Enhancing prediction accuracy for LOCA break sizes in nuclear power plants: A hybrid deep learning method with data augmentation and hyperparameter optimization

Abstract

In order to accurately predict the break size of loss of coolant accident, this study presents a novel break size prediction method based on a hybrid deep learning model that leverages data augmentation and optimization. This approach utilizes an interpolation method to augment limited sample data. By integrating Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks within the hybrid deep learning architecture, accurate extraction of both spatial and temporal information from the data is achieved. The hyperparameters of the hybrid model are optimized using an improved particle swarm optimization (IPSO) algorithm. Furthermore, a comparative analysis between the proposed algorithm and traditional methods is conducted, employing four out-of-sample data derived from novel test cases to evaluate the model's prediction accuracy. The results demonstrate that the proposed method can achieve high-precision deep learning modeling and effectively determine the break size of loss of coolant accident.