ASTEA - Automated Self-Healing and Traffic Engineering Approach to Regulate Seamless Traffic in Emergency Communication Systems

Abstract

Emergency communication systems require persistent high-reliability connectivity, even if public infrastructure is unavailable, overloaded or damaged. Existing SD-WAN deployments, however, tend to respond to congestion after it occurs, rather than proactively avoiding it. In this paper, we propose ASTEA (Automated Self-Healing and Traffic Engineering Approach), a framework that leverages Software-Defined Wide Area Network (SD-WAN) capabilities and machine learning to proactively detect, classify, and mitigate network congestion in emergency situations. Real telemetry—interface statistics, BFD session health, tunnel statistics, and device counters—were collected from a live Cisco DevNet sandbox spanning four WAN edge devices via the vManage REST API and engineered into latency, packet loss, jitter, and a composite congestion score. An LSTM network trained on five-timestep sequences predicted congestion scores, benchmarked against a Random Forest regressor, achieving a 40% improvement in prediction accuracy and a training loss of 0.0775 by epoch 10. These predictions powered a rule-based self-healing engine that could dynamically select between primary, low-latency QoS or backup failover paths. The results show the feasibility of autonomous, policy-driven traffic management for mission-critical emergency communications networks over SD-WAN fabrics.