Reinforcement Learning Based Automatic Synchronization Method for Nuclear Power Digital Twin Model

Abstract

The objective of this research is to tackle the synchronization challenge between digital twin models of nuclear power and their corresponding real-world units, thereby enhancing the safety protocols of nuclear power operations. Central to achieving this synchronization, the study zeroes in on optimizing parameters. This entails conducting parameter optimization on the model whenever discrepancies in synchronization between the digital twin and the actual power units emerge. A novel optimization algorithm, termed TD3PSO, is introduced within this manuscript. Notably, this algorithm possesses the ability to adaptively fine-tune hyperparameters, significantly diminishing the reliance on human intervention. Validation of the proposed method is carried out through experiments using a digital twin model of a nuclear power steam system. The findings underscore that, in comparison to conventional manual adjustments, our method achieves a substantial 93.93% enhancement in accuracy. Further, to ascertain the algorithm’s universality and robustness, it was evaluated against a range of benchmark functions. Across these tests, the TD3PSO algorithm consistently surpassed the performance metrics of traditional Particle Swarm Optimization algorithms. Consequently, this study furnishes a potent solution to the issue of synchronization between nuclear power digital twin models and actual units, marking a significant milestone in terms of its practical and theoretical contributions.

Keywords

• Nuclear power
• automated synchronization
• reinforcement learning
• particle swarm optimization