Abstract:To address indicator heterogeneity, confusion between source segments and affected segments, and delayed threshold-based detection during multi-segment degradation in 5G private networks, a degradation source detection method based on time-lag propagation of segment-level margin loss is proposed. The method converts observations from the radio access, transport, core network user-plane, and MEC/application segments into pressure values, and constructs segment-level service quality margins by combining average pressure and short-board pressure. Taking the steady-state margin as the reference, margin loss is calculated, propagation strength is obtained through time-lag correlation search, and the degradation source is identified by combining topological relationships with dominant bottleneck scores. Disturbance injection experiments show that the proposed method achieves an early-warning lead time of 3.57 s, a detection accuracy of 85.4% under dual-factor coupled disturbances, and a propagation direction determination accuracy of 87.6%. Under compound disturbances, the P95 end-to-end latency is reduced to 7.43 ms and the packet loss rate is reduced to 0.07%. The average computation time per cycle is 43.1 ms, satisfying the 100 ms requirement for online detection and correction.
