To transfer the life prediction model, an LSTM-fine-tune(long short-term memory fine tune) model was proposed. The model was trained by using experimental data under ideal conditions. During the transfer process, part of the LSTM network layer was frozen, and other parts of the network were modified by using data in actual service environment. In order to verify the generalization ability of the model, sinusoidal functions with different phases and amplitudes to generate data were used, obtained the knowledge of the sinusoidal function, and applied it to the regression of other sinusoidal functions. The results show that the LSTM-fine-tune model can be fitted quickly, and the average mean square error is only 1.033 5. It is significantly lower than the direct prediction error 1.536 8. In order to test the generalization ability of this method through actual monitoring data, the data of oxygen concentrators under test conditions and actual service environment respectively is obtained, verifies the generalization ability of the model. The results show that the prediction accuracy of the training set is improved by 43.0% and that of the test set is improved by 20.2%.