Machine learning assisted molecular modeling from biochemistry to petroleum engineering: A review

Abstract

Unconventional reservoirs have emerged as pivotal contributors, responsible for over 50% of total U.S. oil production. Yet, comprehending the intricate mechanisms of fluid transport in ultra-low permeability reservoirs, with characteristic diameters spanning from a few to several hundreds of nanometers, remains a formidable challenge. To address this challenge, molecular dynamics (MD) simulation has evolved as a potent tool for probing fluid transport properties in unconventional reservoirs. Simultaneously, machine learning (ML) has witnessed increasing adoption within the domain of petroleum engineering. The integration of MD simulation and ML has proven highly effective in disciplines like biochemistry and materials science. However, this amalgamation's application in petroleum engineering remains relatively scarce or non-existent. This paper endeavors to present a tutorial review tailored to non-experts, with a particular focus on petroleum engineers, to elucidate the intricacies of merging MD simulation and ML. The tutorial embarks on its journey with an initial introduction, establishing a foundational understanding of both MD simulation and ML. It then proceeds to showcase their applications in petroleum engineering. Furthermore, it acknowledges the constraints associated with MD simulations. Subsequently, the paper delineates how ML techniques can be harnessed to circumvent the limitations intrinsic to MD simulations. Notably, it delves into a comprehensive exploration of the Markov State Model (MSM) to underscore its potential in mitigating the intractable timescales synonymous with MD simulations. The paper culminates by outlining potential applications of this integration, including the in-depth examination of asphaltene behavior, encompassing phenomena like aggregation and self-assembly. This review underscores the feasibility of synergizing MD and ML techniques within the sphere of petroleum engineering. The authors anticipate that the ongoing advancement and deployment of ML-enhanced MD methodologies will usher in a new era of innovation and inquiry within the ambit of petroleum engineering.