Available online at www.sciencedirect.com

SciVerse ScienceDirect

Procedía - Social and Behavioral Sciences 39 (2012) 664 - 676

The Seventh International Conference on City Logistics

A system of models for the simulation of urban freight

restocking tours

Agostino Nuzzoloa, Umberto Crisallib, Antonio Comia*

a Department of Enterprise Engineering, Tor Vergata University of Rome, Via del Politécnico 1, 00133 Rome, Italy b Department of Civil Engineering, Tor Vergata University of Rome, Via del Politecnico 1, 00133 Rome, Italy

Abstract

The paper proposes a modelling system for simulating freight tours within urban and metropolitan areas using the tour-based approach. In particular, the proposed model allows us to define vehicle Origin-Destination (O-D) matrices satisfying a given delivery O-D matrices by modelling the definition of the trip-chain order and the choice of the delivery location. The modelling system has been calibrated and validated on the basis of more than 500 truck driver interviews carried out in the city centre of Rome.

© 2012 Published by El sevier Ltd . Stele ction and/or peer-review under responsibility of 7th International Conference on City Logistic s

Keywords: Urban freight transport; freight demand; restocking; trip-chain order; delivery location

1. Introduction

Today, there is a worldwide focus on setting up a sustainable development strategy to identify and define measures to achieve a continuous long-term improvement in quality of life by creating sustainable communities able to manage and use resources efficiently, tap the ecological and social innovation potential of the economy and ensure prosperity, environmental protection and social cohesion. In this context, great importance is assumed by freight transport which plays a key-role in the day-to-day activities for business and people. It becomes more evident, especially, if we analyse the recent trends of e-commerce, economic globalization, high-tech warehousing and just-in-time production systems. In addition, the partial replacement of the traditional demand for lighter and/or higher-value goods has been

* Corresponding author. Tel.: +39-06-72597059; fax: +39-06-72597053. E-mail address: comi@ing.uniroma2.it

1877-0428 © 2012 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of 7th International Conference on City Logistics doi:10.1016/j.sbspro.2012.03.138

increasing because of the spreading of on-line shopping. This implies an increase in the amount of freight transported by trucks. Moreover, when compared with the passenger vehicle fleet, trucks can have significant impacts in road congestion, greenhouse gas and pollutant emissions and pavement wear.

Thus, in order to make urban freight mobility more sustainable different measures should be implemented. The measures implemented by cities to alleviate the negative effects of freight transport have not been often supported by ex-ante assessment based on quantitative models (Filippi et al., 2010; Russo and Comi, 2011).

The most recent and sophisticated modelling systems for the simulation of the urban freight transport demand are made of the integration of two classes of models: models which simulate the level and spatial distribution of commodity exchanged within the study area (O-D matrices) and models which simulate the route choice and, thus, the traffic assignment to estimate vehicle flows on the transportation network.

For the estimation of O-D matrices several models and methods have been proposed. They usually refer to the sequential modelling approach and can be classified in terms of modelling structure and reference unit given by: truck (Ogden, 1992; Hunt and Stefan, 2007; Wang and Holguin-Veras, 2009), commodity/quantity (Ogden, 1992; Russo and Comi, 2010) and delivery (Routhier and Toiler, 2007; Nuzzolo et al., 2009).

Referring to route choice and traffic assignment, the literature reports models and methods that allow us to obtain vehicle link flows by solving Vehicle Routing Problems (Laporte, 2007; Taniguchi et al., 2007; Russo et al., 2010) derived from the operational research field. The main objective of this type of models is to deal with the vehicle assignments and the optimal routing of a single operator and/or distributor. This disaggregate approach is generally suitable to be used for operative optimization of single restockers. The planning point of view is quite different as it is not important to focus on the behaviour of single restockers but it is important to consider the average behaviour of all restockers (or categories of restockers) serving the study area for which an high level of detail is not necessary.

