Heuristics Acquisition from Problem Solving Process and Knowledge Representation by State-Action Network : Application to Two-Dimensional Packing Problems

Abstract

This paper proposes a new model and method for knowledge acquisition for planning problems in manufacturing, such as production planning, packing problems, and so forth, based on a problem solving process of a planner. In our proposed model, knowledge used for solving the objective problem is obtained through construction of state action network. A state action network consists of states describing some phases of the problem solving process and planner's actions in each phase. Moreover, reinforcement learning is used to refine the obtained state action network and to evaluate the problem solving process. We apply the proposed method to two dimensional packing problems and demonstrate its applicability and effectiveness through some computational experiments.