Intelligent Multi-Objective Cluster-Head Selection with AoI-Awareness and Anomaly-Informed Suspicion for Tactical WSNs

Technical Summary: Intelligent Multi-Objective Cluster-Head Selection with AoI-Awareness and Anomaly-Informed Suspicion for Tactical WSNs

Problem Statement
Tactical Wireless Sensor Networks (WSNs) operate under stringent constraints where energy efficiency, information freshness, and security must be managed simultaneously. Existing clustering protocols typically optimize only one or two of these dimensions. Classical methods like LEACH focus on energy but ignore trust and information staleness; Reinforcement Learning (RL) approaches often prioritize adaptivity and routing efficiency without explicitly integrating Age of Information (AoI) or anomaly-driven suspicion; and fuzzy trust-based methods improve security but rely on fixed rules that lack adaptability and do not explicitly minimize information age. Consequently, a unified framework that jointly addresses energy conservation, real-time data freshness (AoI), and security (trust, suspicion, and attack status) in a single Cluster-Head (CH) selection process is lacking.

Methodology
The authors propose a multi-objective CH selection framework that integrates spatial clustering, temporal anomaly detection, and Q-learning. The system operates within a simulated tactical environment (100–300 nodes) subject to probabilistic attacks (DoS, False Data Injection, Selective Forwarding, and Sinkhole).

1. Spatial Organization: The network is partitioned into 10 spatial groups using K-Means clustering. This is executed once offline by the Base Station to reduce the computational burden on sensor nodes and to localize the CH selection search space.
2. Temporal Anomaly Detection: Each node maintains a sliding history window (6 rounds) of traffic rate, packet drop ratio, AoI, and residual energy. The system computes "temporal jump scores" to detect behavioral deviations. An "edge anomaly score" is derived from immediate and temporal components, accumulating a "risk memory" to penalize persistent minor anomalies.
3. Candidate Filtering: Before the learning phase, nodes are filtered based on minimum residual energy (0.30 J), minimum trust (0.35), and maximum suspicion (0.65). If no candidates meet these criteria, a strict fallback policy is applied to prevent compromised nodes from becoming CHs, or selection is suspended to avoid network poisoning.
4. Q-Learning Selection: A Q-learning agent selects the CH from the filtered candidates within each cluster. The state space is discretized (144 states per cluster) to ensure low memory overhead.
5. Reward Function: The core novelty lies in a multi-objective reward function that balances:
   • Positive factors: Residual energy and trust.
   • Negative factors (Penalties): Node AoI, cluster average AoI, cluster maximum AoI, suspicion score, active attack status, distance from the cluster centroid, and the number of times a node has previously served as CH.
   • Weights: Energy (9.0) and Node AoI (8.0) are weighted highest, followed by Trust and Attack Status (6.0 each).

Key Contributions

• Unified Framework: The paper introduces a single CH selection mechanism that jointly optimizes energy, AoI, trust, suspicion, attack status, spatial suitability, and fairness, addressing a gap where previous works treated these as separate objectives.
• Anomaly-Informed Suspicion: A lightweight, temporal anomaly detection mechanism is embedded into the CH selection process, using risk memory to penalize nodes with recurring suspicious behavior rather than relying solely on static thresholds.
• AoI-Centric Reward: Unlike standard RL approaches that focus on energy or routing, the proposed reward function explicitly penalizes information staleness (Node, Cluster-Average, and Cluster-Max AoI), ensuring that selected CHs deliver fresh data.
• Fair Experimental Evaluation: The study employs a unified simulation environment where all baselines (LEACH, an RL-based method, and a Fuzzy Trust method called RFGD) are subjected to identical network conditions, attack probabilities, and a shared external security evaluator. This prevents bias arising from different evaluation criteria.
• Robustness Validation: The method is tested across 20 random seeds and nine network configurations (varying node counts and simulation lengths), demonstrating consistent performance improvements.

Results
In the primary 100-node, 2500-round scenario, the proposed method achieved:

• Lowest Age of Information (AoI): A final AoI of 2.15, significantly outperforming LEACH (6.99), the RL baseline (18.43), and RFGD (206.78).
• Highest Quality Score: A composite score of 0.567, indicating a superior balance of freshness, security, and energy.
• Energy Efficiency: The lowest final average energy consumption rate (0.0012) among the compared methods.
• Statistical Significance: Paired Wilcoxon signed-rank tests confirmed that the improvements in AoI and energy consumption are statistically significant (p < 0.001) against all baselines.
• Trade-offs: The proposed method achieved lower selected-CH trust and higher suspicion compared to RFGD and the RL baseline. The authors note this is a deliberate trade-off to prioritize freshness and energy continuity, rather than a failure of the security mechanism. Security detection metrics (F1-score, accuracy) remained competitive but were not the primary optimization target.

Significance and Claims
The paper claims that its primary contribution is an AoI-driven integrated decision model that prevents the selection of CHs that are energy-rich but information-stale or suspicious. The authors emphasize that in tactical monitoring, a secure CH providing outdated data is operationally useless. By making AoI a central component of the CH selection reward, the framework ensures that the network remains not only alive and secure but also operationally relevant.

The study modestly acknowledges its limitations: it uses direct CH-to-sink transmission (unlike the multi-hop RL baseline), which affects delay comparisons; the attack model is synthetic; and the method is security-aware but not a standalone intrusion detection system. However, the authors assert that the framework successfully demonstrates that tactical WSN clustering requires a multi-objective approach where information freshness is treated with the same priority as energy and security.