Application of Genetic Algorithm on Data Driven Models for Optimized ROP Prediction

Abstract

The demand for cost-effective drilling operations in oil and gas exploration is ever growing. One of the important aspects to tackling the aforementioned difficulty is determining the optimal rate of penetration (ROP) of the drill bit. The most important optimization objective is to achieve a high optimal rate of penetration in safe and stable drilling conditions. Several machine learning models have been developed to predict ROP, however, there have been few studies that consider the different optimization algorithms needed to optimize the conventional developed models other than the conventional grid search and random search techniques. Genetic algorithm (GA) has gained much attention as methods of optimizing the predictions of machine learning algorithms in different fields of study. In this study, GA optimization algorithm was implemented to optimize 5 machine learning algorithms: Linear Regression, Decision Tree, Support Vector Machine, Random Forest, and Multilayer Perceptron algorithm while using torque, weight on bit, surface RPM, mud flow, pump pressure, downhole temperature and pressure, etc, as input parameters. Three scenarios were analyzed using a train-test split ratio of 70-30, 80-20 and 85-15 percent on all the developed models. The results from the comparative study of all models developed shows that the implementation of the GA optimization algorithms increased the individual ROP models, with the multilayer perceptron model having the highest coefficient of determination of 0.989% after GA optimization.