You are on page 1of 6

A Hybrid Chaotic Quantum Evolutionary Algorithm

Cai Yanguang
Faculty of Automation Guangdong University of Technology Guangzhou, China caiyg99@163.com

Zhang Minjie
Faculty of Automation Guangdong University of Technology Guangzhou, China wdyhdz@126.com

Cai Hao
Department of Development and Planning Aalborg University, 9220 Aalborg, Denmark howard007008@hotmail.com

Abstract-A hybrid chaotic quantum evolutionary algorithm is


proposed to reduce amount of computation, speed up convergence and restrain premature phenomena of quantum evolutionary algorithm. The proposed algorithm adopts the chaotic initialization method to generate initial population which will form a perfect distribution in feasible solution space in advantage of randomicity and non-repetitive ergodicity of chaos, the simple quantum rotation gate to update non-optimal individuals of population to reduce amount of computation, and the hybrid chaotic search strategy to speed up its convergence and enhance the global search ability. A large number of tests show that the proposed algorithm has higher convergence speed and better optimizing ability than quantum evolutionary algorithm, real-coded quantum evolutionary algorithm and hybrid quantum genetic algorithm. Tests also show that when chaos is introduced to quantum evolutionary algorithm, the hybrid chaotic search strategy is superior to the carrier chaotic strategy, and has better comprehensive performance than the chaotic mutation strategy in most of cases. Especially, the proposed algorithm is the only one that has

100%

convergence

rate

in

all

tests.

The

presented

algorithm is applied to urban traffic signal timing optimization and the effect is satisfied.

Keywords-evolutionary algorithm; algorithm; chaos; hybrid chaotic algorithm 1.

quantum quantum

evolutionary evolutionary

INTRODUCTION

In recent years, quantum evolutionary algorithm (QEA) has been widely applied in logistics transportation scheduling, task scheduling of grid, wireless sensor network layout, intrusion detection of network and nonlinear equation solving as in [1-5]. For solving optimization problems, QEA adopts quantum bit encoding to represent individuals, and uses quantum rotation gate and quantum NOT gate to update them. But studies reveal that although QEA shows better global search capability than conventional evolutionary algorithms, it still has some defects [6-7]. Firstly, the workload of lookup operation for quantum rotation angle updating based on binary encoding is too heavy for engineering application. Secondly, when solving high dimensional optimization problems, QEA is prone to cause premature convergence. Therefore, in [8] Gao introduced real coding into QEA, and proposed the real-coded quantum evolutionary algorithm (RQEA). RQEA has higher

convergence speed than QEA. Chaos is a common phenomenon in nature and it has features such as randomness, ergodicity and regularity. Introducing chaos into intelligent algorithms provides a new direction for improving their convergence [9-12]. For example, Chen proposed the real-coded chaotic quantum-inspired genetic algorithm (RCQGA) in [13]. RCQGA has better search capability than QEA. However, RQEA and RCQGA are still subject to premature convergence problem in a certain extent when solving high-dimensional optimization problems. For solving optimization problems, hybrid chaotic quantum evolutionary algorithm (HCQEA) is proposed. HCQEA adopts chaotic initialization method to generate initial population, which will form a perfect distribution in feasible solution space in advantage of randomicity and non repetitive ergodicity of chaos, the simple quantum rotation gate to update non-optimal individual in current population to reduce amount of computation, and the hybrid chaotic search strategy to speed up its convergence and enhance global search capability. A large number of tests show that the proposed algorithm has higher convergence speed and better optimizing ability than QEA, RQEA and hybrid quantum genetic algorithm (HQGA) in [14]. Tests also show that when chaos is introduced to quantum evolutionary algorithm, hybrid chaotic search strategy is superior to carrier chaotic strategy, and has better comprehensive performance than chaotic mutation strategy in most of cases. Especially, the proposed algorithm is the only that has 100% convergence rate in all tests. In addition, HCQEA also has features of easy realization and convenient engineering application. Its performance is satisfied in urban traffic signal timing optimization. II.
ALGORITHM DESIGN

The optimization problem to be discussed is minf (X) , XE S (1)

where X=(xj , X2"", xn)T , S={ (Xl. X2"", xn)T I XiE [ai, bJ, a6bb both a; and bi are real constants, i=1, 2, "', n}.

