Machine Learning and Deep Learning Methods used in Safety Management of Nuclear Power Plants: A Survey

The nuclear power industry is currently a strategic sector in the national economy, along with nuclear energy being considered to be an essential source of national power supply and security. Under such circumstances, nuclear power plants (NPPs) have been constructed globally in recent decades, providing stable and large amounts of electricity to many countries and regions for quite a long time. However, due to the specialty of NPPs, safety management is always the top priority in their daily operations, like fault diagnosis and monitoring. At the same time, with the great development of artificial intelligence, machine learning and deep learning methods have been infiltrating lots of disciplines and resulting in more intelligent transformations in real industries. By the strength of better performance, machine learning and deep learning models have been introduced into research and practice of safety management in NPPs, not only producing more academic papers but also ensuring stable and safe operations of NPPs. Focusing on the safety management of NPPs, this article starts with the common trend of data analytics, also the evolving process of algorithms applied in academia and real practice, which is from model-driven methods to data-driven approaches. Then detailed applications of conventional machine learning, advanced deep learning and other related intelligent approaches used in the safety management of NPPs are comprehensively categorized and reviewed. Further, we make necessary summaries and discussions, proposing new ideas and perspectives to better promote the theoretical and practical development of safety management in NPPs.