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/ACCESS.2019.2928992, IEEE Access

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2017.Doi Number

Research on Task Priority Model and Algorithm

for Satellite Scheduling Problem
JIAN WU1,2, JIAWEI ZHANG2,3, JINGHUI YANG4, and LINING XING1,2
1
School of Logistics & Transportation, Central South University of Forestry & Technology, Changsha 410073, P.R. China
2
College of Systems Engineering, National University of Defense Technology, Changsha 410073, P.R. China
3
College of Information and Communication, National University of Defense Technology, Xian 710106, P.R. China.
4
College of Engineering, Shanghai Polytechnic University, Shanghai 201209, P.R. China

Corresponding author: JIAWEI ZHANG([email protected]) and LINING XING([email protected])

ABSTRACT In satellite scheduling system, one of the most important issues is task scheduling. Most of
the existing algorithms focused on the scenario of normal tasks with fixed priority and don’t consider
emergency tasks. However, emergency tasks in satellite scheduling system have the independent arrival time
and execution deadline. To deal with this challenge, we propose the task scheduling model, task priority
model and task scheduling algorithms based on RHO (Rolling-Horizon Optimization) strategy to manage
the tasks. In task scheduling model, we construct a scheduling model with multiple constraint conditions.
In task priority model, we decompose task priority into four parts, including target priority, imaging task
priority, TT&C (track, telemetry & control) requirement priority, and data transmission requirement priority,
and describe the factors corresponding to each priority based on the actual engineering. The RHO strategy
is designed with a periodical triggering mode, and scheduling process is decomposed into a series of static
scheduling intervals, the dynamic scheduling of emergency tasks is realized by optimizing the scheduling
schemes in each interval. Meanwhile, task scheduling algorithms including Heuristic algorithm, Genetic
algorithm and NSGA- Ⅱ algorithm are designed to solve the proposed task scheduling model. Our
simulation results show that the proposed task priority model and RHO strategy are effective, which can
significantly improve the number of overall tasks and emergency tasks. Meanwhile, each task scheduling
algorithm has different performance in the processing time and optimal objectives.

INDEX TERMS Satellite scheduling system; Task priority model; Rolling-Horizon Optimization strategy;
Task scheduling algorithms

I. INTRODUCTION fixed priority and don’t consider emergency tasks. Since

With the fast development of various fields in satellite emergency tasks in satellite scheduling system have the
scheduling system such as remote telemetry, remote control, independent arrival time and execution deadline. Therefore,
data transmission, the huge growth of space resources and the real-time or near-real-time scheduling strategy is more
users’ demand increase the difficulty of satellite task effective in solving the problem of emergency tasks
scheduling [1]. The satellite task scheduling is to use the scheduling.
effective methods to schedule the multiple satellites and In this paper, we propose the task scheduling model, task
tasks in the dynamic environment, which is constrained by priority model and task scheduling algorithms based on
the multiple factors including tasks scheduling time, task RHO (Rolling-Horizon Optimization) strategy to manage
execution time, the memory capability of the satellites, etc. the tasks. In task scheduling model, we construct a
In the process of task scheduling, task priority has an scheduling model with multiple constraint conditions. In
important influence on the scheduling results, which can the task priority model, The task priority is decomposed
objectively reflect the relative importance of the indicator into four parts, including target priority, imaging task
and subjectively reflect the importance of the decision priority, TT&C (track, telemetry & control) requirement
maker to the indicator. Therefore, it is important to priority and data transmission requirement priority. Then,
determine the task priority in satellite scheduling system. factors corresponding to each priority are described based
However, research on the task priority for satellite on the actual engineering. Finally, the initial priority of the
scheduling are not sufficient, and most of the existing task is determined according to the task priority model. The
algorithms focused on the scenario of normal tasks with dynamic task scheduling algorithms based on RHO strategy

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

are proposed according to the independent arrival time and HVDF algorithm to ensure the task priority. Aldarmi et al.
execution deadline of emergency tasks. The RHO strategy [18] proposed a real-time scheduling strategy to optimize the
is designed with a periodical triggering mode, and system resources. The aforementioned task priority models
scheduling process is decomposed into a series of static are generally designed according to a factor of tasks, various
scheduling intervals, the dynamic scheduling of emergency factors are not considered in the task scheduling problem.
tasks is realized by optimizing the scheduling schemes in Several well-known algorithms like ABCA (Artificial Bee
each interval. Colony Algorithm) [19-22], SA (Simulated Annealing) [23-
The rest of this paper is organized as follows. Section Ⅱ 25], GA (Genetic Algorithm) [26-28] and PSOA (Particle
presents literature review regarding task priority model and Swarm Optimization Algorithm) [29-30] have been applied
algorithm. Section Ⅲ formulates the task scheduling model so far to the task scheduling problem. Kai et al. [31]
with multiple constraints. In section Ⅳ, the task priority proposed a ABCA to solve the relay satellite scheduling,
model is proposed to determine the initial priority of the which can easily get the domain optimal solution. B Deng
task. In section Ⅴ, the RHO strategy and the task et al. [32] proposed a two-phase task scheduling algorithm to
scheduling algorithms are proposed. The performance of improve the performance of scheduling, which can
the task priority model, the RHO strategy and the proposed significantly improve the performance of dynamic
scheduling in DRS (data relay satellite) system. X. Liu et al.
algorithms are evaluated by simulations in Section Ⅵ.
[33] proposed a multi satellites scheduling algorithm to
Finally, Section Ⅶ gives a conclusion and further research
weaken the constraints arising from satellites' poor slew
to this paper.
ability. Keshanchi et al. [34] used heuristic-based HEFT
search to improve GA, which can optimize the total
II. LITERATURE REVIEW
execution time.
In recent years, with the expansion of real-time system
application, people have paid more attention to how to
III. TASK SCHEDULING MODEL UNDER MULTIPLE
schedule task flexibly and efficiently. Therefore, it is CONSTRAINT CONDITION
important to design a reasonable task priority model, which As shown in Figure 1, the satellite scheduling system is
can objectively reflect the relative importance of the mainly composed of multiple satellites, control stations and
indicator and subjectively reflect the importance of the data receiving stations. The satellites use various satellite-
decision maker to the indicator. Several well-known borne sensors to collect the important information, then
researchs like EDF (Earliest Deadline First) [2-4], LSF transmit the data directly or via the relay satellite to the
(Least Slack First) [5-6], HVF (Highest Value First) [7-8], ground. The satellites can only transmit data within the
HVDF (Highest Value Density First) [9-11]. Based on the visible time window between satellites and data receiving
EDF model, Semghouni et al. [12] proposed a new task stations. As shown in Figure 2, the organization of satellite
scheduling model to improve the algorithm performance in scheduling is actually a cyclical process, which contains
handling system overload. Chantem et al. [13] proposed a multiple complex links.
more general-purpose real-time task model, which allowed
the deadline of periodic tasks to change during a certain
period. In order to improve the schedulability of the task and
the efficiency of system. Balbastre et al. [14] proposed a new
algorithm to calculate the earliest deadline by dynamically
adjusting the deadlines and periods of real-time tasks, which
can speed up task scheduling and reduced the waiting time
of system. Moreover, In order to overcome the
shortcomings of EDF algorithm and LSF algorithm, Jin et
al. [15] adopted a comprehensive weighted method to
determine the task priority. To improve the flexibility and
efficiency of real-time system, Burns et al. [16] proposed the
priority scheduling strategy based on value, meanwhile,
they applied the algorithm to the car automatic navigation
system, and use measurement and decision analysis
methods to calculate the task priority. Buttazzo et al. [17] FIGURE 1. The satellite scheduling system
determined the value density according to the value of the
tasks and the maximum execution time, then proposed