A.

The Principle of the Algorithm

HCQEA applies real code to represent individuals. It adopts chaotic initialization method to generate initial population which will form a perfect distribution in feasible

978-1-4244-6585-9/10/$26.00 2010 IEEE

771

solution space in advantage of randomicity and non repetitive ergodicity of chaos, the simple quantum rotation gate to update non-optimal individuals in current population to reduce its amount of computation, the quantum rotation gate based on the gradient of objective function and chaotic sequence to update optimal individual of current population so that convergence speed can be increased. The chaotic mutation is also used to avoid premature convergence. For simplicity, the hybrid chaotic search strategy means the quantum rotation gate based on the gradient of objective function and chaotic sequence, and the chaotic mutation.

where i=l , 2, ... ,n, ()o is the basic rotation angle (specified small positive real constant, generally O<()o<l). Let

(); ' = ();+ l!J. ()i' i=l , 2, ... , n


The i-th gene of e is updated as follows:

(S)

B; < 0 B; > 1
otherwise

, i 1, 2, ..., n
==

(6)

B.

Parameter Design

1) Individual Coding Let X=(x) , X2 , ..., xn)T be any feasible solution of (1). Then for any x; (i=l , 2, ..., n) , there exists a unique real ();E [0 ,1] such that
(2) Obviously, (); and X; are one to one correspondence. Let X=(x) , X2 , ..., xn)T be any feasible solution of (1). If for any i = 1 , 2 , ..., n, there exists a real (); E [O ,l] to satisfy (2), we call e=(), ()2 , "' , ()n)T be the individual code of X (referred to as individual without confusion), and (); the i-th gene of e. 2) The Strategy for Generating Initial Population HCQEA adopts chaotic initialization such that the initial population forms a perfect distribution in feasible solution space because of random and ergodic feature of chaos. The process is as follows: Randomly to generate a real ()j (O<()j<1 , ()1';tO.S) as the initial value of the chaotic sequences, and to calculate the sequences by Logistic mapping:
Definition 1

();+j =4 . ();' (1

();)

(3)

where i=l , 2, ..., n-l. The system is completely chaotic. Let (), ()2 , ..., ()n)T be an individual of the initial population. T Proposition 1 Let O<()j<l and ()j f: O.S. If()j , ()2 , ... , ()n) is generated by (3), then it is an individual of (1). ProofIn fact, we can easily prove O<();<l , i=1 , 2, ... , n. 3) The Updating Strategy for Non-optimal Individual in Current Population Simple quantum rotation gate is used to update non optimal individuals in current population. Let B=()jb , ()2b , ""()nb)T be the global optimal individual, i.e., B is the optimal individual for algorithm to run so far. Let f9=(), ()2 , "" ()n)T be any non-optimal individual in current population, where the optimal individual of current population means that it has the minimum value of objective function in current population. The rotation angle /)B; of quantum rotation gate of i-th gene of e is given by:

where ();" is the new value of ()i' In this way, QEA's lookup operations are replaced by simple arithmetic operations such that amount of calculation significantly reduces and the speed of QEA improves. T Proposition 2 Let B=()jb' ()2b , "" ()nb) be the global T optimal individual, e=(), ()2 , "" ()n) an arbitrary non optimal individual in current population, and O<()o<l. If ()j" , ()/, ... , ()n")T is decided by (4)- (6), then it is an individual of (1). Proof For i=l, 2, ..., n, we have !/)();!<l by (4) since O<()o<l. By (S) and ();E[O,l], we have -1<(); '<2. By (6), we have ();"E [O,l] , i.e., ()j" , ()2" , ... , ()n")T is an individual of (1). 4) The Updating Strategy for Optimal Individual of Current Population Based on the gradient of objective function and chaotic strategy, we will construct the quantum rotation gate to update optimal individual of current population. Let pre Gen be the evolution generations, max Gen the maximum evolution generations, m the invalid evolution generations (i.e., the evolution generations for the optimal individual of current population to be not continuously improved), and M the superior limit of invalid evolution generations (specified positive integer constant, e.g., M=60). Assume pre Gen<max Gen and m<M. Let f9=(), ()2 , ... , ()n)T be the optimal individual of current population. e will be updated as follows: When solving (1), only if the search direction forms an acute angle with its negative gradient, the objective function declines. Therefore we design perturbation parameter fJ; based on the gradient of objective function:

