Abstract
We study a class of chance constrained programs (CCPs) where the underlying distribution is modeled by a Gaussian mixture model. As the original work, Hu et al. (IISE Trans 54(12):1117–1130, 2022. https://doi.org/10.1080/24725854.2021.2001608) developed a spatial branch-and-bound (B &B) algorithm to solve the problems. In this paper, we propose an enhanced procedure to speed up the computation of B &B algorithm. We design an enhanced pruning strategy that explores high-potential domains and an augmented branching strategy that prevents redundant computations. We integrate the new strategies into original framework to develop an enhanced B &B algorithm, and illustrate how the enhanced algorithm improves on the original approach. Furthermore, we extend the enhanced B &B framework to handle the CCPs with multiple chance constraints, which is not considered in the previous work. We evaluate the performance of our new algorithm through extensive numerical experiments and apply it to solve a real-world portfolio selection problem.
Similar content being viewed by others
Data availability
Not applicable.
Notes
Data available at http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.
References
Agnew, N. H., Agnew, R. A., Rasmussen, J., et al. (1969). An application of chance constrained programming to portfolio selection in a casualty insurance firm. Management Science, 15(10), B-512-B-520. https://doi.org/10.1287/mnsc.15.10.B512
Anis, H. T., & Kwon, R. H. (2022). Cardinality-constrained risk parity portfolios. European Journal of Operational Research, 302(1), 392–402. https://doi.org/10.1016/j.ejor.2021.12.045
Bhagat, K., Dai, C., Ye, S., et al. (2022). Stochastic energy management strategy of smart building microgrid with electric vehicles and wind-solar complementary power generation system. Journal of Electrical Engineering & Technology, 18(1), 147–166. https://doi.org/10.1007/s42835-022-01193-1
Bishop, C. M. (2006). Pattern recognition and machine learning. Springer.
Bremer, I., Henrion, R., & Möller, A. (2015). Probabilistic constraints via SQP solver: Application to a renewable energy management problem. Computational Management Science, 12(3), 435–459. https://doi.org/10.1007/s10287-015-0228-z
Charnes, A., & Cooper, W. W. (1959). Chance-constrained programming. Management Science, 6(1), 73–79. https://doi.org/10.1287/mnsc.6.1.73
Charnes, A., Cooper, W. W., & Symonds, G. H. (1958). Cost horizons and certainty equivalents: An approach to stochastic programming of heating oil. Management Science, 4(3), 235–263. https://doi.org/10.1287/mnsc.4.3.235
Chen, Z., Peng, S., & Lisser, A. (2020). A sparse chance constrained portfolio selection model with multiple constraints. Journal of Global Optimization, 77(4), 825–852. https://doi.org/10.1007/s10898-020-00901-3
Chen, Z., Peng, S., & Liu, J. (2018). Data-driven robust chance constrained problems: A mixture model approach. Journal of Optimization Theory and Applications, 179(3), 1065–1085. https://doi.org/10.1007/s10957-018-1376-4
Frick, D., Sessa, P. G., Wood, T. A., et al. (2019). Exploiting structure of chance constrained programs via submodularity. Automatica, 105, 89–95. https://doi.org/10.1016/j.automatica.2019.03.027
Geng, X., & Xie, L. (2019). Data-driven decision making in power systems with probabilistic guarantees: Theory and applications of chance-constrained optimization. Annual Reviews in Control, 47, 341–363. https://doi.org/10.1016/j.arcontrol.2019.05.005
Hanasusanto, G. A., Roitch, V., Kuhn, D., et al. (2017). Ambiguous joint chance constraints under mean and dispersion information. Operations Research, 65(3), 751–767. https://doi.org/10.1287/opre.2016.1583
Hantoute, A., Henrion, R., & Pérez-Aros, P. (2018). Subdifferential characterization of probability functions under Gaussian distribution. Mathematical Programming, 174(1–2), 167–194. https://doi.org/10.1007/s10107-018-1237-9
Henrion, R. (2007). Structural properties of linear probabilistic constraints. Optimization, 56(4), 425–440. https://doi.org/10.1080/02331930701421046
Henrion, R., & Möller, A. (2012). A gradient formula for linear chance constraints under gaussian distribution. Mathematics of Operations Research, 37(3), 475–488. https://doi.org/10.1287/moor.1120.0544
Hu, Z. L., Sun, W. J., & Zhu, S. S. (2022). Chance constrained programs with Gaussian mixture models. IISE Transactions, 54(12), 1117–1130. https://doi.org/10.1080/24725854.2021.2001608
Kataoka, S. (1963). A stochastic programming model. Econometrica, 31, 181–196.
