Mixed-Variable Correlation-Aware Metaheuristic for Deployment Optimization of 3-D Sensor Networks

Deployment optimization of 3-D sensor networks is essential for the overall cost of the system and the downstream tasks performance. The key of establishing realistic deployment is twofold: a high-fidelity mathematical programming model and an efficient algorithm for solving it. In this paper, we revisit the 3-D sensor networks deployment and present a mixed-variable optimization problem (MVOP) which jointly considers the discrete subset selection decision, continuous orientation decision, and decisionmaking under uncertainty. Based on the proposed real-world application, we innovatively design a mixed-variable correlation-aware genetic algorithm as the solver. Different from mainstream two-partition methods in MVOP, our algorithm captures the problem-specific features of deployment optimization and introduces a correlation-aware search paradigm which interactively updates the discrete and continuous decision variables. On the one hand, we update the discrete part (i.e., subset selection of candidate locations) first and then optimize the continuous part (i.e., sensor orientation parameters). On the other hand, we customize a heuristic mechanism to start with continuous part to identify the suitable discrete part. Experiments demonstrate that our approach can improve the performance of small-scale and large-scale scenarios of deployment by up to 55.7% and 56.4%, respectively, compared to state-of-the-art MVOP algorithms.