Embedded Impedance Measurement for DC Microgrid Towards PHM and Control
Embedded Impedance Measurement for DC Microgrid Towards PHM and Control
复制标题
面向 PHM 和控制的直流微电网嵌入式阻抗测量
DOI：
10.1109/pedes56012.2022.10080141
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
H. Krishnamoorthy
中科院分区：
文献类型：
--
作者：
Hussain Sayed;Bharat Bohara;H. Krishnamoorthy
包括微电网在内的新兴直流电网，包括用于连接到不同负载（无源和有源）的分布式能源（DER，如光伏、风能和电池）的多个转换单元。因此，未来直流电网的弹性和实时健康状况必须不断评估。本文提出了一种嵌入式预测健康监测（PHM）和控制方法与电网的功率转换单元，使渐进的串扰之间的转换器在多个节点，以评估电网的健康动态。这项研究推进了数字化工具的使用，以提供准确的在线电网健康监测的机器学习技术的辅助下，现场实施的工具，如现场可编程门阵列（FPGA）。通过测量电网各个节点上功率转换器端子的电网阻抗（幅度和相移），经过训练的机器学习模型有助于评估电网的健康指数并识别潜在的故障易发区。通过特定转换器在宽频率范围内注入小扰动信号（即，0.5至10 kHz），以确定阻抗和相移特性。利用Matlab/Simulink工具对小型电网进行仿真，采集不同条件下的阻抗相关数据，进行机器学习训练。相关的结果验证了所提出的概念的好处。
Emrging DC grids, including microgrids, incorpo-rate multiple conversion units for distributed energy resources (DERs such as photovoltaics, wind, and batteries) connected to different loads (passive and active). Consequently, future DC grids' resilience and real-time health must be continually assessed. This paper presents an embedded Prognostic Health Monitoring (PHM) and control approach associated with the grid power conversion units, which enables progressive crosstalk between the converters at multiple nodes to evaluate the grid's health dynamically. This research advances the use of digital tools to provide accurate online grid health monitoring assisted by machine learning techniques, implemented in situ on tools such as Field-Programmable Gate Arrays (FPGAs). By measuring the grid impedance (magnitude and phase shift) at the terminals of the power converters at the grid's various nodes, the trained machine learning model helps assess the grid's health index and identify the potential fault-prone zones. A small perturbation sig-nal is injected through specific converters over a wide frequency range (i.e., 0.5 to 10 kHz) to determine the impedance and phase-shift characteristics. This paper uses Matlab/Simulink tools to simulate a small grid network to collect impedance-related data under different conditions for machine learning training. The associated results validate the benefits of the proposed concepts.