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

FIGURE 2. The process of satellite scheduling

The problem of satellite task scheduling can be regarded the visible time window. d means that the task execution
as an optimization problem with multiple constraints. time is not lower than the minimum imaging time. e is the
System parameters and decision variables are given as constraint for the tasks storage, which means the tasks
follows. storage cannot exceed the maximum spaceborne memory
• task  [task1，task 2 ....task nt ] is the set of tasks, nt is the storage. f is the constraint for attitude conversion, which
amount of tasks. means the attitude conversion of the satellite in orbital
• Sat  [ Sat1 , Sat1....Sat n ] is the set of satellites , nsat is the circle cannot exceed the maximum numbers of attitude
number of satellites. conversion.
• ai is the time that taskti arrived at the system, ei is the nt nsat qi , j

 x p
qi , j
deadline. TWi  njsat 1 k 1 twi , j is the set of time windows,
k
Max f g ( s )  k
(2)
ij ti
which is an important parameter to describe whether the i 1 j 1 k 1

link is connected or not at a certain instant for taskti , The

1/ 2

 
time window can be defined as an interval  nsat

twik, j  [ wsik, j , weik, j ] , which represents the starting and ending  L( sat j )  L( Sat ) 
 
points for the time window k  TWi on Sat j , qi , j is the number Max f c ( s )  1   j 1
 / L( Sat ) (3)
 nsat 
of the time windows.
 
• etwi , stwi  is the execution time of taskti , stwi is the  
starting time instant of taskti , etwi is the ending time instant
of taskti , pti is the priority of taskti , which represents the  nSat qi , j k
importance level of taskti , d i is the imaging time for
a :


j 1 k 1
xij  1,[ stwi , etwi ] [ stwi* ,etwi* ]  
taskti , hi is the total storage for taskti , rj is the maximum  nSat qi , j

number of attitude conversion, M j is the spaceborne
memory storage.
b : a i  stw i , etw i  ei , if  xijk  1
 j 1 k 1

In addition, the formulation also requires a group of  qi , j

qi , j

decision variables. The task scheduling indicator xij is

k c : [ stwi , etwi ]  twik, j , if
 k 1
 xijk  1
defined by an binary variable. 
k 1
nSat qi , j

1，taskti is executed by satellite sat j in time window k
xijk   (1)
s.t. d : etwi  stwi  d i , if

 j 1 k 1
xijk  1

0，otherwise  nt qi , j

Based on the characteristic and parameters of satellite e :  xijk hi  M j
 i1 k 1
scheduling system ,the task scheduling model can be  nt qi , j

formulated as follows. The objective f g (s ) is the total

revenue of the scheduled tasks, the objective f c (s ) is the
f :


i 1 k 1
xijk  rj

satellite resource load balance. On the other hand, for the i  [1,2,3...nt ], i  [1,2,3...nt ]
*

constraints, a means that each task is only executed once,  j  [1,2,3...nSat ], k  [1,2,3...qi , j ]

and there is no preemption service during execution.
b and c are the constraints for tasks scheduling Where
time, b means that the task should be executed before the
deadline, c means that the task should be executed within

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

nSat
1, factor x is more important than factor y
nt  L(sat ) j 
axy  0.5, factor x is as important as factor y (6)
L( sat j )   x (etw  stw ) , L(Sat ) 
i 1
k
ik i i
j 1

nsat
(4) 0, factor y is more important than factor x

The weight coefficient of the factor i is calculated as
IV. TASK PRIORITY MODEL
follows:
In general, The satellites are used to conduct the imaging  ny 1 a xy
task according to the plan. The plan includes the ground i  , x  1,2    n (7)
 nx1  ny 1 a xy
area to be observed, the starting and ending time of the
corresponding observations, track, telemetry, control, and 2) CALCULATE THE QUANTITATIVE VALUES OF
data downlink. Therefore, the task priority is decomposed FACTORS
into four parts, including target priority, imaging task (1) Type
priority, TT&C (track, telemetry & control) requirement The quantitative value of types are shown in Table 2:
TABLE 2
priority and data transmission requirement priority. Then QUANTITATIVE VALUE OF TYPES
factors corresponding to each priority are described Status
Type
according to the actual engineering. Stationary Moving
Point 1 6
Area
Region 6 10
A. TARGET PRIORITY
(2) Imaging
Target priority is used to assess the importance of the
To satisfy the users’ requirement, the images acquired by
imaging target. Different targets have different priorities
the satellites must reach a certain quality. In this paper, we
since characteristics of each target are different. We study
use NIIRS level of the images as the imaging priority. The
the various factors that affect the target priority in the
formula is as follows:
academic field, and consider the optimal objectives and
constraints in the actual engineering. The factors that affect NIIRS  c  3.32lgGSDGM (8)
the target priority are shown in Table 1.
TABLE 1 Where GSDGM is the ground sample distance. For the
FACTORS THAT AFFECT TARGET PRIORITY visible light image, c  9.1915 . For the infrared image,
Factors c  9.82 . For the SAR image, c  10.5 .
Tar1
Type
area (3) Space-Domain
Tar2 Status Two aspects are considered in the space-domain, one is
Tar3 Image type
Tar4
Imaging
Minimum ground resolution the regional coverage, and its value is in the range of (0,10].
Tar5 Regional coverage The quantitative value is calculated as follows:
Space-Domain
Tar6 Geographical characteristic
Tar7 Observation frequency v Attr3 _ cov  Tar5  10 (9)
Time-Domain
Tar8 Longest observation interval
Each factor in Table 1 has different influence on the target where Tar5 is the demand of the imaging coverage. The
priority, therefore, it needs to be assessed. The weight another one is the geographical characteristic. The
coefficient is usually used to reflect the influence level of the quantitative value v Attr3 _ area is defined as follows:
TABLE 3
factor. The weighted summation method is used to calculate QUANTITATIVE VALUES OF GEOGRAPHICAL CHARACTERISTIC
the target priority, which is shown in Formula 5. Geographical characteristic Quantitative value
Characteristic A 10
Ptarget   i41 i  v Attri (5) Characteristic B 7
Characteristic C 5
Where Ptarget represents the target priority, i represents In conclusion, the quantitative value of space-domain is
the weight coefficient of each factor, v Attri represents the defined as follows:
quantitative value of each factor.
v Attr3  a  v Attr3 _ cov  (1  a )  v Attr3 _ area (10)
1) DETERMINE THE WEIGHT COEFFICIENT
Because users pay different attention to each factor, each In general, a is set as 0.5.
factor plays a different role in the priority evaluation. (4) Time-domain
Therefore, the comparative method is used to determine the The quantitative value of observation frequency v Attr4 _ frep
weight coefficient. First, we list all factors to form a square is defined as follows:
matrix n  n . Then, we invite users to compare and score
each factor through 0-1 scoring method. Finally, we sum the
scores of each factor. The elements are determined as
follows:

