基于XGBoost特征筛选和决策树多标签分类的配电网拓扑辨识方法

A Topology Identification Method for Distribution Networks Based on XGBoost Feature Selection and Decision Tree Multi-Label Classification

中文关键词:配电网  拓扑辨识  特征筛选  多标签分类

keywords:distribution network  topology identification  feature selection  multi-label classification

中文摘要:针对配电网中开关状态遥信丢失或延迟导致配电网拓扑不准的问题，通过引入遥测信息进行开关状态的补全与校正，提出了一种基于极限梯度提升算法（extreme gradient boosting，XGBoost）的特征筛选和决策树多标签分类的配电网拓扑辨识方法。利用XGBoost筛选出最相关、最具信息量的量测数据，在保证拓扑辨识准确度的同时大幅度减轻模型计算负担；以筛选的遥测特征和原始遥信开关状态数据为输入变量，以配电网开关状态为输出，构造了基于决策树的多标签分类模型，用于配电网拓扑辨识。在改进的IEEE 34节点配电系统中进行了算例测试。仿真结果表明：所提方法适用于多分布式电源接入的配电网拓扑辨识，具有很好的抗遥测噪声干扰能力；相比其他同类算法，所提方法的辨识准确性更高、计算耗时更少。

Abstract:For the inaccurate distribution network topology due to the loss or delay of switch state telemetry in distribution networks， this paper proposes a topology identification method for distribution networks based on XGBoost feature selection and decision tree multi-label classification by introducing telemetry information for the complementation and correction of switch states. First， XGBoost is used to filter out the most relevant and informative measurements to substantially reduce the computational burden of the model while ensuring the accuracy of topology identification. Second， a decision tree-based multi-label classification model is constructed using filtered telemetry features and raw telemetry switching state data as input variables and distribution network switching state as output for distribution network topology identification. Finally， case studies are investigated in the modified IEEE 34-node distribution system. The findings show that the proposed method is suitable for distribution network topology identification with multiple distributed generation and is highly resistant to telemetry noise interference. In contrast to other similar algorithms，the proposed method has higher identification accuracy and smaller computational time consumption.