more 26 expensive when it comes to the time required for loading and unloading. 27

• Sender costs account for the value of time of the sender. They occur through loss of 28 shipment value during the transportation of perishable goods and through depreciation when transporting 29 high-value goods. This means that faster (but more expensive) modes of transport are preferred for 30 perishable or high-value goods, while inexpensive (but slow) modes of transport tend to be preferred for 31 long-lasting or low-value goods. 32

33 The longer the transport distance, the more competitive more efficient modes of transportation, such as 34 railways, become. Their lower variable costs even make connections with expensive transshipment 35 transactions more competitive (Figure 4). 36

Page 9: Behavior-Oriented Freight Modelling

9

Figure 4: Effect of different cost structures (distance related and non-distance related costs) of 1

individual network layers. The resultant (continuous line) shows additional 2 transshipments with increasing distance, the selection of more efficient modes of 3 transportation (railways), and a decreasing trend in costs (costs/km). 4

As a result of multi-modal assignment in a multi-layer model, you receive a set of unimodal routes for 5 each layer that can be used to generate mode-dependent demand matrices. For these one can convert the 6 shipments into vehicle units, using standard vehicles with an average capacity and utilization rate. Empty 7 trips are calculated based on the level of the mode of transport, whereby two cases are distinguished: 8

• For trucks with a special purpose body (e.g. for fresh concrete, milk, or heating oil) the 9 option of taking on a load for the return trip is generally excluded. Here the return trip is assumed to be an 10 empty trip. 11

• In other cases, a new load can be taken on for the return trip. However, this is only possible 12 within the same logistics system (to ensure that flour and cement, e.g., which are both transported in 13 pressure-silo trucks, do not form a trip pair). Based upon the total of beginning and ending loaded trips 14 per traffic zone the amount of empty trips is calculated. With the number of empty trips, another freight 15 distribution calculation is performed. 16

17 The result of this calculation provides you with mode-dependent demand matrices. In a one-layer 18

network, these can be assigned in order to calculate link volumes. 19 This means that three different reference systems are successively used throughout the entire 20

modeling process (see TABLE 1). 21

TABLE 1: Reference values of the model 22

Reference value Unit Application

Commodity Tons per year Calculation of freight transport demand per zone and demand per O-

Page 10: Behavior-Oriented Freight Modelling

10

D pair

Logistics system Shipments per year Calculation of transport demand in a multi-layer model and thus of mode and route choice

Transport mode Vehicles per day Calculation of number of trips (including empty trips) for individual modes of transportation and of link volume

1

Instead of directly converting tons into vehicles, the model first looks at the logistics systems. This 2 additional step allows the model to account for logistics processes and regionally differentiated freight 3 flows. 4

APPLICATION 5

Freight model for Switzerland 6

This was the first model for which the behavior-oriented approach was developped. The high spatial 7 resolution with over 3000 traffic zones was given as the model needed to be compatible with an existing 8 passenger transportation model. In this case, it was extremely helpful that high-resolution statistical data 9 on occupation and land use was available in excellent quality. 10

Figure 5 is an overview of the network model and the assignment results obtained for heavy 11 goods vehicles. You can see the high-resolution road network and the dominating freight flows north of 12 the Alps, in the transport corridor Geneva - Lausanne - Bern - Zurich, including the concentration of 13 traffic passing the Alps on limited pass roads. 14

Page 11: Behavior-Oriented Freight Modelling

11

Figure 5: Freight model Switzerland, VISUM assignment results for heavy goods vehicles on 1 road network 2

Figure 6 shows an example of the results obtained when using more detailed commodity groups. You can 3 see the assignment of the logistics system "tank transportation and mineral oil products", with inland 4 waterway transportation (dark blue, on the Rhine river), railway transportation (purple), and fine trip 5 distribution per truck (light blue). 6

Figure 6: Demand assignment of logistics system "tank transportation and mineral oil 7

products" in 5-layer model (Switzerland model). 8

Freight model for the United Arab Emirates 9

This model was part of a national transport development plan. As in the Swiss model, the traffic zones 10 were already defined through a passenger transportation model. In contrast to Switzerland, freight 11 transportation in the Emirates is dominated by the following three sectors: construction, oil and gas 12 industry, and consumer goods. There is no significant production in other sectors and the logistics 13 processes are not that complex. This is why the methods applied in this model are less complicated. 14

