A Layered Quantitative Non-revision Method For Inconsistency Belief

With the development of computer technology, from imitating the human behavior to imitating the human intelligence, people have done a number of attempts. But, the practice of artificial intelligence indicated that today's most intelligent systems are vulnerable to common sense. Namely, it's difficult for the knowledge representation and reasoning. After the constantly practice, people realize that the difficulties and focus of the common sense reasoning is the treatment of inconsistent knowledge-belief revision.According to the different needs of different areas, researchers have proposed many kinds of revision methods about the inconsistent belief, which mainly contain revision and non-revision methods. The belief revision method based on the new belief to rectify the belief base to maintain the inconsistency, in which the researchers have made a lot of achievement. But, nearly all of the belief revision methods exist the following problems:losing the hopeful information, resulting a conclusion which doesn't need or conclusions which are different to choice. The non-revision method deducing based on inconsistency belief base, which doesn't need to rectify the knowledge base. So the non-revision method avoids the problems which generate by the revision method.Based on the above thought, in this paper, a layered approach of quantitative non-revision in Propositional logic is provided.To ensure the Extension has good mathematical properties, we restrict the belief set as the clause and deduce rule as Resolution, stratify the inference results, which make the contradictory information isolated at one level, by this way give a new definition of Extension. In addition, this paper has proved that the Extension has many good properties, such as the consistency, interpretation, etc. If we regard the premise's infinite growth as an epistemic process, when the premise increases, its extension will converge to a definite limit.In order to filter the information, to save much information in Extension, we introduce a weight for any belief, which method optimize the definition of Extension from the quantitative way, and prove the optimized Extension also meet the related mathematical properties. Otherwise, as AGM postulate is the measure for most of belief revisions. So in this paper, we compare this method with AGM postulates.