Emergency communication networks face severe challenges due to unpredictable disasters, such as earthquakes, floods, typhoons, and fire accidents, as well as sudden high-density user demands during large-scale public events. Traditional terrestrial base stations are not always reliable in such scenarios because of infrastructure damage or insufficient capacity. In this context, unmanned aerial vehicles have emerged as a promising technology to provide flexible, rapid, and cost-effective wireless connectivity. Equipped with miniature base stations, unmanned aerial vehicles can be deployed to serve ground users, collect data from wireless sensors, and establish temporary communication links. However, the limited onboard battery capacity of unmanned aerial vehicles remains one of the most critical constraints for their practical applications. Therefore, optimizing the deployment and resource allocation of unmanned aerial vehicles to maximize energy efficiency is of great significance.
In this paper, we study energy-efficient deployment optimization for unmanned aerial vehicles acting as aerial base stations. Two typical emergency communication scenarios are considered. The first scenario involves a single unmanned aerial vehicle that sequentially visits multiple hovering points to collect data from ground wireless devices. The second scenario involves multiple unmanned aerial vehicles deployed cooperatively to provide temporary communication services to ground users with high traffic demands. For both scenarios, we formulate optimization problems that jointly consider device association, three-dimensional placement, bandwidth allocation, and power allocation. The objective is to improve coverage efficiency and reduce total energy consumption while satisfying quality-of-service constraints. We design efficient algorithms based on particle swarm optimization and whale optimization techniques. Simulation results demonstrate that the proposed schemes significantly outperform baseline methods in terms of coverage utility and energy consumption.
1 Introduction
Natural disasters often cause severe damage to terrestrial communication infrastructure. For instance, earthquakes can destroy base stations and fiber links, while floods may interrupt power supply and data transmission. According to statistical reports, China suffers from diverse natural disasters every year, including earthquakes in Sichuan and Yunnan, typhoons in coastal provinces, and floods along the Yangtze River. In such emergencies, rescue teams need timely and reliable communication to coordinate operations and collect information. Traditional emergency communication methods, such as satellite communication and emergency communication vehicles, have limitations in terms of bandwidth, latency, and accessibility. Unmanned aerial vehicles offer a promising alternative because they can fly over damaged areas, carry communication payloads, and establish temporary networks quickly.
With the rapid development of 5G/6G networks, the Internet of Things, and artificial intelligence, unmanned aerial vehicles have been widely used in various fields, including disaster management, surveillance, smart agriculture, logistics, and public safety. In particular, UAV-mounted base stations, also known as aerial base stations, can provide on-demand coverage to ground users. Compared with terrestrial base stations, unmanned aerial vehicles can adjust their positions dynamically to improve channel quality and reduce transmission distance. They can also be deployed in a layered manner with terrestrial and satellite networks to form a heterogeneous communication architecture.
However, unmanned aerial vehicles are typically powered by batteries with limited energy density. The endurance of a commercial rotary-wing unmanned aerial vehicle ranges from tens of minutes to a few hours, depending on payload and flight conditions. Communication payloads, such as base stations, further increase power consumption and reduce endurance. Therefore, energy-efficient deployment is essential to prolong the operational time of unmanned aerial vehicles and ensure sustainable communication services. Many recent studies have focused on optimizing the trajectory, placement, and resource allocation of unmanned aerial vehicles to minimize energy consumption while maximizing coverage and throughput.
Previous research on UAV-assisted communications can be broadly categorized into three types: UAV relays, UAV data collection and dissemination, and UAV-mounted base stations. As relay nodes, unmanned aerial vehicles can extend the communication range between ground devices and base stations. As data collection platforms, unmanned aerial vehicles can fly close to wireless sensors to gather data efficiently. As aerial base stations, unmanned aerial vehicles can provide direct wireless access to ground users in emergency areas or high-traffic events. Most existing works assume that perfect channel state information is available and that the spectrum resources are continuous. However, in practice, resource blocks are discrete and channel conditions are complicated. Moreover, many existing studies neglect the difference in traffic demands among ground users, which may lead to inefficient resource allocation.
In this paper, we address the above limitations by proposing energy-efficient deployment algorithms for two practical scenarios. Our contributions are summarized as follows:
- We investigate a single-UAV data collection scenario where the UAV hovers at multiple points to collect data from ground devices. A joint optimization problem involving device association, hovering positions, and bandwidth allocation is formulated. We propose a particle swarm optimization algorithm with K-means initialization to solve the device association subproblem and a two-stage optimization algorithm based on particle swarm optimization to minimize UAV energy consumption.
