Research on rate of penetration prediction for geothermal wells based on a transformer-GRU model

Abstract

Rate of penetration (ROP) in geothermal drilling is jointly controlled by multiple operational variables and often shows strong fluctuations, which makes accurate prediction difficult. To address this issue, a Transformer-GRU-based prediction method for geothermal well ROP is proposed. First, the publicly available Utah FORGE geothermal drilling data are preprocessed. Then, input features are constructed using depth, weight on bit, pump pressure, rotary speed, torque, flow rate, temperature, and hook load. Finally, a time-series prediction model for ROP is established by introducing a GRU structure into the Transformer model. The evaluation results indicate that the proposed model performs better than both hybrid and single models across several key metrics, effectively improves ROP prediction accuracy, and demonstrates strong adaptability and practical value for handling highly fluctuating and unstable ROP sequences.