Abstract: In order to apply the latest deep learning intelligent algorithm in the field of vibration fault diagnosis of steam turbine generator sets and promote the progress of intelligent diagnosis of vibration faults of steam turbine generator sets, the method of combining deep learning with expert experience is adopted, and according to the experience of vibration experts in on-site vibration fault diagnosis, the timing characteristics of vibration faults and the impact of operating parameters on faults are integrated into the traditional deep learning algorithms, and a vibration fault diagnosis algorithm based on the time series deep integration network is proposed. Experimental data verification results show that the algorithm greatly improves network performance, network robustness and migration ability while improving the accuracy of fault diagnosis.