In recent years, the integration of unmanned aerial vehicles (UAVs) as mobile data mules or aerial base stations within wireless sensor networks (WSNs) has emerged as a transformative paradigm. This is particularly relevant for large-scale environmental monitoring, precision agriculture, and disaster area surveillance, where ground sensor nodes are often sparsely deployed, energy-constrained, and possess limited communication ranges. The inherent mobility and line-of-sight advantage of China UAV drones offer a powerful solution to extend network lifetime, improve data collection reliability, and provide on-demand coverage. However, effectively planning the flight paths for one or multiple UAVs in complex three-dimensional terrains to maximize information gain while minimizing operational costs like time and energy consumption presents a significant multi-objective optimization challenge. Traditional path planning methods often struggle with the dynamic nature of sensory data value and the intricate constraints of 3D environments.
This paper addresses the critical problem of collaborative, multi-objective path planning for China UAV drones within a WSN. We formulate the UAV’s trajectory not merely as a geometric path but as a sequence of optimal sensing positions. At each step, a UAV must decide its next 3D location to efficiently harvest valuable data from ground sensors, considering that the utility of sensed information decays after collection and then recovers over time. To solve this complex problem, we propose a comprehensive strategy centered around a bio-inspired Particle Swarm Optimization (PSO) algorithm. Our approach systematically models the sensor field, defines a multi-purpose fitness function encompassing information value, time, and energy, and employs a constrained PSO routine to derive optimal UAV waypoints. The performance of the proposed system is validated through extensive simulations, demonstrating its effectiveness and superiority over baseline methods.
Related Work and Problem Context
The domain of UAV path planning, especially within the context of WSNs, has been extensively studied, yielding various methodologies ranging from classical optimization to modern machine learning techniques. Early approaches often simplified the problem to coverage path planning or Traveling Salesman Problem (TSP) variants, focusing on visiting all points of interest. More recent works have incorporated dynamic elements and multiple objectives. For instance, some studies employ genetic algorithms or ant colony optimization to find energy-efficient paths. Others utilize reinforcement learning, particularly deep Q-networks (DQN), to enable UAVs to learn optimal policies in unknown environments. A significant limitation in many existing works is the treatment of the sensing field as a 2D plane or a set of discrete points, neglecting the crucial impact of 3D topography, antenna models, and time-varying data value. Furthermore, ensuring persistent communication connectivity among a fleet of China UAV drones and a ground control station during the data harvesting mission adds another layer of complexity often overlooked. Our work distinctively integrates a realistic 3D terrain-aware communication model with a time-dependent information value function, and solves the resulting high-dimensional optimization problem using an efficient PSO variant tailored for multi-UAV systems.
System Model Formulation
The proposed system involves a collaborative network where multiple UAVs operate in a designated 3D airspace to collect data from a static WSN deployed over a geographically complex region. A central ground station orchestrates the mission by computing optimal waypoints and disseminating them to the UAV fleet.
