Improving rate of penetration prediction by combining data from an adjacent well in a geothermal project

Abstract

Drilling costs can account for more than 60% of the total cost of an Enhanced Geothermal System (EGS). Thus, drilling optimization is crucial for the success of EGS, especially with limited drilling data. In this study, the rate of penetration (ROP) was predicted using an Artificial Neural Network (ANN) model and drilling data from two wells situated 6 m apart at the surface in an EGS project. Ten drilling data cases that used either data from one or both wells were compared. The input parameters included depth, pore pressure gradient, equivalent circulating mud density, weight on bit, rotary speed, fraction of tooth worn away, mud density, flow rate, and mud viscosity. The lowest error was obtained for the case that combined a short data section located right before the prediction depth from the adjacent well with data resampled in a root-square-like manner before prediction in the predicted well. This arrangement led to a mean percent error of 18.5%. Besides highlighting the positive impact of root square data resampling, this assessment also showed the negative effect of including a data set from the adjacent well at the predicted depth during training.