Disaggregating Load by Type from Distribution System Measurements in Real Time
Disaggregating Load by Type from Distribution System Measurements in Real Time
复制标题
从实时配电系统测量中按类型分解负载
DOI：
--
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
J. Mathieu
中科院分区：
文献类型：
--
作者：
Gregory S. Ledva;Zhe Du;L. Balzano;J. Mathieu
可以使用不同负载/发电机类型（例如，空调负荷、照明负荷、光伏发电）。这些信息可以从额外的设备级传感器和通信基础设施中收集。或者，可以使用现有的网络测量和底层系统的一些知识来推断该信息。这项工作应用了两种在线学习算法，动态镜像下降（DMD）和动态固定份额（DFS），以分离（或分解），在实时，馈线级主动需求测量成两个组成部分：（1）人口的住宅空调的需求和（2）由馈线服务的剩余负载的需求。在线学习算法包括使用历史建筑物级或设备级数据生成的底层负载类型的模型。我们开发的方法，将模型预测误差统计的算法，DMD和卡尔曼滤波之间的连接，适应算法的能量分解应用程序，目前的案例研究表明，该算法有效地执行分解。
An electricity distribution network’s efficiency and reliability can be improved using real-time knowledge of the total consumption/production of different load/generator types (e.g., air conditioning loads, lighting loads, photovoltaic generation) within the network. This information could be gathered from additional device-level sensors and communication infrastructure. Alternatively, this information can be inferred using existing network measurements and some knowledge of the underlying system. This work applies two online learning algorithms, dynamic mirror descent (DMD) and dynamic fixed share (DFS), to separate (or disaggregate), in real-time, feeder-level active demand measurements into two components: (1) the demand of a population of residential air conditioners and (2) the demand of the remaining loads served by the feeder. The online learning algorithms include models of the underlying load types, which are generated using historical building-level or device-level data. We develop methods to incorporate model prediction error statistics into the algorithms, develop connections between DMD and Kalman filtering, adapt the algorithms for the energy disaggregation application, and present case studies demonstrating that the algorithms perform disaggregation effectively.