This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSC.2020.2964244, IEEE
Transactions on Services Computing

SUBMIT TO IEEE TRANSACTIONS ON SERVICES COMPUTING 1

IoT-enabled service for crude-oil production

systems against unpredictable disturbance
Qianqian Duan, Daniel Sun Member, IEEE, Guoqiang Li Member, IEEE, Genke Yang, WeiWu Yan

Abstract—Internet of Things (IoT) has become a new paradigm In the IoT context, service revolution in the petrochemical
of communication to reform traditional industries, in which dis- industries lead to more uncertainties in scheduling manage-
tributed data automatically collected via IoT in a low cost enables ments. These unexpected uncertainty disruptions make crude-
many new IT services that were even impossible decades ago.
This research reports on an IoT-enabled production management oil scheduling plans infeasible and disturb the initial schedul-
service for crude-oil industry. In practice, even if an optimal ing plans, thereby reducing the production effectiveness and
management decision is achieved, disruptions, such as possible profits of petrochemical enterprises. If a lack in timely, accu-
oil-device failures, inclement weathers and other disturbances, rate, and consistent processing information, the disturbance in
arise frequently and then weaken efficiency and stability of the scheduling process will be aggravated. This results in the
supplement. With the help of IoT, a near-real-time management
service comes into being, although the adoption of IoT brings new essential need for decision-makers to upgrade their disruption
challenges to management of disruptions. The contributions of management with advanced technologies to improve the effi-
this paper are as follows: First, a service framework is proposed ciency and quality. Therefore, how to effectively manage the
for refinery which combines MQTT and Azure cloud, enabling disruption receives attention.
reliable data/command delivery. Secondly, a smart disruption Based on the IoT technologies, real-time data streams from
management service system is developed, which consists of
monitor and alarm module, disruption management module, sensor network challenge traditional disturbance management
and rescheduling procedure module. The rescheduling procedure methods. The smart crude-oil disruption management system
module takes into account the network of the refinery operations, tends to provide the users services rather than set of function
and is easy to accommodate changes in the refinery configuration and equipment system. Smart refinery industry needs one uni-
for unforeseen disruptions. The experimental results show that form standard framework which is able to integrate different
our solution has better efficiency and stability compared to
traditional ones. application services. Therefore, an application platform for
refinery industry on the basis of cloud computer need to be
Index Terms—Disruption Management, Service Framework, built, and all kinds of application can run on it, so as to provide
Reschedule, IoT
better service for factory. These services can be represented
as security service, such as disruption management service,
disruption alarm service and equip monitoring service, etc.
I. I NTRODUCTION In this paper, a smart scheduling service framework for
With the rapid development of modern information tech- petrochemical industries is proposed to satisfy the individual
nologies, Internet of Things (IoT), as a new technology, has service, where several refineries share one service framework,
gained widely attention in the petrochemical industries. Based and select the service based on their own needs. Based on the
on IOT, real-time data can flow from fields to decision-making service framework, a disturbance management service system
center seamlessly , which provides better decision-makings, is developed. The initial schedule, real-time streaming data,
optimization processes and safer working environments. In and details of the disruption are the inputs to the service
an industry that spends trillions of dollars every year on system. The impact of the disruption on the original schedule
exploration, development and maintenance of assets, even a is determined based on the disruption management modules.
single-digit percentage improvement through IoT will have For the rescheduling procedure module in the service system,
a huge impact. IoT can improve operation processes for the rescheduling procedure based on the network of the
companies, especially scheduling and planning processes. In refinery operations is proposed to improve the computation
addition, the integration of IoT with operations technologies efficiency, which is able to guarantee flexibility, accommodate
generates new business forms, resulting in service revolution changes in the refinery configuration during different types of
and innovation in the petrochemical industries. unforeseen disruptions.
The contribution of the research is summarized as follows:
• A smart scheduling service framework for petrochemical
Q. Duan is with Shanghai University of Engineering Science, Shanghai,
China. Email: [email protected] industry is proposed to provide better service for factory.
Daniel Sun is with Data61, CSIRO, ACT, Australia. Email: • A disturbance management service system is developed
[email protected]
Guoqiang Li, is with School of Software, Shanghai Jiao Tong University, to analyze the equipment data and manage the disruption.
Shanghai, China. Email: [email protected]. • The rescheduling procedure module is proposed to im-
Genke Yang is with Shanghai Jiao Tong University, Shanghai, China.. prove the computation efficiency.
Email: [email protected].
Weiwu Yan is with Shanghai Jiao Tong University, Shanghai, China.. Email: The rest of this paper is organized as follows. Section
[email protected]. II reviews the existing work related to application of IOT

1939-1374 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSC.2020.2964244, IEEE
Transactions on Services Computing

