Collaborative Research: Information Geometry for Model Verification in Energy Systems with Renewables

新兴的通信和计算能力有可能深刻地改变和改善电力系统等基础设施。近年来，现代电力系统的结构和组成发生了重大变化。其中包括燃气发电厂和热电联产设施等新能源，以及通过电力电子转换器连接并通过当地通信网络严格控制的新负荷。新源和负载的可编程特性提供了新的功能，但同时需要频繁重复的模型验证。能源工程师首选的模型通常是由组件和子系统的物理特性驱动的。这些模型在参数方面通常是非线性的。然而，从测量中可靠地识别参数是一个具有挑战性的问题，对于非线性模型来说，这在很大程度上是没有解决的。该项目旨在部署新的模型验证工具，将深厚的数学基础（微分几何和信息论）与现代计算算法相结合。这个项目将对其他使用类似模型的工程分支产生直接影响。在能源系统中，该项目通过实现未来电力市场的更精确操作以及实际电厂和客户现场的控制，有可能带来经济、环境和弹性效益。该项目建立在微分几何计算进步的基础上，为系统识别和能源系统模型简化中经常遇到的挑战提供了新的全局表征。这种方法的前提是，具有许多参数的模型是从参数空间到数据或预测空间的映射。处理复杂系统模型的一个关键困难是参数和数据空间之间映射的高度各向异性，这意味着参数空间的微小变化可能导致测量（数据）空间的巨大变化，而参数的其他变化可能导致模型行为的不可察觉的变化。该项目将使用日常操作的事件记录（例如，从线路切换和负载变化后的相量测量单元）来激发新的模型验证和选择算法。长期愿景是开发非线性参数化能源组件和系统的全局和半全局识别程序，建立相量测量单元传感器的性能限制，开发新的模型简化程序，并为大规模能源系统的识别奠定基础。具体目标包括：1)风能和太阳能发电厂的参数识别，包括更详细的流形图；2)常规电源（同步发电机）和负载参数辨识；3)动态研究中常用模型的再参数化和约简。模拟将使用工业标准和定制软件以及硬件实验记录来量化进展。预期结果将与微电网、虚拟实体（虚拟公用事业、能源中心）相关，这些实体通常被认为是智能电网长期发展的关键，而未来的电力市场可能会在更短的时间尺度上运行，因此取决于系统动力学的模型保真度。

Emerging communication and computation capabilities have the potential to profoundly change and improve infrastructures such as electric power systems. The architecture and composition of modern power systems have been undergoing significant changes recently. These include new sources, such as gas-fired plants and co-generation facilities, and from new loads connected through power electronic converters and tightly controlled through local communication networks. The programmable nature of new sources and loads offers new capabilities, but at the same time necessitates frequently repeated model verification. Models preferred by energy engineers are often motivated by the physical properties of components and sub-systems. These models are typically nonlinear in terms of parameters. However, reliable identification of parameters from measurements is a challenging problem that is largely unsolved for the case of nonlinear models. This project aims to deploy new model verification tools that combine profound mathematical foundations (differential geometry and information theory) with modern computational algorithms. This project will have direct implications on other branches of engineering that use similar types of models. Within energy systems, this project has the potential to result in economic, environmental, and resilience benefits by enabling more precise operation of future electricity markets and control in actual power plants and customer sites.This project builds on computational advances in differential geometry, and offers a new, global characterization of challenges frequently encountered in system identification and model reduction of energy systems. The premise of this approach is that a model with many parameters is a mapping from a parameter space into a data or prediction space. A key difficulty in dealing with models of complex systems is the highly anisotropic nature of the mapping between the parameters and data spaces, meaning that small variations in parameter space may lead to dramatic changes in the measurement (data) space while other variations in parameters can lead to no discernable change in the in the model behavior. This project will use event recordings from daily operation (e.g., from phasor measurement units following line switchings and load variations) to motivate new model validation and selection algorithms. The long-term vision is to develop global and semi-global identification procedures for nonlinearly-parametrized energy components and systems, to establish limits of performance with phasor measurement unit sensors, to develop novel model reduction procedures, and to lay the groundwork for identification of large-scale energy systems. Specific goals include: 1) parameter identification for wind and solar plants, including more detailed manifold maps; 2) parameter identification for conventional sources (synchronous generators) and loads; and 3) re-parametrization and reduction for models that are typically employed in dynamic studies. Simulations will use industry-standard and custom software and recordings of hardware experiments to quantify progress. Anticipated results will be relevant for microgrids, virtual entities (virtual utilities, energy hubs) that are often considered essential in the long-term evolution of smart grids, and future electricity markets that will likely operate on shorter time-scales and thus depend on model fidelity of system dynamics.