Abstract
This paper considers the optimization of linearly constrained stochastic problem which only noisy measurements of the loss function are available. We propose a method which combines genetic algorithm (GA) with simultaneous perturbation stochastic approximation (SPSA) to solve linearly constrained stochastic problems. The hybrid method uses GA to search for optimum over the whole feasible region, and SPSA to search for optimum at local region. During the GA and SPSA search process, the hybrid method generates new solutions according to gradient projection direction, which is calculated based on active constraints. Because the gradient projection method projects the search direction into the subspace at a tangent to the active constraints, it ensures new solutions satisfy all constraints strictly. This paper applies the hybrid method to nine typical constrained optimization problems and the results coincide with the ideal solutions cited in the references. The numerical results reveal that the hybrid method is suitable for multimodal constrained stochastic optimization problem. Moreover, each solution generated by the hybrid method satisfies all linear constraints strictly.
Similar content being viewed by others
References
Bazaraa M, Sherali H, Shetty CM (2006) Nonlinear programming: theory and algorithms. Wiley-Interscience, New Jersey
Buriol LS, Hirsch MJ, Pardalos PM (2010) A hybrid genetic algorithm for road congestion minimization. Optim Lett 4(4):619–633
Conn AR, Gould NIM, Toint PL (1991) A globally convergent augmented Lagrangian algorithm for optimization with general constraints and simple bounds. SIAM J Numer Anal 28(2):545–572. doi:10.1137/0728030
Conn AR, Gould NIM, Toint PL (1997) A globally convergent augmented lagrangian barrier algorithm for optimization with general inequality constraints and simple bounds. Math Comput 66(217):261–288
Du D-Z, Zhang X (1986) A convergence theorem of Rosens gradient projection method. Math Program 36(2):135–144. doi:10.1007/BF02592021
Du D-Z, Zhang X (1989) Global convergence of Rosen’s gradient projection method. Math Program 44(1—-3):357–366. doi:10.1007/BF01587098
Du D-Z, Pardalos PM (1999a) Handbook of combinatorial optimization, vol 1-3. Kluwer Academic Publishers, Norwell
Du D-Z, Pardalos PM (1999b) Handbook of combinatorial optimization, supplement volume A. Kluwer Academic Publishers, Norwell
Du D-Z, Pardalos PM (2005) Handbook of combinatorial optimization, supplement volume B. Kluwer Academic Publishers, Norwell
Dupuis P, Kushner H (1987) Asymptotic behavior of constrained stochastic approximations via the theory of large deviations. Prob Theory Relat Fields 75(2):223–244. doi:10.1007/BF00354035
Floudas CA, Pardalos PM (2009) Encyclopedia of optimization, 2nd edn. Springer, New York
Floudas CA, Pardalos PM (2003) Frontiers in global optimization. Kluwer Academic Publishers, Norwell
FengChang M (2009) Optimization method and matlab programming. Science Press, Beijing
Fu M, Hill D (1997) Optimization of discrete event systems via simultaneous perturbation stochastic approximation. IIE Trans 29(3):233–243. doi:10.1023/A:1018523313043
Goldberg DE (1989) Genetic algorithms in search, optimization and machine learning. Addison Wesley, New York
Goldberg DE (2002) The design of innovation: lessons from and for competent genetic algorithms. Kluwer Academic Publishers, Norwell
Giannessi F, Pardalos PM (eds) (2001) Optimization theory: recent developments from Matrahaza. Kluwer Academic Publishers, Norwell
Gupta R, Agarwal M (2006) Penalty guided genetic search for redundancy optimization in multi-state series-parallel power system. J Comb Optim 12(3):257–277. doi:10.1007/s10878-006-9632-1
Hock W, Schittkowski K (1980) Test examples for nonlinear programming codes. J Optim Theory Appl 30(1):127–129. doi:10.1007/BF00934594
Hadjisavvas N, Pardalos PM (2001) Advances in convex analysis and global optimization. Kluwer Academic Publishers, Norwel
