Sensing Data Aggregation in Mobile Crowd Sensing: A Cloud-Enhanced-Edge–End Framework With DQN-Based Offloading

Mobile crowd sensing (MCS) has attracted extensive attention as a promising method for environmental sensing and data collection. However, due to the increasing computational tasks, bandwidth, and computing pressure, traditional “Cloud-Edge–End” MCS is insufficient to efficiently offload sensing data in real-time for data aggregation. Therefore, we consider a novel MCS framework based on the “Cloud-Enhanced-Edge–End,” the framework uses MCS idle users as edge nodes (ENs) to assist edge servers to enhance computing power and reduce delay and energy consumption. To achieve efficient data aggregation in the “Cloud-Enhanced-Edge–End” MCS framework, we first consider the multiobjective optimization of delay and energy consumption to establish a utility function. Second, addressing the shortcomings of traditional optimization algorithms, which often struggle with complex decision spaces and dynamic environments, we propose a MCS offloading (MCSOL) algorithm based on deep Q-network (DQN). The algorithm uses reinforcement learning to adaptively offload computing tasks to the optimal ENs or servers to maximize utility function to reduce delay and energy consumption. Our experimental results reveal that, in comparison with the other five strategies like only offloading data to base stations and ENs, MCSOL enhances data aggregation performance within the range of 30%–60%.