2 Pi -Vf(e) i 1Vf(ei )) + D), i


==

1, 2 , ..., n

(7)

where V j{();) is!,s partial derivative for (); at e (If the partial derivative does not exist, it is replaced by f s change rate for (); at any point of some sufficiently small neighborhood of e). D is a specified positive constant (e.g., D=SOOO). The rotation angle /)(); of quantum rotation gate of i-th gene of e is determined by:
'

Bb =B. I!J.B. = ' Bo . (Bb i -Bl i, i -Bi)/IBb


I

{o

else

(4)

I!J.(); = fJ ; ' ();* , i=l , 2, ..., n

(8)

772

where

e/ = 4 e; (1 - e;), i=l , 2, ... , n


is a chaotic sequence produced by Logistic mapping. Let

(9)

Output: solution of (1). a) Preferences: population size N, maximum evolution generations max Gen (a positive integer), a positive constant D, maximum invalid evolution generation M , basic rotation angle eo.

e;, = e;+ b,B;,


The i-th gene of

i=l , 2, ... , n

(10)

e e; == 2 'e; , B;, ' ,

where e;" is the new value of the i-th gene of e. Thus, the optimal individual of current population will generally evolve along the direction of decreasing objective function value. This has benefit of the whole population's evolving toward the direction of the global optimal solution. Notice that if the number of optimal individual of current population is more than one, each optimal individual will be updated by the above method. Proposition 3 Let B=(e), e2, , en)T be the optimal
.

1-

e is

updated according to:

b) Initialization: set evolution generation pre Gen=O, and invalid generation m=O, and then produce initial population according to the method described in the section 'The Strategy for Generating Initial Population'. c) All individuals of population are decoded according to (2) to get the corresponding feasible solution, and then find the optimal individual in current population.

B; < B; > 1
otherwise

, i == 1, 2, ..., n

(11)

d) If pre Gen=O, set the optimal individual of current population as the global optimal individual B. e) If pre Gen<max Gen , goto f); else output of B and the corresponding objective fuction value and stop the algorithm.

fs,

fs,

j) If pre Gen>O , and if the objective function value of the optimal individual of current population is smaller than set the optimal individual of current population as B, and set m=O; else m=m+1.

individual of current population. If D>1, then (el", e2", en")T decided by (7)-(11) is a valid individual of (1).
Proof Firstly,

. .

g) Update all non-optimal individuals in current population according to the method mentioned in the section 'The Updating Strategy for Non-optimal Individual in Current Population'. h) If m<M, update optimal individuals of current popu lation according to the method given in the section 'The Updating Strategy for Optimal Individual of Current Population'; else (m=M) mutate the whole population according to the section 'The Solution for Premature Convergence'. i) pre Gen=pre Gen+1, goto c).
III.
TESTS AND ANALYSIS

1/3;1<1. In fact, by (7), we have

IPd IVj(Bi J/I D < 1 if 1 V f ee;) 1':::1, and IP;I < IVf(B;)I/Vf(B; l J== lJIVj(B;)1 < 1 if 1 V fie;) I> l.
On the other hand, we have e;*E [0,1]. Since B=(e), e2, ..., en)T is an individual, then e;E [0,1]. We immediately obtain

e/E [0,1] by (9). Therefore Ie; 1<1 holds by (8).


