Well ranking for in-fill drilling using machine learning with production and geological data

In machine learning, support-vector machines (SVM) are supervised learning models with associated learning algorithms that analyze data used for classification analysis. In this paper SVM-rank model is applied for productivity prediction of infill wells for giant Western Siberian oilfield. An essential condition for method’s application is availability of digital databases with representative results which allows adequate model training. Ranking algorithm also uses Voronoi diagram, proven as an approximation to the well drainage area. Complex method allows combine different reservoir and production parameters: productivity and water cut of surrounding wells, frac parameters etc without common reservoir dynamics model, which in this particular case is not able to clarify and confirm the parameters of the reservoir system. There is double model used: the first model utilizes productivity reservoir parameters, the second one uses capacity parameters. The rank of the first model is one of the training options for the second model, and both of them take into account all the geological and production information. The method can be particularly useful in complicated reservoirs, e.g. in dual porosity ones, where the relationship between formation parameters (permeability, porosity, saturation) and production rates is unclear and cannot be set by traditional development analysis, particularly in frac environment.