An emergency task autonomous planning method of agile imaging satellite
Song et al. EURASIP Journal on Image and Video Processing
An emergency task autonomous planning method of agile imaging satellite
Yanjie Song 0
Danjie Huang 2
Ziyu Zhou 1
Yingwu Chen 0
0 College of Systems Engineering, National University of Defense Technology , Changsha 410073, Hunan , China
1 Department of Industrial and Enterprise Systems Engineering, University of Illinois at Urbana-Champaign , 117 Transportation Buld., 104 S. Mathews Ave., MC-238, Urbana, IL 61801-3080 , United States
2 School of Science and Engineering & School of Management and Economics, Chinese University of Hong Kong , Shenzhen 518172, Guangdong , China
As the number of satellite emergency imaging tasks grows, the main goal of satellites becomes putting forward solutions and meeting users' demands in a relatively short time. This study aims to investigate the problem of emergency task planning for agile satellites. Through the analysis of the problem and its constraints, a model of emergency task autonomous planning was implemented. According to the characteristics of emergency tasks, a strategy to deal with the tasks of different emergency levels and various quantities was proposed. We put forward three algorithms for quick insertion of emergency tasks, i.e., emergency task insertion algorithm (ETIA), general emergency task insertion algorithm (GETIA), and general emergency task planning &insertion algorithm (GETPIA). The experimental result showed that the strategy and algorithms can not only respond quickly to observation tasks but also produce effective planning programs to ensure the successful completion of observation tasks.
Emergency task; Autonomous planning; Agile imaging satellite; Replanning
As the satellite imaging technology develops in recent
years, people have become increasingly dependent on it.
For example, China has its own commercial remote
sensing satellite constellation, SuperView-1.
SuperView1 consists of four optical satellites with the resolution of
0.5 m. It will provide customers with remote sensing
data services for multiple fields. Figure 1 is a picture
taken by it. There are a large amount of images that
need to be completed every day. With the emergence of
increasingly strict and meticulous requirements of users
as well as some temporary demands, the traditional
satellite management and control modes that are highly
dependent on the commands from the ground have
experienced difficulties in meeting the needs of future
development of space technology [
]. Such situation is
reflected in the patterns of the traditional model. Due to
the limitation of communication time between ground
stations and satellites, it is very difficult for satellites to
respond in time to the temporary observation demands
of users or to obtain information about some emergency
events, resulting in a lag in task completion . Even if
the tasks could be completed in time, the imaging
quality would be difficult to be guaranteed. Therefore, the
most effective way to reduce the response time of
satellites as well as complete all kinds of emergency tasks of
high quality without delay is to rely on the satellites to
make the task plans and propose solutions through
The satellite image acquisition is an important
scheduling problem in satellite mission planning [
is especially critical as the number of imaging tasks
increases. The autonomous replanning can effectively solve
the problem of limited communication time between
ground and satellite, so as to improve the utilization
efficiency of satellite resources [
]. The CASPER
(Continuous Activity Scheduling Planning Execution and
Re-planning) of NASA has successfully made the E0-1
satellite reprogram online by using a local search algorithm
for iterative repair . The FireBIRD of DLR (German
Aerospace Agency) has carried a VAMOS (Autonomous
Task Planning On-board a Spacecraft) platform, which
uses real-time constraint checking to reprogram tasks and
adjust the follow-up program in time [
]. The AGATA
(Autonomy Generic Architecture-Test and Application)
platform of CNES has adopted the task triggering
mechanism, which can respond quickly to the tasks [
et al. [
] have proposed a set of visualized satellite
autonomous task planning system and demonstrated the
feasibility of the system, but the efficiency of the system
has not been discussed. Chu et al. [
] have used branch
demarcation to effectively solve the real-time scheduling
problem of the dual agile satellite system, and the
autonomous scheduling method of clustering is likely to be a
research direction in the future. Some other studies have
analyzed the online scheduling problem from the
theoretical level, developed the scheduling framework, and
proposed corresponding methods [
]. These studies have
taken into account the online scheduling of routine
observation tasks, but they paid less attention to the
reprogramming of emergency tasks [
]. Additionally, the
realtime dynamic scheduling is more often done offline, so it
is difficult to respond to emergency tasks [
The current study has little research on the replanning
and completion process of emergency tasks and lacks a
method for generalizing problems. Based on previous
studies, we proposed a generalized model for handling
emergency task replanning problem.
