| Title | Development of a Recurrent Neural Network for Formation Temperature Estimation During Drilling |
|---|---|
| Authors | Yuta Takahashi, Shigemi Naganawa, Elvar K. Bjarkason |
| Year | 2026 |
| Conference | World Geothermal Congress |
| Keywords | |
| Abstract | Estimation of the formation temperature in real time during geothermal drilling is technically challenging due to thermal disturbances caused by circulating drilling fluids. This study aims to develop an artificial intelligence-based method for predicting static formation temperature (SFT) using surface and downhole logging data acquired during drilling operations. Synthetic datasets were generated using the GEOTEMP2 wellbore thermal simulator under a range of geothermal gradients and drilling conditions to reflect different geothermal environments. A recurrent neural network (RNN) was trained on these datasets to learn the relationship between the wellbore thermal behavior and the SFT. The input features consist of bottom-hole temperature (BHT) and mud outlet temperature recorded during drilling. The proposed method is designed to estimate SFTs using only temperature data recorded during drilling, without requiring wellbore temperature recovery observations during post-drilling shut-in periods. This study contributes to the integration of artificial intelligence into geothermal drilling workflows and offers potential applications in thermal diagnostics and decision support in high-temperature geothermal operations. |