基于特征迁移和深度学习的配电网故障定位

故障诊断对电力系统的稳定运行至关重要。当配电网的拓扑结构发生较大变化时，难以获取大量带有标签的暂态数据，导致传统的故障预测模型精度难以提高。针对此问题，提出一种将特征迁移和深度学习相结合的配电网故障诊断新方法。首先，采集配电网不同线路的零序电流构造故障特征集；其次，引入加权半监督迁移成分分析方法(semi supervised migration component analysis，SSTCA)，利用混合核函数将不同拓扑结构下的特征样本映射到同一特征空间中，缩小数据间的分布差异性；最后，将映射后的源域样本输入到卷积神经网络中进行分类训练，并测试映射后的目标域样本。通过Simulink仿真表明，在改变配电网拓扑结构的新场景下，文中所提的特征迁移方法与其它方法相比，对目标域故障定位精度最高且达到98%以上。

Fault location in distribution networks based on feature transfer and deep learning

Fault diagnosis is crucial to the stable operation of power systems. When the topology of the distribution network changes significantly, it is difficult to obtain a large amount of transient data with labels, which makes it difficult to improve the accuracy of traditional fault prediction models. To address this problem, a new method of distribution network fault diagnosis that combines feature transfer and deep learning is proposed. First, the zero-sequence currents of different lines in the distribution network are collected to construct fault feature sets; second, a weighted semi-supervised transfer component analysis (SSTCA) method is introduced to map the feature samples under different topologies into the same feature space using a hybrid kernel function to reduce the Finally, the mapped source domain samples are input to the convolutional neural network for classification training, and the mapped target domain samples are tested. The simulation shows that the proposed feature transfer method has the highest target domain fault recognition accuracy of more than 98% compared with other methods under the new scenario of changing the distribution network topology.