Analyzing long-chain rules extracted from a learning classifier system

A learning classifier system (LCS) is a machine learning system with components of a production system, a reinforcement learning mechanism, and a rule generation mechanism by genetic algorithms (GAs). This paper presents a novel method that both extracts effective chunks of knowledge (long-chain rules) frequently used by an LCS and identifies the features of a given dynamic environment. The task is critical when we use a rule-based discrete event simulation system for manufacturing scheduling problems. In such a case, we must detect the features of environmental conditions about the manufacturing system and identify the meanings of the long-chain rules in order to apply the acquired rules to the scheduling system in operation. The key idea of the proposed method is that: (1) the global environmental conditions are detected by a simple fuzzy inference mechanism; (2) we keep track of sequences of rule executions and the corresponding objective function values; and (3) a recursive algorithm is used to extract the effective chunks from the sequences by evaluating the objective functions values