This paper will discuss and analyze the difficulties in
the emergency observation tasks as well as put forward
the task planning model of the emergency tasks based
on various constraints. In the end, the paper will put
forward a combined strategy to cope with several kinds of
tasks and design the corresponding algorithm.
The rest of the article is structured as follows. The
second part will propose the model of the emergency task
planning and specify the corresponding restriction
conditions. Then, the strategies and algorithms to solve this
problem will be presented in the third part. In the fourth
part, the feasibility and effectiveness of the proposed
method will be verified by an example. Finally, the
conclusion is given in the fifth part.
2 Problem description and model
The observation demands of emergency tasks refer to a
series of demands that the imaging satellite needs to
complete the observing tasks due to the temporary
needs of users or some emergencies. Compared with the
general tasks, the emergency tasks have a higher priority
to be completed. However, the insertion of emergency
tasks inevitably conflicts with the original task plan;
thus, the problem arises.
2.1 Problem description
We translated the agile satellite task planning problem
into mathematical models so as to design algorithms
and solve them. In the following, we described the
problem as well as proposed models and constraints based
The problem can be described as follows: A set of
emergency tasks requires observation by the agile
satellites, but their urgency levels may be different in
urgency. The task reprogramming can efficiently
accomplish the observation task without intervention of
the ground. In this way, the satellite can not only
respond quickly to emergency demands, but also perform
as many emergency tasks as possible, providing
maximum observation profits.
The difficulties in this question are as follows:
1. Generally, there are numerous constraints in
satellite task planning, so the reprogramming makes
the tasks more difficult.
2. The task time window and imaging time window
exist at the same time, so the new problem is a lot
more complicated than the general scheduling
3. The operation ability of agile satellite is limited, and
the planning algorithms need to be adapted to the
resources on the satellite.
4. Emergency task reprogramming requires high
efficiency in the process of reprogramming.
5. The emergency tasks are different in the urgency
levels and inconsistent with the requirements of
In this article, we analyzed the previous research and
proposed a model based on the following assumptions:
Assumption 1: The original planning is the best
planning, and emergency replanning could lower the
Assumption 2: The emergency task can be completed
through an observation, and there is only one
Assumption 3: The satellite can communicate with the
ground and other satellites (relay satellites) in time and
receive the pretreatment results.
Assumption 4: The replanning system can
communicate effectively with other systems on the
satellite and execute the corresponding commands
Assumption 5: The higher priority the emergency
task is, the faster the system accomplishes the
Assumption 6: The model only focuses on the satellite
imaging planning and observation process.
2.2 Model input
I represents the set of original tasks, J represents the
set of emergency tasks, K represents the set of tasks
that have been excluded from the original task plan;
Si and Ei represent the start time and completion
time of the task i in the original plan, WSi and WEi
represent the task start time window and end time
window of the ith item, S'i and WE'j represent the
start time window and end time window of task j,
T'j represents the duration of task j, wj represents
the task Priority of the task j, Ti → j represents the
attitude maneuver time from the ith original task to
the jth emergency task, mi represents the storage
consumption of the task i, M represents the
maximum storage capacity that the satellite can use for
replanning, ei represents the power consumption of
task i, E represents the maximum power that the
satellite can be used for replanning, and σ represents
the longest time required for attitude maneuver of
2.3 Model output
xj and yk represent the decision variables in the
model, where Xj indicates that the emergency task j
is selected; otherwise, it equals 0. yk indicates that
the task k is deleted; otherwise, it equals to 0. S’j
and E’j represent the start time and end time of the
emergency task j, SS represents the start time of the
replanning, SE represents the end time of the
replanning, and IG represents the time spent on the
generation of commands.
The objective function is designed to realize the
maximization of the total profit of replanning tasks. In
the model, fi(x) refers to the profit in the original task,
fj(x) is the profit of emergency task, lk(x) refers to the
loss value from the cancel task of the original plan, i, j,
and k refers to the original task sequence, emergency
Ei þ T i→ j ≤ S0j
6. The time window of emergency task j is not
overlapped with the time window of original
½Si; Ei ∩hS0j; E0ji ¼ ∅
7. Each of the emergency tasks should be
executed within the time window allowed
task sequence, and task sequence canceled
correspondingly. The model is as follows:
X f iðiÞ þ
X f jð jÞx j− X lk ðkÞyk
x j ≤ 1
SE þ IG ≤ WS0j
1. Each of the emergency tasks can be executed once
2. The total amount of time spent on replanning
and command generation should not exceed the
start time window of emergency task j.
3. The total storage consumption of all tasks should
not exceed the fixed capacity limit that satellites
can use for replanning.
