Answers are grounded in this paper and cite their sources.
Flood-warning networks require authenticated message delivery under changing traffic loads, link availability, and impersonation risks. This study develops a risk-prioritized constrained Q-learning controller (RPC-Q) that integrates a noisy warning-risk observation, deadline-weighted queue selection, and joint route/authentication decisions. The controller uses event-based action-value learning with a deadline feasibility mask. A synthetic capacity-reservation model compares six controllers across nine traffic-load and attack-parameter combinations using ten independent training seeds and 900 held-out traces. RPC-Q achieves risk-weighted authenticated on-time delivery of 62.20% (95% bootstrap interval: 61.94–62.45%), compared with 41.52% for ordinary Q and 63.37% for a model-aware heuristic. Removing the risk input and deadline mask reduces delivery by 0.55 and 3.17 percentage points, respectively. RPC-Q accepts 11.24% of generated attacks, compared with 12.47% for the heuristic, and incurs a normalized resource cost of 1.82 units per delivered legitimate message. These results quantify the contributions of risk information and deadline constraints and the trade-off between authenticated delivery, attack acceptance, and resource consumption in the simulated network.
No takes yet. Share an insight, caveat, or question.
Niu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: