Professional Documents
Culture Documents
To cite this article: Xian Cheng, Shaoyi Liao & Zhongsheng Hua (2016): A policy of picking
up parcels for express courier service in dynamic environments, International Journal of
Production Research, DOI: 10.1080/00207543.2016.1231431
Download by: [Northern Illinois University] Date: 17 September 2016, At: 06:01
International Journal of Production Research, 2016
http://dx.doi.org/10.1080/00207543.2016.1231431
As a particular logistics service, the express courier service has seen considerable growth recently, which resulted in an
unprecedented fierce competition. Besides, the development of information and communication technologies has enabled
express company to manage their service. With the purpose of improving service quality and operation efficiency for
express company, we focus on the problem of intercity express courier routing in courier-triggered pickup service. A
novel pickup policy for courier routing is proposed based on the idea of centrality measures and the nearest-neighbour
(NN) policy by considering the un-serviced customer requests as a globally coupled network. This policy enables to
dispatch the idle courier to the more central request location, which allows the courier to easily serve the neighbouring
requests around the central request location, thus securing both global and local performance. We also propose a simple
prototype of real-time fleet management system where the proposed pickup policy is embedded into it. To evaluate the
efficiency and practicability of the pickup policy, we conduct comprehensive computational experiments to generate
various testing scenarios; moreover, two widely used dispatching policies – NN and first-come-first-served (FCFS) – are
considered as the benchmark policy. Results show that the proposed pickup policy significantly outperforms the NN and
FCFS policies in terms of waiting time and total service time.
Keywords: express courier service; pickup process; dynamic routing; network analysis; centrality measures; dispatching
policy
1. Introduction
Express courier service pertains to the provision of value-added, door-to-door transport and next-day (or time-definite,
usually several days) delivery of shipments, such as documents, parcels and merchandise goods (henceforth, parcels for
short) (Forecasting 2005). Recently, there has been a growth in the shipment of individual parcels as the development
of diversification of commercial mode, such as business-to-consumer (B2C), consumer-to-consumer (C2C) and
e-commerce. So the demand for express courier service has risen over time. Taking China, for example, according to
the State Post Bureau of China, the annual growth rate of express business volume has been maintained at higher
than 50% for four consecutive years since 2011, and the business volume of express courier service has grown from
1.5 billion in 2008 to 13.96 billion in 2014 (http://www.spb.gov.cn/) (see Figure 1 for detailed information). Moreover,
the intensified competition between express companies is unprecedented. As of 2013, approximately 8000 express com-
panies exist in China. In addition, labour cost is rising dramatically. With these considerations, the express company has
to invite new and efficient technique (e.g. information and communication technologies) for managing express courier
service operations with the aim of gaining competence in the industry, sustaining high-quality service, reducing opera-
tion cost and keeping and attracting more customers.
For next-day or time-definite express courier service, parcels have to be handled and often be transported in large
distances (from one city to another city) in order to be delivered on time. We denote this kind of service as intercity
express courier service.1 Figure 2 shows the typical process of intercity express courier service in which several opera-
tions are included. Following the flow of parcels, these operations can be divided into three subprocesses, namely
pickup, sorting and delivery. In pickup process, the courier service occurs in the place where the parcels originated, the
couriers pick up the parcels at the customer locations after receiving the customer requests, the parcels are consolidated
into conveyance (usually electric bikes or motorcycles) and transferred to the local station. In sorting process, parcels
are sorted at location station based on their destinations, volume and other factors, parcels are then moved to a major
regional sorting facility called ramp or hub for further sorting, usually by truck. Finally, during delivery process, parcels
Figure 2. The intercity express courier service process. Here we highlight the pickup process.
are firstly delivered to the destination ramp by aeroplanes or trains, then to the destination location by truck and ulti-
mately to the end customer by electric bikes or motorcycles. Couriers routing are needed in both pickup and delivery
processes. However, compared with the delivery process, there are more challenges during the pickup process, such as
new customer requests arriving, uncertain service demand, we will describe these challenges in detail in subsection 1.1;
besides, in pickup process the couriers have to meet customers, express company has to pay much attention to improve
the service quality, in order to keep and attract more customers. Motivated by these considerations, we mainly focus on
the pickup process in this paper with the objectives of improving service quality and operation efficiency for express
company.
Two measures can be used to evaluate the express company’s service quality and operation efficiency in the pickup
process. One is customer waiting time, which is defined as the time taken by a courier dispatched to serve the customer
after a request is arrived. Customer waiting time reflects the service quality of pickup for an express company because a
short waiting time denotes a short response time for the company, and further reflects its agility. The other measure is
the total time to serve a set of customer requests, which can be used to evaluate the operation efficiency as spending a
small amount of time to serve a certain request reflects a high operation efficiency. The fleet managers of express com-
pany should formulate a pickup policy that focuses on the manner of dispatching couriers to serve customer requests to
reduce customer waiting time and the total service time. In general, there are two types of decisions for fleet managers
in pickup process: request-triggered decision and courier-triggered decision. When a new customer request arrives and
several couriers are idle, the request triggers a decision of selecting an appropriate courier to serve this request
(request-triggered). On the other hand, if requests can’t be served immediately and should be queued, and a courier who
has just been idle from picking up has to select a request from the waiting queue, a courier-dispatching decision
(courier-triggered) is triggered. The request-triggered decision is more relevant in routine situation where the extent of
busyness of the service district is relatively low, e.g. the suburban district, in request-triggered scenario the customer
International Journal of Production Research 3
requests are served easily during the projected service time as idle couriers are always available. Conversely, in the
condition where the extent of busyness is high, the courier-triggered decisions come into play. This scenario often
occurs in central business districts of a city where the demand of pickup service is always high. The courier-triggered
decisions can also be occurred during the promotion activity for e-commerce platforms, e.g. the promotion of Chinese
Singles Day at Taobao.com, or the Cyber Monday in United States. The demand of shipment of individual parcels
emerges in large numbers during the promotion activity. Compared with request-triggered decision, the courier-triggered
decision challenges the fleet managers as this type of decision is often associated with the effort of minimising response
time to meet the projected service time as the customer request has to be queued, the fleet managers have to adopt
effective pickup policy to serve customer requests. Motivated by this situation, here we consider the situation of cour-
ier-triggered decision in pickup process.
