Supporting Deliberation with Post-Evaluation Analysis of Constrained Optimization Models

Many real life optimization problems cannot be modeled adequately using linear programming. Other techniques such as integer programming and non-linear programming are typically used to solve these problems. However, these techniques provide limited view of the solution space and have serious limitations dealing with the stochastic nature of the constraints. This paper investigates the principles and techniques for providing a systematic, thoroughgoing support for post-evaluation analysis of constrained optimization models that are not (limited to) linear programming models. More specifically, the paper focuses on a challenging class of integer programming problems, the generalized assignment problem (GAP) and information gathered by genetic algorithms. After introducing the larger problem of post-evaluation analysis for constrained optimization models, we examine the performance of random search and two promising implementations of the genetic algorithm (Penalty GA and FI-2Pop GA) on a set of standard GAP test problems. While much work remains to be done to explore the full capabilities of these techniques, we demonstrate that useful information can be obtained through meta-heuristic search techniques and we conduct a systematic comparison of such techniques.