Shadfar Davoudi from Skoltech has developed an AI-based approach to optimize drilling in heterogeneous carbonate reservoirs. The method integrates machine learning, explainable AI, and multi-objective optimization to balance drilling speed, torque, and energy consumption.
The researcher tested models including LSTM, Random Forest, and multilayer neural networks using 5,740 records from three vertical wells. The LSTM model proved most accurate, achieving an R-squared coefficient of 0.846 for penetration rate and 0.995 for torque.
Using SHAP and PDP methods, the study identified drill rotation speed as the most significant parameter. The model was then integrated into the NSGA-III optimization algorithm to find ideal combinations of axial load, rotation speed, and fluid flow rate.
For one well, the model estimated a potential penetration rate increase of 68.4%, alongside a 41.2% reduction in torque and a 17.9% decrease in mechanical energy. Another well showed a 66.9% speed increase and a 59.6% torque reduction.
The findings were published in Results in Engineering. The author notes these are surrogate model estimates based on historical data, and future work will focus on field testing and real-time adaptive optimization.
