The data-driven remaining useful life (RUL) prediction method has attracted extensive attention in recent years. However, the construction of the RUL prediction model based on the data-driven method is required amount number of labeled life-cycle data of mechanical equipment performance degradation. Collecting labeled life-cycle data is very time-consuming. Besides, it is necessary to train different data-driven RUL prediction models for the degradation modes of mechanical equipment under different operating conditions. Aiming to address these defects, an unsupervised domain adaptive transfer learning method based on bidirectional long short-term memory (BiLSTM) is proposed in this paper for RUL prediction. The proposed method is mainly composed of a feature extraction module, domain adaptation module, and regression prediction module. The features of data under different operating conditions are extracted through feature extraction module, and then the features under different operating conditions are aligned through domain adaptation module with multiple-kernel maximum mean discrepancies (MMD) method. Finally, the RUL of cross operating condition mechanical equipment is predicted through the linear regression layer. The prediction performance and effectiveness of the model are verified by the C-MAPSS dataset. The experimental results prove the proposed model provides an effective method for cross operating condition predictive diagnosis.