SUBMIT TO IEEE TRANSACTIONS ON SERVICES COMPUTING 2

and disruption management on crude-oil system. Section III which has three basic features of IOTs (i.e, overall perception,
proposes a smart scheduling service framework for petro- reliable transmission and intelligent processing), and satisfies
chemical industry. In Section IV, the disruption management individual demands of users. As shown in Figure.1, the bottom
service system is presented. Sections V , VI and VII discuss layer is the sensing layer, where the IoT equipment (wireless
the disruption management modules in the service system sensors, wireless actuators, etc.) and non-intelligent appliances
individually. Section VIII explains the computational tests that are connected to form the basic sensing network, communicat-
demonstrate the availability of the disruption-management for ing with external devices. The middle layer, the transmission
the crude-oil problem, and Section IX concludes the paper. layer, offers the reliable massage transmission mechanism.
The top layer is the cloud service layer, where the user can
II. R ELATED W ORK control the real-time and historical scheduling information by
A. IoT and Its Application in crude-oil systems accessing the platform through web, etc..
IoT technologies will provide real-time data and services Sensing layer: The sensing layer involves data acquisition
of systems. Key technologies (e.g., wireless networks) allow and collaboration between device and gateway(s). Each device
networking among data from/to automation devices, operat- is equipped with different types of sensors, like pressure,
ing supplies, and equipment status[1], [2], [3], [4], [5], [6]. temperature, acoustic and flow sensors to detect the disruption.
According to the researches [7], for the decentralized opti- In order to analyze and process the data completely, the data
mizers, the real-time cooperation like cloud computing with gathered must be delivered to the corresponding server.
strong big-data analysis, information sharing and visualization transmission layer: There are three ways of network trans-
[8] will help to achieve decision making intelligence. By mission in the transmission layer, i.e., wireless network, wired
assembling and integrating the real-time data from crude-oil network, and mobile communication network, providing effec-
systems, the device can be monitored effectively. And, refinery tive message transmission. The middleware technology ensure
scheduling and planning can be used as an illustration [9]. the flexibility and reliable transmission requirements. Wireless
The radio communications are applied to transfer the real-time sensor network transmission protocol MQTT is able to meet
information and datas sensed from the sensors for pipeline the timeliness requirement. MQTT broker will be built to
inspection [10]. In [11], sensors are placed along the pipelines ensure reliable information transmission.
to monitor pipelines. cloud service layer: After data collected form sensing layer
and transporting layer, we should use data reasonably. Not only
the acquired message has to be analyzed and integrated, but
B. Disruption management
also the availability of the message has to be improved further.
Recently, there has been an increase in the number of Windows Azure is used as the basic cloud technology for cloud
studies conducted on managing disruptions in airline opera- data management. The accessing method, Web, is provided
tions [12], [13], supply chain [14], [15], and shop scheduling for interaction layer. The Cloud Service Center for Multiple-
[19], [20], [21]. Unexpected disruptions occur during the refinery Management is built, providing better interface and
crude-oil scheduling process such as crude-oil device failures simple operation for users.
and inclement weather [16], [17], [18].The most common As shown in Figure.1, information flow includes two direc-
disruptions in the scheduling process of a refinery include tions, one is data transmission from perception layer to cloud
emergency crude-oil procurement and vessel delay in the service layer and the other is commands transmission from
supply chain. Mishra [16] developed an agent-based support top to bottom. The active and passive control is implemented
system to manage disruptions. Under abnormal supply chain through the overall framework. That is to say, the system can
events, the rescheduling method proposed by Adhitya [22] realize intelligent function of electrical equipment by receiving
used an iterative algorithm to generate a new feasible schedule. passive command control from users and taking initiative
However, the heuristic-based method only works for specific optimization manage.
types of disruptions, and lacks generality; moreover, it is
difficult to modify or extend the method. Nevertheless, to
overcome these shortcomings, a model-based rescheduling IV. S MART D ISRUPTION M ANAGEMENT S ERVICE S YSTEM
method was proposed [23]. The general rescheduling problem A smart disturbance management service system is devel-
has received attention. Airline rescheduling problem [24] that oped, running on Azure cloud. Real-time scheduling cost and
involves many considerations is difficult to describe mathemat- device status information can be easily obtained by users.
ically. The rescheduling of flight legs employs the heuristics Users can also control devices remotely. The disturbance
method [25], [26]. Malucelli [27] proposed a recovery strategy management service system will analyz the equipment data
to generate a final plan, interacting the information obtained delivered by the devices to alarm the disruption, and manage
from the IoT within the airlines schedule. the disruption for maximizing the production profit and mini-
mizing disruptions, thus to better control and minimize risks.
III. T HE I NTELLIGENT S CHEDULING S ERVICE Figure.2 schematically outlines the framework for the pro-
F RAMEWORK posed the disruption management service system. The initial
In this section, a scheduling service framework for petro- schedule, real time streaming data, and details of the disruption
chemical industries is proposed, which provides the hierar- are the inputs to the service system. The impact of the
chical integration and management for users based on IoT, disruption on the original schedule is determined based on the

1939-1374 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSC.2020.2964244, IEEE
Transactions on Services Computing

SUBMIT TO IEEE TRANSACTIONS ON SERVICES COMPUTING 3

SAAS
Application Fronted 1
Refinery
Application Fronted 2
Industry
Application Fronted 3
Custom
....... Application Fronted n Web
command date

ESB/API-Hub
Windows Azure
Service integration
SAAS Service Service 1 Service 2 ....... Service n
management Disruption Alarm Disruption Management

command date

WISE-PasS Cellular
WLAN Internet network MQTT based Infomation Middleware

command
date

Embedded System WISE IoT Automation Controller

IOT Gateway
Gateway Web Access

date date

IOT
Devices & Scheduling
Sensors Device Date Date
IoT
Devices/Sensors/Actuators WISE Devices ADAM Devices

Fig. 1. IoT based architecture

disruption management model module and a new schedule is ment model, the system is rescheduled to cope with these
generated. The steps for disruption management are carried new conditions. In this section, the corresponding effective
out by the following components : rescheduling procedure is proposed to increase the quality of
1. IoT-enabled real time data. the solution. Additional details will be presented in Section
Real-time data from equipment is readily available when VII.
the crude-oil scheduling system includes seamless integration
with installed IoT devices. The information provided is instan- V. T HE KPI M ONITOR AND A LARM M ODULE
taneous regarding scheduling process dynamics and equipment A. IoT-Enabled Real Time Data
performance.
2. Key Performance Indicators (KPI) monitors and alarms With the integration of IoT devices, the crude-oil scheduling
module. system is able to capture a wide variety of required information
and data from person and equipment. This includes but is not
The impact of KPIs and their measurements on the crude-oil
limited to, vessel arrival date, flow rates, personnel location
scheduling system is essential in managing disruptions. They
and routes, inventory profiles, operation schedules, operational
are monitored by comparing values against set predetermined
details, sales information etc. Information related to each and
limits. If a sustained deviation is detected, an alarm is gener-
every event among the entities in the scheduling process is
ated. Additional details will be presented in Section V.
transferred to disruption management system.
3. The initial schedule.
The initial schedule contains the vessel arrival data, the
scheduling targets, and the flow rates based on the scheduling B. Key Performance Indicators (KPI) Monitors and Alarms
targets. KPI refers to the function of several activities in the crude
4. The disruption management model module. oil scheduling process. Flow rate, inventory levels, arrive date
The disruption-management system may suffer from the etc. are examples of KPIs. In order to assess the performance
trade-off between the stability and the quality of the schedule. of the process, KPI must be continuously monitored. For
In this research, a disruption-management model is proposed instance, it is able to monitor the flow rate through the storage,
that uses the performance measures considering both the charging tank, and so on. As the disturbances (deviations of
efficiency and stability. The disruption management model is the defined limit values) on the relevant KPI are discovered,
modified in order to control the disruption process effectively. an alarm is generated.
Based on the refinery configuration, the model is transformed In order to manage the disruptions in the crude oil schedul-
into a set of network flow models. Additional details will be ing process, the KPIs shall be essentially measured, and its
presented in Section VI. influence on the system will be identified. Two types of KPI
5. Rescheduling procedure module. , namely material flow (MF) and information flow (IF), are
Given an initial schedule, a disruption event and manage- selected for the present experimental study. MF is calculated

