Mathematical Sciences: Statistical Strategies for Complex Computer Models

数学科学：复杂计算机模型的统计策略

• 批准号: 9001726
• 负责人: Jerome Sacks
• 金额: $ 11万
• 依托单位: University of Illinois at Urbana-Champaign
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1990
• 资助国家: 美国
• 起止时间: 1990-07-01 至 1991-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9001726&HistoricalAwards=false
• 关键词: Mathematical; Sciences; Statistical; Strategies; Complex

This research provides surrogate statistical models for the usually deterministic output of complex computer models which cannot be directly explored in great detail because of their size and limitations on the number of runs. The models are based on a set of runs at selected inputs to aid in predicting the output at untried inputs, optimizing characteristics of the output, identifying important input factors and tuning the model to physical data. The strategies will be devised in the context of large numbers of input factors. The ingredients will include methods to select inputs (statistical experimental design) and to analyze the output data through the use of stochastic process models for deterministic outputs. Supercomputing power will be necessary to treat large numbers of factors and to extract the most out of an expensive-to-collect set of runs from computer models that themselves may run on supercomputers. A cross- disciplinary atmosphere will be maintained to stimulate the formulation of relevant problems with applicable solutions.

这项研究为复杂计算机模型的通常确定的输出提供了替代统计模型，由于其大小和运行次数的限制，不能直接详细地探索这些模型。这些模型基于选定输入的一组运行，以帮助预测未尝试输入的输出、优化输出的特性、识别重要的输入因素并将模型调整为物理数据。这些战略将在大量投入因素的背景下制定。这些内容将包括选择投入的方法(统计试验设计)和通过使用确定性产出的随机过程模型分析产出数据的方法。超级计算能力将是处理大量因素和最大限度地从从计算机模型收集昂贵的运行集中提取最大价值所必需的，这些计算机模型本身可能在超级计算机上运行。将保持跨学科的氛围，以促进相关问题的制定和适用的解决方案。