Efficient Satellite Scheduling Based on Improved ... - Semantic Scholar

JOURNAL OF NETWORKS, VOL. 7, NO. 3, MARCH 2012

517

Efficient Satellite Scheduling Based on Improved Vector Evaluated Genetic Algorithm Tengyue Mao

1, 2

1 State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, Hubei, China, 430079 2 College of Computer Science, South-Central University for Nationalities, Wuhan, Hubei, China, 430074 E-mail:[email protected]

Zhengquan Xu

State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, Hubei, China, 430079 E-mail: [email protected]

Rui Hou

College of Computer Science, South-Central University for Nationalities, Wuhan, Hubei, China, 430074 E-mail: [email protected]

Min Peng

State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing Wuhan University, Wuhan 430079, China Email: [email protected] Abstract—Satellite scheduling is a typical multi-peak, manyvalley, nonlinear multi-objective optimization problem. How to effectively implement the satellite scheduling is a crucial research in space areas.This paper mainly discusses the performance of VEGA (Vector Evaluated Genetic Algorithm) based on the study of basic principles of VEGA algorithm, algorithm realization and test function, and then improves VEGA algorithm through introducing vector coding, new crossover and mutation operators, new methods to assign fitness and hold good individuals. As a result, the diversity and convergence of improved VEGA algorithm of improved VEGA algorithm have been significantly enhanced and will be applied to Earth-Mars orbit optimization. At the same time, this paper analyzes the results of the improved VEGA, whose results of performance analysis and evaluation show that although VEGA has a profound impact upon multi-objective evolutionary research, multi-objective evolutionary algorithm on the basis of Pareto seems to be a more effective method to get the non-dominated solutions from the perspective of diversity and convergence of experimental result. Finally, based on Visual C + + integrated development environment, we have implemented improved vector evaluation algorithm in the satellite scheduling. Index Terms—satellite scheduling, VEGA, multi-objective optimization, dividing Pareto front

I. INTRODUCTION Remote sensing satellite uses a variety of satellite sensors to photograph targets within a certain range of ground to produce high resolution images. According to imaging mechanism, it can be divided into visible light imaging, infrared imaging and microwave imaging. Since they have some merits such as broad coverage area, long © 2012 ACADEMY PUBLISHER doi:10.4304/jnw.7.3.517-523

duration, effective surveillance and free from airspace boundaries, satellite observations has important applications in military reconnaissance, and environmental protection and enjoy the high priority in the countries around the world[1]. Specifically speaking, the satellite scheduling refers to allocate resources to competing multiple observation missions and determine starting and ending time for the specific activities of missions so as to eliminate the conflicts of resource use among different missions and meet the needs of users to maximum, based on the comprehensive consideration of the ability of remote sensing satellites and user demands on remote sensing images [2, 3]. The satellite scheduling is quite complex and consists of many constraints associated with the specific problems, such as rule constraint of satellite and payload use, visible constraint of satellite and target, capacity constraint of satellite memory, constraint of visibility of satellite and ground receiving station and the capacity of data transmission, and requirement of observation missions on the image type, image resolution, sunlight and cloud thickness. Especially in recent years, dexterity of remote sensing satellites has been constantly increased in order to provide more choices for observation of a given target, which also makes satellite mission planning become more complex[4, 5,6]. The satellite scheduling belongs to multi-objective optimization problem. In MOP (Multi-Objective Genetic Algorithms MOGA), various objectives restrain each other through decision variables, and the objective units are often inconsistent, which makes it difficult to

