The accurate prediction of rate of penetration (ROP) has a crucial role in improving efficiency and minimizing cost in geological drilling process. Considering the drilling characteristics of strong nonlinearity, complexity, multiple variables and drilling conditions in drilling process, an online hybrid prediction model based on the drilling data is developed to achieve high accuracy prediction of the ROP. First, mutual information analysis is used to determine the appropriate model inputs. Then,k-nearest neighbor algorithm and dynamic time warping (KNN–DTW) are combined to identify drilling condition. After that, ROP prediction model is established by support vector regression (SVR) method. The hyperparameters of SVR method are obtained by hybrid bat algorithm (HBA) and nondominated sorting genetic algorithm II (NSGA-II) based on the identified drilling condition. Finally, a modified sliding window method is developed to update the prediction model to deal with complex and variable drilling process. The simulation results show that our method has higher accuracy than other methods, and our method can identify the drilling condition and provide guidance for the drilling operation.

准确预测渗透率（ROP）在提高地质钻探过程的效率和最小化成本方面具有至关重要的作用。考虑到钻井过程中强烈的非线性，复杂性，多变量和钻井条件等特点，建立了基于钻井数据的在线混合预测模型，以实现对ROP的高精度预测。首先，使用相互信息分析来确定适当的模型输入。然后，结合k近邻算法和动态时间规整（KNN–DTW）来识别钻井条件。之后，通过支持向量回归（SVR）方法建立了ROP预测模型。根据确定的钻井条件，通过混合蝙蝠算法（HBA）和非支配排序遗传算法II（NSGA-II）获得SVR方法的超参数。最后，开发了一种改进的滑动窗口方法来更新预测模型，以处理复杂而可变的钻探过程。仿真结果表明，该方法比其他方法具有更高的精度，可以识别钻井条件，为钻井作业提供指导。