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

TABLE 4 follows:
QUANTITATIVE VALUES OF OBSERVATION FREQUENCY
Observation frequency Quantitative value rt Attr1  Ptarget (12)
1 1
2 2 (2) Application scenario
3 4
4 6 The quantitative value of application scenario rt Attr2 is
5 7 defined as follows:
6 9 TABLE 7
≥7 10 QUANTITATIVE VALUES OF APPLICATION SCENARIO
The quantitative value of longest observation interval Application scenario Quantitative value
Scenario A 2
v Attr4 _ itv is defined as follows: Scenario B 4
TABLE 5 Scenario C 6
QUANTITATIVE VALUES OF MAXIMUM TIME INTERVAL
Scenario D 8
Longest observation
Quantitative value Scenario E 10
interval
≤10 10 (3) User
10~20 9 The quantitative value of user rt Attr3 is defined as follows:
20~40 8 TABLE 8
40~60 7 QUANTITATIVE VALUES OF USER
60~90 5 User Quantitative value
90~120 3 User A 10
≥120 1 User B 8
In conclusion, the quantitative value of time-domain is User C 6
User D 4
defined as follows: User E 2
v Attr4  b  v Attr4 _ ferp  (1  b)  v Attr4 _ itv (11) (4) Satellite application
The quantitative value of satellite application rt Attr4 is
In general, b is set as 0.5. defined as follows:
TABLE 9
QUANTITATIVE VALUES OF USE
B. IMAGING TASK PRIORITY Satellite application Quantitative value
Imaging task priority is used to assess the importance of the Application A 10
imaging task. We study the various factors that affect the Application B 5
imaging task priority in the academic field, and consider (5) Working mode
the optimal objectives and constraints in the actual The quantitative value of working mode rt Attr5 is defined
engineering. The factors that affect the imaging task as follows:
priority are shown in Table 6. TABLE 10
TABLE 6 QUANTITATIVE VALUES OF WORKING MODE
FACTORS THAT AFFECT IMAGING TASK PRIORITY
Working mode Quantitative value
Constellation 5
Factors Single star 10
RT1 Source target (6) Type
RT2 Application scenario The quantitative value of type rt Attr6 is defined as follows:
RT3 User TABLE 11
RT4 Satellite application QUANTITATIVE VALUES OF TYPE
RT5 Working mode Type Quantitative value
RT6 Type Emergency task 10
RT7 Satellite attribute Normal task 1
RT8 Execution urgency (7) Satellite attribute
1) IMAGING TASK PRIORITY MODEL BASED ON The quantitative value of satellite attribute rt Attr7 is
IMPROVED TOPSIS METHOD defined as follows:
Since there are many periodic imaging tasks and imaging TABLE 12
task sets with contextual constraints in satellite scheduling QUANTITATIVE VALUES OF SATELLITE ATTRIBUTE
Satellite attribute Quantitative value
system, it is difficult to access the importance of each AW stars 10
imaging task. Therefore, we improve the TOPSIS ES stars 7
(Technique for Order Preference by Similarity to Solution) CU stars 3
method to determine the imaging task priority, which can get (8) Execution urgency
the objective and reasonable results. The procedure of The quantitative value of execution urgency rt Attr8 is
TOPSIS method is shown as follows: defined as follows:
Step 1: Calculate the quantitative values of factors.
(1) Source target
The quantitative value of source target rt Attr1 is defined as

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

rt Attr 8  10 * ettask ettask (13) The imaging task priority rt Attr7 is defined as follows:

Where ettask represents the deadline for the task , and ttask cmax  ci
Prt _ i  10  (19)
represents the current moment. cmax  cmin
Step 2: Construct the factor matrix X for imaging task
priority model. Where cmax  max1in ci , cmin  min1in ci 
Finally, the imaging task priority can be obtained through
 x11  x18  Step1~Step7, and its value is in the range of (0,10]. The
 
X  (xij )m8       (14) larger the value, the higher the imaging task priority.
 x1m    xm 8 
C. TT&C REQUIREMENT PRIORITY MODEL
Where xij (1  i  m,1  j  8) represents the quantitative TT&C (track, telemetry & control) requirement priority is
value of the jth factor in the ith imaging task. to maximize the satellites’ design efficacy through TT&C
Step 3: Standardization process for matrix X and obtain resource scheduling and to support the high-level decision-
the weighting standardization matrix Y. making process of TT&C system. The factors that affect the
TT&C requirement priority are shown in Table 13.
 y11  y18  TABLE13
  TT&C REQUIREMENT PRIORITY
Y  (yij )m8       (15) FACTORS THAT AFFECT
Factors
 y1m  ym 8  TT&C1 Source target
TT&C2 Control circle
xij TT&C3 TT&C resource
Where yij  ,1  i  m,1  j  8 .
10 TT&C4 Service capability
Step 4: Determine the plus ideal solution, minus ideal TT&C5 Event
TT&C6 Flight stage
solution and the weight coefficient of each factor. TT&C7 Available resources
(1) Construct the plus ideal solution Y   (1,1,  1) and TT&C8 Execution urgency
the minus ideal solution Y -  (0,0,  0) .
1) TT&C REQUIREMENT PRIORITY MODEL BASED ON
(2) Determine the weight coefficient based on DELPHI IMPROVED AHP METHOD
method. Each factor in Table 13 has different influence on TT&C
Since the importance of each factor is vague, it is difficult requirement priority , therefore, it needs to be assessed. The
to judge the relative importance according to the user’s weight coefficient is usually used to reflect the influence
subjective opinion. Therefore, we use DELPHI method to level of the factor. The weighted summation method is used
determine the weight coefficient, which can obtain a to calculate TT&C requirement priority, which is shown in
reasonable result. Formula 20.
1) We select a group of experts based on the satellite
scheduling topic. Once all participants are confirmed, each Ptt &c  8i1 i  tt & cAttri (20)
member is sent a questionnaire with instructions to
comment on each factor. Where Ptt &c represents the TT&C requirement priority,
2) We use statistical method to analyze the scores of each i represents the weight coefficient of each factor,
factor, then send the results to each member for a new tt & c Attri represents the quantitative value of each factor.
round. The questionnaire rounds can be repeated as many 1 Determine the weight coefficient
times as necessary to get the consistent results. Since TT&C system is a large and complex system, and
Step 5: Calculate the Euclidean distances d i and d i- that the management has different preferences for each factor.
between each quantitative value and plus ideal scheme Y  or Therefore, we improve the AHP (Analytic Hierarchy Process)
minus ideal scheme Y - . method to determine TT&C requirement priority, which can
solve inconsistencies in the expert evaluation.
 Step 1: Attribute resolution.
8
d i   j ( yij  1) 2 (16)
j 1 Step 2: Construct the decision matrix.
In the decision matrix, it is synthesized decision maker

8
d i-   j ( yij  0) 2 (17) information with resulted elements in pair compared criteria
j 1
with a normalized and reciprocal scale of relative importance
Step 6: Define the relative distance ci .

