Data-Driven Optimal Voltage Regulation Using Input Convex Neural Networks
Data-Driven Optimal Voltage Regulation Using Input Convex Neural Networks
复制标题
DOI：
10.1016/j.epsr.2020.106741
复制
发表时间：
2020-12-01
影响因子：
3.9
通讯作者：
Zhang, Baosen
中科院分区：
文献类型：
--
作者：
Chen, Yize;Shi, Yuanyuan;Zhang, Baosen
需要快速的时间尺度电压调节来实现配电网络中可再生能源的高渗透率。一种很有前途的方法是控制逆变器的无功功率注入以维持电压。然而，现有的电压调节算法需要线路参数的确切知识，这是不知道的大多数配电系统，是难以推断的。在这项工作中，通过利用凸性的电压调节问题的结果，我们设计了一个输入凸神经网络来学习之间的映射的功率注入和电压偏差。通过使用智能电表数据，我们提出的数据驱动方法不仅准确地拟合系统行为，而且还提供了一种易于处理的最佳方法来找到无功功率注入。各种数值仿真证明了所提出的电压控制方案的有效性。
Fast time-scale voltage regulation is needed to enable high penetration of renewables in power distribution networks. A promising approach is to control the reactive power injections of inverters to maintain the voltages. However, existing voltage regulation algorithms require the exact knowledge of line parameters, which are not known for most distribution systems and are difficult to infer. In this work, by utilizing the convexity results of voltage regulation problem, we design an input convex neural network to learn the underlying mapping between the power injections and the voltage deviations. By using smart meter data, our proposed data-driven approach not only accurately fits the system behavior, but also provides a tractable and optimal way to find the reactive power injections. Various numerical simulations demonstrate the effectiveness of the proposed voltage control scheme.