518

JOURNAL OF NETWORKS, VOL. 7, NO. 3, MARCH 2012

objectively evaluate the merits and shortcomings of solutions of MOP. Due to its unique advantage, MOGA (multi-objective genetic algorithm) has now become one of main solutions of MOP. The solution does not require man to define the weight of each objective. On the contrary, the algorithm via operation finds out all Pareto fronts and then decision makers weigh and make a choice. At present, MOGA has aroused a heated discussion in the research field. Multi-objective optimization evolutionary algorithm can be divided into two main categories: multi-objective evolutionary algorithm based on Pareto and non-Paretobased multi-objective evolutionary algorithm. Obviously, VEGA is multi-objective evolutionary algorithm which is not based on Pareto front. In the process of the optimization only use Pareto dominance of individual in groups to evolve so as to approximate Pareto front. It is possible easier to eliminate some dominated solutions at the early stage of the optimization. But at the middle and late stage or much earlier stage of the optimization, such cases are likely to occur. It is difficult to compare their advantages and disadvantages among individuals in the evolutionary group with the limited size. Sometimes all individuals do not dominate, but still far converge to Pareto frontier. Thus, the normal evolutionary optimal selection can not conducted. Without the strategic choice and discarding often result in the loss of VEGA optimal solution and its diversity. Finally, the results VEGA searched basically tend to the optimal solution of single objective functions, and meanwhile diversity and convergence are unsatisfactory. When making improvements, this paper has introduced the fitness which divides Pareto front and niche technology to calculate individuals in order to solve equilibrium of Pareto dominance among individuals happening in the group evolutionary process so that evolutionary search can continue. From the test results we can see that after the algorithm has been improved, individuals are divided evenly in the optimal Pareto front, and convergence and diversity are highly enhanced. But the nature of the improvement, based on the Pareto front, is to have a clear direction for the selecting the elite which is required to retain in each generation so that groups converge rapidly and distribute evenly. At this time, the improved VEGA can not be simply thought to be not based on Pareto front [7, 8, 9]. II. SYSTEM MODEL A. System Description Considering multi-objective minimization problem in the general form, we can use the following set of mathematical formulas to describe [10]: Pinying corresponding Latin characters is displayed: (1) min F ( x) = ( f1 ( x), f 2 ( x), , f n ( x)) (2) s.t. gi ( x) ≤ 0, i = 1, 2, m (3) where x = ( x1 , x2 , xk ), xi ∈ [ ai , bi ], i = 1, 2, k In the above formula, x is a k-dimensional decision variable in the decision space, and F ( x) is an objective

© 2012 ACADEMY PUBLISHER

space. It can be inferred from the above formula that this multi-objective problem has k-decision variables, n objective functions and m inequality constraints. ai and b show that value range of number i decision variable xi in vector x. Definition 1 Pareto Dominance: superior to vector Vector u = (u1 , u2 , , uk ) i

v = (v1 , v2 ,

∃i ∈ (1, 2,

, vk )

means that ∀i ∈ (1, 2, , k ) has

ui ≤ vi

， and

, k ) makes ui < vi record as u ≺ v , also

called vector u dominating vector v 。 Definition 2 Pareto Optimality: The idea that a candidate solution x ∈ Ω is the optimal solution of Pareto means that x ∈ Ω (disorderly code) makes F ( x ' ) ≺ F ( x) . Ω is solution space. Definition 3 Pareto Optimal set: (4) P = {x ∈ Ω ¬∃x ∈ Ω st F ( x ) ≺ F ( x )} '

Definition 4 Pareto Front: PF = {F ( x) x ∈ P}

'

(5)

From the above definition 1&2 we can see that Pareto Front here is possibly satisfactory solution decision makers may need. However, since Pareto solution of MOP is often a set (an area), it is the crux of the problem to find out more and even elements in this set so that decision makers have sufficient alternatives. MOGA is to construct non-dominated set and to enable non-dominated set to approximate constantly the optimal solution set until the optimal. Mission planning problem of satellite observation to Earth can be briefly described as a group of satellites, a group of observation missions, and the fulfillments of each observation mission including data collection and data return. Specify a priority for each observation mission; there are a set of available time window between the satellite and ground targets corresponding to observation mission; a reference time range is viewed as the starting and ending time of mission planning. Satellite observation to Earth need to meet the following constraints: each observation mission must be completed within an available time window; there is sufficient time to adjust between two successive observations; the limit of number of adjustment of satellite side view, storage capacity and energy lead to the limit of the cumulative observation time of each circle. The goal of mission planning problem is to maximize the weighted sum of observation missions, or complete missions of high priority as many as possible, or make satellite energy consumption to minimum on the basis of meeting the needs of mission, or multi-objective combinations forming multi-objective problem. B. Satellite Scheduling model Description A satellite scheduling problem can be expressed by 4tuple Θ = ( A, R, TW, C ) . Set of task is ,set of satellite resources A = ( a1 , a2 ,..., an ) is R = ( R1 , R2 ,..., Rn ) .C is constraint set, including: Mission constraints mainly mean that a mission must be executed or not, using Cl to represent; Capacity constraints: once

