Planning under uncertainty solving large-scale stochastic linear programs

Abstract

For many practical problems, solutions obtained from deterministic models are unsatisfactory because they fail to hedge against certain contingencies that may occur in the future. Stochastic models address this shortcoming, but up to recently seemed to be intractable due to their size. Recent advances both in solution algorithms and in computer technology now allow us to solve important and general classes of practical stochastic problems. We show how large-scale stochastic linear programs can be efficiently solved by combining classical decomposition and Monte Carlo (importance) sampling techniques. We discuss the methodology for solving two-stage stochastic linear programs with recourse, present numerical results of large problems with numerous stochastic parameters, show how to efficiently implement the methodology on a parallel multi-computer and derive the theory for solving a general class of multi-stage problems with dependency of the stochastic parameters within a stage and between different stages. while the use of importance sampling instead of crude Monte Carlo sampling has been recommended by Professor Peter W. Glynn. At the beginning of 1989 the author joined the Department of Operations Research at Stanford University as a visiting scholar from Vienna University of Technology, and started to work closely with Professor Dantzig on the development of techniques for solving large-scale stochastic linear programs. Progress on this topic has led to extensions of the author's visit and to his current position of Senior Research Associate at the Department of Operations Research at Stanford University. It has been an outstanding and influential experience for the author to know and to collaborate with Professor George Dantzig. The author wants to thank Professor Dantzig for everything, all his advice, his outstanding professional support and his great friendship. The author wants to thank Professor Peter Glynn for the many valuable discussions on the topic of planning under uncertainty and importance sampling and his great support. The author is grateful to Professor Michael Saunders and Dr. John Stone for the valuable discussions on the topic of large-scale systems and their important and helpful suggestions concerning previous versions of this paper. The author collaborated with Professor James K. Ho on using parallel processors. The author wishes to thank Professor Ho for this collaboration. Thanks also to David Morton for valuable comments on a previous version of this paper and to Alamuru Krishna, who assisted in preparing some of the test problems. The author is especially grateful to Professor Peter Jansen, his "Doktorvater", for initiating the author's visit to Stanford University and his continuing great support.

Short-term manpower planning for MRT carriage maintenance under mixed deterministic and stochastic demands

“…Taiwan's MRT can use the most suitable numbers of scenarios for their own needs. If the scenario-generation method proposed in this study cannot be used to generate the number of scenarios to represent the population, then other techniques can be used to generate the most suitable number of scenarios (see Kaut and Wallace 2007; scenario reduction by Dupačová et al 2003;Römisch and Heitsch 2003; internal sampling matching by Higle and Sen 1991;Ermoliev and Gaivoronski 1992;Infanger 1994, and so on). This could be a topic of future research.…”

Stochastic Optimization of Grid to Vehicle Frequency Regulation Capacity Bids

“…This is because SLP problems grow exponentially in the number of decision stages. They become very difficult to even approximately solve by using scenario reduction [13] or nested Bender's decomposition [7], [14].…”

“…The constraints in (14) can be thought of as a set of optimality cuts which form a continuous approximate cost to go function of SOC at the next decision epoch. In order to solve for given some state, one must know the inequality constraint coefficients of (14), and , the next conditional expected values of the states, , and the state transition probabilities, , from the last observed exogenous state to the next hour's possible exogenous states. This information can be estimated from historical data as will be described in Section V. Given some state, the recursive equation in Problem 4 can be solved over a continuous space of decisions using linear programming methods.…”

Solving probabilistic programming problems involving multi-choice parameters

“…In the literature of the stochastic linear programming [3,9,10,12,13], various models have been suggested by several researchers. A bibliography has been presented by Stancu-Minasian [15].…”