(see in Table 14). aij represents the relative importance of
the element i to the element j.
d i TABLE 14
ci  (18) COMPARE CRITERIA IN PAIRS
d  d i
i


Serial number level aij

Step 7: Calculate the the imaging task priority. 1 i and j are equally important 1

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

2 i is slightly more important than j 3 Full-featured station 10

3 i is strongly more important than j 5 Multi-function station 7
4 i is very more important than j 7 Telemetry single receiving station 3
5 i is absolutely more important than j 9 (5) Event
6 i is slightly less important than j 1/3
The quantitative value of event tt & c Attr5 is defined as
7 i is strongly less important than j 1/5
8 i is very less important than j 1/7 follows:
9 i is absolutely less important than j 1/9 TABLE 19
QUANTITATIVE VALUES OF EVENT
Step 3: Consistency test. Event Quantitative value
Since the subjective values cause a degree of uncertainty remote control 10
or lack of reliability, therefore, it is necessary to test the telemetry 8
consistency. Measuring track 6
Single data reception 4
1) Calculate the maximum matrix eigenvalues max ; Voice 2
2) Calculate the consistency index CI, (6) Flight stage
CI  (max  n) /( n  1) ; The quantitative value of flight stage tt & c Attr6 is defined
3) Find the average random consistency index RI (see in as follows:
Table 15). TABLE 20
TABLE 15 QUANTITATIVE VALUES OF FLIGHT STAGE
THE ORDER OF THE JUDGMENT MATRIX (1-10) AND ITS RI VALUE Flight stage Quantitative value
Order 1 2 3 4 5 launch 10
RI 0.00 0.00 0.58 0.90 1.12 Incarnation 8
Order 6 7 8 9 10 early stage 6
RI 1.24 1.32 1.41 1.46 1.49 operation 4
4) Calculate the consistency rate CR, CR  CI / RI . The Recycle 2
consistency is acceptable if CR is less than or equal to 0.1, (7) Available resources
otherwise it must be revised. The quantitative value of available resources tt & c Attr7 is
Step 4: Calculate the weight vector. defined as follows:
First, we calculate normalization eigenvector W TABLE 21
QUANTITATIVE VALUES OF AVAILABLE RESOURCES
corresponding to the maximum matrix eigenvalues max . Number of resources available Quantitative value
Then, we average the elements in all weight vectors. 1 10
2 Calculate the quantitative values of factors 2 8
(1) Source target 3 6
4 4
The quantitative value of source target tt & c Attr1 is defined
5 2
as follows: ≥6 1
tt & c Attri  Prt (21) (8) Execution urgency
The quantitative value of execution urgency tt & c Attr9 is
(2) Control circle defined as follows:
The quantitative value of source target tt & c Attr2 is defined durtt &c
tt & c Attr9  10 * (22)
as follows: ettt &c  ttt &c
TABLE 16
QUANTITATIVE VALUES OF CONTROL CIRCLE Where ettt &c represents the deadline for the TT&C
Control circle Quantitative value requirement, ttt &c represents the current moment,
Departure circle 10 and durtt &c represents the minimum duration.
Entry circle 10
Middle circle 5
D. DATA TRANSMISSION REQUIREMENT PRIORITY
(3) TT&C resource
The ground station resources are assigned to the data
The quantitative value of TT&C resource tt & c Attr3 is
transmission tasks in descending order, this order is
defined as follows:
TABLE 17
determined by data transmission requirement priority,
QUANTITATIVE VALUES OF RESOURCE which is composed of multiple indicators, such as the data
TT&C resource Quantitative value quality, the data processing ability, and so on. We study the
Fixed ground station 10 various factors that affect the data transmission requirement
Moving ground station 7
Marine survey ship 3
priority in the academic field, and consider the optimal
Relay satellite 1 objectives and constraints in the actual engineering. The
(4) Service capability factors are shown in Table 22.
TABLE 22
The quantitative value of service capability tt & c Attr4 is FACTORS THAT AFFECT DATA TRANSMISSION REQUIREMENT PRIORITY
defined as follows: Factors
TABLE 18 DTR1 Source target
QUANTITATIVE VALUES OF SERVICE CAPABILITY
DTR2 Way
Service capability Quantitative value

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

DTR3 Data receiving station defined as follows:

DTR4 Downstream data type TABLE 26
QUANTITATIVE VALUES OF AVAILABLE RESOURCES
DTR5 Available resources
Available resources Quantitative value
DTR6 Execution urgency
1 10
Each factor in Table 22 has different influence on data 2 8
transmission requirement priority, therefore, it needs to be 3 6
assessed. The weight coefficient is usually used to reflect the 4 4
influence level of the factor. The weighted summation 5 2
≥6 1
method is used to calculate data transmission requirement
(6) Execution urgency
priority, which is shown in Formula 23.
The quantitative value of execution urgency dtrAttr6 is
Pdtr   i61 i  dtrAttri (23) defined as follows:

Where Pdtr represents the data transmission requirement durdtr

dtrAttr6  10 * (25)
priority, i represents the weight coefficient of each factor, etdtr  tdtr
dtrAttri represents the quantitative value of each factor.
Where durdtr represents the minimum Duration, etdtr
1) DETERMINE THE WEIGHT COEFFICIENT represents the latest ending time, and tdtr represents the
We use the AHP method and Delphi method in Section C to current moment.
determine the weight coefficient.
2) CALCULATE THE QUANTITATIVE VALUES Ⅴ. TASK SCHEDULING ALGORITHMS FOR SATELLITE
(1) Source target SCHEDULING
The quantitative value of source target dtrAttr1 is defined as A. BASIC IDEA OF THE TASK SCHEDULING
follows: ALGORITHMS
Compared to the fixed mode of normal tasks, Emergency
dtrAttri  Prt (24)
tasks arrive at the system in a dynamic mode. Therefore,
(2) Way real-time or near real-time scheduling algorithm is more
The quantitative value of way dtrAttr2 is defined as follows: effective for emergency tasks. Dynamic task scheduling
TABLE 23 algorithm is proposed based on RHO strategy. The RHO
QUANTITATIVE VALUES OF WAY strategy is designed with a periodical triggering mode, and
Way Quantitative value the scheduling process is decomposed into a series of static
Real-time transmission 10
Store and Forward 5
scheduling intervals, the dynamic scheduling of emergency
(3) Data receiving station tasks is realized by optimizing the scheduling schemes in
The quantitative value of data receiving station dtrAttr3 is each interval.
defined as follows:
TABLE 24 B. RHO STRATEGY
QUANTITATIVE VALUES OF RECEIVING STATION 1) ROLLING WINDOW
Data receiving station Quantitative value
Station A 10
In RHO strategy, the selection of the rolling window is very
Station B 9 important, which determines the response of the scheduling
Station C 7 system to the emergency tasks. The rolling window is
Station D 5 mainly used to store the currently scheduling tasks, which
Station E 3
includes normal tasks and emergency tasks.
Station F 2
No requirement 1 2) PERIODIC TRIGGERING MODEL
(4) Downstream data type (1) Uniform fixed mode
The quantitative value of downstream data type dtrAttr4 is As shown in Figure 3, the task priority changes does not
defined as follows: depend on any other external behavior, it only related to its
TABLE 25 own rolling cycle. It is important to select the fixed cycle,
QUANTITATIVE VALUES OF DOWNSTREAM DATA TYPE
the shorter the fixed cycle, the better the scheduling result
Downstream data type Quantitative value
Optical 10 is.
electronic 8 (2) Dynamic changing mode
radar 5 As shown in Figure 4, the task priority changes with
(5) Available resources rolling rescheduling of the emergency tasks.
The quantitative value of available resources dtrAttr5 is

