Shiny web applications for unsupervised learning optimization applied to diagnostic fracture injection test event detection

Lukas Sadownyk, Marcelo Guarido, Danial Zeinabady, Erfan Sarvaramini, Zhenzihao Zhang, Farshad Tabasinejad, Hashem Salari, Kristopher A. Innanen, Christopher R. Clarkson

Diagnostic Fracture Injection Tests (DFIT), are commonly used to derive key parametersand other parameters for hydraulic fracture design and modeling. Although this processcan identify properties needed for well optimization, it is also time intensive and affectedby human interpretation bias. In this report, we address this adversity by applying unsupervised clustering methods: K-Means, DB-Scan, Hierarchical modeling, and Gaussian mixture models to identify point density variation that correlates to key parameters on a DFIT curve. A R-Studio Shiny Web App® is developed to apply these methods and provide a user-friendly platform for adjusting input variables and hyperparameters. Exploring the clustering approach emphasizes the importance of different variable combinations as well as noise considerations when interpreting a DFIT curve with clustering methods. Principle Component Analysis (PCA) further demonstrates why clusters occur where they do along a DFIT curve. Unsupervised clustering applied to DFIT data achieves an unbiased and quick workflow for event identification that can be scalable to large datasets.