Enhancing the Drilling Rate of Penetration Prediction by Coupling Mud Logging Data and Gene Expression Programming

The drilling rate of penetration (ROP) is essential in the oil and gas industry, influencing both efficiency and cost-effectiveness. Traditional methods, anchored in empirical correlations and basic mathematical models, frequently fail in capturing the intricate dynamics of the drilling process. However, the integration of mud logging data and machine learning (ML) techniques, including gene expression programming (GEP), marks a transformative advancement. This approach provides enhanced accuracy and stability in predictive models, indicating improved drilling performance and cost management. This study aims to i) develop a GEP-based predictive ROP model and ii) compare it with other ML-based models. The ML and GEP-based models are constructed using a database of 500 datasets of mud logging parameters, with 325 (65%) for training and 175 (35%) for validation. Through extensive analysis and experimentation, the GEP model demonstrated its effectiveness, showing a high precision with a correlation coefficient of 99.31%, least root mean square of 0.8457, and maximum relative error of 2.28 %, compared to the other ML-based ROP models, including support vector machine, neuro-fuzzy inference system, functional neural networks, random forest. The study concludes that ML techniques are capable of predicting ROP and GEP offers higher accuracy in ROP estimation compared to traditional models, offering promising guidelines for model development in the oil and gas industry.