- We investigate a multi-UAV deployment scenario where multiple UAVs provide temporary communication services to ground users with high traffic demands. A traffic-priority deployment strategy is proposed to minimize the number of UAVs while satisfying a traffic coverage ratio. Furthermore, a joint bandwidth, power, and three-dimensional position optimization algorithm is developed to minimize the total energy consumption of the system.
- We conduct extensive simulations to evaluate the performance of the proposed algorithms. Results show that our algorithms achieve higher coverage utility and lower energy consumption compared with baseline schemes such as K-means, standard particle swarm optimization, differential evolution, and artificial hummingbird algorithm.
2 System Model and Problem Formulation
In this section, we introduce the general system model for UAV-assisted emergency communication networks. We consider a set of ground wireless devices or users randomly distributed in a target area. Let \(\mathcal{N} = \{1, 2, \dots, N\}\) denote the set of ground nodes. The location of node \(n\) is given by \(\mathbf{l}_n^{GU} = (x_n, y_n, 0)\). We consider a set of rotary-wing unmanned aerial vehicles, denoted by \(\mathcal{M} = \{1, 2, \dots, M\}\), which are deployed as aerial base stations. The three-dimensional position of the \(m\)-th UAV is denoted by \(\mathbf{l}_m^U = (x_m^U, y_m^U, h_m^U)\). Each UAV is equipped with a directional antenna with half-power beamwidth \(2\theta_B\). The antenna gain can be approximated by
$$
G = \begin{cases}
G_{3dB}, & -\theta_B \le \phi \le \theta_B, \; -\theta_B \le \psi \le \theta_B, \\
g, & \text{otherwise},
\end{cases}
$$
where \(G_{3dB} \approx \frac{29000}{(2\theta_B)^2}\) is the main-lobe gain, \(\phi\) and \(\psi\) are the azimuth and elevation angles, and \(g\) is the side-lobe gain which is assumed to be negligible.
The air-to-ground channel between a UAV and a ground node is modeled as a line-of-sight (LoS) dominated link. The path loss depends on the distance and the environment. For a UAV at height \(h\) and a ground node at horizontal distance \(r\), the elevation angle is \(\theta = \frac{180}{\pi} \arctan(h/r)\). The probability of LoS is given by
$$
P_{LoS}(\theta) = \frac{1}{1 + a \exp(-b (\theta – a))},
$$
where \(a\) and \(b\) are environment-dependent parameters. The path loss for LoS and NLoS links can be expressed as
$$
PL_{LoS} = 20 \log_{10} \left(\frac{4\pi f_c d}{c}\right) + \eta_{LoS},
$$
$$
PL_{NLoS} = 20 \log_{10} \left(\frac{4\pi f_c d}{c}\right) + \eta_{NLoS},
$$
where \(f_c\) is the carrier frequency, \(d\) is the distance between the UAV and the node, \(c\) is the speed of light, and \(\eta_{LoS}\) and \(\eta_{NLoS}\) are excessive losses. For simplicity, many studies assume a LoS channel with free-space path loss, which is reasonable when the UAV flies high enough. In this paper, we adopt the following channel gain model:
$$
h_{mn} = \frac{\beta_0}{\| \mathbf{l}_m^U – \mathbf{l}_n^{GU} \|^2},
$$
where \(\beta_0\) is the path loss at a reference distance of 1 meter. The achievable downlink rate from UAV \(m\) to node \(n\) is given by
$$
R_{mn} = B_{mn} \log_2 \left(1 + \frac{p_{mn} G h_{mn}}{B_{mn} \sigma_n^2} \right),
$$
where \(B_{mn}\) is the bandwidth allocated by UAV \(m\) to node \(n\), \(p_{mn}\) is the transmission power, and \(\sigma_n^2\) is the noise power spectral density.
The total energy consumption of a rotary-wing UAV consists of communication energy and propulsion energy. The communication energy is usually much smaller than the propulsion energy. According to the power consumption model in the literature, the horizontal flight power \(P_{for}\), hovering power \(P_{hov}\), vertical ascent power \(P_{asc}\), and vertical descent power \(P_{des}\) can be expressed as functions of the UAV’s velocity and physical parameters. For example, when the UAV flies horizontally at constant speed \(V\), the flight power is
$$
P_{for}(V) = c_1 c_2 (mg)^{3/2} + c_3 V^3,
$$
and the hovering power is
$$
P_{hov} = c_1 c_2 (mg)^{3/2},
$$
where \(m\) is the mass of the UAV, \(g\) is the gravitational acceleration, and \(c_1, c_2, c_3\) are constants related to the UAV’s aerodynamics. The total energy consumption of a UAV completing a mission can be modeled as the sum of energy consumed during vertical takeoff, horizontal flight, hovering, and data transmission.
