Framework for automated generation of real-time rate of penetration models

Abstract

Due to highly uncertain underground conditions, most estimations of drilling and geological parameters must be performed using numerical modeling, thus by fitting empirical models to measurement data or by using an approach based on data analysis. In this paper, a combination of these two solutions is proposed for a realistic replication of historical drilling data in a drilling software simulator environment e.g., the DrillSIM:600 at Drilling Simulator Celle (Clausthal University of Technology). The whole process ranging from the modeling over the model evaluation and selection to the integration in the simulator environment is integrated within the presented framework. For this study several empirical models and five machine learning algorithms were applied to benchmark the performance on predicting the drilling parameter Rate of Penetration (ROP) using data from a single geothermal well located in northern Germany. The results of the comparison between the classical and the data driven models prove that the use of data analysis methods is a very valuable alternative to explore, describe, diagnose, and predict the underground conditions and drilling parameters. Those can be used to optimize the entire drilling process by replicating critical situations, recommending better practices or emphasizing the proper performance of a whole drilling project. Thus, by turning raw historical data in valuable insights, the further education of decision makers and rig crews can be significantly enhanced.