Method and practice of deep favorable shale reservoirs prediction based on machine learning

Abstract

A set of methods for predicting the favorable reservoir of deep shale gas based on machine learning is proposed through research of parameter correlation feature analysis principle, intelligent prediction method based on convolution neural network (CNN), and integrated fusion characterization method based on kernel principal component analysis (KPCA) nonlinear dimension reduction principle. (1) High-dimensional correlation characteristics of core and logging data are analyzed based on the Pearson correlation coefficient. (2) The nonlinear dimension reduction method of KPCA is used to characterize complex high-dimensional data to efficiently and accurately understand the core and logging response laws to favorable reservoirs. (3) CNN and logging data are used to train and verify the model similar to the underground reservoir. (4) CNN and seismic data are used to intelligently predict favorable reservoir parameters such as organic carbon content, gas content, brittleness and in-situ stress to effectively solve the problem of nonlinear and complex feature extraction in reservoir prediction. (5) KPCA is used to eliminate complex redundant information, mine big data characteristics of favorable reservoirs, and integrate and characterize various parameters to comprehensively evaluate reservoirs. This method has been used to predict the spatial distribution of favorable shale reservoirs in the Ordovician Wufeng Formation to the Silurian Longmaxi Formation of the Weirong shale gas field in the Sichuan Basin, SW China. The predicted results are highly consistent with the actual core, logging and productivity data, proving that this method can provide effective support for the exploration and development of deep shale gas.

Keywords

• Sichuan Basin
• Ordovician–Silurian
• shale gas
• reservoir prediction
• machine learning
• convolution neural network