This paper focuses on models for vehicle O-D matrices estimation within the commodity/quantity approach. Given the complexity of representing the restocking phenomenon, the estimation of vehicles OD matrices from a given quantity or delivery O-D matrices is quite complex, and the literature only reports some applications to test cases (Raothanachonkun et al., 2007; Hunt and Stefan, 2007; Wang and Holguin-Veras, 2008) or to real cases under strong assumptions (Nuzzolo et al., 2009). The former studies propose to obtain the number of stops per tour by an incremental growth for which, at each stop, the option to come back to the base (warehouse) is considered. This approach implies relevant approximations in some real cases, especially when restockers plan tours with a pre-fixed number of stops. Referring to the real case experiences, in order to obtain the vehicle O-D matrices, Nuzzolo et al. (2009) propose to use a particular zoning of the study area that allows us to assume that, when a zone is served by a tour, all deliveries are done in the same traffic zone (average quantity approach). This assumption is based on some survey results that pointed out the restocking strategy for which decision-makers organize tours trying to serve all customers located in the nearby.

The proposed modelling system allows us to overcome these limits pointing out the pre-trip definition (i.e. before starting the journey) of the number of stops for restocking journey and, successively, the choice of delivery point location. We will also consider that restockers can have a different behaviour in relation to journey characteristics (round trip vs. trip chain).

The here proposed modelling system is a component of a general framework for modelling urban freight transport developed by authors (Nuzzolo et al., 2010). It is structured in two levels and allows us to integrate the advantages of the three recalled simulation approaches for O-D matrix estimation: commodities/quantities (to capture the mechanisms underlying the generation of freight demand), deliveries (to follow the decisional and logistic process of restocking) and trucks (input for traffic assignment models, Fig. 1):

• demand sub-system; it allows us to estimate O-D matrices both in terms of quantities and deliveries; this level concerns the calculation of

O quantity O-D matrices: freight O-D flows in quantities (e.g. tons per day) are estimated starting

from socio-economic data, O quantity O-D matrices per transport service type: freight O-D flows in quantities are split in terms of who can be assumed as the decision-maker of the restocking process and of the used transport service: receiver (e.g. retailer) in own account, sender (e.g. producer, wholesaler) in own account, third party by transport company and third party by courier, O delivery O-D matrices per transport service type: the quantity O-D flows per transport service type are converted into delivery flows; the delivery step is modelled through the evaluation of freight quantity (shipment size) delivered by different identified transport service types, O delivery O-D matrices per time slice: the delivery O-D flows are characterised for target time, O delivery O-D matrices per vehicle type: the delivery O-D flows per target time are split in terms of used vehicle type (e.g. Light Goods Vehicles, Heavy Goods Vehicles);

• logistic sub-system; it allows delivery flows to be converted into vehicle flows (both loaded and empty) used for restocking the study area for different freight types; this level concerns the calculation of

O delivery O-D matrices for n-th order trip chain: the delivery O-D flows obtained from the demand

sub-system are split in terms of journey characteristics (i.e. number of stops for journey); O vehicle O-D matrices: the delivery O-D matrices for n-th order trip chain are allocated to each trip within the considered trip chain or tour.

Fig. 1. System models architecture for simulating urban goods movements

The paper focuses on the logistic sub-system. Section 2 describes the modelling structure of the recalled sub-system focusing on the modeling of restocking journey (i.e. round trips and trip-chain/tour). Section 3 presents the results of calibration using data collected in the city center of Rome. Finally, in Section 4 some conclusions are given.

2. The logistic sub-system

The demand sub-system allows us to estimate the delivery O-D matrices per restocking type, delivery time and vehicle type. However, the logistic sub-system allows us to convert these matrices into vehicles.

In order to obtain the link flows on the road network, the estimation of vehicle O-D matrices and the relative used paths for each od pair are required. The vehicle O-D matrices are made of elements VCOdh[r,x, v, n] which represents the average vehicle flow in time period h (e.g. working day) on the O-D

pair od for transport service type r and vehicle type v moving freight type s in time slice z (e.g. 10:00 -10:15 a.m.). They can be obtained starting from the output of the demand sub-system (delivery O-D matrices per vehicle type). In the following, for semplicity of notation, the class indexes s (freight type) and h (time period) will be omitted unless otherwise stated.

