With the development of the power system, the number of devices connected to the power communication network has been increasing, leading to heightened security risks. Traditional manual registration methods in power communication networks struggle to cope with the expanding scale of the power Internet of Things (IoT). To address this issue, a power communication network traffic classification method based on multi-modal competitive feature selection is proposed, reducing security risks through automated registration. This method first uses various feature selection techniques to assess feature importance and determines the optimal feature subset using a random forest. Based on this subset, multiple ensemble learning models are compared. Experimental results show that this method can efficiently and accurately classify and register connected devices. Compared to other methods, it achieves higher accuracy and shorter processing time, making it more suitable for the power communication network environment.