We propose a novel UAV swarm cluster-based data collection (USCDC) algorithm for wireless sensor networks (WSNs) that addresses the challenge of data collection in geographically isolated sensor sub-regions. Traditional approaches suffer from path redundancy and energy waste because each UAV drone independently returns to the base station after completing its collection task. Our algorithm introduces a two-layer UAV architecture: cooperative UAV drones at the bottom layer handle data collection within each sub-region, while a single sink UAV drone at the top layer aggregates data from cooperative UAV drones and delivers it to the base station. This hierarchical cooperation mechanism eliminates the need for cooperative UAV drones to return to the base station, significantly reducing path overhead and energy consumption. We design a UAV temporal cooperative scheduling (UTCS) algorithm to synchronize the operations of different UAV drones using a unified time reference and event-triggered mechanisms. For path optimization within each sub-region, we develop an improved ant colony optimization (IACO) algorithm that incorporates elite ant strategy and position-state transition. For the sink UAV drone’s visiting sequence, we propose an improved particle swarm optimization (IPSO) algorithm with adaptive inertia weight and dual-objective fitness function. Extensive simulations show that the USCDC algorithm reduces path length by 52% and energy consumption by 35% compared to the parallel data collection approach without clustering. The scalability of the algorithm is validated across different network sizes.
The rapid expansion of the Internet of Things has led to large-scale deployments of wireless sensor networks in environmental monitoring, smart agriculture, and industrial automation. These networks often exhibit spatial geographical distribution characteristics, resulting in multiple isolated sub-regions that traditional multi-hop routing cannot effectively connect. Unmanned aerial vehicles (UAV drones) offer a promising solution due to their high mobility and flexible deployment. However, existing works on UAV-assisted WSN data collection either focus on single-UAV scenarios or assume that all UAV drones work independently and return to the base station separately. This independent working model leads to significant path redundancy and energy waste. To address this, we introduce a cluster-based approach where multiple cooperative UAV drones operate within separate regions, and a dedicated sink UAV drone collects data from them. The key innovations are: (1) a two-layer UAV architecture that decouples data collection from data delivery; (2) a temporal coordination algorithm that synchronizes UAV drone operations; (3) enhanced metaheuristic algorithms for path planning at both layers. Our approach is particularly beneficial for large-scale WSNs where the number of sub-regions is high.
System Model
We consider a large-scale WSN divided into \(W\) isolated sub-regions. Each sub-region contains \(A\) cluster head nodes selected by a low-energy adaptive clustering hierarchy (LEACH) protocol. All sensor nodes have identical initial energy \(E_n\) and communication radius \(R_n\). The system comprises three layers:
- Ground sensor network layer: Cluster heads collect data from ordinary nodes within their cluster.
- Cooperative UAV layer: \(W\) cooperative UAV drones, indexed by \(U = \{u_1, u_2, \ldots, u_W\}\), are deployed to collect data from cluster heads in their assigned sub-regions. Each cooperative UAV drone has initial energy \(E_c\), flight height \(H_c\), and speed \(V_c\).
- Sink UAV layer: One sink UAV drone \(u_s\) collects data from cooperative UAV drones and delivers it to the base station. It has initial energy \(E_s\), flight height \(H_s\), and speed \(V_s\).