1.2 Contribution
As characterised by: (1), dynamic changing of the courier routing schedule as the real-time information of new requests
arriving; and (2), considering the uncertainty of parcels and customer service demands, the considered courier routing
4 X. Cheng et al.
problem is distinct from the traditional vehicle routing problems (VRPs) (Tsai, Tsai, et al. 2003; Pillac, Gendreau, et al.
2013). Indeed, the arrival of new request is considered in the dynamic VRPs (Yan, Lin, et al. 2013; Ferrucci and Bock
2014; Lin, Choy, et al. 2014). However, to the best of the author’s knowledge, the uncertainty of parcels and customer
service demands has not been previously explored in the literature.
In this paper, we propose a novel pickup policy for real-time courier routing in an effort to help fleet managers make
effective decisions in reducing customer waiting time and total service time, and based on this pickup policy, we
develop a simple prototype of real-time fleet management information system. In detail, concentrating on the courier-
triggered decision situations in which many requests are queued, the proposed pickup policy considers these un-serviced
requests as a globally coupled network according to the request’s geographical location. It is a dynamic network because
of the arriving of new requests and removing of serviced requests. And then the centrality measures, which are adopted
from network science (Friedkin 1991; Leem and Chun 2014), are used to measure the centrality of the un-serviced
requests. The centrality measures prioritise the un-serviced requests based on the centrality of requests as it evaluates
the efficiency of a request in reaching to other requests. However, if we dispatch the courier merely by the centrality
measures, the courier may always excessively move to the most central requests thus would ignore to serve the neigh-
bouring requests, potentially leading to waste the travelling time. In order to deal with this problem, we integrate the
nearest-neighbour (NN) policy into centrality measure. The NN policy, which is widely used in traditional VRPs
(Egbelu and Tanchoco 1984; Bertsimas and van Ryzin 1991; Mantel and Landeweerd 1995; Dean 2008), is defined as
dispatching a courier to the closest request, thus the NN policy has the capacity of pursuing the local performance.
Based on the integration of the centrality measures and the NN policy, the proposed pickup policy enables to dispatch
the idle courier to the more central request location, which allows the courier to easily serve the neighbouring requests
around the central request location, thus securing both global and local performance. The main contributions of this
paper are as follows:
(1) we construct a dynamic globally coupled network of queued requests which enables us to measure the centrality
of those requests. The centrality measure provides a global view of request distribution and thus secures the glo-
bal performance. At the same time, in order to pursue the local performance, we also integrate the NN policy
into the proposed pickup policy, as the NN policy always tries to minimise the current travel time. Moreover,
the new arriving requests are added into the globally coupled network and served requests are removed from it,
so the dynamic globally coupled network integrates the real-time information.
(2) we develop a probability model to cope with the uncertainty of parcels and customer service demands. As we
mentioned, the uncertainty of volume and quantity of the parcels and the uncertainty of customer service
demands determines the possibility that the courier goes back to local station to place the parcels, once he fin-
ished serving a current request. These uncertainty events exhibit a stochastic nature, thus can be model based on
a probability of going back to local station for a courier once he finished serving the current request. Indeed, it
is not necessary for a courier to always go back to the local location for each pickup service. However, once the
vehicle is filled with parcels, or the courier encounters special customer service or special parcel (e.g. fragile
products, fresh fruits), he has to go back to the local location. So the probability model is well suited to model
these stochastic events.
(3) we propose a simple prototype of real-time fleet management system in which the proposed pickup policy is
embedded. There are two main modules in this fleet management system: data processing module and dynamic
routing module. As the dynamically arriving of new customer requests, the data processing module is used to
construct the globally coupled network by recording the location for new requests and the courier current loca-
tion, and removing the location for served requests. And then the globally coupled network is considered as the
input of the dynamic routing module in which the proposed pickup policy is integrated. Based on this policy,
the suitable dispatch information is obtained and then sent to idle courier by a real-time data receiver.
(4) we design various test scenarios to evaluate the efficiency and practicability of the proposed pickup policy by
compared with two widely used dispatching policies – the NN policy and the first-come-first-served (FCFS)
policy (Chen 1999; Kuo 2013).
1.3 Organisation
The remainder of this paper is organised as follows. Section 2 briefly reviews the related work. Section 3 describes the
proposed pickup policy and proposes simple prototype of real-time fleet management system. Section 4 performs the
computational experiments and analyses the results. Section 5 summarises the paper and presents the direction for future
work.