X mi þ
X m j− X mk ≤ M
j∈ J k∈K
4. The total power consumption of all tasks should
not exceed the amount of power the satellite can
be used in replanning.
X ei þ
X e j− X ek ≤ E
j∈ J k∈K
5. There must be sufficient time from original task i to
emergency task j.
WS0j ≤ S0j < E0j ≤ WE0j
8. There is no overlapped time among emergency
hS0j1 ; E0j1 i∩hS0j2 ; E0j2 i∩…∩hS0jn ; E0jn i
¼ ∅; j1; j2; …; jn∈ J
These constraints make the problem optimization
more complicated. To solve this problem effectively, it is
very important to put forward a set of efficient solutions.
Next, we started from the characteristics of emergency
tasks and considered the solutions of emergency task
reprogramming in different circumstances.
The importance of the emergency task is regarded as
the weight, which is related to the priority of the task.
The mathematical expressions are as follows:
8 1:5 5 ≤ priority
wi ¼ < 1:3 3 ≤ priority < 5
: 1:0 0 ≤ priority < 3
It is obvious that the higher the task priority, the
greater impact on the final profit of the function. Above
all, we have described the model and the constraints
completely. In the next chapter, we designed a set of
algorithms to solve this problem.
Fig. 2 The overall flow of the algorithm
Due to the limited computing resources on the satellite,
the design of the algorithm must be compatible with the
], requiring the program to be as simple
and small as possible. At the same time, the replanning
of emergency task requires high efficiency. In this paper,
we proposed a set of emergency task planning
algorithms for combinatorial strategy selection. Through the
algorithms, we classified different tasks in the emergency
task sequence and ensured the high quality and
efficiency of solution. The algorithms were designed into
three kinds respectively: the emergency tasks with high
priority, general emergency tasks with fewer tasks, and
general emergency tasks with more tasks.
3.1 The overall process of the algorithm
Step 1: Tasks classification. Classify the tasks in
accordance with the priority and urgency degrees, and each
priority consists of a kind of ambiguous priority, giving
priority to the first priority order and calculating the
coverage of emergency task time windows. If time
window overlapped with that of the task of the first priority
exceeds the threshold, then analyze the combination of
the two time windows;
Step 2: Strategy selection. If it is an emergency tasks
with high priority, use (emergency task insertion
algorithm) ETIA. If it is the general emergency tasks with
fewer tasks, use general emergency task insertion
algorithm (GETIA). If it is the general emergency tasks with
more tasks, use general emergency task planning
&insertion algorithm (GETPIA);
Step 3: Constraint checking. Check the results of the
programming generated by the respective algorithms
and check whether they satisfy the constraints. If they
fail to pass, try the program with the second total
profit. If they still fail to pass, try the third one, so
on and so forth;
Step 4: Try more emergency tasks. Check if there is
still time, storage, and remaining power to join more
emergency tasks, and if so, return to the second step
and add the next priority emergency task; if there is no
time, storage or remaining power, then directly carry out
the fifth step;
Step 5: Profit calculation. Calculate the profit of each
part and finally determine the overall benefit using the
objective function. The overall flow chart of the
algorithm is shown in Fig. 2.
The whole algorithms contain three sub-algorithms.
The three sub-algorithms are mainly designed according
to the urgency of the emergency tasks and the number
of emergency tasks in the same category. First of all, we
designed the emergency task insertion algorithm for
emergency tasks with high priority.
The emergency task insertion algorithm (ETIA)
mainly solves the problem of the emergency tasks
with high priority, and the pseudo-code of the
algorithm is shown in Table 1. The basic idea is rapidly
positioning correspondingly through the location of
the percentage and then searching backward, until
there is enough time to insert the emergency tasks.
Insert the next task after a task is inserted. Since
there are no many high-priority emergency tasks, we
could insert the tasks one by one (Table 2).
In general emergency tasks insertion algorithm
(GETIA), the processing of tasks is relatively more
complicated than that of high-priority emergency
tasks. The algorithm is applied with the idea of
Greedy Algorithm [
]. For the original task plan
within the time window of emergency tasks, the task
sequence should be sorted according to the profit
sequence. The task elimination should be performed in
the direction of the maximum return of the new task
sequence. Therefore, the task with the lowest return
is supposed to be removed firstly (Table 2).
