JournalVolume 4, №1Optimization under uncertainty via random sampling of scenarios, II Louis (pages 20-28) Anthony Cox, Djangir A. Babayev Abstract. In many decision and risk optimization problems under uncertainty, the set of possible future scenarios determining the eventual consequences of current actions cannot be explicitly specified or enumerated in advance, but can only be generated by stochastic simulation modeling of the system or situation being acted on. This is the second paper under the same name (Cox and Babayev, 2004) exploring how far one can get in such problem by the conceptually simple device of randomly sampling (or generating) a set of deterministic scenarios (each of which resolves all uncertainties by sampling values for all random variables) and then solving the resulting sampled deterministic optimization problems. The result is a general, computationally practical approach, SORS - Stochastic Optimization Via Random Sampling of Scenarios, to solving such optimization problems under uncertainty that enjoys the theoretical benefits of sampling-based methods and that appears to work surprisingly well in practice based on computational experiments on several well-known problems. The method can fully exploit the immense variety of existing methods, algorithms and software for solving deterministic optimization problems, while extending them to a broad class of optimization problems under uncertainty. The SORS method is applicable whenever a computationally efficient method is available for solving the sampled deterministic versions of the original stochastic problem. The SORS method provides solutions that asymptotically converge with probability 1 to the exact solutions as the number of sampled scenarios increases. Computational experiments demonstrate the high efficiency of the method on a variety of stochastic optimization problems.
|
|