Multi-UAV Assisted Data Gathering in WSN: A MILP Approach For Optimizing Network Lifetime

TL;DR

This paper presents MILP models and heuristics for optimizing data gathering in large-scale WSNs using multiple UAVs, significantly extending network lifetime and solving problems with up to 50 nodes efficiently.

Contribution

The paper introduces two MILP formulations and a heuristic to optimize UAV-assisted data collection, addressing the entire problem holistically for the first time.

Findings

Optimal solutions for up to 50 nodes in under 30 minutes.

Heuristic approach extends network lifetime more effectively.

MILP models outperform existing methods in network longevity.

Abstract

In this paper, we study the problem of gathering data from large-scale wireless sensor networks using multiple unmanned air vehicles (UAVs) to gather data at designated rendezvouses, where the goal is to maximize the network lifetime. Previous proposals often consider a practical approach where the problem of determining a data gathering scheme is decomposed into 2 sub-problems: i) partitioning the networks into clusters for determining the rendezvouses as these obtained cluster heads; and ii) determining the paths for a set of a given number of UAVs to come gathering data at these rendezvouses which have been harvesting data within each local clusters, respectively. We try to deal with this as a whole optimization problem, expecting a significant increase in computation complexity which would bring new challenge in creating practical solutions for large-scale WSNs. We introduce two alternatives mixed-integer linear programming (MILP) formulations, namely the 2-index model with $O(n^2)$ variables and the 3-index model that has $O(n^3)$ variables, where $n$ denotes the number of sensor nodes. We show that our best model could solve optimally the problem instances with up to 50 sensor nodes in less than 30 minutes. Next, we propose a heuristic idea to reduce the number of variables in implementing the 3-index model to effectively handle larger-scale networks with size in hundreds. The experiment results show that our heuristic approach significantly prolongs the network lifetime compared to existing most efficient proposals.