Enhanced Gold Mining Optimization with Dual-Layer Information Sharing and Dynamic Control for Multi-Objective Scheduling of Heterogeneous UAV Swarms

Technical Summary: Enhanced Gold Mining Optimization for Heterogeneous UAV Swarms

Problem Statement
The paper addresses the complex challenge of multi-objective collaborative scheduling for heterogeneous Unmanned Aerial Vehicle (UAV) swarms. This problem involves the tightly coupled optimization of task allocation and trajectory planning for fleets comprising diverse UAV types (e.g., fixed-wing, rotary-wing, and hybrid-wing). The objective is to simultaneously optimize three conflicting goals under strict operational constraints:

1. Minimizing mission completion time (makespan).
2. Minimizing total energy consumption.
3. Maximizing workload balance (minimizing the standard deviation of individual UAV completion times).

The problem is constrained by task uniqueness, energy feasibility, time windows, and dynamic feasibility (kinematic limits like turning radius and climb angles). The authors argue that existing meta-heuristics often suffer from structural decoupling of task and trajectory planning, static algorithmic designs that lack adaptability to real-time changes, and suboptimal Pareto front uniformity (gaps or clustering). Furthermore, the original Gold Mining Optimization (GMO) algorithm, while effective in global exploration, lacks sufficient population diversity, adaptive step-size control, and dedicated mechanisms for multi-objective handling, leading to premature convergence and poor Pareto front approximation in this specific domain.

Methodology: IGMO-DP
To overcome these limitations, the authors propose the Improved Gold Mining Optimization with Dual-layer Information Sharing and Dynamic Pareto Dominance (IGMO-DP). The methodology integrates three core enhancements into the standard GMO framework:

1. Dual-Layer Information Sharing Mechanism:

   • Local Neighborhood Enhancement: Utilizes a topological neighborhood structure where individuals exchange information with their $K_{neigh}$ nearest neighbors. A weighted average fusion strategy assigns higher weights to neighbors with better non-dominated ranks, preserving local diversity and preventing premature convergence.
   • Global Elite Guidance: Directs the population toward the Pareto-optimal front using an external archive of elite solutions.
   • Adaptive Fusion: A time-varying coefficient $\alpha(t)$ dynamically balances the influence of local and global vectors, prioritizing global exploration in early iterations and local refinement in later stages.

2. Dynamic Position Update Mechanism:

   • Adaptive Step Size: The step size $\eta_i(t)$ is adjusted based on the individual's fitness improvement and the iteration count. This allows for large steps during global exploration and smaller, precise steps during local exploitation.
   • Directional Adjustment: A gradient-based factor is introduced to adjust the search direction based on the local terrain of the objective function, balancing information guidance with gradient descent.

3. Dynamic Pareto Dominance and Archive Maintenance:

   • Dynamic Dominance Threshold: A threshold $\tau(t)$ decreases over iterations. In early stages, it relaxes dominance criteria to retain diverse, slightly inferior solutions; in later stages, it tightens to standard Pareto dominance to accelerate convergence.
   • Elite Archive: Maintains a non-dominated solution set, using crowding distance to prune the archive when capacity is exceeded, ensuring uniform coverage of the Pareto front.

The algorithm encodes solutions as task allocation matrices and execution sequences, utilizing Dubins path models to ensure kinematic feasibility during fitness evaluation. Final decision-making among the Pareto set is performed using the TOPSIS method.

Key Contributions

• Algorithmic Innovation: The development of IGMO-DP, which specifically adapts the Gold Mining Optimization algorithm for multi-objective heterogeneous UAV scheduling through dual-layer information sharing and dynamic control mechanisms.
• Problem Formulation: A rigorous mathematical model that jointly optimizes time, energy, and load balance while enforcing strict kinematic and temporal constraints via Dubins paths.
• Mechanism Analysis: The introduction of a dynamic Pareto dominance strategy that self-adjusts selection pressure without manual parameter tuning, effectively balancing convergence speed and solution diversity.

Experimental Results
The IGMO-DP algorithm was evaluated against four baselines (original GMO, NSGA-II, MOPSO, and HMOEA) across three scenarios of increasing complexity (S1: 5 UAVs/10 tasks; S2: 10 UAVs/20 tasks; S3: 20 UAVs/40 tasks).

• Convergence and Quality: IGMO-DP consistently outperformed all baselines. In the most complex scenario (S3), it achieved an Inverted Generational Distance (IGD) improvement of 12.7%–18.3% and a Hypervolume (HV) improvement of 8.2%–14.6% compared to the second-best algorithm (HMOEA).
• Uniformity: The algorithm produced Pareto fronts with significantly better spacing (SP), showing a 15.1%–22.4% improvement in solution distribution uniformity.
• Stability: IGMO-DP demonstrated superior stability across 30 independent runs, with lower standard deviations and fewer outliers compared to other algorithms.
• Ablation Studies: Experiments confirmed that removing any single component (local sharing, global sharing, adaptive step size, or dynamic dominance) resulted in measurable performance degradation, validating the necessity of each module. For instance, removing local sharing degraded the SP metric by 21.3%.
• Computational Efficiency: While IGMO-DP incurred a modest runtime increase (approx. 15% higher than NSGA-II in S2), the authors note this is acceptable for offline scheduling applications where solution quality is paramount.

Significance and Claims
The paper claims that IGMO-DP provides a robust, scalable solution for the multi-objective scheduling of heterogeneous UAV swarms. Its primary significance lies in its ability to:

1. Simultaneously optimize conflicting objectives (time, energy, load) while respecting complex physical constraints.
2. Maintain high diversity in the solution set, offering decision-makers a wide range of well-distributed trade-off options (e.g., prioritizing speed vs. energy in disaster response scenarios).
3. Operate autonomously without the need for manual parameter calibration for balancing convergence and diversity, thanks to its dynamic dominance and adaptive mechanisms.

The authors conclude that the algorithm is feasible for engineering applications, citing a specific case study where it reduced task completion time by 23% and energy consumption by 18% compared to traditional greedy methods in a disaster response scenario. Future work is suggested to focus on online scheduling via reinforcement learning, distributed architectures, and hardware-in-the-loop testing.