In modern warfare, the integration of drone technology has revolutionized combat strategies, particularly through the deployment of Unmanned Aerial Vehicle (UAV) clusters. These clusters leverage distributed computing to enhance operational efficiency, but they face challenges such as limited onboard computational resources and the need for real-time data processing in dynamic environments. To address these issues, we propose a novel network architecture that synergizes cloud, fog, and edge computing paradigms. This architecture optimizes resource utilization across dispersed nodes within and outside UAV clusters, ensuring robust performance in bandwidth-constrained scenarios. By incorporating dynamic load balancing strategies, our system mitigates load imbalances and environmental sensitivities, thereby improving overall computational efficiency and resilience. In this paper, we delve into the internal composition and network structure of UAV clusters, analyze the integration of distributed computing modes, and validate the approach through simulations that demonstrate enhanced robustness and reduced latency.
The UAV cluster combat system operates on a hierarchical command structure, forming a multi-tiered interconnected network from basic combat units to advanced command layers. Large UAV clusters, controlled by command centers, provide signal coverage and resource allocation for smaller clusters managed by individual soldiers. These large clusters incorporate various functional UAVs, such as those equipped with fog computing servers, communication relays, and combat support units, to meet the demands of acting as aerial fog nodes. Small UAV clusters, consisting of portable multi-functional drones, offer tactical support to soldiers. This design ensures a scalable and adaptive system, as illustrated in the overall system diagram. The cloud-fog-edge collaborative computing network architecture comprises three layers: the cloud computing center at the top, which handles large-scale data analysis and decision support; the fog computing layer in the middle, formed by large UAV clusters that process and filter data; and the edge computing layer at the bottom, consisting of UAVs and soldier terminals that perform real-time data preprocessing. This layered approach minimizes latency and bandwidth usage by leveraging local computation where possible.
For computing tasks that do not require cloud offloading, the system employs local computation, edge cloud computing, and fog-edge collaborative computing. The edge cloud computing network involves small multi-functional UAVs offloading tasks to high-performance mobile devices within their detection range, forming a localized network that reduces energy consumption and extends operational duration. In contrast, the fog-edge collaborative computing network relies on large UAV clusters acting as fog nodes to process data from small clusters, utilizing multi-hop communication links to maintain stable connectivity and load distribution. This network architecture simplifies UAV entry and exit procedures and ensures communication stability despite cluster density fluctuations. To model the distributed computing processes, we represent the network as an undirected cyclic graph $G = (V, E)$, where $V = \{v_0, v_1, \dots, v_p\}$ denotes the set of nodes (e.g., UAVs and computing devices), and $E = \{(v_i, v_j) | v_i, v_j \in V \text{ and } i \neq j\}$ represents the edges (communication links). The computational capacity of each node $v_i$ is denoted by $C_i$, and the bandwidth of edge $(v_i, v_j)$ is $b_{ij}$. The total delay $t$ for task processing includes communication delay $t_p$, load diffusion delay $t_d$, computation delay $t_c$, and result feedback delay $t_b$, expressed as:
$$t = t_p + t_d + t_c + t_b$$
Where $t_p$ is bounded by $0 < t_p \leq (2R_G + n) \cdot \tau$, with $R_G$ as the graph diameter and $\tau$ as the node communication delay; $t_d = \sum_{k=1}^n \max_{(v_i,v_j) \in E} (|\Delta_{ij}^k| \div b_{ij})$ for $n$ diffusion steps; $t_c = \max_{i=0,1,\dots,p} (l_i \div b_{ij})$; and $t_b = \max_{i=0,1,\dots,p} (t_i)$. Simplifying by ignoring small terms, the optimization problem minimizes:
$$\min\left[ \sum_{k=1}^n \max_{(v_i,v_j) \in E} (|\Delta_{ij}^k| \div b_{ij}) + \max_{i=0,1,\dots,p} (l_i \div b_{ij}) \right]$$
Subject to $\sum_{i=0}^p l_i = L$, where $L$ is the total load. We implement the Generalized Diffusion Algorithm (GDA) for dynamic load balancing, which uses a diffusion matrix $M = M_0^1(\epsilon)$ for fast convergence in heterogeneous environments. The algorithm involves nodes exchanging load information with neighbors, computing load transfers, and iterating until equilibrium is reached, as detailed in Algorithm 1. This approach ensures efficient task distribution across the UAV cluster combat system, enhancing performance in unpredictable warfare scenarios.
Simulations were conducted to evaluate the system’s robustness to cloud service bandwidth variations and its performance stability. The setup included 5 edge computing nodes, 6 fog nodes, and 1 cloud node, with computational capacities ranging from 48 MIPS for soldier-carried devices to 5000 MIPS for the cloud center. Link bandwidths were randomly assigned between 80-110 Mbps. Assumptions included 20 machine instructions per byte of data for delay calculation. Results compared processing delays across different computing modes: edge cloud computing versus local and cloud-terminal computing showed that edge computing outperforms others for small data sizes, requiring at least 109.7 Mbps cloud bandwidth for parity. Fog-edge collaborative computing demonstrated lower latency than cloud-terminal computing, necessitating over 207.5 Mbps cloud bandwidth for equivalence. The cloud-fog-edge approach exhibited superior robustness, with slower performance degradation under bandwidth fluctuations. Fault tolerance analysis revealed that edge computing improves resilience as the number of connected edge servers increases, while fog-edge computing depends heavily on robust fog nodes. Comparative tests of load balancing algorithms—GDA, Smooth Weighted Round Robin (SWRR), Pick-KX, and Greedy Load Balancing (GreedyLB)—confirmed GDA’s efficiency in reducing task processing delays across all computing modes.
| Node | $v_0$ | $v_1$ | $v_2$ | $v_3$ | $v_4$ | $v_5$ | $v_6$ | $v_7$ | $v_8$ | $v_9$ | $v_{10}$ | $v_{11}$ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| $C_i$ (MIPS) | 102 | 50 | 51 | 52 | 48 | 94 | 500 | 500 | 93 | 104 | 97 | 5000 |
The integration of cloud, fog, and edge computing in UAV cluster combat systems represents a significant advancement in drone technology. By employing distributed computing and dynamic load balancing, our architecture enhances computational efficiency, reduces latency, and improves adaptability to environmental changes. Future work could focus on real-world deployment and further optimization of algorithms for specific military applications. The proposed system underscores the potential of Unmanned Aerial Vehicle clusters in modern warfare, ensuring reliable performance even in communication-limited environments.
In summary, this research contributes to the evolution of drone technology by addressing key challenges in UAV cluster operations. The cloud-fog-edge network architecture, combined with effective load balancing, paves the way for more resilient and efficient Unmanned Aerial Vehicle systems in diverse combat scenarios.