Rule Based Fuzzy Cognitive Maps (RBFCMs) have been developed for modelling non-monotonic, uncertain, cause-effect systems. However, the standard reasoning and impact accumulation mechanisms developed for RBFCMs assume that the level of variation that a fuzzy set represents is directly linked with the shape of the fuzzy set. It poses a big restriction on how the corresponding fuzzy sets have to be constructed and limits flexibility of knowledge representation. This thesis presents a critical analysis of standard RBFCMs and defines its limitations. To address these limitations a new reasoning and impact accumulation mechanisms are proposed. They increase flexibility of the method, by reducing overall number of constraints of RBFCMs. The new mechanisms take into consideration standard semantics of fuzzy sets, where their uncertainty is measured by fuzziness and specificity. Introduction of new methods allows development of new type of complex fuzzy relationships and reasoning on them. Thanks to new type of relationships, RBFCMs can model complex systems, where a joint impact of several causal nodes on one effect node needs to be considered. Increased flexibility and modelling capability is achieved using mechanisms which are significantly less computationally demanding. New algorithms reduce by over 80% the number of operations that need to be performed to calculate one impact between two nodes. Advantages of using new RBFCMs are demonstrated using two new complex case studies: the modelling of the resource allocation on military capability of military units and the impact of investments into cyber security on the risk to the enterprise’s business. Both case studies could not be modelled using standard RBFCMs as they require development of complex relationships. To demonstrate advantages of new RBFCMs a dedicated software packaged was developed for standard and new reasoning and accumulation mechanisms. Additionally, a new direction for the development of RBFCMs is outlined, i.e. integration with Discrete Event Simulation (DES), that allows combining abstract RBFCMs models with the operational perspective represented by DES.
|Date of Award||Aug 2017|
|Supervisor||Dobrila Petrovic (Supervisor) & Ammar Al Bazi (Supervisor)|