Abstract: The 3D seismic data-processing assembly with the merit of high signal-to-noise ratio and the obvious wave characteristic is used to select high-reliability seismic reflection internal texture and external shape from Xiazijie formation of Permian in Wu-Xia area of Junggar basin. By means of such seismic attributes as instantaneous amplitude, instantaneous frequency and persistence length, the seismic facies is quantitatively analyzed and nominated using Kohonen neural network method. The results from well log interpretation, core analysis and palaeontologic evidence are applied to conversion of the seismic facies into sedimentary facies, thus well solving the problem unable to classify the sedimentary facies with few deep well in the past.

Key words: Junggar basin, seismic attribute, Kohonen neural network, deep stratum, sedimentary facies