For reader convenience, it is important to introduce some definitions and notations used in the following of this paper. A trip is a vehicle movement between two stops (one origin and one destination zone) for picking up/delivering. A journey is a sequence of trips starting and ending at base (e.g. warehouse). Journey with a single destination is a round trip; a journey made of a sequence of trips is a tour or a trip chain. A tour can also have one destination zone but many stops; it happens if all shops to be served/restocked are located within the same traffic zone d (i.e. stops within the same zone are also considered). These tours are simulated as round trips. Furthermore, we assume that a single delivery is performed at each delivery point (stop).

The reader should consider that restocking deliveries are usually planned by tours made of multiple trips among stops, and trips of a given journey are defined according to logistic decisions. In order to simulate the existing dependences among successive trips (of the same journey) according to a spatial-temporal connection among activities of different trips (e.g. with respect to time constraints), the tour-based approach has to be used. The tour-based approach allows us to define trips of a journey assuming that the choices for each trip influence other trips belonging to the same journey. Hence, tour-based models ably reproduce the choice structure of freight transport simulating the existing dependences among successive trips of the same journey (i.e. sequence of intermediate stops between restockers and retailers). It implies that each stop is chosen according to the previous and the next stops. In order to implement this model it is essential to understand the constraints underlying all the decision-makers' choices. This approach is relatively new; it has been mainly developed to simulate passenger mobility (Hunt and Stefan, 2007) and only few examples have been implemented for freight mobility in extraurban areas (Russo and Carteni, 2006). We propose to apply the tour-based approach to urban and metropolitan areas aiming at estimating O-D matrices in vehicles. Considering the different decision-makers of each identified transport service, as well as following the sequence of choices done by each of them, the proposed modelling structure receives as input the delivery O-D matrices per vehicle types and gives as output freight journeys satisfying the input matrices, through which the vehicle O-D matrices are carried out. The O-D matrices in deliveries are quite different from O-D matrices in vehicles. In fact, the O-D matrices in vehicles are similar to the O-D matrices in deliveries in the cases of round trips because the origin and destination of the movement of both deliveries and vehicles are the same. However, the delivery O-D matrices are also relatively different when the vehicles move through different stops along the tour. For this reason, the estimation of vehicle O-D matrices requires the definition of freight journeys

that can be carried out by modelling the definition of the trip chain order (that is the number of stops in a tour departing from a given origin zone) and the choice of the delivery location.

The reader should consider that the restocker knows the total number of deliveries for each destination. The deliveries are subjected to some constraints, such as number and type of available vehicles and time windows specified for each customer to be restocked. Thus, the decision-maker organizes the restocking journey for each available vehicle optimising the total perceived costs. This disaggregate process is generally approached through operative research methods that are suitable to be used for operative optimization of single restockers. As previously described, the planner point of view is quite different, and the analyst can usually approach this problem in two different ways:

• the first is based on results of behavioural surveys to restockers in order to investigate how socioeconomic and level-of-service attributes influence tour definition (i.e. before the decision);

• the latter is based on results of surveys to drivers (i.e. after the decision, capturing the results of the decision), that is the approach used in this paper to simulate the restocking process.

We assume that at each stop (delivery point) we have only one delivery (i.e. one vehicle trip is equal to one delivery trip).

Hence, given the flows of deliveries performed by journeys of order n (i.e. n deliveries per journey) departing from origin zone o operated by transport service r and vehicle type v in time slice r, the number of vehicles on each (dt, dj) pair, VCdd [r,t, v, n], can be calculated solving the following system of linear equations:

VCo.[r,x, v, n] = Y,VCod, [r,x, v,n] = Y,NDodj [r,x, v, n]/n = £NDodj [r,x, v] • p[n/ vxro]Wn

dj ' dj ' / dj ' /

VC.' [r,X, v, n] = XVCd^' [r,X, v,n] = NDodj [r,x, v, n]

