Intelligent Routing to Optimize Energy Consumption in Wireless Sensor Networks: A Distributed Approach with Reinforcement Learning

Energy depletion heterogeneity — not total energy consumption — determines when wireless sensor networks fail: nodes at traffic hotspots die while peripheral nodes retain 60–80% residual charge, collapsing coverage prematurely. Existing clustering protocols such as LEACH and HEED address this by optimizing cluster head selection globally, yet their episodic reformation cycles (every 20 rounds) leave them structurally blind to topology changes between updates. No published distributed protocol simultaneously optimizes residual energy, transmission distance, and node load using only one-hop information while adapting those weights online through reinforcement learning. This paper presents DEAR (Distributed Energy-Aware Routing), a protocol that makes per-transmission forwarding decisions via a three-term cost function C (n) = α·Eᵣ (n) + β·d (n) + γ·L (n), requiring no global state. DEAR-RL extends this by embedding Q-learning at cluster heads to update α, β, γ each round based on observed network conditions. Both protocols were evaluated in MATLAB R2023a across six competitors, four network scales (N ∈ 50, 100, 200, 300), and five environmental scenarios, with 50 independent runs per configuration and two-sample t-tests for significance verification. DEAR raises HND from 938 to 1, 127 rounds versus HEED at N=100 in baseline conditions — a 20. 1% gain (t=8. 4, p<0. 001, d=1. 31) — while reducing energy balance variance from σE=0. 041 J to σE=0. 023 J, a 44% improvement in distribution uniformity. Under node mobility, the advantage grows to 35. 2% (1, 014 vs 750 rounds), confirming that continuous local adaptation outperforms periodic global reformation under dynamic conditions. DEAR-RL adds a further 8. 2% over DEAR in baseline (1, 219 vs 1, 127 rounds, p<0. 001) and 13. 2% under mixed interference-mobility conditions, at 1. 1% computational overhead. Local, continuous, adaptive routing eliminates the reformation-lag bottleneck that constrains globally-informed protocols, delivering measurable lifetime gains deployable on resource-constrained hardware without infrastructure modification.