1939-1374 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSC.2020.2964244, IEEE
Transactions on Services Computing

SUBMIT TO IEEE TRANSACTIONS ON SERVICES COMPUTING 4

Vessel Storage tank Charging tank CDU

Pump Pump

Date
node
Real Time data

Product Product Crude Crude Product Crude Crude

Inventory High Inventory Low Inventory High Inventory Low Inventory Low Inventory High Inventory Low

KPI KPI KPI

KPI KPI KPI KPI
monitor monitor monitor
monitor monitor monitor monitor

Alarm

Disruption Rescheduling
Management procedure
Model

New Schedule
Corrective Actives

Fig. 2. The disruption-management framework

1939-1374 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSC.2020.2964244, IEEE
Transactions on Services Computing

SUBMIT TO IEEE TRANSACTIONS ON SERVICES COMPUTING 5

based on the material flow values, which can be detected between the initial and new schedules, while limiting the
through IoT devices. IF is calculated based on the corre- number of changes to the plan as much as possible. The
sponding information flow values. Once the disruptions occur following method is employed:
during the scheduling time, its influence on the system will be For time t:
identified. In this paper, the MˆF k and IFˆ k refer respectively The configuration of the schedule:
to the abnormality of MF and IF. They are calculated as
following: SCH = {SCH1 , . . . SCHt . . . , SCHT }
Abnormality of material flow: where,SCHt is to denote which type of operation is ex-
MˆF k = 1 if M Fk > F MkU or M Fk < M FkL else MˆF k = ecuted at time t. There are three types of SCH: crude oil
0 unloading, transportation, and feeding, denoted by SCHt,v
Abnormality of information flow: ,SCHt,s and SCHt,c respectively;
ˆ k = 1 if IFk 6= IF Initial or IFk = IF Initial else
IF Define:
k k
ˆ
IF k = 0 SCH Initial : the configuration of the initial schedule
When MˆF k = 1 or IF ˆ k = 1, a disruption is detected and SCH N ew : the configuration of the new schedule
an alarm is generated.
T VX
X type
VI. T HE D ISRUPTION M ANAGEMENT MODEL M ODULE Stability : Cd = C CON F |SCHt,v
Initial N ew
−SCHt,v |
We assume that the operations of crude-oil problem have t=1 v=1
(2)
been scheduled well beforehand and that they are now dis-
rupted. Once the disruptions occur, the initial schedule may T Stype
X X
be infeasible. The disruption management is a real-time pro- + C CON F |SCHt,s
Initial N ew
− SCHt,s |
cedure, which can obtain the new schedule quickly and reduce t=1 s=1
the negative impact of the disruption on the system. T Ctype
X X
To control the disruption process effectively, the disruption + C CON F |SCHt,c
Initial N ew
− SCHt,c |
model is modeled as a set of network-flow models, which t=1 c=1
consist of arcs and nodes that represent possible cause and
3) Disruption Management Objective: The disruption man-
effect relationships among the crude-oil entities. Moreover,
agement objective is utilized as the performance measure to
they guarantee the flexibility to accommodate changes in the
construct the schedules. In particular, the objective function
refinery configuration during unforeseen types of disruptions.
incorporates efficiency and stability, which are expressed by
the operating cost and the sequence deviation respectively.
A. Rescheduling performance measures Eqs. (1) and (2) are combined to form the objective function
In this section, two types of performance measures of the expressed as Eq. (3). The following is the disruption manage-
disruption-management system are considered. ment objective that contains both the efficiency and stability
1) Evaluation of efficiency: The performance of efficiency measures, which allows balancing between the two measures.
refers to the optimality of the schedule with respect to the
operating cost. The lower the operating cost, the higher the C = α · Ccost + (1 − α) · Cd (3)
efficiency. The operating cost is one of the significant factors
in the scheduling process. Hence, it is considered in the B. The scheduling operation
disruption-management process; the operating cost needs to The operation decision is defined as Dect = (SCHt , Ft ),
be reduced as much as possible. where SCHt denotes the type of operation that is executed,
and Ft is the corresponding flow rate. The operation decision
Ef f iciency: Ccost = C1 + C2 + C3 + C4 + C5 (1) SCHt contains the source and start time; these variables are
PN V discrete scheduling variables. However, Ft is a continuous
C1 = ν=1 CνU N LOAD (TνL − TνF ) scheduling variable.
PN V
C2 = ν=1 CνSEA (TνF − TνARR ) Given the states of the system, the crude-oil problem is to
PN S PT V S +V S determine a series of Dect :
C3 = i=1 CiIN V ST t=1 ( i,t 2 i,t−1 )
PN B PT V B +V B Decision = {Dec1 , . . . , Dect , . . . , DecT }
C4 = j=1 CjIN V BL t=1 ( j,t 2 j,t−1 )
PT PN B PN B SET U P PN CDU
C5 = t=1 j=1 j 0 =1 Cj,j 0 l=1 Zj,j 0 ,l,t Where
2) Evaluation of stability: Disruptions occurring lead to Dect = {SCHt , Ft } =
frequent changes in the initial schedule, which causes instabil- {SCHt,v , SCHt,s , SCHt,c , Ft,v , Ft,s , Ft,c };
ity with the scheduling process. This will also incur additional discrete state SCHt = {SCHt,v , SCHt,s , SCHt,c };
costs in re-sequencing of the operations. The configuration of continuous state Ft = {Ft,v , Ft,s , Ft,c };
the schedule is defined as the types of executed operations discrete schedule SCH =
according to the set of discrete scheduling variables. Stable {SCH1 , . . . , SCHt , . . . , SCHT };
performance refers to the complete configuration deviation continuous schedule F = {F1 , . . . , Ft , . . . , FT }.

