Considerin gthe Dynamic in Knowledge Based Configuration

Learning from previous solutions could be a key for improving both the solution process and the quality of the solution. Unfortunately knowledge (i.e. components) in technical configuration domains changes quickly and learning brings with it a certain amount of conservatism. For example a component that was learned to be good for a certain task should no longer be chosen, if a newer version appears on the market. On the other hand a certain amount of conservatism is often desired since uncontrolled innovation is as a rule also detrimental, i.e. the newer component mentioned above is not obviously better. This article presents Relevant Knowledge First (RKF) as a method and heuristic respectively. It tries to find a good compromise between conservatism and innovation based on statistical values and aging of knowledge.