Building intermodal freight transport services

Chapter 6

Building intermodal freight transport services Chapter outline 6.1 Assumptions for the model 6.2 Model architecture 6.2.1 Environment 6.2.2 Agents 6.2.3 Engines 6.2.4 Interactions

161 163 164 166 182 187

6.3 Verification and validation of ABM models References

192 195

This chapter introduces TransFreight, a generic freight transport market simulator. TransFreight incorporates most of the conceptual elements discussed in Chapter 3 with regard to freight transport agents and freight transport chains. It was created using the principles of agent-based modeling as described in Chapter  6. The presentation follows the natural organization of agent-based models into environment, agents, and interactions. The environment is defined in accordance with the Russel and Norvig (2003) taxonomy. The description of the agents is effected in two steps—attributes and tasks—instead of simply presenting each basic building block. One should note that although every replication of an agent shares the same attributes (and tasks), some of them are randomly generated at the moment of creation, which may result in agents having different behaviors. The description of the interactions will be divided into four streams, each one corresponding to a type of flow that occurs on an intermodal transport chain, which are information, liability, physical, and financial.

6.1  Assumptions for the model As discussed in Chapter  6, every model is a simplified representation of the reality. It is up to the modeler to choose the elements that properly represent the reality, in accordance with the purpose and goals of the model and project. Hence, the description of any model should begin with a presentation of (i) the purpose and objectives of the model; (ii) the principles governing design of the model; and (iii) the assumptions and simplifications. The model is expected to replicate the dynamics of a freight transport market, in which intermodal transport services take place. It should i­ ncorporate the Intermodal Freight Transportation. https://doi.org/10.1016/B978-0-12-814464-0.00007-4 © 2019 Elsevier Inc. All rights reserved.

161

162  Intermodal freight transportation

fundamental properties of the conceptual framework presented in Chapter 3. The conceptual framework explains the mechanisms of integration in intermodal freight transport services. It depicts five dimensions of fitness, which are physical, logical, liability, financial, and strategic. For purposes of testing, and in order to keep the level of complexity to a minimum, only the following fitness dimensions were analyzed: physical fitness, logical fitness, and strategic fitness. Based on these decisions, the specifications and assumptions of the model could be elaborated as follows: ●

●

●

●

●

The freight transport market should be divided into two regions (let us call them Origin Region and Destination Region), where land-based transport services ensure the connection between the origin/end point and a hub. Let us assume, for the sake of simplicity, that the land-based services are road and rail services. The markets are connected through long-distance services linking the hubs of the origin and the destination regions. Let us assume, again for the sake of simplicity, that the long-distance services are air transport services. The model is populated with an arbitrary number of agents, which is fixed for each simulation. In accordance with the conceptual framework, the level of fitness may evolve with market dynamics, and to recreate the market dynamic, multiple agents are required. Moreover, the level of fitness may have an impact on the performance; thus, multiple agents are required to recreate a transport chain with potentially different levels of fitness. The agents to be included are shippers, freight forwarders, carriers, vehicles, and terminals. Different agents will have different levels of complexity and detail, depending on their role and level of influence in the organization and management of intermodal freight transport services. Accordingly, agents may be cognitive: shippers, freight forwarders and carriers; or noncognitive: vehicles or terminals. As far as the cognitive agents are concerned, their behavior is time and path dependent. Memory is built from own experiences and perceptions from the environment. It fades with time. Their main roles are: – Shipper agents generate the daily demand of freight transport services— shipment orders; – Freight Forwarder agents organize and manage the transport services on behalf of the Shipper agent. Each Freight Forwarder agent works with a varying number of Carrier agents, depending on the geographic area they work in; – Carrier agents provide transport services between designated geographic locations. Each one uses one mode of transport. Carrier agents control the behavior of the noncognitive Vehicle agents; – Carrier agents either work with fixed timetables or do not, and are designated as fixed or flexible Carrier agents, respectively.

Building intermodal freight transport services  Chapter | 6  163

●

●

●

●

On a given transport service, the vehicles of fixed Carrier agents may convey goods from several Freight Forwarder agents, whereas the vehicles of flexible Carrier agents solely convey goods from one Freight Forwarder agent. The model simulates a spot market with varying daily and weekly bids and shipments. The model should simulate both the decision-making process and the production of intermodal freight transport services. Because the former takes place when the fitness is taken into consideration, whereas the latter occurs when the fitness impacts the performance of the service. The Shipper agent and the Freight Forwarder agent contracts are based on a set of factors, such as pricing, service conditions, and perceptions.

6.2  Model architecture The conceptual structure of TransFreight is based on system engineering principles that have been widely adopted in the literature (Tavasszy et  al., 2010; Manheim, 1979; Jensen, 1990; Woxenius, 1998a, b; D’Este, 1996; Graham, 1990) (Fig. 6.1). Typical conceptualization organizes the transport chains into two layers: the administrative layer and the physical layer. Four types of flows have been identified within and across these layers: information flow; financial flow; physical flow; and liability flow.

FF0

C

C

FFi

Ci

FFk

Origin region

Destination region

O0

TO

Oi

D0

Point-to-point TAj

O1

D1 TD

Di Di+1

On–1

Point-to-door TA r

Dt

Door-to-point TA p Dt–1

On Legend: C—Customer;

O—Origin;

FF—Freight forwarder;

D—Destination

FIG. 6.1  TransFreight’s virtual freight transport market.

164  Intermodal freight transportation

In this market, the Shippers generate the demand for freight transport services. The Freight Forwarders organize and manage their transport services. The freight transport services are provided by multiple Freight Transport Companies. The Freight Transport Companies convey cargo either between origin and terminal (door to airport); or terminal and destination (airport to door); or between the terminals (airport to airport). Each transport company uses one mode of transport. Competition takes place at two levels: firstly, between freight forwarders when competing for the shippers’ transport services; and secondly, between transport agents when competing for the freight forwarders’ transport services. Freight forwarders compete based on price, transit time, and shipper’s perception (of them); while transport agents compete based on price, transit time, and freight forwarder’s perception (of them). Transport companies are not fully nor equally reliable; therefore, failure (either in terms of delay or damage) may occur. Penalties are associated with failure that, ultimately, results in loss of competitiveness, for both the transport agent and the freight forwarder (that has chosen that transport agent). Additionally, every freight forwarder and transport agent adopts a specific price strategy aiming to leverage their competitive position (which is a function of intrinsic properties, external pressures, and agreements between them). With regard to the organization of the intermodal freight transport services, TransFreight considers two levels: the administrative level and the physical level (Fig. 6.2). The administrative level encompasses both the activities carried out prior to the physical transport (Subprocess 1, Fig. 5.6) and the management activities carried out during the physical transport (Activity 9 of Subprocess 2, Fig. 5.6). It thus includes all the activities not directly related with the physical transport of the freight, such as negotiation between shipper and freight forwarders, and communication between freight forwarder and other transport agents. The physical level encompasses the activities carried out during the physical transport of the freight (Activities 1–5 and 6–8 of Subprocess 2, figure other chapter).

6.2.1 Environment Using the taxonomy put forth by Russel and Norvig (2003), TransFreight’s environment can be classified as follows: ●

Partially observable: akin to the real-world freight transport market, where transport operators keep most of the information and do not disclose to the market (and, consequently, to the other transport agents), TransFreight’s agents only have access to the information they can capture from the market, such as the number of agents or schedules and routes. Both private information on agents (such as pricing strategies, reliability levels, or financial performance) and information on negotiation details (such as pricing, volumes, or quantities) are restricted to the agents directly involved.

Building intermodal freight transport services  Chapter | 6  165

C

Ci

C

Administrative level FF0

Physical levels

Door-to-point TAp

FFi

FFk

Point-to-door TAr

Point-to-point TAj

TO

TD Dt

Oi

Legend: C—customer; O—Origin; FF—Freight forwarder; D—Destination; TA—Transport agent; T—Terminal Relation (information, legal, etc.) between agents Physical transport

FIG. 6.2  Conceptual market structure for TRANSFREIGHT.

●

●

●

●

●

Stochastic: TransFreight is inherently random. At the beginning of each run, each agent’s internal properties are randomly generated (within a certain interval); consequently, any given action by an agent rarely produces the same outcome. Sequential: the environment subsists throughout the entire lifespan of a simulation run. Static: the environment’s properties only change through the intervention of agents (before the first order being placed on the market, there is no change in the environment). Moreover, agents’ decision-making processes are instantaneous (take zero time); therefore, while simultaneous actions may occur, the environment’s properties do not change during the process, meaning that agents’ expectations are not affected. Discrete: agents can only assume, at each given time, one of a predefined set of states. Consequently, the environment also has finite number of states (although the number increases with the number of agents). Multiagent: there is at least one element of each agent.

