A method and device for multi-target monitoring oriented unmanned aerial vehicle network cooperative trajectory planning and energy efficiency optimization

Abstract

This invention relates to a method and apparatus for collaborative trajectory planning and energy efficiency optimization of UAV networks for multi-target surveillance. The method achieves deep integration of routing pairing, trajectory planning and power control by constructing a multi-target collaborative optimization framework, designing a hierarchical linkage algorithm, and enhancing scenario adaptability and dynamic response capabilities. This solves the problem of balancing network performance, lifetime and energy efficiency in multi-target surveillance scenarios and promotes the large-scale application of UAV swarms in complex surveillance tasks.

Description

Technical Field

[0001] This invention relates to the field of UAV swarm cooperative control and network optimization technology, specifically to a method and apparatus for UAV network cooperative trajectory planning and energy efficiency optimization for multi-target surveillance. Background Technology

[0002] With the deep integration of drone technology, communication technology, and artificial intelligence, drone swarms have become a core piece of equipment for wide-area multi-target surveillance missions. In scenarios such as border control, disaster relief, and urban security, multiple tracking drones are needed to approach targets to collect real-time data such as images and locations. Then, relay drones construct a multi-hop transmission network to transmit the dispersed data back to remote base stations, forming a complete workflow of "target acquisition - relay transmission - base station processing." This heterogeneous architecture of "tracking + relay" can overcome the limitations of the coverage and communication distance of a single drone, meeting the core requirements of large-scale multi-target surveillance.

[0003] However, existing UAV network optimization technologies still suffer from the following key shortcomings in practical applications: First, the optimization objectives are singular, making it difficult to meet the needs of multiple tasks. Traditional technologies often focus on a single performance indicator, either maximizing data transmission rate or minimizing energy consumption. This leads to contradictions such as "high throughput accompanied by high energy consumption," "low energy consumption at the expense of transmission efficiency," and "long endurance but frequent disconnections," failing to balance the comprehensive demands of wide-area multi-target surveillance for "fast transmission, long service life, and low energy consumption." Second, the hierarchical optimization is disconnected. UAV routing pairing, trajectory planning, and power control often adopt independent design patterns. For example, trajectory planning does not consider the energy consumption constraints of pairing relationships, and power adjustment does not adapt to link changes in flight trajectories, resulting in low overall system optimization efficiency and difficulty in coping with complex scenarios involving multiple moving targets and dynamic changes in network topology. Third, the dynamic adjustment mechanism is imperfect and lacks robustness: the quantitative conditions for triggering scheme adjustments are not clearly defined, such as target movement distance and energy thresholds, and priority strategies are not established. When sudden situations such as insufficient energy, decreased connectivity, or failure to meet throughput occur, the system cannot respond quickly, easily leading to monitoring interruptions or task failures. Furthermore, in multi-target surveillance scenarios, issues such as dispersed and mobile targets, communication links being susceptible to environmental interference, and the limited energy of drones further exacerbate the complexity of network optimization.

[0004] In summary, existing technologies have failed to fully integrate the advantages of multiple algorithms to solve the aforementioned pain points, resulting in the drone swarm's monitoring coverage integrity, data transmission reliability, and mission duration failing to meet actual needs. Summary of the Invention