1939-1374 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSC.2020.2964244, IEEE
Transactions on Services Computing

SUBMIT TO IEEE TRANSACTIONS ON SERVICES COMPUTING 6

n
vs vd Flow-in node:
n
Node 𝑛 𝐷𝑗: It represents the demand of the mixed oil j during the period, 𝑏𝑛𝐷𝑗 = 𝐷𝑗 .

Flow-out node:
n D1
Node 𝑛𝑡𝑣 : 𝑉 (𝑡
𝑡 = 1, it represents the initial oil volume in the crude vessel v, 𝑏𝑛𝑣1 = 𝑉𝑣,1 , = 1).
ncdu1
1 n
n cdu1 ncdu1 𝑠 𝑆 (𝑡
2 3
Node 𝑛𝑡 𝑖: 𝑡 = 1, it represents the initial oil volume in the storage tank i, 𝑏𝑛𝑠𝑖 = 𝑉𝑖,1 , = 1).
1
n1v1 nv1
2 nv1
3 𝑐𝑗
Node 𝑛𝑡 : 𝑡 = 1 , it represents the initial oil volume in the charge tank j ,𝑏 𝑐𝑗 = 𝑉𝑗𝐵,1 , (𝑡 = 1).
𝑛𝑡

Node 𝑛 𝑣𝑠: It represents the virtual demand node of the virtual inventory, 𝑏𝑛𝑣𝑠 = 0.
State node:
j=1 n s1
1
s1
n2
s1
n3
Node 𝑛𝑡𝑣 : 𝑡 ≠ 1. It represents the state of the crude vessel v in the period t. Based on the
s2 ns22
2 n1 ns23
principle of mass conservation, 𝑏𝑛𝑣𝑡 = 0, (𝑡 ≠ 1).
3 c1 c1 c1
n1 n n 𝑠
2 3
4 Node 𝑛𝑡 𝑖: 𝑡 ≠ 1, it represents the state of the storage tank i in the period t. Based on the
t  0 t  1 n c2
1 t  2 n c2 t  3 n c2
2 2 principle of mass conservation, 𝑏𝑛𝑠𝑖 = 0, (𝑡 ≠ 1).
1
𝑐𝑗
Node 𝑛𝑡 : 𝑡 ≠ 1 it represents the state of the charge tank j in the period t. Based on the
Fig. 3. An example for the network principle of mass conservation, 𝑏 𝑐𝑗 = 0, (𝑡 ≠ 1).
𝑛𝑡

C. The rescheduling sub-problem Fig. 4. The three types of nodes

A solution for the crude oil problem is characterized by

Operation arc:
the schedule configuration (implying discrete state SCH) and Arc (𝑛 𝑣𝑠, 𝑛1𝑣 )
𝑣
:𝑣 ∈ 𝑽，representing the oil volume from the virtual inventory to the vessel
continuous state F , both of which are assigned to a specific
v.
schedule configuration. The cost of a solution is the sum of 𝑠
Arc (𝑛 𝑣𝑠, 𝑛1𝑖 ) :𝑣 ∈ 𝑽，representing the oil volume from the virtual inventory to the storage
operating costs and deviation cost, which refers to the measure
of efficiency and stability performance. If the discrete schedule tank i.
𝑐𝑗
SCH is fixed, and the according discrete state is known, then Arc (𝑛 𝑣𝑠, 𝑛1 ) :𝑣 ∈ 𝑽，representing the oil volume from the virtual inventory to the charge

the problem of determining the continuous state that fits to the tank j.
𝑣 𝑠
discrete schedule, results in a minimal loss. This is considered Arc (𝑛𝑡 𝑣 , 𝑛𝑡 𝑖 ) :𝑣 ∈ 𝑽, 𝑖 ∈ 𝑰, representing the oil volume from the vessel v to the storage tank
i in the time t.
a minimize cost network flow problem (MCFP).
𝑠 𝑐𝑗
Arc (𝑛𝑡 𝑖 , 𝑛𝑡 ) : 𝑖 ∈ 𝑰, 𝑗 ∈ 𝑱 , representing the oil volume from the storage tank i to the
Procedures to solve the problem are essentially comprised
charging tank j in the time t .
of the two following tasks : 𝑐𝑗 𝑐𝑑𝑢𝑙
Arc (𝑛𝑡 , 𝑛𝑡 ) :𝑗 ∈ 𝑱, 𝑙 ∈ 𝑳, representing the volume of the mixed oil from the charging tank
· Task1 : “Fixing the discrete schedule SCH ”. j to the CDU l in the time t.
· Task2 : “Solving the remaining problem”(the MCFP). Production arc
In order to adapt the solution procedures , these two tasks Arc (𝑛𝑡
𝑐𝑑𝑢𝑗
, 𝑛 𝐷𝑗 ) :𝑗 ∈ 𝑱, representing the demand of mixed oil j in the time t.
have to be thought over. 𝐷𝑗 vd
Arc (𝑛 , 𝑛 ) :𝑗 ∈ 𝑱, representing the demand of mixed oil j.
The MCFP with fixed discrete state: To evaluate the
Inventory arc
minimal costs attainable for the already fixed discrete schedule 𝑣
Arc (𝑛𝑡 𝑣 , 𝑛𝑡+1
𝑣 𝑣
) :𝑡 ∈ 𝑻\{𝑇}, 𝑣 ∈ 𝑽，representing the inventory of the vessel v at the end of
SCH; the following form of MCFP must be solved to obtain
time t.
optimality on the network: 𝑠 𝑠
Arc (𝑛𝑡 𝑖 , 𝑛𝑡+1
𝑖
) :𝑡 ∈ 𝑻\{𝑇}, 𝑖 ∈ 𝑰，representing the inventory of the storage tank i at the end
MCFP X
min ckl xSkl 𝑐𝑗 𝑐𝑗
of time t.
Arc (𝑛𝑡 , 𝑛𝑡+1) :𝑡 ∈ 𝑻\{𝑇}, 𝑗 ∈ 𝑱，representing the inventory of the charge tank j at the end
(k,l)∈AS
of time t.
s.t. 𝑣
Arc (𝑛 𝑇𝑣 , 𝑛 𝑣𝑑 )
X X : 𝑣 ∈ 𝑽，representing the inventory of the vessel v at the end of the period.
xSkl − xSlk = bk ,
𝑠
{l:(k,l)∈AS } {k:(l,k)∈AS } Arc (𝑛 𝑇𝑖 , 𝑛 𝑣𝑑 ) :𝑖 ∈ 𝑰，representing the inventory of the storage tank i at the end of the period.
𝑐
Arc (𝑛 𝑇𝑖 , 𝑛 𝑣𝑑 ), :𝑗 ∈ 𝑱， representing the inventory of the charge tank j at the end of the period.
LSkl ≤ xSkl ≤ Ukl
S
∀(k, l) ∈ AS
The network g = (N , A) consists of a set of nodes N and a Fig. 5. The three types of arcs
set of arcs AS . The problem is the identification of a flow xSkl
for each arc (k, l) ∈ AS . Moreover, ckl denotes the per unit
profit of arc (k, l). The flow xSkl on arc (k, l) is bound by the A directed arc (k, l) from the tail k to the head l connects
lower boundary LSkl and the upper boundary Ukl S
. the node k ∈ N with the node l ∈ N , there are three types
Figure.3 shows an example of a network g = (N , A) with of arcs: operation arc, production arc, and inventory arc. The
J = 4 and T = 3 macro-periods (plus the additional fictitious arcs of AS are summarized in Fig.5.
period 0). A network can represent all the possible material
flows between the crude-oil entities.
Node: D. The whole disruption management model
There are three types of nodes.The nodes N are summarized The general idea: If the solution of the problem
in Figure.4. M CF P (SCH) can be determined, the corresponding discrete
Arc: schedule SCH is feasible. The set S of the SCH is limit-

