A Generic Planning Method with Knowledge Acquisition Function

在提出的状态选择规划方法的基础上，设计了几种规划知识的自动学习方法。它们使我们能够构建一个能够处理一般规划问题的规划系统。将典型的归纳学习方法ID3应用于规划知识获取，提出了能够学习规划知识的训练数据格式。针对归纳学习存在如何收集大量训练数据的问题，提出了递归式训练数据收集方法。该方法将知识划分为多个阶段，从最新阶段的知识逐级递归再现训练数据的收集和归纳学习的应用，便于训练数据的收集。将这些方法应用于流程车间问题和作业车间问题。通过实验证明，这些方法可以产生比人类知识更好的知识。规划知识的自动学习方法减少了规划的时间，因此对于需要在短时间内对大规模问题进行规划的CIM来说，这种方法是有效的。如上所述，我们通过引入规划知识的自动学习特性，实现了一个全面支持规划的通用规划体系结构。

We devised several kinds of automatic learning method for planning knowledge which are based on Status Selection Planning Method that we had proposed. They enable to constract a planning systems which can deal with generic planning problems. ID3, which is one of typical inductive learning methods, was applied to planning knowledge acquisition, and format of training data, which enables to learn planning knowledge was proposed. Because the inductive learning have a problem of how to gather a great deal of training data, recursive type training data gathering method was proposed. In the method, knowledge is divided into several stages of knowledge, gathering training data and application of inductive learning are recursively repreated one by one from the latest stage knowledge, in which it is easy to gathering training data. These methods were applied to flow shop problems and job shop problems. From the experimentation, it was confirmed that these methods could generate the superior knowledge than human knowledge. The automatic learning methods for planning knowledge decreases time for planning, so that it is obvious that these method is effective for CIM,in which short time planning for a large scale problem is necessary.As mention above, by introducing automatic learning feature of planning knowledge, we actualized a generic planning architecture which supported planning comprehensively.