In the context of large-scale Internet of Things deployments, wireless sensor networks frequently encounter spatial segmentation, resulting in multiple isolated sub-regions due to geographical constraints. This characteristic presents significant challenges for conventional data collection methods. The advent of drone technology has provided a novel perspective on this issue, leveraging its high maneuverability and deployment flexibility. However, existing drone-assisted data collection schemes often require each drone to operate independently and return to the base station for data delivery, leading to considerable path redundancy and energy inefficiency. To address these limitations, we propose the UAV Swarm Cluster-based Data Collection (USCDC) algorithm, which is specifically designed to optimize the performance of drone technology in wireless sensor network data acquisition. This algorithm employs a two-layer drone architecture: cooperative drones at the lower layer are responsible for data collection within individual regions, while a sink drone at the upper layer aggregates and transmits the data to the base station. This hierarchical collaboration mechanism allows cooperative drones to complete data transmission without returning to the base station, thereby substantially reducing path overhead and energy consumption. This paper presents a comprehensive analysis of the USCDC algorithm, including system models, problem formulation, algorithmic details, and simulation results, all from our firsthand perspective in developing and evaluating this approach.
Our research focuses on the core challenge of minimizing total energy consumption in a system comprising multiple drones and a vast number of sensor nodes. We have designed a sophisticated framework that integrates path planning and temporal coordination. The system model details the network architecture, communication links, and energy consumption patterns. The problem is formulated as a non-convex optimization, which we decompose using a hierarchical strategy. The proposed solution encompasses three key algorithms: the UAV Temporal Cooperative Scheduling (UTCS) algorithm for time synchronization, the Improved Ant Colony Optimization (IACO) algorithm for cooperative drone path planning, and the Improved Particle Swarm Optimization (IPSO) algorithm for sink drone trajectory optimization. Simulation results demonstrate that the USCDC algorithm significantly outperforms existing methods, achieving a 52% reduction in path length and a 35% reduction in energy consumption. This validates the effectiveness of our approach, which is a testament to the advanced application of drone technology in this domain.
System Model
Network Model
In our system, the sensor networks are geographically divided into several isolated sub-regions. We deploy a sub-sensor network in each sub-region and assign a cooperative drone to it. The architecture is three-tiered:
- Ground Sensor Network Layer: Sensor nodes within each sub-region are clustered using an algorithm. In each round, cluster heads are elected to collect data from their members. The set of cluster heads in sub-region $o$ is denoted as $Z_o = \{z_{o,1}, z_{o,2}, \dots, z_{o,A}\}$, and the data collected by the $a$-th cluster head is $D_{o,a} = \{D_1, D_2, \dots, D_K\}$. All sensor nodes have the same initial energy $E_n$ and communication radius $R_n$.
- Cooperative Drone Layer: This layer consists of $W$ cooperative drones, $U = \{u_1, u_2, \dots, u_W\}$. Each cooperative drone $u_j$ is responsible for collecting data from the cluster heads in its assigned sub-region. It plans an optimal path to visit all cluster heads. The initial energy, flight altitude, and flight speed of a cooperative drone are $E_c$, $H_c$, and $V_c$, respectively. The use of drone technology at this layer enables efficient, line-of-sight communication with ground nodes.
- Sink Drone Layer: One sink drone $u_s$ is responsible for collecting data from all cooperative drones and transmitting it to the base station. After all cooperative drones finish data collection, the sink drone departs from the base station, visits the cooperative drones in an optimized sequence, and returns to the base station. Its initial energy, flight altitude, and flight speed are $E_s$, $H_s$, and $V_s$, respectively. This hierarchical design is a key innovation in our application of drone technology.
Communication Model
We adopt a line-of-sight (LoS) propagation model for both drone-to-ground and drone-to-drone communications, as drones generally operate at high altitudes with fewer obstacles. The path loss between cooperative drone $u_j$ and cluster head $z_{o,a}$ is given by:
$$l_{LoS}^{u_j, z_{o,a}} = 20 \lg \left( \frac{4\pi f_c d_{u_j, z_{o,a}}}{c} \right) + \eta_{LoS}$$
where $f_c$ is the carrier frequency, $c$ is the speed of light, $d_{u_j, z_{o,a}}$ is the distance, and $\eta_{LoS}$ is the additional path loss for LoS links.
