Methods And Applications Of The CBR-ANN Intelligent Decision Support

Solving decision problems by artificial intelligence methods such as case-based reasoning (CBR), artificial neural network (ANN) and so on belongs to a significant research field of intelligent decision support system (IDSS). So CBR-ANN based IDSS is studied by this dissertation to develop integration theory of CBR and ANN and to explore new IDSS method for solving engineering decision problems, especially agricultural decision problems.First, relations among instance-based learning (IBL), CBR and ANN are analyzed, and some hybrid models of CBR and ANN and their possible integration researches are discussed. Based on design characteristics of CBR and ANN, a design framework consisted of"pattern space-representation","integration method selection"and"model and algorithm"for CBR-ANN integration is proposed. Then a design framework for IDSS is presented with a viewpoint of IDSS design tasks consisting of"problem analysis and representation","method analysis and design","system analysis and planning"and"system design".This research proposes a general framework for CBR-ANN IDSS and analyzes principia of its components. Three case representation methods, character vector, category-exemplar and dynamic storage are discussed, and CR4 process model is proposed for CBR-ANN integration. For construction of casebase, similarity relations on possible problem categories should defined and analyzed, mapping relations from possible problem categories to possible solution categories should be studied, and casebase should be repartitioned and refined. This research presents knowledge representation methods suitable for CBR-ANN IDSS, analyzes different similarity evaluation measures for attributes and cases, and designs experimental methods for performance evaluation of several typical case similarity measures.This dissertation proposes a vantage case based indexing mechanism and corresponding case retrieval algorithm for assisting rapid retrieval of similar cases and improving retrieval efficiency in large casebases. At the same time, two competitive ANN (Adaptive Resonance Theory and Self-organizing Maps) suitable for case classification and retrieval are discussed. Based on production representation method, design method and reasoning process framework for system reasoner are presented.Knowledge-based high dimensional case reasoning can be performed through parallel computation capacity of ANN, and Bayesian reasoning by ANN is discussed. Then theoretical CBR decision method is discussed thoroughly.Recommendation methods based on target case information and user historical request information are proposed. Case retainment method is thought of as relating to the Recall of reasoning, and the case retainment and learning strategies are presented.Finally, combining corresponding national policy and a research project of the author, CBR-ANN decision support model and methods are applied to the IDSS for crop disease prevention and treatment, and a concrete cucumber disease (Cucumber Fusarium Wilt, CFW) is taken as research object for the exploration of integration of CBR and ANN.Based on the traits of a CFW case-base, the optimal case classification number is determined by analyzing the relations between case classification and case retrieval efficiency, and the optimal case dissimilarity threshold is figured out through analysis of reasoning effectiveness evaluated by cross-validation experimental method. In the application of CBR-ANN intelligent decision support model and methods, Adaptive Resonance Theory-Kohonen Neural Network (ART-KNN) are integrated into CBR based IDSS to perform task of case recognition, and the corresponding classification accuracy and the prediction performance of some predicted attributes in case solution are analyzed. The reasoning effectiveness of the applied system is validated by further actual case test and analysis.This dissertation has systematically studied and established the CBR-ANN intelligent decision support model and methods, and made true the application of them. This shows theoretical and application significance. It is no doubt that this important research field of CBR-ANN IDSS will be continuously enriched and developed, resulting in more effective solutions for practical problems through artificial intelligence and decision support methods.