Efficient Monte Carlo Methods for Characterization of Safety Margins of Nuclear Power Plants

该奖项的目标是设计出有效的计算方法，用于对核电站进行风险和安全分析。核能机构（Nuclear Energy Agency）最近的一项国际努力制定了一个新的框架，即风险知情的安全边际特征，以解决当今核工业面临的几个关键问题。这些发展的影响，包括延长老化设施的许可证和以更高的功率运行，可能会深刻影响安全边际，迫切需要更好地了解这些变化的风险。事实上，2011年日本福岛第一核电站（Fukushima Daiichi）的灾难性事故清楚地表明，准确衡量风险至关重要。除了进行有效的核安全分析之外，这些技术还可以应用于土木和机械工程、灾难建模和金融方面的攻击问题。本研究将理论和实践相结合，包括设计新的蒙特卡罗算法，正式建立其有效性，并实施它们来分析现有核电站的风险。这项工作的核心将集中于减少方差的技术，以便有效地模拟核设施的风险和安全分析。由于当前的模拟需要大量的计算工作，应用方差减少对于执行安全分析是必不可少的。如果成功，新方法将大大减少计算时间，也许是数量级上的减少。为了进一步解释结果估计的统计误差，研究者还计划产生渐近有效的置信区间。

The goal of the award is to devise efficient computational methods for performing risk and safety analyses of nuclear power plants. A recent international effort of the Nuclear Energy Agency formulated a new framework, known as risk-informed safety-margin characterization, to address several critical issues facing the nuclear industry today. The impacts of these developments, which include extending licenses for aging facilities and operating them at higher power levels, can profoundly affect safety margins, and there is an urgent need to better understand the risks in the changes. Indeed, the catastrophic 2011 accident at the Fukushima Daiichi plant in Japan clearly shows the utmost importance of accurately measuring risk. In addition to performing efficient nuclear safety analyses, the techniques may also be applied to attack problems in civil and mechanical engineering, catastrophe modeling, and finance.This research combines theoretical and practical components, including designing new Monte Carlo algorithms, formally establishing their validity, and implementing them to analyze risks of existing nuclear plants. The core of the work will focus on variance-reduction techniques for efficient simulation for performing risk and safety analyses of nuclear facilities. Because current simulations can require enormous computational effort, applying variance reduction is essential for performing safety analyses. If successful, the new methods will profoundly reduce computation time, perhaps by orders of magnitude. To further account for the statistical error of the resulting estimators, the investigator also plans to produce asymptotically valid confidence intervals.