With the continuous popularization of information technology in the power system, it is possible to propose protection schemes for smart microgrids. Taking typical line protection as an example, a deep learning protection method based on hollow convolution is proposed to ensure the feasibility of intelligent protection schemes. Firstly, the local position and remote current and voltage signals of the protection device are collected through network optical fibers, and then the data is standardized and processed. Then, through the hollow convolutional network module, signal features are extracted to complete data analysis and type judgment. Finally, real-time intelligent protection of the line is completed. Using PSCAD simulation software, the modeling of typical substation transmission line intervals was completed, and the effectiveness and accuracy of the proposed scheme were verified. Good discrimination results can be achieved under high sampling errors and data loss, with good fault tolerance performance.