The communication model between a cooperative UAV drone \(u_j\) and a cluster head \(z_{o,a}\) in sub-region \(o\) assumes line-of-sight (LoS) propagation due to high altitude. The path loss is:
$$l_{\text{LoS}}^{u_j,z_{o,a}} = 20 \lg\left(\frac{4\pi f_c d_{u_j,z_{o,a}}}{c}\right) + \eta_{\text{LoS}},$$
where \(f_c\) is the carrier frequency, \(c\) is the speed of light, \(d_{u_j,z_{o,a}}\) is the distance, and \(\eta_{\text{LoS}}\) is the excess loss. The channel gain is \(h_{u_j,z_{o,a}} = 10^{-l_{\text{LoS}}^{u_j,z_{o,a}}/10}\). The achievable data rate is:
$$r_{u_j,z_{o,a}} = B \log_2\left(1 + \frac{P_t h_{u_j,z_{o,a}}}{N_0 B}\right),$$
with transmit power \(P_t\), bandwidth \(B\), noise power spectral density \(N_0\). Similarly, the air-to-air communication between sink UAV drone \(u_s\) and cooperative UAV drone \(u_j\) follows:
$$l_{\text{LoS}}^{u_s,u_j} = 20 \lg\left(\frac{4\pi f_c d_{u_s,u_j}}{c}\right) + \eta_{\text{LoS}},\quad h_{u_s,u_j} = 10^{-l_{\text{LoS}}^{u_s,u_j}/10},\quad r_{u_s,u_j} = B \log_2\left(1 + \frac{P_t h_{u_s,u_j}}{N_0 B}\right).$$
The energy consumption of a UAV drone consists of hovering power \(P_h\) and flight power \(P_f\). Hovering power is:
$$P_h = \sqrt{\frac{(m_{\text{tot}} g)^3}{2\pi \sigma_p^2 \delta_p \rho}},$$
where \(m_{\text{tot}}\) is total mass, \(g\) gravitational acceleration, \(\sigma_p\) propeller radius, \(\delta_p\) number of propellers, \(\rho\) air density. Flight power scales linearly with speed \(v_f\):
$$P_f = \frac{P_{\max} – P_{\text{idle}}}{v_{\max}} v_f + P_{\text{idle}}.$$
For a cooperative UAV drone \(u_j\) in round \(x\), the total energy is:
$$E_{u_j}^x = E_{u_j,\text{fly}}^x + E_{u_j,\text{coll}}^x + E_{u_j,\text{wait}}^x + E_{u_j,\text{trans}}^x,$$
where fly energy \(E_{u_j,\text{fly}}^x = \frac{L_{u_j}^x}{V_c} (P_f + P_h)\), collection energy \(E_{u_j,\text{coll}}^x = \frac{D_{u_j}^x}{r_{u_j,z_{o,a}}}(P_h + P_{\text{com}})\), wait energy \(E_{u_j,\text{wait}}^x = t_{\text{wait}}^x P_h\), and transmission energy \(E_{u_j,\text{trans}}^x = \frac{D_{u_j}^x}{r_{u_s,u_j}}(P_h + P_{\text{com}})\). The sink UAV drone energy is:
$$E_{u_s}^x = E_{u_s,\text{fly}}^x + E_{u_s,\text{trans}}^x,$$
with fly energy \(E_{u_s,\text{fly}}^x = \frac{S_{u_s}^x}{V_s} P_f\) and transmission energy similar to above. Total energy over \(x_{\max}\) rounds is:
$$E_{\text{total}} = \sum_{x=1}^{x_{\max}} \left( \sum_{j=1}^W E_{u_j}^x + E_{u_s}^x \right).$$
Problem Formulation
Our objective is to minimize the total energy consumption \(E_{\text{total}}\) subject to constraints: (C1) each cooperative UAV drone’s cumulative energy does not exceed \(E_c\); (C2) sink UAV drone’s cumulative energy does not exceed \(E_s\); (C3) each cooperative UAV drone stays within its designated region; (C4) the sink UAV drone starts and ends at the base station; (C5) all cluster head data must be collected. The problem is non-convex, so we decompose it into two subproblems: cooperative UAV path planning and sink UAV trajectory optimization.
Proposed USCDC Algorithm
UAV Temporal Cooperative Scheduling (UTCS)
The UTCS algorithm establishes a unified timeline for heterogeneous UAV drones. The sink UAV drone’s departure time is set as the maximum completion time of data collection among cooperative UAV drones:
$$t_d = \max_{j \in W} t_{u_j}^{\text{coll}}.$$
Each cooperative UAV drone has two waiting phases: wait1 (after collection, before sink arrival) and wait2 (after transmission, until system end). The waiting times are computed via event-triggered mechanisms. This ensures minimal idle energy consumption while maintaining synchronization. The algorithm pseudocode is summarized in the following table.