The TransFreight environment recreates in an identical way the real-world physical properties, namely, time, distance, volume, or weight. Additionally, real-world objects, such as vehicles (trucks or aircrafts) or containers, also exhibit similar physical properties in terms of capacity, speed, or other properties. With regard to the geographic-related dimension, TransFreight recreates a market geography where freight transport services are offered (Fig.  6.1).

166  Intermodal freight transportation

The market is represented by two separate regions: Origin Region and Destination Region. Each region has a set of facilities (Origin Region: O0 to On; and Destination Region: D0 to Dt, n, t ∈ N), and one Terminal (Origin Region: TO; and Destination Region: TD). Facilities are places of either origin or destination of freight. Terminals correspond to the airports, where cargo is transferred between modes of transport (land and air transport modes). TransFreight considers there is enough cargo flow to support air transport links. Another way to look at these regions is to consider them as the airports’ hinterlands. This does not imply, however, that some parts of these regions do not belong to other airports’ catchment areas. Freight transport services occur in one direction from the origin region to the destination region. As far as the agent’s mobility is concerned, there is no actual movement of agents. The transport of freight is simulated using time-based referential: calculated with an average speed and the distance between points, plus a stochastic variable (to incorporate factors that may cause delays or result in earlier arrivals).a Time-related dimensional properties of TransFreight are: ● ● ● ● ●

Time is measured in hours; 1 year is considered to have 8736 h and 52 weeks; 1 week has 7 days or 168 h; Time 0 (zero) corresponds to Monday midnight; The working week consists of 5  days or 120 h, extending from hour 0 (Monday midnight) to 120 (Saturday midnight).

Distances are measured in kilometers. An object’s dimensions, volume, and weight are measured in meters, cubic meters, and metric tonnes, respectively. Freight is considered to have a volume, weight, and a certain level of fragility. The level of fragility impacts the likelihood of damage: the higher the level of fragility, the higher the probability of damage will be. Freight rates are given on a basis of equivalent weight (EW), which is computed as follows: EW = max ( weight ,volume / 6 ) .

6.2.2 Agents The following cognitive agents have been developed in TransFreight for simulating the freight transport agents: ● ●

Shipper: places the orders on the market; Freight forwarder: assembles and manages the freight transport services on behalf of shippers;

a. More information will be provided later as part of the description of the agents.

Building intermodal freight transport services  Chapter | 6  167

●

Freight transport companies: – Air transport company: provides the air transport services between the airports; – Land transport company: provides the land transport services between the origin and the airport (door to terminal) or between the airport and the destination (terminal to door).

Noncognitive agents have also been accounted for. These are entities with simple internal structures that perform simple tasks and do not exhibit all the features of an agent. They only present part of the agent’s properties mentioned on Chapter 6. These entities are: ●

●

Terminal Handling Company: provides transshipment services, between land and air transport; Vehicle (aircraft, truck, or multimodal vehicle): conveys one or more orders between two points.

6.2.2.1  Cognitive agents 6.2.2.1.1 Shippers The initialization attributes of shippers are presented in Table  6.1. Shippers perform two main tasks: the tendering procedure (Fig. 6.3); and payment and indemnities (Fig. 6.4). The tendering procedure is the task whereby the shipper determines the freight forwarder that will manage the transport service. Fig. 6.3 presents the structure of this task. The tendering procedure is simulated through an auction process based on “first-price sealed-bid” protocol (Sandholm, 1999). This is a one-round auction protocol. Each bidder (freight forwarder) presents one bid without knowing the others’ bids. The best bid wins and the auction process ends. Orders are randomly generated in accordance with the agent’s initial attribute: Shipment per week; and are placed on the market at midnight (00:00) to be dispatched at 18 hundred hours (18:00) the next day. An order consists of the following information: weight (Weight per Shipment), volume (Volume per Shipment), origin and destination (Geographic coverage), pick up time and maximum transit time (Transit time). The shipper invites a fixed number of freight forwarders to bid (which is defined by the attribute: Number of bids). Real-world shippers are generally not experts in transport market issues, which is why they hire a freight forwarder. Shippers will certainly only engage in negotiations with, and eventually entrust their goods to, those freight forwarders that they believe (or trust) offer the conditions for providing an adequate transport service. Belief (or trust) has a multidimensional nature, encompassing factors such as capacity to deliver on time and damage-free; availability to execute unpredictable or unusual demands; willingness to deliver tailored transport services; and openness to giving discounts. It is built up over time, as freight forwarders consistently provide solid

168  Intermodal freight transportation

TABLE 6.1  Shipper agent’s attributes Attribute

Description

Fuzzy Weights

Weights for the computation of final fuzzy value (summing weight is equal to one)

Weight per Shipment

Interval of generation of orders’ weight

Volume per Shipment

Interval of generation of orders’ volume

Maximum Unitary Price

Maximum acceptable unitary price

Maximum Transit time

Maximum acceptable transit time

Geographic coverage

Fixed distances of origin to airport (of origin) and airport (of destination) to destination

Trust Positive Reward

Increase in shipper’s level of trusta in relation to the freight forwarder for noncompliance with initial conditions (delay or damage)

Trust Negative Reward

Reduction in shipper’s level of trusta in relation to the freight forwarder for noncompliance with initial conditions (delay or damage)

φ

Memory effect on the level of trust

Number of Bids

Number of Freight Forwarders asked to tender a bid

Shipments per Week

Maximum number of shipments generated by week, with indication of day of the week

a

Level of Trust will be further explained in the description of a shipper’s tendering procedure.

Stand by

Generation of orders

Invitation to freight forwarders

Receiving bids

Decision making

FIG. 6.3  Customer’s tendering procedure.

Building intermodal freight transport services  Chapter | 6  169

Stand by

No Accomplishment

Negative reward

Yes Positive reward

Delay

Delay Updating LoT

FIG. 6.4  Customer’s payment and indemnity procedure.

proof of quality and reliable service. Of course, belief (or trust) can be eroded or destroyed when a freight forwarder begins to register failures to deliver the services expected of it. It would also be impractical and time consuming to invite every freight forwarder in the market to bid. Accordingly, shippers tend to only invite the most trustworthy freight forwarders. In TransFreight, the nature of the relationship between a customer and a supplier (shipper and freight forwarder; freight forwarder and transport agent) is simulated by a variable called: Level of Trust (LoT). The LoT of a customer i in relation to supplier j, for the time t, is calculated as followsb: LoTij ( t ) = LoTij ( t − 1) ∗ (1 − ϕ ) + Reward The LoT, for time t, is a function of the past LoT, for t − 1, plus the amount of Rewards, obtained between time t and t − 1. In TransFreight, LoT is updated every week, on Sundays (interval of time of 168 h). Reward is given by the number of successfully accomplishedc transport services times the parameter Trust Positive Reward, plus the amount of nonsuccessfully accomplished transport orders times the parameter Trust Negative Reward.d The parameter φ simulates the memory effect. The memory effect represents the fading of the perception (either positive or negative) of something (or somebody) over time. This mechanism is the result of several factors, namely, the fact that people tend to forget past events over time; and the fact that a shipper’s employees change; also, newcomers do not have a developed memory base. b. The formulation was inspired by the early work about learning and reinforcement by Roth and Erev (1995). c. A transport service is successfully accomplished when cargo is delivered without damage and within the time window initially agreed; otherwise, it is nonsuccessfully accomplished. d. The memory effect has an asymptote for LoT = 0; thus, a supplier with negative LoT would converge to zero (but always be negative). As, in order to be called to submit a bid, a supplier needs to have a positive LoT value, 0.1 is added to any negative LoT, to enable it to become a positive value.