[0005] To address the aforementioned problems, the present invention aims to provide a method and apparatus for collaborative trajectory planning and energy efficiency optimization of UAV networks for multi-target surveillance. By constructing a multi-target collaborative optimization framework, designing a hierarchical linkage algorithm, and enhancing scenario adaptability and dynamic response capabilities, the invention achieves deep integration of routing pairing, trajectory planning, and power control, thereby solving the problem of balancing network performance, lifespan, and energy efficiency in multi-target surveillance scenarios and promoting the large-scale application of UAV swarms in complex surveillance tasks.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a method for cooperative trajectory planning and energy efficiency optimization of unmanned aerial vehicle (UAV) networks for multi-target surveillance, characterized by the following steps: At the routing and pairing layer, feasible pairing of tracking drones and relay drones is performed based on the three-dimensional distance between the tracking drone and the relay drone, LOS or NLOS path determination, channel gain quantization and rate constraint screening. The globally optimal pairing combination is found through the Hungarian algorithm, while energy consumption and rate constraints are verified, and the pairing results are output. The pairing results include a pairing relationship table and a charging request for energy-crisis drones. At the topology and trajectory layer, the energy replenishment mode is prioritized based on the charging urgency. The single-hop or multi-hop transmission mode is adaptively switched based on link quality. A smooth trajectory is generated by using multi-segment Dubins paths and rolling time-domain optimization. Kalman filter is fused to predict the target position and velocity obstacle model to achieve dynamic obstacle avoidance. The system plans charging and return paths for low-energy drones and outputs the trajectory dynamics. Combining the pairing results with the trajectory dynamics, a full-dimensional calculation model integrating flight energy consumption and transmission energy consumption is constructed at the power control layer. The transmit power is adaptively adjusted based on link rate, channel gain and remaining energy through a time slot iteration mechanism to achieve total energy consumption optimization while meeting communication quality constraints. Based on total energy consumption optimization, a weighted sum method is used to define the comprehensive objective function. The weights of each objective are dynamically allocated according to the proportion of the UAV's remaining energy. A linkage mechanism is established between the weights and the dual-objective optimization precise pairing algorithm, the throughput-aware hybrid trajectory planning algorithm, and the multi-constraint adaptive power optimization algorithm to ensure that the optimization actions at each level conform to the overall optimal requirements of the system. Based on the linkage mechanism, the system extracts task, UAV status and environmental features through a multi-head heterogeneous attention strategy network. The REINFORCE algorithm is used to train the pairing parameters of the routing and pairing layer, the trajectory planning and charging scheduling parameters of the topology and trajectory layer, and the power control parameters of the power control layer. The trained model is deployed on the cloud platform, and data is collected in real time. The system calls the dual-objective optimization precise pairing algorithm, throughput-aware hybrid trajectory planning and multi-constraint adaptive power optimization algorithm to perform collaborative calculations and output structural optimization schemes. The system status is monitored based on quantized thresholds and dynamically adjusted, forming a closed-loop execution mechanism of "real-time perception-policy reasoning-dynamic deployment".

[0007] This invention also provides a device for cooperative trajectory planning and energy efficiency optimization of unmanned aerial vehicle (UAV) networks for multi-target surveillance, comprising: The first processing unit is used to perform feasible pairing of tracking drones and relay drones at the routing and pairing layer based on the three-dimensional distance between the tracking drone and the relay drone, LOS or NLOS path determination, channel gain quantization and rate constraint screening, find the globally optimal pairing combination through the Hungarian algorithm, and at the same time verify energy consumption and rate constraints, and output the pairing result. The pairing result includes a pairing relationship table and a charging request marked for the energy-crisis drone. The second processing unit is used to prioritize energy replenishment mode based on charging urgency at the topology and trajectory layers, adaptively switch between single-hop and multi-hop transmission modes based on link quality, generate smooth trajectories using multi-segment Dubins paths and rolling time-domain optimization, and achieve dynamic obstacle avoidance by integrating Kalman filter prediction of target position and velocity obstacle models. It plans charging and return paths for low-energy drones and outputs trajectory dynamics. The third processing unit is used to combine the pairing results with the trajectory dynamics, and in the power control layer, to construct a full-dimensional calculation model that integrates flight energy consumption and transmission energy consumption. Through a time slot iteration mechanism, the transmission power is adaptively adjusted based on the link rate, channel gain and remaining energy to achieve total energy consumption optimization under the premise of meeting communication quality constraints. The fourth processing unit is used to define a comprehensive objective function based on the total energy consumption optimization using a weighted sum method, dynamically allocate the weights of each objective according to the proportion of the UAV's remaining energy, and establish a linkage mechanism between the weights and the dual-objective optimization precise pairing algorithm, the throughput-aware hybrid trajectory planning algorithm, and the multi-constraint adaptive power optimization algorithm to ensure that the optimization actions at each level conform to the overall optimal requirements of the system. The fifth processing unit, based on the linkage mechanism, extracts task, UAV status, and environmental features through a multi-head heterogeneous attention strategy network. It uses the REINFORCE algorithm to train the pairing parameters of the routing and pairing layer, the trajectory planning and charging scheduling parameters of the topology and trajectory layer, and the power control parameters of the power control layer. The trained model is then deployed to the cloud platform, where data is collected in real time. The unit calls upon the dual-objective optimization precise pairing algorithm, the throughput-aware hybrid trajectory planning algorithm, and the multi-constraint adaptive power optimization algorithm for collaborative computation, outputting a structural optimization scheme. Based on a quantized threshold, the system status is monitored and dynamically adjusted, forming a closed-loop execution mechanism of "real-time perception - strategy reasoning - dynamic deployment".

