In response to the challenges encountered during the drilling process, such as difficulties in increasing speed, unpredictable formation drillability, and high costs, a hybrid deep neural network model for predicting formation drillability measures was proposed. Given that multiple evaluation metrics were available for rock drillability, a novel metric for evaluating rock drillability was redefined as the drilling speed per unit of energy. The performance of three artificial intelligence models was compared using the coefficient of determination(R²), root mean square error(RMSE), and mean absolute error(MAE) as evaluation criteria. The direct impact of rock properties on drilling efficiency was recognized, and the lithology was categorized into five distinct grades and digitally encoded. The inclusion of lithological data is found to improve the model's predictive accuracy significantly. Among the models, the convolutional neural networks (CNN) -bidirectional gated recurrent unit (BIGRU)-Attention model demonstrates the most substantial enhancement, with its R² value increasing from 0.91 to 0.97, marking a 0.06 improvement. The results underscore the critical role of lithology in predicting formation drillability, and this the CNN-BIGRU-Attention model exhibits the highest predictive accuracy. It is recommended that this model be extensively applied in analogous geological settings to enhance drilling efficiency.