Full text loading...
-
Predicting variability in well performance using the concept of shale capacity: An application of machine learning techniques
- Source: First Break, Volume 38, Issue 9, Sep 2020, p. 43 - 49
-
- 01 Sep 2020
Abstract
The production from a shale well depends on how accurately that well has been placed through zones with superior reservoir quality and completion quality. To be able to locate such pockets, an integration of different types of reservoir properties such as organic richness, frackability, fracture density and porosity is essential. One way of achieving this is by using cut off values for the different reservoir properties and generating a shale capacity volume. However, in this regard, a couple of things need to be borne in mind. Not only seismic data provide indirect measurements of the individual reservoir properties, but different seismically derived attributes may contribute to the computation of the individual components of the shale capacity volume. Therefore, the definition of the different reservoir property cut off values is subjective and difficult to finalize. What may be desirable is that the different seismically derived attributes are interpreted simultaneously for predicting the performance of a well, which may not be a straightforward task. The application of machine learning techniques seems like an attractive alternative here, which we explore in this article. For doing so, data examples from the Delaware Basin as well as the Western Canadian Sedimentary Basin (WCSB) have been considered. The dataset from the Delaware Basin is used to gain confidence in applying unsupervised machine learning techniques as one of them is used thereafter on the dataset from WCSB, where production data associated with different wells were available.