Using Machine Learning to Classify Phase Behavior of Oil/Water/Surfactant Systems

According to BP Statistical Review of World Energy in 2019, total oil production from the US in 2018 was about 15 million barrels per day (MBPD). Thanks to the booming shale oil production, United States has become the world’s top oil producer. In fact, estimates show that up to two thirds of conventional crude oil in mature fields remains unproduced due to the physics of fluid flow. The techniques of chemical enhanced oil recovery could overcome the physical force holding hydrocarbons, and turn these accumulations into oil reserve, which would enhance the US energy security and maintain economic growth. For the oil/water/surfactant system, the goal is to form a microemulsion phase achieving the lowest interfacial tension, which increases the capillary number and dramatically recovers the remaining oil fraction within the pore. Therefore, it is critically important to understand the phase behavior for the oil/water/surfactant systems. In collaboration with Pioneer Oil Company, our current effort is to optimize the selection of surfactants and the constituents of the surfactant blend, which turns to be a high dimensional problem. In addition to the conventional analysis, we employ machine learning techniques to solve the system as a multinomial classification problem.