ML Boosts Tabu Search for Wireless Network Design

This research introduces a novel framework that combines machine learning with the classical metaheuristic Tabu Search to optimize the design of tactical wireless networks. Designing these networks under real-world constraints presents complex combinatorial optimization challenges, often requiring extensive computation to evaluate potential solutions. The proposed method aims to overcome this by using a data-driven approach to guide the move selection process within Tabu Search. The framework records search trajectories, including both successful and unsuccessful transformations, along with descriptive features of the network's structural, geometric, and performance characteristics. This data is then used to train a Graph Neural Network (GNN) that predicts the quality of candidate moves. By integrating this trained GNN, the Tabu Search algorithm can prioritize promising transformations, thereby reducing the number of costly objective function evaluations. Experimental results on benchmark instances demonstrate that this learning-assisted Tabu Search not only significantly cuts down computation time but also consistently yields higher-quality network designs than the traditional algorithm.