2.1 3D Sensor Field and Grid-Based Representation
We consider a three-dimensional sensor field spanning $$[0, R_x] \times [0, R_y] \times [0, R_z]$$ in Cartesian coordinates. The terrain is not flat; each (x, y) coordinate has a corresponding terrain height. To make the problem computationally tractable, we discretize the (x, y) plane into $$N_C$$ uniform grid cells. Each cell $$C_i$$ is defined by its center coordinates $$(x_i, y_i)$$ and its surface height $$z_i = f_{terrain}(x_i, y_i)$$. The cell size $$S_C$$ is given by:
$$S_C = \frac{R_x \times R_y}{N_C}$$
Within this field, a heterogeneous WSN is deployed. Let $$G = \{g_1, g_2, …, g_{N_g}\}$$ denote the set of $$N_g$$ different sensor types (e.g., temperature, humidity, acoustic). The deployment is non-uniform; each grid cell $$C_i$$ may contain a certain number of sensors of each type. We define $$N_{g_j}^{C_i}$$ as the number of type-$$g_j$$ sensors located in cell $$C_i$$. The following table exemplifies the sensor distribution for a subset of cells:
| Grid Cell $$C_i$$ | Coordinates (x, y, z) | Number of Temp. Sensors ($$g_1$$) | Number of Humidity Sensors ($$g_2$$) | Other Sensors… |
|---|---|---|---|---|
| $$C_1$$ | (150, 300, 25) | 3 | 1 | … |
| $$C_2$$ | (155, 305, 28) | 0 | 2 | … |
| $$C_3$$ | (160, 310, 22) | 1 | 0 | … |
2.2 UAV-Assisted Time-Varying Data Harvesting Model
The core objective of a China UAV drone is to collect valuable sensory information. A key insight is that the value of data from a sensor is not static. Immediately after collection, the information’s utility drops, as it represents a known state. Over time, as the environment changes, new measurements become valuable again. We model this time-dependent value $$V_{g_j}^{C_i}(t)$$ for sensor type $$g_j$$ in cell $$C_i$$ at time $$t$$ using an exponential recovery function:
$$
V_{g_j}^{C_i}(t) =
\begin{cases}
a \cdot \exp(t – t_{g_j}^{C_i}) + b, & \text{if } t – t_{g_j}^{C_i} \leq R_{g_j} \\
V_{g_j}^{max}, & \text{otherwise}
\end{cases}
$$
where:
- $$t_{g_j}^{C_i}$$ is the last time data from type $$g_j$$ sensors in cell $$C_i$$ was collected by any UAV.
- $$R_{g_j}$$ is the recovery time constant for sensor type $$g_j$$.
- $$V_{g_j}^{max}$$ and $$V_{g_j}^{min}$$ are the maximum and minimum value indices.
- Constants $$a$$ and $$b$$ are calculated as: $$a = \frac{V_{g_j}^{max} – V_{g_j}^{min}}{e^{R_{g_j}} – 1}$$ and $$b = V_{g_j}^{min} – a$$.
Furthermore, the total value obtainable from a cell also depends on the density of sensors. A diminishing returns effect is modeled via a sensor quantity function $$f_{g_j}^{ns}(N)$$:
$$
f_{g_j}^{ns}(N_{g_j}^{C_i}) =
\begin{cases}
0, & N_{g_j}^{C_i} < 1 \\
\frac{(N_{g_j}^{max} – N_{g_j}^{min})N_{g_j}^{C_i} + N_{g_j}^{max}(N_{g_j}^{min} – 1)}{N_{g_j}^{max} – 1}, & 1 \leq N_{g_j}^{C_i} \leq N_{g_j}^{max} \\
N_{g_j}^{max}, & N_{g_j}^{C_i} > N_{g_j}^{max}
\end{cases}
$$
Here, $$N_{g_j}^{min}$$ and $$N_{g_j}^{max}$$ are thresholds defining the linear saturation of value with sensor count.
2.3 3D Communication and Coverage Model
For a China UAV drone located at $$U_k = (x_k^u, y_k^u, z_k^u)$$ to collect data from a ground sensor in cell $$C_i$$, a line-of-sight (LoS) communication link must exist. We assume UAVs are equipped with directional antennas for downlink sensing (to focus energy) and omnidirectional antennas for inter-UAV communication. The received power $$P_r$$ at a distance $$d$$ under free-space path loss is:
$$P_r = \frac{\lambda^2 G_r G_t}{(4\pi)^2 d^2} P_t$$
where $$\lambda$$ is wavelength, $$G_r$$ and $$G_t$$ are receive/transmit gains, and $$P_t$$ is transmit power. The maximum communication distance $$d_{max}^{U \leftrightarrow S}$$ between a UAV and a ground sensor is determined by the minimum decodable power $$P_r^{min}$$:
$$d_{max}^{U \leftrightarrow S} = \sqrt{ \frac{\lambda^2 G_r G_t P_t^{U \rightarrow S}}{(4\pi)^2 P_r^{min}} }$$
A similar equation defines $$d_{max}^{U \leftrightarrow U}$$ for UAV-to-UAV links. A ground cell $$C_i$$ is considered within the coverage of UAV $$U_k$$ only if: 1) The distance $$d_{k-i} \leq d_{max}^{U \leftrightarrow S}$$, 2) The cell lies within the directional antenna’s beamwidth $$\theta$$, and 3) An unobstructed LoS exists, meaning no terrain point between them has a height greater than the line connecting the UAV and the cell center.
