Real-Time Prediction of Rate of Penetration Using Multi-Gene Genetic Programming

The rate of penetration (ROP) is a crucial metric for assessing drilling efficiency. However, the ROP prediction models based on conventional machine learning have high accuracy but poor interpretability, and the offline models make it difficult to capture the dynamic impact of formation and drilling condition changes on ROP. In this paper, a real-time prediction method for ROP using multi-gene genetic programming (MGGP) algorithm is proposed. First, outliers and non-drilling conditions data are eliminated. Through the multicollinearity analysis, seven mud logging variables, weight on bit, rotary speed, torque, inlet flow, inlet density, standpipe pressure, and mud pit volume, are taken as input variables and filtered. Then the explicit expression between variables and ROP is constructed based on the MGGP algorithm, and the variables sensitivity is analyzed. Finally, a real-time update framework is proposed to capture the dynamic trend of ROP, with a coefficient of determination of 0.89 and a mean absolute percentage error of 5.44%. This study verifies the effectiveness of the MGGP algorithm in the prediction of ROP and improves the interpretability of the model, which provides a basis for complex geological drilling process.