Abstract: In view of the lithology identification technology problems encountered by mine drilling rigs in the process of drilling construction, a lithology identification method based on Ensemble Empirical Mode Decomposition (EEMD) and Product Neural Networks (PNN) optimized by Differential Evolution (DE)algorithm is proposed, which combines the vibration signals collected behind the bit when cutting rock. Firstly, Synthesizing drilling simulation signals by using sinusoidal signals with different amplitude-frequency characteristics, the high-frequency high-energy signal with amplitude of 10 g and frequency of 50 Hz and the low-frequency low-energy signal with amplitude of 5 g and frequency of 20 Hz are respectively selected to represent the rock and coal seam encountered by drilling. The modal decomposition results of Empirical Mode Decomposition (EMD) and EEMD with multiple Gaussian white noise are compared, verifying the application effect of EEMD in solving modal aliasing; Then, the 1 s-long vibration signals of coal cut, mudstone cut and sandstone cut from a mine in guizhou province, China were decomposed by EEMD, the spectrum analysis was carried out to obtain the corresponding amplitude-frequency characteristics, The first 7 Intrinsic Mode Function (IMF) components were selected as the characteristic modes, the characteristic vectors of 14 parameters were constructed by normalizing the energy and singular value decomposition. Finally, 160 groups of 1 s-long vibration signals of cutting coal, cutting mudstone and cutting sandstone from a mine in guizhou province, China were constructed with characteristic vectors, and the DE-PNN was input for training and verification of identification effect, the optimized neural network has significantly improved the lithology recognition rate of the remaining samples. The results show that the vibration signal of cutting rock can be processed by probability neural network optimized by EEMD decomposition and DE algorithm, and lithology identification can be carried out efficiently, quickly and accurately, providing new reference significance for further lithology identification and better guiding the development of mine drilling. The results show that the vibration signal of cutting rock is processed by EEMD decomposition and probability neural network optimized by DE algorithm, which can identify lithology efficiently, quickly and accurately. It provides a new reference for further lithology identification and better guide the development of mine drilling.