Lagoa, C. M., Li, X., & Sznaier, M. (2005). Probabilistically constrained linear programs and risk-adjusted controller design. SIAM Journal on Optimization, 15(3), 938–951. https://doi.org/10.1137/s1052623403430099
Lodi, A., Malaguti, E., Nannicini, G., et al. (2019). Nonlinear chance-constrained problems with applications to hydro scheduling. Mathematical Programming, 191(1), 405–444. https://doi.org/10.1007/s10107-019-01447-3
Masmoudi, M., & Ben Abdelaziz, F. (2017). A chance constrained recourse approach for the portfolio selection problem. Annals of Operations Research, 251(1–2), 243–254. https://doi.org/10.1007/s10479-015-1844-2
Maugis-Rabusseau, C., & Michel, B. (2013). Adaptive density estimation for clustering with Gaussian mixtures. ESAIM-Probability and Statistics, 17, 698–724. https://doi.org/10.1051/ps/2012018
Miller, B. L., & Wagner, H. M. (1965). Chance constrained programming with joint constraints. Operations Research, 13(6), 930–945. https://doi.org/10.1287/opre.13.6.930
Nemirovski, A., & Shapiro, A. (2006a). Convex approximations of chance constrained programs. SIAM Journal on Optimization, 17(4), 969–996. https://doi.org/10.1137/050622328
Nemirovski, A., & Shapiro, A. (2006b). Scenario approximations of chance constraints (pp. 3–47). Springer.
Pang, X., Zhu, S., & Hu, Z. (2023) Chance constrained program with quadratic randomness: A unified approach based on gaussian mixture distribution https://doi.org/10.48550/arXiv.2303.00555
Prékopa, A. (2003). Probabilistic programming. In Handbooks in operations research and management science (Vol. 10, pp. 267–351) Elsevier. https://doi.org/10.1016/s0927-0507(03)10005-9
Schildbach, G., Fagiano, L., & Morari, M. (2013). Randomized solutions to convex programs with multiple chance constraints. SIAM Journal on Optimization, 23(4), 2479–2501. https://doi.org/10.1137/120878719
van Ackooij, W. (2020). A discussion of probability functions and constraints from a variational perspective. Set-Valued and Variational Analysis, 28(4), 585–609. https://doi.org/10.1007/s11228-02”0-00552-2
van Ackooij, W., Aleksovska, I., & Munoz-Zuniga, M. (2018). (Sub-)Differentiability of probability functions with elliptical distributions. Set-Valued and Variational Analysis, 26(4), 887–910. https://doi.org/10.1007/s11228-017-0454-3
van Ackooij, W., & Henrion, R. (2014). Gradient formulae for nonlinear probabilistic constraints with Gaussian and Gaussian-like distributions. SIAM Journal on Optimization, 24(4), 1864–1889. https://doi.org/10.1137/130922689
Vielma, J. P., Ahmed, S., & Nemhauser, G. L. (2008). A lifted linear programming branch-and-bound algorithm for mixed-integer conic quadratic programs. INFORMS Journal on Computing, 20(3), 438–450. https://doi.org/10.1287/ijoc.1070.0256
Wächter, A., & Biegler, L. T. (2005). On the implementation of an interior-point filter line-search algorithm for large-scale nonlinear programming. Mathematical Programming, 106(1), 25–57. https://doi.org/10.1007/s10107-004-0559-y
Wang, S., Li, J., & Mehrotra, S. (2021). Chance-constrained multiple bin packing problem with an application to operating room planning. INFORMS Journal on Computing, 33(4), 1661–1677. https://doi.org/10.1287/ijoc.2020.1010
Zeevi, A. J., & Meir, R. (1997). Density estimation through convex combinations of densities: Approximation and estimation bounds. Neural Networks, 10(1), 99–109. https://doi.org/10.1016/s0893-6080(96)00037-8
Funding
Jinxiang Wei and Zhaolin Hu were supported by the National Natural Science Foundation of China under Grant No. 72071148. Shushang Zhu was supported by the National Natural Science Foundation of China under Grant No. 72271250. Jun Luo was supported in part by the National Natural Science Foundation of China under Grant No. 72031006.
Author information
Authors and Affiliations
Contributions
JW: Methodology, Software, Investigation, Writing (original draft). ZH: Conceptualization, Methodology, Investigation, Writing (review and editing). JL: Conceptualization, Investigation, Writing (review and editing). SZ: Conceptualization, Methodology, Investigation, Writing (review and editing).
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Code availability
The code used in this paper is available from the corresponding author on reasonable request.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: The B &B algorithm
Appendix B: Proof of Proposition 3
The proof of Proposition 3 is adapted from the proof of Proposition 3 in Pang et al. (2023).
For any \({\varvec{x}}\in {\mathcal {X}}\) and \(y_{pj}\in \Delta ^0\),
where \({\hat{y}}_{pj}\in l _{y^0_{pj}}\). The last inequality holds since the derivative of \(\Phi ^{-1}(y)\) is \(\frac{1}{\phi \left( \Phi ^{-1}(y)\right) }\) and \(\lambda ^{\star } I-\Sigma _j\) is positive semi-definite and thus \(\lambda ^{\star } f({\varvec{x}})_{\max }^2 \ge \lambda ^{\star }\Vert f^p({\varvec{x}})\Vert ^2 \ge f^p({\varvec{x}})^{\top }\Sigma _j f^p({\varvec{x}})\).