International Journal of Production Research 5
2. Related literature
Due to the challenges as mentioned in former section, the considered problem – intercity express courier routing in cour-
ier-triggered pickup service – has not been previously studied in the literature. However, ignoring the probability of going
back to the local station, this problem has been considered as the VRP (Coelho and Laporte 2013; Yousefikhoshbakht,
Didehvar, et al. 2014; Mavrovouniotis and Yang 2015). Besides, the centrality measures are adopted in the proposed
pickup policy. So this section we focus on the related literature of VRP (especially the dynamic VRP) and the centrality
measures.
Sungur, Ren, et al. One-day service; pickup No real-time situation Maximising customer coverage;
(2010) and delivery minimising total time
Chang and Yen One-day service; pickup No real-time situation Minimising total time; minimising
(2012) and delivery unbalanced workload
Lin, Yan, et al. Several-days service; Stochastic travel time Minimising operating costs
(2013) pickup
Yan, Lin, et al. Several-days service; Stochastic travel time; New requests; Minimising operating costs
(2013) pickup
Lin, Choy, et al. Several-days service; New requests; Requests cancellations Minimisng total travel distance
(2014) pickup
Ferrucci and Bock One-day service; pickup New requests; traffic congestion; Minimisng lateness; minimising vehicle
(2014) and delivery Vehicle disturbance operating costs
6 X. Cheng et al.
Guéret, et al. (2012), the authors design a Java event-driven framework which focuses on the highly flexible online opti-
misation in dynamic operating environments. Concentrating on the public logistics information service management and
optimisation, Hu and Sheng proposes a decision support system (DSS) for fleet management (Hu and Sheng 2014). This
DSS employs the intelligent positioning technologies to acquire and manage the vehicle status. In the field of express
courier service, Lin and his co-authors develop a DSS to deal with real-time customer requests (including new customer
orders and order cancellations) by integrating with the dynamic courier routing model (Lin, Choy, et al. 2014). Many
other real fleet management systems are also published in the literature, such as Ninikas, Athanasopoulos, et al. (2014),
Barceló, Grzybowska, et al. (2007) and Crainic, Gendreau, et al. (2009). Following this fashion, here we also propose a
prototype of real-time fleet management system in which the proposed pickup policy is embedded.
!w
X 1
ci ¼ (1)
j2V ;j6¼i
1 þ t ij
(A) (B)
(D)
(C)
A globally coupled network is shown in Figure 3(B). Here we set the parameter w as 2. So, the centrality for each
un-served request is calculated.
The results are 2 shown in the second row in Figure 3(C). Taking u1 as an example,
c1 ¼ 7:29 ¼ ð2:7Þ2 ¼ 12 þ 13 þ 12 þ 12 þ 13 þ 13 þ 15 .
Step 3: Computing for the travelling time t ai , which denotes the time for an idle courier a reaching an un-serviced
request i.
The travelling time for courier a to each un-served request is shown in the third row in Figure 3(C).
Step 4: Identifying the current waiting time for each un-served request, which is denoted as Ti (at most time the T i
cannot exceed T max ).
In the illustrative example, we assume that the current waiting time for each un-served request is not exceeded the
half of the maximum waiting time, so T i T max
2 .
( )
ci
; 0 \ T i T max
f ai ¼ 1þt ai Tmax 2 (2)
ci
1þt ai þ T maxT T
i
i
; 2 \ T i \ T max
The fitness of the eight un-served requests is shown in the fourth row of Figure 3(C).
Step 6: Dispatching the idle courier a to the request i that maximise the fitness of f ai .
In Figure 3(D), the idle courier a is dispatched to either u4 or u5, due to the fitness for these two requests are the high-
est among the eight un-served requests.
In the illustrative example, it is worth to highlight that the NN policy dispatches the courier a to serve request u8
due to the courier is the closest to the request u8, whereas the proposed pickup policy dispatches the courier a to request
u4 (or u5). Indeed, the current travelling time to u4 (or u5) is larger than the travelling time to u8; however, the courier
a can serve more quickly to the requests nearby the u4 (or u5), e.g. u1, u3 and u6, after it finishes serving the request u4
(or u5).
We label the proposed pickup policy as centrality-related policy. Steps 1 to 5 is repeated once any courier is idle,
and the whole process ends when all un-serviced requests have been addressed. Besides, as mentioned earlier, the
courier may go back to the local station once he finished serving the current request as the uncertainty of parcels and
customer service demands. Here we adopt the probability concept to model the uncertainty, which means that the
courier will go back to the local location at a certain probability once he finished serving the current request. The
probability of going back to the local station is denoted as Back Prob.
International Journal of Production Research 9
4. Computational experiments
In this section, we examine the performance of the proposed pickup policy for dispatching courier in various scenarios
by conducting computational experiments. The effect of this pickup policy is compared with two benchmark policies –
the NN and FCFS policies.
(Bertsimas and van Ryzin 1991; de Koster, Le-Anh, et al. 2004). However, although the NN policy enables to minimise
the waiting time for the current request, the waiting time for the next request may be prolonged owing to the short-
sightedness of this policy. Both policies are considered as the benchmark policies in this paper.