PSO-Based Decision and Optimization Algorithm
The decision-making core of our strategy is a multi-objective fitness function optimized using a constrained PSO algorithm to determine the next best position for each China UAV drone.
3.1 Multi-Objective Fitness Function
When a UAV at its $$m$$-th position plans its $$(m+1)$$-th waypoint, it evaluates candidate positions based on three competing objectives: Maximizing Sensed Value, Minimizing Operation Time, and Minimizing Energy Consumption. These are combined into a single aggregate fitness function $$F$$:
$$F = \omega_S F_S + \omega_T F_T + \omega_E F_E$$
where $$\omega_S, \omega_T, \omega_E$$ are weighting coefficients that prioritize the objectives.
1. Sensing Value Fitness ($$F_S$$): This reflects the total time-varying information value available within the UAV’s coverage area $$COV_k$$ at the estimated arrival time, excluding cells recently covered by other UAVs.
$$
F_S = \frac{ \sum_{\forall C_i \in COV_k, \ C_i \notin COV_{\forall l \neq k}} \sum_{j=1}^{N_g} V_{g_j}^{C_i}(t + \hat{T}_M^k(m+1)) \cdot f_{g_j}^{ns}(N_{g_j}^{C_i}) }{NOR_S}
$$
Here, $$\hat{T}_M^k(m+1)$$ is the estimated movement time to the candidate position, and $$NOR_S$$ is a normalization factor.
2. Time Cost Fitness ($$F_T$$): This penalizes the total time required for movement and data harvesting at the new position.
$$
F_T = – \frac{ [\hat{T}_M^k(m+1) + \hat{T}_S^k(m+1)] }{NOR_T}
$$
$$\hat{T}_S^k(m+1)$$ is the estimated data collection time, which depends on the number of sensors and the transmission protocol’s packet time $$T_{g_j}^{pkt}$$ and success probability $$p_{g_j}^{C_i}$$:
$$\hat{T}_S^k(m+1) = \sum_{\forall C_i \in COV_k} \sum_{j=1}^{N_g} \left( N_{g_j}^{C_i} \times T_{g_j}^{pkt} \times \frac{1}{1 – p_{g_j}^{C_i}} \right)$$
3. Energy Cost Fitness ($$F_E$$): This penalizes the energy consumed for movement $$\hat{E}_M^k$$, hovering for collection $$\hat{E}_S^k$$, and communication $$\hat{E}_C^k$$.
$$
F_E = – \frac{ [\hat{E}_M^k(m+1) + \hat{E}_S^k(m+1) + \hat{E}_C^k(m+1)] }{NOR_E}
$$
Each energy component is modeled based on UAV dynamics and communication power consumption.
3.2 Constrained Particle Swarm Optimization
We employ PSO to find the candidate position (a particle in 3D space) that maximizes the fitness function $$F$$. Each particle’s position represents a potential next waypoint $$U = (x, y, z)$$ for a UAV. The standard PSO velocity and position update equations for particle $$p$$ in dimension $$i$$ at iteration $$q$$ are:
$$
v_{pi}^{q} = w v_{pi}^{q-1} + c_1 r_1 (pbest_{pi}^{q-1} – u_{pi}^{q-1}) + c_2 r_2 (gbest_{i}^{q-1} – u_{pi}^{q-1})
$$
$$
u_{pi}^{q} = u_{pi}^{q-1} + v_{pi}^{q}
$$
where $$w$$ is inertia weight, $$c_1, c_2$$ are acceleration coefficients, and $$r_1, r_2$$ are random numbers. $$pbest$$ is the particle’s personal best position, and $$gbest$$ is the swarm’s global best position.