170  Intermodal freight transportation

Shippers will choose the highest number of freight forwarders, a number equal to the parameter Number of bids. Shippers only choose freight forwarders with a positive LoT. The freight forwarders may or may not offer a bid for the transport in question, depending on their strategy. The shipper’s decision-making process is emulated through the engine-­ controlled decision-making process, based on fuzzy set theories and fuzzy control. The engine considers two decision variables in the modal choice process, which are price and transit time. These are the most common and relevant variables in the mode choice process (Reis, 2014). The payment and indemnities task are carried out upon conclusion of every transport service. The structure of the task is represented in Fig. 6.4. Details on how TransFreight handles and considers financial flows (related with payments and indemnities) are provided in the following chapter. One of two situations may occur at the end of a transport service: accomplishment or nonaccomplishment of the initial request. In the case of accomplishment, the freight forwarder receives a Trust Positive Reward; whereas, in the event of nonaccomplishment, the freight forwarder receives a Trust Negative Reward. These rewards are memorized to be used afterwards while updating the shipper’s LoT. 6.2.2.1.2  Freight forwarders The Freight forwarder’s initialization attributes are presented in Table  6.2. Freight forwarders perform two main tasks: order processing (Fig.  6.5) and payment and indemnities task (Fig. 6.4). By means of the order processing task, the freight forwarder agent builds the transport solution to be offered to the shipper. The solution corresponds to an intermodal transport solution with three transport agents: two land legs and one air leg. Step one—determination of time window for the air leg. In TransFreight, both the air transport services and the land transport service in the destination region have fixed schedules.e Land transport services in the origin region, as they do not work on fixed schedules, match their services to times of the air leg. Time windows for the air leg and land leg (destination) are computed as belowf (Fig. 6.6). Air leg early departure time is equal to shipper’s pick-up time plus expected land-based transport time plus expected transfer time at terminal. Land leg (destination) late departure time is equal to shipper’s delivery times minus expected land-based transport time. Air leg late arrival is equal to land leg (destination) late departure minus expected transfer time at terminal. e. The purpose was to explore in greater detail the influence of fitness, in particular, strategic fitness (schedule coordination). Fixed schedules may simulate either rail services and road services (some medium-to-long-distance road services have fixed schedules). f. The determination of the time windows consists in a PERT analysis of the transport process.

Building intermodal freight transport services  Chapter | 6  171

TABLE 6.2  Freight forwarder agent’s attributes Attribute

Description

Profit Margin

Initial profit margin applied on top of transport agent’s price, when service is negotiated

Profit Margin Variation

Profit margin variation amount. Whenever freight forwarder decides to change its profit margin, the amount of variation is defined by this variable

Profit Margin Interval

Minimum and maximum admissible margin profits

Fuzzy Weights

Weights for the calculation of final fuzzy value (summing weight is equal to one)

Number of Air Bids

Number of air transport companies to be invited to submit bids during the order processing task

Number of land Origin Bids

Number of land transport companies in region of origin to be invited to submit bids during the order processing task

Number of Land Destination Bids

Number of land transport companies in region of destination to be invited to submit bids during the order processing task

Trust Positive Reward

Increase in shipper’s level of trust in relation to the freight forwarder for noncompliance with initial conditions (delay or damage)

Delay Trust Negative Reward

Reduction in shipper’s level of trust in relation to the transport agent for noncompliance with initial conditions (delay)

Damage Trust Negative Reward

Reduction in shipper’s level of trust in relation to the freight forwarder for noncompliance with initial conditions (damage)

φ

Memory effect on the level of trust

∂

Level of Trust spillage effect

Expected times are computed on the basis of average times, which are publicly available on the market. Step two—invitation of transport agents. The freight forwarder’s invitation process is similar to that of the shipper. Each freight forwarder has a built-in LoT for every transport agent. In the moment of invitation, it invites the most trustworthy agents (with a positive LoT), to the maximum amount defined by the parameters: Number of air bids, Number of land origin bids, Number of land destination bids. For the purposes of this model, the freight forwarder only invites transport agents with a positive LoT. The rationale underlying this process is similar to that presented for the shippers’ invitation process. In the real world, freight forwarders have preferential relationships with certain transport agents, which normally stem from either

172  Intermodal freight transportation

Stand by

Invitation transport agents

Receiving bids

Identification of possible chains Decision making

Communication with customer

FIG. 6.5  Freight forwarder’s order processing.

Early departure time

Pick up time

Leg 1 Transit time

Transfer time

Late arrival time

Leg 2 Transit time

Transfer time

Delivery time

Leg 3 Transit time

Customer’s transit time

FIG. 6.6  Intermodal transport service time intervals.

past positive experiences or can be attributed to market recognition. Freight forwarders are certainly unlikely to entrust untrustworthy or untruthful transport agents with cargo, as that would involve high risks of noncompliance, with negative consequences for their own image from the viewpoint of the customers. The consequences of such action could include a reduction in business operations with the shipper in question and, ultimately, going out of business. Furthermore, it would be unmanageable and costly to ask for bids from every transport agent in the market.

Building intermodal freight transport services  Chapter | 6  173

The information conveyed to all transport agents includes the goods’ volume and weight; the geographic location, specifically for land transport agents; and the time windows, specifically for air transport agents and land (destination) agents. In the opposite direction, the information conveyed to the freight forwarder includes the quote (price), transit time, and possible schedules (in the case of air transport and land-based [destination] agents). The land-based transport agents may decline to submit a bid. If they all decline, the freight forwarder then invites other transport agents with a positive LoT. If there are no more transport agents, the freight forwarder declines the shipper’s request. It is assumed the freight forwarder would not risk choosing transport agents with a negative LoT, as the quality of the transport solution would be undermined and the likelihood of noncompliance too high. The freight forwarders’ LoT is updated on a weekly basis (every Sunday), based on the same mechanism as presented for the shippers. However, in this situation, a distinction between delay and damage is made and two different negative reward attributes are considered: Delay Trust Negative Reward and Damage Trust Negative Reward. The rationale is that a freight forwarder may see a delay as being more significant in terms of trust than damage. In the real world, freight forwarders carry out market intelligence. They are constantly scanning the market for better transport providers for future services or, alternatively, seeking to identify those that underperform so that they can avoid them in the future. Each freight forwarder has normally been able to indicate the shippers and transport companies that its competitors have worked with and has also been aware of any recent faults (such as accidents, delays, strikes, etc.) or successes (such as proven reliability, flexibility, vehicle acquisitions, etc.) on the part of most transport companies. Market intelligence is necessary for agents to remain competitive and to avoid being overrode by competitors. In order to simulate the transport agents’ image in the market, a spillover effect was implemented in TransFreight. The change in the LoT of freight forwarder j toward transport agent k due to a change in the LoT of freight forwarder i is given by: LoTjk ( t ) = ∂ ∗ Reward ik ( t ) + LoTjk ( t − 1) The factor ∂ is the spillover factor and represents the impact of a change that is conveyed in the market. Step three—identification of feasible transport chain solutions. A feasible transport chain is a transport chain whose pick-up and delivery times are within the time windows demanded by the shipper. The pick-up and delivery times are computed by subtracting and adding the land transport times and transfer times to the air leg’s schedule, respectively. However, not all possible transport chains are considered during the decision-making process; the worst solutions are immediately discarded. A solution is worse than another if for the same price its transit time is longer, or if for the same transit time it is costlier. The rationale for this action is that it is implausible that a human decision maker would take

174  Intermodal freight transportation

into consideration solutions that are, from the outset, worse than others. Bearing in mind that in the real world this process is to be carried out repeatedly and rapidly, it is most likely that the decision maker will use any strategy to reduce the complexity of the decision making. So, at the end of this process, the freight forwarder has a list of all possible transport chains that meet the shipper’s time (and volume and weight) demands. Step four—identification of the best transport solution. The freight forwarder’s decision-making process is emulated in the same engine as the shipper’s. Moreover, it is assumed that a freight forwarder’s decision-making process is based on the same attributes and has the same fuzzy weights as the shippers. The consideration of the same attributes by the freight forwarder is based on the need that to assemble the transport chain that best fits the shipper’s requirements, the freight forwarder will use the same attributes (priceg and transit timeh) and weights. Otherwise, it would end up with a chain that, regardless of its value and performance, would not match the shipper’s demands. The assumption that the freight forwarder uses the same weights is based on the fact that the shipper presents its demands to the freight forwarder. Step five—determination of final price. Final price is calculated as the sum of the price of each leg plus a profit margin. The initial value of the profit margin is given by the attribute Profit Margin. Real-world markets are highly competitive, with freight forwarders constantly looking for sources of competitive advantage. However, freight forwarders have a reduced scope for determining their competitiveness level, as they are pure service providers (at least, in TransFreight); indeed, the only mechanism at their disposal is to change their own profit margins and, thus the final price. In the real world, freight forwarders are constantly adapting their profit margin, in order to offer better prices than competitors and to match a shipper’s willingness to pay.i In order to simulate the market’s pricing and competition dynamics, TransFreight has implemented dynamic profit margin calculator for the freight forwarder. The mechanism works as follows: ●

●

●

If a previous bid (for a specific shipper) was lost, then the profit margin is reduced by the amount defined in the attribute Profit Margin Variation; If past two bids (for a specific shipper) were won, then the profit margin is increased by the amount defined in the attribute Profit Margin Variation; Otherwise, the profit margin does not change.

