Optimization-Based Decision Support Software for a Team-in-the-Loop Experiment: Multilevel Asset Allocation

摘要

Motivated by the Navy's interest in decision support tools that augment planning activities within a maritime operations center (MOC), we have developed a multilevel resource allocation model that is capable of interacting with human planners to dynamically allocate hierarchically-organized assets to process interdependent tasks in order to accomplish mission objectives. The planning problem is formulated as a mixed-integer nonlinear programming (MINLP) problem of minimizing the overall difference between the human-specified desired task accuracy performance criteria and the expected performance outcomes, the latter being based on how well the assigned resources match the required resources, subject to a number of real-world planning constraints. To solve the resulting large-scale MINLP problem, we propose two methods: 1) a Lagrangian relaxation method that solves the multilevel asset allocation problem with a measure of sub-optimality in terms of an approximate duality gap and 2) a dynamic list planning heuristic algorithm that provides high-quality sub-optimal solutions rapidly (less than 10 s for the scenarios considered here). Finally, we verify our methods using realistic MOC planning scenarios, provide a comparative evaluation of the performance measures of the two proposed methods, and investigate the value of information via human-in-the–loop experiments.