Real-time and multi-objective optimization of rate-of-penetration using machine learning methods

Rate of penetration and mechanical specific energy are two widely used objectives when optimizing the drilling process, yet a simultaneous optimization of both is still a challenge due to their strong non-linear relationship. In this paper, a framework for real-time and multi-objective optimization of the two targets using artificial intelligence techniques is proposed. The framework combines all relevant steps into a single, real-time workflow, including training the prediction model, performing multi-objective optimization and making a decision. Random Forest, Support Vector Regression, and Neural Network are selected to predict rate of penetration and calculate mechanical specific energy. The Non-dominated Sorting Genetic Algorithm II is then used for multiobjective optimization with rate of penetration and mechanical specific energy as the optimization objectives, and the Technique for Order Preference by Similarity to Ideal Solution approach is used for decision-making. To realize real-time optimization, the flexible window method is employed to dynamically update the framework using real-time data streams. Finally, five field drilling datasets are used to test the proposed framework. Results indicate that Random Forest performs better in real-time prediction and optimization than Neural Network and Support Vector Regression. The performance of the real-time rate of penetration prediction is satisfying with an average relative error of 6.65%-12.54%. Additionally, the multi-objective optimization, which simultaneously improves rate of penetration and lowers mechanical specific energy, results in significant gains in drilling efficiency. More specifically, at least 70% of the rate of penetration are improved and it can be increased by 1.2 times, while more than 80% of the mechanical specific energy is lowered by as much as 30% in the best case. This work offers a practical method that significantly improves drilling efficiency in the field.