Abstract: In the substation production operations,the manual supervision method is mainly used to detect the illegal behaviors that ignore the power rules and regulations.The manual supervision method has the disadvantages of large workload,low efficiency,poor real-time performance and high security risks, and it cannot achieve all-round effective management of the safety of operators at substation sites.Therefore,a "deep learning+edge computing" approach to substation safety management was proposed.First,a dataset of inspector behaviour was filmed at the substation site,and the deep learning algorithm model was trained and tested on a server computing platform.Then,the substation safety monitoring edge computing device was developed to meet the needs of substation safety management applications.Finally,the trained deep learning algorithm model was deployed on the edge computing device to achieve real-time accurate detection and early warning of construction violations by inspectors.The results show that the detection accuracy of the method can reach 93.80%,which can realize the real-time accurate detection and early warning of construction violations by inspectors in complex construction scenarios in substations.