An Online Approach to Balancing the Energy Consumption for IoT Sensors in UAV-based MEC Systems

In this paper, we consider an unmanned aerial vehicle-enabled (UAV-enabled) Mobile Edge Computing (MEC) Internet of Things (IoT) system consisting of multiple energy harvesting devices. In the system, the UAV acts as the edge computing platform providing computation service to the devices. We assume that the energy and sensing data arrive at the devices stochastically. As the amount of energy harvested by devices can differ significantly, we attempt to balance the energy consumption among devices thus avoiding the loss of devices due to the depletion of the battery energy. To ensure energy fairness, we employ an a-utility function, which incorporates the device's remaining energy and energy expenditure. Without any future knowledge of the data and energy arrivals, our objective is to design an online algorithm to jointly optimize the long-term system throughput and utility, while satisfying the stability requirement of the data queue. We formulate the problem as multi-stage stochastic programming that involves functions of time averages and propose an online algorithm based on the perturbed Lyapunov optimization technique. Due to the time-coupling of different variables, we first transform the original problem into a multi-stage stochastic optimization problem involving only time averages. Then, we utilize the Lyapunov optimization method to decouple the transformed problem into a series of per-slot deterministic subproblems. In each slot, we alternatively optimize the auxiliary variables, the data sensing rate, the computation offloading, the communication resource allocation, and the trajectory of the UAV. Simulation results show that the online algorithm can maintain the stability of the data queue. Additionally, we observe a tradeoff between the system's lifetime and throughput when adjusting the parameter <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\alpha$</tex>, as well as a tradeoff between throughput and delay for the Lyapunov method.