1939-1374 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSC.2020.2964244, IEEE
Transactions on Services Computing

SUBMIT TO IEEE TRANSACTIONS ON SERVICES COMPUTING 7

Solution Procedure
Disruption
First Task: Fix the discrete schedule Alarm
Aim Seek the best schedule

Initial Schedule
Second Task: Solve the MCFP problem

Generating the initial feasible schedule

Fig. 6. The whole optimal process

Hierarchical searching procedure

ed,then the disruption management optimization problem P is
to seek the best schedule SCH in the S . It is done as follows. Large-scale search Small-scale search

min{C(SCH))|SCH ∈ S}

C(SCH) determines the scheduling cost ,where

C(SCH) = α· CCost (SCH) + β· Cd (SCH) .
Figure.6 shows the control process. Although the combi- New Schedule
nations of the discrete scheduling are limited, its cardinal
numbers are still considerable, and it is still very difficult to
solve this disruption-management optimization problem. In the Fig. 7. The framework of the rescheduling procedure
following section, a rescheduling process is presented to obtain
the new schedule quickly.
Based on the real time data and KPI monitors, when
a sustained deviation is detected , the disruption alarm is
VII. T HE RESCHEDULING PROCEDURE MODULE detected.
Given an initial schedule and a disruption alarm, the system Step 2: Initial feasible schedule generation
is rescheduled to cope with these new conditions. The charac- A feasible initial schedule represents the chosen sequences
teristic of the disruption management is that it is a real-time of operations that optimize the performance measures, without
procedure, which should obtain the new schedule quickly and violating the technological constraints. The method of gener-
reduce the negative impact of the disruption on the system. ating the initial feasible schedule for rectifying the violation is
In this section, the corresponding effective rescheduling designed, resulting in a sufficient starting point in the optimal
procedure is proposed to increase the quality of the solution, control process.
which is designed based on the characteristics of the problem. Step 3: Hierarchical searching procedure
It shall be pointed out that this new schedule is not re- The searching procedure based on the hierarchical frame-
optimized using the rescheduling process, but it is optimized work is proposed to determine the new optimal schedule.
based on the initial schedule. Moreover, the new schedule Each of these steps will be discussed further in more detail.
needs to be generated quickly, and it should significantly
reduce the disruption impact on the system. B. Detection of disruption alarm
The disruptions typically arise from incomplete unloading
A. The framework of the rescheduling procedure of the vessel, transfer flow limits, and underproduction or over-
production. Once the disruption alarms have been identified,
The framework of the rescheduling procedure is proposed in
they are imposed on the network and their effect is determined
this section. Figure.7 shows the rescheduling process. The key
by graph propagation. Figure.8 provides a summary of the
component is the network of the refinery operations, described
different arcs, nodes, and disruptions in which they may occur.
in Section VI, which can represent the cause and effect rela-
Arc:
tionships between the variables, and all the possible material
• Disruption: Violating the limits of the volume of the tank.
flows between the crude-oil entities. This guarantees flexibility
Node:
in accommodating changes in the refinery configuration during
• Disruption: changes in timing (e.g., ship arrival delays),
unforeseen types of disruptions. Once the disruptions have
unavailability of the equipment (e.g., pumps, tanks, CDUs,
been identified. The disrupted schedule becomes the base-
etc). For example, the CDU fails to work during the scheduling
case for rescheduling, resulting in a sufficient starting point in
period.
the optimal control process. Then, the rescheduling procedure
based on the hierarchical framework is repeat until the stop
criterion is satisfied. And, the best schedule that optimizes C. Feasible initial schedule generation
the user-specified criteria is thus selected. The following steps Once the disruptions have been identified, the disrupted
formalize the rescheduling procedure: initial schedule becomes the base-case for generating a new
Step 1: Detection of disruption alarms feasible schedule. The feasible initial schedule optimizes the

1939-1374 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSC.2020.2964244, IEEE
Transactions on Services Computing

