PTFlash : A vectorized and parallel deep learning framework for two-phase flash calculation

Phase equilibrium calculations are an essential part of numerical simulations multi-component multi-phase flow in porous media, accounting for the largest share computational time. In this work, we introduce a fast and parallel framework, PTFlash, that vectorizes algorithms required two-phase flash calculation using PyTorch, can facilitate wide range downstream applications. Vectorization promotes parallelism consequently leads to attractive hardware-agnostic acceleration. addition, further accelerate design two task-specific neural networks, one predicting stability given mixtures other providing estimates distribution coefficients, which trained offline help shorten computation time by sidestepping analysis reducing number iterations reach convergence. The evaluation PTFlash was conducted on three case studies involving hydrocarbons, CO2 N2, phase tested over large temperature, pressure composition conditions, Soave–Redlich–Kwong (SRK) equation state. We compare with in-house thermodynamic library, Carnot, written C++ performing CPU. Results show speed-ups up order magnitude scale calculations, while maintaining perfect precision reference solution provided Carnot.