【作者】 赵亮；

【导师】 刘友波；

【作者基本信息】 四川大学 ， 电气工程（专业学位）， 2021， 硕士

【摘要】 随着配电网自动化的发展和配电网量测体系的完善,越来越多的配电网实时运行数据的潜在价值有待进一步开发。但由于配电网监测信息缺失或不准确、配电网运行方式变换频繁等原因造成配电网实时拓扑信息不可靠,配电网拓扑结构不完全可观。同时配电网侧运行交互方式愈发多样化,配电网的运行方式和物理特性变得更加复杂,传统的配电网拓扑可观性分析方法难以适应当前主动配电网的运行要求。因此研究如何通过实时量测数据实现配电网拓扑完全可观并且基于配电网拓扑可观性分析对量测装置布点进行优化具有重要研究意义。本文提出了基于机器学习的配电网拓扑在线辨识和量测布点优化方法。针对配电网拓扑辨识问题,提出了一种鲁棒的PCA-DBN耦合模型用于主动配电网运行拓扑在线识别。首先采用主成分分析法(Principal Component Analysis,PCA)对配电网节点电压幅值数据进行去噪和降维处理,同时进行初步的鲁棒特征选择,这显著降低了深度信念网络((Deep Belief Network,DBN)训练的复杂性的同时又不损失拓扑识别的准确性。然后利用DBN映射配电网节点电压幅值数据特征与支路分断开关和联络开关的0/1状态之间的非线性关系,从而对配电网拓扑进行在线辨识。经实验验证,所提算法能够实现小型主动配电网和大规模配电网的拓扑在线识别。与随机森林(Random Forest,RF),多输出回归(Multiple Output Regression,MOR)以及传统DBN等机器学习方法相比,该方法可以实现更高的拓扑识别精度。另外也分析了该方法的鲁棒性、兼容性和准确性,结果表明所提拓扑识别方法对样本数据的噪声和样本数据缺失具有较强的鲁棒性,对配电网不同负荷类型和不同的DERs配置方案具有较强的兼容性。针对配电网量测布点优化问题,提出了基于深度学习和决策树模型驱动的配电网可观性量测布点优化研究方法。首先提出了基于决策树的配电网量测特征属性选择方法,通过决策树对配电网量测特征进行筛选并对节点特征重要度进行分析计算,然后基于所选择的重要节点特征对配电网拓扑可观性进行分析,最后确定配电网量测装置布点优化方案。经实验验证分析,在保证配电网拓扑可观性的前提下所提方法能够得出配电网最佳的量测装置布点方案。更多还原

【Abstract】 With the development of distribution network automation and the improvement of distribution network measurement systems,the potential value of distribution network real-time operation data remains to be developed.However,the real-time topology information of the distribution network is unreliable.The topology of the distribution network is not completely observable due to the lack or inaccuracy of the monitoring information and frequent changes in the operation of the distribution network.At the same time,the operation interaction of the distribution network is becoming more and more diversified,and the operation mode and physical characteristics of the distribution network have become more complicated.The traditional distribution network topology observability analysis method is difficult to adapt to the operating requirements of the active distribution network.Therefore,it is of great significance to study how to realize a completely observable distribution network topology through real-time measurement data.Based on the observability analysis of distribution network topology,how to optimize the layout of measurement devices is also a problem that needs to be studied.This paper presents a method for online topology identification of distribution network and optimization of measurement points based on machine learning.Aiming at the problem of distribution network topology identification,a robust principal component analysis coupled deep belief network(PCA-DBN)model is proposed for on-line topology identification of active distribution network.First,PCA is used to denoise and reduce dimensionality of the voltage amplitude data.At the same time,robust feature selection is performed preliminary,which significantly reduces the complexity of DBN training without losing the accuracy of topology identification.Then DBN is used to map the non-linear relationship between the voltage amplitude and the binary states of switchable connections.Experiments show that the proposed method can realize the online topology identification of small-scale distribution network and large-scale distribution network.Compared with random forest(RF),multiple output regression(MOR)and traditional DBN methods,the proposed method can achieve a much higher topology identification accuracy.In addition,the robustness,compatibility and accuracy of the method are also analyzed.It is shown that the proposed topology identification method has strong robustness to the measurement noise and missing data,and has strong compatibility with different load types and various penetration levels of DERs.Aiming at the optimization problem of distribution network measurement points,a research method based on deep learning and decision tree model is proposed for the optimization of distribution network observability measurement points.Firstly,decision tree is proposed to select measurement characteristics of distribution network.The measurement characteristics of the distribution network are screened by decision and the importance of node characteristics is analyzed.Then,the layout of distribution network measurement devices are optimized based on the distribution network topology observability analysis method.Through the experimental verification and analysis,the optimal layout scheme of measuring devices can be obtained on the premise of ensuring the topology observability of distribution network.更多还原