Figure 3 showed the process from considering the
inserted position to eliminating the conflicting tasks
in the original task plan. Firstly, we determined the
scope of tasks that may need to be eliminated to
resolve the task conflict according to the time window
of the emergency task. After comparing the profit of
the tasks, we selected one or more tasks with the
lowest level of profit, then replaced it with the
emergency task (Table 3).
The general emergency task planning &insertion
algorithm (GETPIA) uses the idea of Pareto
], which requires that any
improvement needed should not damage the profit of
other parts. For each task j added to the task
sequence, we suppose that j dominates i, which is
written as j ≺ i, subject to f(j) ≤ f(i). Pareto solving
method can not only effectively reduce the number of
arithmetic operations, but also complete the
emergency task insertion and task re-planning in one
algorithm flow at the same time with the resource
constraints of autonomous programming on the
satellite (Table 3).
Figure 4 described the process of replanning the task
plan after choosing the location of emergency task in
Algorithm III. The data from l1to l5 represent the loss
value of the five task execution paths. In the process of
selecting the path of the algorithm, assume that l3 is the
maximum loss value, then traverse all the situations
from emergency task with high priority to task two.
Since l3 is larger than l5, it is impossible to execute task
1 and then task 2 after executing the emergency task
with high priority according to the Pareto principle
because l3 has been removed from the operation. This
will help to improve the operation efficiency of the
4 Experimental analysis
The proposed algorithms are implemented by
Matlab2016b on a laptop with Core I5-3337U 1.
8 GHz CPU, 4 GB memory, and Windows 8.1
This part of the experiment was designed to verify the
feasibility of the algorithm. Since agile satellite task
planning has no baseline, so we did the simulation
Fig. 3 The diagram of task insert selection
experiments with scenario simulation and data under
agile satellite observation constraints generated
randomly. We analyzed generalization issues and did not
specify the type of satellites.
The main points of simulation experiments are as
1. The satellite task planning scenarios have been
2. A series of emergency tasks with different priorities
were randomly generated according to the task
requirements, and the time windows of these tasks
may overlap with those of each other.
We designed experiments to verify the three
subalgorithms in the emergency task planning algorithms
for combinatorial strategy selection and prove the overall
feasibility of the algorithm respectively.
5 Results and discussion
Experiment 1 was conducted to verify ETIA and the
result is shown in Table 4.
In the experiment 1, we simulated different original
tasks and considered different positions of emergency
tasks in the original task planning sequence. Besides, we
compared ETIA with the traditional emergency insertion
algorithm. Taking the calculation efficiency of satellite
into consideration, we assumed that the calculation time
of satellite was 500 times of the simulation experiment.
The experimental results showed that the average
running time of ETIA was less than that of the traditional
insertion algorithm. As the number of original task plans
increased, the completion time of the algorithm tended
to remain stable. However, ETIA is not ideal for
inserting multiple emergency tasks, partly because of its
greater volatility in profit, and on the other hand,
inserting tasks one by one is inefficient for multiple tasks.
When the task scale was 100, the emergency insertion
algorithm could save 10.306 s than the conventional
one. Therefore, we concluded that Algorithm I could
effectively solve the problem of reprogramming a few
emergency tasks with high priority.
In experiment 2, we discussed the GETIA. Firstly, we
considered the scale of different tasks and the number of
different emergency tasks; chose the task scales of 100,
150, and 200 for analysis; and selected 5, 10, 15, and 20
emergency tasks, respectively, and we considered the
three types of high density, low density, and high density
for each emergency task. Figure 5 represents the original
task scale of 50, 100, 150, and 200 from Fig. 5a–d,
respectively. In Fig. 5a, the GETIA has a good effect on
low-density emergency tasks when the original task scale
is 50. When the task scale rises to 100, the GETIA
performs well at the general density, which is shown in Fig.
5b. In Fig. 5c, after the task scale reaches 150, the
GETIA has higher profit when the scale of emergency
task with high density is 5 and 10, then the profit shows
a significant drop, which may be due to the excessive
number of original tasks deleted. When the scale of
original tasks is 200 and the number of emergency tasks
with low density is 20, the profit of GETIA with high
density is much higher than that of general density and
Another experiment was taken. We respectively
compared the completion of 5, 10, 15, and 20 emergency
tasks as the original task scale increases, and the result
is shown in Fig. 6. Figure 6a–d shows cases when the
number of emergency tasks is 5, 10, 15, and 20,
respectively. When the scale of emergency tasks is small,
the completion rate of tasks is greatly affected by the
distribution density of tasks. When the numbers of
emergency tasks and the original task continue to
increase, the completion rate of emergency tasks at
different densities shows a high degree of consistency. The
result also indicates that the GETIA can complete the
emergency tasks well.