Figure 5. Customer requests arriving patterns. The red vertex denotes the local station.
International Journal of Production Research 11
To generate various test scenarios, five factors are considered: (1), the customer request distribution patterns; (2), the
number of couriers; (3), the maximum desired waiting time T max ; (4), the Back Prob and (5), the weight of centrality
w. We consider four patterns for request distribution which are shown in Figure 5(A)–(D): uniform, bipartite, centripetal
and anti-centripetal. The value in Figure 5 denotes the probability of request arriving at a corresponding vertex. Taking
the bipartite patterns, for example, the black vertex in the left gets an arriving request with probability 0.0225, so there
are about 90% requests will be arrive at the left part in the service area (0.0225 × 40 = 0.9), and only 10% request will
be arrive at the right part. The purpose of designing these four distribution patterns is to reflect various possible scenar-
ios in reality. The number of courier can be 1, 2, 3, 4 or 5; the maximum waiting time is set as 30, 60 or 120 min; the
Back Prob can be {0, 0.01, 0.02, 0.03, 0.05, 0.07, 0.09, 0.1, 0.15, 0.2, 0.25, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1}; and
the weight of centrality varies is set as {0.01, 0.02, 0.03, 0.06, 0.09, 0.1, 0.3, 0.6, 0.8, 1, 1.5, 3, 4.5, 6}. As a result, a
total of 15,960 (4 × 5 × 3 × 19 × 14) test scenarios are generated. The proposed pickup policy and the two benchmark
policies are applied for each test scenario. For each test scenario, we conducted 50 simulation experiments to obtain the
robustness of results.
Waiting time for NN policy Waiting time for the pickup policy
Average reduction of wating time ¼ (4)
Wating time for NN policy
Figure 6 shows the average reduction in waiting time for the centrality-based pickup policy over the NN policy. In each
subfigure, the horizontal axis denotes the probability of going back to the local station (Back Prob), and the vertical
axis presents the value for the average reduction in waiting time; the subfigures in each row reflect the condition of dif-
ferent customer request distribution patterns, and the subfigures in each column denote the number of couriers. As
shown in Figure 6, the centrality-based pickup policy dominantly outperforms the NN policy in almost all test scenarios.
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
T-max
100% 100% 100% 100%
=30
Courier=3
T-max
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
=60
T-max
50% 50% 50% 50% =120
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Figure 6. The average reduction of waiting time for the CP over the NN policy.
12 X. Cheng et al.
This result supports the potential of the centrality-based pickup policy in reducing the waiting time for customer request
under different conditions.
In detail, on one hand, focusing on the subfigures in each row (e.g. the first row), we can find that the performance
of centrality-based pickup policy in the situation where the customer request distribution is centripetally optimal than
other situation, such as uniform, bipartite and anti-centripetal. Taking the first row in Figure 6 as an example, when
Back Prob approaches 0, the reduction in waiting time is up to 90, 80, 70 and 50% for the condition of customer
request distribution is centripetal, uniform, bipartite and anti-centripetal, respectively. There is one probable reason.
When customer request distribution follows the centripetal patterns, there are many requests arrive at the central vertex,
so it is very easy to conduct the CP.
On the other hand, focusing on the subfigure in each column (e.g. the column of Uniform), when the number of
courier is small (e.g. courier = 1, 2, 3), the performance enhancement for the pickup policy is significant as Back Prob
approaches 0, which means the performance enhancement decreases as the value of Back Prob increases because the
increase in Back Prob reduces the benefit of dispatching to the central request (the courier always get back to the local
station). However, these trends do not exist when the number of courier is big (e.g. courier = 4, 5). Two probable expla-
nations are offered. First, there are so many couriers that the service area is not busy, and every courier can easily get
to the closest request site, which means that the NN policy is also suitable when many couriers are available. Second,
the weight of centrality denoted as w also influences the performance of the pickup policy: the value of w influences the
value of Cu, such that the performance of the pickup policy is significant when Cu is large, based on Equation (2), for
more detail discussion about the influence of w, please refer to supplementary information.3 Moreover, a threshold value
exists for Back Prob; the performance of the pickup policy is significantly changed when Back Prob exceeds this
threshold. For example, when courier = 1 and T max = 30 min, the threshold value is 0.4. Another feature is that the
number of courier influences this threshold value. For example, the threshold value is 0.4 when courier equals one and
0.9 when courier is 5. Besides, the maximum desired waiting time denoted as T max influences the performance of the
pickup policy. For example, when courier equals one, the reduction in waiting time is up to 70, 80, and 90% when
T max is 30, 60 and 120 min, respectively. This feature indicates that a high T max can amplify the contribution of
centrality measures.
Similarly, Figure 7 shows the average reduction in waiting time for the centrality-based pickup policy over the FCFS
policy, in which we can obtain same conclusions. In all of the test conditions, the centrality-based pickup policy is
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 T-max
=60
100% 100% 100% 100%
Courier=4
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Back-Prob Back-Prob Back-Prob Back-Prob
Figure 7. The average reduction in waiting time for the centrality-based pickup policy over the FCFS policy.
International Journal of Production Research 13
significantly optimal compared with the FCFS policy, and the performance enhancement is weakened with an increase
in the number of couriers; a threshold value exists for Back Prob, and this value changes along with the changing num-
ber of couriers. Besides, the performance enhancement under the scenario where customer request distribution follows
the centripetal pattern is optimal than other distribution patterns.
