EAGER: Real-Time: Collaborative Research: Unified Theory of Model-based and Data-driven Real-time Optimization and Control for Uncertain Networked Systems

该项目寻求找到一个通用的决策框架，无缝集成离线数据和计算、实时数据和计算、学习和概率预测决策。它为不确定网络系统基于模型和数据驱动的实时优化与控制提供了统一的理论。集成强化学习是集成实时数据驱动方法、基于模型的方法和物理约束的关键。我们将探索整体强化学习的结构，以准确地研究在具有多个嵌套学习回路的架构中如何以及在何处使用深度学习神经网络。提出了一种概率时空情景数据驱动框架，用于不确定条件下网络化工程系统的多尺度顺序控制。开发的算法和工具将用于为直流配电网络中的电力电子换流器塑造最优功率分布，并帮助减轻间歇性电源、不确定负载需求或故障的不利影响。该项目代表着从现有的大数据和决策研究的根本转变，转向为规模不断增长和时间关键的任务要求的系统开发不确定结构下的自主决策。开发的算法和工具可以扩展到其他智能和互联领域，例如空中交通管理、网络交通排和传感器网络。预计到2020年，美国微电网的装机容量将达到4.3千兆瓦。直流配电网络是交流配电网络的新兴替代品，对于可再生能源和电气化运输车队的可扩展整合至关重要。研究成果将被移植到强化学习、最优控制、网络控制系统、数据驱动分析和决策以及电力电子系统等主题。该项目协同了德克萨斯大学阿灵顿分校(UTA)和德克萨斯A&A；M-Corpus Christi(TAMUCC)之间的研究活动，这两所大学都是HBCU/MI西语裔服务机构，涉及电气工程和计算机科学背景的学生。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The project seeks to find a common decision-making framework that seamlessly integrates offline data and computing, real-time data and computing, learning, and probabilistic predictive decision. It provides a unified theory of model-based and data-driven real-time optimization and control for uncertain networked systems. Integral Reinforcement Learning holds the key to integrating real-time data-driven methods, model-based methods, and physical constraints. The structure of Integral Reinforcement Learning will be explored to investigate exactly how and where to use Deep Learning neural networks in architectures that have multiple nested learning loops. A probabilistic spatiotemporal scenario data-driven framework will then be developed for multi-scale sequential control of networked engineering systems under uncertainty. The algorithms and tools developed will be used to sculpt optimal power profiles for power electronics converters in a DC distribution network and help mitigate the adverse effects of intermittent sources, uncertain load demand, or faults. The project represents a radical departure from the exiting big data and decision-making research, toward developing autonomous decision-making under uncertainty constructs for systems of growing scales and time critical mission requirements. Algorithms and tools developed can be extended to other smart and connected domains, e.g., air traffic management, networked traffic platoons, and sensor networks. US microgrid capacity is expected to reach 4.3 GW by 2020. DC distribution networks are emerging alternatives to AC distribution ones, and are critical to the scalable integration of renewable energy resources and electrified transportation fleets. Research results will be ported into topics in reinforcement learning, optimal control, networked control systems, data-driven analysis and decision-making, and power electronics systems. This project synergizes research activities between University of Texas at Arlington (UTA) and Texas A&M-Corpus Christi (TAMUCC), both HBCU/MI Hispanic Serving Institutions, and involves students from Electrical Engineering and Computer Science backgrounds.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.