基於觀念模糊集的觀念溝通

Abstract

觀念溝通是人機介面中一個很重要的部分，它可以讓使用者與系統之間順暢地交換訊息。本論文引入了觀念模糊集來表示現實世界中抽象與具體的觀念，並提出了幾種對映的機制來建立這些觀念之間的關係。觀念的對映必須是雙向的，由抽象觀念對映至具體的觀念是一種觀念解釋的過程，而由具體觀念對映至抽象的觀念則是一種觀念辨識的過程。首先我們提出以模糊關係方程式來達成觀念對應的方法，觀念的雙向對映是以不同的關係方程式來達成，我們引用了基因演算法以及Fuzzy Delta Rule來學習這些關係矩陣，但是其效果則是無法令人滿意。因此我們又使用了MLP神經網路，同樣以不同網路的來記憶觀念的雙向對應，在此我們引入了BP演算法來學習MLP神經網路的權重矩陣，最後以MLP神經網路來達成觀念溝通，它的效果經過驗證是相當令人滿意的。

Concept communication is an important issue of man machine interface. It provides the smooth communication between man and system. This thesis introduces conceptual fuzzy set (CFS) to represent the abstract concepts and concrete concepts in the real world. Concept mapping must be bidirectional. Mapping from abstract concepts to concrete concepts is considered as concept recognition. Mapping from concrete concepts to abstract concepts is considered as concept interpretation. We propose several mapping schemes to relate these two type concepts. The fuzzy relation equation approach is first applied for the concept mapping. The forward and backward mappings of concepts are archived by adopting two different fuzzy relation equations, respectively. We apply the genetic algorithm and fuzzy delta rule to learn the relation matrix of fuzzy relation equation. Their performances are not acceptable. In a functional mapping respective instead, the multilayer perceptron neural network is utilized to the concept mapping problem. BP algorithm is adopted to learn the weight matrix in the multilayer perceptron neural network. The backward mapping of concepts is achieved by adopting another MLP neural network. The result of concept mapping by MLP neural network has demonstrated that the MLP network is an effective scheme for concept communication. ABSTRACT II ACKNOWLEDGEMENTS III CONTENTS IV LIST OF FIGURES VI LIST OF TABLES X CHAPTER 1. INTRODUCTION 1 1.1. RESEARCH BACKGROUND 1 1.2. INTRODUCTION TO CONCEPTUAL FUZZY SET 2 1.3. CONCEPTUAL FUZZY SET MAPPING VIA A FUZZY RELATION EQUATION 4 1.4. CONCEPTUAL FUZZY SETS MAPPING WITH MULTILAYER PERCEPTRON NEURAL NETWORK 6 CHAPTER 2. FUZZY RELATION EQUATION FOR CONCEPT MAPPING 7 2.1. INTRODUCTION TO FUZZY RELATION EQUATION 7 2.2. SOLUTION EXISTENCE OF A FUZZY RELATION EQUATION 8 2.2.1. Single-input-single-output Case 8 2.2.2. Multi-input-multi-output Case 10 CHAPTER 3. GENETIC ALGORITHM TO SOLVE FUZZY RELATION EQUATION FOR CONCEPT MAPPING 13 3.1. INTRODUCTION TO GENETIC ALGORITHM 13 3.2. GENETIC ALGORITHMS 14 3.2.1. Encoding and Decoding 15 3.2.2. Fitness Function Definition 16 3.2.3. Reproduction 16 3.2.4. Crossover 18 3.2.5. Mutation 20 3.3. EXPERIMENT RESULTS 21 CHAPTER 4. FUZZY DELTA RULE TO SOLVE FUZZY RELATION EQUATION FOR CONCEPT MAPPING 45 4.1. INTRODUCTION 45 4.2. FUZZY DELTA RULE ALGORITHM 47 4.3. EXPERIMENT RESULTS 50 CHAPTER 5. MULTILAYER PERCEPTRON NEURAL NETWORK APPROACH TO CONCEPT MAPPING 53 5.1. INTRODUCTION 53 5.2. BP LEARN ALGORITHM AND MLP NEURAL NETWORK 54 5.3. EXPERIMENT RESULTS 56 CHAPTER 6. CONCLUSION 60 REFERENCES 61