The Improvement Of Supervised Fuzzy ART Neural Network And Its Application On The Identification Of The Lithology And The Fluid Type In The Reservoir

The Adaptive Resonance Theory is putted forward by an American scholar named Carpenter in 1979, later on; he cooperated with Grossberg and putted forward the Adaptive Resonance Theory Network. The learning style of the Adaptive Resonance Theory Network belongs to the unsupervised, and its input is the binary pattern. its Working principle is that the network gets the input from the outside, according to amount of similarity between the new sample and the characteristic pattern (the Representative Value of the class's nodes which output terminal corresponds to )storing in the network ,Compare the amount of similarity with the reference threshold value designed before and decide the new inputting pattern is to the Class Model stored in the network. If the sample whose amount of similarity is greater than the reference threshold values of some classes, will be classified the class whose characteristic pattern is most similar with the sample. The Fuzzy Adaptive Resonance Theory Network is putted forward based on the Adaptive Resonance Theory Network, and its input is the fuzzy mode. Dased on upon The Fuzzy Adaptive Resonance Theory Network, the Pan Ziwei and the Xu Jinwu combine the network'supervised learning style and the unsupervised, and put forward the Supervised Fuzzy ART Neural Network.Dased on the Supervised Fuzzy ART Neural Network putted forward by Pan Ziwei , when training the network , the warning parameters of the classes will be trained too but not be decided by people , and the learning style of the outstar vector and the instar vector are changed ,then we get a improved supervised fuzzy ART neural Network. We get the warning parameters because of the crossed samples among the classes, so the network's fault tolerance is better. And the network will be more appropriate to the samples which there are same fault ones in them like the logging data. The input style of the network is added encoding expansion vector. In supposed of that the inputting vector has n dimensions; we can get the geometrical meaning of the network. Its geometrical meaning is that when the network working, it help us find the n dimensions cubes in the n dimensions volume coordinate, and the samples of every close will fielded in the cubes that the class corresponding to. When training the instar vector, it means locate the cubes in the volume coordinate. When training the warning vectors, it mans control the size of the cubes. The new network'learning style is unsupervised. But combining with the fuzzy C- means clustering, the whole algorithm combine two learning styles.The new network is used to identify the Lithology in Hailaer Basin Uerxun sag and the fluid in the Eerduosi Basin Tabamiao sag in the article. When identifying the Lithology, we have to decrease the fluid's influence to the logging data. on the contrary when identifying the fluid , we have to decrease the Lithology's influence to the logging data. In the study area of the Uerxun sag, the Lithology of the layer is pyroclastic rock, and Porosity of the layer is low, and the sort of the fluid is unitary, so the fluid's influence to the logging data is weak. While the Lithology of the layer in the Tabamiao sag is Sedimentary rock, the Lithology of the reservoir is mainly sandstone, so the fluid's influence is strong. The Hailaer Basin, in northern part of the Inner Mongolia Autonomous Region and to the west side of the Da Hinggan mountains, was a Late Mesozoic-Paleogene basin, with the basement being Hercynides. The area of Urxun sag is big, and the difference of diagenesis is also big. The main types of rock in the research region are: Sedimentary rocks, volcanic clastic sedimentary rocks, volcanic sedimentary rocks, volcanic rocks and lava. Rocks in the area also use the classification criteria of the sandstone, and tuff, melting guitar-tuff and pyroclastic tuff are subdivided into coarse grain tuff / ignimbrite / bedded tuff, granule tuff / ignimbrite / bedded tuff,silt tuff / ignimbrite / bedded tuff. Eerduosi basin is a platform-type structure of sedimentary basin, originally belonged to a part of north china platform and in early Caledonian is unified north china platform, accepted the thick sedimentary carbonate. Tabamiao block is located in the northern Ordos Basin, in the block there are many gas layers. The Xiahezi formation and the Shanxi formation are the two study horizon. The Lithology of the reservoirs in the two formations is mainly sandstone. According to the interpretation of the logging data, the fluids in the study block can be classified to seven kings, they are gas layer /containing gas layer / containing gas dense layer / poor gas layer /the layer without gas/dense layer /coal layer.The procedure of the network is finished with the procedure for c++ Editor. According to the logging data and the geological data, extract and screen the training samples, and use them to train the network. Use the network that has been trained to identify the Lithology or the fluids, and the outcome will be showed with the carbon software, at the same time compare the outcome with the geological data. At last we can get the statistics results that in the study blocks the average of coincidence rate is almost 70%.