[0008] The present invention has the following advantages due to the adoption of the above technical solutions: This invention innovatively designs a three-layer architecture of "route pairing - topology trajectory - power control", which solves the problem of disconnect in traditional layered optimization and achieves "precise pairing, adaptive trajectory, and intelligent power control". This invention constructs a comprehensive objective function using a weighted sum method, and adaptively adjusts the objective weights based on the remaining energy state of the UAV to balance the comprehensive requirements of "fast transmission, long service life, and low energy consumption". This invention integrates the advantages of multiple algorithms such as the Hungarian algorithm, Dubins path planning, Kalman filtering, and reinforcement learning. All core parameters and constraints are quantitatively defined, and can be directly reproduced by ordinary technicians. This invention clarifies the quantitative thresholds for triggering adjustments, such as target movement, insufficient throughput, insufficient energy, and connectivity imbalance, and establishes a priority strategy. Attached Figure Description

[0009] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. In the drawings: Figure 1 This is a flowchart of the UAV network cooperative trajectory planning and energy efficiency optimization method for multi-target surveillance as described in this invention; Figure 2 This is a schematic diagram of the system architecture of the present invention. Detailed Implementation

[0010] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0011] This invention provides a method and apparatus for collaborative trajectory planning and energy efficiency optimization of UAV networks for multi-target surveillance. The method first constructs a three-dimensional joint optimization framework centered on "throughput-lifetime-energy," defining a well-defined comprehensive objective function through a weighted sum method. It quantifies fundamental constraints such as network connectivity, minimum throughput requirements, single-unit energy limits, and kinematics, clarifying the computational logic and boundaries of each objective. Furthermore, it models the complex network optimization problem as a Markov decision process, innovatively designing a three-layer collaborative optimization algorithm: a routing and pairing layer, a topology and trajectory layer, and a power control layer. The routing and pairing layer employs a throughput maximization method, using multi-stage quantization modeling and the Hungarian algorithm to achieve optimal pairing between tracking and relay UAVs. The topology and trajectory layer uses a throughput-aware hybrid trajectory planning algorithm, adaptively switching between single-hop and multi-hop modes, combining multi-segment Dubins paths, rolling temporal optimization, and dynamic obstacle avoidance techniques to generate energy-aware optimized trajectories. The power control layer uses an adaptive power adjustment algorithm, adjusting the transmission power in stages based on link quality, throughput gap, and remaining energy. Key parameters of the entire algorithm are optimized through reinforcement learning training to improve scenario adaptability. Finally, the trained model is deployed on a cloud platform, and the monitoring target, UAV status and environmental information are input in real time. The model quickly outputs the optimal pairing scheme, relay trajectory sequence and power control strategy, and dynamically adjusts based on a clear threshold during task execution. In this way, in dynamic and complex scenarios, the model can simultaneously maximize the total network throughput, extend the network lifetime and reduce the total energy consumption, thus solving the key problem of balancing network performance, lifetime and energy efficiency in multi-target monitoring.