JOURNAL OF NETWORKS, VOL. 7, NO. 3, MARCH 2012

519

the mission is planned, it must be assigned certain resources, including satellite and satellite visible time window, and the assigned number of resources can not exceed the capacity of the satellite resources, using Cb to represent; Compatibility constraints: the same satellite time window can not be assigned to two or more missions, using Ce to represent. Time requirements described the mission chiefly means that the mission must be finished within the established time period, using Chorizon to represent. TW: time window set. For each satellite each mission has a series of time window or not. Suppose starting time of scheduling is 0, the ending time Thorizon . Mission decision variables: is T = (T1 , T2 ,..., Tn ) , ∀i ≤ n, Ti = {0,1} , Ti = 1 Which means it is planned; Ti = 0 which means it is not planning. Demand variables: Qij signifies the demand of the number i mission to time window of the number j satellite; Time window (virtual resources): 1 1 TW = {TW1,11 , TW1,21 , , TW1,11 , TW2,1 , TW2,2 , TW2,1P ; ;TW1,1m , TW1,2m , ,TW1,mi } , TWi1,k represents the number j satellite is the number k

time window of the number i mission. The range of j ≤ m, i ≤ n, k is obtained by actual calculation, where the time window corresponds to satellite resources. The time period of mission planning is registered as TWi ,jk . P = { P1 , P 2 , , Pn } is the priority of the mission.

According to the above constraints, regarding completing the sum of mission priority as the largest target, we set up the initial problem model of satellite scheduling: (6) max : ∑ Ti Pi i

st : Ti = {0,1} ； P = {1, 2,

(7)

, P} ；

(8)

⎧⎪Cl = (Tn = 1) Cl : ⎨ , n′ ∈ N ′ ⊆ N ； ⎪⎩Cl = ∧ i∈( N − N ′) (T1 = 1)

(9)

⎧⎪Cb (T1 ) = ∧(Ti = 0) ∨ (∪ RS ≠ Φ) Cb : ⎨ , i = 1, 2, ⎪⎩Cb (T1 ) = ∧(Ti = 0) ∨ (∪Qi , j ≠ Φ )

Old population

n

(10)

f1

P1

f2

P2

f3

P3

f4

P4

f5

P5

Ce : Ce (twij, k ) = ∧((Qi ,jk = twij, k ) ∨ (Qqj, j = twij, k ))

if

∃q, r , st , tw ∩ tw j i,k

≠Φ；

Ctemp : 0 ≤ si < Thorizon , 0 ≤ ei ≤ Thorizon

(11) (12) (13)

