With the development of the Internet of Things (IoT), ensuring efficient and stable network operation has become increasingly important. Performing network traffic identification on edge servers can reduce transmission latency and bandwidth consumption of network traffic data while alleviating dependence on network links, making it an effective method for real-time network traffic identification. However, edge environments are often widely distributed and complex, and the decentralized network traffic management approach makes it difficult to support centralized management of network data on a larger scale and unified allocation of network resources. In contrast, cloud servers, which aggregate network traffic from various edges and have abundant computational resources, can handle large-scale network data and perform complex analysis and identification tasks. This makes them suitable for unified identification and control of network traffic across the entire network. However, cloud servers struggle to meet the real-time requirements of network traffic identification and also face the issue of unstable cloud-edge links. To address these issues, this paper designs a cloud network traffic identification model updating algorithm based on multi-teacher reverse distillation. This algorithm utilizes the output information from multiple edge models to quickly update the cloud model. Additionally, to address the issue of abnormal edge network traffic data aggregation caused by unstable cloud-edge links, this paper designs an edge feature correction algorithm suitable for cloud-edge collaborative scenarios, which can identify and correct abnormal edge data at the cloud level, ensuring the effective update of the cloud network traffic identification model.