3 Energy-Efficient Trajectory Optimization for Single-UAV Data Collection
3.1 Scenario Description
In the first scenario, we consider a post-disaster area where a set of ground wireless devices (WDs) are deployed for environmental monitoring and rescue assistance. A single rotary-wing unmanned aerial vehicle is dispatched from a base station at the origin. The UAV flies at a fixed altitude \(H\) and visits a set of \(M\) hovering points sequentially. At each hovering point, it serves a subset of WDs that are within its maximum coverage radius \(r_{\max}\). After collecting all data, the UAV returns to the starting point. The objective is to maximize the data collection rate (i.e., coverage utility) and minimize the total energy consumption.
Let \(q_n\) denote the data amount to be collected from WD \(n\). The coverage utility is defined as the ratio of collected data to the total data:
$$
\alpha = \frac{\sum_{m \in \mathcal{M}} \sum_{n \in \mathcal{N}} e_{mn} q_n}{\sum_{n \in \mathcal{N}} q_n},
$$
where \(e_{mn}\) is a binary association indicator. Each WD can be served at most once, and the number of WDs served by each UAV is limited by \(N_{\max}\). The bandwidth allocation must satisfy
$$
\sum_{n \in \mathcal{N}} e_{mn} B_{mn} = B_W, \quad \forall m,
$$
where \(B_W\) is the total bandwidth of the UAV. In addition, the horizontal distance between the UAV at hovering point \(m\) and a served WD must not exceed \(r_{\max}\).
3.2 Proposed Optimization Framework
We formulate the joint optimization problem as
$$
\begin{aligned}
(P3-1): \quad & \min_{\mathbf{e}, \mathbf{B}, \mathbf{L}} \; F = – \alpha + \lambda E_{total}, \\
\text{s.t.} \quad & e_{mn} \in \{0,1\}, \\
& \sum_{m \in \mathcal{M}} e_{mn} \le 1, \quad \forall n \in \mathcal{N}, \\
& \sum_{n \in \mathcal{N}} e_{mn} \le N_{\max}, \quad \forall m \in \mathcal{M}, \\
& \sum_{n \in \mathcal{N}} e_{mn} B_{mn} = B_W, \quad \forall m \in \mathcal{M}, \\
& r_{mn} \le r_{\max}, \quad \forall m, n,
\end{aligned}
$$
where \(\mathbf{L} = \{\mathbf{l}_m^U\}\) is the set of hovering positions and \(\lambda\) is a weighting factor. This problem is NP-hard and non-convex. To tackle it, we decompose the problem into two subproblems: device association and energy minimization.
3.2.1 User Association with K-Means Initialized PSO
To optimize the device association, we propose a Particle Swarm Optimization algorithm enhanced with K-means initialization (PSO-KM). Standard PSO initializes particles randomly, which may degrade convergence performance. Instead, we first apply a weighted K-means algorithm to partition the WDs into \(M\) clusters. The clustering center of cluster \(m\) is computed as
$$
\mathbf{c}_m = \frac{\sum_{n \in \mathcal{C}_m} q_n \mathbf{l}_n}{\sum_{n \in \mathcal{C}_m} q_n},
$$
where \(\mathcal{C}_m\) is the set of WDs assigned to cluster \(m\). The cluster centers are used as an initial solution for PSO. Each particle position represents the coordinates of the \(M\) hovering points. For a given set of positions, we assign WDs to UAVs using a greedy approach: each WD is associated with the nearest UAV that has available capacity and is within \(r_{\max}\). The fitness function is set to maximize \(\alpha\). After iteratively updating velocities and positions, the algorithm converges to the optimal association and initial hovering positions.
