Chapter 5 - Robust geomechanical characterization by analyzing the performance of shallow-learning regression methods using unsupervised clustering methods

Abstract

Sonic logging tool can acquire compressional and shear travel-time logs (DTC and DTS), which aid geomechanical characterization. Six shallow-learning regression models were trained to process 13 “easy-to-acquire” conventional logs for synthesizing the DTC and DTS logs. The six shallow-learning models used in this study are ordinary least squares (OLS), partial least squares (PLS), elastic net (EN), LASSO, multivariate adaptive regression splines (MARS), and artificial neural network (ANN). The shallow-learning models were trained and tested on logs from a 4240-ft depth interval in a shale reservoir in Well 1. Five clustering methods, namely, K-means, hierarchical clustering, DBSCAN, self-organizing map, and Gaussian mixture model, were applied on the 13 “easy-to-acquire” logs. Clusters identified using the K-means clustering have a strong correlation with the performance of the shallow-learning regression models. Dimensionality reduction technique is used to visualize the performance of the clustering methods. In conclusion, sonic logs can be synthesized using shallow-learning regression models, and K-means clustering can be used to determine the reliability of the sonic-log synthesis using the shallow-learning methods.