Sequential Detection and Prediction for Solar Situation Awareness in Power Networks

该项目的总体研究目标是开发统计工具，以帮助电网运营商提高太阳能发电机的态势感知能力。目标包括使用顺序数据从实时电力供应和消耗数据推断太阳能发电机，其位置和状态。我们采用了一种新颖的方法，利用太阳能发电机的开关将在电力供应和消费的不同时间序列中引入变化点，并通过检测变化点并从序列数据中估计其大小，提出了一个太阳能态势感知的统计框架。我们使用了一种自下而上的方法，这是一个自然适合电网的分层结构：我们在单元级别检测变化点，然后在网络级别上聚合它们，使用多维点过程模型，并考虑到观测的固有稀疏和低维性质。变化点检测将与多保真度模型相结合，实现发电的网络级预测。结果将在动态电网的模拟高保真时间序列和真实数据上进行验证。住宅和商业领域的太阳能装置一直在增加。然而，很难知道这些光伏电池板的确切数量、位置、装机容量和生产状况，特别是由于光伏电池板的表后安装数量不断增加，以及太阳能生产的内在随机性。不了解太阳能发电机组在电网中的状态，将对配电系统的稳定性和安全性构成重大挑战。准确了解配电系统中住宅光伏发电机组的数量、位置、容量和运行状态对配电系统的日常运行和规划以及最终对输电系统的运行和规划至关重要。目前，没有一种有效的方法可以实时地推断大型电网中太阳能发电机组的状态。拟议的研究将改变现状，并显著推进使用统计方法进行太阳能态势感知的最新技术。公用事业公司、行业监管机构和太阳能电池板营销人员将受益于为态势感知开发的算法。例如，获得特定社区太阳能光伏生产状况的详细信息，可以使当地公用事业公司更好地平衡该地区的电力供需，提高对最终用户的电力服务质量，并提高配电电网的可靠性和安全性。拟议的研究将与教育组成部分紧密结合，研究结果将作为开放源码软件提供给国家实验室，研究结果将通过会议和期刊出版物传播。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 overarching research objective of this project is to develop statistical tools to help power network operators in increasing the solar generator situational awareness. The goals include inferring the solar power generators, their locations, and status, from the real-time power supply and consumption data using sequential data. We take a novel approach leveraging the fact that the on and off of the solar generators will introduce change-points in the difference time series of the power supply and consumption, and propose a statistical framework for solar situational awareness by detecting change-points and estimating their magnitudes from sequential data. We use a bottom-up approach, which is a natural fit to the hierarchical structure of the power networks: we detect change-points at the unit level, and then aggregate them on a network level, using multi-dimensional point process model, as well as considering the inherent sparse and low-dimensional nature of the observations. The change-point detection will be combined with multi-fidelity models to achieve network level prediction of power generation. The results will be verified on simulated high-fidelity time-series and real-data over dynamic power networks.Solar power installations have been increasing in both the residential and commercial areas. However, it is very difficult to know the exact numbers, locations, installed capacities, and production status of these PV panels, especially due to the increasing number of behind-the-meter installations of PV panels and the intrinsic stochasticity in solar production. Not knowing the status of the solar generators in the network can pose a significant challenge to stability and security of power distribution and transmission systems. The precise knowledge of the number, location, capacity, and operational status of residential PV units within a distribution system will be very critical to the daily operation and planning of distribution system and eventually that of transmission systems. Currently, there is no effective way to infer in real-time the solar generators' status in a large-scale power network. The proposed research will change the landscape and significantly advance the state-of-the-art in using statistical methods for solar situational awareness. Utility companies, industry regulators and solar panel marketers are a few of the groups that will benefit from the algorithm developed for situational awareness. For example, having access to detailed information of the status of solar PV production in a given neighborhood can enable local utilities to better balance the area's power supply and demand, improve the quality of electricity service to end users, and increase the reliability and security of distribution power grids. The proposed research will be tightly integrated with education components, and the research results will be made available to national labs as open source software and research findings will be disseminated via conferences and journal publications.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.