Ji M, Jin Z, Tang H (2006) An improved simulated annealing for solving the linear constrained optimization problems. Appl Math Comput 183(1):251–259
Kushner H, Clark D (1978) Weak convergence: constrained systems. Springer, New York
Kushner H, Yin GG (1997) Weak convergence methods for general algorithms. Springer, New York
Kushner HJ, Gavin T (1974) Stochastic approximation type methods for constrained systems: algorithms and numerical results. IEEE Trans Autom Control 19(4):349–357. doi:10.1109/TAC.1974.1100580
Kushner HJ, Lakshmivarahan S (1977) Numerical studies of stochastic approximation procedures for constrained problems. IEEE Trans Autom Control 22(3):428–439. doi:10.1109/TAC.1977.1101505
Kushner HJ, Sanvicente E (1974) Penalty function methods for constrained stochastic approximation. J Math Anal Appl 46(2):499–512. doi:10.1016/0022-247X(74)90256-X
Kushner HJ, Sanvicente E (1975) Stochastic approximation of constrained systems with system and constraint noise. Automatica 11(4):375–380. doi:10.1016/0005-1098(75)90086-2
Michalewicz Z (1996) Genetic algorithms+data structures=evolution programs. Springer, Verlag Berlin Heidlberg
Okamoto T, Hirata H (2009) Constrained optimization using the chaotic Lagrangian method and the simultaneous perturbation gradient approximation. ICCAS-SICE, pp 741–746
Pardalos PM, Resende MGC (eds) (2002) Handbook of applied optimization. Oxford University Press, New York
Pardalos PM, Rajasekaran S (eds) (2001) Handbook of randomization (2 volumes). Kluwer Academic Publishers, Norwell
Pardalos PM, Du D-Z (2013) Handbook of combinatorial optimization, 2nd edn. Springer, New York
Pardalos PM, Romeijn E (2002) Handbook of global optimization: heuristic approaches, vol 2. Kluwer Academic Publishers, Norwell
Rosen JB (1960) The gradient projection method for nonlinear programming. Part I. Linear constraints. J Soc Ind Appl Math 8(1):181–217
Sadegh P (1997) Constrained optimization via stochastic approximation with a simultaneous perturbation gradient approximation. Automatica 33(5):889–892. doi:10.1016/S0005-1098(96)00230-0
Spall JC (1998) Implementation of the simultaneous perturbation algorithm for stochastic optimization. IEEE Trans Aerosp Electron Syst 34(3):817–823. doi:10.1109/7.705889
Wang IJ, Spall JC (1998) A constrained simultaneous perturbation stochastic approximation algorithm based on penalty functionsIntelligent Control (ISIC), 1998. Held jointly with IEEE International Symposium on Computational Intelligence in Robotics and Automation (CIRA), Intelligent Systems and Semiotics (ISAS), Proceedings pp 452–458. doi:10.1109/ISIC.1998.713704.
XiaoPing W (2002) Genetic algorithm: theory, application and programming. Xian Jiaotong University Press, Xian
Acknowledgments
Thanks for the supports of National Natural Science Foundation of China (Grant Nos. 61273174, 61034006, and 60874047).
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
-
1.
Test function 1
$$\begin{aligned} \begin{array}{l@{\quad }l} \min &{} f = - 3{\left( {1 - {x_1}} \right) ^2}{e^{ - x_1^2 - {{\left( {{x_2} + 1} \right) }^2}}} + 10\left( {0.2{x_1} - x_1^3 - x_2^5} \right) {e^{ - x_1^2 - x_2^2}}\\ &{} + \frac{1}{3}{e^{{{\left( {1 - {x_1}} \right) }^2} - x_2^2}}\\ \hbox {Subject to: } &{} {x_1} - {x_2} \ge - 6,\, {x_1} + {x_2} \ge - 6,\, - 4 \le {x_1} \le 4,\, - 4 \le {x_2} \le 4. \end{array} \end{aligned}$$The global solution is \({x^ * } = \left( {\begin{array}{cc} {0.000474}&{1.580535} \end{array}} \right) \), and \(f\left( x^* \right) = - 8.105417\).
-
2.
Test function 2
$$\begin{aligned} \begin{array}{ll} \min &{} f = \left( {1 + {{\left( {{x_1} + {x_2} + 1} \right) }^2}\left( {19 - 14{x_1} + 3x_1^2 + 6{x_1}{x_2} + 3x_2^2} \right) } \right) \\ &{} \times \left( {30 + {{\left( {2{x_1} - 3{x_2}} \right) }^2}\left( {18 - 32{x_1} + 12x_1^2 + 48{x_2} - 36{x_1}{x_2} + 27x_2^2} \right) } \right) \\ \hbox {Subject to:} &{} - 2 < {x_i} < 2,\, i = 1,2. \end{array} \end{aligned}$$The global solution is \({x^ * } = \left( {0, - 1} \right) \), and \(f\left( {{x^ * }} \right) = 3\).