3.2.2 Two-Stage Optimization for Energy Minimization
After the device association is determined, the next step is to minimize the total energy consumption by optimizing the hovering positions and bandwidth allocation. We propose a two-stage optimization algorithm based on PSO (TOA-PSO). In the first stage, given the hovering positions, we optimize the bandwidth allocation to minimize the hovering time. Since the hovering time at each point is determined by the maximum transmission time among the served WDs, the problem can be decomposed into \(M\) independent subproblems. For each hovering point, we use PSO to find the optimal bandwidth vector. In the second stage, given the bandwidth allocation, we optimize the hovering positions using a PSO variant enhanced with Gaussian perturbation and differential evolution mechanisms (PSO-GD). The total energy consumption includes flight energy and hovering energy:
$$
E_{total} = P_{for} \frac{D_{for}}{V} + P_{hov} \sum_{m \in \mathcal{M}} T_m^{hov},
$$
where \(D_{for}\) is the total horizontal flight distance and \(T_m^{hov}\) is the hovering time at point \(m\). To minimize \(D_{for}\), the order of visiting hovering points is optimized as a traveling salesman problem. The PSO-GD algorithm initializes particles around the previously obtained initial positions using Gaussian perturbation, and then applies crossover and mutation operations from differential evolution to improve global search capability. The two stages are iterated alternately until convergence.
3.3 Simulation Results
We evaluate the proposed algorithm in a square area of 500 m × 500 m. The UAV has a total bandwidth of 1 MHz, flying power 240 W, hovering power 200 W, maximum service capacity \(N_{\max}=10\), and flight speed 15 m/s. The number of hovering points varies from 2 to 8.
We compare our PSO-KM algorithm with standard K-means and standard PSO in terms of coverage utility. The following table shows the coverage utility \(\alpha\) for different numbers of hovering points when the number of WDs is 50.
| Algorithm | M = 2 | M = 4 | M = 6 | M = 8 |
|---|---|---|---|---|
| Proposed PSO-KM | 45.58% | 77.97% | 89.54% | 95.61% |
| K-Means | 35.76% | 74.88% | 79.80% | 82.69% |
| PSO | 42.04% | 75.01% | 82.51% | 88.55% |
It is observed that the proposed PSO-KM achieves the highest coverage utility for all values of \(M\). Moreover, when the number of WDs increases to 100, the coverage utility decreases for all algorithms due to the limited service capacity, but our algorithm still outperforms the baselines.
We also compare the total energy consumption under different optimization strategies. The results are summarized in the table below.
| Optimization Strategy | M = 2 | M = 4 | M = 6 | M = 8 |
|---|---|---|---|---|
| Joint optimization (Proposed TOA-PSO) | 57.27 kJ | 91.39 kJ | 102.66 kJ | 106.93 kJ |
| Initial positions, equal bandwidth | 83.71 kJ | 134.69 kJ | 145.37 kJ | 161.55 kJ |
| Only bandwidth optimization | 63.31 kJ | 101.99 kJ | 114.99 kJ | 121.20 kJ |
| Only position optimization | 75.20 kJ | 126.19 kJ | 137.41 kJ | 143.00 kJ |
Clearly, the joint optimization strategy achieves the lowest energy consumption in all cases. For example, when \(M=8\), the proposed algorithm saves about 34% energy compared to the case with initial positions and equal bandwidth allocation. The convergence analysis shows that the TOA-PSO algorithm converges within about 80 iterations depending on the number of hovering points. Compared with differential evolution, standard PSO, and artificial hummingbird algorithm, the proposed PSO-GD converges faster and yields a better solution.
4 Traffic-Priority-Based Multi-UAV Energy-Efficient Deployment
4.1 Scenario Description
In the second scenario, we consider a large public event where a large number of ground users are concentrated in a certain area. The traffic demand of users varies significantly. Some users may have high data rate requirements, while others may have low traffic. To avoid network congestion, multiple unmanned aerial vehicles are deployed to provide temporary communication services. Each UAV is dispatched from a launch point, flies to a designated position, and hovers to serve its associated users. The goal is to cover at least 90% of the total traffic demand with the minimum number of unmanned aerial vehicles and the minimum total energy consumption.
Let \(\mathcal{N}\) denote the set of ground users with traffic demand \(q_n\). The association between UAV \(m\) and user \(n\) is denoted by \(e_{mn}\). Each user can be served by at most one UAV. Each UAV can serve at most \(N_{\max}\) users and at most \(Q_{\max}\) total traffic. The total bandwidth and transmit power of each UAV are limited to \(B_W\) and \(P_d\), respectively. The traffic coverage ratio is defined as
$$
\alpha = \frac{\sum_{m \in \mathcal{M}} \sum_{n \in \mathcal{N}} e_{mn} q_n}{\sum_{n \in \mathcal{N}} q_n}.
$$
The optimization problem is formulated as follows.