The channel gain $h_{u_j, z_{o,a}}$ is:
$$h_{u_j, z_{o,a}} = 10^{-\frac{l_{LoS}^{u_j, z_{o,a}}}{10}}$$
The average data transmission rate between the cooperative drone and cluster head is:
$$r_{u_j, z_{o,a}} = B \log_2 \left( 1 + \frac{P_t h_{u_j, z_{o,a}}}{N_0 B} \right)$$
where $P_t$ is the transmission power, $B$ is the system bandwidth, and $N_0$ is the noise power spectral density.
Similarly, the path loss between sink drone $u_s$ and cooperative drone $u_j$ is:
$$l_{LoS}^{u_s, u_j} = 20 \lg \left( \frac{4\pi f_c d_{u_s, u_j}}{c} \right) + \eta_{LoS}$$
The channel gain $h_{u_s, u_j}$ and the data transmission rate are:
$$h_{u_s, u_j} = 10^{-\frac{l_{LoS}^{u_s, u_j}}{10}}$$
$$r_{u_s, u_j} = B \log_2 \left( 1 + \frac{P_t h_{u_s, u_j}}{N_0 B} \right)$$
Drone Energy Consumption Model
The energy consumption of a drone is composed of three parts: hovering power, flight power, and communication power. The hovering power $P_h$ is defined as:
$$P_h = \sqrt{\frac{(m_{tot} g)^3}{2\pi \sigma_p^2 \delta_p \rho}}$$
where $m_{tot}$ is the total mass of the drone, $g$ is gravitational acceleration, $\sigma_p$ is the propeller radius, $\delta_p$ is the number of propellers, and $\rho$ is air density.
The horizontal flight power $P_f$ is proportional to flight speed $v_f$:
$$P_f = \frac{P_{\text{max}} – P_{\text{idle}}}{v_{\text{max}}} v_f + P_{\text{idle}}$$
where $v_{\text{max}}$ is the maximum speed, $P_{\text{max}}$ is the hardware power at full speed, and $P_{\text{idle}}$ is the hardware idle power.
Cooperative Drone Energy Consumption
The total energy consumption of cooperative drone $u_j$ in round $x$ is:
$$E_{u_j}^x = E_{u_j, \text{fly}}^x + E_{u_j, \text{coll}}^x + E_{u_j, \text{wait}}^x + E_{u_j, \text{trans}}^x$$
The flight energy is:
$$E_{u_j, \text{fly}}^x = \frac{L_{u_j}^x}{V_f} (P_f + P_h)$$
where $L_{u_j}^x$ is the total flight path length.
The data collection energy is:
$$E_{u_j, \text{coll}}^x = \frac{D_{u_j}^x}{r_{u_j, z_{o,a}}} (P_h + P_{\text{com}})$$
where $D_{u_j}^x$ is the collected data amount and $P_{\text{com}}$ is the communication power.
The waiting energy is:
$$E_{u_j, \text{wait}}^x = t_{\text{wait}}^x P_h$$
where $t_{\text{wait}}^x$ is the waiting time.
The data transmission energy to the sink drone is:
$$E_{u_j, \text{trans}}^x = \frac{D_{u_j}^x}{r_{u_s, u_j}} (P_h + P_{\text{com}})$$
The total energy consumed by all cooperative drones in one round is:
$$E_W^x = \sum_{j=1}^W E_{u_j}^x$$
Sink Drone Energy Consumption
The total energy consumption of the sink drone $u_s$ in round $x$ is:
$$E_{u_s}^x = E_{u_s, \text{fly}}^x + E_{u_s, \text{trans}}^x$$
The flight energy is:
$$E_{u_s, \text{fly}}^x = \frac{S_{u_s}^x}{V_s} P_f$$
where $S_{u_s}^x$ is the total flight path length.
The data transmission energy is:
$$E_{u_s, \text{trans}}^x = \frac{D_{u_s}}{r_{u_s, u_j}} (P_h + P_{\text{com}})$$
The total energy consumed by all drones in one round is:
$$E_{\text{total}}^x = E_W^x + E_{u_s}^x$$
Given $x_{\text{max}}$ rounds, the total energy consumption of the drone swarm is:
$$E_{\text{total}} = \sum_{x=1}^{x_{\text{max}}} E_{\text{total}}^x = \sum_{x=1}^{x_{\text{max}}} (E_W^x + E_{u_s}^x)$$
To summarize the key parameters and models, we present the following table.
