POWRE: Exploration of Applying Statistical Process Control Strategies to Simulated Annealing Optimization Techniques

POWRE：统计过程控制策略应用于模拟退火优化技术的探索

• 批准号: 9805972
• 负责人: Linda Schmidt
• 金额: $ 7.48万
• 依托单位: University of Maryland, College Park
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1998
• 资助国家: 美国
• 起止时间: 1998-08-01 至 2000-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9805972&HistoricalAwards=false
• 关键词: POWRE; Exploration; Applying; Statistical; Process

9805972 Schmidt This research will apply statistical process control (SPC) techniques, commonly used to monitor quality in manufacturing systems, to the control of simulated annealing optimization algorithms. Simulated annealing is a popular technique for complex problem optimization because it only requires calculations of the value of each potential solution. Other techniques calculate evaluation function derivatives or do matrix manipulations. Simulated annealing's drawback is that thousands or millions of solutions must be evaluated during the optimization process. Preliminary research supports the use of SPC methods to detect productive search. Productive search is defined as search resulting in a period of significant change in the average value of the most recent set of proposed solutions. If detectable, algorithms can be directed to run during only while the search is productive. This work is to develop a refined implementation of a prototype detection of productive search annealing algorithm and supporting foundational theory. The algorithm will be tested on selected problems from the Multidisciplinary Optimization Branch of NASA's Langley Research Center's test suite of engineering problems. This grant enables a more thorough investigation of this fledgling idea, allowing the researcher to bridge current work in design theory and designer assistance tools into operations research.

9805972施密特本研究将应用统计过程控制（SPC）技术，通常用于监控制造系统的质量，模拟退火优化算法的控制。 模拟退火是一种用于复杂问题优化的流行技术，因为它只需要计算每个潜在解决方案的值。 其他技术计算评估函数导数或进行矩阵操作。 模拟退火的缺点是在优化过程中必须评估数千或数百万个解决方案。 初步研究支持使用SPC方法来检测生产性搜索。生产性搜索被定义为导致最近一组建议解决方案的平均值发生显著变化的搜索。 如果可检测到，则可以指示算法仅在搜索有效时运行。 本文的工作是开发一个生产性搜索退火算法的原型检测的精细实现和支持的基础理论。 该算法将测试选定的问题，从多学科优化分支的美国宇航局的兰利研究中心的测试套件的工程问题。这笔赠款使这个羽翼未丰的想法进行更彻底的调查，使研究人员能够将当前的设计理论和设计师辅助工具的工作与运筹学联系起来。