Then by following the second half part of the proof of Proposition 2, we can easily prove that (el", e2", ... , en")T is an individual of (1). 5) The Solutionfor Premature Convergence Assume pre Gen<max Gen and m=M. It means that evolution converges to some local extreme points. Here, we adopt Logistic mapping to chaotically mutate the gene of each individual in current population so that the distribution of population can be altered and premature convergence can be overcome. Let (ek), ek2, , e/m)T be the k-th individual in current population, k=1, 2, ... , N , where the positive integer N is the size of population.
.

A.

Test Functions

QEA, HQGA, RQEA and HCQEA are used to solve famous test functions in papers [15-16], and then their convergence and optimizing ability are compared. Results show that the performance of HCQEA is excellent. Due to limitation, we only list the test results for 10 functions among them. The dimension n of high dimensional function is set to 30, x* is the theoretical global optimal solution. 1) Sphere Model minfl (x)

;=1 * T S=[-100, 100)", x*=(O, 0, ..., O) ,./i (X )=O. 2) Cosine Mixture Problem min i=1 i=1 * S=[-I, 1]", x*=(O , 0, ..., 0) T,./2 (X )= - O.ln. 3) Exponential Problem

== L X /

(12) where ek;' is the new value of the i-th gene of individual k.
C.

fz (x) L X / - 0.lL cos(511Xi)


==

Steps ofH CQEA


Input: optimization problem (1).

773

min f3 (x) =-(exp(-0.5 L x/ ))

S=[-1, It, x*=(0, 0, ..., 0) T, h(x*)= - 1. 4) Griewank Problem 1 X 2 min f4 (x)=1 + -- L x; - n cos( f:) 4000 ;=1 V i ;=1
n n-l 2 minfs(x)=0.I{sin2 (37lXl ) + L (x; -1)2[1 + sin (37lX;+1 )] n

;=1

S=[-600, 600t, x*=(O, 0, ... , 0) \f4(x*)=O. 5) Levy and Montalvo Problem


;=1

S=[-5, 5t, x*=(I, 1, ..., 1) \fs(x*)=O. 6) Schwefel's Problem


n

+ (xn _1)2 [1 + sin 2(27lXn )] )


;

algorithm is less than 0.001. Each algorithm runs 50 times for each test function, and the test result is denoted by the average of the results of all running. The result is shown in Table I that is used to compare of convergence speed, best optimal value and average optimal value, and Table II for the comparison of convergence rate. Here, convergence speed of algorithm is described by the average evolution generations when its all running converges (referred to as average convergence generations). Optimizing ability of algorithm is represented by best optimal value (the best objective function value in all running), average optimal value (the average objective function value of all running) and convergence rate (the number of convergence with the ratio of the number of running).
TABLE!.
TF AL

COMP ARISON OF ALGORITHMS


CS BOV AOV

n S=[-100, 100] , x*=(O, 0, ... , 0) \f6(x*)=O. 7) De Jong Function


min f7 (x)=100(x2 - x/ )2 + (x2 _1)2 S=[-2.048, 2.048f, x*=(1, 1) T, h(x*)=O.

min f6 (x) =L( L xj)2 ;=1 j=l

minii

minii

8)

Sinusoidal Problem
n n
_

minij

9)

min fs(x)=-[2.5n sin(x; - 7l") + n sin 5(x; 7l")] 6 6 ;=1 ;=1 S=[O, 7tt, x*=( 27t/3, 27t/3, ... , 27t/3) T, .fs(x*)= - 3.5.

Peaks Function
-x/ -(x, +l)

min./4

min f9 (Xl,x2)=10 - 3(1- Xl)2e

' + lOc

Xl -X2 )e

-Xj--X,-

i+

+ "3 e

minis

1 -(Xj+l) 2 -x,-

S=[-3, 3f, x*=(-0.009, 1.581) T, .f9(x*)=1.8938. 10) Six-Hump Camel-Back Function 6 4 Xl . 2 mm.!iO(xl,X2)=4xl -2.1Xl + - + XIX2 -4x22 + 3 4 S=[-5, 5]2, x*=( -0.0898, 0.7126) T or (0.0898, fi o(x*)=1.0316285.

mini6

minf,

4X2

_ 0.7126) T,
mini,

B.