III. PROPOSED EFFICIENT SATELLITE SCHEDULING The ultimate goal of genetic algorithm is to obtain a group of sets with good diversity and convergence. However, since Pareto solution of MOP is often a set (an area), it is the crux of the problem to find out more and even elements in this set so that decision makers have sufficient alternatives. MOGA is to construct nondominated set and to enable non-dominated set to approximate constantly the optimal solution set until the optimal. A. The thought of VEGA The algorithm evaluates an objective vector and every element of vector represents an objective function [7]. VEGA is a straightforward extension of a single objective genetic algorithm for multi-objective optimization. When a series of objectives (set M) need to deal with, Schaffer considers that the population in each generation is randomly divided into M sub-populations. Each sub-population based on different objective function is assigned fitness. So, each of M target functions is used to assess members of the population. In this paper, Figure 1 illustrates the fitness assessment program of five objective functions. Population in each generation is equally divided into five parts. Every individual in the first population is assigned fitness only on the basis of the first objective function, while every individual in the second population is assigned fitness only on the basis of the second objective function and so on. In order to reduce the positional bias in the population, it would be better to mix up the location of individual in the population before being split into five equal-sized sub-populations. After each individual has been assigned fitness, selection operators which are limited in each sub-population are employed until the fill of sub-populations is completed.

Crossover & mutation

Mating pool Figure 1 VEGA Diagram

© 2012 ACADEMY PUBLISHER

j q,r

New population

520

It is particularly useful when dealing with the issues that there is great difference in value range of objective functions. Since all the members in the sub-population are assigned fitness based on their own objective functions, to constrain selection operators stresses good populations corresponding to their own objective function only in sub-populations. In addition, because no two individuals based on different objective functions are compared, inconsistency within a series of different objective functions will not cause any difficulties. What Schaffer uses is to select operators according to proportion [7, 8]. VEGA algorithm is described as follows (take M objective functions as an example, M = 2): Step1 Using the traditional binary coding, algorithm first randomly generates initial population and its size is popsize. Step2 During operation in each generation (1) Set counter i = 1, and define q = popsize / M (2) For all individuals ranging from j = 1 + (i-1) * q to j = i * q, assign the fitness as: (14) F ( j ) = fi max − fi ( j ) max f1 is the maximum value of number i objective function, while fi ( j ) is the value of number i objective function. If i = M, go to (3), otherwise i = i + 1, go to (2). As for two objective functions, put the former popsize/2 individuals in sub-population 1 and regard f1max － f1( j ) as the value of fitness of corresponding individual j, and then put the latter popsize/2 individuals in subpopulation 2 and regard f 2max － f 2( j ) as the value of the fitness of corresponding individual j. (3) Select each sub-population according to proportion, then substitute individuals with the best fitness for individuals with the worst fitness in the subpopulation and combined all sub-population into one, that is the current population oldpop. Step3 Make use of roulette wheel selection mechanism oldpop, binary single-point crossover and single-point mutation for the current population until the number of the new population newpop arrives at popsize. Step4 Save non-inferior solutions of every generation. Methods: Non-inferior solutions of new newpop generated by each generation are kept in defined nonpoor population, and then select non-inferior solutions in non-poor population. Define N as the number of noninferior solution set of the final output, and define number as the current total number of individuals in the non-poor population nond. If the number is greater than N, put them in order according to the sum of the value of objective function, remove the latter (number-N ) / 3 and the former number-N-(number-N) / 3 individuals, making the current total number of individuals in nond is still N. Step5 Run until terminate algebra and output nond solution set. Otherwise go to Step 2. VEGA emphasizes on the population which is beneficial for the single objective function. The results finally found are mostly the optimal solution of a single objective.

