Green Multi-Objective Simulation Optimization in Parallel Computing Environments

This project will advance US prosperity, welfare, and science by developing efficient and rigorous techniques for solving complex decision-making problems when multiple conflicting objectives must be weighed, the value of potential solutions is uncertain, and error is costly. Such decision-making problems occur in all sectors of the US economy, including health and defense, and are very difficult to solve, both conceptually and algorithmically.

This award supports research that investigates a novel formulation of complex, multi-objective decision-making problems under uncertainty aimed at facilitating tradeoffs among multiple objectives while allowing for statistically valid algorithms that achieve high efficiency via novel comparison techniques, recycling data, computer simulation, and parallel computing. Through educational and outreach activities, including providing implementations of the approaches in a free public domain, this award will increase US productivity and help decision makers manage tradeoffs and meet contractual and managerial obligations when solving complex problems under uncertainty.

This research will develop a novel approach to large-scale multi-objective simulation optimization that involves placing constraints on all performance measures. The constraints are stochastic in that performance measures can be estimated through stochastic simulation, and subjective in that their thresholds can be adjusted as necessary. Subjective constraints allow for the development of a multi-pass approach that uses feasibility with respect to more strict thresholds to prune inferior systems.

To efficiently identify the preferred system with statistical validity, solution procedures that use multiple indifference-zone parameters to reduce conservativeness while avoiding the collection of unnecessary observations for detecting small differences between performance measures and thresholds. For additional efficiency, the new procedures employ green simulation with recycled observations for different thresholds.

This approach also avoids direct pairwise comparisons between systems, leading to reduced communication and synchronization among processors and facilitating parallel computing implementations.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.