Subsequently, we analyzed the influence of GETIA on
the original task. As shown in Fig. 7, we compared the
completion rate of the original tasks under different task
scales and different emergency task densities. Figure 7a–
d represents the case of 5, 10, 15, and 20 emergency
tasks, respectively. Figure 7 shows that the influence of
GETIA on the original task has great uncertainty. In the
best case, the original task sequence can continue
implementing the original planning scheme. In the worst case,
the original task can be only completed 68%, that is to
say, the task is influenced greatly. The result also
explains why great changes happen in the returns of the
new task sequence above.
We also conducted a comparison experiment between
GETIA and GETPIA. The experiment was based on
200 original tasks, and the emergency task scales were
set as 40, 60, 80, and 100, respectively, taking both
long-term and short-term emergency tasks into
account. The result is shown in Fig. 8. Figure 8a indicates
that when dealing with a short-term emergency task,
the profit of GETPIA is lower than that of GETIA.
However, in Fig. 8b, GETPIA performs better than
GETIA when the emergency task takes a long time.
The reason is that the GETPIA is more concerned with
the loss profit of the task when the number of tasks
changes, compared with GETIA. When the number of
emergency tasks is large and the duration is longer, the
addition of an emergency task may result in the
deletion of tasks in multiple original task sequences,
resulting in a decrease in profit.
We increased the scale of the original tasks to 300 and
the number of emergency tasks to 100, 120, 140, and
160 respectively. The results are shown in Fig. 9.
Figure 9a shows a short-term emergency task and Fig.9b
indicates a long-term emergency task. Dealing with 200
original tasks, GETPIA has higher profit than GETIA
for long-term emergency tasks. As the scale of the
original task increases, the gap between two algorithms also
In this research, a model was established to solve the
problem of autonomous programming for the
emergency missions of agile imaging satellites. We proposed
a combination of three strategies for the algorithm based
on the number and characteristics of emergency
missions. After that, those three strategies in the algorithm
were verified by three experiments. We find that the
algorithm could meet the requirements of autonomous
programming for a satellite as a whole. The three
subalgorithms had stringent requirements for the
characteristics of emergency missions, and the quality of solution
to the algorithm would be declined when it was beyond
their scopes of application. ETIA has high computational
efficiency. GETIA has a good performance in solving
small-scale and short-term emergency missions.
GETPIA is more suitable for large-scale missions and when
the duration of emergency missions is longer. The
conclusion demonstrates that these three algorithms can
solve the problem of autonomous replanning for
emergency missions on the satellite. As a result, a more
advanced study of satellite autonomous mission
programming is necessary, and some learning methods can
be used to enhance the forecasting ability of satellites.
AGATA: Autonomy Generic Architecture-Test and Application;
CASPER: Continuous Activity Scheduling Planning Execution and
Replanning; DLR: German Aerospace Agency; ETIA: Emergency task insertion
algorithm; GETIA: General emergency tasks insertion algorithm;
GETPIA: General emergency task planning & insertion algorithm;
RPA: Routine Planning Algorithm; VAMOS: Autonomous Task Planning
Onboard a Spacecraft
Availability of data and materials
Due to legal constraints the data for the research is not publicly available
upon publication. The data can be requested from the corresponding author
on reasonable request.
JYS and JDH, YZZ, and WYC designed the research. JYS performed the
research. JYS wrote the paper. JDH adjusted the format. YZZ processed data.
WYC guided the paper’s idea. All authors read and approved the final
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published
maps and institutional affiliations.
1. M Lema ̂ ıTre, G Verfaillie , F Jouhaud , JM Lachiver , N Bataille , Selecting and scheduling observations of agile satellites . Aerosp. Sci. Technol . 6 ( 5 ), 367 - 381 ( 2002 ).
2. Rabideau , G. , Tran , D. , Chien , S. , Cichy , B. , Shenvood , R. , & Mandl , D. , et al. ( 2006 ). Mission Operations of Earth Observing-1 with Onboard Autonomy . Proceedings of IEEE International Conference on Space Mission Challenges for Information Technology , 2006 . Smc-It (pp. 7 pp.- 373 ). IEEE.
3. ZY Lian , YJ Tan, ZZ Yan, Temporal reasoning technology for aeos scheduling . Syst. Eng. Electron . 35 ( 6 ), 1206 - 1211 ( 2013 ).
