Research And Realization Of Fuzzy Reasoning Algorithm Based On Belief Rules

With the propulsion of informatization and intellectualization, artificial intelligence haspenetrated into daily life. Reasoning technology became one of the most active research directions.Although many achievements on production rules and fuzzy reasoning methods were arised, classicsfuzzy rule cannot meet the needs of many applications because of its defect in explaining theuncertainty of knowledge. Recently, belief-rule has been proposed, it can show both fuzziness andprobatilistic uncertainty of knowledge. But currently specific study on fuzzy reasoning based onbelief-rule was very little. In order to address the fuzzy reasoning problems in complex systems, it isnecessary to make more systematic and in-depth research on fuzzy reasoning algorithms which werebased on belief-rule.In the view of this three most typical reasoning method, Mamdani method, DS method based onsimilarity and cloud-based, the mechanisms of these fuzzy reasoning algorithms were summarized, Inthis case, fuzzy reasoning algorithms based on belief-rule were put forward. Firstly, some analyseswere made on the reasoning algorithm that based on fuzzy relation synthesis, and their limitations onrule forms and sensitiveness on inputs indicated. After that, the calculate methods of matching degreeof the whole antecedent and that of distributed conclusions were given, on this basis, a fuzzyreasoning algorithm that suitable for belief-rule was proposed. Also, the the steps and graphicalexplanation of this algorithm were given. The results of experiment showed that the inferences of thisalgorithm were closer to the expert opinion than the original algorithm. Secondly,“MembershipCloud” was introduced to make fuzziness and randomness of natural language combining in thereasoning process, meanwhile make the algorithm applying to complex systems that were describingby belief-rules. Then another new algorithm was put forward, it was based on belief-rule and“Membership Clouds”, and its inputs must be quantitatively. With the purpose of making the inputscan be qualitative value, the other new algorithm were proposed. Before this, a new method wasintroduced to calculating the similarity of clouds. The definitions and proffs of cloud’s minimum andmaximum boundaries were given. Then the two algorithms’ relations and differences were pointed out,their validity and advantages were verified by some experiments. Finally, these algorithms wereimplemented in the FuzzyCLIPS expert system tools, and the extended FuzzyCLIPS were introducedto the torpedo evasion system. Some comparative analyses on the results were given to show that thenew fuzzy reasoning algorithms based on belief-rule were workable and advantageous.