A binary encoded incidence matrix representation for KBS Verification

Abstract

An advanced binary encoded incidence matrix representation, to support KBS Verification and Validation (KBS V & V), is described. The representation has two very important advantages over traditional incidence matrix representation techniques. Firstly the storage capacity required by the binary encoded matrices is substantially less (97%) than that required using more conventional incidence matrix representations. Secondly the representation supports much more efficient processing. Both a traditional and the advanced incidence matrix representation are described as built into two KBS V & V systems, IMVER-1 and IMVER-2. The similarity between the two systems provides an appropriate mechanism for comparing the two representations. Two algorithms, one for inference verification and one for structural verification, are analysed with respect to the IMVER representations. Consequently some time complexity measures are determined based on the number of matrix element comparisons required when the algorithms are tested against a range of “balanced” rulebases. The overall conclusion is that the advanced binary representation, as implemented in the IMVER-2 system, when tested against realistically sized rulebases, utilizes significantly fewer (around 96%) comparisons than that required by the traditional representation technique incorporated into the IMVER-1 system.