VCJlj2 [r, X, v, n] = VC Jt [r, X, v, n] • p\d2/d1 nvro 1

d d od 1 L J

VCdl [r, X, v, n] = VCod} [r, X, v, n] • p [dk/d1 nvro] VCdldl+1 [r, x, v, n] = VCdld [r, x, v, n] • p [dk+1/dk nvro] where,

VCo [r, t, v,n]: is the number of vehicles departing from origin zone o (wholesaler zone);

VC. d [r, t, v, n]: is the number of vehicles arriving to deliver in zone dj;

ND^d [r, t, v, n]: is the number of deliveries performed in zone d' by journeys of order n

departing from zone o; NDod [r, x, v]: is the number of deliveries performed on odj pair;

p[n / vxro] : is the probability that deliveries are performed by a journey departing from

zone o with n deliveries; it can be obtained by a trip chain order model; p [dk+1 / dknvro~j : is the probability to perform the delivery (k+1) in the zone dk+1, conditioned

to a previous delivery in zone d k within a journey with n deliveries by

transport service r departing from zone o; it can be obtained by a delivery location choice model.

2.1. Trip chain order model

The trip chain order model allows us to define the average number of stops that characterize the undertaken journey. It allows us to obtain the delivery O-D matrices per n-order trip chain order by disaggregate the flow of deliveries NDod [r, x, v]. Let be p[n / vxro] the probability that deliveries are performed by a journey with n stops departing from zone o. In the sphere of the Random Utility Theory, the trip chain order model can be specified assuming that the random residuals are i.i.d. (s) as Gumbel variables. Under this hypothesis the probability p[n / vxro] can be calculated as:

p [n/vxm ] = exp (Vn y X exp (V.)

where Vn is the systematic utility for a tour with n stops departing from zone o.

This step allows us to assess the city logistics measures impacting on Level-of-Service attributes, and thus the number of stops per tour. The number of deliveries per journey could be modified by the implementation of city logistics solutions pushing towards more efficient transport services in terms of transported load and generalized travel cost; for example, the time windows and pricing modify the accessibility increasing the number of journeys, which are usually non-optimized in terms of transported load, the vehicle constraints (e.g. weight) reduce the transported load implying tours with few deliveries, the incentives to third party force to optimized tours. At the other hand, the Intelligent Transportation Systems (ITSs) can play a key-role. The interest in Intelligent Transportation Systems (ITS) has been growing in recent years. The ITCs are used to improve route and trip planning as well as services provided to customers. Some of these systems have been implemented by local authorities in order to support the traffic management of the city, others are privately operated. The privately-operated systems are mainly concerned with optimising logistics and distribution processes, hence contributing to a cost optimisation of the supply chain and optimising the sequence of stops for loading and unloading.

2.2. Delivery location choice model

As previously described, a journey can be characterised by one or more than one stops, and the choice process of stop location is quite different. The delivery location choice model allows us to define the sequence of zones performed during the tour. The probability p ^dk+1 / dk nvro'j can be simulated by

Random Utility Models (e.g. multinomial logit) for which p[dk+1 / dk nvro] can be written as:

p [dk+l /dk nvro] = exp(v^ ysexp (Vr)

where Vkis the systematic utility of delivering in zone dk+1 conditioned to have previously delivered in

zone dk within a tour with n deliveries that departs from zone o.

The systemic utility could include two groups of attributes: the first considers all variables associated with an alternative destination such as accessibility; the second includes the memory variables representing the history of a tour such as the cumulative distance covered up to the current location. Thus,

the delivery location model allows us to point out the impacts due to implementation of city logistics that can modify the generalized travel cost and, thus, the accessibility: time windows, pricing and vehicle constraints modify the accessibility pushing to long trips within the same tour, ITS (e.g. traffic information system) pushes to choose close destinations reducing the travelled distance.

3. Specification and calibration

The calibration has been carried out using a dataset of more than 500 interviews to truck drivers. These interviews are part of a survey campaign in the city of Rome consisting of traffic counts and interviews to retailers and to truck drivers (Comi et al., 2011).

The municipality area has been divided into 99 traffic zones with a higher level for the inner area, which has been derived by aggregating 760 census parcels.

