Application Research Of Seismic Reservoir Prediction Based On Convolutional Neural Network

Seismic reservoir prediction obtains fault,reservoir lithology or reservoir parameters by using seismic data or optimized seismic attributes.However,with the deepening of oil and gas exploration and development,exploration targets have gradually changed from conventional reservoirs to fractured reservoirs and lithologic reservoirs.The accuracy and efficiency of traditional reservoir prediction methods can not meet the prediction requirements of these reservoirs.Therefore,the development of high-precision automatic fault identification and reservoir lithology prediction technology is of great significance for oil and gas exploration and development.Deep learning algorithm is a very popular research direction in the field of artificial intelligence.Convolutional neural networks only need to train convolutional networks with known patterns.The network has the ability of mapping between input and output,and can well establish the complex non-linear relationship between input and output data.In this paper,the convolution neural network algorithm is applied to fault identification and reservoir prediction.The fault identification based on Le Net convolution neural network is studied,and the fault fracture development zone is determined.Then the reservoir prediction based on U-Net convolution neural network is studied,and the reservoir is searched in the determined fault fracture development zone.Traditional coherence attributes have some shortcomings in fault prediction,such as fault artifacts and vulnerability to noise.Firstly,this paper presents a method of extracting multispectral coherence attributes based on S-transform,which effectively solves the shortcomings of traditional coherence attributes,and also provides input data for fault prediction based on Le Net convolution neural network.Then,according to the characteristics of faults in the work area,the relevant parameters of the network are determined through experimental analysis,and good training results are obtained.The Le Net network model is used to predict faults in the work area,and the accuracy is obviously higher than traditional methods.According to the characteristics of the lithology data of the work area,the residual module is added to the U-Net convolutional neural network,and the input information is directly bypassed to the output,which protects the integrity of the information,so that the entire network only needs to learn the input and output differences.It effectively solves the problem of gradients that are prone to occur in deeper networks.Through actual data testing,it is found that the improved U-Net convolution neural network can achieve good results in reservoir lithology prediction and "sweet plot" reservoir prediction.