1. College of Earth Sciences，Jilin University，Changchun 130061，China; 2. Norheast Oil and Gas Branch of SINOPEC，Changchun 130062，China; 3. College of Geo-exploration Science and Technology，Jilin University，Changchun 130026，China

Due to the complex lithology and severe diagenetic alteration，the volcanic rocks of the Huoshiling Formation from the Changling fault depression in the southern Songliao Basin can't be effectively identified by conventional logging cross-plot methods，which seriously hinders the exploration and development process of the volcanic oil and gas reservoirs in the study area． Based on volcanic rocks collected by 17 wells of the Huoshiling Formation in the Changling fault depression，the volcanic rock sequences of the core section are established，and 8 kinds of volcanic rocks are selected to extract 258 sets of conventional logging data ( GＲ，LLS，LLD，CNL，DEN，AC) ．The logging response characteristics of different volcanic rocks are summarized． The extracted data are randomly divided into training data ( 70%) and prediction data ( 30%) ． The training data is used to build up a BP neural network lithological prediction model，and the Dropout mechanism is introduced to reduce overfitting． The prediction data is used to verify the lithological prediction coincidence rate of the model． The research results show that the highest coincidence rate of lithological prediction by this model can reach 89. 03%，which can effectively distinguish the main volcanic rock types in the study area．