$$
\begin{aligned}
(P4-1): \quad & \min_{\mathbf{e}, \mathbf{B}, \mathbf{P}, \mathbf{L}} \; \left( M, \; \sum_{m \in \mathcal{M}} E_{TEC}^{(m)} \right), \\
\text{s.t.} \quad & e_{mn} \in \{0,1\}, \\
& \sum_{m \in \mathcal{M}} e_{mn} \le 1, \quad \forall n \in \mathcal{N}, \\
& \sum_{n \in \mathcal{N}} e_{mn} \le N_{\max}, \quad \forall m \in \mathcal{M}, \\
& \sum_{n \in \mathcal{N}} e_{mn} q_n \le Q_{\max}, \quad \forall m \in \mathcal{M}, \\
& \sum_{n \in \mathcal{N}} e_{mn} B_{mn} = B_W, \quad \forall m \in \mathcal{M}, \\
& \sum_{n \in \mathcal{N}} e_{mn} p_{mn} = P_d, \quad \forall m \in \mathcal{M}, \\
& \alpha \ge 90\%.
\end{aligned}
$$
The total energy consumption of UAV \(m\) includes vertical ascent energy, horizontal flight energy, hovering energy, and communication energy:
$$
E_{TEC}^{(m)} = P_{asc} \frac{h_m^U}{V_{asc}} + P_{for} \frac{\| \mathbf{l}_m^{U} – \mathbf{l}_m^{start} \|}{V} + P_{hov} \max_{n \in \mathcal{C}_m} \left\{ \frac{q_n}{R_{mn}} \right\} + \sum_{n \in \mathcal{C}_m} p_{mn} \frac{q_n}{R_{mn}},
$$
where \(\mathbf{l}_m^{start}\) is the launch point and \(\mathcal{C}_m\) is the set of users served by UAV \(m\).
4.2 Proposed Deployment Algorithm
4.2.1 Traffic-Priority Algorithm for User Association
To minimize the number of unmanned aerial vehicles, we propose a step-by-step deployment method called the Traffic-Priority Algorithm (TPA). For each newly deployed UAV, the algorithm selects a cluster center that maximizes the total traffic demand covered by that UAV while satisfying the service constraints. To reduce the search space, we introduce the concept of traffic density. For each uncovered user \(n\), the traffic density is defined as the sum of traffic demands of all uncovered users within a distance of \(2 r_{\max}\):
$$
\xi_n = \sum_{n’ \in \mathcal{U}} q_{n’} \cdot \mathbb{1}(d_{nn’} \le 2 r_{\max}),
$$
where \(\mathcal{U}\) is the set of uncovered users and \(d_{nn’}\) is the Euclidean distance between users \(n\) and \(n’\). The user with the maximum traffic density is selected as a reference point. Only users within \(2 r_{\max}\) from this reference are considered as candidate users for the current UAV, which significantly reduces the solution space.
Then, a particle swarm optimization algorithm is applied to find the optimal cluster center that maximizes the total traffic demand of users within the coverage radius. The fitness value for each particle is computed by selecting the nearest \(N_{\max}\) users within \(r_{\max}\) and ensuring the total traffic does not exceed \(Q_{\max}\). Once the cluster is determined, the minimum enclosing circle of the cluster is computed to obtain the coverage region. The covered users are removed from the set \(\mathcal{U}\). This process is repeated until the traffic coverage ratio reaches 90%.
4.2.2 Joint Bandwidth, Power, and 3D Position Optimization
After the user association is determined, we optimize the bandwidth, power, and three-dimensional position of each UAV to minimize its total energy consumption. We decompose the problem into two subproblems: resource allocation and position optimization, and solve them alternately using the block coordinate descent method.
Resource Allocation: Given the UAV position, the optimal bandwidth and power allocation minimizes the maximum transmission time among the served users. Since bandwidth and power resources are discretized into resource blocks (RBs), we propose an iterative allocation algorithm. Initially, each user receives one RB. At each step, the user with the longest transmission time is assigned an additional RB. This process is repeated until all RBs are allocated. Power allocation is performed in a similar manner with discrete power levels.
