In the context of mature oil fields, a long history of production data is available together with a good knowledge of geological parameters of these wells. This data can be used to predict the cumulative production curve for a well depending on its physical and geological properties.
The objective is to classify the wells among different classes of production behavior using information on the well, such as position, reservoir depth, rock type…
The solution is two-part. It starts with an unsupervised classification of the wells depending of their cumulative production curve, sorting wells into different clusters. Then a supervised classification is run, aiming at predicting the cluster of a well depending on its physical and geological parameters, and finding the maximal number of different clusters allowing good classification performances.
Tested on an operational data set from a producing field onshore North Africa, the methodology classified the wells with a F1-score of 95% and 7 different clusters. Besides, it led to the ranking of the physical and geological variables in terms of impact on the well production profiles.
This methodology could be applied to have a fast estimate of the cumulative production curve of a well to be drilled. Further developement could focus on packaging the algorithms in a web service and integrating them within an interactive tool for well data screening.