© 2012 ACADEMY PUBLISHER

JOURNAL OF NETWORKS, VOL. 7, NO. 3, MARCH 2012

Although VEGA is simple and easy to achieve, there are some restrictions in working principles. Disadvantages are also obvious. Each individual in VEGA is evaluated only in one objective function. Thus, each individual is not evaluated by other (M-1) objective functions, but these objective functions are also very important in MPO. In the process of simulation run of VEGA, it is most likely that in the sub-populations the set of a single objective function which gets close to the best will be selected by selection operators. This preference selection will similarly occur in other objective function in different sub-populations. In the initial study of VEGA, it is assumed that the crossover operator can combine the optimal solution set of a single objective function together and find compromising solutions in Pareto optimal region. However, as Schaffer has found, even if in the convex search space problems, the intersection between the optimal sets of a single objective function can not find diversified set in populations, that is to say, diversity is poor. Consequently, VEGA can only search the optimal set of a single objective function [7]. B. The Imporvement of VEGA The ultimate goal of genetic algorithm is to obtain a group of sets with good diversity and convergence. In this paper, some improvements have been made as follows. (1) Make use of allocation strategy of fitness value of new individuals and compare the degree of congestion of individuals in the same front through niche technology so as to maintain even distribution of individuals in populations and together with Pareto strength value apply to the definition of the fitness value of individuals. (2) To retain the elite, the main idea of the paper is that from the current populations and new populations generated after hybrid select the best popsize individuals (the contemporary population size) to be involved in the operation of the next generation. Only half of the worst individuals will be eliminated. Due to less pressure to be eliminated, elite individuals are automatically retained. Multi-objective minimization problem in the general form is still taken into consideration. First describe the fitness assignment [7, 9]. The first step of fitness assignment algorithm is to divide population P into different Pareto fronts depending on dominance degree. Individuals in each sub-population P j (front) are non-inferior solutions of this subpopulation. No individual is more dominated than the other individuals (m is the number of sub-populations). P = ∪mj=1 Pj (15) The set of the first front is that P1 individual is the best non-dominated set in population P. The secondary non-dominated set in population P belongs to P2, and so on. For any two solutions i and j in any front (take P1 for example), their normalized distance d ij can be calculated as follows:

JOURNAL OF NETWORKS, VOL. 7, NO. 3, MARCH 2012

dij =

521

2

⎛ f k( i ) − f k( j ) ⎞⎟ ∑ ⎜⎜⎜⎝⎜ f max ⎟⎟ − f max ⎠⎟ M

k =1

k

(16)

k

In the equation, f kmax and f kmin are the maximum value and the minimum value of k objective function value. ( j) f k( i ) is k objective function value of individual i, and f k is k j objective function value of individual for the individual target function values j. Once the distance is calculated, we may make use of them to calculate and share function through the following formula. ⎧1− dij / rz d ij < rz ⎪ sh( dij ) = ⎪ ⎨ ⎪ 0 otherwise ⎪ ⎩

(17)

P1

nci = ∑ sh(dij )

(18)

j =1

P1 is the total number of individuals in the front.

Specific steps are as follows: Step1 Individuals in the population P are classified by degree of dominance, forming a series of sub-populations ( P1 , P2 , Pm ); Step2 Choose a shared parameter rz and a small positive number w, initialize Fmin = N + w, (N is the population size) and set the counter j = 1; Step3 for individual q in each sub-population Pj . 1) Assign fitness F = Fmin – w; 2) calculate value ncq of niche of each individual in Pj ; 3) Calculate shared fitness value F (q) = F / ncq . Step4 Fmin = min (F (q): q is any individual in Pj , and set j = j + 1; Step 5 If j ≤ m, go to Step 3, otherwise, the process ends. Using arithmetic hybridization and arithmetic mutation, the method is as follows: 1) Hybridization. The process of hybridizing individual x1and individual x2 so as to generate crossover operator of new individuals c1 , c2 is as follows ( x1 and x2 in the following equations are vectors, a ∈ [0,1] ). c1 = ax1 + (1− a ) x2 c2 = ax2 + (1− a) x1

(19)

2) Mutation. The process of making individual x mutate so as to generate mutation operator of new individuals u is as follows[13], where u and x are vectors, x (j) and u (j) are number j component of the vectors u and x, x( j ) and x( j ) are respectively the lower and upper bounds of x (j), a is random number, a ∈ [−1,1] , b is nonuniform coefficient, b ∈ [2, 7] and gen are the current operation algebra, and mgen is the maximum operating algebra. ⎧⎪x( j) + a( x( j) − x( j))(1− gen / mgen)b a < 0 u( j) = ⎪⎨ ⎪⎪x( j) + a( x( j) − x( j))(1− gen / mgen)b a ≥ 0 ⎪⎩

© 2012 ACADEMY PUBLISHER

(20)

