Research Of Knowledge Acquisition For Expert System Based On Rough Sets And ANN

As one of the important applications of artificial intelligence, expert system (ES) has developed in recent several decades and matured in both theory and technology, wide and successful application of it has also been made in many fields. Knowledge acquisition is an important step in development process of ES, and the low efficient makes it the bottleneck of ES development. Because of good complemental characteristics of Rough sets and Artificial Neural Networks (ANN), their combination provides a good thought for solving the bottleneck-problem of ES. The main thought of their combination is to use rough sets as a pre-processor for neural networks. The knowledge-reduce function of rough set can eliminated the redundant attributes and so make the dimension of neural networks' input reduced, as a result of which, the scope of neural network become small, generalization capability and precision of prediction become more satisfying.The thesis makes some researches on the knowledge acquisition problem for expert system based on rough sets and ANN according to this train of thought. It makes introduction of several basic concepts in this thesis and makes some analysis to the defects and superior of them. Then it makes some researches in the method of knowledge acquisition based on rough sets and ANN which includes several problems, such as coding of data in decision tables, evaluate of output results, rough sets multi-reduction's processing to artificial neural networks and the composite evaluate of multi-reduction. Corresponding resolvents is put forward, and a kind of BP artificial neural network is designed and implemented to test the resolvents which is proved to be valid.