电工技术学报2024，Vol.39Issue(8)：2281-2293,2324,14.DOI:10.19595/j.cnki.1000-6753.tces.231393
基于机器学习的带被动阻尼直流微电网系统的稳定性检测
Stability Detection of DC Microgrid Systems with Passive Damping Based on Machine Learning
摘要
Abstract
In DC microgrids,the negative impedance characteristics of constant power loads(CPL)can reduce system damping,leading to system collapse.To enhance the stability of DC microgrid systems,researchers propose various active damping addition methods,including virtual active resistance compensation,synchronous buck circuit,and output current feedforward,which require complex control algorithms and signal processing.In addition,these methods may reduce load performance and affect the system's dynamic response.Compared with active dampers,passive damping only requires simple passive components without complex control algorithms and signal processing technology.Accordingly,the implementation is simple,and the cost is low.To analyze the stability of the DC microgrid after adding dampers,traditional methods based on the mechanism model ignore the structure influence,and the model error gradually increases with the increase of system dimension and complexity.Thus,this paper proposes a machine learning-based method to detect the stability of DC microgrids with passive damping.This method can effectively improve the error problem of the mechanism model,comprehensively consider the relationship between local and global stability,and realize the rapid and accurate online detection of the passive damped DC microgrid stable state. Firstly,this paper establishes an impedance model for the DC microgrid system with passive damping,using the Routh-Hurwitz criterion to determine the parameters that influence system stability.Secondly,simulation scenarios are conducted based on the selected system parameters to obtain sample data for training the machine learning algorithm.A stability detection model based on the light gradient boosting machine(LGBM)is introduced for DC microgrids.The influence of the selected parameters on the LGBM prediction and the stability of the DC microgrid system is further explored using Shapley additive explanations(SHAP).Finally,the effectiveness and superiority of the proposed method are validated through simulations and hardware-in-the-loop experiments. The results show that the CPL critical value is increased by 31 kW,15 kW,and 22 kW,respectively,when RC parallel,RL parallel,and RLC series dampers are added to the single energy storage DC microgrid.The RL series damper only improves the CPL critical value by 3kW.After establishing six simulation scenarios with RC parallel,RL parallel,and RLC series dampers,the proposed LGBM achieves prediction accuracy of more than 96.5%and a false alarm rate of less than 2.5%when there is uncertainty in source and load.Compared with traditional mathematical,machine learning,and deep learning methods,the proposed method has advantages in prediction accuracy and computational speed.Among the selected system parameter features,CPL,line inductance,and damper inductance are negatively correlated with the stability of the DC microgrid.In contrast,line capacitance and damper capacitance are positively correlated with the stability of DC microgrids. The following conclusions can be drawn.(1)RC parallel dampers,RL parallel dampers,and RLC series dampers significantly enhance the stability of the DC microgrid system with CPL.However,the RL series damper does not show an ideal improvement effect.(2)The proposed LGBM model consistently achieves the highest prediction accuracy across multiple scenarios and reduces computational time.(3)The SHAP method effectively explains the impact of each input feature on the stability of the LGBM detection DC microgrid,providing interpretability to the LGBM model.关键词
直流微电网/稳定性检测/被动阻尼/轻量型梯度提升机/沙普利加解释Key words
DC microgrid/stability detection/passive damping/light gradient boosting machine/Shapley additive explanations分类
信息技术与安全科学引用本文复制引用
刘笑,杨建,李力,董密,宋冬然..基于机器学习的带被动阻尼直流微电网系统的稳定性检测[J].电工技术学报,2024,39(8):2281-2293,2324,14.基金项目
国家自然科学基金(52177204)和湖南省自然科学基金(2021JJ30875)资助项目. （52177204）
