Machine learning model for real-time prediction of bottom-hole circulating temperature in geothermal wells

Background

Drilling high-temperature geothermal wells presents significant technical and economic challenges due to the extreme conditions encountered deep underground. Effective thermal management is crucial to ensure that the bottom-hole circulating temperature (BHCT) remains within the operational limits of downhole tools, which is essential for the safe and economical exploitation of geothermal resources. While some oil and gas wells encounter temperatures up to 150°C, geothermal wells often exceed these temperatures, necessitating precise control and understanding of BHCT to prevent damage to drilling equipment and avoid unplanned trips. Accurate real-time prediction of BHCT is therefore paramount for implementing proactive temperature management strategies during geothermal drilling operations.

Historically, challenges such as temperature modeling and control, drilling rate optimization, and cuttings transport phenomena have been addressed using physics-based models. While these models are reliable, they are also computationally intensive and generally not suited for real-time applications. Existing machine learning (ML) models have emerged as a promising alternative, offering potentially more efficient means to model and predict BHCT. However, these ML models primarily focus on steady-state conditions with constant drilling parameters, overlooking the dynamic nature of drilling operations. Most numerical models, although precise, require extensive computational times and involve complex simulation processes, making them impractical for real-time decision-making.

Consequently, there is an urgent need for advanced real-time temperature models that can capture the full-scale transient behavior of BHCT, allowing for continuous monitoring and dynamic adjustment of cooling strategies to meet operational demands.

Technology overview

The technology described focuses on the real-time prediction of bottom-hole circulating temperature (BHCT) in geothermal wells using advanced machine learning models. This is particularly vital for managing the high temperatures encountered during geothermal drilling, which can exceed the operational limits of downhole tools. The technology employs a range of machine learning models, including deep neural networks (DNN), support vector machines (SVR), random forests (RF), extreme gradient boosting regressor (XGBoost), long-short term memory (LSTM) networks, and ensemble (Stacked) algorithms. These models are trained on a large dataset generated from an integrated thermo-hydraulic model validated with real field data from the Utah FORGE project. The LSTM model excels in predicting the transient behavior of BHCT with high accuracy and low computational cost, making it suitable for real-time applications.

What differentiates this technology is its ability to predict BHCT dynamically and accurately in real-time, which is crucial for proactive temperature management during geothermal drilling. Traditional physics-based models, while precise, are computationally intensive and not suitable for real-time applications. In contrast, the LSTM model offers excellent generalization capability, stability, and precision without the high computational costs.

This model can adapt to both stationary and dynamic drilling conditions, effectively preventing temperature-related drilling problems and enhancing operational efficiency. The use of machine learning models, particularly the LSTM, represents a significant advancement over existing methods by providing a robust, efficient, and scalable solution for managing thermal dynamics in geothermal wells.

Benefits

Accurate real-time prediction of bottom-hole circulating temperature (BHCT)
Enhances proactive temperature management during geothermal drilling
Mitigates thermally induced challenges and prevents downhole tool failures
Reduces unplanned and unnecessary bit/BHA trips
Improves drilling efficiency and reduces operational costs
Provides a computationally efficient alternative to traditional numerical models
Adapts to dynamic drilling conditions with high accuracy
Facilitates optimal decision-making during drilling operations