Drilling rate of penetration (ROP) is an important indicator for achieving cost reduction in drilling operations and formulating plans. Accurately predicting the ROP is the first step to improving the ROP and optimizing the drilling plan. In view of the current problem of low accuracy in predicting ROP, this paper establishes a prediction model based on Transformer, performs time series processing on the data set, and conducts simulation experiments on single-well data sets, block data sets, partial block data (data below 3000 m), ascending order data sets, and sequential order data sets. It was found that the ascending order dataset had the highest prediction accuracy. To further enhance the prediction ability, the temporal convolutional network (TCN) is introduced to improve the accuracy of prediction. The actual drilling data of a certain block in the Sichuan Basin was selected for training. The data set was arranged in ascending order of blocks, and the prediction accuracy of the TCN-Transformer model reached 0.989. In addition, new well data outside the dataset was used for verification, and the prediction accuracy reached 0.979. It is believed that this method can be extended to the entire block and has good prediction accuracy, which can meet the field application requirements.
Yuehao et al. (Thu,) studied this question.