LAUSR

Abstract Due to the mounting concern in the oil and gas industry about creating well-designed drilling plans and efficiently drilling new wells, analysis of offset well data has received extensive attention over the past few decades. By doing this, bottom hole assemblies, drill bit for a given section, drilling fluid, and the operational parameters such as weight on bit (WOB) and drilling string rotational speed (RPM) can be accurately determined. This makes it possible to achieve the most favorable Rate of Penetration (ROP) without causing drilling problems or non-productive Time (NPT). The present paper proposes a unique formation-based Artificial Neural Networks (ANN) model of ROP, which suits deviated wells with the application of Polycrystalline Diamond Compact (PDC) bits. In this study, nine different drilling functions that account for various factors affecting ROP are considered. Unlike the conventional ANN-ROP modeling technique, which directly considers raw drilling variables as inputs into ANN, the drilling functions are considered as inputs to ANN in the proposed ANN-ROP model. The drilling functions for this study are available in literature. The proposed ANN-ROP model is trained and tested using data collected from the 12 ¼ inches drilling section, which is the most extended drilling section of these wells. Furthermore, the prediction performance of the proposed models is evaluated in various formations. The results indicate that the proposed model is able to predict drilling ROP with high accuracy with the average Absolute Percentage Error (AAPE) being 0.83% for the training wells and 4.5% for the testing wells and with the R-squared obtained being 0.97 for the training wells and 0.93 for the testing wells. The proposed formation-based ANN-ROP methodology is shown to surpass the conventional ANN-ROP in terms of both accuracy and efficiency and the principle-based methodology in terms of accuracy.