Research on Case Intelligent System Based on Rough Sets and Multilayer Feedforward Neural Networks
|School||Hefei University of Technology|
|Keywords||Rough Set Feedforward neural networks Case-Based Reasoning Synthesis Reasoning Constructive Neural Networks Knowledge Granularity Covering Algorithm Case Knowledge Base Maintenance|
Learning is the most important manifestation of intelligent behavior, knowledge and explore the ancient Greeks began, has been a human pursuit of the goal. Analogies are important human cognitive method, a technique that allows knowledge with reasoning learning strategies in the field of a similar nature, the overall performance in the form of human intuition, logic, and creative ways of thinking, but also people experience the decision-making process used way of thinking. Case-based reasoning technology as a new reasoning method of intelligent system, the analogical learning computer in the human brain to achieve, but also the pioneer of this field of research and successful practitioners. Case intelligent systems research, contribute to imitate human thinking and human intelligence. Access to knowledge from large amounts of data, knowledge representation and reasoning decision rules, is the primary task of intelligent information processing, especially for practical problems uncertain, incomplete knowledge processing, rough set theory and artificial neural network technology to demonstrate the amazing data processing capabilities. This paper focuses on case-based reasoning cycle process, the main rough set reasoning and large-scale data processing feedforward neural network-based research to improve the accuracy and efficiency of case-based reasoning system, and enhance the flexibility and robustness of the system. The main contents of this paper are as follows: (1) The case of intelligent systems based knowledge representation case, the case may be semi-structured or unstructured, or even use natural language to express; analogy feasibility of case-based reasoning, knowledge structure of the logical basis of case-based reasoning and reasoning, and given some of the confusion in the reasoning of human knowledge; case smart aspects of the knowledge structure must handle the problem. Build a suitable case base, and how it is organized and maintain, and rapid retrieval of the complete case is very important and have a direct impact on problem solving performance. (2) from the point of view of knowledge reasoning the uncertain knowledge model and its relationship analysis, a detailed study of the combination of rough set with case-based reasoning reasoning techniques, many ways to find the theoretical model fusion technology and methods, so as to facilitate our from a higher level of understanding of the human mind and its problems processing method, launched the nature of knowledge from incomplete knowledge, which is essential for high-level decision-making. Problem solving actual needs, The integrated reasoning technology can make full use of multi-level knowledge, to integrate various reasoning techniques, the use of different knowledge granularity, and improve the efficiency of the system reasoning. Ultimately build a unified, more effective, able to handle complex and ambiguous information theory of granular computing platform, which provided for the case of intelligent systems to achieve a reliable technology foundation, and greatly improve the the problem processing capacity of the system; Finally, a case of intelligent decision support system architecture. (3) study the intelligent knowledge retrieval technologies, and a detailed study of feedforward neural networks as a case retrieval algorithm; pointed out that is still widely used BP algorithm, simulated annealing algorithm and its improved algorithm, difficult to overcome the slow local extreme disadvantage. Subsequent comparative study of radial basis function algorithm, it is according to the spatial distance mode vector multidimensional nonlinear alluding to the identification and classification of a similar detector. The study shows that the RBF network has a fast learning, global convergence; then retrieve similar cases based on RBF network model. (4) detailed study of the structure of neural network algorithm, consider the pros and cons of a performance of the network as one of the pursuit of the goal of the algorithm, from the global to examine the neural network learning process of studying the structure of the network structure large-scale problem solving has important significance. From the start with the MP neurons geometric significance, contrast the early constructive neural network algorithms - FP algorithm, and then launched covering algorithm. Studies have shown that the technology is easy modular building has the advantages of fast, high recognition rate, the last areas of coverage based on constructive neural network algorithm case intelligent systems. (5) a detailed study of case knowledge base maintenance technology, and put forward the principle of case knowledge base maintenance. Case-Based Reasoning incremental learning system, similar to the accumulation of human knowledge. Case base maintenance is the core of the CBR knowledge systems research, the CBR reasoning knowledge representation, adaptation and rewriting process; Although CBM as an important branch of CBR research, has been widely studied and developed in a different case base maintenance strategy; However, in different environment, because of the different characteristics of the CBR system size, timeliness, and applications, there is a greater difference case knowledge base maintenance tools and maintenance performance. Redundant and inconsistent case knowledge, or due to the changing environment of the areas of knowledge, and put forward a case based on the similar rough set knowledge base maintenance technology, knowledge-based particle size, so that the real-time monitoring of controllable threshold to achieve real-time maintenance. For actual field applications such as interactive state of fault diagnosis, online help, e-commerce, the size of the case base can easily reach hundreds of thousands of non-reduction system performance and maintenance problems, the case based on the covering algorithm knowledge base maintenance, the system is two-pronged approach to the completion of the case base maintenance: on the one hand, the coverage algorithm in the case library is divided into areas covered, the case of selective filtering; the other hand, application of multilayer feedforward neural network improved case matching, improve retrieved efficiency. Experiments show that the method can be used to deal with a flood of high-dimensional data, and to ensure the availability of the system.