Machine Learning of Spatially Varying Decline Curves for the Duvernay Formation

The focus of this paper is on Duvernay shale formation in Alberta, Canada. The objective is to provide, based on existing data of production, completion and geological parameters, an automated machine- learning approach to determine the spatial variation in decline type curves for gas production. This model will enable the prediction and uncertainty quantification of production profiles for new target wells or areas in the basin. The project is based on publicly available monthly production data from most of the producing wells of the Duvernay formation. We use k-means to cluster 273 wells, using geological parameters (thickness, porosity, etc.), completion parameters (horizontal section length, proppant volume, etc.), spatial location, fluid window, and production curves. Based on the clustering results, a machine learning classification is used to draw distinct geographic regions, within which the combination of geological, completion, and production factors is fairly similar. A support vector machine approach is used to create maps of clusters and quantify its uncertainty. In addition, functional classification and regression trees (CART) is used to indicate the most important/sensitive factors that should be used for clustering. The results show that the unsupervised method, k-means, performs equally as well as the supervised CART method. The methodology is flexible and allows for quick changes in the variables used in clustering; the transfer to another dataset or basin is straightforward.