An improved deep residual network(ResNet)fault identification method for flexible direct current(FD)transmission lines is proposed to address the issues of poor anti-noise performance and insufficient performance with limited historical samples in existing deep learning methods.Firstly,the residual unit is enhanced by integrating attention mechanisms and regularization methods,and the model is optimized in terms of width and depth.Then,feature screening of the fault electrical quantities is conducted using recursive feature elimination(RFE)within cross-validation to define key features based on feature attributes.Finally,a simulation model of the Zhangbei-Beijing four-terminal±500 kV FD transmission network is established on the PSCAD simulation platform.The effectiveness of the proposed method under anti-noise interference and small sample size conditions is verified through simulations.
关键词
柔性直流输电系统/ResNet/注意力机制/特征筛选/故障辨识
Key words
flexible DC transmission system/ResNet/attention mechanism/feature selection/fault identification
维普
万方数据