Deep learning method for a dynamic adsorption-diffusion model

John (Zehao) Yang


This work shows that the deep learning method can solve the problem of a PDE system effectively and accurately (error: less than 0.7%); The deep learning method can accurately handle the estimation problem of a PDE system (error: less than 0.5%); Shale gas production can be divided into three stages, and they are dominated by free gas, adsorbed gas and dissolved gas, respectively; The sequence of the amount of stored shale gas in GM reservoir: free gas > dissolved gas > adsorbed gas.