Evaluation of fuzzy rule bases under delayed reinforcement

This article concerns the problem and solution of judging fuzzy rule bases according to environmental reinforcements. We propose an on-line, incremental credit assignment algorithm, which takes environmental reinforcement as input and assigns credit to individual rules. The proposed approach adopts a simple updating policy based on recency-weighted average, and demands only small amount of memory. We also contribute to the problem of delayed reinforcement. In the case of delayed reinforcement, the state preference function is constructed iteratively during the exploration phase