CAREER: Extensions of Stochastic Programming: Models, Algorithms, and Applications

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

This Faculty Early Career Development (CAREER) award is to study new models, algorithms, and applications of stochastic programs with alternative risk-averse objectives, subjective probabilities, and decision-dependent uncertainties. Although stochastic programming has now evolved as a viable paradigm for planning and decision-making under uncertainty, much of the progress in this area has been made at the expense of some simplifying assumptions. For example, traditional stochastic programming is concerned with optimizing an expected objective function. Other common assumptions include precise knowledge of a static underlying probability distribution. However, these risk ignoring assumptions can be quite undesirable in many practical applications. Unfortunately, these generalizations typically lead to non-convex optimization models. Consequently, traditional decomposition algorithms for (convex) stochastic programs are inapplicable. The research will investigate the integration of decomposition principles within non-convex optimization algorithms in order to attack large-scale instances of these general stochastic programs. The developed concepts will be applied to planning problems in important economic sectors such as process industries, engineering design, and utility industries. The educational component of this career development plan is aimed at popularizing stochastic programming based planning and decision making in engineering education and practice. Towards this goal, user-friendly stochastic programming modeling and solver tools, electronic tutorials, and real world case studies will be developed. The operations research community recognizes Stochastic Programming as a valuable quantitative technique for decision support in the face of uncertainty. However, this tool has not achieved widespread use in practical planning and decision-making. Two reasons for this are: traditional stochastic programming models can often be overly simplistic for real-life applications and the lack of exposure to practical stochastic programming concepts in engineering education. The proposed research program will extend stochastic programming paradigm beyond some of the traditional impractical assumptions. These generalizations will require the development of entirely new stochastic programming models and algorithms, and their application to relevant practical problems. On the education side, user friendly stochastic programming solver, electronic tutorials, and industrial case-studies will be developed to facilitate the introduction of applied stochastic programming concepts in undergraduate and graduate engineering education.