These simple rules allow for each freight forwarder to adapt to the shippers’ willingness to pay, and they are expected to emulate real-world behavior. g. The time variable corresponds to the door-to-door transport time, equal to the expected delivery time of leg 3 minus expected pick-up time of leg 1. h. The price variable corresponds to the sum of the price of the three transport legs. i. Willingness to pay represents the maximum price a shipper is willing to pay for a product of service. Willingness to pay defines, therefore, the upper threshold for the price and the maximum amount a supplier can get. Please see Breidert (2006) for more detailed information on this matter.

Building intermodal freight transport services  Chapter | 6  175

TransFreight considers a maximum and minimum value for the profit margin. In the real world, the profit margin moves between certain limits, which are dynamically adjusted by the market conditions. Progressive growth in profit margin will provide an opening for market newcomers or, eventually, force authorities to intervene in the market. Conversely, a continuous reduction in profit margin may signify overcompetition and a lack of capacity on the part of the agent. The outcome is typically bankruptcy of one or more agents or, eventually, the intervention of the authorities. In both cases, natural mechanisms are deployed to counteract abnormal fluctuations in the market. Although TransFreight simulates market conditions, in some situations the market dynamics may not be realistic (e.g., simulation with a single freight forwarder), and thus artificial boundaries are introduced to prevent agents’ profit margins reaching implausible values. Step six—communication of the offer. If the offer wins then the transport agents are notified, otherwise the freight forwarder waits for a new request. The last task carried out by a freight forwarder is the payment and indemnities task. This task is identical to the task performed by the shipper presented here, so no further explanations are necessary at this point. 6.2.2.1.3  Air transport company The air transport company’s initialization attributes are presented in Table 6.3. TransFreight considers that no costs are due for empty flights. Let us assume that the air transport company is a combination company. Combination companies are passenger airlines that also provide cargo services; therefore, flights will take place regardless of the existence of cargo or not. Essentially, it is assumed that the air transport companies have adopted a unit business strategy for their cargo division. Moreover, it is not easy to compute and allocate costs to the freight segment. The tasks produced by an air transport company are order processing (Fig. 6.7) and physical transport (Fig. 6.8). The order processing task starts with a request from the freight forwarder to submit a proposal. For the time window indicated by the freight forwarder, the air transport company determines the possible schedules. Secondly, it checks if there is available space (in terms of weight and volume) to transport the freight. In case of availability, the market price is obtained by multiplying the Unitary cost of production with the freight’s equivalent weight, plus the Profit Margin. The information sent to the freight forwarder includes price and the identified schedules. If the order is won, a physical transport service is scheduled for that date. TransFreight does not implement any revenue management scheme. However, as in the case of the freight forwarder agent, a mechanism for determining the freight forwarder’s willingness to pay is implemented. The rationale and rules of the mechanism are similar to those presented here for the freight forwarder.

176  Intermodal freight transportation

TABLE 6.3  Air transport company agent’s attributes Attribute

Description

Transit Time

The transit time between two airports

Profit Margin

Initial profit margin applied on top of production costs, when service is negotiated on the market

Profit Margin Variation

Profit margin variation amount. Whenever the transport agent decides to change its profit margin, the amount of variation is defined by this variable

Interval Profit Margin

Minimum and maximum admissible margin profits

Unitary Cost of Production

Cost of transporting equivalent to 1 kg

Schedule

Weekly schedule of the flights

Reliability Damage

Spectrum of probability of induced damage. It is considered that important freight forwarders (in terms of weight transported) receive dedicated treatment and, thus, higher resources are put at their disposal, with a corresponding reduction in the probability of damage

Reliability Delay

Probability of induced delay. This is considered to be fixed, as the transport agent’s core business is passenger transport. Thus, reliability depends on factors that do not relate to the cargo business

Maximum Volume Capacity

Maximum volume capacity of the aircraft

Maximum Weight Capacity

Maximum weight capacity of the aircraft

Stand by

Pricing

Scheduling selection

Sending offer

FIG. 6.7  Air transport company agent’s order processing.

Building intermodal freight transport services  Chapter | 6  177

Stand by

Loading

Flight

Unloading

FIG. 6.8  Air transport company agent’s physical transport.

The final task deployed by an air transport company agent is the physical transport of freight between the two airports. On every flight, there is a list of booked orders from the various freight forwarders; however, for various reasons (e.g., delays), orders may not arrive on time. Accordingly, the orders that are actually loaded may not entirely correspond to the list of booked orders. At departure time,j those orders are loaded onto the aircraft; they include booked orders and, eventually, delayed cargo (that missed previous flights and has meanwhile arrived). If the aircraft’s capacity is not enough to accommodate all the cargo, then booked orders have priority over delayed cargo. Loading operations are carried out by the handling agent and simulated through the noncognitive agent: terminal. The flight is simulated by a cognitive agent vehicle designated as aircraft. The attributes of these cognitive agents are explained further later in this chapter. At the end of the transport, orders are unloaded by the handling operators and made available for land transport. 6.2.2.1.4  Land-based transport company The land-based transport company agent’s initialization attributes are presented in Table 6.4. It performs two main tasks: order processing (Fig. 6.9) and physical transport (Fig. 6.10). The task order processing is triggered by the receipt of a request from the freight forwarder. The agent starts by determining if there are available resources (in function of the Refusal order rate). If so, the transport agents in the destination regions compute the possible schedules in line with the j. Actually, this is not actual aircraft departure time, but the flight closing time, which normally is 90–120 min before departure time. During this period, freight is loaded onto unit load devices or onto containers and transported to the ramp where the aircraft will depart. In TransFreight, and for simplicity reasons, the air transport company’s agent transit time encompasses this period.

TABLE 6.4  Road transport company agent’s attributes Attribute

Description

Unitary Fixed Price

Price per hour of transport

Unitary Marginal Price

Price per kilogram and per kilometer of transport

Profit Margin

Initial profit margin applied over production costs

Profit Margin Variation

Profit margin variation amount. Whenever transport agent decides to change its profit margin, the amount of variation is defined by this variable

Interval Profit Margin

Minimum and maximum admissible margin profits

Reliability Damage

Spectrum of percentage of induced damage. It is considered that important freight forwarders (in terms of weight transported) receive dedicated treatment and, thus, higher resources are put at their disposal, with a corresponding reduction in the probability of damage

Reliability Delay

Spectrum of percentage of induced delay. It is considered that important freight forwarders (in terms of weight transported) receive dedicated treatment and, thus, higher resources are put at their disposal, with a corresponding reduction in the probability of damage

Order Refusal Rate

Probability of orders that are not accepted. For simplicity reasons, it is assumed road transport companies have infinite capacity (in practical terms, they can go to the market and get as many trucks as they want). However, in the real world, for the most varied reasons, these agents do not respond positively to every request (e.g., no available trucks or available trucks too far away). This order refusal rate is a solution to simulating reality, avoiding the need to simulate a road transport company’s fleet management

Schedulea

Weekly schedule of the transport services

a

Only transport agents in the destination region have fixed scheduling.

Stand by

Pricing

Sending offer

FIG. 6.9  Land-based transport company agent’s order processing.

Building intermodal freight transport services  Chapter | 6  179

Stand by

Loading

Transport

Unloading

FIG. 6.10  Land-based transport company agent’s physical transport.

time window provided by the freight forwarder. Transport agents in the origin region do not have fixed schedules, so do not need to perform this task. Finally, in case of availability of resources and schedules (in case of agents in destination region), the final price is calculated as: Price = Profit margin × ( Unitary fixed price × Transit time + Unitary marginal price × Length × Weight ) The profit margin is calculated as for freight forwarders and air transport agents. The rationale is also the same as that presented for those agents. The physical transport of freight encompasses the land transport of freight. In the case of the origin region, at the time of booking the transport service is initiated. In the case of the destination region, the transport service is initiated in accordance with the transport agent’s schedule, and the orders available at the airport are conveyed until destination. Delayed orders are kept at the airport waiting until the next schedule. As for the air transport company, transport is produced by a cognitive agent vehicle: truck, about which further detail is provided further here. At the end of the transport service, the cargo is delivered and the vehicle agent is terminated.

