A Hybrid Integrated Compositional Reservoir Simulator Coupling Machine Learning and Hard Computing Protocols

Abstract

Abstract The compositional representation of a petroleum system offers a greater detail in the phase equilibria calculations when compared to the black-oil approach. However, the slow and iterative nature of the numerical EOS based flash calculation is the primary drawback of compositional modeling. In this work we propose a hybrid model coupling machine learning and hard computing algorithms to accelerate the coupled wellbore hydraulic and numerical reservoir simulation processes. In this work, a robust compositional articificial neural network (ANN) based wellbore hydraulics tool is successfully coupled with a numerical compositional reservoir simulation model. The proposed hybrid simulation protocol is validated by comparing the results generated from the coupled ANN-numerical model against the fully numerical model, and a commercial compositional numerical simulation software package using a single-phase liquid, a single-phase gas and a two-phase liquid/gas case. Also, a comprehensive gas lift case study is discussed to compare the results and computational efficacy between the coupled ANN- numerical model and the full numerical model. It is observed that the mean deviation in the total oil production of the hybrid simulation protocol when compared to the full numerical model under the entire range of gas lift injection rates is approximately 5%, while the computational time taken by the coupled ANN-numerical model is 160 times less than the corresponding full numerical model. The proposed model can be employed as a practical gas lift optimization tool. More importantly, it gives the opportunity for simultaneously exploiting the strengths of both hard-computing and soft-computing algorithms.