Time series and graph neural network-based traffic anomaly warning for different service nodes

Aiming at the problem that artificial intelligence algorithms have difficulty in detecting abnormal traffic of service nodes in large-scale networks. Therefore, the study proposes an improved way of fusing time series based on graph sampling and aggregation algorithms, splicing the time series with multiple features of nodes as inputs and introducing time-aware weights to achieve dynamic aggregation among neighboring nodes. Simulation results show that the research model achieves 98.1% accuracy in 600 iterations, and the recall rate increases from 63.8% to 96.8%. In the throughput comparison, the research model achieves a processing capacity of 18.2 × 103 samples/second. The loss value decreases to 0.09 at the end of iteration, which is better than that of the comparison model, which is 0.31. And in the operational efficiency, its detection delay is 13.7 s. The above results demonstrate the practicality of the research model in different service node traffic detection, and provide theoretical foundations and practical bases for the application of the intelligent technology in service node traffic monitoring.