Investigating the performance of multi-objective reinforcement learning techniques in the context of IoT with harvesting energy

In the realm of IoT, wireless sensor networks (WSNs) play a crucial role in efficient data collection and task execution. However, energy constraints, particularly in battery-powered WSNs, present significant challenges. Energy harvesting (EH) technologies extend battery life but introduce variability that can impact quality of service (QoS). This paper introduces QoSA, a reinforcement learning (RL) agent designed to optimize QoS while adhering to energy constraints in IoT gateways. QoSA employs both single-policy and multi-policy RL methods to address trade-offs between conflicting objectives. This study investigates the performance of these methods in identifying Pareto front solutions for optimal service activation. A comparative analysis highlights the strengths and weaknesses of each proposed algorithm. Experimental results show that multi-policy methods outperform their single-policy counterparts in balancing trade-offs, demonstrating their effectiveness in real-world IoT applications.