Effective retrieval and new indexing method for case based reasoning: Application in chemical process design

In this paper we try to improve the retrieval step for case based reasoning for preliminary design. This improvement deals with three major parts of our CBR system. First, in the preliminary design step, some uncertainties like imprecise or unknown values remain in the description of the problem, because they need a deeper analysis to be withdrawn. To deal with this issue, the faced problem description is soften with the fuzzy sets theory. Features are described with a central value, a percentage of imprecision and a relation with respect to the central value. These additional data allow us to build a domain of possible values for each attributes. With this representation, the calculation of the similarity function is impacted, thus the characteristic function is used to calculate the local similarity between two features. Second, we focus our attention on the main goal of the retrieve step in CBR to find relevant cases for adaptation. In this second part, we discuss the assumption of similarity to find the more appropriated case. We put in highlight that in some situations this classical similarity must be improved with further knowledge to facilitate case adaptation. To avoid failure during the adaptation step, we implement a method that couples similarity measurement with adaptability one, in order to approximate the cases utility more accurately. The latter gives deeper information for the reusing of cases. In a last part, we present a generic indexing technique for the base, and a new algorithm for the research of relevant cases in the memory. The sphere indexing algorithm is a domain independent index that has performances equivalent to the decision tree ones. But its main strength is that it puts the current problem in the center of the research area avoiding boundaries issues. All these points are discussed and exemplified through the preliminary design of a chemical engineering unit operation. (C) 2010 Elsevier Ltd. All rights reserved.