4. Wolfe , W. J. , & Sorensen , S. E. Three scheduling algorithms applied to the earth observing systems domain . Management Science . 46 ( 1 ), 148 - 166 ( 2000 ).
5. Y Min , Z Min , Small satellite task autonomous scheduling design method based on fuzzy neural network . J. Astronaut . 28 ( 2 ), 385 - 291 ( 2007 ).
6. JC Agn , N Bataille , D Blumstein , E Bensana , G Verfaillie , Exact and approximate methods for the daily management of an earth observation satellite . RAIRO - Oper. Res . 41 ( 4 ), 381 - 398 ( 2007 ).
7. S Chien , R Knight , A Stechert , R Sherwood , G Rabideau , Using Iterative Repair to Improve the Responsiveness of Planning and Scheduling , International Conference on Artificial Intelligence Planning Systems (AAAI Press, 2000 ), pp. 300 - 307 .
8. B Wille , MT Wörle, C Lenzen , Vamos-verification of autonomous mission planning on-board a spacecraft . IFAC Proc . Vol. 46 ( 19 ), 382 - 387 ( 2013 ).
9. MA Gleyzes, L Perret , P Kubik , Pleiades system architecture and main performances . ISPRS - Int. Arch. Photogramm. Remote Sens. Spati. Inf. Sci. XXXIX-B1 , 537 - 542 ( 2012 ).
10. SW Baek, SM Han, KR Cho, DW Lee, JS Yang, PM Bainum , et al., Development of a scheduling algorithm and GUI for autonomous satellite missions . Acta Astronaut . 68 ( 7-8 ), 1396 - 1402 ( 2011 ).
11. X Chu , Y Chen , Y Tan , An anytime branch and bound algorithm for agile earth observation satellite onboard scheduling . Adv. Space Res . 2017(1) , 1 - 15 ( 2017 ).
12. Augenstein , S. , Mann , J. M. , & Berkenstock , D. ( 2016 ). Satellite Scheduling System .
13. B Sun , L Mao , W Wang , X Xie , Q Qin , Satellite mission scheduling algorithm based on GA . Proc. SPIE 6795 , 30 ( 2007 ).
14. Maillard , A. , Verfaillie , G. , Pralet , C. , Jaubert , J. , Fontanari , F. , & Sebbag , I. ( 2015 ). Postponing decision-making to deal with resource uncertainty on earth-observation satellites . International Workshop on Planning and Scheduling for Space..
15. C. Pemberton A , J & G. Greenwald B , L. On the need for dynamic scheduling of imaging satellites . 34 . ( 2012 )
16. Q Dishan , H Chuan , L Jin , M Manhao , A dynamic scheduling method of earth-observing satellites by employing rolling horizon strategy . ScientificWorldJournal 2013 ( 3 ), 304047 ( 2013 ).
17. B Song , Y Li , Y Chen , F Yao , Y Chen , A repair-based approach for stochastic quadratic multiple knapsack problem . Knowl.-Based , 145 , 145 - 155 ( 2018 ).
18. P Wojtaszczyk , Greedy algorithm for general biorthogonal systems . J. Approx. Theory 107 ( 2 ), 293 - 314 ( 2000 ).
19. C Kahraman , O Engin , İ Kaya, RE Öztürk, Multiprocessor task scheduling in multistage hybrid flow-shops: a parallel greedy algorithm approach . Appl. Soft Comput . 10 ( 4 ), 1293 - 1300 ( 2010 ).
20. HE Romeijn, DR Morales, A class of greedy algorithms for the generalized assignment problem . Discret. Appl . Math. 103 ( 1 ), 209 - 235 ( 2000 ).
21. H Li , Q Zhang , Multiobjective optimization problems with complicated pareto sets, MOEA/D and NSGA-II . IEEE Trans. Evol. Comput . 13 ( 2 ), 284 - 302 ( 2009 ).
22. MP Andersson , T Bligaard , A Kustov , KE Larsen, J Greeley , T Johannessen , et al., Toward computational screening in heterogeneous catalysis: paretooptimal methanation catalysts . J. Catal . 239 ( 2 ), 501 - 506 ( 2006 ).
23. D Lei , A pareto archive particle swarm optimization for multi-objective job shop scheduling . Comput. Ind. Eng . 54 ( 4 ), 960 - 971 ( 2008 ).
24. PK Shukla , K Deb , On finding multiple pareto-optimal solutions using classical and evolutionary generating methods . Eur. J. Oper. Res . 181 ( 3 ), 1630 - 1652 ( 2007 ).