【摘要】 针对带标签(类别已知)的电压暂降历史样本数据有限且不易获得的情况,引入基于标签传播半监督学习的电压暂降源识别方法。首先从电压暂降信号中提取了五类暂降信号特征,建立了K-近邻图模型,并实现了图模型上的标签传播。分析了图模型参数k、α对标签传播结果的影响,同时与神经网络、最小二乘支持向量机等监督学习算法的识别结果进行了对比。仿真结果表明,在历史数据较少的情况下,标签传播算法比传统监督学习算法具有更高的识别准确率且实时性好。更多还原

【Abstract】 In view of the situation that data for the historical voltage sags sample is limited and difficult to obtain,an approach for voltage sag sources identification based on label propagation semi-supervised learning is introduced.Firstly,five classes of voltage sag features are extracted from voltage sag information,and K-nearest neighbors graph model is constructed,finally label propagation on the graph model is realized.Then,the influences of parameters k and α on identification accuracy and running time are analyzed.At the same time,the identification results obtained through the proposed approach are compared with the supervised learning algorithm of neural network and least squares support vector machine.The simulation results verify that compared with traditional supervised learning algorithm,the approach based on label propagation algorithm has higher identification accuracy and runs faster with few labeled samples.更多还原