Standard deviation for NN policy Standard deviation for the pickup policy
Average reduction of variation ¼ (5)
Standard deviation for NN policy
Figure 8 shows the average reduction in standard deviation for the centrality-based pickup policy over NN policy. The
vertical axis denotes the value for average reduction in standard deviation for waiting time. The results show that the
centrality-based pickup policy can significantly reduce the variation under different test conditions. First, when the cus-
tomer request distribution is centripetal, the performance of the CP is optimal than other request distribution patterns,
such as uniform, bipartite and anti-centripetal, due to the average reduction in standard deviation being always bigger
than other scenarios. Second, the average reduction in standard deviation decreases when Back Prob increases under
the conditions when the number of courier is small (courier = 1, 2, 3), but this phenomenon is not observed when the
courier number is high. When many couriers service the same area, the load of that service area is so low that every
courier can easily be dispatched to the closest request site, thereby diminishing the contribution of CP. Third, a threshold
value for Back Prob also exists under each test condition. This threshold value is very close to the threshold value in
the situation of average reduction of waiting time in the former subsection (Section 4.2). Fourth, the reduction in stan-
dard deviation is improved as T max increases. For example, when T max = 30 min in subfigure 1, the standard deviation
reaches up to 70%; when T max = 120 min, the maximum of standard deviation is up to 80%.
Similarly, Figure 9 shows the average reduction in standard deviation for the centrality-based pickup policy over
FCFS policy; the same results have also been obtained. For example, when the system is busy and courier = 1, 2 or 3,
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
T-max
100% 100% 100% 100% =30
Courier=3
T-max
0% 0% 0% 0% =60
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
T-max
50% 50% 50% 50% =120
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Back-Prob Back-Prob Back-Prob Back-Prob
Figure 8. The average reduction in standard deviation for the centrality-based pickup policy over NN policy.
14 X. Cheng et al.
Uniform Bipartite Centripetal Anti-centripetal
100% 100% 100% 100%
Courier=1
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 T-max
=30
100% 100% 100% 100%
Courier=3
T-max
50% 50% 50% 50% =120
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Back-Prob Back-Prob Back-Prob Back-Prob
Figure 9. The average reduction in standard deviation for the centrality-based pickup policy over FCFS policy.
the performance of reduction in standard deviation diminishes as the Back Prob increases. In addition, when the
customer request arrives at more central place, the performance of the CP is optimal. Besides, the value of T max can
magnify the performance of the pickup policy.
Reduction in total time NN policy Reduction in total time for the Pickup policy
Reduction in total time for NN policy ¼
Reduction in total time for NN policy
(6)
The performance of reduction in total time for the centrality-based pickup policy compared with the NN policy is shown
in Figure 10. When the number of courier is small (e.g. courier = 1, 2), which means that the service area is very busy,
the performance of reduction in total time is significant. However, the performance is not significant when the number
of courier is high. The reduction can be up to 30% when courier = 1 but can be only up to 4% when courier = 4, partic-
ularly, there are almost no reduction in the total time when the courier equals 5. In addition, a threshold value was
observed for Back Prob; the feature of this threshold value is mostly similar with that in Subsection of 4.2. For a small
number of couriers, this value is changed with the variation. Furthermore, the centrality-based pickup policy is best
when the customer requests distribution is centripetal. For example, under the scenario when the number of courier
equals one and the Back Prob approaches zero, the performance of reduction in total time of the centripetal scenario is
up to 20%, but performance for the other three scenarios are all below 20%. Besides, the value of T max also influences
the performance of reduction in total time. For example, up to 30% reduction can be achieved when T max = 120 min
but only up to 18% when T max = 30 min, as shown in Subfigure 1.
Figure 11 shows the performance of reduction in total time for the centrality-based pickup policy over the FCFS pol-
icy. The results indicates that the centrality-based pickup policy is optimal than the FCFS policy in almost all test sce-
narios. Similar results can be obtained. Such as, the performance of reduction is significant when the number of courier
International Journal of Production Research 15
Uniform Bipartite Centripetal Anti-centripetal
40% 40% 40% 40%
Courier=1
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 T-max
=30
20% 20% 20% 20%
Courier=3
T-max
0% 0% 0% 0% =60
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
6% 6% 6% 6%
Courier=4
4% 4% 4% 4%
T-max
2% 2% 2% 2% =120
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Back-Prob Back-Prob Back-Prob Back-Prob
Figure 10. The reduction in total time for the centrality-based pickup policy over the NN policy.
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 T-max
=30
40% 40% 40% 40%
Courier=3
T-max
10% 10% 10% 10% =120
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
3% 3% 3% 3%
Courier=5
2% 2% 2% 2%
1% 1% 1% 1%
0% 0% 0% 0%
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Back-Prob Back-Prob Back-Prob Back-Prob
Figure 11. The reduction in total time for the centrality-based pickup policy over the FCFS policy.
is small. For example, when courier equals 1, the reduction can up to 65%, but when courier equals 5, the reduction is
only up to 2%. In addition, a threshold value also exists and the performance of the pickup policy is also influenced by
the value of T max .
