AI-Driven SDN Strategies for Energy-Efficient Heterogeneous Network Selection in Low-Memory IoT Edge Nodes

Green IoT nodes deployed outdoors face connectivity challenges due to weather conditions, foliage, and non-line-of-sight scenarios. To address these issues, we use heterogeneous wireless technologies such as Wi-Fi and LoRa at the IoT edge. Wi-Fi operates in the 2.4 GHz/5 GHz bands, supporting higher payload capacity but is prone to signal attenuation. In contrast, LoRa, operating in the 868 MHz band, supports lower payloads but excels in signal penetration. These networks have distinct energy consumption profiles influenced by outdoor environmental conditions. This highlights the need for AI-driven software-defined networking (SDN) strategies at IoT edge for dynamic heterogeneous network selection to optimize energy efficiency. Our green IoT nodes are powered by ambient energy sources like solar and utilize low-power hardware with limited memory. Therefore, the proposed solution must also be suitable for low-memory hardware. This paper introduces a low-memory SDN controller using AI-based Multi-Armed Bandit (MAB) policies for dynamic heterogeneous network selection to reduce energy consumption. We formulated and evaluated three MAB policies—approximate, contextual, and adversarial bandits—using real field data. The results show that approximate and contextual bandits outperform adversarial bandits, making them ideal for low-memory IoT edge devices.