A molecular pump is a high vacuum acquisition piece of equipment that provides a clean vacuum environment for the Experimental Advanced Superconducting Tokamak (EAST) device. Its running state affects the smooth development of the EAST experiment. Because of fatigue degradation of internal components of the molecular pump, vacuum leakage may occur during long-term operation, causing secondary hazards to the device. In order to improve the accuracy of molecular pump fault prediction, based on the long short-term memory network (LSTM), the deep long short-term memory network (DE-LSTM) and the bidirectional long short-term memory network (Bi-LSTM) are combined. The deep bidirectional long short-term memory network (DE-Bi-LSTM) algorithm is proposed, and the piecewise linear degradation model is introduced to predict fault of the molecular pump. By collecting the vibration signals leaked in the atmosphere and running to the fault time series on the destructive test platform simulating molecular pump fault, data were extracted in the time domain. Finally, the obtained feature vector set was used as the input of the DE-Bi-LSTM algorithm through data standardization to train the model and realize the prediction of molecular pump fault. The experimental results show that the proposed method is optimal to LSTM, DE-LSTM, and Bi-LSTM in predicting performance.