Position Optimization: Given the allocated resources, the UAV can adjust its horizontal position and height to reduce energy consumption. Since the coverage region of a UAV is circular in the horizontal plane, the UAV does not need to hover exactly above the center. By tilting its directional antenna, the coverage area becomes an ellipse. The minimum height required to cover the circular region of radius \(r_m\) is derived from the ellipse geometry. According to the literature, the minimum height is
$$
h_m^{U,min}(x_m^U, y_m^U) = -\frac{D}{2} + \sqrt{\frac{E(x_m^U, y_m^U) + D^2}{2} + \frac{F(x_m^U, y_m^U)}{2}},
$$
where
$$
D = r_m \cos \theta_B, \quad E = (x_m^U – x_m^c)^2 + (y_m^U – y_m^c)^2, \quad F = \sqrt{E^2 + 2 E D \sin \theta_B + D^4},
$$
and \((x_m^c, y_m^c)\) is the center of the coverage region. By substituting this minimum height into the energy expression, the three-dimensional optimization problem is converted into a two-dimensional positioning problem. We adopt the Whale Optimization Algorithm (WOA) to find the optimal horizontal position that minimizes the total energy consumption. The BPL algorithm iterates between resource allocation and position optimization until the energy difference between consecutive iterations is below a threshold.
4.3 Simulation Results
We consider a square area of 1 km × 1 km with 200 ground users randomly distributed. The maximum coverage radius is 200 m, each UAV has a bandwidth of 1 MHz, total transmit power 1 W, and service limit \(N_{\max}=20\), \(Q_{\max}=10\) Gbits. The proposed TPA algorithm requires 10 unmanned aerial vehicles to achieve a traffic coverage ratio of 91.2%. The total energy consumption is \(5.91376 \times 10^6\) J.
We compare TPA with two existing methods: weighted K-means (WKM) and minimum degree priority (MDP). The following table shows the required number of unmanned aerial vehicles for different deployment area sizes and user counts.
| Area Side Length / Users | TPA | WKM | MDP |
|---|---|---|---|
| 1 km, 300 users | 12 | 16 | 15 |
| 2 km, 300 users | 29 | 41 | 36 |
| 3 km, 300 users | 58 | 88 | 69 |
| 1 km, 500 users | 19 | 25 | 23 |
It is evident that TPA requires the fewest unmanned aerial vehicles in all configurations. We also examine the impact of service capability. When \(Q_{\max}=40\) Gbits and \(N_{\max}\) increases from 10 to 50, the required number of UAVs decreases because each UAV can serve more users. However, when \(Q_{\max}=10\) Gbits, increasing \(N_{\max}\) has limited effect due to the traffic constraint.
For energy consumption, we compare the proposed BPL algorithm with three baseline strategies: (1) no bandwidth/power/position optimization (NBPL); (2) only position optimization (OL); and (3) only bandwidth and power optimization (OBP). The total energy consumption for different \(N_{\max}\) values is listed below.
| \(N_{\max}\) | Number of UAVs | NBPL (10^6 J) | OL (10^6 J) | OBP (10^6 J) | Proposed BPL (10^6 J) |
|---|---|---|---|---|---|
| 20 | 13 | 2.3374 | 2.1017 | 1.4312 | 1.3028 |
| 30 | 9 | 2.2398 | 2.1390 | 1.4002 | 1.3276 |
| 40 | 8 | 2.2817 | 2.2127 | 1.4534 | 1.4015 |
The BPL algorithm achieves the lowest total energy consumption among all strategies. For example, with \(N_{\max}=20\), BPL reduces total energy consumption by 44.3% compared to NBPL and by 10.4% compared to OBP. Figure 7 in our simulation shows that the average energy consumption per UAV increases with \(N_{\max}\), mainly due to the larger coverage radius and higher flight height. Figure 8 demonstrates that BPL maintains the best energy efficiency as the number of ground users increases.
5 Conclusion
In this paper, we have investigated energy-efficient deployment optimization for unmanned aerial vehicles acting as aerial base stations in emergency communication scenarios. Two typical scenarios were studied: single-UAV multi-point data collection and multi-UAV cooperative coverage. For the single-UAV scenario, we proposed a particle swarm optimization algorithm with K-means initialization for device association and a two-stage optimization algorithm for joint hovering position and bandwidth allocation. For the multi-UAV scenario, we introduced a traffic-priority algorithm to minimize the number of unmanned aerial vehicles while satisfying the traffic coverage requirement, and a joint bandwidth, power, and three-dimensional position optimization algorithm to minimize the total energy consumption. Simulation results showed that the proposed algorithms significantly improve coverage utility and reduce energy consumption compared with various baseline methods. Future work will consider obstacle-aware path planning, connectivity constraints among unmanned aerial vehicles, and dynamic user mobility in real-world deployments.