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

FIGURE 3. Uniform fixed mode

FIGURE 4. Dynamic changing mode

New tasks

Input: Return
Ttasks

Priority rolling change Emergency task rolling rescheduling

External conditions

Calculate task priority Realculate current task priority and start

the solving algorithm
Waiting

Yes
External conditions

Yes No

Whether trigger Whether trigger task

priority rolling rolling rescheduling
change

Waiting
No
Output：
Output：
Unscheduled tasks
Scheduling schemes
Store current
task priority

FIGURE 5. Algorithm framework

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

(3) Algorithm framework 1 Individual coding

As shown in Figure 5, the algorithm framework includes Satellite imaging task scheduling is actually a two-level
two parts: priority rolling change module and task rolling decision problem, therefore the two-stage coding method is
rescheduling module. applied for all individuals in the Genetic algorithm, which
1) In the priority rolling change module, first, we use the is recorded as the task execution sequence segment and the
task priority model to calculate the priority of scheduling satellite resource allocation segment.
tasks. Then, if the scheduling tasks trigger the condition of •Task execution sequence segment
priority rolling change, we recalculate the priority; if not, Random number coding method is applied for task
we temporarily store the current task priority. execution sequence segment. First, we assume a set of
2) In the task rolling rescheduling module, if the tasks Task  T1 , T2 .....Tn  .Then, random numbers between 0
scheduling tasks trigger the condition of task rolling and 1 are generated , which are corresponding to each task
rescheduling, we assign the resources and time windows to (see in Figure 6). During the process of the algorithm, the
the tasks and form the scheduling schemes. The task sequence corresponding to Task remains unchanged,
unscheduled tasks will be scheduled with the next group of but it has multiple random number sequences. Therefore,
tasks. each random number sequence can be regarded as an
3) HEURISTIC ALGORITHM individual, and the position corresponding to each random
To achieve the two objectives and find the optimal solutions, number in the random number sequence can be regarded as
we propose a Heuristic algorithm based on the one gene position, and the random value is the value of the
aforementioned algorithm framework. The specific solution gene.
process is described as follows:
Step 1: Initialize the tasks, and calculate each task
priority.
Step 2: If entering a new rolling rescheduling window,
the task rolling rescheduling is started, then go to Step 3; if
not, continue to wait for the rolling window, the task sets
will be updated continuously as the new tasks added. Figure 6 . task execution sequence segment coding
Step 3: Recalculate the tasks priority based on the task In this paper, we sort the random numbers in descending
priority model in section Ⅳ. order, which is regarded as the order of satellite resources
Step 4: Sort the tasks by their priority in descending and execution time for the tasks. If the random numbers
order. corresponding to the two tasks are the same, satellite
Step 5: WCI (Windows Conflict Index) method is resources and execution time are assigned to the task
applied to select the resources and time windows,and assign preferentially, which is ahead of another task. We can get
them to the tasks according to the order. the task execution order through decoding the individual in
Step 5.1: Calculate all visible time windows between the Figure 6.
tasks and their available resources.
Step5.2: Calculate the windows conflict index for T4→T2→T3→T1→T5
taski according to the order. •Satellite resource allocation segment
Step 5.2.1: Assign task i to the visible time windows TWk , the satellite resource allocation segment coding is
and determine the relevant resource S j . determined by the corresponding task execution sequence
Step 5.2.2: Take out all tasks after taski , and determine segment. First, all time windows and satellite resources are
all the visible time window sets {TWs j } on the resource S j . assigned to the tasks, which belong to a certain gene
Step 5.2.3: Calculate number of time windows in position in the task execution sequence segment. Then, an
conflict between {TWs j } and the time window TWk . element is selected randomly to be the value of the gene in
Step 5.2.4: If all time windows for taski have been the satellite resource allocation segment. As shown in
traversed, then go to Step5.3; if not, then go to Step 5.2.1. Figure 6, we assume that there are 3, 2, 5, 4, and 7 time
Step 5.3: Select the time window with the least number windows and corresponding satellite resources for the tasks
of conflicts and its corresponding resources for this task. {T1,T2,T3,T4,T5}, respectively, a satellite resource
Step 5.4: Update the visible time of all tasks after taski . allocation segment coding is shown in Figure 7, it can
Step 5.5: If all the tasks are scheduled, then go to Step 6; decode that the tasks {T1,T2,T3,T4,T5} are scheduled on
if not, then go to Step 5.2. their 2nd, 1st, 3rd, 2nd, and 6th time windows and
Step 6: Update the scheduling scheme. corresponding satellite resources.
4) GENETIC ALGORITHM
(1) Main operators
The main operators of the Genetic algorithm are as
follows:

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

•Satellite resource allocation segment

For the satellite resource allocation segment, the value of
each gene represents the execution time window and
corresponding satellite resources for the tasks, in order to
avoid unreasonable gene coding, we adopt the two-point
Figure 7 . satellite resource allocation segment coding crossover method. As shown in Figure 9 ， First, two
2 initial population crossover sites are randomly generated, and two parent
Initial population is broken down into two parts based on individuals are simultaneously divided into three segments
the above two-level structure of individual coding. First, all with the same length, and swap their positions. At the same
individual satellite resource allocation segment coding are time, we adopt the one-point mutation method.
initialized according to the procedure , which is described
in the second part above. The task execution sequence
segment coding is initialized as follows.
•Ruler 1
In the initial population, First, the tasks are in increasing
order for earliest start time. Then, the random numbers
between 0 and 1 are assigned to each gene position for half
of the individuals. The same random numbers are assigned
to the tasks that the earliest start time is the same.
•Ruler 2
In the initial population, the residual individual coding
randomly generated follow the process, which described in
the first part.
3 crossover and mutation
•Task execution sequence segment
For the task execution sequence segment, the random
numbers size on each gene only represent task execution
order, therefore we adopt the one-point crossover method.
As shown in Figure 8 ，First, a crossover site is randomly
generated, and two parent individuals are simultaneously
divided into two segments with the same length, and swap Figure 9. Crossover and mutation in Satellite resource allocation
their positions. At the same time, we adopt the one-point segment
mutation method. A random number between 0 and 1 is (2) Solution process
generated to replace the original value in the gene, which is The specific solution process of the Genetic algorithm is
randomly selected in the parent individuals. described as follows:
Step 1: Initialize the tasks and calculate each task
priority.
Step 2: If entering a new rolling rescheduling window,
the task rolling rescheduling is started, then go to Step3; if
not, continue to wait for the rolling window, the task sets
will be updated continuously as the new tasks added.
Step 3: Recalculate the tasks priority based on the task
priority model.
Step 4: Initialize the populations.
Step 5: WCI method is applied to assign the resource
and time windows to the tasks according to the order, and
calculate the fitness values.
Step 6: The parent populations generate the offspring
populations through selection, crossover and mutation.
Step 7: If the termination condition is reached, then go to
step 8; if not, then go to step 5.
Step 8: Update the scheduling scheme.
5) NSGA-Ⅱ ALGORITHM
(1) Main operators
Figure 8 . Crossover and mutation in task execution sequence segment

