The Study On The Application Of Deep Learning Into Seismic Reservoir Prediction

The use of seismic data for oil and gas reservoir prediction has always been a hot and difficult point in oil and gas exploration and research.The main reason is that the relationship between seismic data and reservoir parameters is relatively complicated.Especially in the current oil and gas reservoirs from conventional oil and gas to unconventional oil and gas,from sandstone reservoirs to tight oil and gas reservoirs and complex carbonate reservoirs,the application of geophysical methods in reservoir prediction becomes more difficult.It is of great significance to accurately establish the correspondence between the original massive data and the relevant parameters of the seismic reservoir.In recent years,with the rapid development of artificial intelligence technology,machine learning algorithms that can solve strong nonlinear problems have been well applied in signal analysis and big data processing.Therefore,the paper is based on the rapid development of artificial intelligence technology,and carries out application research on the use of machine learning algorithms to predict oil and gas reservoirs.Deep learning is a rapidly developed method in the field of machine learning in recent years.Compared with traditional linear and shallow machine learning methods,deep learning methods have the advantages of deep extraction of large data features and high accuracy.This thesis completed the research work on reservoir prediction using deep learning algorithms.Firstly,basic research on deep learning methods was carried out.A deep residual network model was designed and written using the Tensorflow library.The model was tested using cifar10 data,and the influence of different parameters on the model results was analyzed.Secondly,the non-linear relationship between the conventional logging curve and porosity is established using the depth residual network,and the prediction of the well porosity is completed.Finally,through the deep residual network technology and the seismic attribute technology,the fracture prediction of the indoor physical model data and the reservoir parameter prediction of the actual seismic data are completed.In the application of fracture prediction,the seismic attribute technology is used to extract the seismic attributes related to fractures in the Longgang area,and the K-fold cross-validation and Focal?loss loss function are applied to the residual network to predict the fracture.There are good test results in the prediction.In the application of reservoir parameter prediction,the actual complex carbonate reservoir is selected for the network test of parameter prediction.The non-linear correspondence between seismic data and porosity is established using the depth residual network method,and the prediction of the porosity of the target layer in the work area is completed.From the results of the continuous well section and porosity slice,it can be seen that the use of earthquakes to predict porosity Consistent with the well porosity,the range of porosity on the slice is consistent with the low porosity characteristics of the study area.Research through model data and actual data shows that:(1)Compared with the traditional convolutional neural network,the deep residual network method has a better convergence effect.(2)The deep learning method can deeply mine the deep features of the data,and shows a very strong ability in characterizing the nonlinear relationship between seismic data and reservoir parameters.(3)Combining deep learning methods with well-seismic data for reservoir prediction is feasible and has good application prospects.