Application of machine learning and well log attributes in geothermal drilling

As the world is moving towards carbon-neutral energy resources, geothermal has potential to become one of the most compatible avenues. The geothermal exploration requires drilling operations in high-temperature environments and has hard-rock formations. The traditional drilling systems have been designed for optimal performance in oil and gas reservoirs. However, the unique characteristics of the geothermal reservoirs require a fresh look on the available systems. One approach can be to use the data generated during field operations and then design, test, and validate the operational processes. Here, we have applied machine learning algorithms and different architectures of deep-learning to capture the patterns in the operational parameters and understand its implications on the real-time monitoring systems. Also, a robust framework is proposed for building classification predictors for lost circulation problems. In this paper, we used the FORGE well logging data and synthesize the evolution of dynamic data. We used the surface drilling data as input for supervised machine learning algorithms such as k-Nearest Neighbors, Random Forest, Decision tree, Gradient Boosting, and Deep Learning models with hidden layers. Finally, we identified hazardous zones using the classification models. Additionally, the log attribute based on change point analysis, is used as an additional input parameter to train these classification models. Result analysis and comparative studies of these models show that the random forest and gradient boosting outperforms the other classifiers. Based on this study, we developed and demonstrated the suitability of a incremental learning framework for real-time monitoring systems.