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

The main operators of the NSGA- Ⅱ algorithm are as follows:

follows: 1) pre_prob1 is the user's preference coefficients for the
1 Fast non-dominated sorting method objective f g (s ) , pre_prob2 is the user's preference
First, we find out the non-dominated solution sets in the coefficients for the objective f c (s ) , and
group, which are called the first level F1. The non- pre_prob1  pre_prob2  1 .
dominant rank irank  1 is assigned to all the individuals, and 2) The user's preference point ( f1 pre , f 2pre ) is calculated
delete it from the whole population. Then, we continue to based on the ideal point ( f1* , f 2* ) , which is in the optimal
find out the non-dominated solution sets in the remaining solution set P - O set .
groups, which are called the second level F2. The non-
dominant rank irank  2 is assigned to all the individuals. f1*  max{ f i f i  P - O set}, i  1,2 (27)
Follow this cycle until the entire population is stratified, the
f i pre  f i * pre _ probi , i  1,2 (28)
individuals have the same non-dominated rank irank , which
are in the same level. The Euclidean distances are calculated
2 Calculate the crowding distance between ( f1 pre , f 2pre ) and all solutions in P - O set , the
The crowding distance calculation is improved to keep minimum distance corresponding to the optimal solution is
the population diversity. selected to get the scheduling scheme Pi * .
•Method 1 (2) Solution process
If the number of optimal solutions is less than 10, The specific solution process of the NSGA-II
the repetitions of each individual satellite resource ALGORITHM is described as follows:
allocation segment coding in the population are employed Step 1: Initialize the tasks, and calculate each task
to calculate the crowding distance. The specific formula is priority.
as follows: Step 2: If entering a new rolling rescheduling window,
nPop  T [i ]SRassignment the task rolling rescheduling is started, then go to Step3; if
I [i ]distance  (26) not, continue to wait for the rolling window, the task sets
nPop
will be updated continuously as the new tasks added.
where I [i ]distance is the crowding distance for individual Step 3: Recalculate the tasks priority based on the task
i, nPop is the size of current population, T [i ]SRassignment is the priority model.
same number of individuals, which are in current population Step 4: Initialize the populations and conduct non-
and satellite resource allocation segment coding. dominated sorting.
•Method 2 Step 5: WCI method is applied to assign the resource
In addition, the traditional operator in the NSGA- Ⅱ and time windows to the tasks, calculate the objective value,
algorithm is employed to calculate the crowding distance. and generate the dominance between candidate solutions.
The calculation process of crowding distance for the point n Step 6: The first offspring populations are generated
is shown in Figure 10, a rectangle is constructed, whose through selection, crossover and mutation.
diagonal point is the point n-1 and the point n+1, the Step 7: Use the fast non-dominated sorting method,
crowding distance for the point n is the sum of length and meanwhile, calculate the crowding distance for each
width. Pareto optimality in Figure 10 represents the optimal individual, select appropriate individuals to form generate
solution. parent populations.
Step 8: Generate new offspring populations through
Genetic algorithm;
Step 9: If the termination condition is reached, then go to
step 10; if not, then go to step 7.
Step 10: Update the scheduling scheme.

Ⅵ. NUMERICAL EXPERIMENTS AND PERFORMANCE

ANALYSIS

A. DESCRIPTION OF FUNCTIONS AND PARAMETERS

1) SATELLITE DESIGN
Figure 10. Crowding distance for individual n In order to reduce the complexity of the problem, the
3 Demand preference imaging targets are set as the static point targets, which are
In this paper, if the NSGA-II algorithm is employed to randomly generated in the global region. All satellites in
solve the satellite scheduling, multiple scheduling schemes the scenarios are set as AW stars, the orbital height is set as
are generated. We must select the optimal solution based
on the user's demand preference. The specific process is as

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

800 km, the number of satellites is set as 6, the satellite

orbit parameters are shown in Table 28.
TABLE 27
SATELLITE PARAMETERS
Semi major axis Inclination Argument of Right ascension of Trueanomaly
Number Eccentricity
(km) (deg) perigee(deg) ascending node(deg) (deg)
Sat1 7171.393 0 96.576 0 175.72 0.075
Sat2 7171.393 0 96.576 0 145.72 30.075
Sat3 7171.393 0 96.576 0 115.72 60.075
Sat4 7171.393 0 96.576 0 85.72 90.075
Sat5 7171.393 0 96.576 0 55.72 120.075
Sat6 7171.393 0 96.576 0 25.72 150.075
TABLE 28
COMMON PARAMETERS OF GENETIC ALGORITHM AND NSGA-Ⅱ ALGORITHM
S Population Generations Crossover rate Mutation rate
S  50 10 20
50  S  100 20 40
0.8 0.1
100  S  200 30 60
200  S 40 80
TABLE 29
EXAMPLES OF IMAGING TASK
task1 task 2 task 3 task 4 task 5
longitude 130.25 -50.65 78.93 -56.35 -110.58
latitude 60.25 30.17 -17.35 -54.63 63.12
Source target priority 7.3 6.7 5.1 4.9 5.8
Application scenario Scenario A Scenario B Scenario C Scenario D Scenario E
User User A User B User A User C User A
Application Application A Application A Application B Application A Application B
Satellite working mode Single star Single star Single star Single star Single star
Type Emergency Normal Emergency Emergency Normal
Satellite's properties AW stars AW stars AW stars AW stars AW stars
flight stage Operation Operation Operation Operation Operation
Execution urgency 0.68 0.56 0.81 0.33 0.57

2) SCENARIO DESIGN the number of original emergency tasks is set as 100, the
(1) Scenario for the task priority model and the RHO arrival rate of emergency tasks is set as 15/hour, 30/hour,
strategy 60/hour, 90/hour, 120/hour, respectively. The simulation
To further verify the feasibility of the priority model and period is set as 2004/06/01 00:00:00~2004/06/02 00:00:00.
the RHO strategy in solving the problem of satellite Meanwhile, the common parameters of NSGA-II
scheduling, we design six simulation scenarios SCEN_1~ algorithm and Genetic algorithm are shown in Table 28.
SCEN_6, each scenario has a corresponding satellite The preference coefficients for both objectives are set as
Sat1~Sat6. The number of normal tasks is set as 700, and 0.5.
the number of original emergency tasks is set as 300. (3) Test task sets
First, we conduct three guiding experiments to improve To better verify the experiment, the normal tasks
the efficiency of experiments. The result shows that the library Nor _ Task and the emergency tasks
NSGA-II algorithm and Genetic algorithm can only library Emer _ Task are created, including the number of
schedule all the normal tasks over one day when the normal tasks 4000, and the number of emergency tasks
number of tasks more than 100. Therefore, the Heuristic 6000, respectively. The deadline of normal tasks is
algorithm is selected to verify the feasibility of the priority randomly chosen from 2004/06/02 00:00:00 to 2004/06/04
model and the dynamic changing strategy. The arrival time 00:00:00, and the deadline of emergency tasks is randomly
interval of the emergency task is set as one minute, period chosen from the arrival time to 2004/06/03 00:00:00. Some
for the emergency task rolling scheduling driven is set as imaging tasks are shown in Table 29.
one hour. The simulation period is set as 2004/06/01
00:00:00~2004/06/02 00:00:00. B EXPERIMENTAL RESULTS AND ANALYSIS
(2) Scenario for the task scheduling algorithms The algorithms are programmed in Matlab R2013a, and the
To further verify the feasibility of three priority experiments are implemented on a PC with an Intel(R)
algorithms in solving the problem of satellite scheduling, Core(TM) i5-3210M with 2.50 GHz CPUand 4 GB of
we design five simulation scenarios SCEN_7~SCEN_11, RAM.
each scenario has six corresponding satellites. Meanwhile, (1) The task priority model and RHO strategy

