Robustness Analysis of Data-Driven Self-Learning Controllers for IoT Environmental Monitoring Nodes based on Q-learning Approaches

Today we are seeing a significant need for efficient control of operating cycles to deliver improvements in services provided by Internet of Things (IoT) devices embedded with environmental monitoring. To design an algorithm which provides sufficient duty-cycle control, we can apply machine learning approaches. The present study investigates the reinforcement learning (RL) algorithm family, especially Q-learning (QL) and Double Q-learning (DQL) algorithms and their suitability for devices deployed in a range of locations. We present a comprehensive analysis of the implemented RL approaches for regulating data-driven self-learning (DDSL) controllers. We tested QL and DQL algorithms on various datasets and evaluated their performance with a statistical analysis. The results indicated that the QL and the DQL approaches were highly dependent on the nature of the environmental parameters which the DDSL controller detected and recorded.