Smart IoT Devices Using AI-Based Energy-Efficient Edge Computing Architecture

Abstract

As the number of IoT-connected smart objects, including environmental, wearable, smart-home, and industrial sensors, keeps growing, delivering low-latency, privacy-preserving intelligence at the network edge is no longer just convenient; it is becoming a requirement. Conventional edge deployments typically rely on fixed resource allocations that pay little attention to fluctuating workloads, device heterogeneity, or battery constraints, so energy ends up being spent needlessly. This paper introduces an edge computing framework built around four coordinated, energy-aware components: a reinforcement-learning (RL) taskoffload scheduler, an adaptive dynamic voltage and frequency scaling (DVFS) controller, a dynamic model quantization/pruning engine, and a duty-cycled sensor gating manager. These sit within a four-tier architecture spanning the cloud, edge orchestration, edge compute, and device layers, extending the threetier structure common in industry practice by introducing an explicit orchestration tier with clearly defined responsibilities. We describe the resulting system architecture and interlayer coordination protocol and evaluate the approach through a discrete-event simulation calibrated against representative embeddedinference power profiles. The simulated results show a 77.6% reduction in energy per inference relative to cloud-only offloading, a 57.0% reduction relative to a statically configured edge baseline, and a 3.7x improvement in projected device battery life, with end-to-end latency scaling sub-linearly as concurrent device load increases. An ablation study shows removing the RL-based offload scheduler causes the largest single drop in performance, with dynamic quantization contributing the next-largest share. We close by addressing real-world deployment challenges, particularly scale and security, and point to federated edge learning and standardized, cross-layer energy orchestration as key directions for future work.