| Natural fractures are developed in the Paleogene Oligocene-Neocene Miocene carbonate reservoirs of Asmari formation in A Oilfield of Iraq.Fractures are effective reservoir space and main seepage channels,which have important influences on oilfield development.Because there are few cores and image logs in this oilfield,and the response characteristics of conventional logs are complex and have multiple solutions,fracture identification by well logs is difficult.To address this problem,this work establishes an intelligent identification method for fractures in carbonate reservoirs based on bi-directional recurrent neural network.The basis inlcudes analysis of conventional logs response feature against fractures,reconstruction of logs for distinguishing fractures and selection of optimal feature attribute.According to the distribution of fractures and the characteristics of logs responses,a detection method of logs response against fracture based on window sliding is proposed,which realizes the detection of conventional logs responses due to fractures and establishes sequential logs for input of neural network.A comprehensive optimization method of logs feature attributes was formed,which integrates algorithms of random forest,autocorrelation analysis and hierarchical clustering.7 sensitive curves suitable for fracture identification in carbonate reservoirs were selected.A data balance method using under-sampling was used to deal with the problem of data imbalance in fracture samples.Through many experimental analyses and verifications,the neural network parameters and the optimal network structure adapted to fracture identification in carbonate reservoirs were optimized.A bi-directional recurrent neural network model for the identification of well logging fractures in carbonate reservoirs is constructed,and an intelligent identification method for fractures in carbonate reservoirs based on recurrent neural network is formed.Through the blind well test by core and image logs,the accuracy of fracture identification can reach about 90%.Through the identification and interpretation of fractures for 40 wells in 7 oil-bearing intervals of the Asmari Formation in A Oilfield of Iraq,and combined with the main controlling factors such as sedimentation and structure,the vertical distribution of fractures is clarified.The fracture identification results by logs show that the fractures in A section are more developed than those in B section.In the southern part of the study area,the fractures in A1 and A2 oil-bearing intervals are more developed,and fractures in the A and B oil-bearing interval are relatively developed in the north. |