The search is not unconstrained. We impose several practical constraints on the particle positions to ensure feasible and safe operation for the China UAV drone fleet:
| Constraint # | Description | Mathematical Formulation |
|---|---|---|
| 1 | Stay within mission area. | $$U \in [0, R_x] \times [0, R_y] \times [0, R_z]$$ |
| 2 | Align (x,y) with grid centers, fly above terrain. | $$(u_x, u_y) \leftarrow (x_i, y_i), \quad u_z > z_i$$ |
| 3 | No collision with other UAVs. | $$U \neq U_l \quad \forall l \neq k$$ |
| 4 | Avoid no-fly zones. | $$U \notin \{PZ_1, PZ_2, …, PZ_{N_{PZ}}\}$$ |
| 5 | Maintain network connectivity: At least one UAV must connect to the ground station. | $$\exists k \ | \ d(U_k, GS) \leq d_{max}^{U \leftrightarrow S}$$ |
| 6 | Maintain swarm connectivity: Each UAV must connect to at least one other UAV. | $$\forall k, \ \exists l \neq k \ | \ d(U_k, U_l) \leq d_{max}^{U \leftrightarrow U}$$ |
| 7 | Respect energy & time budget for safe return. | $$\hat{T}_M^k + \hat{T}_S^k + \hat{T}_R^k < T_{max}^k$$ $$\hat{E}_M^k + \hat{E}_S^k + \hat{E}_C^k + \hat{E}_R^k < E_{max}^k$$ |
Our algorithm evaluates the fitness $$F$$ for a particle only if it satisfies all relevant constraints. The PSO algorithm efficiently navigates this complex, constrained search space to find high-utility waypoints for the China UAV drones.
Simulation Results and Performance Analysis
We developed a comprehensive simulation environment in Python to validate the proposed PSO-based 3D path planning strategy. The sensor field was set to $$2km \times 2km \times 1km$$ with realistic mountainous terrain generated using fractal algorithms. A total of 20,000 sensors of 3 different types were non-uniformly deployed. A fleet of 5 China UAV drones was simulated, starting from a ground station at (0,0,0). Key simulation parameters are summarized below:
| Parameter | Value |
|---|---|
| Field Dimensions ($$R_x, R_y, R_z$$) | 2000 m, 2000 m, 1000 m |
| Number of Grid Cells ($$N_C$$) | 10,000 |
| Total Number of Sensors | 20,000 |
| Number of Sensor Types ($$N_g$$) | 3 |
| Number of China UAV Drones | 5 |
| UAV Max Speed ($$v_u$$) | 20 m/s |
| UAV Initial Energy | 400 kJ |
| UAV Max Operational Time | 360 s |
| PSO Particles per Decision | 5 |
| PSO Iterations per Decision | 50 |
| Fitness Weights ($$\omega_S, \omega_T, \omega_E$$) | 0.7, 0.15, 0.15 |
4.1 Convergence and Effectiveness of PSO
The first experiment analyzes the convergence property of our PSO optimizer. We tracked the global best fitness value over 50 iterations for different swarm sizes (3, 5, and 10 particles). The results show that the proposed method converges rapidly towards a high-quality solution. While a larger swarm size (10 particles) converges slightly faster, even a small swarm of 5 particles—as used in our main simulations—achieves over 95% of the final fitness within 15-20 iterations. This demonstrates the efficiency of our PSO formulation, making it suitable for real-time or near-real-time replanning for China UAV drones. The computational cost of 250 fitness evaluations (5 particles * 50 iterations) per UAV movement is significantly lower than a brute-force global search, which would require evaluations at all potential grid points (tens of thousands).
