Research on Key Parameter Prediction Technology of Small Modular Pressurized Water Reactor Under Ocean Conditions

Abstract Small modular pressurized water reactors can currently be installed in floating equipment such as ships and submarines. However, due to factors such as waves and grounding, their operation process is complicated. Long-term prediction of key transient parameters is required. By predicting the reactor over a long period of time Whether the operating status is normal, so as to take measures to correct its operating status and effectively reduce safety accidents, which helps to improve its safety. This article is based on the modeling of a small modular pressurized water reactor designed by Harbin Engineering University. During the modeling process, this article uses Multiple heat transfer tubes of a DC steam generator are used instead of multiple heat transfer tubes to simulate the heat transfer process, which simplifies the model, speeds up calculations, and improves transient accuracy. Later, the relap5 program developed under ocean conditions was used to simulate the rocking under ocean conditions. After obtaining the constant power transient data, this paper made predictions based on the data under ocean conditions. Because under undulating conditions, the key parameters in the reactor are affected by periodic gravity. Field-driven influences exhibit cyclical fluctuations. The greater the fluctuation amplitude, the larger the fluctuation period, the higher the parameter fluctuation range, the greater the power. The sensitivity of key parameters to changes in ocean motion conditions also increases. At present, predictions for small modular pressurized water reactors mostly stay at the direct analysis of data. Forecasting, since steady-state operating data has periodic fluctuations, there is a certain mathematical relationship between data and time. In order to improve the accuracy of forecasting, a joint model of VARIMA model and LSTM is proposed. Among them, VARIMA model can not only capture the trend changes of data, but also Can handle linear and periodic data. LSTM is a long short-term memory network that effectively captures the important long-term dependencies in the sequence and solves the gradient problem by introducing the concepts of memory cells, input gates, output gates and forgetting gates. Therefore, after dividing the data into periodic data and non-periodic data, VARIMA is used to predict the periodic data, and LSTM is used to predict the non-periodic data. After adding the two prediction values, the prediction result of the model is obtained., and compare the prediction results of the VARIMA+LSTM joint model with the optimized LSTM model to verify whether it can improve the prediction results.