| 引用本文: | 郑翔,王慧芳,姜宽,曹建伟,刘莹,陈永炜.机理与数据融合驱动的含IIDG配电网短路电流计算方法[J].电力自动化设备,2021,41(1): |
| ZHENG Xiang,WANG Huifang,JIANG Kuan,CAO Jianwei,LIU Ying,CHEN Yongwei.Mechanism and data-driven combined short circuit current calculation method for distribution network with IIDG[J].Electric Power Automation Equipment,2021,41(1): |
|
| 摘要: |
| 目前含逆变器型分布式电源(IIDG)的配电网短路电流计算主要采用物理建模方法，在IIDG高渗透情况下其在计算速度、准确性、通用性等方面存在不足。因此提出一种机理与数据融合驱动的适用于含IIDG和IIDG高渗透下配电网的短路电流计算方法。为与数据驱动建模的计算方法进行对比，在对含IIDG的配电网进行特征分析的基础上，提出了反映短路电流的2种特征组合方式：一种使用蕴含机理的特征，即将不接入IIDG时配电网的短路电流作为关键特征，另一种则使用配电网特征。通过运行MATLAB/Simulink上搭建的仿真模型自动积累样本集合，使用机器学习中的集成方法(包括随机森林、极限随机树、XGBoost、LightGBM)进行2种特征下的模型训练。在IEEE 34节点系统上验证了集成方法建模的可行性和有效性，同时对比了不同集成方法以及不同特征组合方式的计算误差，结果表明，各集成学习方法均能够准确地进行短路电流计算，机理与数据融合的驱动方法在机理未失效情况下，比单纯的数据驱动模型计算更准确。与物理建模方法的对比结果也验证了所提方法的准确性和快速性。 |
| 关键词: 逆变器型分布式电源 配电网 短路电流计算 特征分析 机器学习 集成方法 |
| DOI：10.16081/j.epae.202010018 |
| 分类号:TM713 |
| 基金项目:国网湖州供电公司科技项目(2019-HUZJTKJ-17) |
|
| Mechanism and data-driven combined short circuit current calculation method for distribution network with IIDG |
|
ZHENG Xiang1, WANG Huifang1, JIANG Kuan1, CAO Jianwei2, LIU Ying2, CHEN Yongwei2
|
|
1.College of Electrical Engineering, Zhejiang University, Hangzhou 310027, China;2.State Grid Huzhou Power Supply Company, Huzhou 313000, China
|
| Abstract: |
| The short circuit current in a distribution network with IIDGs(Inverter-Interfaced Distributed Gene-rators) is mainly calculated by using physical modeling at present, which has drawbacks in terms of calculation speed, accuracy, and versatility in the case of high penetration of IIDG. A mechanism and data-driven combined short circuit current calculation method for distribution network with IIDGs and high penetration of IIDGs is proposed. To compare with simple data-driven calculation method, based on the feature analysis of distribution network with IIDGs, two kinds of feature combinations reflecting the short circuit current are introduced. In one of them, mechanism is considered and the short circuit current when IIDGs are not connected into the system is used as the key feature, while the other only uses features of the distribution network. The data sets are accumulated by running the simulation model on MATLAB/Simulink, and the machine learning models are trained with two feature combinations using the ensemble methods, which include random forest, extremely randomized trees, XGBoost and LightGBM. The feasibility and effectiveness of models based on ensemble methods are verified on the IEEE 34-bus system, and the calculation results of different ensemble methods and different feature combinations are compared. The results show that the ensemble learning methods are capable of correctly predicting the short circuit current, and the mechanism and data-driven combined method outperforms the simple data-driven method if the corresponding mechanism is still available. In addition, the comparison with the physical modeling method validates the accuracy and rapidity of the proposed method. |
| Key words: inverter-interfaced distributed generator distribution network short circuit current calculation feature analysis machine learning ensemble methods |
