Exploiting Deep Reinforcement Learning for Coverage Maximization and Cost Minimization in Air Quality Monitoring Systems

Abstract

Air-quality monitoring is highly desired in modern life where an environmental problem has become more and more serious. Such a task requires continuous surveillance over large urban areas, which is costly both in infrastructure and sensor resources. Hence gives rise to the Vehicular Mobile Networks (VMNs), in which mobile vehicles play the role of sensor devices and constantly monitor the area. However, with extensive constraints, the optimization of both maximizing the coverage and minimizing the sensory costs is vastly challenging. In this research, we resolve the problem in terms of a learning process. Applying deep reinforcement learning, we outperform more than 1.65% in terms of coverage, compared to common setups while remaining considerably small sensory costs in terms of sensor activation rate. We conduct extensive experiments for a better understanding of the behavior of the deep reinforcement learning model.