According to these observations from subsection 4.3 to 4.5, we can find that the reduction in both waiting time
(mean and standard deviation) and total time over the NN and FCFS policy reflects the effectiveness of the proposed
pickup policy. First, the performance of the pickup policy is influenced by the value of Back Prob. The Back Prob
16 X. Cheng et al.
indicates the probability of going back to local station, so under the situation where this probability is small, the courier
can effectively serve the neighbouring requests when he is dispatched to the central requests, which amplifying the effi-
cacy of the centrality-based pickup policy. So the performance of the proposed policy is significant when the probability
is small. Second, the customer request distribution pattern also influences the performance of the CP. The experiments
result shows that this policy is more suitable for the situation where the customer request distribution follows the cen-
tripetal pattern. Indeed, when customer request arrives at more central among the local station, it is very easy to conduct
the CP. This conclusion also indicates the importance of the selection of local station for the express courier company.
However, the detail analysis of station/location selection exceeds the scope of this paper; one can refer to the literature
(Kahraman, Ruan, et al. 2003) and (Silva and de la Figuera 2007). Third, the performance of the centrality-based pickup
policy is also influenced by the numbers of couriers and the T max . The number of couriers reflects the busyness of the
service area, if this number is big, the courier can effectively be dispatched to the request location based on NN or
FCFS policy, as the request always closes to the courier. So the performance of CP is not significant when number of
couriers is big. Moreover, the T max also can amplify the efficacy of the pickup policy.
based on big data analysis for historical information as the computer-based system can store millions of historical
customer requests. As the development of sophisticated methods and intelligent algorithm, the expected future dynamic
events can be predicted. For example, the future expected customer requests can be generated by analysing past request
data. Overall, this kind of research is significant and meaningful for the development of express industry.
Disclosure statement
No potential conflict of interest was reported by the authors.
Funding
This work was supported by the The development grants from Shenzhen Science, Technology and Innovation Commission [grant
number JCYJ20140630144136828]; The Guangdong Provincial Technology Scheme [grant number 2015A020217007]; National
Science-technology Support Plan [project number 2015BAK18B02]; The development grants from Shenzhen Science, Technology
and Innovation Commission [grant number ZDSYS20140509155229805].
Notes
1. Another kind of courier service is intra-city express courier service which only contains the pickup process and delivery process
due to the short distance between origin and destination. The courier routing for intra-city service is the classic pickup and
delivery problem which has been well studied, such as D’Souza, Omkar, et al. (2012), Masson, Ropke, et al. (2014), Nanry and
Barnes (2000) and Ropke and Pisinger (2006).
2. The idea of the Equation (1) is from the centrality measures, particularly, the degree centrality. We consider the reciprocal of t ij
as the weight of edges. However, in order to deal with the situation where t ij equals zero, we add 1 in the denominator.
3. As the space restriction of the format of this paper, we investigate the influence of w on the performance of the centrality-based
pickup policy in the Supplementary Information.
References
Alanis, R., A. Ingolfsson, and B. Kolfal. 2013. “A Markov Chain Model for an EMS System with Repositioning.” Production and
Operations Management 22 (1): 216–231.
Andersson, T., and P. Värbrand. 2007. “Decision Support Tools for Ambulance Dispatch and Relocation.” Journal of the Operational
Research Society 58 (2): 195–201.
Barceló, J., H. Grzybowska, and S. Pardo. 2007. “Vehicle Routing and Scheduling Models, Simulation and City Logistics.” In
Dynamic Fleet Management, 163–195. Springer.
Bard, J. F., and A. I. Jarrah. 2009. “Large-Scale Constrained Clustering for Rationalizing Pickup and Delivery Operations.” Trans-
portation Research Part B: Methodological 43 (5): 542–561.
Batallas, D. A., and A. A. Yassine. 2006. “Information Leaders in Product Development Organizational Networks: Social Network
Analysis of the Design Structure Matrix.” IEEE Transactions on Engineering Management 53 (4): 570–582.
Bendle, L. J., and I. Patterson. 2010. “The Centrality of Service Organizations and Their Leisure Networks.” The Service Industries
Journal 30 (10): 1607–1619.
Bertsimas, D. J., and G. van Ryzin. 1991. “A Stochastic and Dynamic Vehicle Routing Problem in the Euclidean Plane.” Operations
Research 39 (4): 601–615.
Bezuidenhout, C. N., S. Bodhanya, T. Sanjika, M. Sibomana, and G. L. N. Boote. 2012. “Network-analysis Approaches to Deal with
Causal Complexity in a Supply Network.” International Journal of Production Research 50 (7): 1840–1849.
Bock, S. 2010. “Real-time Control of Freight Forwarder Transportation Networks by Integrating Multimodal Transport Chains.”
European Journal of Operational Research 200 (3): 733–746.
Chang, T.-S., and H.-M. Yen. 2012. “City-courier Routing and Scheduling Problems.” European Journal of Operational Research
223 (2): 489–498.
Chen, Y. 1999. “Banking Panics: The Role of the First-come, First-served Rule and Information Externalities.” Journal of Political
Economy 107 (5): 946–968.
Coelho, L. C., and G. Laporte. 2013. “A Branch-and-cut Algorithm for the Multi-product Multi-vehicle Inventory-routing Problem.”
International Journal of Production Research 51 (23–24): 7156–7169.
