期刊文献+

反应堆过程控制强化学习可靠性评估方法研究

Research on Reliability Assessment Methodology for Reinforcement Learning of Reactor Process Control
下载PDF
导出
摘要 人工智能技术的发展推动了强化学习(RL)算法在工业过程控制中的应用。在航空、航天和核能等行业中,不仅要求控制算法准确性高,还对控制算法的可靠性提出高要求。对RL领域的可靠性评估方法进行了研究,提出了反应堆过程控制RL可靠性评估方法。该方法通过分析模型的推理性能以评估RL模型的决策能力,并创新性地提出了针对传感⁃控制系统的对抗样本(AE)生成方法。通过AE测试评估RL模型的抗干扰能力。利用所提评估方法对反应堆过程控制RL模型进行了可靠性评估。试验结果表明,所提方法可以评估实际应用中RL模型的可靠性。该研究为人工智能技术在反应堆控制应用的可靠性评估工作以及RL领域对抗测试方法研究提供参考。 The development of artificial intelligence technology promotes the application of reinforcement learning(RL)algorithms in industrial process control.In industries such as aviation,aerospace and nuclear energy,not only the control algorithms are required to be highly accurate,but also the reliability of the control algorithms is highly demanded.The reliability assessment method in the field of RL has been investigated,and a RL reliability assessment method for reactor process control is proposed.The method analyzes the inference performance of the model to assess the decision⁃making ability of the RL model,and innovatively proposes an adversarial examples(AE)generation method for the sensing⁃control system.The anti⁃interference capability of the RL model is evaluated by AE testing.The reliability of the RL model for reactor process control is evaluated using the proposed evaluation method.The test results show that the proposed method can assess the reliability of the RL model in practical applications.The study provides a reference for the reliability assessment work of artificial intelligence technology in reactor control applications and the research of adversarial testing methods in the field of RL.
作者 乔宇 颜瑞 黄百健 姚文卿 王少华 陈日罡 QIAO Yu;YAN Rui;HUANG Baijian;YAO Wenqing;WANG Shaohua;CHEN Rigang(School of Software and Microelectronics,Peking University,Beijing 102627,China;Chian Nuclear Power Engineering Co,.Ltd,.Beijing 100840,China)
出处 《自动化仪表》 CAS 2023年第S01期361-365,375,共6页 Process Automation Instrumentation
关键词 反应堆过程控制 可靠性评估 强化学习 人工智能 对抗性测试 对抗样本 Reactor process control Reliability assessment Reinforcement learning(RL) Artificial intelligence Adversarial testing Adversarial examples(AE)
  • 相关文献

参考文献1

二级参考文献3

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部