Drilling technology is an indispensable technical support for deep resource exploration， and the prediction of drilling efficiency is an important way to improve drilling technology. In response to the requirements for drilling speed-up in a certain block in the South China Sea， this paper collected actual drilling data from 10 wells， and these data were first interpolated and normalized. In order to eliminate the high correlation among different parameters， the initial 43 parameters were reduced to 21 common factors based on factor analysis， where there was no correlation between the 21 factors. Based on well number and depth， combining with a 10-fold cross-validation scheme， stratified sampling and grouping were performed on the original data. Through an optimized structure with a single hidden layer and 15 neurons， two neural network models were established on the basis of whether a hydraulic motor was used， and they both achieved an accuracy of over 96%. The model prediction shows that the use of speed-up tools in the target block with low silica content can effectively improve the drilling efficiency. At the same time， the model also predicts that for the high silicon content section， the use of speed-up drilling tools will increase wear on the drilling tools and cause a decrease in drilling speed. The results of the study show that the drilling speed prediction model based on a neural network can effectively make up for the differences among wellbores. Through accurate drilling speed prediction， it is possible to efficiently evaluate the effect of using speed-up tools and improve drilling efficiency.