Research on Edge Intelligence-based Security Analysis Method for Power Operation System

Overall architecture



260
At this stage, the training of deep learning is placed in the cloud center, and the training data is on the edge. This mode is not suitable for all deep learning application scenarios, especially for some applications that require local information and continuous iterative training. Mass data transmission needs to occupy the resources of the communication channel, which will bring not only great network resource consumption but also challenging to ensure the reliability of information trans­mission. Besides, the part of the data on the edge side involves the privacy of end-users in edge nodes, and uploading all data to the cloud center is not a practical approach. Therefore, the edge computing nodes with stable computing resources should be regarded as multiple training centers, collecting information locally and performing data pre-processing and model training. This training method needs to combine edge-oriented cloud- edge collaboration and edge-federal training collaboration. The overall architecture diagram is shown in Figure 1, which mainly has the following steps: 1) The cloud center sends the preliminary training deep learning model completely to an edge node, which can be called an aggregation server (AS). 2) Edge nodes participate in As model training and use their local data to train local models. 3) The edge computing node sends the updated local model to AS to obtain the updated global model. This model training method is based on the premise of protecting the data privacy and security of edge nodes. It




Fig. 2. Edge security detection model.


Fig. 3. Application framework of edge computing platform in power system.
reduces the communication pressure of the entire system and increases the reliability of model training

2.