| Step | Description |
|---|---|
| 1–2 | Initialize start times for cooperative UAV drones and time dictionary for sink UAV drone. |
| 3–5 | For each cooperative UAV drone \(u_j\), compute its data collection completion time \(t_{u_j}^{\text{coll}}\). |
| 6 | Set sink departure time \(t_d = \max_j t_{u_j}^{\text{coll}}\). |
| 7–12 | For each cooperative UAV drone \(u_j\): compute sink arrival time \(t_{u_j}^{\text{us,arr}}\), transmission end time \(t_{u_j}^{\text{trans,end}}\), and first waiting time \(t_{u_j}^{\text{wait1}}\). |
| 13 | Compute system end time \(t_{\text{sys,end}}\) (when sink returns to base station). |
| 14–17 | For each cooperative UAV drone, compute second waiting time \(t_{u_j}^{\text{wait2}} = \max(0, t_{\text{sys,end}} – t_{u_j}^{\text{trans,end}})\). |
| 18 | Return time link and total task duration. |
Improved Ant Colony Optimization (IACO) for Cooperative UAV Drones
We enhance the classic ant colony optimization by introducing an elite ant strategy and a position-state transition mechanism for multi-round tasks. The transition probability from node \(a\) to node \(b\) for ant \(k\) is:
$$p_{ab}^k(t) = \begin{cases} \frac{[\tau_{ab}(t)]^\alpha [\epsilon_{ab}(t)]^\beta}{\sum_{i \in \text{allowed}_k} [\tau_{ab}(t)]^\alpha [\epsilon_{ab}(t)]^\beta}, & i \in \text{allowed}_k, \\ 0, & \text{otherwise}. \end{cases}$$
Pheromone update includes evaporation and deposition. For elite ants (top \(\lambda\) fraction), extra reinforcement is added:
$$\Delta\tau_{ab}^k = \begin{cases} (1+\xi) \frac{Q}{C_k}, & \text{if } k \in \lambda, (a,b) \in F_k, \\ \frac{Q}{C_k}, & \text{if } (a,b) \in F_k, \\ 0, & \text{otherwise}. \end{cases}$$
The position-state transition mechanism uses the end position of round \(x-1\) as the start for round \(x\): \(T_{\text{start}}^x = (X_{\text{end}}^{x-1}, Y_{\text{end}}^{x-1}, H_c)\). This reduces inter-round flight distance. The algorithm is summarized below.
| Step | Description |
|---|---|
| 1–3 | Set ant population \(n_{\text{ant}}\), maximum iterations \(M\), pheromone factor \(\alpha\), heuristic factor \(\beta\), evaporation rate \(\mu\). |
| 4–8 | If first round, start from initial position; else start from previous round’s end position. |
| 9–14 | For each ant \(k\): build complete path using transition probability, compute path length and energy, evaluate fitness \(\psi_k\). |
| 15–22 | Pheromone evaporation; all ants deposit pheromone; elite ants deposit extra pheromone. |
| 23 | Return best path, length, energy. |
Improved Particle Swarm Optimization (IPSO) for Sink UAV Drone
The sink UAV drone must visit cooperative UAV drones in an optimal order. We design an IPSO algorithm with adaptive inertia weight and a dual-objective fitness function. The velocity update is:
$$v_i = \omega v_i + \theta_1 \text{rand}() (\phi_{\text{best}}^i – \iota_i) + \theta_2 \text{rand}() (\varphi_{\text{best}} – \iota_i),$$
where \(\omega\) is adaptive inertia weight:
$$\omega = \omega_{\max} – (\omega_{\max} – \omega_{\min}) \frac{\kappa_{\text{iter}}}{M}.$$
The fitness function combines normalized path length and energy:
$$\psi = w_1 \frac{d}{d_{\max}} + w_2 \frac{e}{e_{\max}}.$$
The algorithm steps are:
| Step | Description |
|---|---|
| 1–6 | Initialize particle population, inertia weight bounds \(\omega_{\max}, \omega_{\min}\), learning factors \(\theta_1, \theta_2\), positions \(\iota_i\), velocities \(v_i\), and global best. |
| 7–24 | For each iteration: update \(\omega\); for each particle, generate visit sequence starting and ending at base station, compute normalized path length and energy, compute fitness; update personal and global bests; update velocity and position. |
| 25 | Return best path, length, energy. |
Simulation Results and Analysis
We implemented simulations in PyCharm over a 2000 m × 2000 m area divided into 6 sub-regions. Key simulation parameters are listed in the table below.
