Research on influencing factors for drilling rate in coal mines and its intelligent prediction methods

Abstract

In the field of drilling in coal mine underground, drilling rate (DR) is one of the most effective indicators for assessing drilling operations. Accurate prediction of DR is a prerequisite for the realization of intelligent drilling in coal mines, which is of great significance for optimizing drilling parameters, reducing operational costs, and ensuring safe and efficient drilling. In the present research, the influence of drilling parameters on DR are investigated and the intelligent prediction methods are developed to achieve accurate DR predictions with different machine learning algorithms based on several drilling parameters, including weight on bit, rotation speed, torque, and drilling depth. Initially, micro-drilling experiments are conducted to analyze the impact of coal rock mechanical properties, weight on bit, rotation speed, and drilling depth on torque and drilling rate. The experimental results indicate that during the underground coal mining drilling process, drilling rate gradually rises with the increasing weight on bit. Under higher rotation speed conditions, the influence of weight on bit on drilling speed becomes more pronounced. Increasing the rotational speed is advantageous for improving drilling speed, but the impact of rotational speed on drilling speed is more significant in softer coal seams. Subsequently, three different machine learning algorithms, namely K–Nearest Neighbors (KNN), Support Vector Regression (SVR), and Random Forest Regression (RFR), are used to build drilling rate prediction models based on on-site drilling data from underground coal mining, and the hyperparameters of those models are optimized with Particle Swarm Optimization (PSO) method. Finally, the prediction results of three drilling rate models, PSO–KNN, PSO–SVR, and PSO–RFR, are analyzed comparatively. The results show that PSO–RFR model offers the highest accuracy, with R2 of 0.963, MSE of 29.742. On the other hand, the PSO–SVR model exhibits the best robustness, with minimal changes in evaluation metrics after withstanding adversarial attacks. This study is beneficial for the realization of precise DR prediction, providing theoretical support for the optimization of intelligent drilling parameters in underground coal mining operations.

Keywords

• drilling parameter
• k-nearest neighbors
• random forest regression
• support vector regression
• particle swarm optimization
• rate of penetration prediction model