Abstract: As the nerve center of nuclear power plants, the stability and reliability of the instrumentation and control (I&C) system are the key to ensuring the safe operation of nuclear power plants. Predicting the remaining life of nuclear power I&C system circuit boards can reduce the maintenance and acquisition costs caused by direct replacement and reduce the risk of damage to key equipments, which is an important means to improve system reliability. Data-driven and physics of failure (PoF) are the two main methods in the field of remaining life prediction, but both methods have limitations: The method based on PoF requires less data information, but it is difficult to establish a complex system fault physical model. The accuracy of data-driven method depends on large data. Only by learning and training with enough data can obtain better prediction accuracy. The existing fusion algorithms are difficult to achieve accurate remaining life prediction for nuclear power products with diverse stresses and complex structures. In addition, the current research focuses on the fusion of single fault mechanism and data-driven algorithm, and the fault physical model of single fault mechanism is difficult to meet the solving requirements of complex systems with variable environmental stresses. A method for predicting residual life of nuclear power products with diverse stresses and complex structures is urgently needed. In order to solve the above problems, in this paper a remaining life prediction method for nuclear power products was proposed based on PoF and data-driven fusion algorithms, which uses the generalized Arrhenius model for PoF modeling, the Wiener process was used to model the degradation process and generate degradation data, and the effective fusion of the PoF and data-driven method was achieved with long short-term memory (LSTM) neural network. Based on the proposed method, the remaining service life prediction of DC-DC board of the instrument control system of a nuclear power plant was completed. In the prediction of the remaining life of DC-DC board, it is found that more accurate prediction effect will be obtained after the degraded data generated by Wiener process are used for prediction. The prediction effect of LSTM, bi-directional long short-term memory (BiLSTM) and support vector machines (SVM) were compared. Considering the prediction effect and prediction error comprehensively, LSTM algorithm achieves the best prediction effect. This further verifies the validity and accuracy of the proposed method. The research results can be used to guide the preventive maintenance of instrumentation and control equipment in nuclear power plants, and also provide reference and direction for equipment reliability management.