(685f) Prediction of Production Performance of Unconventional Reservoirs Using Data-Driven Methods

Evaluation of the early production data plays a vital role in minimizing uncertainties associated with developing unconventional reservoirs. The traditional decline curve analyses are not reliable for analyzing the early production data due to the long transient flow in unconventional reservoirs. Other challenges of developing unconventional reservoirs include multiphase flow, uncertain geological properties, complex fluid flow mechanisms, and variable completion properties, which cannot be easily modeled using empirical decline models. Fundamental equations and reservoir simulations can also be used for modeling the production performance of unconventional reservoirs, which are effective but also time-consuming. Furthermore, it doesn’t satisfy the current need to drill many wells in a short amount of time to improve the return on investment.

This study presents a data-driven approach to forecast the production from unconventional reservoirs with limited production histories using multivariate statistics and traditional decline curve analysis. We use an unsupervised machine learning algorithm to identify the patterns and regularities on a large production data set. This algorithm is then applied to large sets of field data to create an uncorrelated weighted sum of ‘representative curves,’ the truncated summation of which yields an approximation of the original data.

The method presented is also applicable to estimate the gas to oil ratio (GOR) for liquid-rich shale reservoirs with complex phase behavior. As the underlying data capture the effect of multiphase flow automatically, this method presents a reliable alternative to estimate the secondary phase from volatile oil and gas condensate reservoirs.