考虑样本类别不平衡的电网故障事件智能识别方法

电网中不同设备的故障概率存在差异,影响智能诊断技术的准确性.为解决此问题,提出了一种基于代价敏感学习和模型自适应选择融合的电网故障事件智能识别方法.首先,利用Word2vec模型将预处理后的电网告警信息向量化,并搭建2个双向长短期记忆网络作为基础分类器;然后,设计代价敏感损失函数,将交叉熵损失函数与代价敏感损失函数分别应用于2个分类器中;最后,提出一种模型自适应选择融合法,融合上述分类器,得到故障事件识别结果.实际数据测试表明,所提方法能够有效降低故障事件识别中样本类别不平衡的影响.

Intelligent identification method of power grid fault events considering sample classification imbalance

In order to solve the problem that the difference of the failure probability of different equipments in the power grid affects the accuracy of fault intelligent diagnosis technology, an intelligent identification method of power grid fault events based on cost-sensitive learning and model adaptive selection fusion is proposed. Firstly, the Word2vec model is used to vectorize the pre-processed power grid alarm information, and two bidirectional long-short-term memory networks are established as basic classification models. Then, the cost-sensitive loss function is designed. The cross-entropy loss function and cost-sensitive loss function are respectively applied to the two classification models. Finally, a model adaptive selection fusion method is proposed to fuse the above classification models, so as to obtain the identification results of fault events. Actual data test shows that the proposed method can effectively reduce the impact of sample classification imbalance in the fault event identification.