6.2.2.2  Noncognitive agents The Terminal agent is responsible for two real-world processes that occur practically simultaneously: the handling and customs clearance processes. The former corresponds to the modal transfer of freight, while the latter corresponds to the legal process that is necessary for authorizing freight to leave the airport. There are two terminal agents, each one operating at a terminal. The attributes of this pseudoagent are presented in Table 6.5. Terminal agents perform two main tasks: freight handling services and short-term storage. Such tasks are required typically in four moments along an

180  Intermodal freight transportation

TABLE 6.5  Terminal agent’s attributes Attribute

Description

Damage Reliability

Probability of damage to cargo during handling process

Handling Transfer Time

Required time for processing the handling process

intermodal transport chain: at the airport of origin, for unloading land transport vehicles (and freight stored at the terminal) and for loading aircraft; and at the airport of destination, for unloading aircraft (and freight stored at the terminal) and for loading land transport vehicle. The time to carry out these activities depends on both the terminal’s productivity and customs clearance time. Since these two processes may run simultaneously to some extent, the overall time is not the sum of the two times but likely the maximum value. TransFreight entirely simulates the terminal’s tasks. The main actions are: ●

● ● ●

Unloading of land transport vehicle upon request and storage of freight at the terminal; Loading of aircraft upon request; Unloading of aircraft upon request and storage of freight at the terminal; Loading of land transport vehicle upon request.

Handling time is considered constant over time and defined by the attribute: Handling transfer time. In the real world, air transport companies impose deadline times for the arrival time of cargo at an airport (unless subject to special conditionsk). Some reasons can be indicated for this: the handling procedures take time and, in the case of unexpected peak periods, the handling operator needs to ensure cargo is prepared on time; some customs procedures may require cargo to be physically at the terminal before final clearance; or the cargo leaves the terminal to go to the aircraft some time before its departure.l In terms of behavioral properties, TransFreight considers that the handling and the storage processes may cause damage to cargo. The level of reliability is given by the attribute: Damage reliability. In every transfer process a random number is generated, following a uniformly distributed probability function, to determine whether cargo was damaged or not. If the random number is greater than the reliability level, then damage occurs; otherwise, there is no damage. Any damage caused by the terminal agent is considered as occurring during the air leg. The handling operator works for the air transport company; k. For example, in cases where cargo arrives at the airport already unitized and there is no need for handling procedures. l. Not only because there is some distance between the terminal and the aircraft location. For example, at Lisbon Airport cargo takes 40 min to arrive at the aircraft, while in Brussels that time is around 60 min.

Building intermodal freight transport services  Chapter | 6  181

therefore, the liability for any misconduct within the terminal lies with the air transport company. Indeed, the handling operator is invisible for the freight forwarder (or the shipper), which only has a legally binding agreement with the air transport company. The following cognitive agent is the vehicle, which is responsible for simulating the physical transport of freight between locations. There are two types of (cognitive agent) vehicles in TransFreight: the (passenger) aircraft and land transport vehicle (truck). All the properties of the cognitive agent vehicles are embedded in the respective transport agent, as they are inherited during the generation process.m The following table presents the attributes of the pseudoagent or vehicle. Vehicles are the only agents that do not run through the lifespan of a simulation run. They solely exist during the period of transport: they are generated at the moment the transport service commences and are terminated in the moment the cargo is unloaded. This is the reason for the aforementioned fact that all their attributes are inherited from the parent transport agent (Table 6.6). The following behaviors are implemented: delay and damage. A vehicle may suffer a delay in relation to the initial schedule for the most varied reasons, of which congestion is the most frequent. Delays are simulated in a two-step approach. First of all, it is determined whether the service will be delayed or not. A random number between zero (0) and one (1) is generated, following a uniform distribution function; if the random number is greater than the Delay Reliability attribute then there is a delay, otherwise the transport service is delivered on schedule. Secondly, the amount of delay is computed. The amount of delay is randomly generated following a lognormaln distribution function. Moreover, a minimum delay of 15 min for land transport and 30 min for air transport is c­ onsidered. Damage is computed following the same rules already described for the terminal agent and utilizing the attribute Damage reliability. Bear in mind that delay reliability (Table 6.4) and damage reliability (Tables 6.3 and 6.4) are arrays of values. Each value represents the reliability for a certain

TABLE 6.6  Vehicle agent’s attributes Attribute

Description

Damage Reliability

Probability of transport services causing damage on freight

Delay Reliability

Probability of transport services being delayed

m. They are presented separately for clarity purposes. n. Lognormal function was found to be the most suitable distribution function, as it does not assume negative values (there are no negative delays) and it is skewed to the right (most delays occurs around a certain value close to the minimum, and long delays tend to be rare).

182  Intermodal freight transportation

type of shipper. Three types of shippers (freight forwarders) are considered: nonrelevant if it accounts for less than 5% of transport agent’s total volumes; medium-relevant if it accounts for less than 20%; and relevant otherwise. The lower bound is the reliability considered for the nonrelevant shipper, the middle value the reliability for the medium relevant shipper, and the upper bound the reliability for the relevant shipper.

6.2.3 Engines Engines are complex and autonomous processes that involve two or more agents. They are presented separately for clarity purposes. TransFreight has one engine: the decision-making process. One of the assumptions is that TransFreight models a freight transport spot market. In a spot market, decision making is performed immediately upon a request. This is a very typical situation for the freight forwarding business, where a middle-level manager is required to organize the transport service in a short period of time (e.g., spot markets) and often with incomplete (or little) information. These situations are not compatible with complex, multivariate decision-making processes. Instead, the decision maker is more likely to adopt a more straightforward and efficient reasoning process (Beuthe et al., 2008). D’Este (1996) has proposed a three-step approach. The decision maker begins with a pool of all transport operators with which he has commercial agreements. In the first phase, he selects those options that are simultaneously feasible from a technological and service point of view (e.g., feasible schedules or capacity). The purpose of this step is to reduce the decision-making process to a manageable level of complexity by limiting the amount of possibilities. It is then time to select the most interesting solution. The choice is based on two main factors, which are: ●

●

Familiarity with the agent—there is a greater likelihood of selecting an agent with which the decision maker has already worked in the past; Past positive experiences—the decision maker’s perception of the agent, which is built over time and which includes many of the qualitative factors referenced the literature (such as reliability, flexibility, safety, and frequency).

The second level of decision making encompasses a more in-depth analysis of the various agents. Eventually, the decision maker may enter into contact with them (requesting information about prices and transit times). And, the final choice falls on the option that presents the best combination of the decision factors. In summary, the first phase reduces the universe of options to a ­manageable set of trustful agents and the second phase enables identification of the best choice. TransFreight’s decision-making process was developed to mirror this d­ ecisionmaking process. The first phase is conducted by means of consideration of a variable designated Level of Trust (LoT). This variable represents the shipper’s ­trust ­toward a freight forwarder and the freight forwarder’s trust toward a transport

Building intermodal freight transport services  Chapter | 6  183

agent. Trust building (or erosion) is a cumulative process of acquisition of rational and nonrational (subconscious) perceptions about something (or someone) (Kramer, 2009, p. 70), which include both qualitative and nonqualitative attributes. The second phase of the decision-making process is emulated through a fuzzy logic inference mechanism (FLIM) based on fuzzy sets theory and fuzzy control. FLIM considers two modal choice factors (price and transit time), which can have different impacts in the decision-making process (and which are defined through the attributes Fuzzy weights in Tables 6.1 and 6.2). The decision-making process engine is divided into two phases: ●

●

Phase 1—Fuzzy logic inference mechanism (Fig. 6.11): – Fuzzification—conversion of real-world variables (normally referred to as crisp variables) into fuzzy input variables; – Fuzzy inference—computation of a fuzzy output variable (or variables); Phase 2—Identification of the transport solution winner.