From

a ( x( j ) − x( j ))(1− gen / mgen)b

or

a ( x( j ) − x( j ))(1− gen / mgen)b

in the above formula, we can see that as the increase of operating algebra gen, the probability close to 0 will increase, which makes the operator at the beginning (when the gen is very small) may evenly search in the solution space so as to increase the opportunity to find non-inferior solution. But in later times (when gen is very large) has local search nature. At this time, the solutions in the set are mostly close to or have been non-inferior solutions. There is great possibility for local search to find non-inferior solutions. If even search continues, it is likely to search the solution farther away from noninferior solutions. We have elaborated in detail the literature related to the improvement of the algorithm [14], so we will not repeat them. Improved VEGA (IVEGA) is described as follows (take M objective function as an example, M = 2): Step1 Using vector coding, randomly generate initial population and assign each individual a same initial fitness. Step2 Operation of each generation. 1) Set the counter i = 1; 2) Randomly divide populations into M subpopulations; 3) Next, sort individual in each sub-population according to some corresponding objective function value of from small to large. For instance, the number i subpopulation sequences its individuals from small to large according to objective function value. If i = M, go to 3, otherwise i = i + 1, go to 2). Step3 Every time using wheel selection mechanism, select one from two sub-populations, crossover and mutate until the number of new population newpop achieves to popsize. Step4 Integrate oldpop and new generated population newpop into a newly defined population where 2*popsize individuals can be put. Step5 Assign 2*popsize individuals in the population cpop fitness value according to a certain method (The algorithm process of assigned fitness will be discussed in later chapters), and select from popsize individuals with the best fitness to put in oldpop. Step6 Judge whether running algebra terminates. If so, the operation ends, and output all the individuals in oldpop, that is non-inferior solutions we seek. Otherwise, go to Step 2. IV. Simulation results Experiment function T1： Minimize T1 ( x) = ( f1 ( x), f 2 ( x)) f1 ( x) = x1

(21)

f 2 ( x) = (1 + x2 ) / x1 x = ( x1 , x2 ), x1 ∈ [0.1,1], x2 ∈ [0,5]

This function is put forward by Deb [7]. The optimal Pareto front lies in f 2 = 1/ f1 (0.1 ≤ f1 ≤ 1) . Experiment function T2：

522

JOURNAL OF NETWORKS, VOL. 7, NO. 3, MARCH 2012

Minimize T2 ( x) = ( f1 ( x ), f 2 ( x)) 3

f1 ( x) = 1 − exp(−∑ ( xi −1 / 3) 2 )

(22)

i =1 3

f 2 ( x) = 1− exp(−∑ ( xi + 1 / 3) 2 ) i =1

x = ( x1 , x2 , x3 ), xi ∈ [−4, 4], i = 1, 2,3

The Pareto optimal set of experiment function T2 is ： x1 = x2 = x3 ∈ [−1/ 3,1/ 3] . The experiment is completed in the VC + +. After several numerical experiments the improved algorithm determines the appropriate parameters. The parameter of function T1 is selected as follows: population size N = 100, the neighborhood radius rz = 0.15, pcross = 0.95, pmutation = 0.1,and running algebra is 1000 generations. The parameter of function T2 is selected as follows: population size N = 100, the neighborhood radius rz = 0.05, pcross = 0.99, pmutation = 0.1, running algebra is 3,000 generations. Figure 2 shows the evolutionary result before and after the improvement of objective function of experiment function Tl. Figure 3 shows the evolutionary result before and after the improvement of objective function of experiment function T2l. In the figure, the abscissa is considered as objective function f1, and the vertical axis is considered as objective function f2.

Figure 2: the comparison of the operating result before and after the improvement of algorithm of experiment function Tl.