Cortés, C. E., A. Núñez, and D. Sáez. 2008. “Hybrid Adaptive Predictive Control for a Dynamic Pickup and Delivery Problem
including Traffic Congestion.” International Journal of Adaptive Control and Signal Processing 22 (2): 103–123.
Crainic, T. G., M. Gendreau, and J.-Y. Potvin. 2009. “Intelligent Freight-transportation Systems: Assessment and the Contribution of
Operations Research.” Transportation Research Part C: Emerging Technologies 17 (6): 541–557.
Crucitti, P., V. Latora, and S. Porta. 2006. “Centrality Measures in Spatial Networks of Urban Streets.” Physical Review E 73 (3):
036125.
18 X. Cheng et al.
D’Souza, C., S. N. Omkar, and J. Senthilnath. 2012. “Pickup and Delivery Problem Using Metaheuristics Techniques.” Expert
Systems with Applications 39 (1): 328–334.
Dantzig, G. B., and J. H. Ramser. 1959. “The Truck Dispatching Problem.” Management Science 6 (1): 80–91.
Dean, S. F. 2008. “Why the Closest Ambulance Cannot Be Dispatched in an Urban Emergency Medical Services System.” Prehospi-
tal and Disaster Medicine 23 (2): 161–165.
Durugbo, C., and J. C. K. H. Riedel. 2013. “Readiness Assessment of Collaborative Networked Organisations for Integrated Product
and Service Delivery.” International Journal of Production Research 51 (2): 598–613.
Egbelu, P. J., and J. M. Tanchoco. 1984. “Characterization of Automatic Guided Vehicle Dispatching Rules.” International Journal of
Production Research 22 (3): 359–374.
Ferrucci, F., and S. Bock. 2014. “Real-time Control of Express Pickup and Delivery Processes in a Dynamic Environment.” Trans-
portation Research Part B: Methodological 63: 1–14.
Forecasting, O. E. 2005. The Impact of the Express Delivery Industry on the Global Economy. St Aldates: Abbey House.
Freeman, L. C., S. P. Borgatti, and D. R. White. 1991. “Centrality in Valued Graphs: A Measure of Betweenness Based on Network
Flow.” Social Networks 13 (2): 141–154.
Friedkin, N. E. 1991. “Theoretical Foundations for Centrality Measures.” American Journal of Sociology 96 (6): 1478–1504.
Gerkey, B. P., and M. J. Mataric. 2002. “Sold!: Auction Methods for Multirobot Coordination.” IEEE Transactions on Robotics and
Automation 18 (5): 758–768.
Ghiani, G., E. Manni, and C. Triki. 2009. “Anticipatory Algorithms for Same-Day Courier Dispatching.” Transportation Research
Part E: Logistics and Transportation Review 45 (1): 96–106.
Giaglis, G. M., I. Minis, A. Tatarakis, and V. Zeimpekis. 2004. “Minimizing Logistics Risk through Real-time Vehicle Routing and
Mobile Technologies.” International Journal of Physical Distribution & Logistics Management 34 (9): 749–764.
Gloor, P. A., M. Paasivaara, D. Schoder, and P. Willems. 2008. “Finding Collaborative Innovation Networks through Correlating
Performance with Social Network Structure.” International Journal of Production Research 46 (5): 1357–1371.
Gómez, D., J. R. Figueira, and A. Eusébio. 2013. “Modeling Centrality Measures in Social Network Analysis Using Bi-criteria
Network Flow Optimization Problems.” European Journal of Operational Research 226 (2): 354–365.
Hu, Z.-H., and Z.-H. Sheng. 2014. “A Decision Support System for Public Logistics Information Service Management and Optimiza-
tion.” Decision Support Systems 59: 219–229.
Hu, D. N., J. L. Zhao, Z. M. Hua, and M. C. S. Wong. 2012. “Network-based Modeling and Analysis of Systemic Risk in Banking
Systems.” MIS Quarterly 36 (4): 1269–1291.
Kahraman, C., D. Ruan, and I. Doǧan. 2003. “Fuzzy Group Decision-making for Facility Location Selection.” Information Sciences
157: 135–153.
Kiss, C., and M. Bichler. 2008. “Identification of Influencers – Measuring Influence in Customer Networks.” Decision Support
Systems 46 (1): 233–253.
de Koster, R., T. Le-Anh, and J. R. van der Meer. 2004. “Testing and Classifying Vehicle Dispatching Rules in Three Real-world
Settings.” Journal of Operations Management 22 (4): 369–386.
Kuo, Y. 2013. “Optimizing Truck Sequencing and Truck Dock Assignment in a Cross Docking System.” Expert Systems with Appli-
cations 40 (14): 5532–5541.
Larsen, A., O. B. Madsen, and M. M. Solomon. 2008. “Recent Developments in Dynamic Vehicle Routing Systems.” In The Vehicle
Routing Problem: Latest Advances and New Challenges, 199–218. Springer.
Lee, S. 2012a. “The Role of Centrality in Ambulance Dispatching.” Decision Support Systems 54 (1): 282–291.
Lee, S. 2012b. “Centrality-based Ambulance Dispatching for Demanding Emergency Situations.” Journal of the Operational Research
Society 64 (4): 611–618.
Li, J.-Q., P. B. Mirchandani, and D. Borenstein. 2009. “Real-time Vehicle Rerouting Problems with Time Windows.” European
Journal of Operational Research 194 (3): 711–727.