4.2 Impact of Key System Parameters
Varying the Number of UAVs: We evaluated system performance with 1 to 5 drones. The cumulative sensing fitness $$F_S$$ increased linearly with more drones, as a larger area could be covered concurrently. However, the total fitness $$F$$, which includes time and energy penalties, peaked at 3 drones for our specific scenario. Beyond this, the added coordination overhead, longer paths to avoid interference, and increased communication energy began to outweigh the gains in information collection. This highlights the importance of our multi-objective optimization in determining the cost-effective fleet size.
Varying Sensor Density: The performance was tested with total sensor counts of 1,000, 5,000, 10,000, and 20,000. As expected, the cumulative sensing fitness $$F_S$$ scaled directly with sensor density. Crucially, the time and energy fitness components ($$F_T$$, $$F_E$$) remained relatively stable, as the planning algorithm focuses on coverage geometry and UAV dynamics, which are less affected by the number of data points within a covered cell. This shows the robustness of our China UAV drone path planner across different WSN deployment densities.
4.3 Comparative Analysis with Benchmark Methods
To establish a baseline for comparison, we implemented four alternative strategies:
- Random Position: UAV moves to a random valid location maintaining connectivity.
- Global Search: Exhaustively evaluates all discrete candidate positions (22,832 points) and selects the best.
- Optimal Then Hover: Uses PSO once to find an initial optimal point; the UAV then hovers there.
- Random Then Hover: UAV moves to a random initial point and hovers.
The cumulative fitness values at the end of the mission for all methods are compared in the following analysis. Our proposed PSO-based method achieves a cumulative sensing fitness value within 95% of the theoretically optimal but computationally prohibitive Global Search method. It outperforms the Random Position method by over 80-fold in sensing value gained. The Optimal Then Hover strategy performs better than random strategies but fails over time as it doesn’t adapt to the changing information value field. Our dynamic, step-by-step PSO replanning strategy for China UAV drones effectively balances the need to explore new high-value areas with the costs of movement, demonstrating clear superiority in the multi-objective context. The ability to maintain persistent network connectivity throughout all maneuvers was successfully validated for our method and the Global Search, while random methods occasionally failed this constraint.
Conclusion and Future Work
In this paper, we have presented a comprehensive framework for multi-objective, three-dimensional path planning of China UAV drones within wireless sensor networks. By modeling the sensor field as a grid with time-varying information value and realistic 3D communication constraints, we formulated a sophisticated optimization problem. The core of our solution is a multi-purpose fitness function that quantifies the trade-off between information gain, mission time, and energy expenditure. A constrained Particle Swarm Optimization algorithm was developed to efficiently solve for the optimal UAV waypoints that maximize this fitness while adhering to practical flight and network constraints.
Simulation results confirm the efficacy of our approach. The PSO optimizer demonstrates rapid convergence, making it suitable for dynamic mission planning. Our method achieves performance nearly equivalent to an exhaustive global search but with a fraction of the computational cost. It significantly outperforms naive random strategies and static hovering approaches by dynamically adapting to the evolving state of the WSN. The successful integration of 3D terrain, directional antenna models, and time-dependent data value represents a step forward in realistic UAV path planning for environmental monitoring and data harvesting applications.
Future work will focus on several extensions. First, we plan to implement a fully decentralized version of the algorithm where China UAV drones collaboratively make decisions using distributed PSO or consensus algorithms, reducing reliance on the ground station. Second, incorporating real-time sensor fault models and dynamic obstacle avoidance (e.g., other air traffic) will enhance robustness. Finally, field experiments with physical China UAV drone platforms and sensor nodes are planned to validate the simulation results in real-world scenarios, further bridging the gap between theoretical models and practical deployment.