关键词: 暂降源辨识, 集成学习, 梯度提升树, 概率分类器

Abstract: In this paper, a method based on gradient boosting decision tree for voltage sag sources probabilistic recognition is proposed. The residuals are continuously fitted through the training of multiple decision trees， and the sag sources are recognized according to the recognition probability. The effectiveness and accuracy of the proposed method are verified by using various types of sag source waveform obtained by simulation, and compared with the traditional support vector machine method. Compared with the support vector machine, this method has higher accuracy under the same sample number, and the various types of sag source identification probability information can better reflect the credibility of the model identification， which is helpful for decision makers to assist decision making.

Key words: recognition of voltage sag sources, integrated learning, gradient boosting decision tree, probability classifier