Lin, J. R., S. Yan, and C. W. Lai. 2013. “International Express Courier Routing and Scheduling under Uncertain Demands.”
Engineering Optimization 45 (7): 881–897.
Lin, C., K. L. Choy, G. T. S. Ho, H. Y. Lam, G. K. H. Pang, and K. S. Chin. 2014. “A Decision Support System for Optimizing
Dynamic Courier Routing Operations.” Expert Systems with Applications 41 (15): 6917–6933.
Mantel, R. J., and H. R. Landeweerd. 1995. “Design and Operational Control of an AGV System.” International Journal of Produc-
tion Economics 41 (1–3): 257–266.
Masson, R., S. Ropke, F. Lehuédé, and O. Péton. 2014. “A Branch-and-cut-and-price Approach for the Pickup and Delivery Problem
with Shuttle Routes.” European Journal of Operational Research 236 (3): 849–862.
Mavrovouniotis, M., and S. Yang. 2015. “Ant Algorithms with Immigrants Schemes for the Dynamic Vehicle Routing Problem.”
Information Sciences 294: 456–477.
Mezgár, I., and U. Rauschecker. 2014. “The Challenge of Networked Enterprises for Cloud Computing Interoperability.” Computers
in Industry 65 (4): 657–674.
Nanry, W. P., and J. W. Barnes. 2000. “Solving the Pickup and Delivery Problem with Time Windows Using Reactive Tabu Search.”
Transportation Research Part B: Methodological 34 (2): 107–121.
International Journal of Production Research 19
Ninikas, G., T. Athanasopoulos, V. Zeimpekis, and I. Minis. 2014. “Integrated Planning in Hybrid Courier Operations.” The
International Journal of Logistics Management 25 (3): 611–634.
Pillac, V., C. Guéret, and A. L. Medaglia. 2012. “An Event-driven Optimization Framework for Dynamic Vehicle Routing.” Decision
Support Systems 54 (1): 414–423.
Pillac, V., M. Gendreau, C. Guéret, and A. L. Medaglia. 2013. “A Review of Dynamic Vehicle Routing Problems.” European Jour-
nal of Operational Research 225 (1): 1–11.
Ravindran, K., A. Susarla, D. Mani, and V. Gurbaxani. 2015. “Social Capital and Contract Duration in Buyer-supplier Networks for
Information Technology Outsourcing.” Information Systems Research 26 (2): 379–397.
Ropke, S., and D. Pisinger. 2006. “An Adaptive Large Neighborhood Search Heuristic for the Pickup and Delivery Problem with
Time Windows.” Transportation Science 40 (4): 455–472.
Ruhnau, B. 2000. “Eigenvector-Centrality – A Node-centrality?” Social Networks 22 (4): 357–365.
Silva, F. J. F., and D. S. de la Figuera. 2007. “A Capacitated Facility Location Problem with Constrained Backlogging Probabilities.”
International Journal of Production Research 45 (21): 5117–5134.
Sosa, M. E., S. D. Eppinger, and C. M. Rowles. 2007. “A Network Approach to Define Modularity of Components in Complex
Products.” Journal of Mechanical Design 129: 1118.
Sosa, M., J. Mihm, and T. Browning. 2011. “Degree Distribution and Quality in Complex Engineered Systems.” Journal of Mechani-
cal Design 133: 101008.
Sungur, I., Y. Ren, F. Ordóñez, M. Dessouky, and H. Zhong. 2010. “A Model and Algorithm for the Courier Delivery Problem with
Uncertainty.” Transportation Science 44 (2): 193–205.
Tsai, C.-F., C.-W. Tsai, and C.-C. Tseng. 2003. “A New and Efficient Ant-based Heuristic Method for Solving the Traveling Salesman
Problem.” Expert Systems 20 (4): 179–186.
Urrutiaguer, D. 2004. “Programme Innovations and Networks of French Public Theatres.” The Service Industries Journal 24 (1):
37–55.
Chaiken, J. M., and R. C. Larson. (1972). “Methods for Allocating Urban Emergency Units: A Survey.” Management Science
19 (4-part-2): P110–P130.
Durugbo, C. (2013). “Modelling Information for Collaborative Networks.” Production Planning & Control 26 (1): 34–52.
Lee, S. 2014. “Role of Parallelism in Ambulance Dispatching.” IEEE Transactions on Systems, Man, and Cybernetics: Systems PP
(99): 1–1.
Leem, B., and H. Chun. 2014. “An Impact of Online Recommendation Network on Demand.” Expert Systems with Applications
41 (4, Part 2): 1723–1729.
Ostergaard, E. H., M. J. Mataric, M. J. Mataric, and G. Sukhatme. (2001). “Distributed Multi-robot Task Allocation for Emergency
Handling.” 2001 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE.
Yan, S., J.-R. Lin, and C.-W. Lai. 2013. “The Planning and Real-time Adjustment of Courier Routing and Scheduling under
Stochastic Travel times and Demands.” Transportation Research Part E: Logistics and Transportation Review 53: 34–48.
Yousefikhoshbakht, M., F. Didehvar, and F. Rahmati. 2014. “Solving the Heterogeneous Fixed Fleet Open Vehicle Routing Problem
by a Combined Metaheuristic Algorithm.” International Journal of Production Research 52 (9): 2565–2575.