Development of System-Type Neural Network Architectures for Distributed Parameter Systems Using Algebraic Decomposition

使用代数分解开发分布式参数系统的系统型神经网络架构

• 批准号: 0501305
• 负责人: Kwang Lee
• 金额: $ 24万
• 依托单位: Pennsylvania State Univ University Park
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2005
• 资助国家: 美国
• 起止时间: 2005-09-01 至 2007-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0501305&HistoricalAwards=false
• 关键词: Development; System; Type; Neural; Network

Proposal Number: ECS-0501305Proposal Title: Development of System-Type Neural Network Architectures for Distributed Parameter Systems Using Algebraic Decomposition PI Name: Lee, KwangPI Institution: Penn State University Intellectual Merit: This project will investigate the use of symmetry properties over space in order to train neural networks to analyze large, complex systems distributed over space. To exploit spatial symmetry, the PI will use algebraic decomposition to obtain an analytic description of empirical data in a specific form, called the semigroup form, which involves the product of a coefficient vector and a basis set of vectors. Additionally, each component of the coefficient vector and each component of the basis set of vectors can be described individually, allowing each component to be modeled separately. A combination of RBF neural networks and time-lagged recurrent neural networks is used in implementing these principles.Broader Impacts: The project will use, as its initial testbed, the challenge of monitoring temperature distributions across space in a boiler furnace. More accurate monitoring of boilers may be useful not only in diagnostics and maintenance but also in control; more effective boiler control can improve efficiency somewhat and reduce pollution substantially. Testbeds involving jet engines and flexible structures will also be considered. Boilers, furnaces and jet engines account for a majority of major pollutants like NOx around the world. The project will also advance crossdisciplinary education at Penn State.

提案号：ecs -0501305提案题目：使用代数分解开发分布式参数系统的系统型神经网络架构PI名称：Lee， KwangPI机构：宾夕法尼亚州立大学智力优势：该项目将研究空间对称性的使用，以训练神经网络来分析分布在空间上的大型复杂系统。为了利用空间对称性，PI将使用代数分解来获得特定形式的经验数据的解析描述，称为半群形式，它涉及系数向量和向量基集的乘积。此外，系数向量的每个分量和向量基集的每个分量可以单独描述，从而允许对每个分量单独建模。RBF神经网络和时滞递归神经网络的结合用于实现这些原理。更广泛的影响：该项目将利用监测锅炉炉内空间温度分布的挑战作为其最初的试验台。对锅炉进行更精确的监测不仅可以用于诊断和维护，还可以用于控制；更有效的锅炉控制可以在一定程度上提高效率，并大大减少污染。还将考虑涉及喷气发动机和柔性结构的试验台。锅炉、熔炉和喷气发动机是世界上氮氧化物等主要污染物的主要来源。该项目还将推进宾夕法尼亚州立大学的跨学科教育。