Knowledge Discovery And Control Rule Extraction Based Fuzzy Neural Network

Knowledge Discovery and Data Mining (KDD) is an intelligent techniquefor information automatic pick-up from data-base to acquire unknown, usefulknowledge, which emerges as the time require and develops very quickly withinfinite vital force and wider foreground. Up to the present, there isn't a universal algorithm for knowledgediscovery and data mining. The algorithms in existence are aimed atidiographic targets and objectives in their respective fields. For the industrycontrol purpose, this paper presents an approach to make data mining usingfuzzy neural network for rule extraction. As it is well known that the constringency of traditional BP algorithm ispoor and a hybrid supervised/unsupervised algorithm is discussed in thispaper to improve the learning rate. In fact, this hybrid algorithm is combinedthe neural computation with conventional computation to learn from others'strong points to offset one's weakness. The learning of neural network can bedeemed as the construction process of neural network which configurationincluding the number of rule nodes and parameters can be first acquiredthrough unsupervised algorithm and then optimize by supervised BPalgorithm from the training data. It is tested by experiment that the hybridalgorithm with higher learning rate compared with traditional BP algorithm. To advance the adaptation, the fuzzy adaptive neural networks (FANN)with respective algorithms for learning, aggregation, rule insertion, ruleextraction and rule adaptation is further developed here. FANN is aimed at 第- III -页北京化工大学硕士学位论文building adaptive intelligent systems that evolve their structure and parametervalues through increment learning. They can accommodate new input datathrough local element tuning and new connections and new neurons arecreated during the operation of the system for the 'opening' structure. The effectiveness of the two model and respective algorithms in this paperare clearly demonstrated by the experiment results. Further more, theapplicability of FANN on an evaporator's level control system discussed hereillustrates the feasibility of data mining with fuzzy neural network for eitherprediction or control tasks. This paper presents fuzzy neural network as a realization for KDD. Someprimary research works are developed and a series of conclusions are drawnin this paper and all this lay a foundation for further research works in KDD.