Results and Analysis


minfg

The parameters of QEA are set as: each component of feasible solution is represented by 20 binary bits, population size N=10, basic rotation angle 80=;r;/50. The parameters of HCQEA are set as: N=10, M=60, D=5000, 80=0.005. The parameters of HQGA are set the same as [14]: cross probability Pc=l, mutation probability Pm=0.05, N=30. The parameters of RQEA are set the same as [8]: N=10, 80=0.4;r;, evolution scale ,0.05, number of times of continuous refinement search mj=6, number of times of continuous extensive search m2 =2. The maximum evolution generation of all the algorithms is maxGen=5000. Evolution is regarded to be convergent if the relative error between the theoretical global optimal value and the best optimal value obtained by

minfio

QEA HQGA RQEA HCQEA QEA HQGA RQEA HCQEA QEA HQGA RQEA HCQEA QEA HQGA RQEA HCQEA QEA HQGA RQEA HCQEA QEA HQGA RQEA HCQEA QEA HQGA RQEA HCQEA QEA HQGA RQEA HCQEA QEA HQGA RQEA HCQEA QEA HQGA RQEA HCQEA

n 1420 1637 1358 n 226 482 162 n 160 328 52 n 2110 2217 2081 n 284 447 226 n 1720 1860 1579 622 122 155 76 n 340 523 360 816 89 328 63 724 81 203 75

15.628 7.3e-8 4.1e-20 2.ge-11 -l.8253 -2.9996 -3 -3 -0.5527 -1 -1 -1 4.9911 8.3e-4 1.8e-5 1.4e-5 64.971 6.5e-1O 2.5e-12 7.0e-18 10.853 3.2e-5 6.5e-7 4.3e-7 2.6e-4 2.2e-11 6.1e-1O 4.1e-12 -l.0 -3.5 -3.5 -3.5 1.8943 1.8938 l.8938 l.8938 -l.032 -l.032 -l.032 -l.032

113.41 2.2e-7 l.6e-15 3.0e-11 -0.5712 -2.9960 -2.9826 -3 0.1116 -1 -1 -1 102.61 9.6e-4 8.ge-4 6.7e-5 71.146 2.7e-8 l.ge-1O l.8e-17 100.27 3.8e-5 1.1e-6 4.4e-7 0.0726 9.6e-9 5.6e-9 2.5e-9 0.1365 -3.5 -3.5 -3.5 2.3363 l.9002 1.8940 1.8940 -0.9772 -l.031 -l.032 -l.032

Note: IF represents test function, AL represents algorithm, CS represents convergence speed, BOV represents best optimal value of objective function, AOY represents average optimal value of objective function, and 'n' represents that algorithm is not convergent in all running.

771

TABLE II.
CR (%) QEA

COMPARISON OF CONVERGENCE RATE


HQGA RQEA HCQEA TF

TABLE III.
CR(%)

TEST RESULT OF RCQEA


CS BOV AOV

min/i minJi mini; min!. mini; mini, minf, minis minfo min./io

0 0 0 0 0 0 30 0 8 92

100 96 100 86 100 90 100 96 96 96

100 94 100 82 100 92 100 92 100 98

100 100 100 100 100 100 100 100 100 100

min/i minJi mini; min!. mini; mini, minf, minis minfo min./io

20 24 30 0 0 0 68 12 72 98

2062 478 343 n n n 533 682 216 274

8.2e-6 -3 -1 0.0116 3.7802 9.9503 l.3e-5 -3.5 1.8940 -1.03I

4.7e-5 -2.9987 -I 0.1141 20.332 71.101 5.6e-5 -3.4980 1.8952 -1.031

Note: CR represents convergence rate

It is shown in Table II that QEA isn't convergent for solving high dimensional optimization problems, and thus it seems to be not suitable for solving such kind of problems. Neither is it satisfied for solving two dimensional optimization problems such as.h and .19' The main reason is that binary coding affects the precision so that the algorithm is difficult to converge to the global optimal solution. To some extent, HQGA restrains premature convergence and has higher convergence rate than QEA by the strategy of quantum gates-based updating search in longitudinal direction and real coding-based genetic search in transverse direction. But the strategy spends long running time. RQEA adopts real coding such that it can avoid premature convergence caused by binary coding scheme. Both its convergence rate and convergence speed are roughly the same as HQGA. Comparing with QEA, RQEA and HQGA, HCQEA has the best convergence speed and the best optimizing ability. It is also shown in Table II that HCQEA is the only algorithm that has 100% convergence rate for all tests. IV.
COMPARATION OF CHAOTIC STRATEGIES