-
3.
Test function 3
$$\begin{aligned} \begin{array}{ll} \min &{} f = {\left( {{x_2} - \frac{5.1}{4{\pi ^2}}x_1^2 + \frac{5}{\pi }{x_1} - 6} \right) ^2} + 10\left( {1 - \frac{1}{{8\pi }}} \right) \cos \left( {{x_1}} \right) + 10 \\ \hbox {Subject to:} &{} - 5 \le {x_1} \le 10,\, 0 \le {x_2} \le 15 \end{array} \end{aligned}$$The problem has three global solutions:
$$\begin{aligned} x_1^ * = \left( { - \pi ,12.275} \right) ,\, x_2^ * = \left( {\pi ,2.275} \right) ,\, x_3^ * = \left( {3\pi ,2.475} \right) . \end{aligned}$$In all cases \(f\left( {x_i^ * } \right) = \frac{5}{{4\pi }},\, i = 1,2,3\).
-
4.
Test function 4
$$\begin{aligned} \begin{array}{ll} \min &{} f = \left\{ {\begin{array}{l} {{x_2} + {{10}^{ - 5}}{{\left( {{x_2} - {x_1}} \right) }^2} - 1,0 \le {x_1} < 2}\\ {\frac{1}{{27\sqrt{3} }}\left( {{{\left( {{x_1} - 3} \right) }^2} - 9} \right) x_2^3,2 \le {x_1} < 4}\\ {\frac{1}{3}{{\left( {{x_1} - 2} \right) }^3} + {x_2} - \frac{11}{3},4 \le {x_1} < 6} \end{array}} \right. \\ \hbox {subject to:} &{} \frac{1}{\sqrt{3} }{x_1} - {x_2} \ge 0,\, - {x_1} - \sqrt{3} {x_2} + 6 \ge 0,\, 0 \le {x_1} \le 6,\, {x_2} \ge 0. \end{array} \end{aligned}$$The problem has three global solutions: \(x_1^ * = \left( {0,0} \right) ,\, x_2^ * = \left( {3,\sqrt{3} } \right) ,\, x_3^ * = \left( {4,0} \right) \). in all cases \(f\left( {x_i^ * } \right) = - 1,\, i = 1,2,3\).
-
5.
Test function 5
$$\begin{aligned} \begin{array}{ll} \min &{} f = \left( {4 - 2.1x_1^2 + \frac{1}{3}x_1^4} \right) x_1^2 + {x_1}{x_2} + \left( { - 4 + 4x_2^2} \right) x_2^2 \\ \hbox {Subject to:} &{} - 3 \le {x_1} \le 3,\, - 2 \le {x_2} \le 2. \end{array} \end{aligned}$$The problem has two global solutions: \(x_1^ * = \left( { - 0.0898,0.7126} \right) ,\, x_2^ * = \left( {0.0898, - 0.7126} \right) \), and \(f\left( {x_1^ * } \right) = f\left( {x_2^ * } \right) = - 1.0316\).
-
6.
Test function 6
$$\begin{aligned} \begin{array}{ll} \min &{} f = x_1^{0.6} + x_2^{0.6} - 6{x_1} - 4{x_3} + 3{x_4} \\ \hbox {Subject to:} &{} - 3{x_1} + {x_2} - 3{x_3} = 0,\, {x_1} + 2{x_3} \le 4,\, {x_2} + 2{x_4} \le 4,\, {x_1} \le 3,\\ &{} {x_4} \le 1,\, {x_i} \ge 0,\, i = 1,2,3,4. \end{array} \end{aligned}$$The best known global solution is \({x^ * } = \left( {\frac{4}{3},4,0,0} \right) \), and \(f\left( {{x^ * }} \right) = - 4.5142\).
-
7.
