A Polynomial-Decay and Pinhole-Imaging Whale Optimization Algorithm for UAV Relay Communication Deployment

Technical Summary: A Polynomial-Decay and Pinhole-Imaging Whale Optimization Algorithm for UAV Relay Communication Deployment

Problem Statement
The paper addresses the joint optimization of Unmanned Aerial Vehicle (UAV) relay deployment, a critical task for providing flexible, on-demand wireless coverage in scenarios such as disaster recovery and tactical communications. The objective is to simultaneously determine the UAV's 3D position (horizontal coordinates $x, y$ and altitude $h$), transmit power ($p_{tx}$), and bandwidth ($B$). This constitutes a non-convex, heavily constrained black-box optimization problem. The deployment must satisfy five specific inequality constraints: minimum Shannon capacity, line-of-sight clearance relative to altitude, a power-bandwidth budget, and horizontal range limits for two ground nodes. The goal is to minimize a service cost function (accounting for power, altitude, and bandwidth) while maximizing capacity and adhering to constraints, typically handled via a quadratic penalty function.

Methodology: The PWOA Algorithm
The authors propose the Polynomial-decay and Pinhole-imaging Whale Optimization Algorithm (PWOA). Rather than a loose accumulation of variants, PWOA is designed as a role-coupled optimizer where specific operators address distinct failure modes of the standard Whale Optimization Algorithm (WOA). The methodology integrates three complementary improvements:

1. Good Nodes Set (GNS) Initialization:
   To counteract the uneven coverage and clumping caused by random initialization—particularly problematic in narrow feasible regions—the authors introduce a deterministic GNS initialization. This method spreads the initial population uniformly across the search space with a small jitter to preserve diversity. This ensures better space-filling properties compared to pseudo-random sampling, providing a more balanced starting point for the swarm.

2. Polynomial Nonlinear Decay of the Convergence Factor ($a$):
   The standard WOA uses a linear decay for the convergence factor $a$, which often forces premature exploitation. PWOA replaces this with a polynomial nonlinear schedule ($a(t)$) where the exponent $\mu > 1$. This design prolongs the exploration phase (retaining larger $|A|$ values) during the first half of the iterations and sharpens the exploitation phase (rapidly decreasing $a$) in the second half. This adaptive scheduling aligns the search behavior with the landscape of the problem, delaying the transition to local search until sufficient global exploration has occurred.

3. Stagnation-Triggered Pinhole-Imaging Opposition-Based Learning (POBL) with Elite Local Search:
   To escape local optima without disrupting the search unnecessarily, PWOA employs a conditional recovery mechanism triggered only when the global best score stagnates (exceeding a threshold $\sigma_{max} = 15$).

   • Elite Gaussian Local Search: At every iteration, a local search is performed around the current leader ($X^*$) using a shrinking Gaussian step.
   • Pinhole-Imaging POBL: Upon stagnation, a pinhole-imaging transformation is applied to the leader. The search space center acts as a "pinhole," reflecting the leader to generate a candidate solution. The magnification factor $k(t)$ is adaptive: it is small in early iterations to allow aggressive long-range jumps for escaping local minima, and large in later iterations to map candidates closer to the search space centroid for controlled diversification. The resulting candidate is injected into the worst individual in the population to break stagnation.

Key Contributions
The paper outlines three primary contributions:

1. Problem Formulation: A rigorous five-dimensional constrained UAV relay deployment model incorporating five inequality constraints and a quadratic penalty function.
2. Algorithm Design: The development of PWOA, which uniquely combines GNS initialization, polynomial nonlinear decay, and a stagnation-triggered POBL operator with elite Gaussian local search.
3. Empirical Validation: A comprehensive evaluation against three baselines (standard WOA, Sine Cosine Algorithm (SCA), and an Improved Particle Swarm Optimization (IPSO)) over 30 independent runs.

Experimental Results
The experiments were conducted with a population size ($N$) of 30 and a maximum of 500 iterations. The results demonstrate that PWOA outperforms all baselines across all metrics:

• Accuracy and Reliability: PWOA achieved the lowest Best, Worst, Mean, and Standard Deviation (Std) values simultaneously.
• Performance Gains: Compared to the baselines, PWOA reduced the Mean fitness by 1.4% (vs. SCA), 2.9% (vs. WOA), and 18.5% (vs. IPSO). It reduced the Standard Deviation by 15–87%, indicating significantly higher stability and reliability.
• Convergence Speed: PWOA exhibited the fastest average convergence, dropping below the WOA curve within the first 30 iterations and maintaining the lowest fitness for the remainder of the run. In contrast, SCA showed initial stagnation, and IPSO was the slowest due to its operators pushing the swarm too far from the feasible region early on.

Significance and Claims
The paper claims that PWOA represents a significant advancement in swarm intelligence for UAV deployment by addressing the specific limitations of standard meta-heuristics in constrained, non-convex environments. The authors emphasize that the algorithm's success stems from its "coverage–transition–recovery" chain: GNS ensures initial coverage, polynomial decay manages the exploration-exploitation transition, and the stagnation-triggered POBL provides a targeted recovery mechanism for trapped leaders.

The significance of the work lies in its applicability to "one-shot or limited-data" relay deployment scenarios where the objective function is directly evaluable but difficult to convexify or differentiate, making model-driven or learning-based approaches less suitable. The paper concludes that PWOA is a robust, role-coupled optimizer that effectively balances global search and local refinement, offering a superior alternative to existing WOA variants and other swarm algorithms for complex UAV deployment tasks. Future work is noted to extend this approach to multi-UAV cooperative deployment and closed-loop scenarios with online channel estimation.