Skip to main content
Log in

An accelerated L-shaped method for solving two-stage stochastic programs in disaster management

  • S.I.: Decomposition Methods for Hard Optimization Problems
  • Published:
Annals of Operations Research Aims and scope Submit manuscript

Abstract

Mitigating the disastrous effects of natural disasters by performing preparation activities is one of the main purposes of relief organizations. However, the high degree of uncertainty associated with disasters impedes the work of aid agencies considerably. In this regard, two-stage stochastic programs are often used in the relevant literature to support decision making in these situations. An accelerated L-shaped method is proposed in this work, which solves realistic large-scale two-stage stochastic problems within a reasonable time-frame, allowing relief organizations to react to short-term forecasts, as e.g. available in case of hurricanes or floods. In particular, computation times needed for solving the resulting sub-problems via a specialized interior-point method are significantly reduced by exploiting the specific structure of second-stage constraints. To show the superiority of this approach with respect to solution times, a realistic large-scale case study is developed for America’s hurricane-prone south-east coast. The accelerated L-shaped method outperforms the standard L-shaped method significantly whereas a commercial solver failed to solve the case study within an acceptable time-frame.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Subscribe and save

Springer+ Basic
$34.99 /Month
  • Get 10 units per month
  • Download Article/Chapter or eBook
  • 1 Unit = 1 Article or 1 Chapter
  • Cancel anytime
Subscribe now

Buy Now

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6

Similar content being viewed by others

Notes

  1. Warm-starting is to use the solution to the previous problem as a starting point for the next problem. Especially simplex algorithms benefit from such warm-start points and can find the next solution after only few iterations. However, developing efficient warm-start strategies for interior-point methods is an active research area nowadays, see e.g. Cay et al. (2017).

  2. Note that it is very unlikely that a scenario will occur exactly as predefined. Moreover, penalty costs for unsatisfied demand are only fictitious and used to minimize unmet demand.

  3. The reader is referred to Birge and Louveaux (2011) for a thorough description of two-stage stochastic programs in general.

  4. Note that simplex multipliers are also known as shadow prices which are available, e.g. in the final tableau of the simplex algorithm (Diwekar 2008, p. 23). However, the dual sub-problem will be important for the primal–dual interior-point method described in the next section.

  5. For two-stage stochastic models where no complete recourse is given, so-called feasibility cuts have to be included to the master problem in addition to the optimality cuts (Birge and Louveaux 2011, Ch 5).

  6. The approach is presented for one sub-problem such that the subscript s is omitted for now.

  7. See Ferris et al. (2007) for a general description of the Newton method for interior-point algorithms.

  8. An approach for setting the step length \(\alpha \) can be found in Ferris et al. (2007).

  9. A detailed description of the algorithm and how to reduce \(\mu ^{\mathbf {k}}\), can be found in Ferris et al. (2007) or Wright (1997).

  10. In such cases the matrix \(W D^{-1} W^T\) is often dense leading to an increased effort when computing its inverse.

  11. Distances are determined by the ArcGIS software.

  12. The built-in function linprog uses a predictor-corrector primal–dual interior-point method by default.

  13. Such termination criteria are common for interior-point methods (Wright 1997, p. 226).

  14. By default, Gurobi solves the MIP root node via the dual simplex method and run the barrier and simplex method on multiple threads concurrently for LP problems (Gurobi Optimization 2017).

  15. For comparison reasons Gurobi was terminated if the relative optimality gap fell below 0.009. However, the relative optimality gap jumped from 0.173 to 0.004 in the last iteration of Gurobi.

  16. As mentioned above, column ’L-Shaped’ in Table 6 refers to the method where MATLAB’s linprog and ’Accelerated L-Shaped’ where SIPM is used for solving the sub-problems. Since computation times for solving the master problem in each iteration are independent of the sub-problems and insignificant in comparison to the solution times of the sub-problems, the overall time for the L-shaped method consists mainly of computation times for the sub-problems. Therefore, solution times given in Table 7 have a major influence on the overall solution times of the L-shaped method as given in Table 6.

References

Download references

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Emilia Grass.

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Grass, E., Fischer, K. & Rams, A. An accelerated L-shaped method for solving two-stage stochastic programs in disaster management. Ann Oper Res 284, 557–582 (2020). https://doi.org/10.1007/s10479-018-2880-5

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10479-018-2880-5

Keywords

Navigation