Основні поняття
The key idea is to efficiently solve the NP-hard problem of optimizing the movements of multiple aerial base stations to maximize the coverage rate for mobile ground users in site-specific environments with obstacles, by constructing a global connectivity map and using a fast online algorithm.
Анотація
The paper addresses the challenge of optimizing the movements of multiple aerial base stations (ABSs) to maximize the coverage rate (CR) for mobile ground users (GUs) in site-specific environments with obstacles. The key contributions are:
- Construction of a global connectivity map (GCM) that captures the site-specific LoS/NLoS propagation characteristics between ABS and GU locations.
- Formulation of the ABS movement optimization problem as a binary integer linear programming (BILP) problem based on the GCM.
- Proposal of a fast online algorithm that solves the BILP problem efficiently, achieving near-optimal performance with significantly reduced running time compared to using an open-source solver directly.
- Evaluation of the proposed algorithm's performance in dynamic scenarios with moving GUs, demonstrating superior CR and time efficiency compared to state-of-the-art deep reinforcement learning and evolutionary algorithm methods.
The algorithm partitions the ABS movement optimization problem into a series of ABS placement sub-problems, each solved using the fast online algorithm that conducts projected stochastic subgradient descent within the dual space. This approach can rapidly find the near-optimal ABS locations to maximize the CR of mobile GUs, while accommodating the site-specific propagation environment and GU mobility constraints.
Статистика
The paper reports the following key metrics:
The average coverage rate (ACR) achieved by the proposed algorithm is close to that obtained by the open-source solver SCIP, but with significantly reduced running time.
For a scenario with N=2 ABSs and M=20 GUs, the ACR of the proposed algorithm is 0.84 compared to 0.90 for SCIP, while the average planning time is 0.92s versus 1.47s for SCIP.
For a larger scenario with N=5 ABSs and M=100 GUs, the ACR of the proposed algorithm is 0.91 compared to 0.95 for SCIP, while the average planning time is 1.79s versus 5.56s for SCIP.
Цитати
"Our proposed algorithm achieves a high CR performance close to that obtained by the open source solver (SCIP), yet with significantly reduced running time."
"The algorithm also notably outperforms one of the state-of-the-art deep reinforcement learning (DRL) methods and the K-means initiated evolutionary algorithm in terms of CR performance and/or time efficiency."