Nuclear power plant is an efficient energy conversion facility, which provides stable power and reduces greenhouse gas emissions. However, it still faces challenges such as high safety risk and complex nuclear waste disposal. Today's fault diagnosis mostly relies on manual experience and periodic detection. This method has some shortcomings such as slow response and limited accuracy. Therefore, a fault diagnosis model of nuclear power plant based on CNN-LSTM is proposed in this study. The purpose of this study is to improve the fault diagnosis ability of nuclear power plant, ensure that the fault can be diagnosed as early as possible, and ensure that the operator of nuclear power plant can get the fault information in time, thus improving the safety and stability of nuclear power system. In this paper, a fault diagnosis method based on time sliding window, CNN and LSTM is constructed. The time information of time series is retained by time sliding window, the data features are extracted by CNN, and finally the feature information is diagnosed by LSTM model. This method is verified based on Qinshan simulator data. The results show that the fault diagnosis method based on CNN-LSTM has higher fault diagnosis accuracy than the traditional LSTM, and the model can diagnose the approximate location of the accident. Therefore, the working condition of the data can be diagnosed with high accuracy by the fault diagnosis method of nuclear power plant based on CNN-LSTM, which can provide strong support for the safety and stability of nuclear power system.

nuclear power plant,fault diagnosis,Long Short-Term Memory Networks,convolutional neural network