AMPS: Online and Model-Free Optimization of Power and Energy Systems

由于基于可再生能源的发电和分布式能源(DER)的激增，电力系统正在经历着快速的变化。当前最先进的网络监视和控制方法需要适应这种新兴的设置，尤其是它们必须能够应对更多的可控点和更快的网络级动态，这些动态可以在不到一秒的时间内发生变化。该项目提出了一个计划，以设计新的算法，运行速度更快，从而能够在快速的时间尺度上优化电网，并能够处理模型中的不确定性和误差。这些算法将适用于能源以外的其他关键基础设施，包括交通和工业自动化。对电力和能源部门的直接影响，如交流电网和风力发电场的运营，将是更有效地利用电力和更可靠地执行电压和电流限制。该项目的间接影响是通过本科生参与、研究生指导和课程开发、针对初中和高中科学、技术、工程和数学(STEM)夏令营的外展活动、广泛的传播活动和工业外展来培养新一代学生。现有的方法不足以在高度动态的环境中运行的复杂电力系统中进行数据处理和决策，因为以下缺点：i)与监测和控制任务相关的优化问题的批量解决方法可能无法在时间尺度上提供与分布级能源资源和不可控资产的可变性相匹配的解决方案；Ii)基于模型的优化方法需要准确的网络知识；以及iii)为批量/静态解决方案设计的算法的幼稚在线实现可能不能提供最优性和收敛保证。本项目中拟议的工作将大大扩展加速方法的理论和应用--成功地应用于批量优化--到网络和联网系统的时变优化，其中成本、约束和问题输入在算法步骤的执行过程中发生变化。该项目进一步研究了在线无导数策略的发展，以处理成本函数的未知梯度，这些策略适用于网络模型未知或难以准确估计的情况。主要努力致力于具有挑战性的优化设置，这些优化设置涉及时变目标中的约束和不可微项。拟议的努力侧重于严格的收敛分析；目的是根据算法执行期间跟踪最优解的误差(而不是渐近界)得出结果，并在时变设置中实现线性收敛速度。在电力系统领域，拟议的研究回顾了在动态运行条件下的经典监测和优化任务--状态估计、交流最优潮流和风力发电场的偏航控制，仅举几例；拟议的理论方法使这些任务能够实时执行。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。

Power systems are undergoing a rapid change due to the proliferation of renewable-based generation and distributed energy resources (DERs). Current state-of-the-art methods for network monitoring and control need to adapt this emerging setting, and in particular they must be able to cope with an increased number of controllable points and faster network-level dynamics that can change in less than a second. This project advances a program to design new algorithms that will run faster, and therefore be able to optimize the power grid at fast time scales, and that will also be able to handle uncertainty and errors in the model. The algorithms will apply to other critical infrastructures beyond energy, including transportation and industrial automation. The direct impact to the power and energy sector, such as the AC power grid and the operation of wind farms, will be a more efficient use of power and more reliable enforcement of voltage and current constraints. An indirect impact of the project is the training of a new generation of students through undergraduate student involvement, graduate student mentoring and curriculum development, outreach activities targeted at middle and high school Science, Technology, Engineering and Mathematics (STEM) camps, broad dissemination activities, and industrial outreach.Existing approaches are not adequate for data-processing and decision-making in complex power systems operating in highly dynamic environments because of the following drawbacks: i) batch solution approaches for optimization problems associated with monitoring and control tasks may fail to provide solutions at a time scale that match variability of distribution-level energy resources and non-controllable assets; ii) model-based optimization approaches require accurate knowledge of the network; and, iii) a naive online implementation of algorithms that are designed for a batch/static solution may not provide optimality and convergence guarantees. The proposed work in this project will significantly extend theory and application of accelerated methods -- successfully applied in batch optimization -- to time-varying optimization of networks and networked systems where costs, constraints and problem inputs evolve during the execution of the algorithmic steps. The project further investigates the development of online derivative-free strategies to deal with unknown gradients of the cost function, which apply when the network model is unknown or difficult to estimate accurately. Major efforts are devoted to the challenging optimization settings that involve constraints and non-differentiable terms in the time-varying objective. The proposed efforts focus on rigorous convergence analysis; the aim is to derive results in terms of error in the tracking of an optimal solution during the execution of the algorithms (rather than asymptotic bounds), and enable linear convergence rates in time-varying settings. Within the power systems area, the proposed research revisits classical monitoring and optimization tasks -- state estimation, AC optimal power flow, and yaw control in wind farms just to mention few -- under dynamic operating conditions; the proposed theoretical approach enables a real-time execution of these tasks.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.