| Fault identification based on seismic data is an important reference basis for seismic structures interpretion,reservoir characteristics and determining well position determination.With the increase of society’ s demand for oil and gas resources and the difficulty of exploitation,traditional fault identification methods are time-consuming and subjectively affected,which can no longer meet the requirements for effective and accurate identification of faults.In recent years,deep learning has emerged in the field of geophysical exploration by virtue of its powerful feature learning ability and efficient expression of data,providing a new solution to the problem of fault identification.Based on the research and summary of the fault identification technology and the principles of deep learning models,this paper made an in-depth research on the use of convolutional neural network models to solve fault identification problems.In the fault identification,aim at the problem of scarcity of data and obvious differences in the characteristics of faults in different work areas,a method of artificially synthesizing seismic faults is designed.By adding a two-dimensional Gaussian function and a linear scaling function to the sine model,and convolving with Ricker wavelet,Create synthetic seismic images and corresponding fault labels,which effectively improves the generalization ability of the model.To solve the problem that convolutional neural network is suitable for extracting local features and difficult to extract non-local relationship between fault features,a fault identification model based on VGG16 is designed and constructed,the model reduces the number of convolution layers and the difficulty of network training.The 1*1 convolution kernel is added to control the dimension of the feature graph in the channel direction.Finally,a Softmax classification layer is used to obtain the fault probability.At the same time,a quasi-balanced binary cross-entropy loss function is used to adjust the imbalance of faults and non-faults labels in the training set.The model parameters of the improved CNN(Convolutional Neural Networks)fault identification model are pre-trained by the supervised method,and the multi-layer model is effectively trained.Aim at the problem of lack of continuity constraint of fault detection results and noise in identification results,a set of fault image post-processing process with experience of geological experts is proposed.The 3D Log-Gabor filter array is used to remove the scattered noise,and the seismic image data is enhanced by combining with expert experience to optimize the fault surface,suppress noise and other artifacts,and improve the accuracy of fault identification results.The improved CNN structural model was tested for several times,and the experimental results show that the fault identification accuracy of the model on the composite data set reaches98.75%.The model is compared with coherence body,ant tracking technique and VGG16 model respectively.Compared with the coherence volume method,ant tracking algorithm and VGG16 model,the recognition accuracy of the model presented in this paper is higher than that of the coherent volume method,ant tracking algorithm and VGG16 model in the actual data set of the working area.The results show that the deep learning-based fault identification method designed in this paper eliminates the impact of seismic noise on fault identification,avoids the problem of complex parameter adjustment of VGG16 model,and has higher accuracy and clarity of seismic fault identification.It is beneficial to explore the application of deep learning technology in specific fields such as oil and gas resource exploration. |
