Maximizing Sensing and Computation Rate in Ad Hoc Energy Harvesting IoT Networks

This article considers the collection and processing of data by solar-powered servers operating in an Internet of Things (IoT) network. Specifically, these servers aim to cooperatively maximize the amount of data collected from devices and computed over multiple time slots. To achieve this aim, they must consider computation deadline, time-varying energy arrivals at sensor devices and other servers. To this end, this article outlines a mixed-integer linear program (MILP), which can be used to optimize the sensing time of sensor devices, offloading decision of each server, and the number of virtual machines (VMs) assigned to each device. Further, this article proposes a multiagent co-operative
$Q$
-learning approach coupled with the Hungarian algorithm to assign VMs to devices. It allows servers to learn when to share their energy and VMs with neighbor servers using only noncausal energy and channel gain information. The simulation results show that the amount of processed data by co-operative
$Q$
-learning is 93% that of MILP.