Finding Effective Courses of Action in a Complex Situation Using Evolutionary Algorithms

摘要

This paper proposes an Evolutionary Algorithm (EA) based approach for finding effective courses of action (COAs) in a complex uncertain situation. The complex situation is modeled using a probabilistic modeling and reasoning framework, referred to as Timed Influence Nets (TINs). The TIN-based framework helps a system modeler in connecting a set of actionable events and a set of desired effects through chains of cause and effect relationships. Once a TIN is built, the optimization task confronted by the modeler is to identify a course of action that would increase the likelihood of achieving the desired effects over a pre-specified time interval. The paper uses Evolutionary Algorithms to accomplish this task. The proposed approach generates multiple COAs that are close enough in terms of achieving the desired effect. The purpose of generating multiple COAs is to give several alternatives to a decision maker. Moreover, the alternate COAs could be generalized based on the relationships that exist among the actions and their execution timings. While determining an effective course of action in a given situation, a system modeler has to consider several temporal/causal constraints that are present in a problem domain. The paper also proposes a constraint specification language that would help a system modeler in specifying these constraints.