The identification of rock types is of great significance to the safety and economic benefits of drilling engineering. Real-time rock type during drilling. The identification is mostly through logging while drilling， but it is rarely used in the field of geological exploration due to the high cost of logging while drilling. In this paper， the drilling parameters （ROP， WOB， rotational speed， bit torque， pump pressure， pump volume） of Jinchuan Scientific Drill are used to identify rock types through a fusion model. First， the noise of the drilling parameter data is reduced by the Savitzky-Golay smoothing filter， and then the data is normalized. Finally， the fusion model is used to predict rock types. The primary learners of the fusion model are Support Vector Machines， Artificial Neural Networks and Random Forests， and the weights of each model are calculated by the secondary learner Bayesian model averaging algorithm. The results show that the accuracy of the multi-model fusion algorithm is 0.9686， which is higher than that of each individual algorithm.