From the start, the main focus was to build up a forecast of transport development for the next 15 few decades, as well as to estimatethe possibility of establishing a railroad network. In both cases, 16 detailed commodity groups were needed (cp. Figure 7). 17

Page 12: Behavior-Oriented Freight Modelling

12

Figure 7 shows an overview of the model, including the results for truck assignment. You can 1 easily recognize Dubai's residential area, with Sharjah and Ajman in the middle, Abu Dhabi in the south-2 west, and AlAin in the south. The freight flows run parallel to the coast, but more towards the back 3 country than passenger car traffic does (which is not depicted here). In the north, they are determined by 4 trips with construction material to the construction sites in Dubai. The south is dominated by trips towards 5 Saudi Arabia and to the oil and gas production facilities in the country's west. 6

Figure 7: Freight model UAE, VISUM assignment results for truck transportation 7

Freight model for analysis of the Traceca corridor 8

The name Traceca stands for the EU-funded analysis of a transport corridor that reaches from Eastern 9 Europe, via the Baltic States to the central Asian states. The main point of interest was the freight flows 10 running across a large area. Modeling was difficult due to the poor quality of data available, the large 11 differences in economic development of the individual countries, and their different consumer behavior 12 based on culture, religion and geography. Moreover, wars and conflicts between the corridor countries led 13 to shifts in freight flows that cannot be derived based on transport facts alone. 14

Page 13: Behavior-Oriented Freight Modelling

13

Figure 8: TRACECA freight model, network with roads, railroads and shipping routes. 1

Figure 8 shows the large geographical area of the network model, stretching from Central Europe via the 2 Baltic and Caspian Seas to China. Although this model strongly differed from the others in terms of level 3 of detail, input data, and objective, the modeling process described could still be successfully applied. The 4 Traceca model was also different in another respect, namely that it required further methodical 5 development: In the Swiss and the Emirates model, capacity restraints did not play a role. In the Traceca 6 model, however, these had to be accounted for, in particular for harbor transshipments. To account for 7 this fact, capacity-dependent cost supplements and time supplements for capacity overload were allocated 8 to the network elements. As a result, in an iterative assignment process, this led to a shifting of the freight 9 flows. 10

Previous experience 11

The Swiss model was the first model based on the approach described and was thoroughly checked by the 12 project group after completion. These checks revealed the following primary findings (11): 13

• After performing a sequence of sensitivity analyses using different input values (partly 14 exaggerated on purpose), one could see that all effects produced by the model pointed in the "right", i.e. 15 the plausible direction. 16

• Data collection and estimation of the effect parameters represented an enormous effort, 17 particularly when it came to updating the model according to development or to performing a forecast or 18 scenarios. 19

• Effects models for converting tons, shipments and trips are not based on decision models, but 20 mainly on simple breakdowns. Particularly the assignment of commodities to logistics systems has a 21 major impact on mode choice. 22

The modeling process was already standardized with the Swiss model. It was realized in VBA 23 script, which also triggers PTV VISUM to calculate transport distribution and assignment. Its modular 24 structure allowed for a relatively smooth adoption of the approach in the other two models, Emirates and 25 Traceca, although they differ largely in terms of data quality, data availability, area size, and objectives. 26

Page 14: Behavior-Oriented Freight Modelling

14

The modular structure further allowed for a fairly easy extension of the method in order to account for 1 capacity restraints. 2

The architecture of the network model also enables the modeling of various multi-modal 3 transport chains and cost structures. Additional criteria though, such as reliability or punctuality cannot be 4 modeled. This would require additional decision models. 5

