A Multiobjective Matrix Particle Swarm Optimization Algorithm Integrating Graph Neural Networks and Multilayer Perceptrons for IRS-Assisted Drone Deployment

Abstract

The evolution of 6G networks toward integrated air–ground–space systems has highlighted drones as a key solution for enhancing communication in complex terrains. However, their practical deployment faces multiple challenges, including non-line-of-sight (NLoS) propagation, cochannel interference, and energy constraints. To address these issues, this article proposes a graph neural network (GNN)- and multilayer perceptron (MLP)-assisted multiobjective matrix particle swarm optimization (GM-MOPSO) algorithm. We construct a 3-D optimization model that integrates signal coverage, interference suppression, and energy efficiency, aiming to overcome the shortcomings of traditional methods, such as slow convergence and susceptibility to local optima in high-dimensional nonconvex problems. By introducing a physics-informed GNN surrogate model to accelerate evaluation and an MLP-based adaptive parameter controller, the proposed algorithm effectively balances global exploration and local exploitation. Simulation results demonstrate that GM-MOPSO significantly improves Pareto frontier quality and computational efficiency in real urban scenarios. The algorithm achieves collaborative optimization of drone 3-D positioning, intelligent reflecting surface (IRS) phase shifts, and power allocation, providing a new paradigm for green and efficient wireless network deployment. The code related to this algorithm is available at https://github.com/Donglin0730/GM-MOPSO