AMPS: Compositional Data-Driven Modeling, Prediction and Control for Reconfigurable Renewable Energy Systems

现代电网正迅速向以分布式发电、插电式电动汽车、储能和需求响应资源为代表的可再生能源为主导的分布式、可重构系统发展。该项目的目标是开发计算工具，以解决在受异构干扰的分布式和可重构电力系统建模和控制方面出现的新挑战。这一目标将通过发展新的数学算法和理论来解决，这些算法和理论将部署在电力系统应用中，利用机器学习，动力系统和控制理论的基础知识。该项目将通过培养两名研究生和通过设计关于网络物理微电网和动力系统机器学习的课程开发，为NSF推进STEM的使命做出贡献。本项目旨在设计组成数据驱动的建模、预测和控制方法，以确保分布式和可重构的可再生能源主导电力系统的暂态稳定，该系统具有固有的非线性、高维、部分可观测和异构不确定性。该项目将阐明机器学习的进展，以开发可扩展和内聚的方法来解决系统运行的基本挑战。具体来说，主要研究者（pi）将(1)开发一种抗噪声组合双线性算子理论方法，以确定可重构可再生能源系统的瞬态动力学控制模型；(2)通过整合严格的统计闭包公式和物理拓扑感知数据驱动模型，设计部分观测系统的随机动力学模型；(3)将所建立的模型与最优控制算法相结合，以预测的方式提高分布式可重构系统的暂态稳定性，使其具有实时自主运行能力。pi预计这些成果将极大地丰富和扩展当前对大规模互联系统动态建模和控制的研究，并支持这些技术在下一代配电网应用中的发展。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。

The modern power grid is rapidly evolving towards a distributed and reconfigurable system dominated by renewable energy resources, represented by distributed generation, plug-in electric vehicles, energy storage, and demand-response resources. The goal of this project is to develop computational tools to address new challenges arising in modeling and control of the distributed and reconfigurable power systems subject to heterogeneous disturbances. This objective will be addressed by the development of new mathematical algorithms and theory that will be deployed in power system applications, leveraging the fundamental knowledge from machine learning, dynamical systems, and control theory. This project will contribute to the NSF mission of advancing STEM through the training of two graduate students and curricular development through the design of courses on the topics of cyber-physical microgrids and machine learning for dynamical systems. This project aims to devise compositional data-driven modeling, prediction, and control methods to ensure the transient stability of the distributed and reconfigurable renewable-energy-dominant power systems, which are inherently nonlinear, high dimensional, partially observed, and subject to heterogeneous uncertainties. This project will illuminate the machine learning advances for developing scalable and cohesive approaches to solve the fundamental challenge of in system’s operation. Specifically, the principal investigators (PIs) will (1) develop a noise-resilient compositional bilinear operator theoretic method to identify a control-amenable model for the transient dynamics of reconfigurable renewable energy systems; (2) devise a stochastic dynamics model for the partially-observed system by integrating a rigorous statistical closure formulation and a physics-informed topology-aware data-driven model; and (3) integrate the developed models with the optimal control algorithms to improve the transient stability of the distributed and reconfigurable system in a predictive manner towards a real-time autonomous operation capability. The PIs anticipate that these outcomes will substantially enrich and expand the current research on dynamic modeling and control of large-scale interconnected systems and support the development of these techniques for applications of the next-generation distribution grids.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.