| Symbol | Description | Expression/Value |
|---|---|---|
| $P_h$ | Hovering power | $\sqrt{\frac{(m_{tot} g)^3}{2\pi \sigma_p^2 \delta_p \rho}}$ |
| $P_f$ | Flight power | $\frac{P_{\text{max}} – P_{\text{idle}}}{v_{\text{max}}} v_f + P_{\text{idle}}$ |
| $E_{u_j}^x$ | Coop. drone energy in round $x$ | $E_{u_j, \text{fly}}^x + E_{u_j, \text{coll}}^x + E_{u_j, \text{wait}}^x + E_{u_j, \text{trans}}^x$ |
| $E_{u_s}^x$ | Sink drone energy in round $x$ | $E_{u_s, \text{fly}}^x + E_{u_s, \text{trans}}^x$ |
| $r_{u_j, z_{o,a}}$ | Data rate (drone to CH) | $B \log_2 \left( 1 + \frac{P_t h_{u_j, z_{o,a}}}{N_0 B} \right)$ |
| $r_{u_s, u_j}$ | Data rate (sink to coop) | $B \log_2 \left( 1 + \frac{P_t h_{u_s, u_j}}{N_0 B} \right)$ |
| $L_{u_j}^x$ | Coop. drone path length | Optimized by IACO |
| $S_{u_s}^x$ | Sink drone path length | Optimized by IPSO |
Problem Formulation
The objective function of our algorithm is to minimize the total energy consumption of the sink drone and cooperative drones. The formal optimization problem is expressed as:
$$\min E_{\text{total}}$$
subject to the constraints:
$$\text{C1}: \sum_{x=1}^{x_{\text{max}}} E_{u_j}^x \leq E_c, \quad \forall j \in [1, W]$$
$$\text{C2}: \sum_{x=1}^{x_{\text{max}}} E_{u_s}^x \leq E_s$$
$$\text{C3}: p_{u_j}^x(t) \in O_o, \quad \forall j \in [1, W]$$
$$\text{C4}: p_{u_s, \text{end}}^x = p_{u_s, \text{start}}^x, \quad \forall x \in [1, x_{\text{max}}]$$
$$\text{C5}: D_{u_s}^x = D_W^x = \sum_{o=1}^W \sum_{a=1}^A D_{o,a}$$
Constraints C1 and C2 ensure that the total energy consumption of each drone across all rounds does not exceed its initial energy. C3 restricts cooperative drones to their assigned regions. C4 mandates that the sink drone starts and ends at the base station. C5 ensures all cluster head data is collected. This problem is non-convex due to its dependence on multiple variables such as drone positions, speeds, hovering times, and data volumes. To solve it, we adopt a hierarchical optimization strategy, decomposing the problem into two sub-problems: cooperative drone path planning and sink drone trajectory optimization, which are solved sequentially using heuristic algorithms.
Algorithm Description
The USCDC algorithm comprises three core components: the UAV Temporal Cooperative Scheduling (UTCS) algorithm, the Improved Ant Colony Optimization (IACO) algorithm, and the Improved Particle Swarm Optimization (IPSO) algorithm. These components work together to leverage drone technology for efficient data collection.
UAV Temporal Cooperative Scheduling (UTCS) Algorithm
In our two-layer drone architecture, different types of drones operate on different time scales. The UTCS algorithm solves this by establishing a unified time coordinate system with relative-to-absolute time conversion. This algorithm uses an event-triggered scheduling mechanism. The sink drone uses a time reference relative to its departure, while the system operates on absolute time. When a key event occurs (e.g., all cooperative drones complete data collection), the algorithm triggers a scheduling decision. The waiting time is divided into two phases: the first wait is the time after a cooperative drone finishes collection until the sink drone arrives; the second wait is the time after data transmission until the system cycle ends. This refined time management is crucial for the synergy of drone technology components. The departure time of the sink drone is:
$$t_d = \max_{j \in W} t_{u_j}^{\text{coll}}$$
The first waiting time for a cooperative drone is:
$$t_{u_j}^{\text{wait1}} = \max(0, t_{u_j, \text{arr}}^{u_s} – t_{u_j}^{\text{coll}})$$
The second waiting time is:
$$t_{u_j}^{\text{wait2}} = \max(0, t_{\text{sys, end}} – t_{u_j}^{\text{trans, end}})$$
The UTCS algorithm is outlined in the following steps.
