基于深度-迁移学习的输电线路故障选相模型及其可迁移性研究

In order to improve the transferability of transmission line fault diagnosis models,the transmission lines are divided into source lines and target lines based on transfer learning theory,then a method based on deep-transfer learning for identifying transmission line fault types is proposed. The time series data during transmission line faults is generated by combining different fault conditions,and the input data samples of CNN(Convolutional Neural Network) are obtained by data preprocessing. Then the initial CNN is pre-trained by using the source domain data to obtain a pre-trained model of the source line fault type identification. Next,the maximum mean difference method is used to test the similarity of source and target lines,and the source domain pre-trained model to be migrated is screened out. The target domain data is used to fine-tune the migration training to obtain the final target domain fault diagnosis model. The simulative results show that by using the target domain data of 5 % of the source domain data to fine-tune the migration training of the pre-trained model,the target line fault diagnosis accuracy of the target domain model can reach more than 99 %.