It can be seen from Figure 2 that although VEGA obtains three optimal solutions when operating once, it is clear that the revised VEGA, the results not only converge to the optimal Pareto front, but also present an even distribution so that diversity has been greatly improved. Figure 3 is the evolutionary results before and after the improvement of the algorithm of experimental function of T2. Obviously, convergence and diversity of the solution obtained in such a power function where binary encoding VEGA is applied is not satisfactory. However, the improved results basically fall in the optimal Pareto front, and the results are evenly distributed and have satisfactory diversity. The improved algorithm has high stability. For two test functions, the solutions obtained every time are fundamentally distributed in the optimal Pareto front. © 2012 ACADEMY PUBLISHER

Figure 3: the comparison of the operating result before and after the improvement of algorithm of experiment function T2.

V. CONCULSION VEGA is multi-objective evolutionary algorithm which is not based on Pareto front. In the process of the optimization only use Pareto dominance of individual in groups to evolve so as to approximate Pareto front. It is possible easier to eliminate some dominated solutions at the early stage of the optimization. But at the middle and late stage or much earlier stage of the optimization, such cases are likely to occur. It is difficult to compare their advantages and disadvantages among individuals in the evolutionary group with the limited size. Sometimes all individuals do not dominate, but still far converge to Pareto frontier. When making improvements, this paper has introduced the fitness which divides Pareto front and niche technology to calculate individuals in order to solve equilibrium of Pareto dominance among individuals happening in the group evolutionary process so that evolutionary search can continue. From the test results we can see that after the algorithm has been improved, individuals are divided evenly in the optimal Pareto front, and convergence and diversity are highly enhanced. But the nature of the improvement, based on the Pareto front, is to have a clear direction for the selecting the elite which is required to retain in each generation so that groups converge rapidly and distribute evenly. At this time, the improved VEGA can not be simply thought to be not based on Pareto front. Acknowledgment The project was supported by the Morning Programmer for Young Scientists of Wuhan, China, under grant of 201150431076, and the Special Fund for Basic Scientific Research of Central Colleges, SouthCentral University for Nationalities under grant of CZZ10004. References [1] Y. W. Chen, B.B. Cun, H.R. Jie and L.J. Fang, “A Survey on Mission Planning for Remote Sensing Satellites,”

JOURNAL OF NETWORKS, VOL. 7, NO. 3, MARCH 2012

[2]

[3] [4]

[5]

[6] [7] [8] [9]

[10]

[11]

[12]

[13]

[14]

Journal of Spacecraft TT&C Technology, Vol.27, No.5, pp. 1–7, Oct. 2008. E. Bensana,G. Verfaillie and G. Agnese, “Exact and ap proximate methods for the daily management of an earth observation satellite,” Proc of the 4th International Symposium on Space Mission Operations and Ground Data System, 1996: 3-12. G. Halln and J. Magazinem, “Maximizing the value of a space mission,” European Journal of Operations Research, Vol.78, No.2, pp. 224-241, Dec. 1994. G. Verfaillie and M. Lemaitre, “Selecting and scheduling observations for agile satellites: some lessons from the constraint reasoning community point of view,” Proc of the 7th International Conference on Practical and Principles of Constraint Programming, Berlin, Springer, 2001: 670-684. M. Lemaitre, G. Verfaillie,F. Jouhaud, ,et al, “Selecting and scheduling observations of agile satellites,” Aerospace Science and Technology, Vol.6, No.5, pp. 367-381, Oct. 2002. P. Wang and T.Y. Jin, “Mission planning problem for earth observing satellites: a survey,” Application Research of Computers, Vol.25, No.10, pp. 2893–2897, Oct. 2008. K. Deb, “Multi-Objective Optimization using Evolutionary Algorithms,” USA:JH ON WILEY & SONS.LTD,2001. J. D. Schaffer, “Multiple Objective Optimization With Vector Evaluated Genetic Algorithms,” PhD thesis, Vanderbilt University, Nashville, TN, USA, 1984. N. Srinivas and K. Deb, “Multi-Objective Optimization Using Non-dominated Sorting in Genetic Algorithms,” Evolutionary Computation, Vol.2, No.3, pp. 221-248, Jan. 1995. J. Qin and L.S. Kang, “a Convergence Analysis Framework for Multi-objective Optimization Genetic Algorithm,” Computer Applications and Software, Vol.23, No.1, pp. 1–2, Jan. 2006. Y. Liu, “Research on Dynamic Rescheduling Model, Algorithm and Its Applications of Imaging Reconnaissance Satellite Scheduling Problem,” Graduate School of National University of Defense Technology Doctor’s degree paper, April.2004. Q. M. Yuan, “Research on Photo-reconnaissance Satellites Scheduling Problem for Area Targets Observation,” Graduate School of National University of Defense Technology Doctor’s degree paper, April.2006. J.Z. Chen,J. Wang,J. Li and N. Jing, “Research on Comprehensive Evaluation Technology of Satellite Mission Schedule Algorithm,” Computer Engineering, Vol.35, No.20, pp. 59–61, Jan. 2009. V. Gabrel,D. Vanderpooten, “Enumeration and Interactive Selection of Efficient Paths in a Multiple Criteria Graph for Scheduling an Earth Observing Satellite,” European Journal of Operational Research, Vol.139, No.3, pp. 533– 542, Jan. 2002.

