Study On Reservoir Parameters Prediction Method Based On Deep Learning

Reservoir parameter prediction needs to make full use of multi-disciplinary knowledge such as seismic,logging and geology to carry out comprehensive research.It mainly analyzes the reservoir characteristics by predicting the spatial variation of various parameters such as v_p/v_s,porosity,lithology and other physical properties and elastic properties.Deep learning technology is one of the most successful technologies in the field of artificial intelligence.It is essentially a neural network with multiple hidden layers and has a strong ability to mine high-dimensional spatial data structures.Reservoir parameter prediction needs to solve the nonlinear mapping problem between multi-dimensional parameters,the introduction of deep learning technology will help to promote the development of reservoir parameter prediction in the direction of automation and intelligence.According to the convolution mapping relationship between seismic data and wave impedance(elastic impedance),combined with the high efficiency of convolutional neural network parameter sharing,a one-dimensional convolutional neural network structure that can realize multi-dimensional regression between seismic reflection sequence and impedance sequence is constructed.It realizes multi-dimensional regression from multi-dimensional input to multi-dimensional output between seismic trace and impedance.Through network training,reliable impedance information can be directly obtained from seismic data,which lays the foundation for reservoir prediction using prestack elastic impedance.In the aspect of reservoir parameter prediction,some fully connected deep neural network structures with different complexity are designed to regression and classification problems,such Dropout and other network model optimization techniques are also used in network training.Establish a nonlinear mapping relationship between multi logging information such as natural gamma,neutron density,sonic velocity and reservoir parameters such as v_p/v_s,density,shale content,porosity,lithology and reservoir type.The network model performance with different hidden layers is tested and a high prediction accuracy is achieved in the logging data.On this basis,the deep neural network is used to construct the nonlinear mapping relationship between the elastic parameters and the three-angle elastic impedances that from pre-stack seismic data.The reservoir parameter regression prediction and reservoir type classification based on pre-stack elastic impedance inversion are realized,a variety of three-dimensional information that representing reservoir characteristics are obtained,which achieves good results in actual data application.Based on the different target forms of reservoir parameters prediction,the paper combines deep neural network and pre-stack elastic impedance inversion to form regression and classification prediction from multi logging information and seismic elastic impedance to reservoir parameters.The good application effect of deep learning in reservoir parameter prediction has certain reference significance for promoting the practical application of artificial intelligence technology in seismic interpretation.