6.2.3.1  Phase 1—Fuzzy logic inference mechanismo For the fuzzification step, TransFreight considers two fuzzy input variables: price and time. Each of these variables is graded into three levels, which are: Price = {low, medium, high} Time = { poor , medium, good} Fig. 6.12 outlines the shape of the membership functions for the input v­ ariables; the specific valuation depends on the mode of transport they refer to (road or air). The output variable ranges between zero (0) and one (1). Zero denotes a bad option; while one denotes maximum values and thus an excellent option. Fuzzy logic Input variables (crisp values)

Fuzzifier (fuzzification process)

Real world

Fuzzy world

Fuzzy inference

Output fuzzy value

FIG. 6.11  Fuzzy logic system.

o. There are several techniques for implementing Fuzzy Logic. Herein, the technique was chosen based on the appropriateness in relation to the real-world reasoning process and the number of variables, and on the ease of implementation in the agent-based model. For more information on this aspect, please see Jang (1993).

184  Intermodal freight transportation

m (x)

mLow(x) mPoor (x)

mMedium

mHigh (x)

Price

mMedium

mGood (x)

Time

(min. price + max. price)/2 (min. time + max. time)/2

max. price max. time

1

0 min. price min. time

x

FIG. 6.12  Fuzzy membership functions for input variables.

Fuzzy membership functions are defined for every decision-making process. The dynamic nature of these functions is justified by the fact that each transport service is unique both in terms of prices and times.p Thus, for every transport service, transport agents provide a unique pair of time and price values, rendering it impossible to define static membership functions that could prove reliable for every possible situation. Moreover, in the real world, while freight forwarders (and shippers) are aware of the typical rates and prices for each destination and amount of cargo, in practical terms they have to decide on the basis of what is actually provided by the transport agents (and not on what they are expecting or would like to have). If a price is too high or a transit time too long, the freight forwarder (or shippers) may abandon the intention of forwarding cargo. However, TransFreight makes allowance for this situation through the attributes: Maximum Unitary Price and Maximum Transit Time. Calculation of the membership functions for the Price variable is computed as follows: x  1 − P − P , x ∈ [ PMin ,PMed ] µ Low ( x ) =  Med Min  0, x ∈ [ PMed ,PMax ]  x   P − P , x ∈ [ Pmin ,PMed ]  min µ Medium ( x ) =  Med x 1 − , x ∈ [ PMed ,Pmax ]  Pmax − PMed  0, x ∈ [ Pmin ,PMed ]  µ High ( x ) =  x  P − P , x ∈ [ PMed ,Pmax ] Med  max p. It is important to bear in mind that each shipper forwards cargo from a specific location to a specific destination (which are randomly generated at the start of each run). The same occurs in the real world, since shippers are scattered throughout the territory.Moreover, land-based transport rates are also a function of the length of the transport service; thus, once again, rates are shipper specific.

Building intermodal freight transport services  Chapter | 6  185

where PMin = Min ( Pi ) , i = 0,1,… n PMax = Max ( Pi ) , i = 0,1,… n PMed = 0.5 × ( PMin + PMax ) P is the price of the option; n is the number of possible options (transport chains, in the case of a freight forwarder’s decision-making process, or number of bids, in the case of shipper’s decision-making process). The computation of the membership functions for the Time variable is computed as follows: x  1 − TT − TT , x ∈ [TTmin ,TTMed ] µ Poor ( x ) =  Med min  0, x ∈ [TTMed ,TTmax ]  x   TT − TT , x ∈ [TTmin ,TTMed ]  Med min µ Medium ( x ) =  x 1 − , x ∈ [TTMed ,TTmax ]  TTmax − TTMed  0, x ∈ [TTmin ,TTMed ]  µGood ( x ) =  x  TT − TT , x ∈ [TTMed ,TTmax ] Med  max where TTmin = Min ( TTi ) , i = 0,1,… n TTmax = Max ( TTi ) , i = 0,1,… n TTMed = 0.5 × ( TTmin + TTmax ) TT is the transit time for the option; n is the number of possible options (transport chains, in the case of a freight forwarder’s decision-making process, or number of bids, in the case of shipper’s decision-making process). As far as the fuzzy inference step is concerned, the process computes the final output fuzzy value for every option. Thus, for every option i, the final output fuzzy value is computed as follows:

µi ( x ) = WeightTime × µTime + Weight Price × µ Price

186  Intermodal freight transportation

i = 0, 1, …, n n is the total number of options. The fuzzy weights represent the behavior of the shipper (and freight forwarder). They are defined at the outset and they remain constant during a simulation.

6.2.3.2  Phase 2—Identification of the transport solution winner The final step of the decision-making process is determining the actual contract winner option. This step is conducted outside the fuzzy engine. TransFreight determines the winner randomly, based on the propensity of each final output fuzzy value. The reason for utilizing a random process instead of a deterministic one is the fact that human judgment is not strictly rational. In the case of a deterministic situation, the option with the higher output fuzzy value should be the chosen one. However, it is to be expected that a human decision maker does not always select the highest ranked option. Firstly, let us consider a case where more than one option has similar output fuzzy values. In a deterministic situation, the choice would always be for the higher value, even when the difference is between the values is very low. However, a human decision maker may consider them to be practically identical and, therefore, select one or another. Secondly, for some particular reason, the decision maker may opt for a low-valued option over the highest-ranked option (either because he believes in the transport agents in question, or because he is [in some way] displeased or resentful toward the people working for the high-ranked options). This may occur only sporadically, but it is nevertheless a distinct possibility. The procedure that is built into TransFreight is explained now and consists of four steps. Step one consists of determining the propensity of each option, in relation to the set of options. The propensity is determined as follows: Propensityi =

Final Output Fuzzy Valuei n

∑ Final Output Fuzzy Value

j

j =0

Step two consists of forming a vector of propensities. Each position is calculated as the sum of the propensities of the previous positions. Thus, the first position is equal to the propensity of option 0, the second position is equal to the sum of the propensity of option 0 and option 1, and so on and so forth. i   List of Propenities =  Propensity0 ,…,∑ Propensityk ,…,1 k =0  

Step three consists of drawing a random number, r, with uniform distribution, between 0 and 1.

Building intermodal freight transport services  Chapter | 6  187

And finally Step four consists of determining the winner option. The winner option is the one that meets the following condition: i

Option i is winner if

∑ Propensity

k

≤r

k =0

This procedure ensures, firstly, the random nature of the decision-making process, which is inherent in every human decision. Secondly, it provides a similar probability of selection to options with similar output fuzzy values, like in the real world. And, thirdly, it provides for the probability of even options with low output fuzzy values being selected, albeit a low probability, again similar to the real world.

6.2.4 Interactions The presentation of the TransFreight interactions will be done independently, in order to establish a parallel to the fitness dimensions. This is for reasons of simplicity and clarity, as it provides a simpler and direct comparison with the real world. The relevant flows are: ●

●

●

●

Physical: corresponding to the transfer of the goods between freight transport agents (unidirectional from origin to destination); Informational: corresponding to the exchange of information between freight transport agents; Legal: corresponding to a freight transport agent’s liability for carrying freight; Financial: corresponding to the payments (or indemnities) for providing the transport service.

6.2.4.1  Physical Fig.  6.13 presents the physical interactions between transport agents in TransFreight for an intermodal freight transport service. The physical interactions arise from the transfer of freight between transport agents. The sequence of interactions is: 1. Freight transfer from origin point (shipper) to leg 1 vehicle; 2. Freight transfer from leg 1 vehicle to terminal of origin; 3. Freight transfer from terminal of origin to leg 2 vehicle (aircraft); 4. Freight transfer from leg 2 vehicle (aircraft) to terminal of destination; 5. Freight transfer from terminal destination to leg 3 vehicle; 6. Freight transfer from leg 3 vehicle to destination point (shipper).