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

We conduct 12 sets of comparative experiments in

simulation scenarios SCEN_1~SCEN_6, the parameters are
given as follows.
• DP : The dynamic priority strategy
• SP : The static priority strategy
• NTotal _ init : The number of original normal tasks
• ETotal _ init : The number of original emergency tasks
• Total : The number of original tasks
• SUM _ init : The number of scheduled tasks in the
normal task scheduling stage
• NC _ init : The number of scheduled normal tasks in the
normal task scheduling stage Figure 11. Scheduling rate of tasks

• EC _ init : The number of scheduled emergency tasks in From Table 30, we can see that the number of normal
the normal task scheduling stage tasks less than the number of emergency tasks, meanwhile,
• NR _ init : The scheduling rate of normal tasks in the the of normal tasks less than the scheduling rate of
normal task scheduling stage, the formula is as follows: emergency tasks. On the other hand, the experimental results
in SCEN_1~SCEN_6 are generated by the priority in
NR _ init  NC _ init / NTotal (29) descending order. Therefore, the task priority model is
• ER _ init : The scheduling rate of emergency tasks in the effective, it can ensure that most emergency tasks is high
normal task scheduling stage, the formula is as follows: than the normal tasks in priority , which is consistent with
actual engineering.
ER _ init  EC _ init / ETotal (30) Comparing with the static priority strategy, the RHO
• SUM : The number of scheduled tasks strategy has better scheduling performance in the number of
• NC : The number of scheduled normal tasks scheduled tasks, and the improved tasks are all the
• EC : The number of scheduled emergency tasks emergency tasks. Therefore, the RHO strategy is effective.
TABLE 30 (2) The task scheduling algorithm
EXPERIMENTAL RESULTS OF SCEN_1~SCEN_6 NSGA-II algorithm and Genetic algorithm are run for 10
Scenarios 1 2 3 4 5 6 times in simulation scenarios SCEN_7~SCEN_11, the
NTotal _ init 700 iterative process of a period is shown in Figure 12 and
ETotal _ init 300 Figure 13. The parameters are given as follows.
Total 2400 • Total : The number of original tasks
SUM _ init SP 75 70 65 73 72 69
• Velo : The arrival rate of tasks
DP 76 72 73 73 72 70
• NT : The number of scheduled tasks
SP 5 4 7 8 5 7
NC _ init • NR : The scheduling rate of tasks
DP 5 3 7 7 4 6
SP 70 66 58 65 67 62
EC _ init NR  NT / Total (31)
DP 71 69 66 66 68 64
NR _ init
SP 0.007 0.005 0.010 0.011 0.007 0.01
DP 0.007 0.004 0.010 0.010 0.005 0.009 • ExeTime : The processing time of the algorithms
SP 0.23 0.22 0.19 0.21 0.22 0.20 （minute）
ER _ init
DP 0.24 0.23 0.22 0.22 0.23 0.21
SP 1344 1357 1349 1347 1355 1349
SUM
DP 1346 1360 1370 1361 1363 1351
SP 258 254 252 245 248 258
NC
DP 252 251 251 254 249 252
SP 1086 1103 1097 1102 1107 1091
EC
DP 1094 1109 1119 1107 1114 1099

Figure 12. Iterative process of NSGA-II algorithm

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

TABLE 32
OBJECTIVES OF ALGORITHMS
Heuristic Genetic NSGA-Ⅱ
algorithm algorithm algorithm
f g (s ) f c (s ) f g (s ) f c (s ) f g (s ) f c (s )
7 14642 0.582 15066 0.560 14313 0.732
8 26974 0.553 29298 0.627 27127 0.854
9 55411 0.664 57236 0.668 55913 0.879
10 91879 0.716 92657 0.752 87401 0.909
11 129372 0.795 134424 0.811 127800 0.938
Comparing with NSGA-II algorithm and Genetic
algorithm, Heuristic algorithm has a poor performance in
the revenue of the scheduled tasks and satellite resource
load balance. Genetic algorithm has a better performance in
Figure 13 Iterative process of Genetic algorithm the revenue of the scheduled tasks, but far lower than
TABLE 31
NSGA-II algorithm in the satellite resource load balance.
EXPERIMENTAL RESULTS OF SCEN_7~SCEN_11
Scenarios 7 8 9 10 11 Therefore, NSGA-II algorithm is superior to Genetic
Total 460 820 1540 2260 2980 algorithm and Heuristic algorithm in terms of the optimal
Velo 15 30 60 90 120 objectives.
HEURISTIC NT 347 591 1138 1791 2518
ALGORITHM NR 0.754 0.721 0.739 0.792 0.845
ExeTime 0.27 3.69 7.47 11.90 16.48 Ⅶ. CONCLUSIONS AND FURTHER RESEARCH
Velo 15 30 60 90 120 In this paper, we have have investigated the task
GENETIC NT 350 580 1158 1791 2522 scheduling in satellite scheduling system, and meanwhile
ALGORITHM NR 0.761 0.707 0.752 0.792 0.846 present a task priority model and dynamic task scheduling
ExeTime 2.90 5.42 12.93 36.25 79.32
Velo 15 30 60 90 120
algorithms based on RHO strategy to optimize the
NSGA-II NT 345 549 1138 1771 2507 scheduling performance under the condition of various
ALGORITHM NR 0.750 0.724 0.739 0.783 0.841 dynamic disturbance factors.
ExeTime 2.91 5.51 15.10 33.84 83.73 (1) The task priority model is reasonable and effective.
The simulation results in Table 31 show that the (2) The RHO strategy is effective, it can improve the
scheduling rate of each algorithm is higher than 70%. Note number of completed tasks.
that in the case of the arrival rate of emergency tasks is (3) Heuristic algorithm has a better performance in
120/hour, the scheduling rate is higher than 80%. processing time, Genetic algorithm has a better
On the other hand, It is obviously that Heuristic performance in the revenue of the scheduled tasks, but far
algorithm has a better performance than NSGA-II algorithm lower than NSGA-II algorithm in the satellite resource load
and Genetic algorithm in processing time. The processing balance, and NSGA-II algorithm is superior to Genetic
time of NSGA-II algorithm and Genetic algorithm algorithm and Heuristic algorithm in terms of the optimal
increases largely with the increasing arrival rate of objectives.
emergency tasks. According to the analysis and In further research, more factors will be considered in the
simulations, Heuristic algorithm utilizes low processing task priority model. At the same time, Heuristic algorithm,
time to schedule the imaging tasks. Genetic algorithm, and NSGA-II algorithm are traditional
optimization algorithms. Therefore, we will try a new
algorithm, and consider three or more objectives.

