UNIK-PMA

Abstract

We attempt to unify an optimization model with a rule-based system via overlapped decision variables. The unified model is a kind of multiobjective decision making (MODM) model that encompasses the objectives in both the optimization model and the rule-based system, and formulated as MODM(1, M). The Post-model Analysis approach is adopted to solve the unified model, and three types of tradeoffs are classified as a structure to support nondominated tradeoffs. Specific structure and methodology to solve the unified model is adopted and developed, including: semantic representation of an optimization model, extraction of relevant rule base, rule-based evaluation of the solution of an optimization model, support of nondominated tradeoffs by automatic model formulation, and constraint generation from rule-based goals. Optimization models must be represented semantically so that an optimization model and a rule-based system can understand each other. The knowledge- assisted optimization modeler UNIK-OPT is adopted to represent an optimization model semantically. The rules relevant to an optimization model are identified by overlapped decision variables and used to evaluate the optimal solution of the optimization model in terms of the objectives in the relevant rule base. Since the index manipulation is one of the intrinsic parts of optimization models, the inference engine of the rule-based system must be able to understand and manipulate indices. So, the backward inference engine UNIK-BWD which has the index manipulation capability is used for rule-based systems. A set of methods to compute the nondominated solutions for each of the three types of tradeoffs is developed. The essence of the methods is automatic model formulation by constraint generation from rule-based goals. The developed methodology is implemented in a Decision Support System called UNIK-PMA (UNIfied Knowledge-Post-Model Analysis) on MS- WINDOWS.