A Data Augmentation Method Based on TimeGAN for Air Absorbed Dose Rate Prediction

Air absorbed dose rate is a crucial indicator of radiation intensity, accurate prediction of air absorbed dose rate is crucial for nuclear safety assurance, energy management optimization, and nuclear facility risk management. Although deep learning-based prediction models have shown promising potential, their performance is limited by the quality and quantity of training data. To address the scarcity of nuclear radiation data, this paper proposes a data augmentation method based on TimeGAN, replacing the recurrent neural network generator with a Transformer-based architecture to more effectively capture the periodicity and long-term dependencies of radiation data. Furthermore, this paper employs a long short-term memory (LSTM) model to predict the air absorbed dose rate. Experimental results demonstrate that the improved method significantly improves the prediction performance of the LSTM model, achieving results comparable to real data, validating the effectiveness of this method in improving prediction accuracy and robustness. This study provides a feasible solution for nuclear radiation monitoring data augmentation, with potential applications in risk assessment and early warning.