Dynamic similarity metric using fuzzy predicates for case-based planning

Case-based planning (CBP) is a knowledge-based planning technique which develops new plans by reusing its past experience instead of planning from scratch. The task of CBP becomes difficult when the knowledge needed for planning can not be expressed precisely. In this paper, we tackle this issue by modeling imprecise information using fuzzy predicates; and accordingly, we present a dynamic similarity metric for efficient and effective retrieval of relevant cases from a library of cases. We also present weight adaptation algorithm to allow improving the performance of the metric overtime. We use and compare the performance of Tabu search, simulated annealing, and exhaustive search algorithms in instantiating fuzzy predicates to achieve maximum similarity between a new problem and a case. Our experiments show that the proposed metric is sound. The metric along with the adaptation algorithm have been shown to be promising when compared to others. Experiments also show that simulated annealing is more efficient than Tabu search and exhaustive search in fuzzy predicates instantiation.