| Step | Action | Description |
|---|---|---|
| 1 | Initialize | Initialize start times for all cooperative drones $U$ and a time point dictionary $tp$ for the sink drone. |
| 2-5 | Compute collection end times | For each $u_j$ in $U$, calculate its data collection completion time $t_{u_j}^{\text{coll}}$. |
| 6 | Set sink departure time | Calculate $t_d = \max_j t_{u_j}^{\text{coll}}$. |
| 7-12 | Sink visit phase | For each $u_j$ in $U$, calculate sink arrival time $t_{u_j, \text{arr}}^{u_s}$, data transmission end time $t_{u_j}^{\text{trans, end}}$, and first waiting time $t_{u_j}^{\text{wait1}}$. |
| 13 | Compute system end time | Calculate $t_{\text{sys, end}}$, the time the sink returns to base. |
| 14-17 | Compute second wait times | For each $u_j$ in $U$, calculate the second waiting time $t_{u_j}^{\text{wait2}}$. |
| 18 | Return | Return the system time chain $t_{ch}$ and total task time $t_{\text{total}}$. |
Improved Ant Colony Optimization (IACO) for Cooperative Drone Path Planning
For cooperative drone path planning, we introduce the IACO algorithm. This algorithm enhances the classic ant colony optimization by incorporating an elite ant strategy and a position state transition mechanism. The probability that ant $k$ moves from node $a$ to node $b$ is:
$$p_{ab}^k(t) = \begin{cases} \frac{\tau_{ab}^\alpha(t) \epsilon_{ab}^\beta(t)}{\sum_{i \in \text{allowed}_k} \tau_{ab}^\alpha(t) \epsilon_{ab}^\beta(t)}, & \text{if } i \in \text{allowed}_k \\ 0, & \text{otherwise} \end{cases}$$
The pheromone concentration update is:
$$\tau_{ab}(t+1) = (1-\mu) \tau_{ab}(t) – \Delta \tau_{ab}(t, t+1)$$
With the elite ant strategy, the pheromone increment is modified to:
$$\Delta \tau_{ab}^k = \begin{cases} (1+\xi) Q / C_k, & \text{if } k \in \lambda, (a,b) \in F_k \\ Q / C_k, & \text{if } (a,b) \in F_k \\ 0, & \text{otherwise} \end{cases}$$
The position state transition mechanism for multi-round missions is defined as:
$$T_{\text{start}}^x = T_{\text{end}}^{x-1} = (X_{\text{end}}^{x-1}, Y_{\text{end}}^{x-1}, H_c)$$
This mechanism leverages the continuous nature of drone technology tasks to reduce redundant travel. The IACO algorithm proceeds as follows.
| Step | Action | Description |
|---|---|---|
| 1-3 | Initialize | Set population $n_{\text{ant}}$, max iterations $M$, pheromone factors $\alpha, \beta$, and evaporation rate $\mu$. Initialize pheromone $\tau$ and heuristic $\epsilon$. |
| 4-8 | Position state transition | If first round, set start at base. Otherwise, set start from previous round’s end position. |
| 9-14 | Path construction | For each ant $k$, build a complete path using probability $p_{ab}^k$, and compute path length and energy. |
| 15-18 | Pheromone evaporation and deposit | Evaporate pheromone globally, then each ant deposits pheromone on its path. |
| 19-22 | Elite ant reinforcement | Select the top $\lambda n_{\text{ant}}$ elite ants and add extra pheromone to their paths. |
| 23-24 | Iterate and return | Repeat until max iterations, then return the best path $G_{\text{best}}$, length $L_{\text{best}}$, and energy $E_{\text{best}}$. |
Improved Particle Swarm Optimization (IPSO) for Sink Drone Trajectory Optimization
For the sink drone, we propose the IPSO algorithm, which features adaptive parameter adjustment and bi-objective optimization. The velocity update formula for particle $i$ is:
$$v_i = \omega v_i + \theta_1 \text{rand}() (\phi_{\text{best}}^i – \iota_i) + \theta_2 \text{rand}() (\varphi_{\text{best}} – \iota_i)$$
We use an adaptive inertia weight:
$$\omega = \omega_{\text{max}} – (\omega_{\text{max}} – \omega_{\text{min}}) \kappa_{\text{iter}} / M$$
The bi-objective fitness function is:
$$\psi = w_1 \frac{d}{d_{\text{max}}} + w_2 \frac{e}{e_{\text{max}}}$$
This function normalizes path length $d$ and energy $e$ to balance optimization goals. The IPSO algorithm steps are listed below.
