Abstract: The deep reservoir (Permian) of Wu-Xia fault belt in Junggar basin is characterized by the deep buried, developed faults, random seismic reflection and complex structures, etc. These make the prediction of the deep reservoir more difficult. The drilling data show that the developed volcanic reservoirs of Permian serve as the favorable zone for hydrocarbon accumulation. This paper presents the geologic and geophysical features of the volcanic rocks in Wu-Xia fault belt and their fine horizon calibration and interpretation, based on which several seismic attributes are picked up along the horizons, of which three attributes are selected such as average absolute amplitude, arc length and energy gradient at half (life) period. And the comprehensive classification given by means of Kohonen neural network is put forward, by which the lateral prediction of the volcanic reservoirs in Fengcheng formation of Permian is conducted and the next drilling well sites are determined, considering previous drilling, geologic and geophysical research results. The study shows that the seismic attributes are of good effects on the deep volcanic reservoir predictions.

Key words: Junggar basin, Wuerhe-Xiazijie (Wu-Xia) fault belt, volcanic rock, seismic attribute, reservoir prediction, deep stratum, Kohonen neural network