Application Of Neural Network And Fuzzy Logic Cooperation To Fault Diagnosis

The fuzzy neural netwaork, the integration of the neural network and the fuzzy logical system, has become the study focus in computational intelligence at present. The Neuron-Fuzzy Cooperation (NFC) is a mode of further integration of the neural network and the fuzzy logical system. This paper, taking the recovery equip in Maoming Petrochemical Corporation as the study object, deals with the technique of the NFC and its application in the fault diagnosis of the recovery equip. In the paper, Chapter One makes an introduction of some basic concepts and principles on intelligent diagnosis, illustrates the diagnosis technique's invention and development and the diagnosis's process and essence, draws the conclusion by comparing the advantages and disadvantages of several present intelligent diagnosis methods that the diagnosis method based on the NFC is superior to the others. Chapter Two briefs the recovery equip's technological process and automatic control, demonstrates the necessity and possibility in setting up the real time intelligent diagnosis system to the recovery equip and puts forth the solutions and key techniques. Chapter Three deals with the theories relevant to the NFC, such as the fuzzy set, the fuzzy operator, the cooperation principle, the topological structure and the tiered classified diagnosis model, etc. Chapter Four probes the learning algorithm, which firstly makes an analysis of the principle and disadvantages of the traditonal BP algorithm and puts forth the improved algorithm– the NFC-BP algorithm. The arithmetic adopts the method of dynamic compensation. Namely, when error spread reversely, error act on not only the connecting weight, but also the compensate parameters. This method result in that the compensate parameters change close with optimization. This method can redound to quicken studying of connecting weight and leap over the partial minimum value. So, the network studying performance can be improved.Then the further analysis of the standard genetic algorithm's operating principles and disadvantages is made, finally in a simulating way demonstratates the efficiency of the EGA-NFC-BP composite algorithm by integrating the improved genetic algorithm based on the self-adaptive control parametres of the experimental formulation with the NFC-BP algorithm. Because of the global searching function of EGA, we can get one individual which close with optimization base on previous study. Then, we use individual as original value of NFC-BP algorithm and be adjusted by NFC-BP. EGA-NFC-BP hybrid algorithm integrates characteristic of EGA and NFC-BP. Though, the opportunity to get in local solutions of EGA is less than it of NFC-BP, but it's studying time is longer than that of NFC-BP. So, in...