Browsing by Author "Carvalho, Rodrigo Zuolo"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Mobility planning in edge assisted low power wide area networksPublication . Carvalho, Rodrigo Zuolo; Correia, Noélia Susana Costa; Al-Tam, FaroqEdge computing infrastructures are being integrated with Internet of Things (IoT) systems to facilitate time-critical applications. These systems often require data to be processed within a specific time window, so the edge becomes vital in developing reactive IoT applications with time restriction requirements. Although different architectural designs will always have advantages and disadvantages, mobile gateways are particularly relevant in enabling this integration with the edge, particularly in the context of wide-area networks with occasional data generation. In these scenarios, mobility planning is necessary, as aspects of the technology need to be aligned with the temporal needs of an application. This dissertation intersects machine learning techniques and mathematical models to establish a framework that solves the problem of mobility planning for LoRaWAN gateways when cooperating with edge system architectures. Throughout the pipeline for attaining this objective, some sideline contributions are yielded, such as machine learning agents to improve the Adaptive Data Rate (ADR) mechanism, mathematical models to estimate the gateways’ journey time, and machine learning agents that meet the constraints on valid data collection and delivery to edge systems. The nature of the problem at hand makes cutting-edge Deep Reinforcement Learning (DRL) techniques helpful in solving inherited issues, such as dealing with multiple dimensions in the action space while aiming for optimum system performance. This dissertation culminates in a novel scalable DRL model incorporating a pointer network (Ptr-Net) and an Actor-Critic (AC) algorithm to handle complex action spaces. The model synchronously determines the gateway location and visit time. Ultimately, the gateways can achieve a trajectory planning fulfilling all requirements while reducing latency and energy waste.