ERI: Intelligent Modeling and Parameter Selection in Distributed Optimization for Power Networks

这项工程研究启动（ERI）赠款将通过资助研究解决优化大规模网络系统（如国家电力系统）的挑战，为促进国家繁荣、经济福利和国家安全做出贡献。由于可再生能源和电动汽车的日益普及，国家电力系统正在迅速发展。该奖项支持研究创新算法的发展，使多个代理能够有效地沟通和协作，解决复杂的网络优化问题，同时保护个人隐私。通过整合数学优化和工程，这个跨学科项目的结果将使学术和工业研究人员、数据科学家和政策制定者受益。教育和推广活动旨在激励STEM学生，特别是那些来自代表性不足群体的学生，通过暑期研究项目等举措，探索科学和工程领域的数据驱动方法。本研究创建了智能建模和参数选择方法，以解决分布式优化（DO）中的重要挑战，分布式优化（DO）是一种有前途的方法，适用于广泛的复杂网络系统，如快速发展的现代电网。本研究设计了创新的分割技术，以改善DO算法的缓慢收敛。这种方法不仅通过减少子问题的数量来简化DO的定制，而且还在子问题上施加了理想的结构。PI设计了自适应和基于学习的策略，使用经典参数选择技术和基于学习的方法（如物理信息深度学习）来解决DO算法参数选择的普遍挑战。该研究还将检查DO算法对数据不确定性和通信错误的鲁棒性。结果将通过开源优化软件包传播，促进这些创新算法在现实世界中的实施。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This Engineering Research Initiation (ERI) grant will contribute to the advancement of national prosperity, economic welfare, and national security by funding research that addresses the challenge of optimizing large-scale networked systems such as nation’s power system, which is rapidly evolving due to the growing prevalence of renewable energy sources and electric vehicles. This award supports research examining the development of innovative algorithms enabling multiple agents to communicate and collaborate effectively in solving complex network optimization problems while safeguarding individual privacy. By integrating mathematical optimization and engineering, the results of this cross-disciplinary project will benefit academic and industrial researchers, data scientists, and policymakers. The educational and outreach activities seek to inspire STEM students, especially those from underrepresented groups, to explore data-driven methodologies in science and engineering through initiatives like summer research programs.This research creates intelligent modeling and parameter selection methods to address important challenges in distributed optimization (DO), a promising approach for a broad class of complex networked systems such as the quickly evolving modern power networks. This research designs innovative partitioning techniques to improve the slow convergence of DO algorithms. This approach not only simplifies the customization of DO by reducing the number of sub-problems but also imposes desirable structures on the sub-problems. The PI devises adaptive and learning-based strategies to address the pervasive challenge of parameter selection for DO algorithms using both classical parameter selection techniques and learning-based methods, such as physics-informed deep learning. The research will also examine the robustness of DO algorithms to data uncertainty and communication errors. Results will be disseminated through open-source optimization software packages, facilitating the real-world implementation of these innovative algorithms.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.