SUBMIT TO IEEE TRANSACTIONS ON SERVICES COMPUTING 8

Arc Disruption A 40 B A 70 B

30 30
Vessel-to-storage tank connection ：Violation of minimum/maximum vessel-to- storage tank transfer rate 110
90
40
80
80 110
90
40
80
80
C D E C D E
Storage tank to charging tank connection ：Violation of minimum/maximum storage tank to charging tank transfer rate 60 50 30
20
20
60 50 30
20
20
F G H F G H

CDU flow-in connection ：Violation of minimum/maximum CDU flow-in 10 30 30 10 30 30

10 20 50 10 20 50
Node Disruption I G K I G K

10 20 10 20
Vessel ：Changes in timing (e.g. vessel arrival delays) 90
L
80 M
50 N
70 90
L
80 M
50 N
70

Storage tank ：Unavailability of storage tank 60 60

60 30 60 30
O 30 O 30
Charging tank ：Unavailability of charging tank P X P X

30 30
CDU ：Unavailability of CDUs 30
R
30
S
60
T
60 30
R
0
S
30
T
30

60 30 90
90 90 90 60 90 90
U V W U V W

t=1 t=2 t=3 t=1 t=2 t=3

Fig. 8. Detection of disruptions a b

A 70 B A 100 B

30 30 40
110 40 80 110 80
90 80 90 80
C C D E
D E
20 30 20
30 60 50
performance measures, without violating the technological 60
F
50
G
20
H
F

10 30
G
20
H

30
10 30 30

constraints. However, it is difficult to generate the initial 10

I
20
G
50
K
10
I
20
G
50
K

10 20 10 20 70
feasible schedule because of the disruption and numerous con- 90
L
80 M
50 N
70 90
L
80 M
50 N

60 60

straints. The method of generating the feasible initial schedule O

60
P
30
X 30 O
60
P
30
X 30

30
proceeds by constructing the augmenting paths until the flow 30
R
0
S
30
30
T
30
30
R
0
S
30
T
0

90 90
satisfies the constraints, corresponding to the procedure of 60
30
U
90 V
90 W
30
30
U
60 V
60
30
W

t=1 t=2 t=3 t=1 t=2 t=3

partially changing the discrete-schedule configuration. When c i
flow j
d

the feasible discrete-schedule configuration is obtained, the

negative-cycle cancelling algorithm [29] is then used to obtain Fig. 9. The evolution process (a → b → c → d)
the optimal schedule.
The disruption occurring from the violations of the limits of Schedule

the volume of the tank is taken as an example to explain the where

steps to obtain a feasible initial schedule, with the disrupted

schedule shown in Figure.9a.
• Disruption:
The flow Q02 , on a production arc (A, U ) = (nvs , nS1 2 ) ,
represents the initial inventory of the storage tank S2 . When v1
Operation Executions
the limits of the production arc (A, U ) change, the maximum V2
flow Q02 is limited to 30. However, the flow Q02 in the initial S1
S2
schedule is equal to 90, which is infeasible .
S3
• The initial feasible schedule generating method S4
The method proceeds by constructing the augmenting paths C1

[29] until the flow Q02 decreases to a feasible range. The C2

following is the evolution process (a → b → c → d):

Step 1: Figure.9a shows the initial schedule. The corre- Fig. 10. The corresponding discrete schedule of initial schedule
sponding discrete schedule showing in Figure.10 is assigned
SCH = {SCHt=1 , SCHt=2 , SCHt=3 } , the flow Q02 is
equal to 90, which is infeasible. coverage of the search space, the small-scale search method
Step 2:As there are no augmenting paths decreasing the is conducted. The process would repeat until the stop criterion
flow Q02 , the arc (R, U ) is added to improve the produc- is satisfied. Finally, the best schedule is output.
tion ability in the period 1, and the new augmenting path • The large-scale search
(A, B, T, S, R, U, A) is constructed. Then, the flow Q02 can The large-scale search is applied to exclude the inferior
reduce up to 30 for the capacity constraint, and the network solutions by establishing an evaluation criterion. First, the
graph Figure.9b is generated. population of the candidate discrete schedule S is randomly
Setp 3:The arc (T, W ) is added to improve the produc- generated using the method introduced by Van[28],which
tion ability in the period 3.Based on the augmenting path complies with the sequencing rules and operation condition.
(A, B, T, W, V, U, A), the network graph Figure.9d is gener- Then, the corresponding costs of the schedules are calculated
ated. At this moment, the new schedule is feasible. using the utility function. The best schedule that optimizes the
The new feasible schedule is constructed based on the initial utility function values is then selected.
schedule. It is important to note that finding an initial feasible For the discrete scheduling SCH,the utility function is as
schedule quickly is necessary to allow pruning of the graph at follows:
an early stage.
C(SCH) = α· CCost (SCH) + β· Cd (SCH)
D. The rescheduling procedure Where, CCost (SCH) determines the operating cost;
The rescheduling procedure based on the hierarchical frame- Cd (SCH) determines the deviation cost.
work is proposed. First, the large-scale search method is used The schedules will be selected based on the utility functions.
to exclude the inferior solutions. Second, to obtain a better The whole set of the feasible schedules are considerable to be

1939-1374 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSC.2020.2964244, IEEE
Transactions on Services Computing

SUBMIT TO IEEE TRANSACTIONS ON SERVICES COMPUTING 9

Rescource categories Vessel 1 to Storage tank 1

Uploading oil

Vessel 2 to Storage tank 2

Uploading oil

100% Vessel 3 to Storage tank 3

Uploading oil

Vessel 4 to Storage tank 4

Uploading oil
80%
Storage tank 1 to Charging tank 1 Transfering
oil

60% Storage tank 1 to Charging tank 2 Transfering

oil

Storage tank 2 to Charging tank 1 Transfering

40% oil

Storage tank 2 to Charging tank 3 Transfering

oil
20%
Storage tank 3 to Charging tank 2 Transfering
oil

0% Storage tank 3 to Charging tank 3 Transfering

oil
CPU RAM Network Message size Storage tank 4 to Charging tank 4 Transfering Transfering
oil oil

