On How Solution Populations Can Guide Revision of Model Parameters

Post-evaluation analysis of the model of a constrained optimization problem is conducted after obtaining preliminary optimal or heuristically good solutions. The primary goal of post-evaluation analysis is to reconsider assumptions made in the model in the light of information generated while finding the good solutions as well as information not previously detailed in the model. We seek extensions of the techniques presently available for the special case of linear programming problems because these special problems allow excellent post-evaluation analysis as a side-effect of seeking solutions. Unfortunately, more general problem solvers presently provide little if any information for post-evaluation analysis. We consider general metaheuristic methods that evolve populations of settings of the decision variables. These methods can contribute greatly to reconsideration of modeling assumptions. This is because the evolving populations taken in total provide a great number of samples for conducting post-evaluation analysis in a data-driven fashion. This is a very general claim. It is illustrated in this paper by a single, rather simple, constrained optimization problem.