Research On Logging Facies Recognition Method Based On Convolutional Neural Network

Logging information is a reflection of the stratigraphic rocks' physical properties,and sedimentation is an important controlling factor for the physical properties of stratigraphic rocks.Logging data is the basic and important source of information in the study of hydrocarbon reservoir sedimentology.The logging facies is the bridge between logging information and reservoir sedimentology.However,the well logging method is either using a manual method,manually comparing the log graph for identification analysis,or establishing a mathematical model for a certain research area.The process has some problems including inefficiency,low precision,and poor generalization.Due to the inevitable systematic errors in the actual logging exploration process,there are abnormal data and continuous data missing problems in the logging data.In this thesis,the abnormal datas are deleted and interpolated,the logging data is corrected and normalized,and a gamma data prediction algorithm based on random forest is established.The missing data is effectively filled to establish training.The convolutional neural network is a nonlinear model,which is continuously optimized by feedback algorithms.It has been applied in many practical problems and achieved good effects.According to the characteristics of the special pattern of the logging curve including box,bell and funnel,this thesis uses the image form for logging facies identification,constructs 4 types logging facies data sets,and carries out data enhancement through multi-method fusion.Based on the excellent small sample training ability and generalization ability of support vector machine,we combined convolutional neural network to construct logging facies identification model,and optimize the model,improve efficiency,and enhance generalization ability.Aiming at the problem of recognition fineness in actual geological research,this thesis proposes a multi-scale logging facies identification unit partitioning method based on wavelet transform algorithm to realize multi-scale transformation of logging curve.Aiming at the problem of logging facies unit division in the identification process,a data partitioning method based on extremum segmentation is designed to realize the test data segmentation.In order to verify the effectiveness of the proposed logging facies recognition method,a topic is established.The logging facies-deposit micro-correlation table of the data deposition environment realizes the discrimination of the logging facies to the sedimentary microfacies.Experimental analysis shows that the gamma data prediction method based on random forest can fill the missing data.The logging facies identification model based on convolutional neural network has higher accuracy and stronger generalization ability than other algorithms.The multi-scale logging facies identification unit partitioning method based on wavelet transform can solve the practical problems in geological research.The model recognition results are good and can support the actual research work.