A Unsupervised Pattern Recognizing Method And Application Of Gas Prognosticate

In this dissertation, after researching the raw SOFM arithmetic, we improve it andgain the self organizing neural tree--SONT structure. Finally, combining seismictechnique , implement the specific arithmetic of recognition of oil & gas.Firstly, we introduce the basic theory and methods for realization of SOFM. Subsequently we improve the arithmetic and implement the feature extraction of raw data using K-L translation, select the eigenvector. Combining C-avarage and ISODATA arithmetic, classify the eigenvector, according to the methods of this dissertation, in lower layer, set nerve cell and unite or delete nerve cells in middle layer, to improve the anti-huise- and robust.Using the networks structure to recognition problem of oil & gas. And compare the method with general and regression analysis ,gain perfect effect. The exampleshow SONT is one of effective methods for classification.In the course of the research, explore an application program of SONT and bring forward point of view of preset nerve cell and auto-adjust the structural of network. It not only appear guide effect of gas prognosticate, but also provide new methods for reference of other field of unsuperintend pattern recognize..