| Parameter | Value |
|---|---|
| Number of sub-regions \(W\) | 6 |
| Sub-region radius \(R_r\) | 200 m |
| Sensor initial energy \(E_n\) | 1 J |
| Sensor communication radius \(R_n\) | 50 m |
| Sink UAV initial energy \(E_s\) | 2,700,000 J |
| Sink UAV height \(H_s\) | 75 m |
| Sink UAV speed \(V_s\) | 25 m/s |
| Cooperative UAV initial energy \(E_c\) | 1,500,000 J |
| Cooperative UAV height \(H_c\) | 50 m |
| Cooperative UAV speed \(V_c\) | 15 m/s |
| Flight power \(P_f\) | 90 W |
| Hovering power \(P_h\) | 30 W |
| Communication power \(P_{\text{com}}\) | 0.5 W |
| Ant population \(n_{\text{ant}}\) | 100 |
| Pheromone importance \(\alpha\) | 1.5 |
| Heuristic factor \(\beta\) | 3.0 |
| Evaporation coefficient \(\mu\) | 0.6 |
| Pheromone intensity \(Q\) | 100 |
| Particle population \(n_{\text{part}}\) | 100 |
| Inertia weight max \(\omega_{\max}\) | 0.95 |
| Inertia weight min \(\omega_{\min}\) | 0.4 |
| Cognitive factor \(\theta_1\) | 2.05 |
| Social factor \(\theta_2\) | 2.05 |
We compare our USCDC algorithm with a baseline where all cooperative UAV drones fly independently back to the base station after collection (USPDC). Results over 10 rounds show that the total path length of USCDC is about 10 km, while USPDC requires 21 km—a reduction of 52.4%. The energy consumption per round for USCDC is approximately \(8.5 \times 10^4\) J, compared to \(13.5 \times 10^4\) J for USPDC, a 37% reduction. Energy efficiency increases from 6.5 bit/J to 10 bit/J, a 53.8% improvement. The following tables summarize these findings.
| Metric | USCDC | USPDC | Improvement |
|---|---|---|---|
| Total path length (km) | 10 | 21 | 52.4% reduction |
| Per-round energy (\(10^4\) J) | 8.5 | 13.5 | 37% reduction |
| Energy efficiency (bit/J) | 10 | 6.5 | 53.8% increase |
We also tested scalability across three network sizes: small (1 km × 1 km, 3 sub-regions), medium (2 km × 2 km, 6 sub-regions), and large (3 km × 3 km, 12 sub-regions). The results are shown below.
| Network Scale | USCDC Path Length (km) | USPDC Path Length (km) | Path Length Reduction | USCDC Energy (\(10^4\) J) | USPDC Energy (\(10^4\) J) | Energy Reduction |
|---|---|---|---|---|---|---|
| Small | 5 | 6.94 | 28% | 4.2 | 5.32 | 21% |
| Medium | 10 | 21 | 52% | 8.5 | 13.5 | 35% |
| Large | 18 | 53.5 | 66.3% | 15.3 | 25.5 | 40% |
The path length advantage becomes more pronounced as network size grows: 28%, 52%, and 66.3% reduction, respectively. Energy reduction, though lower in relative terms (21%, 35%, 40%), still demonstrates consistent improvement. The discrepancy arises because the sink UAV drone incurs extra hovering waiting time as the number of cooperative UAV drones increases. This suggests future optimization of the sink visit sequence to minimize waiting energy.
Conclusion
We proposed the USCDC algorithm for efficient data collection in large-scale wireless sensor networks using a two-layer UAV drone architecture. By decoupling data collection (cooperative UAV drones) from data delivery (sink UAV drone), we significantly reduce redundant flight paths and energy consumption. The UTCS algorithm ensures proper temporal coordination among UAV drones, while IACO and IPSO optimize path planning at the cooperative and sink layers, respectively. Simulation results confirm that our approach reduces total path length by up to 66% and energy consumption by up to 40% compared to the baseline. Future work will focus on reducing the waiting time of cooperative UAV drones by optimizing the sink UAV drone’s trajectory more aggressively, and extending the algorithm to continuous sensor field scenarios.