Tengyue Mao received the B.E. degree in electronic technology and the M.S. degree in computer application from Wuhan University (WHU), Hubei, China, in 1998 and 2002, respectively, and he is currently pursuing the Ph.D. degree in communication and information system at WHU. Since 2006, he has been on the faculty in computer school at South-Central University for Nationalities, and he is currently a lecturer in the Department of Computer Application. His research interests include

© 2012 ACADEMY PUBLISHER

523

multimedia information security, and multimedia network communication technology. Mr. Mao was awarded two scientific and technological progress prizes and one invent patent applying. Zhengquan Xu received the B. E. degree in radio technology and information system and M.S. degree in communication and electronic system form Tsinghua University, China, in 1985 and 1988, respectively, and the Ph.D degree in biomedicine engineering form Hong Kong Polytechnic University, in 1998.From 1988 to 1999, he was on the faculty in the department of Electronics and Information at the Huazhong University of Science and Technology. In 1995, he was selected to the Hong Kong Polytechnic University (HKPU) by the State Education Ministry. During the period, he was selected to Stanford University as a short-term visiting scholar by the” Areas of Excellence” plan of the HKPU. Since 1999, he has been on the faculty in Wuhan University (WHU).He is currently the director of multimedia communication engineering center of state key laboratory of information engineering in surveying, mapping and remote sensing at WHU. He is the expert of government affairs information construction engineering of Wuhan City, and the expert of government affairs information construction engineering of Hubei Province. His research interests include space and multimedia information processing, multimedia information security, and multimedia network communication technology. Mr. Xu has presided or participated in a variety of more than 20 research projects and was awarded five scientific and technological progress prizes: three Ministerial-level and two Municipal-level awards respectively. Rui Hou received the B.M.E., B. L., and M.E. degree in mechanics, law school, and physical electronic from Wuhan University, Wuhan, China, in 2000, 2003, and 2003, respectively, and the Ph.D degree in optical engineering from Huazhong University of Science and Technology, Wuhan, China, in 2006. Between the years 2003 and 2006, he was a Research Staff Member in the Wuhan National Laboratory for Optoelectronics. He is currently an assistant professor, College of Computer Science, South-Central University for Nationalities, Wuhan, China. He has published over 60 papers. His main research interests include WDM networks, optical switching, optical waveguide, optical fiber techniques and MPLS/GMPLS. Dr. Rui Hou is a Member of the IEEE and a Member of the OSA. Min Peng received the B.E. degree in computer technology from Central China Nomal University and the M.S. degree in computer application from hubei university of technology, Hubei, China, in 2003 and 2007, respectively, and he is currently pursuing the Ph.D. degree in communication and information system at WHU. Since 2007, His research interests include MAS Social Relationship Model and Trust Model.