| Expert system has been the hot field of artificial intelligence for a long time. Many expert systems have been applied in the fields such as agriculture, medicine and geography, and have acquired satisfied results. But in some domains, especially in agriculture domain, Lots of information is inexact and uncertain. In order to process the uncertainty caused by the lacking of information to make a decision, uncertain inference (also called inexact inference) is needed.Dempster-Shafer Theory (also called Evidence Theory, D-S Theory), belongs to a module-based intentional approach method in the uncertain inference, offers a module-based knowledge representation and several learning and inference methods base on it. Such module can answer any query related to the knowledge belongs to the field and make auxiliary prediction,decision and analysis. With the development of the computer science and many research work has been done, D-S Theory has been apply in many fields and make a good result, it becomes one of the most powerful methods for uncertainty process.This thesis is finished during the research of "Integrate Platform for Developing Intelligent Agriculture Information System", The National High Technology Research and Development Program of China (863 Program). The main purpose is to introduce D-S Theory method and Convex Evidence Theory to the "Basic Platform for Developing Expert Systems-Integrated Development Environment" (BAPDES-IDE), by two components: "D-S Theory-based inference module" and "Convex Evidence Theory-based inference module", to offer a method solving the uncertain problem in agriculture.For the sake of the success of introducing D-S Theory to BAPDES-IDE and solving the Zadeh's absurdity, this thesis adopted three methods with different view—modify D-S's formulary, disturb the mass functions before combining, and learn before combining. Subsequently three results will be combined by voting, in order to gain the more credible result."D-S Theory-based inference module" and "Convex Evidence Theory-based inference module" have three parts: "D-S Theory-based Inference", "Convex Evidence Theory-based Inference", and "The System of building repository" that fulfilling the needs of the front two inference.The function of "The System of building repository" is that communion with experts, for obtaining large numbers of knowledge, experience and information about the domain. In the course of building repository, experts input knowledge, which need fulfill the syntax rules that defined by the complexion of difference inference. Experts can choose any editor of xml to build the repository file (*.xml), only using the xsd file that defined by us as the Schema of repository file, and adding valid every element value.The function of "D-S Theory-based Inference" and "Convex Evidence Theory-based Inference" is drawing an inference from the fact values that were inputted by users and the knowledge of repository, and returning a reference to page layout of outputting inference. The page layout will parse the inference and output the conclusion to users. Both of the above-mentioned inference components were implemented in the language C#, and run under Microsoft .NET Frame environment. These two components were built to be .DLL file. The feature of .DLL file is can encapsulate its functions, only offer some interfaces that can be transferred. Therefore, the implementation under the specification of BAPDES-IDE makes it easy to link these components into the BAPDES-IDE.The repository file is stored in xml format. As fulfilling the syntax rules, the repository file can easily used by reference components. Furthermore, as the specification of xml is becoming the standard of industry, the data structure is convenient for interact with other system. As D-S Theory can grasp more unknown and uncertain, it is better than traditional probability theory. D-S Theory can amalgamate lots of evidences, and fulfill more weakly axiom system, and it is regarded as a sp... |
PDF Full Text Request