This project proposes an innovative edge computing-based traffic detection solution to enhance autonomous optical network operations. A lightweight plugin is deployed on home gateways to collect real-time flow-level data with minimal resource consumption. Through advanced AI models—including RNN and GNN—it identifies 65 service categories and 2,277 subtypes, quantifies key QoE metrics, and predicts anomalies before customer complaints arise. Integrated into the CSP’s AI-driven O&M framework, the system enables proactive ticket dispatching across home, access, and IP layers. Over 1.26 million tickets were automatically issued in one year, reducing user complaint rate by 14%. The project builds a standardized dataset, applies ITU P.1203-based video QoE models, and leverages edge-cloud hybrid computation for near real-time analysis.
The project offers a full-stack intelligent traffic sensing framework based on edge computing. Key innovations include: 1. Lightweight Gateway Plugin: Runs on home gateways with <10% resource usage. Captures per-flow metrics (loss, RTT, retransmissions) with no impact on user experience. 2. Advanced Service Classification: RNN + UA info enables accurate identification of 2,000+ apps and 65 business types. 3. Segmented RTT Algorithm: Differentiates between in-home and upstream network delay, enhancing fault localization precision. 4. ITU-Based QoE Modeling: ITU P.1203-compliant video QoE model runs at the edge, enabling precise video quality classification. 5. Cloud-Edge AI Loop: Combines edge data collection with cloud-based GPU analysis, allowing near real-time predictions. 6. Integration into 4PS O&M: AI auto-dispatch system issued 1.26M tickets/year, with proactive alerts reducing complaints by 14%. This architecture enables true self-intelligent operations: continuous flow detection → real-time insight → automatic action.