The focus regards the inner city area in order to analyze the restocking process of the area (6 km2), with more than 50,000 inhabitants and less than 24,000 employees related to trade. Access to inner city is allowed exclusively to vehicles with certain emissions characteristics (no access to pre-Euro vehicles) and with a gross laden weight of less than 3.5 tones. The access is also allowed to vehicles with a gross laden weight less than 8.5 tones only in the night hours and restricted to some specific roads.

This survey also allows us to characterize the freight quantity moved within the study area. The composition of freight flows in tonnes has been estimated on the basis of counts and truck-driver surveys. The study area is a trading area which is mainly interested by attraction freight flows. The analysis highlights freight movements in the study area for about 15,000 tonnes per day. Considering the revealed freight segmentation, 36% consists of food (about 16% is dispatched to restaurants and bars, and 14% to retailer) and the remaining 70% are made of other end-consumer products (e.g. household and health products).

In order to define the relevant attributes which may characterise restocker behaviour, a detailed analysis on socio-economic and level-of-service attributes has been carried out. The analysis pointed out that restockers prefer to undertake round trips if they are located in a zone with high level of accessibility because round trips allow them to reduce journey operation planning. Thus, as each restocker has a prefixed working time, he could do many trips without losing time for travelling if he is located in a zone with high accessibility. The retailer prefers round trips, but third party prefers long tours (more than 2 stops). The survey analysis also allowed us to understand that the trucks moving foodstuffs are characterised by tour with many stops. It is strictly related to the freight type that is characterised by daily consumption products coming from different companies. For what concerns the vehicle type analysis, survey highlights the use of light goods vehicles for tours with few stops.

Referring to transport freight types, available data allowed us to investigate the joint share in terms of: receiver in own account, sender in own account and third party. We revealed that the average number of stops is about 2 and that it is quite sensitive to freight segmentation. The retailer tends to undertake tours with fewer stops per tour. A detailed analysis pointed out a higher value of stops per tour for foodstuffs (about 2.4 stops per tour), while the lower values refer to home accessories and building materials (about 1.8 stops per tour). This analysis allowed us to identify four alternatives for the trip chain order model: one stop, two stops, three stops and more than three stops.

The stop location choice model has to consider a huge number of elemental alternative destinations, which makes the estimation process next to impossible. To overcome this problem, the definition of the available alternatives for the stop location has been defined in relation to the maximum distance revealed in the sample. The distribution of length tour in the sample is reported in Fig. 2. We can see that the 70% of distance from the current location to next one is less than 5 km (including the distance from warehouse to first stop). This analysis allowed us to insert in the choice set only the locations with a distance from

current location less than 25 km. Further analysis is in progress in order to specify and calibrate model for the choice set definition (i.e. choice set modelling).

Fig. 2. Length of stop-to-stop trips

The specification and calibration of the two models presented previously is here reported. Models have been calibrated using the Maximum Likelihood (LM) estimator within the classic theory of statistical inference. The presented models are the result of several specifications and calibrations based on different combinations of possible attributes. In the following, the models that performed the best statistical significances are reported.

3.1. Trip chain order model

The calibrated model considers four alternatives: one stop/delivery (1), two stops/deliveries (2), three stops/deliveries (3) and more than three stops/deliveries (3+). The data analysis pointed out that the probability of having a journey with n stops does not depend on destination zone attributes (Nuzzolo et al., 2011) but by the zone where the warehouse is located (origin zone of journey). They investigated the attributes impacting on trip chain order and presented a first aggregate model for simulating this choice considering a smaller number of alternatives with respect to the four above.

In order to overcome this limit, this paper presents a system of models, which includes both the trip chain order choice and the delivery location one.

