Multi-Feature Fusion Multi-Step State Prediction of Nuclear Power Sensor Based on LSTM

• DOI: 10.13832/j.jnpe.2021.04.0208
• Publication Date: 2021-08-15
• Authors: Siyuan Zhang; Lu Tianyu; Hui Zeng; Xu Chun; Zhuo Zhang; Qingyu Huang; Zhang Yaoyi; Yuanmei Wang
• Keywords: Feature (linguistics); Univariate

Aiming at the problem in prediction of nuclear power working condition parameter, this paper uses a large number time series collected by plant sensor detection system to propose multi-feature fusion multi-step state model based on long short-term memory network (LSTM). This takes SG1 steam pressure data real-time parameter plants as research object. Firstly, is preprocessed for problems missing and inconsistent sampling scales, then structural design modeling completed LSTM. Finally, proposed compared with models such Recurrent Neural Network (RNN), Gated Unit (GRU), Model-S1 layer univariate Experimental results show that fitting performance are overall optimization, it also verifies applicability deep learning method LSTM field operation safety assurance.