A novel time-series rate of penetration model utilizing bat algorithm and long short-term memory neural network

As drilling rate of penetration serves as a critical metric in assessing deep geological drilling, establishing an appropriate rate of penetration model holds significant importance for optimizing and controlling the drilling process. However, the intricate nonlinearity and time-series characteristic pose challenges in modeling a reliable drilling rate model, while the model parameter design exhibits high-dimensional variability. To address these issues, a new model for rate of penetration is proposed to tackle nonlinear and temporal challenges. First, the support vector regression (SVR) method was utilized to construct the rate of penetration model, and a new improved bat algorithm (IBA) was devised to determine the hyperparameters of the rate of penetration model. Subsequently, the long short-term memory neural network (LSTMNN) was used to fine-tune the model’s output using time series data. The IBA underwent five enhancements aimed at improving global search capabilities, outperforming other algorithms in the IEEE CEC2005 benchmark functions. The rate of penetration model, developed based on SVR, LSTMNN, and IBA, underwent testing using actual drilling site data. The simulation results demonstrated that the developed model exhibited higher prediction accuracy compared to other methods.