75 54 23 7 Charging tank 3 charges CDU 1 Charging

oil
Charging oil

Charging tank 4 charges CDU 1

Charging oil Charging oil

Charging tank 1 charges CDU 2

Charging oil

Fig. 11. Comparison of different device groups Charging tank 2 charges CDU 2
Charging oil Charging oil

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

processed simultaneously. To obtain a better coverage of the Fig. 12. Initial schedule for disruption case 1
search space, the ‘small-scale search ’method is designed with
minimal effort and does not require the problem to be solved
completely. because the total rescheduling strategy has a shortcoming, i.e.
• The small-scale search the ever-changing schedule which can give rise to instability.
The Neighbors search method: Let us suppose two neighbor In practice, this is simply unacceptable. This leads to the con-
0 0 0
operations π and π , when π → π , with G → G . The clusion that the proposed rescheduling method is the method
0
slight differences between G and G is the few arcs 4π of choice when higher stability is the primary concern.
(Lπkl , Ukl
π
, (k, l) ∈ 4π ). 2) Efficiency:: The performance of efficiency refers to the
The priority of the two operations can be determined by optimality of the schedule with respect to the operating cost.
using the dual network flow algorithm [30], whereby shorter Efficiency is a cost function and is considered high if the
computation times are attainable because highly expensive operating cost is low.
candidates have already been rejected. The results are presented in Figure.14, in which the efficien-
cy performance measures of ten disruption events are plotted.
VIII. E XPERIMENTAL S TUDY Regarding the efficiency measure, the results show that, al-
though the operating cost for the schedules obtained using
In this section, SimpleAgentPro is used to simulate one or the proposed rescheduling were often higher than the total
more IoT devices. And the Grafana tool is used to monitor the rescheduling method, there is no significant difference in cost
resource consumption concerning memory, CPU and network when using the total rescheduling and proposed rescheduling
usage. Figure.11 shows the average results for the device methods at the 4% significance level. This result suggests that,
groups and the appropriate resources. We can see that as the for the crude oil problem, the proposed rescheduling method
number of devices grows, the resource utilization also gets degrades efficiency much less than it improves stability.
higher. 3) Analysis: Two analyses are performed on the simulation
To illustrate the performance of the proposed rescheduling results:
method in smart disruption management service system , the 1. There is an improvement in the stability performance
following two methods are compared: when using the proposed rescheduling compared to the total
1. Proposed rescheduling method: The disruption manage- rescheduling, which is statistically significant. Although the
ment method described in Section VII. proposed method generates more operating cost associated
2. Total rescheduling: To optimize all the operations, the with total rescheduling, producing costs are not statistically
rescheduling method proposed by Reddy [32] is used. higher.
The schedule given in Figure.12 is an initial schedule for 2. The rescheduling method proposed in this paper opti-
the problem obtained by Reddy [32]. The according operation mizes both the two measures: efficiency and stability. There-
results are shown in Figure.13 and 14. fore the new schedules generated by it are not only efficient
1) Stability:: The performance of stability means the op- but also stable. In practice, the decision maker can find a
timality of the schedule in terms of the deviation cost. The trade-off between higher efficiency and stability. Therefore,
lower the deviation cost, the higher the level of stability. the proposed rescheduling method is more suitable for a
The results are presented in Figure.13, in which the stability disruption-management problem.
performance measures of ten disruption events are plotted.
The results indicate that the proposed rescheduling method
decreases the deviation cost compared to the total rescheduling IX. C ONCLUSION
method. The total rescheduling process gives the less optimal Recently, IoT technologies have been widely implemented
performance with respect to the stability measure. This is in crude-oil industries. Such an automatic data collecting

1939-1374 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSC.2020.2964244, IEEE
Transactions on Services Computing

SUBMIT TO IEEE TRANSACTIONS ON SERVICES COMPUTING 10

50
48.21
Proposed rescheduling method
45 Total rescheduling method [8] S. J. Qin. “Process data analytics in the era if big data,” AICHE J., vol.
40 39.21
38.45
40.29 60, no. 9, pp. 3092-3100, 2014.
Stability: deviation cost

35
36.54 36.68
35.94

32.75
34.63 [9] M. Joly (2012). “Refinery production planning and scheduling: The
30
30.86
refining core business,” Braz. J. Chem. Eng., vol. 29, no. 2, pp. 371-384,
27.39

25
23.47
24.82
23.46
2012.
22.45
20.86
20
18.45
18.86
[10] A. Sabata, B. Sean. “Remote monitoring of pipelines using wireless
15 15.52

13.46 sensor network,” U.S. Patent 7,526,944., issued May 5, 2009.

10
0 Event 1 Event 2 Event 3 Event 4 Event 5
Disruption
Event 6 Event 7 Event 8 Event 9 Event 10 [11] P. S. Allison, E. C. Charles, L. Bryan . “Acoustic impact detection and
monitoring system,” U.S. Patent 7,607,351., issued October 27, 2009.
[12] J. A. Filar, P. Manyem, and K. White, “How airlines and airports recover
Fig. 13. Performance of the stability
from schedule perturbations: a survey.” Ann Op Res., vol. 108, no. 1,
390
pp. 315-333, 2001.
387.23
Proposed rescheduling method
Total rescheduling method
[13] J. M. Rosenberger, E. L. Johnson, and G. L. Nemhauser, “Rerouting
385 aircraft for airline recovery.” Transport Sci., 2003, vol. 37, no. 4, pp.
408-421, 2003.
Efficiency: operating cost

381.84

380 380.25
379.35

[14] M. Xu, X. Qi, G. Yu, H. Zhang, and C. Gao, “The demand disruption
375
373.52
374.83
375.56
management problem for a supply chain system with nonlinear demand
373.23

370.73
372.93

370.73
372.98
371.97

370.75
372.74 372.49
functions.” J Syst Sci Syst Eng., vol. 12, no. 1, pp. 82-97, 2003.
370 369.49
369.85
369.39
[15] A. S. Kunnathur, P. S. Sundararaghavan, and S. Sampath,“Dynamic
366.92

