The Design and Implementation of an LSTM-Based Steam Generator Level Prediction Model

The Long-Short Term Memory (LSTM) model is applied to the Steam Generator (SG) water level prediction in this work. The model is designed and implemented within the SIMULINK environment, where a real-time validation platform is also constructed using traditional PID controller. The SG water level feedback signal is fed to both the LSTM model and the PID controller, allowing the prediction to be generated online and compared to the actual controlled value. The results have demonstrated the functionality and advantages of the LSTM model, such as the high accuracy with prediction error of −1.1887 × 10–4, low loss value with MSE of 1.4130 × 10–8, and quick convergence during the simulation. Discoveries found in this work could enable future exploration of developing deep-learning-based control strategy for nuclear power plants.