Data-Driven production optimization of a mature asset by development of a WLS regression-based solution

Austrian oil & gas operations are typical of a mature asset ( discovered in 1934 ), with nearly a thousand oil & gas wells in operation, with a mix of various artificial lift systems and reservoir drive mechanisms.

The heterogeneity of the well stock limits the possibility to apply some machine learning methods, such a Deep Neural Networks, but the low individual rates of most of the wells means that building individual physical simulation models is uneconomical.

We therefore structured the available production data to contextualize it, and calculated a reference production rate value that can be operationally reached: This reference is based on a Weighted Least Square model of linear regression, applied on an automatically selected subset of the production timeseries. This reference is then directly representative of the Best Day of production from the past period and is compared automatically with the actual measured production of the current day. The resulting deviations are then investigated by engineers to find the root cause and resolve the underlying issues.

The results of the application of this solution to the oil& gas fields of Austria are also presented in this abstract, highlighting the significant improvement of this application compared to the historical method