A Deep Reinforcement Learning-Based Multi-objective Optimization for Crowdsensing-Based Air Quality Monitoring Systems

Abstract

Global air pollution is becoming increasingly severe. In this context, monitoring air quality at all times and locations is necessary. Traditionally, air quality is monitored using stationary monitoring stations. However, this approach has an inherent shortcoming: limited monitoring locations. Crowdsensing-based air monitoring has recently emerged as a promising alternative that expands monitoring coverage in both temporal and spatial dimensions through the collaboration of numerous participants. Typically, participants in crowdsensing systems are compensated for the data they provide. One of the critical challenges in handling a crowdsensing system is minimizing the cost while guaranteeing the quality of the data collected. For crowdsensing-based air monitoring systems, data quality refers to the temporal and spatial coverage corresponding to the locations and times the data was collected. In this study, we propose a solution based on deep reinforcement learning that simultaneously optimizes two goals: maximizing coverage range and minimizing costs. Our proposed solution is one of the first attempts to optimize both of these objectives for crowdsensing-based air monitoring systems. Compared to other algorithms, experimental results indicate that the proposed solution can increase coverage by more than 30% and reduce cost by more than 70%.

N. D. Tran and M. C. Dao—The two authors contributed equally to this paper.