Simulated Annealing and Neurla Networks for Large-Scale Optimization Problems

用于大规模优化问题的模拟退火和神经网络

• 批准号: 8814035
• 负责人: Michael Shanblatt
• 金额: $ 12.33万
• 依托单位: Michigan State University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1988
• 资助国家: 美国
• 起止时间: 1988-09-01 至 1991-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=8814035&HistoricalAwards=false
• 关键词: Simulated; Annealing; Neurla; Networks; Large

Control strategies for many large-scale engineering applications are often formulated as constrained optimization problems as exemplified by the optimal power flow (OPF) formulation for power system security. At present, only simplistic models of the OPF, using d.c. load flows, can be solved in a time-frame fast enough for real-time control. When reactive power and voltage along with more complex constraints are considered, solving the OPF problem becomes too time consuming. The goal of this project is to enhance the computational throughput of OPF and similar problems by applying two new approaches to constrained optimization which can how be implemented efficiently using nascent ASIC (application-specific integrated circuit) technology. These new optimization approaches are the simulated annealing algorithm (Astochastic state-space search algorithm) and the computational concept of neural networks (massively parallel, redundant, computational structures which are very conducive to VLSI implementation.) The project objectives are to first find a strategy for mapping the OPF and similar problems into the energy and constraint functions for simulated annealing and neural networks. Then, simulations for extracting the performance and cost benefits of these approaches will be developed. Finally, implementation of these systems in dedicated hardware using the latest ASIC technology will be assessed.

在许多大规模工程应用中，控制策略常常被描述为约束优化问题，例如电力系统安全的最优潮流(OPF)公式。目前，OPF的模型只有简单化，采用的是DC。潮流，可以在一个时间范围内以足够快的速度进行实时控制。当考虑无功和电压以及更复杂的约束时，求解最优潮流问题变得非常耗时。这个项目的目标是通过应用两种新的约束优化方法来提高OPF和类似问题的计算吞吐量，这两种方法可以使用新兴的专用集成电路(ASIC)技术有效地实现。这些新的优化方法是模拟退火法(随机状态空间搜索算法)和神经网络的计算概念(大规模并行、冗余的计算结构，非常有利于VLSI实现)。该项目的目标是首先找到一种策略，将最优潮流和类似问题映射到模拟退火法和神经网络的能量和约束函数中。然后，将开发用于提取这些方法的性能和成本效益的模拟。最后，将评估这些系统在使用最新ASIC技术的专用硬件中的实施情况。