For what concerns the trip chain order model, the systematic utilities of each alternative are expressed as a function of attributes relative to origin zone, freight, transport service and vehicle types:

V = Paa • In (IAA0) + ßr • RET + ßVi • VEH + ft • ASA

V = ßfe -q + ßv2 •VEH + ß/2-FGT + ft • ASA V3 =ßft •? + ßc% •CT + ß/3 • FGT + ft • ASA,

V3+=ßfe +-q + * 3 + ^/3 ,'FGT

where,

Vi : is the systematic utility for a journey with one delivery (i.e. round trip),

V2 : is the systematic utility for a journey with two deliveries,

V3 : is the systematic utility for a journey with three deliveries,

V3+ : is the systematic utility for a journey with more than three deliveries,

IAAo : is the retailer accessibility index of zone o, from which the tour departs (e.g. warehouse

location),

q : is the average quantity of freight delivered at each stop along the tour, expressed in

RET: is a dummy variable equal to 1 if the transport service is retailer in own account, 0

otherwise,

CT : is a dummy variable equal to 1 if the transport service is third party, 0 otherwise,

VEH: is a dummy variable equal to 1 if the used vehicle is a light goods vehicles, 0

otherwise,

FGT: is a dummy variable equal to 1 if the delivered freight belongs to the foodstuffs class, 0

otherwise,

A5A1: is the Alternative Specific Attribute equal to 1 for a tour with one stop/delivery, 0

otherwise,

ASA2: is the Alternative Specific Attribute equal to 1 for a tour with two stops/deliveries, 0

otherwise;

ASA3 : is the Alternative Specific Attribute equal to 1 for a tour with three stops/deliveries, 0

otherwise.

The retailer accessibility index IAAo is calculated as:

IAA„ =

AA - min (

(AA)]/[ max {AA)" min (AA,)_

where AAx is the retailer accessibility of zone x estimated as:

= X {ULj Y ■ eXP [a2 • distxj ]

ULf the number of retail establishments of zone j to be restocked,

distj the distance between zone x and j, calculated on the road network according the path of

minimum generalised travel cost, a¡ and a2: calibration parameters.

The results of the calibration are reported in Table 1. We can see that all parameters have the expected signs and most of them are statistically significant. Parameter analysis shows the important role of the accessibility; in fact, we can see that increasing the accessibility, the number of stops per tour decreases. It confirms that the restocker prefers to do round trips if warehouse is located in a zone with a high accessibility as it allows them to reduce the operation complexity of tour management. The probability to have tours with more than 2 stops increases for foodstuffs. Finally, as expected, the positive sign of vehicle type parameters confirm that the probability to have round trips increases for light goods vehicles and small delivered quantities (negative values of average delivered quantity at stop). Referring to transport service type, the positive values of parameters reflect that the probability to do round trips grows

for retailer in own account, while the probability to have many stops per journey increases for third party. Even though they are the first results of the calibration of the trip chain order model, the value of p2 is promising.

Table 1. Trip chain order model: calibration results

Attribute Parameter Alternatives

1 delivery 2 deliveries 3 deliveries 3+ deliveries

Retailer accessibility (lAAo) 0.088 (3.2)

Delivered quantity (qx) ß, -0.014 (-1.1) -0.024 (-1.3) -0.672 (-1.4)

Retailer (RETx) ßr 0.202 (2.7)

Third party (CTX) pa 0.193 (2.8) 0.382 (2.9)

Freight type (foodstuffs; FGTX) ß / 0.556 (1.9) 1.182 (3.7) 2.067 (4.1)

Vehicle type (Light Goods ßv 1.123 0.879 0.978

Vehicle; VEHX) (2.6) (2.0) (2.1)

Alternative Specific Attribute Px 2.102 2.069 1.109

(ASAX) (3.6) (3.9) (2.0)

Accessibility ai =3.871 (2.7); = -4.953 (-2.2)

(-) t-st value

3.2. Delivery location choice model

The survey has revealed that different behaviours could be followed by a restocker in the choice of first destination according to journey with one destination (round trip) or more than one destinations (tour or trip chain). Thus, two different models for the choice of delivery location have been calibrated. The calibrated models has been developed within the Random Utility Theory as multinomial logit in which the systematic utility function related to zone d, Vd, has been expressed as a linear functions of two sets of attributes: all variables associated with a destination alternative (e.g. accessibility) and "memory" variables representing the history of tour (e.g. cumulative distance covered up to the current location).

