核反应堆传感器智能在线监测技术与鲁棒性 自校正方法的研究

A novel training dataset processing method with high robustness was proposed to address the poor robustness in data-driven analytic redundant sensor models, and an sensor online monitoring and self-correction method was developed based on a auto-associative multivariate Long Short-Term Memory (LSTM) artificial neural network model. The method was validated using actual sensor measurement data retrieved from a pressurized water reactor engineering test facility. The results indicate that this research method can achieve high-precision and robust reconstruction of sensor signals, hence meets the requirements of online monitoring and robust self-correction of nuclear reactor sensors.