Dynamic Prediction and Optimization Method of PRP Network Congestion Based on GAT Under Linux

Aiming at the congestion control delay caused by the static link strategy of the Parallel Redundancy Protocol (PRP) in the industrial redundant network, this paper proposes a two-stage dynamic optimization method based on Graph Attention Network (GAT). In the first stage, the graph attention mechanism is used to analyze the spatial dependencies of network nodes and combine the time series characteristics of historical traffic to achieve accurate prediction of link congestion probability. The second stage adopts a hierarchical optimization strategy based on the prediction results. When the primary link congestion probability exceeds a preset threshold, it triggers a priority switching mechanism and quickly switches to the redundant path. For low-risk links, traffic weight (70% for primary path and 30% for backup path) was dynamically allocated according to real-time bandwidth utilization to achieve load balancing. Experimental results show that compared with the traditional PRP static strategy, the proposed method reduces the average transmission delay by 28.5%, and reduces the burst traffic packet loss rate by 25%.