Machine Learning for Estimating Methane Distribution in 2-D Nano-pores

Increasing production of methane in shale reservoir is one of the major perspectives for gas resource exploration. It is very essential to understand the flow mechanism in shale reservoir at different scales. Molecular dynamics simulation of methane at nano-pores scales provides understanding of methane flow properties, e.g. methane molecular distribution. However, molecular dynamics simulation at nano-proes is very complicated and computational expensive resulting from the complicated geometry of shale pores, such as the size and shape of pore cross section. The application of machine learning method is a feasible way to improve the simulation of methane flow properties in shale reservoir. In this research, the objective is to study the methane distribution in shale nano-pores with different geometrical features and different methane bulk densities. We utilized Convolutional Neural Nets (CNN) method to estimate the distribution of methane in nano-pores. CNN is a technological approach to capture spatial features of image data and is effective for prediction problem involving image data as an input. The methane molecular distribution was predicted applying CNN method, which captured pore geometry and combined the bulk density as input. The results show that methane molecules accumulate at the wall of shale pores no matter the shape and size of the pores. Moreover, the methane distribution is related to the bulk density of methane in the shale system. In addition, we investigated the accuracy of the prediction results, and the predicted error is less 5%, which is highly acceptable in gas exploration field. As a result, this study provides understanding for methane flow mechanism in shale reservoir at larger scales.