REFERENCES
[1] Deng, Boyu et al., "Preemptive dynamic scheduling algorithm for
data relay satellite systems." 2017 IEEE International Conference
on Communications (ICC). IEEE, 2017.
[2] Sun, Youcheng and G. Lipari. "Response Time Analysis with
Limited Carry-In for Global Earliest Deadline First
Scheduling." Real-time Systems Symposium, 2016.
[3] Ali, Ahmed A et al., "An Efficient Scheduling Scheme for
OFDMA Resource Blocks with Joint User Scheduling based on
Earliest Deadline First with Carrier Aggregation (CA) in LTE-A
Figure 14. Processing time of algorithms System." Wireless Personal Communications., vol. 88, no. 2, pp.
173-183, 2016.
In addition to the scheduling rate of tasks and processing [4] Xia, Qin, D. Zhu, and H. Aydin. "Work-in-Progress: Preference-
time, the optimal objectives should be employed to verify Oriented Scheduling in Multiprocessor Real-Time Systems." 2018
the feasibility of the algorithms. The simulation results in IEEE Real-Time Systems Symposium (RTSS), 2019.
Table 32 show that the value of two optimal objectives [5] Ren, Xiao Xi, and G. Y. Zhao. "Least Slack First Schedule
Algorithm Based on Dynamic Preemption Threshold." Computer
increases with the increasing arrival rate of emergency tasks. Engineering., vol. 38, no. 4, pp. 275-275, 2012.

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

[6] Wang, Shuju, T. Zhang, and G. Zhang. "Scheduling of vehicle cyanotoxin content from experimental cyanobacteria
body CAN-Bus with event-triggered least slack time first concentrations in the Trasona reservoir (Northern Spain)." Journal
technique." Intelligent Control & Automation, 2010. of Computational & Applied Mathematics., vol. 309, no.1, pp. 587-
[7] Maesano, M et al., "First mapping of the main high conservation 602, 2017.
value forests (HCVFs) at national scale: The case of [23] Huang, Kuo-Chan, Ying-Lin Tsai, and Hsiao-Ching Liu. "Task
Italy." Giornale Botanico Italiano., vol. 150, no. 2, pp. 208-216, ranking and allocation in list-based workflow scheduling on
2016. parallel computing platform." The Journal of
[8] Wang, Li et al., "One‐Step Reforming of CO2 and CH4 into Supercomputing .,vol.71, no.1, pp.217-240, 2015.
High‐Value Liquid Chemicals and Fuels at Room Temperature by [24] Wang, Jianqin Wang Jianqin et al., "A new algorithm for grid
Plasma‐Driven Catalysis." Angewandte Chemie., vol. 56, no. 44, independent task schedule: Genetic simulated
pp. 13679-13683, 2017. annealing." Computer & Computing Technologies in Agriculture-
[9] Burkert, Andreas, M. R. Bate, and P. Bodenheimer. "Protostellar ⅰ- Third Ifip International Conference on Computer & Computing
fragmentation in a power-law density distribution." Monthly Technologies in Agriculture 2009.
Notices of the Royal Astronomical Society., vol. 289, no. 3, pp. [25] Wei, Jiang , and P. Xiuli . "Multi-agent based dynamic mission
497-504, 2018. planning for satellite data transmission." International Conference
[10] Biswas, Sohag, D. Chakraborty, and B. S. Mallik. "Interstitial on Natural Computation IEEE, 2016.
Voids and Resultant Density of Liquid Water: A First-Principles [26] Zhan, Shi-hua et al., "List-based simulated annealing algorithm for
Molecular Dynamics Study." ACS Omega., vol. 3, no. 2, pp. 2010- traveling salesman problem." Computational intelligence and
2017, 2018. neuroscience, pp. 1-12, 2016.
[11] Burkert, Andreas, M. R. Bate, and P. Bodenheimer. "Protostellar [27] Shi, Jinhua, Hui Lu, and Kefei Mao. "Solving the test task
fragmentation in a power-law density distribution." Monthly scheduling problem with a genetic algorithm based on the scheme
Notices of the Royal Astronomical Society., vol. 289, no. 3, pp. choice rule." in International Conference on Swarm Intelligence.
497-504, 2018. Springer, Cham, pp. 19-27, 2016.
[12] Semghouni, Samy et al., "On new scheduling policy for the [28] Xu, Yuming et al., "A genetic algorithm for task scheduling on
improvement of firm RTDBSs performances." Data & Knowledge heterogeneous computing systems using multiple priority
Engineering .,vol. 63, no. 2, pp. 414-432, 2007. queues." Information Sciences .,vol. 270, pp. 255-287, 2014.
[13] Chantem, Thidapat et al., "Period and Deadline Selection for [29] Delice, Yılmaz et al., "A modified particle swarm optimization
Schedulability in Real-Time Systems." Euromicro Conference on algorithm to mixed-model two-sided assembly line
Real-time Systems IEEE, 2008. balancing." Journal of Intelligent Manufacturing.,vol. 28, no.1, pp.
[14] Balbastre, Patricia, Ismael Ripoll, and Alfons Crespo. "Minimum 23-36, 2017.
deadline calculation for periodic real-time tasks in dynamic priority [30] Nouiri, Maroua et al., "An effective and distributed particle swarm
systems." IEEE Transactions on computers.,vol. 57, no. 1, pp. 96- optimization algorithm for flexible job-shop scheduling
109, Jan. 2008. problem." Journal of Intelligent Manufacturing., pp. 1-13, 2018.
[15] Jin, et al., "An integrated design method of task priority." Journal [31] Kai, C. H. , Y. Xiao , and Q. Fang . "Relay satellite scheduling
of Software .,vol. 14, no. 3, pp. 376-382, 2003. based on Artificial Bee Colony algorithm." International
[16] Burns, Alan et al., "The meaning and role of value in scheduling Symposium on Wireless Personal Multimedia
flexible real-time systems." Journal of systems architecture., vol. Communications.IEEE, 2015.
46, no. 4, pp. 305-325, 2000. [32]Deng, Boyu, et al. "Two-phase task scheduling in data relay satellite
[17] Buttazzo, Giorgio, Marco Spuri, and Fabrizio Sensini. "Value vs. systems." IEEE Transactions on Vehicular Technology., vol. 67,
deadline scheduling in overload conditions." in Proceedings 16th no.2, pp. 1782-1793, 2018.
IEEE Real-Time Systems Symposium. IEEE, pp. 90-99,2015. [33]Xiaolu, Liu, et al. "Multi satellites scheduling algorithm based on
[18] Aldarmi, Saud A. , and A. Burns . "Dynamic Value-Density for task merging mechanism." Applied Mathematics and
Scheduling Real-Time Systems." Euromicro Conference on Real- Computation., vol. 230, no.2, pp. 687-700, 2014.
time Systems IEEE, 1999. [34] Keshanchi, Bahman, Alireza Souri, and Nima Jafari Navimipour.
[19] Khomri, Bilal, et al. "Retinal blood vessel segmentation using the "An improved genetic algorithm for task scheduling in the cloud
elite-guided multi-objective artificial bee colony algorithm." IET environments using the priority queues: formal verification,
Image Processing., vol. 12, no.12, pp. 2163-2171, 2018. simulation, and statistical testing." Journal of Systems and
[20] Gao, W. F et al., "Artificial Bee Colony Algorithm Based on Software .,vol. 124, pp.1-21, 2017.
Information Learning. " IEEE Transactions on Cybernetics., vol.
45, no.12, pp. 2827-2839, 2017.
[21] Yang, Liran, and Q. Zhou. "An improved artificial bee colony
algorithm for solving constrained optimization
problems." International Journal of Machine Learning &
Cybernetics., vol. 8, no.3, pp. 1-16, 2017.
[22] Nieto, P. J. García et al., "A hybrid wavelet kernel SVM-based
method using artificial bee colony algorithm for predicting the

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 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/ACCESS.2019.2928992, IEEE Access

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.