Fracture Intelligent Identification By Well Logs Based On Recurrent Neural Network

Natural fractures are developed in the reservoirs the seventh member of the Upper Triassic Yanchang Formation（Chang 7）in Qingcheng Oilfield,Ordos Basin.Natural fractures are the main seepage channels and effective storage spaces of reservoirs.It is of great guiding significance for petroleum exploration and development to clarify the development characteristics and distribution rules of natural fractures in Qingcheng Oilfield.The Chang 7 section includes three lithologic combination reservoirs: thick sandstones interbedded with thin mudstones,sandstones interbedded with mudstones,and thick mudstones interbedded with thin sandstones.Sandstone and mudstone are frequently interbedded,and the fracture scale is generally small,making logging fracture response weak,multiple solutions strong,and fracture identification difficult.In response to this technical challenge,on the basis of clarifying the genetic types and development characteristics of natural fractures in three lithologic combination reservoirs,combined with core data,starting from the analysis of fracture response characteristics of conventional logging,reconstruction of conventional logging curves,and optimization of sensitivity of each curve,this paper establishes an intelligent identification method of reservoir fractures based on Bidirectional Short-term Memory Neural Networks,starting with analysis of fracture response characteristics from conventional logging,reconstruction of conventional logging curves,and optimization of sensitivity of each curve,This method utilizes network memory ability to capture the vertical correlation of logging data,improving the accuracy of relative anomaly detection of logging data.Through multiple experiments to adjust the internal structure and parameters of the network model,A bi-directional recurrent neural network model for intelligent identification of logging fractures was constructed,and an intelligent identification method for fractures based on the model was formed.The blind well recognition effect is verified by using core and production data,and the recognition accuracy is up to 90%;Taking Qingcheng Oil Field Z well area as an example,combined with the analysis of the main control factors of fractures such as sedimentation and lithology in the study area,the vertical distribution law of natural fractures is revealed: in the longitudinal direction,the natural fractures are the most developed in the small layer of Chang 7₃,followed by the small layer of Chang 7₂,and the fractures in the small layer of Chang 7₁ are weaker than the other two layers.