Algorithm Research And Realization Of A Fault Diagnosis Based On Improved SOM-BP Composite Neural Network

Nowadays, the scale of computer network continues to expand, the structure of it becomes more and more complex. To ensure the reliable and stable operation of the network system, rapid network fault location and diagnosis are the two important factors. Aiming at the requirement of quick network fault diagnosis, I research the application of artificial neural network technology in this thesis. The results are as follows:(1)A new kind of compound neural network model, namely the SOM-BP neural network model, has been proposed. This model can increase the degree of separation between abnormal characteristic values, which is more effective.(2)Using self-organizing map networks, fault characteristic values are mapped to a higher class space, to enhance the class separability. And then, using back propagation network (BP network) to classify the abnormal characteristic parameters. In this way, the fault diagnosis and location can be realized.(3)The samples to train the SOM-BP neural network, are the abnormal interface data of communication network. By the MATLAB simulation, the author compared the SOM-BP algorithm with the single BP neural network simulation results. Experiments show that this algorithm effectively improves the effectiveness of network fault diagnosis.(4)In this thesis, by writing the diagnostic system on the WEB, the author realized the function of the algorithm.The results of this article research need to be further perfected and more training sample, in order to improve the practicability of the algorithm. This algorithm also needs to be consummated in the actual network maintenance, accepts a large number of experimental data to verify the stability and accuracy.