A set of multinomial logit models which include different combinations of independent variables were tested. The best statistically results have been obtained considering as attributes:

• the retailer accessibility index of zone d (IAAd);

• the wholesaler accessibility index of zone d (IAPd);

• the natural logarithm of ratio between the covered distance up current location and the distance from the current location to next stop; it is applicable from the second stop/delivery (HTd);

• a dummy variable equal to 1 if the next stop is within the same current zone, 0 otherwise (ASAod).

The retailer accessibility index, IAAd, has been calculated as previously described (see eq. 1). The wholesaler accessibility index, IAPd, has been calculated as:

IAPd =

APd - min (

in (APz ) j max (APz )- min (APz )

where APx is the wholesaler accessibility of zone x estimated as:

APx f • exp[«4 • distx]

WHf: the number of warehouses of zone i,

distix: the distance between zone i and x, calculated on the road network according the path of

minimum generalised travel cost, a3 and a4: calibration parameters.

The estimated models both for round trip and trip chain are reported, respectively, in Table 2 and 3. As shown, all variables are statistically significant. The results show that both retailer and wholesaler accessibility influence the choice of delivery location. In particular, in the round trip the choice of destination depends on wholesaler accessibility, while in trip chain both the wholesaler and the retailer accessibility play a key-role. In fact, we have that the wholesaler accessibility weights more than retailer one in the systematic utility of delivery location. In general, it means that the tour planner uses to choose for the next delivery a zone that is easy to be reached and is well connected to the others. The sign of the memory parameter demonstrates that the systematic utility of choosing a destination is a cost function of the distance from the current location. It implies that farer destinations are those have a lower probability to be chosen. The good values of p2 demonstrate the goodness of proposed approach.

Table 2. Delivery location choice: calibration results for round trips

Attribute Symbol Parameter Value

Wholesaler accessibility IAP Pap, 5.452 (14.7)

Alternative Specific Attribute ASAod Px 1.618 (4.2)

Accessibility a3 CLf 0.43 (-2.0) -1.26 (-1.8)

p2 0.30

(-) t-st value

Table 3. Delivery location choice: calibration results for trip chains

Attribute Symbol Parameter Value

Retailer accessibility IAA Plaax 2.148 (10.4)

Wholesaler accessibility IAP Pap, 2.638 (12.6)

Covered distance HT Pht.< 0.320 (4.1)

Alternative Specific Attribute ASAod Px 0.847 (5.9)

Accessibility a3 CLf 0.43 (-2.0) -1.26 (-1.8)

(-) t-st value

4. Conclusions

This paper presents a modelling system for simulating urban freight restocking journeys. It consists of two models developed within the tour-based approach: trip chain order and delivery location choice models. The former is used to characterize the journeys serving on O-D pair od in terms of total number of stops. The latter is used to define the delivery locations of a given journey. A dataset of urban freight data collected in the centre of Rome has been used to support the model specification and calibration. In particular, the calibration of the trip chain order model pointed out the key-role of the warehouse location, for which round trips are preferred for highly accessible zones. The delivery choice location evidences the importance of retailer and wholesaler accessibility of zones to be served.

In particular, the trip chain order model allows us to investigate the trip chain characteristics. Given a trip chain, the number of stops can be modified by the implementation of city logistics solutions pushing towards more efficient transport services in terms of transported load and generalised travel cost. In this field, the role of Intelligent Transportation Systems (ITS) plays a key-role. They can be used to improve route and trip planning services as well as information to customers, provided by public authorities and private operators. Local authorities usually support the city traffic management while private operators usually optimize logistics and distribution processes, aiming at reducing costs of supply chain and optimising the sequence of stops for loading and unloading. Implementation of measures affecting the types of permitted vehicle could influence the transported load and, thus, the number of stops on the delivery tour.

The calibrated model has been developed to test the goodness of the proposed approach, which explicitly considers the interaction between demand and logistics. Although the obtained statistics confirm the goodness of the proposed approach, further developments are required to specify, calibrate and validate the modelling system by using larger and more detailed samples.

