Lithology identification in the drilling process is a complex and constantly changing nonlinear process, and it is difficult to establish an accurate mathematical model. Due to the low interpretation accuracy of traditional identification methods and the difficulty of obtaining or missing logging curves, the ROP, weight on bit and other drilling parameters that are directly or indirectly related to lithology obtained through sensors are directly used for rock prediction. Based on machine learning technology, using BP neural network learning algorithm, according to the characteristics of stratum lithology, a neural network recognition lithology model was established, drilling parameter samples was constructed and the neural network toolbox in the software MATLAB was used to perform lithology recognition and analysis the effect of the number of samples in the training set on the accuracy of model recognition. The research results show that the output of the BP neural network is very accurate, describing the relationship between the collected drilling parameters and the lithology, and reflecting the superiority of the neural network. It has a positive effect on the identification of lithology in the drilling process, which is conducive to the reasonable selection of drill bit types, timely adjustment of drilling parameters and improvement of drilling efficiency. The lithology identification method is applied to bionic PDC drill bits, etc., while experiment and theory complement and support each other, it is convenient to use this method to carry out deeper research on intelligent oil drilling rigs.