6.2.4.2  Informational The production of an intermodal freight transport service requires an intensive exchange of information among transport agents. Informational interactions

188  Intermodal freight transportation Customer

Freight forwarder

Leg 1 Transport agent

Terminal origin

Leg 2 Transport agent

Terminal destination

Leg 3 Transport agent

1 2 3 4 5 6

FIG. 6.13  Sequence of the physical flow.

occur in two contexts involving different agents and conveying different ­contents, which are chain assemblage process (subprocess 1 in figure in other chapter) and physical transport process (subprocess 2 in figure in other chapter). The first event in informational interaction takes place during assemblage of the freight transport solution. Fig. 6.14 presents the sequence of the flow of information during this process. The sequence of interactions is as follows: 1. Shipper agent invites a set of freight forwarders to submit quotes for a future transport service; 2. Freight forwarder agent invites a set of land-based transport companies (in origin region) to submit quotes for transport service; 3. Freight forwarder agent invites a set of land-based transport companies in destination region) to submit quotes for transport service; 4. Land-based transport companies in origin region submit bids; 5. Land-based transport companies in destination region submit bids; 6. Freight forwarder agent invites a set of air transport company agents to submit bids for transport service; 7. Air transport company agents submit bids; 8. Freight forwarder submits its best bid; 9. Shippers notifies each freight forwarder of its decision (win or lose); 10. Freight forwarder winner books land transport company agent winner (origin region); 11. Freight forwarder winner books air transport company agent winner; 12. Freight forwarder winner books land transport company agent winner (destination region). Fig.  6.15 presents the flow of information during the physical transport process. 1. At start of leg 1, land transport company agent generates vehicle and loads information about the transport service;

Building intermodal freight transport services  Chapter | 6  189 Customer

Freight forwarder

Terminal origin

Leg 1 Transport agent

Leg 2 Transport agent

Terminal destination

Leg 3 Transport agent

1 2 3 4 5 6 7 8 9 10 11 12

FIG. 6.14  Sequence of the information flow—assemblage of transport solution. Customer

Freight forwarder

Leg 1 Transport agent

Terminal origin

Leg 2 Transport agent

Terminal destination

Leg 3 Transport agent

1 2 3 4 5 6 7 8 9

Leg 1 Vehicle

Leg 2 Vehicle

FIG. 6.15  Sequence of the information flow—physical transport process.

Leg 3 Vehicle

190  Intermodal freight transportation

2. On arrival at terminal of origin, vehicle notifies terminal agent to initiate unloading of freight; 3. Notification of parent agent and consequent termination; 4. At start of leg 2, air transport company agent generates vehicle and loads information about the transport service; 5. Air transport company agent notifies terminal agent to load cargo onto vehicle; 6. On arrival at terminal of destination, vehicle notifies terminal agent to initiate unloading of freight; 7. Notification of parent agent and consequent termination; 8. AT start of leg 3, land transport company agent generates vehicle and loads information about the transport service; 9. On arrival at destination, vehicles notifies parent agent and terminates;

6.2.4.3  Legal Fig.  6.16 presents the hierarchical structure of accountability and liability in TransFreight. The liability hierarchy follows a customer-service provider relationship; accordingly, each agent is accountable for both its own actions and its service providers’ actions. As a result, for the shipper the freight forwarder is the sole entity liable for any noncompliance in the freight transport service. The freight forwarder in turn considers the transport company agent that causes the noncompliance liable. Land transport company agents are liable for their vehicles’ operations, while air transport agents are liable for both aircrafts and the terminal operator’s operations. The air transport company agent bears the liability of the terminal operator since this agent works on its behalf. In the real world, liability claims lead to the payment of an indemnity to the shipper after investigation, dispute, and negotiation. In TransFreight, liability claims are accounted for by means of a change in the Level of Trust variable. TransFreight consider two types of noncompliance sources: delay and damage. A delay occurs when cargo is delivered outside the time window defined

Land transport company

Customer

Vehicle

Freight forwarder Air transport company

FIG. 6.16  TransFreight’s hierarchical structure of liability relationships.

Terminal operator Vehicle

Building intermodal freight transport services  Chapter | 6  191

by the shipper (i.e., after the shipper’s deadline). Damage is given when cargo is delivered in a damaged state. Delay and damage may occur simultaneously. In the real world, allocation of liability requires the unequivocal determination of the responsible agent. TransFreight works in the same manner. As far as delay is concerned, the allocation of liability is immediate, as the transit times are already known: arrival at terminals and arrival at destination. Accordingly, whenever a transport company agent has a delay, it is automatically recorded. A shipper may claim for liability of a freight forwarder whenever transit time is longer than the original request (delivery is after deadline). The freight forwarder in turn may claim for liability of the transport agent(s) that has/have caused a delay. However, a situation may occur where overall transit time is lower than the shipper’s original request (there is no delay), but one or more transport leg provided has registered a delay. This may occur in situations where the transit time is lower than the shipper’s time windows. Delays in individual legs may be offset by that buffer time. In such situations, the freight forwarder does not lodge a claim for liability of the transport agent (or agents) that have caused a delay (similar to the current procedure in the real world). With regard to damage, the principle is the same as adopted for accountability for delays: However, identification of damage is not so straightforward as delay. TransFreight simulates real-world complexity based on a set of rules. In the event of damage, a shipper can, of course, lodge a claim for liability of the freight forwarder, as this is the sole agent responsible for the transport service. The freight forwarder, however, can only claim for the liability of a transport agent if damage is unequivocally determined. Intermodal transport services remain a set of individual transport services today and have a clear duration: they start once the transport agent accepts cargo for loading and end once the next transport agent (or shipper, in the case of leg 3) accepts cargo. Consequently, the acceptance of the freight by the transport agent (or shipper in the case of leg 3) denotes in legal terms that the freight is in a good condition (i.e., there is no damage). Obviously, in the event of damage, the transport agent (or shipper, in the case of leg 3) may refuse to accept freight, or accept it but adding a remark about the existence of damage. Therefore, at both airports, the freight is conveniently checked for damage (both obvious external and internal damage). Any damage caused during transport is thus detected. If undetected, the following transport agent (or shipper in the case of leg 3) assumes full liability.

6.2.4.4  Financial Financial interaction corresponds to the monetary transactions between agents. TransFreight does not explicitly consider the financial flow. It is therefore assumed that at the end payments or indemnities are duly processed. Payments are due for compliance with the initial conditions and money flows from the customer to the transport provider. Indemnities are due for noncompliance with the

192  Intermodal freight transportation

initial conditions and money flows in the opposite direction from the transport provider to the customer. Financial flows are only considered indirectly, since the existence of an indemnity implies noncompliance and thus a negative reward, while the existence of a payment implies compliance and thus a positive reward.

6.3  Verification and validation of ABM models Verification and validation are essentials steps in the model development process if it is to be accepted and used. The outcomes of an untested model have no value and, obviously, should not be used. North and Macal (2007) put it bluntly by arguing that “before appropriate verification and validation, models are toys; after appropriate verification and validation, models are tools.” The key issue lies, however, in what is meant by appropriate, or in other words: how many and what kind of tests are appropriate? A caveat should be made at this point, regardless of the amount and ­nature of tests carried out, any model is hardly ever verified or validated; at best, one achieves confidence as to a model’s outcome (North and Macal, 2007; Sterman, 2004; Carson, 2005). The reason lies in the fact that models are ­ill-­representations of the reality. As such, verification and validation are always a matter of judgment and credibilityq building. Although verification and validation are commonly carried out simultaneously, they refer to different concepts (Carson, 2002, p. 52). Verification pertains to the steps, processes, or techniques that the modeler deploys to ensure that the model behaves in accordance with the initial specifications and assumptions (North and Macal, 2007; Carson, 2005). Validation pertains to the steps, processes, or techniques that the modeler (and any other interested party) deploys to ensure that the model adequately represents and reproduces the behaviors of real-world phenomena (North and Macal, 2007; Carson, 2005). The verification and validation of agent-based models (or other dynamic modeling methodologies) entails different challenges than traditional parametric or equation-based models, namely, the verification and validation of agents’ behaviors, interaction mechanisms, and the processes and structures that emerge within the model. This is particularly complex, as there is currently no satisfactory theory of human behavior, and often agents do represent human behavior or activities. This deficiency undermines our ability to conveniently verify and validate agent-based models. Furthermore, unexpected outcomes in agentbased models may give rise to the following doubt: are they the result of a flaw in the model or a major scientific breakthrough? This is of particular importance because emergence phenomenon requires tackling the model as a whole (looking at particular aspects is irrelevant). However, often models are too complex for human understanding, rendering that requirement unattainable. These facts q. Credibility refers to how peers, users, or other interested parties see the model (North and Macal, 2007; Carson, 2005).

