Maximizing Drilling Efficiency in Deviated Wells Using AI Based ROP Optimization

Résumé fourni par la source

Abstract This research presents a machine learning-based approach aimed at optimizing the rate of penetration (ROP) in S-shaped wells. The method uses historical drilling data from wells in Southern Iraq to find out the best drilling operating variables. Dynamic Time Warping (DTW) is applied to prioritize data from wells with similar characteristics to enhance relevance. The framework is adaptable, incorporating new field data and expert insights as necessary. The performance of the method was evaluated by comparing the results to existing well data, with clear improvements identified. Initially, two machine learning models—Extra Trees Regressor and XGBoost—were utilized to predict ROP using 17 different input factors, such as torque and SPP. The accuracy of these models was assessed using correlation coefficients, and Extra Trees Regressor, the more accurate model, was selected to proceed into the second phase. Next, Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Firework Algorithm (FWA) were used to fine-tune key parameters like WOB, RPM, surface pump pressure (SPP), and flow rate to further improve ROP. The dataset comprised data from 6 wells, the data was partitioned to 75% to train the ML models and 25% to test them. In the prediction phase, the correlation coefficients for the models were: Extra Trees Regressor (0.99) and XGBoost (0.97) for the training set, and Extra Trees Regressor (0.96) and XGBoost (0.94) for the testing set. The average optimization improvements were 30.66% for GA, 22.33% for PSO, and 14.5% for FWA.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.