365
0 Event 1 Event 2 Event 3 Event 4 Event 5 Event 6 Event 7 Event 8 Event 9 Event 10
rescheduling using a simulation-based expert system.” J Manuf Technol
Disruption Manage., vol. 15, no. 2, pp. 199-212, 2004.
[16] M. Mishra, R. Srinivasan, and I. A. Karimi, “Managing disruptions in
Fig. 14. Performance of the efficiency refinery supply chain using an agent-based decision support system,” In
AIChE annual meeting., San Francisco, pp. 16-21, 2003.
[17] C. Baumeister, G. Peersman, and “The role of timevarying price elas-
ticities in accounting for volatility changes in the crude oil market.”
approach brings new challenges. For example, one of the Journal of Applied Econometrics., vol. 28, no. 7, pp. 1087-1109, 2013.
challenges is how to use the real-time and multi-source data [18] T. A. Oddsdottir, M. Grunow, and R. Akkerman, “Procurement planning
to management disruption for improving scheduling perfor- in oil refining industries considering blending operations.” Comput.
mance. In this work, a smart disruption management service Chem. Eng., vol. 58, no. 11, pp. 1-13, 2013.
[19] R. J. Abumaizar, and J. A. Svestka,“Rescheduling job shops under
system has been proposed to provide a new solution for random disruptions.” Int J Prod Res., vol. 35, no. 7, pp. 2065-2082,
crude-oil system to enhance the efficiency and accuracy of 1997.
critical event monitoring and exception control. In such a [20] J. Roslöf, I. Harjunkoski, J. Björkqvist, and S. Karlsson, “An MILP-
based reordering algorithm for complex industrial scheduling and
service system, a disruption-management module with IoT is rescheduling.” Comput Chem Eng., vol. 25, no. 4, pp. 821-828, 2001.
proposed to measure both efficiency and stability, and thereby [21] C. A. Mendez, and J. Cerd, “An MILP framework for batch reactive
develop new schedules that are more suitable to the disruption scheduling with limited discrete resources.” Comput Chem Eng., vol. 28,
no. 6, pp. 1059-1068, 2004.
management. The disruption management requires efficient [22] A. Adhitya, R. Srinivasan, and I. A. Karimi, “Managing abnormal events
algorithms; moreover, a new schedule should be generated in refinery supply chains by heuristic rescheduling,” Technical report.
quickly and have a less disruptive impact on the system. Our Laboratory for Intelligent Applications in Chemical Engineering., Na-
tional University of Singapore, 2006.
proposed techniques can accelerate the solution procedures. [23] A. Adhitya, R. Srinivasan, and I. A. Karimi, “A model-based reschedul-
The results show that there is an improvement in performance ing framework for managing abnormal supply chain events,” Comput.
when using the proposed solution compared to the traditional Chem. Eng., vol. 31, no. 5, pp. 496-518, 2007.
[24] G. Yu, and X. Qi, “Disruption management: framework, models and
ones, and this improvement is statistically significant. applications,” World Scientific., 2004.
[25] G. S . Jo, J. J. Jung, J. H. Koo, and S. H. Hyun, “Ramp activity expert
R EFERENCES system for scheduling and coordination at an airport,” Ai Mag., 2000,
vol. 21, no. 4, pp. 75-82, 2000.
[1] A. Kevin, “That Internet of things thing,” RFID Journal., 2009.
[26] J. M. Rosenberger, E. L. Johnson, and G. L. Nemhauser, “Rerouting
[2] D. Guinard, V. Trifa, S. Karnouskos, P. Spiess, and D. Savio, “Interacting
aircraft for airline recovery,” Transport Sci., vol. 37, no. 4, pp. 408-421,
with the soa-based internet of things: Discovery, query, selection, and
2003.
ondemand provisioning of web services,” IEEE Trans. Serv. Comput.,
[27] A. Malucelli, A. Castro, E. Oliveira, “Crew and aircraft recovery
vol. 3, no. 3, pp. 223235, Jul./Sep. 2010.
through a multi-agent airline electronic market,” IADIS International
[3] Vogler M, Schleicher J, Inzinger C, et al, “Optimizing elastic IoT
Conference-Commerce., Barcelona, 2006, pp. 51-58.
application deployments,” IEEE Trans. Serv. Comput., 2016, in press.
[28] Van. Noord. G, “FSA Utilities: A toolbox to manipulate finite-state
[4] I. R. Chen, J. Guo, and F. Bao, “Trust Management for SOA-based IoT
automata,” International Workshop on Implementing Automata. Springer
and Its Application to Service Composition,” IEEE Trans. Serv. Comput.,
Berlin Heidelberg, pp. 87-108, 1996.
vol. 9, no. 3, pp. 482-495, 2016.
[29] R. K. Ahuja,T. L. Magnanti,J. B. Orlin,“Network flows: theory, algo-
[5] D. Zhang, X. Wang, X. Song, et al. “A novel approach to mapped
rithms, and applications[J].”1993.
correlation of ID for RFID anti-collision,” IEEE Trans. Serv. Comput.,
[30] O. Stursberg, “A graph search algorithm for optimal control of hybrid
vol. 7, no. 4, pp. 741-748, 2014.
systems,” In Decision and Control, CDC. 43rd IEEE Conference on.,
[6] T. S. Lopez, D. C. Ranasinghe, and M. Harrison, “Adding sense to the
vol. 2, pp. 1412-1417, 2004.
internet of things: An architecture framework for smart object systems,”
Pers. Ubiquit. Comput., vol. 16, no. 3, pp. 291308, 2012. [31] D. P. Bertsekas, “Network Optimization: Continuous and Discrete Meth-
[7] J. D. Kelly, D. Zyngier. “Hierarchical decomposition heuristic for ods,” Athena Scientific., Vol. 8, 1998.
scheduling: Coordinated reasoning for decentralized and distributed [32] P. Reddy, I. A. Karimi, and R. Srinivasan, “Novel solution approach for
decision-making problems,” Comput. Chem. Eng.,vol. 32, no. 11, pp. optimizing crude oil operations,” AICHE J., vol. 50, no. 6, pp. 1177-
2684-2705, 2008. 1197, 2004.

1939-1374 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.