It can be seen from Table III that the performance of RCQEA for solving high dimensional optimization problems is bad. Its convergence rate is very low, and its convergent speed is slower than HCQEA. Although RCQEA has better performance for solving two dimensional optimization problems such as h, hand fio than high dimensional optimization problems, its performance is worse than HCQEA. We can see that the advantage of hybrid chaotic search strategy is its optimal individual updating strategy and the solution of premature convergence. The former conducts that population evolves toward the theoretical global optimal solution according to the gradient information of the objective function, which can effectively reduce invalid evolution and increase convergence speed. The latter enhances global search ability by chaotically mutating all individuals when evolution falls into local optimal.

B.

Chaotic Mutation

Carrier chaos and chaotic mutation are two common methods to introduce chaos. In this section, two chaotic evolutionary algorithms obtained by respectively employing carrier chaos and chaotic mutation will be compared with HCQEA.

In order to evaluate the effects of introducing hybrid chaotic search strategy and chaotic mutation into evolutionary algorithm, we compare HCQEA with RCQGA. RCQGA adopts real probability crossover and chaotic mutation to generate offspring. The foregoing test functions are still used to test RCQGA. The parameters of RCQGA are set the same as [13], and test result is shown in Table IV.
TABLE IV.
TF

TEST RESULT OF RCQGA


CS BOV AOV

A.

Carrier Chaos

Carrier is often used when chaos is introduced to intelligent algorithms. By replacing hybrid chaotic search strategy of HCQEA with carrier chaotic strategy, we can obtain a new algorithm, real-coded chaotic quantum evolutionary algorithm (RCQEA). We still use the same test functions given in the above section to test RCQEA. Preferences are the same as HCQEA. The algorithm independently runs 50 times for each test function. The result is shown in Table III.

min./i min/i mini; mini4 minis mini, mini; minis minig min/io

CR(%)

100 96 100 90 88 92 100 88 100 98

1476 192 89 2106 186 1627 95 405 88 128

5.1e-9 -3 -I 2.6e-5 6.7e-8 2.4e-6 6.3e-8 -3.5 1.8938 -1.032

1.ge-8 -3 -I 6.7e-5 1.ge-7 3.ge-6 8.5e-8 -3.5 1.8938 -1.032

It can be seen from Table IV that RCQGA has higher convergence rate than RCQEA. However By comparing Table IV with Table I and Table II, HCQEA has higher convergence rate than RCQGA. And in most of cases, HCQEA has better best optimal value, better average optimal value and fewer average convergence generations.

775

v.

CONCLUSION

By constructing the chaotic initialization method to generate initial population, and the hybrid chaotic search strategy to increase convergence speed and enhance global search ability, the hybrid chaotic quantum evolutionary algorithm is presented. A large number of tests show that the proposed algorithm has higher convergence speed and better optimizing ability than QEA, RQEA and HQGA. The proposed algorithm is applied to urban traffic signal timing optimization and the effect is satisfied.
ACKNOWLEDGMENT

E1ectronica Sinica, vol. 36, Nov. 2008, pp. 2120-2124, doi:cnki: sun:dzxu.0.2008.11.007. [13] H. Chen, J. Zhang and C. Zhang, "Real-coded chaotic quantum inspired genetic algorithm," Control and Decision, vol. 20, Nov. 2005, pp. 1300-1303, doi:cnki:issn:1001-0920.0.2005.11.021. [14] L. Wang, F. Tang and H. Wu, "Hybrid genetic algorithm based on quantum computing for numerical optimization and parameter estimation," Applied Mathematics and Computation, vol. 171, Feb. 2005, pp. 1141-1156, doi:I O.1016/j.amc.2005.01. 115. [15] L. Wang, Intelligent Optimization Algorithms with Applications. Beijing, China:Tsinghua University Press, 2001. [16] D. Kusum and T. Manoj, "A new mutation operator for real coded genetic algorithms," Applied Mathematics and Computation, vol. 193, Oct. 2007, pp. 211-230, doi:10.10I 6/j.amc. 2007.03.046.