Test function 7
$$\begin{aligned} \begin{array}{ll} \min &{} f = 6.5{x_1} - 0.5x_1^2 - {x_2} - 2{x_3} - 3{x_4} - 2{x_5} - {x_6} \\ \hbox {Subject to:} &{} {x_1} + 2{x_2} + 8{x_3} + {x_4} + 3{x_5} + 5{x_6} \le 16,\, - 8{x_1} - 4{x_2}\\ &{}\quad - 2{x_3} + 2{x_4} + 4{x_5} - {x_6} \le - 1\\ &{} 2{x_1} + 0.5{x_2} + 0.2{x_3} - 3{x_4} - {x_5} - 4{x_6} \le 24,\, 0.2{x_1}\\ &{}\quad + 2{x_2} + 0.1{x_3} - 4{x_4} + 2{x_5} + 2{x_6} \le 12\\ &{} - 0.1{x_1} - 0.5{x_2} + 2{x_3} + 5{x_4} - 5{x_5} + 3{x_6} \le 3,\, {x_4} \le 1,\\ &{}\quad {x_5} \le 1,\, {x_6} \le 2,\, {x_i} \ge 0,\, i = 1,2,3,4,5,6. \end{array} \end{aligned}$$The global solution is \({x^ * } = \left( {0,6,0,1,1,0} \right) \), and \(f\left( {{x^ * }} \right) = - 11\).
-
8.
Test function 8
$$\begin{aligned} \begin{array}{ll} \min &{} f = \sum \nolimits _{j = 1}^{10} {{x_j}\left( {c_j} + \ln \frac{x_j}{\sum \nolimits _{i = 1}^{10} {{x_i}}} \right) } \\ \hbox {subject to:} &{} {x_1} + 2{x_2} + 2{x_3} + {x_6} + {x_{10}} = 2,\, {x_4} + 2{x_5} + {x_6} + {x_7} = 1, \\ &{} {x_3} + {x_7} + {x_8} + 2{x_9} + {x_{10}} = 1,\, {x_i} \ge 0.000001,\, i = 1, \cdots ,10. \end{array} \end{aligned}$$Where \(c = ( - 6.089, - 17.164, - 34.054, - 5.914, - 24.721, - 14.986, - 24.1, - 10.708, - 26.663, - 22.179)\). The global solution is
$$\begin{aligned} {x^ * }&= (0.04034785,0.15386976,0.77497089,0.00167479,0.48468539,\\&\quad 0.00068965,0.02826479,0.01849179,0.03849563,0.10128126),\\&\quad \hbox { and } f\left( {{x^ * }} \right) = - 47.760765. \end{aligned}$$ -
9.
Test function 9
$$\begin{aligned} \begin{array}{ll} \min &{} f\left( x \right) = \sum \nolimits _{k = 0}^4 \left( 2.3{x_{3k + 1}} + 0.0001x_{3k + 1}^2 + 1.7{x_{3k + 2}}\right. \\ &{}\quad \quad \quad \quad \ \left. +\, 0.0001x_{3k + 2}^2 + 2.2{x_{3k + 3}} + 0.0001x_{3k + 3}^2\right) \\ \hbox {subject to:} &{} 0 \le {x_{3k + 1}} - {x_{3k - 2}} + 7 \le 13,\, 0 \le {x_{3k + 2}} - {x_{3k - 1}} + 7 \le 14 \\ &{} 0 \le {x_{3k + 3}} - {x_{3k}} + 7 \le 13,\, {x_1} + {x_2} + {x_3} - 60 \ge 0 \\ &{} {x_4} + {x_5} + {x_6} - 50 \ge 0, {x_7} + {x_8} + {x_9} - 70 \ge 0 \\ &{} {x_{10}} + {x_{11}} + {x_{12}} - 85 \ge 0, {x_{13}} + {x_{14}} + {x_{15}} - 100 \ge 0 \\ &{} 8 \le {x_1} \le 21, 43 \le {x_2} \le 57, 3 \le {x_3} \le 16 \\ &{} 0 \le {x_{3k + 1}} \le 90, 0 \le {x_{3k + 2}} \le 120, 0 \le {x_{3k + 3}} \le 60, k = 1, \cdots ,4 \end{array} \end{aligned}$$The global solution is
$$\begin{aligned} {x^ * } = \left( {8,49,3,1,56,0,1,63,6,3,70,12,5,77,18} \right) , \hbox { and } f\left( {{x^ * }} \right) = 664.82045. \end{aligned}$$
Rights and permissions
About this article
Cite this article
Huajun, Z., Jin, Z. & Hui, L. A method combining genetic algorithm with simultaneous perturbation stochastic approximation for linearly constrained stochastic optimization problems. J Comb Optim 31, 979–995 (2016). https://doi.org/10.1007/s10878-014-9803-4
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10878-014-9803-4