<![CDATA[The rapid deployment of Internet of Things (IoT) networks enables environmental monitoring, but conventional fixed-interval sensing causes excessive energy consumption, communication overhead, and limited adaptability. This study proposes an adaptive edge intelligence framework using Deep Q-Network (DQN) for energy-efficient air quality monitoring. The system integrates environmental sensors with an ESP32-based edge node to measure PM2.5, PM10, CO₂, NOx, SOx, NHx, temperature, humidity, and traffic-related information. A DQN agent dynamically selects sensing and communication actions using nine environmental and contextual state variables. A customized reward function balances air quality conditions, energy consumption, communication costs, and monitoring requirements. Experimental data were collected through an IoT platform and stored in Supabase. The agent was trained using 103 valid observations over 200 episodes and independently evaluated on 114 observations. Training rewards increased from −43.62 to approximately 270 after convergence. Evaluation achieved a total reward of 316.092, an average reward of 2.773 per decision step, and an average energy cost of 0.20 per step. Under stable environmental conditions, the agent consistently selected SLOW_SAMPLE, reducing unnecessary sensing operations. These findings demonstrate the feasibility of DQN-based adaptive monitoring and provide a scalable foundation for energy-efficient, self-optimizing IoT environmental sensing systems under varying environmental conditions.]]>
Copyrights © 2026
