摘要
【目的】随着以储能、光伏为代表的电力电子器件在微电网的占比不断增加,其低惯性和低阻尼特性给微电网的安全稳定运行带来了严峻挑战。【方法】针对微电网稳定性问题,提出了一种新的基于数据驱动的逆变器参数调节方法,该方法基于有限系统量测数据,优化逆变器的多种参数,实现微电网小干扰稳定性的快速提升。首先,构造基于特征值的离线优化问题,通过鱼鹰优化算法(osprey optimization algorithm,OOA)计算微电网在不同运行场景下的多个最优控制参数。其次,为了减少参数优化模型对全局节点数据的依赖,利用多标签特征选择算法对系统节点数据进行特征筛选。最后,将筛选后的节点数据作为输入变量,多个最优控制参数作为输出变量,基于北方苍鹰优化算法(northern goshawk optimization,NGO)和双向门控循环单元(bidirectional gated recurrent unit,BiGRU)训练并得到参数优化模型。【结果】实验结果表明,所设计的参数优化模型可以基于在线量测数据快速调节控制器参数来提升微电网稳定性;所设计的深度学习算法比传统的神经网络在训练参数优化模型上具有更高的精确度;所得到的参数优化模型在参数优化上具有更快的计算速度。【结论】所提方法仅需局部的系统量测数据,通过在线动态优化逆变器的多个控制参数,便能实现微电网小干扰稳定性的快速提升。
[Objective]With the increasing penetration of power electronic devices,such as energy storage and photovoltaics,in microgrids,their low inertia and low damping characteristics pose challenges to the stable operation of microgrids(MGs).To enhance the stability of inverter-based MGs,this study introduces a novel data-driven method for the coordinated and rapid local adjustment of inverter multicontrol parameters.[Methods]An offline eigenvalue-based optimization problem was formulated to compute the optimal multicontrol parameters using the osprey optimization algorithm(OOA)under various operating conditions.Subsequently,to minimize the reliance on global system information,a multilabel feature selection algorithm is employed to identify the most relevant local measurements that influence the adjustment of each control parameter.Finally,local measurements are treated as input variables and optimal control parameters as output variables.A novel deep learning algorithm based on northern goshawk optimization(NGO)and a bidirectional gated recurrent unit(BiGRU)is proposed to train the local parameter optimization model(LPOM)by learning the input–output mapping.[Results]The case study demonstrates that the designed LPOM can swiftly adjust controller parameters based on online measurement data,thereby enhancing microgrid stability.It also establishes that the proposed deep learning algorithm achieves higher accuracy in training the LPOM compared to traditional neural networks.The LPOM delivers faster computation speeds for parameter optimization.[Conclusions]The proposed method only requires local measurement data and rapidly enhances the small-signal stability of microgrids through online dynamic optimization of multiple inverter control parameters.
作者
庞凯
唐志远
高红均
刘友波
刘俊勇
PANG Kai;TANG Zhiyuan;GAO Hongjun;LIU Youbo;LIU Junyong(College of Electrical Engineering,Sichuan University,Chengdu 610065,China)
出处
《电力建设》
北大核心
2025年第8期67-77,共11页
Electric Power Construction
基金
国家自然科学基金项目(52207127)
中央高校基本科研业务费专项资金资助项目(YJ2021163)。
