APPLICATION OF EXTREME LEARNING MACHINE IN ANALYSIS OF CLAY MINERALS

The study on low permeability reservoirs in the 8th member of Lower Shihezi Formation, Upper Paleozoic in northern Sulige of Ordos Basin show that diagenesis is one of the main factors controlling the distribution of gas reservoirs, while clay minerals affect the type and intensity of diagenesis and serve as an important indicator in the classification of diagenesis. This study attempts to use natural gamma-ray spectral logging combined with neural network (extreme learning machine, ELM) to accurately calculate the content of clay minerals in reservoir, providing support for automatic identification of diagenesis in the whole well. In the analysis of clay minerals, to avoid the influence of difference of rock skeleton and particle compositions on logging information, the neural network of clay mineral analysis is established respectively for lithic sandstone and lithic quartz sandstone in the study area to improve calculation accuracy. The neural network adopts the ELM with lower probability of trapping into low efficiency and local optimum to ensure the speed and stability of analysis result. On this basis, 15 X-ray diffraction analysis samples from the 8th member of Lower Shihezi Formation in northern Sulige area are used to compare the calculated results of differential and indifferential lithology, proving the effectiveness of the method.