Suppose that a domain \(\Delta =\prod _{p=1}^{P}\prod _{j=1}^{K} l _{pj}\) satisfies
If Problem (13) formulated over \(\Delta \) is feasible, denote the optimal solution as \(({{\varvec{x}}}^c,{{\varvec{y}}}^c)\). Since \(F_{pj}({{\varvec{x}}}^c,{{{\overline{y}}}}_{pj})-F_{pj}({{\varvec{x}}}^c,{{\underline{y}}}_{pj})\le \epsilon \) for all \(j\in {\mathbb {K}}\) and \(p\in {\mathbb {P}}\), the solution is an \(\epsilon \)-optimal solution. If Problem (13) formulated over \(\Delta \) is infeasible, Problem (2) formulated over \(\Delta \) must be infeasible. In both cases, the domain \(\Delta \) will not be divided into two smaller subdomain.
According to the original B &B algorithm, the initial domain \(\Delta ^0\) will not be divided into two smaller rectangles from the middle point of the longest edge if this edge is not larger than \(\frac{\epsilon }{d^{\star }f({\varvec{x}})_{\max }\sqrt{\lambda ^{\star }}}\). Now, we consider the p-jth edge of a subdomain with the length of \(\vert l _{y_{pj}}\vert ={{\overline{y}}}_{pj}-{\underline{y}}_{pj}> \frac{\epsilon }{d^{\star }f({\varvec{x}})_{\max }\sqrt{\lambda ^{\star }}}\) and then the partitioning continues. At the beginning of the B &B algorithm, we note that there is only one domain with \(\vert l _{y^0_{pj}}\vert \) edge, i.e., the initial domain \(\Delta ^0\). According to the original branching rule, partitioning in the middle to generate all domains with the length of this edge equal to \(\left( \frac{1}{2}\right) ^{1}\vert l _{y^0_{pj}}\vert \), it is only required \(2^0=1\) time of division. There are two subdomains generated with \(\left( \frac{1}{2}\right) ^{1}\vert l _{y^0_{pj}}\vert \). If all subdomains on this edge are equal to \(\left( \frac{1}{2}\right) ^2\vert l _{y^0_{pj}}\vert \), it is required \(2^0+2^1=3\) times of division. It is not difficult to conclude that \(2^n-1\) times of division can obtain all subdomains that the length of this edge is equal to \(\frac{1}{2}^{n}\vert l _{y^0_{pj}}\vert \).
Let \(n^{\star }=\log _2\left\lfloor \frac{d^{\star }f({\varvec{x}})_{\max }\sqrt{\lambda ^{\star }}\left| { l _{y^0_{pj}}}\right| }{\epsilon }+1\right\rfloor \). Then the subdomians with the length of the p-jth edge are equal to
Therefore this edge is no longer divided and the corresponding number of the partition is
Note that the each division corresponds to one computation of an edge of \(\Delta ^0\) for Problem (13). This implies that the number of division is equal to the number of edges. Therefore, the number of Problem (13) solved is at most
The proof is completed.
Appendix C: Comparing the performance of B &B and enhanced B &B algorithms
In Fig. 8, the y-axis represents the computation time, with the blue dotted line and orange line marked with stars stand for the B &B algorithm and enhanced B &B algorithm, respectively. The computation times of the B &B algorithm and the enhanced B &B algorithm under \(K\in \{2, 5\}\), \(d=50\) are \(\{0.2\ \textrm{s}, 0.5\ \textrm{s}\}\) and \(\{30.1\ \textrm{s}, 171.4\ \textrm{s}\}\) separately. The two lines almost coincide, indicating that each iteration of the two algorithms take roughly the same amount of time. However, the enhanced B &B algorithm avoids solving for redundant solutions, thus the number of iterations is largely reduced, resulting in significant advantage in computation time.
Appendix D: The statistics information of 48 Industry Portfolios
In Table 12, the abbreviations “M”, “S.D.”, “SK” and “K” refer to “Mean”, “Standard deviation”, “Skewness” and “Kurtosis”. We employ the normal distribution to model the distribution of the average monthly return rates of the 48 Industry Portfolio, as depicted in Fig. 9. It is noteworthy that the probability density function (pdf) manifests a long-tail characteristic, a feature not entirely captured by a normal distribution fit.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Wei, J., Hu, Z., Luo, J. et al. Enhanced branch-and-bound algorithm for chance constrained programs with Gaussian mixture models. Ann Oper Res (2024). https://doi.org/10.1007/s10479-024-05947-0
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10479-024-05947-0