Nuclear energy can make an important contribution to low-carbon energy supply for Industry 4.0, while Industry 4.0 can reform this industry in return. As a typical and complex man-machine-network integration system, various faults, insufficient automation and stressed human operators limit the further popularization of nuclear power plants (NPPs) while these issues can be addressed by the aid of artificial intelligence (AI) technologies. In this work, we try to present a systemic review of how AI can benefit NPPs in a top-to-down fashion. We discuss limitations in current NPPs and introduce the concept of Nuclear Power Plant Human-Cyber-Physical System (NPPHCPS) as the top-level design. Then, we category AI-related nuclear power applications into Physical-Plant-Centered and Human-Operator-Centered technologies and review research works from 7 typical NPP functional scenarios in the recent two decades. In each NPP functional scenario, how researchers integrate AI into NPPs is presented following timeline. We hope this review can be used as the guideline for NPPs' Design in the future and contribute to green Industry 4.0.

How to obtain good value estimation is a critical problem in Reinforcement Learning (RL). Current value estimation methods in continuous control, such as DDPG and TD3, suffer from unnecessary over- or under- estimation. In this paper, we explore the potential of double actors, which has been neglected for a long time, for better value estimation in the continuous setting. First, we interestingly find that double actors improve the exploration ability of the agent. Next, we uncover the bias alleviation property of double actors in handling overestimation with single critic, and underestimation with double critics respectively. Finally, to mitigate the potentially pessimistic value estimate in double critics, we propose to regularize the critics under double actors architecture. Together, we present Double Actors Regularized Critics (DARC) algorithm. Extensive experiments on challenging continuous control benchmarks, MuJoCo and PyBullet, show that DARC significantly outperforms current baselines with higher average return and better sample efficiency.

No abstract

No abstract