Case-based modification for optimization agents: AGENT-OPT

For the effective implementation of an inter-organizational supply chain on the Web, many optimization model agents need to be embedded in the distributed software agents. For instance, many suppliers make requests to a delivery scheduler who manages a model warehouse at the e-hub. The scheduler deals with the scheduling of many truckers and each trucker's agent must have its own routing optimization models. Since the formulations in the model warehouse vary depending upon the requirements, it is impossible to formulate all combinations in advance. Therefore, we need a case-based model modification scheme that can generate the required formulation from the semantically specified requirement in the agent communication language. This research deals with the issues of the architecture of an optimization model agent system AGENT-OPT, modeling request language in XML, optimization model representation in semantic-level objects using UNIK-OPT, a method of selecting a base model, an optimization model modification language (OMML), and rule-based modification reasoning. The approach is applied to the delivery scheduling to study the effect of base model selection policies on the modification effort. To determine whether to start with a primitive model, full model, or the most similar model, we experimented with the sensitivity of proximity to the primitive model on 24 cases and discovered the threshold for choosing the most efficient base model.