Building intermodal freight transport services  Chapter | 6  193

emphasize the need to carry out the verification and validation endeavors more carefully and in depth. Verification is concerned with the inner part of the model, i.e., if it runs well, with no mistakes (bugs or other inconsistencies), and if it performs every task as initially specified. The literature is abundant in recommendations and examples of tests for the verification of models. The process of verification of TransFreight included several of those tests that were repeatedly performed throughout the development of the model. The following is a brief list of plausible tests, along with an explanation of what was performed on TransFreight: Stress testing and testing with a wide range of parameters and different random numbers (Carson, 2005; Sterman, 2004). Each component of the model and the model itself should be tested under extreme or very distinct situations. In the case of TransFreight, this included considering 0% reliability, zero speed, zero agents, multiple simultaneous requests of shippers, and ever-growing demand and capacity. In such extreme conditions, the behavior of the model becomes foreseeable and, thus, any errors or flaws (such as negative transit times) are easier to detect. Thorough review of all model outputs, not just the primary measures of performance, but numerous secondary measures (Carson, 2005, p. 21). The purpose of these tests is to increase the model’s transparency. The variables presented in the discussion of results are only, but a few of the total amount of variables collected during the simulations. Debugging the model through automatic procedures (such as the software’s debugger); this ensures no bug is overlooked (North and Macal, 2007). Most software tools have built-in routines for detecting and cleaning up bugs and other errors. TransFreight was developed using AnyLogic Software, which contains a debugging add-on. Adoption of unit testsr while programming (North and Macal, 2007; Castle and Crooks, 2006). The model should be built incrementally and gradually, so that any error or bug can be traced back to the respective source. TransFreight can be broken down into several functional parts, such as agents (each agent was developed independently), communication, decision making, etc. Documentation of the model (Sterman, 2004; Peterson and Eberlein, 1994). The process of documentation should start at the very inception and continue throughout every stage of model development until the final tests. Documentation contains every detail on the model, such as variables, functions, arguments, assumptions, outputs, and structure. This chapter is a good example of such documentation. This process ensures the model and results can be understood, replicated, reviewed, and extended by others. Replication is of interest r. A unit test is a component of either the model (such as decision-making engine, communication protocols, allocation resources, or loading vehicles) or the software (such as function, method, module, or class).

194  Intermodal freight transportation

as it allows others to rebuild the model, checking for errors, in addition to introducing transparency and increasing the utility of the modeling work. Review by a more senior or other simulation professionals (North and Macal, 2007; Carson, 2005). Modelers frequently suffer from myopia,s reducing their ability to detect bugs and other inconsistencies. However, first-time observers do not suffer from that phenomenon and are more likely to identify limitations and other shortcomings. The modeler has several options, such as discussing with other people, including scholar and practitioners, participation in conferences, or publication in peer-reviewed publications. After concluding the verification, the model should be subject to validation. The process of validation is meant to ensure that the model adequately represents the real-world and that its results are meaningful. North and Macal (2007) propose an interesting and comprehensive range of validation tests: ●

●

●

●

●

●

Requirements validation: the model should answer to clear requirements and questions about the real world. The purpose and specification of TransFreight are stated at the beginning of this chapter. Data validation: the data in the model should be valid. Real data should be used wherever possible. When there are constraints to using real data, then plausible data can be formulated. In the case of TransFreight, the data are plausible, as the model describes a synthetic market. Face validation: the assumptions of the model should be valid. The assumptions were described at the beginning of this chapter and they were drawn from both state of the art and state of practice. Process validation: agent and interaction structure and steps in the model have to be clear, meaningful, and correspond to real-world processes. The structure of TransFreight replicates the process of organizing and managing an intermodal freight transport service. Agent validation: agents’ behaviors, relationships, or interactions have to correspond to real-world actions. The purpose of TransFreight was to recreate in a virtual environment a realworld freight transport market. Accordingly, inclusion and adoption of realworld action was always the most important aspect. Theory validation: the model’s theories (either about agents or process) have to be valid and used correctly. FreightTrans was derived from the conceptual discussion in Chapter  3. Moreover, the different functionalities were also derived from the literature, such as memory effect (based on Roth and Erev, 1995) and the decisionmaking engine (based on Zadeh, 1965).

s. Myopia refers to the phenomenon whereby errors and flaws in an object become invisible to those who are continuously looking at it or experimenting with it (Levitt, 1960).

Building intermodal freight transport services  Chapter | 6  195

References Beuthe, M., Bouffioux, C., Krier, C., Mouchart, M., 2008. A comparison of conjoint, multicriteria, conditional logit and neural network analyses for rank-ordered preference data. In: Ben-Akiva, M.E., Meersman, H., Van de Voorde, E. (Eds.), Recent Developments in Transport Modelling. Emerald Group Publishing Limited, Bingley, UK, pp. 157–178. https://doi. org/10.1108/9781786359537-009. Breidert, C., 2006. Estimation of Willingness-to-Pay: Theory, Measurement, Application. DUV, Wiesbaden, Germany. ISBN: 978-3835003996. Carson II, J., 2002. Model verification and validation. In: Yücesan, E., Chen, C-H., Snowdon, J.,and Charnes, J., Proceedings of the 2002 Winter Simulation Conference. pp. 52–58. https://doi. org/10.1109/WSC.2002.1172868. Carson II, J.S., 2005. Model verification and validation. In: Proceedings of the Winter Simulation Conference. vol. 1. IEEE, pp. 52–58. https://doi.org/10.1109/WSC.2002.1172868. Castle, C., Crooks, A., 2006. Principles and concepts of agent based modelling for developing geospatial simulations. University College of London. Working Paper 110. D’Este, G., 1996. An event-based approach to modelling intermodal freight systems. Int. J. Phys. Distrib. Logist. Manag. 26 (6), 4–15. https://doi.org/10.1108/09600039610145899. Graham, M., 1990. Effective Intermodalism. Lloyd’s of London, London. Jang, J.-S.R., 1993. ANFIS: adaptive-network-based fuzzy inference system. IEEE Trans. Syst. Man Cybern. 23 (3), 665–685. https://doi.org/10.1109/21.256541. Jensen, A., 1990. Combined Transport: Systems, Economics and Strategies. Transportforksningsberedningen, Stockholm, Sweden. Kramer, R., 2009. Rethinking trust. Harv. Bus. Rev. 87, 69–77. Levitt, T., 1960. Marketing myopia. Harv. Bus. Rev. 38 (4), 45–56. Manheim, M.L., 1979. Fundamentals of Transportation Systems Analysis: Basic Concepts. MIT Press. http://books.google.com/books?id=F36zLQAACAAJ&pgis=1. North, M.J., Macal, C.M., 2007. Managing Business Complexity: Discovering Strategic Solutions With Agent-Based Modeling and Simulation. Oxford University Press, Oxford, England. https://doi.org/10.1093/acprof:oso/9780195172119.001.0001. Peterson, D., Eberlein, R., 1994. Reality check: a bridge between systems thinking and system dynamics. Syst. Dyn. Rev. 19 (2–3), 254. Reis, V., 2014. Analysis of mode choice variables in short-distance intermodal freight transport using an agent-based model. Transp. Res. A Policy Pract. 61 (March), 100–120. https://doi. org/10.1016/j.tra.2014.01.002. Roth, A.E., Erev, I., 1995. Learning in extensive-form games: experimental data and simple dynamic models in the intermediate term. Games Econom. Behav. 8 (1), 164–212. https://doi. org/10.1016/S0899-8256(05)80020-X. Russel, S., Norvig, P., 2003. Artificial Intelligence—A Modern Approach, second ed. Prentice Hall, Upper Saddle River, NJ. Sandholm, T., 1999. Distributed rational decision making. In: Weiss, G. (Ed.), Multi-Agent Systems: A Modern Approach to Distributed Artificial Intelligence. MIT Press, Boston, MA, pp. 1–58. Sterman, J.D., 2004. Business Dynamics: Systems Thinking and Modeling for a Complex World. McGraw-Hill Higher Education. Tavasszy, L.A., Janssen, R., Van der Lugt, L., Hagdom, L., 2010. Verkenning Synchromodaal Transportsysteem. Delft, The Netherlands. https://mkb-mainports.hva.nl/Bibliotheek/2010/2010-­ Eindrapportage-Synchromodaliteit-TNO.pdf.

196  Intermodal freight transportation Woxenius, J., 1998a. Development of Small Scale Intermodal Freight Transportation in a Systems Context. Chalmers University of Technology. Woxenius, J., 1998b. Development of a Small-Scale Intermodal Freight Transportation in a Systems Context. University of Göteborg. http://www.fek.handels.gu.se/digitalAssets/1344/ 1344712_1998_dissertation_woxenius.pdf. Zadeh, L., 1965. Fuzzy sets. Inf. Control 8, 338–353.