SUMMARY AND OUTLOOK 6

Taking a specific use case as a starting point, a process for modeling freight transportation was 7 developped with a focus on behavior-oriented modeling using effects models to convert structural data, 8 ton flows, shipments, and trips. As for every transport model, the quality of the results produced largely 9 depends on its input parameters, i.e. the network and structural data. This is true even when bringing in a 10 new model architecture. Based on long years of experience with traffic/transport modeling, the authors 11 believe that models that combine "high-quality input data with a pragmatic estimation procedure for 12 effects models" generally produce better results than highly devised effects models with an insufficient 13 data basis. Therefore, when first implementing the described approach, the authors focused on the 14 understanding of real economic processes as well as on the readiness to produce professional estimates to 15 create a good data basis for transport generation, wherever there was a lack of data. Priority wasto 16 improve the first steps of the modeling process. Of course, this does not mean that there is no room for 17 improvement in the other steps of the model, e.g. through more complex methods, particularly for mode 18 and route choice. First steps in this direction have already been taken by accounting for capacity 19 restraints. 20

That difficulties with this approach arise in connection with forecasts or scenario calculations is 21 less due to the approach itself than to the fact that it reveals the general problems of freight transport 22 forecasts, since it tries to realize connections causally instead of functionally. Further development of this 23 approach will show where its possible applications and limits lie. 24

25

REFERENCES 26

1. J.D. Ortúzar, L.G. Willumsen, Modelling Transport, Thrid Edition, Wiley, Chichester, 2001 27

2. Markus Friedrich, Thomas Haupt, Klaus Nökel, Freight Modelling: Data Issues, Survey Methods, 28 Demand and Network Models, Conference Paper, 10th International Conference on Travel 29 Behaviour Research, Lucerne, 2003 30

3. José Holguín-Veras, Gopal R. Patil, An integrated Commodity based / empty Trip Freight Origin-31 Destination Synthesis Model, CD-ROM. Transportation Research Board of the National 32 Academies, Washington D.C., 2007 33

4. Gernot Liedtke, Jola Babani, Hanno Friedrich, Identifikation von Tourtypen in 34 Fahrzeugtagebüchern, Wirtschaftsverkehr 2011: Modelle – Strategien – Nachhaltigkeit, Verlag 35 Praxiswissen, Dortmund 2011, pp. 55-75 36

5. Carlos Bastida, José Holguín-Veras, Freight Generation Models Comparative Analysis of 37 Regression Models an Multiple Classification Analysis, In Transportation Research Record: 38 Journal of the Transportation Research Board, No. 2097, Transportation Research Board of the 39 National Academies, Washington D.C., 2009, pp.51-61 40

6. Michael D. Anderson, Gregory A. Harris Kevin Harrison, Using Aggregated Freight Data to 41 Model Freight in a Medium-Sized Community, In Transportation Research Record: Journal of the 42 Transportation Research Board, No. 2174, Transportation Research Board of the National 43 Academies, Washington D.C., 2010, pp.39-43 44

Page 15: Behavior-Oriented Freight Modelling

15

7. Werner Schnabel, Dieter Lohse, Grundlagen der Straßenverkehrstechnik und der 1 Verkehrsplanung, Bd.2 Verkehrsplanung, 3.Auflage, Beuth Verlag / Kirschbaum Verlag 2011, 2 pp. 459-488 3

8. Jose A. Sorratini, Robert L. Smith Jr., Development of a Statewide Truck Trip Forecasting Model 4 Based on Commodity Flows and Input-Output Coefficients. In Transportation Research Record: 5 Journal of the Transportation Research Board, No. 1707, Transportation Research Board of the 6 National Academies, Washington D.C., 2000, pp.49-55 7

9. Hani S. Mahmassani, Kuilin Zhang, Jing Dong, Chung-Cheng Lu, Vishnu Charan-Arcot, Elise 8 Miller-Hooks, A Dynamic Network Simulation-Assignment Platform for Multi-Product 9 intermodeal Freigth Transportation Analysis. CD-ROM. Transportation Research Board of the 10 National Academies, Washington D.C., 2007 11

10. José Holuín-Veras, Miguel Jaller, Lisa Destro, Xuegang Ban, Catherine Lawson, Herbert S. 12 Levinson, Freight Generation, Freight Trip Generation, and the Perils of using constant Trip 13 Rates, Transportation Research Board of the National Academies, Washington D.C., 2011 14

11. Bundesamt für Raumentwicklung (ARE), Eidg. Departement für Umwelt, Verkehr, Energie und 15 Kommunikation (UVEK), Nationales Güterverkehrsmodell des UVEK Modellbericht und 16 Validierung, Abschlußbericht, Bern 2011. 17