Further developments of this research are mainly related to: calibration of further trip chain order and stop choice location models for each transport service type; investigation of the influence of shipment size, dimension of retail and zone to be served for the definition of the order trip chain; modeling of choice set generation within the delivery location choice model. Finally, further research perspectives could aim to develop this modelling system within a Decision Support System to be used for evaluating impacts of urban freight transport measures.

References

[1] Comi A, Delle Site P, Filippi F, Nuzzolo A. Ex-post assessment of city logistics measures: the case of Rome. In: Mussone L,

Crisalli U, editors. Transport management and land-use effects in presence of unusual demand, Franco Angeli, Milan, Italy, 2011, p. 235-252.

[2] Filippi F, Nuzzolo A, Comi A, Delle Site P. Ex-ante assessment of urban freight transport policies. Procedia - Social and

Behavioral Sciences 2010; 2(3): 6332-6342.

[3] Hunt JD, Stefan KJ. Tour-based microsimulation of urban commercial movements. In: Transportation Research Part B 2007;

41(9): 981-1013.

[4] Laporte G. What you should know about the vehicle routing problem. Les Cahiers du GERAD, G-2004-33, HEC Montreal;

[5] Nuzzolo A, Crisalli U, Comi A. A delivery approach modeling for urban freight restocking. Proceedings of the 12th International

conference on Travel Behaviour Research (IATBR), Jaipur, India; 2009.

[6] Nuzzolo A, Crisalli U, Comi A, Galuppi S. Demand models for the estimation of urban goods movements: An application to the

city of Rome. In: Viegas JM, Macario R, editors. Selected Proceedings of 12th World Conference on Transportation Research-WCTR 2010, Lisbon, Portugal.

[7] Nuzzolo A, Crisalli U, Comi A. A trip chain order model for simulating urban freight restocking. In: European Transport,

Special issue on: Freight transport analysis: New trends and methodologies, Trieste, Italy, 2011.

[8] Ogden KW. Urban goods movement. Ashgate, Hants, England; 1992.

[9] Raothanachonkun P, Sano K, Wisetjindawa W, Matsumoto S. Truck trips origin destination using commodity based model

combined with an empty trip model. Proceedings of the 86th Transportation Research Board Annual Meeting, Washington DC, U.S.A., 2007.

[10] Routhier JL, Toilier F. FRETURB V3, a policy oriented software of modelling urban goods movements. Proceedings of the 11th World Conference on Transport Research, Berkeley CA, U.S.A., 2007.

[11] Russo F, Carteni A. Application of a tour-based model to simulate freight distribution in a large urbanized area. In: Taniguchi E, Thompson RG, editors. Recent advances in city logistics, Elsevier Ltd., United Kingdom; 2006, p. 31-46.

[12] Russo F, Comi A. A modelling system to simulate goods movements at an urban scale. In: Transportation 2010; 37(6): 9871009. DOI: 10.1007/s11116-010-9276-y.

[13] Russo F, Comi A. A model system for the ex-ante assessment of city logistics measures. Research in Transportation Economics 2011, 31 (1), DOI: 10.1016/j.retrec.2010.11.011, Elsevier Ltd., 81-87.

[14] Russo, F., Vitetta, A. and Polimeni, A. (2010). From single path to vehicle routing: the retailer delivery approach. In: Procedia - Social and Behavioral Sciences 2011; 2(3): 6378-6386.

[15] Taniguchi E, Ando N, Okamoto M. Dynamic vehicle routing and scheduling with real time travel times on road network. Journal of the Eastern Asia Society for Transportation Studies 2007; 7: 1138-1153.

[16] Wang Q, Holguin-Veras J. An investigation on the attributes determining trip chaining behavior in hybrid micro-simulation urban freight models. Proceedings of the 87th Transportation Research Board Annual Meeting, Washington DC, U.S.A., 2008.

[17] Wang Q, Holguin-Veras J. Tour-based entropy maximization formulations of urban freight demand. Proceedings of the 88th Transportation Research Board Annual Meeting 2009, Washington DC, U.S.A.