The research is supported by the National Natural Science Foundation of China under grant No. 61074147 and No. 60374062, the Team Project of the Natural Science Foundation of Guangdong Province under grant No. 8351009001000002, the Science and Technology Program of Guangdong Province under grant No. 2008BOI0200005, and the Key Discipline of 211 Engineering of Guangdong Province under grant No. Yuefagai 431.
REFERENCES [I ] A. Narayanan and M. Moore, "Quantum-inspired genetic algorithms," Proc. IEEE Symp. Evolutionary Computation (ICEC 1996), IEEE Press, May 1996, pp. 61-66, doi:! 0.1109/ICEC.1996. 542334. H. Xing, X. Liu, X. Jin, L. Bai and Y. Ji, "A Multi-granularity evolution based quantum genetic algorithm for QoS multicast routing problem in WDM Networks," Computer Communications, vol. 32, Feb. 2009, pp. 386-393, doi:10.1016/j.comcom.2008.I 1.00 9. Y. Zhao, D. Peng, 1. Zhang and B. Wu, "Quantum evolutionary algorithm for capacitated vehicle routing problem," Systems Engineering-Theory & Practice, vol. 29, Feb. 2009, pp. 159-166, doi:cnki:sun:xt11.0.2009.02.022. R. Jozsa, "Quantum computation in algebraic number theory: Hallgren's efficient quantum algorithm for solving Pell's Equation," Annals of Physics, vol. 306, Feb. 2003, pp. 241-279, doi:! 0.1016/S0003-4916(03)00067-8. Y. Gao, K. Lu and Z. Shen, "Gradual target elimination quantum genetic algorithm," Signal Processing, vol. 25, Feb. 2009, pp. 238242, doi:cnki:sun:xxcn.0.2009.02.016. G. Zhang, W. Jin and L. Hu, "Quantum evolutionary algorithm for multi-objective optimization problems," Proc. IEEE Symp. Intelligent Control (ISIC 03), IEEE Press, Oct. 2003, pp. 703-708. K. Han and 1. Kim, "Quantum-inspired evolutionary algorithms with a new termination criterion, HE gate, and two-phase scheme," IEEE Trans. Evolutionary Computation, vol 8, Apr. 2004, pp. 156169,doi:I O.1109/TEVC.2004.823467. H. Gao, G. Xu, R. Zhang and Z. Wang, "Real-coded quantum evolutionary algorithm," Control and Decision, vol. 23, Jan. 2008, pp. 87-90, doi:cnki:sun:kzyc.0.2008.01.018.

[2]

[3]

[4]

[5]

[6]

[7]

[8]

Y. Cai and M. Wei, "Self-adaptive chaos particle swarm optimization for allied vehicle routing problems," Systems Engineering, vol. 26, Aug. 2008, pp. 32-36, doi:cnki:sun:gcxt.O. 2008.08.005. [10] H. Zhang, G. Han and Y. Pan, "Chaos immune evolutionary algorithm and its applications to function optimization," Pattern Recognition and Artificial Intelligence, vol. 20, Feb. 2007, pp. 225229, doi:cnki:sun:mssb.0.2007.02.013. [I I ] S. Gao, "Solving traveling salesman problem by chaos ant colony optimization algorithm," Systems Engineering-Theory & Practice, vol. 25, Sep. 2005, pp. 100-104, doi:cnki:issn:1000-6788. 0.2005.09.016. [12] W. Ma and Q. Zhu, "fast continuous ant colony optimization algorithm for solving function optimization problems," Acta

[9]

776

You might also like