| Step | Action | Description |
|---|---|---|
| 1-6 | Initialize | Set population $n_{\text{part}}$, max iterations $M$, learning factors $\theta_1, \theta_2$, inertia bounds $\omega_{\text{min}}, \omega_{\text{max}}$. Initialize particle positions $\iota_i$, velocities $v_i$, and global best $\varphi_{\text{best}}$. |
| 7-24 | Iterate | For each iteration $\kappa_{\text{iter}}$: update $\omega$; for each particle $i$, construct a full visit sequence, compute $d_i$, $e_i$, and fitness $\psi_i$; update individual and global bests; update velocity $v_i$ and position $\iota_i$. |
| 25 | Return | Return the best path $G_{\text{best}}$, length $S_{\text{best}}$, and energy $E_{\text{best}}$. |
Simulation Results and Analysis
We conducted simulations to validate the performance of the USCDC algorithm, comparing it with the UAV Swarm Parallel Data Collection (USPDC) algorithm. The simulations were performed in a PyCharm environment, with 100 independent runs to ensure reliability. The simulation parameters are summarized in the table below.
| Parameter | Value | Parameter | Value |
|---|---|---|---|
| Sub-regions $W$ | 6 | Node initial energy $E_n$ | 1 J |
| Sub-region radius $R_r$ | 200 m | Node communication radius $R_n$ | 50 m |
| Sink drone initial energy $E_s$ | 2,700,000 J | Sink drone altitude $H_s$ | 75 m |
| Sink drone speed $V_s$ | 25 m/s | Coop. drone initial energy $E_c$ | 1,500,000 J |
| Coop. drone altitude $H_c$ | 50 m | Coop. drone 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 | Ants $n_{\text{ant}}$ | 100 |
| $\alpha$ (pheromone) | 1.5 | $\beta$ (heuristic) | 3.0 |
| $\mu$ (evaporation) | 0.6 | $Q$ (intensity) | 100 |
| Particles $n_{\text{part}}$ | 100 | $\omega_{\text{max}}$ | 0.95 |
| $\omega_{\text{min}}$ | 0.4 | $\theta_1, \theta_2$ | 2.05 |
Path Length Comparison
We compared the total path length per round for both algorithms over 10 rounds. The results show that the USCDC algorithm has a total path length of approximately 10 km, while the USPDC algorithm requires about 21 km. This represents a 52.4% reduction in path length. This significant improvement is due to the hierarchical architecture, where cooperative drones only operate locally, and the sink drone handles long-distance data delivery, avoiding redundant round trips to the base station. This is a direct benefit of our advanced drone technology deployment.
Energy Consumption and Efficiency
The energy consumption per round is also compared. The USPDC algorithm consumes approximately $13.5 \times 10^4$ J per round, while the USCDC algorithm consumes only about $8.5 \times 10^4$ J, a 37% reduction on average. The energy efficiency, measured in bits per joule, is 10 bit/J for USCDC and 6.5 bit/J for USPDC. This represents a 53.8% improvement in energy efficiency. The hierarchical collaboration and optimized path planning are the main contributors to these gains, showcasing the effective use of drone technology to minimize energy waste.
Scalability Analysis
We evaluated the algorithms in three different WSN scales: small (1000 m x 1000 m, 3 sub-regions), medium (2000 m x 2000 m, 6 sub-regions), and large (3000 m x 3000 m, 12 sub-regions). The results for total path length and total energy consumption are summarized in the table below.
| Scale | Metric | USPDC | USCDC | Improvement |
|---|---|---|---|---|
| Small | Path Length (km) | 10.5 | 7.6 | 28% |
| Small | Energy (10^5 J) | 4.5 | 3.6 | 21% |
| Medium | Path Length (km) | 21.0 | 10.0 | 52% |
| Medium | Energy (10^5 J) | 13.5 | 8.5 | 35% |
| Large | Path Length (km) | 50.0 | 16.8 | 66.3% |
| Large | Energy (10^5 J) | 30.0 | 18.0 | 40% |
The results demonstrate that the USCDC algorithm’s advantage increases with network scale. In large networks, the path length is reduced by 66.3% and energy consumption by 40%. The reduction in energy is less pronounced than path length due to additional waiting and hovering energy in the hierarchical scheme, but the overall trend confirms the excellent scalability of our approach.
Conclusion
In this paper, we present the USCDC algorithm, a novel solution for energy-efficient data collection in large-scale wireless sensor networks. By employing a two-layer drone architecture and advanced heuristic algorithms for path planning and temporal coordination, our method significantly reduces path redundancy and energy consumption. The integration of the UTCS, IACO, and IPSO algorithms demonstrates a sophisticated application of drone technology that is highly effective and scalable. Simulation results confirm that the USCDC algorithm achieves up to a 52% reduction in path length and a 35% reduction in energy consumption compared to existing methods. Future work will focus on optimizing the sink drone’s visit strategy to minimize waiting times and extending the algorithm to continuously distributed sensor networks, further advancing the role of drone technology in IoT data acquisition.