[0012] like Figure 1 As shown, the UAV network cooperative trajectory planning and energy efficiency optimization method for multi-target surveillance provided by the present invention is characterized by the following steps: S1. At the routing and pairing layer, with the dual objectives of maximizing throughput and minimizing energy consumption, a pairing optimization model for tracking UAVs and relay UAVs is established to realize a precise pairing algorithm for dual-objective optimization and output the pairing results: Based on three-dimensional distance calculation, LOS or NLOS path determination, channel gain quantization and rate constraint screening, feasible pairing of tracking UAVs and relay UAVs is performed. The globally optimal pairing combination is found through the Hungarian algorithm, while verifying energy consumption and rate constraints, and the pairing results are output. The pairing results include a pairing relationship table and a charging request for marking energy-crisis UAVs.

[0013] S2. At the topology and trajectory layer, with the goal of "extending network lifetime", a relay UAV trajectory planning and charging scheduling model is established to realize the throughput-aware hybrid trajectory planning algorithm and output dynamic trajectory: the energy replenishment mode is prioritized based on the charging urgency, the single-hop or multi-hop transmission mode is adaptively switched based on link quality, and a smooth trajectory is generated by using multi-segment Dubins path and rolling time domain optimization. Kalman filter is integrated to predict the target position and velocity obstacle model to realize dynamic obstacle avoidance, and to plan charging and return paths for low-energy UAVs.

[0014] S3. Based on the dual-objective optimization precise pairing algorithm and the throughput-aware hybrid trajectory planning algorithm, at the power control layer, with the goal of "minimizing the total system energy consumption", a dynamic adjustment model for transmit power is established to realize a multi-constraint adaptive power optimization algorithm: combining the pairing results and the trajectory dynamics, a full-dimensional calculation model integrating flight energy consumption and transmission energy consumption is constructed. Through the time slot iteration mechanism, the transmit power is adaptively adjusted according to the link rate, channel gain and remaining energy to achieve total energy consumption optimization under the premise of meeting communication quality constraints.

[0015] S4. Based on the multi-constraint adaptive power optimization algorithm, a three-dimensional global collaborative optimization framework of "throughput-lifetime-energy consumption" is established to realize the collaborative balance mechanism of throughput-lifetime-energy consumption: the weighted sum method is used to define the comprehensive objective function, the weight of each objective is dynamically allocated according to the proportion of the UAV's remaining energy, and a linkage mechanism is established between the weight and the dual-objective optimization precise pairing algorithm, the throughput-aware hybrid trajectory planning algorithm and the multi-constraint adaptive power optimization algorithm to ensure that the optimization actions of each layer conform to the overall optimal requirements of the system.

[0016] S5. Based on the throughput-lifetime-energy consumption collaborative balance mechanism, a three-layer collaborative algorithm training of the routing and pairing layer, topology and trajectory layer, and power control layer is implemented, and a closed-loop execution mechanism is deployed in the cloud: Task, UAV state, and environmental features are extracted through a multi-head heterogeneous attention strategy network. The REINFORCE algorithm is used to train the pairing parameters of the routing and pairing layer, the trajectory planning and charging scheduling parameters of the topology and trajectory layer, and the power control parameters of the power control layer. The trained model is deployed on the cloud platform, data is collected in real time, and the dual-objective optimization precise pairing algorithm, throughput-aware hybrid trajectory planning, and multi-constraint adaptive power optimization algorithm are called for collaborative calculation to output a structural optimization scheme. The system state is monitored based on the quantization threshold and dynamically triggered for adjustment, forming a closed-loop execution mechanism of "real-time perception-policy reasoning-dynamic deployment".

[0017] In the above embodiment, preferably, step S1 involves performing feasible pairing of the tracking UAV and the relay UAV based on three-dimensional distance calculation, LOS or NLOS path determination, channel gain quantization, and rate constraint screening. The globally optimal pairing combination is found using the Hungarian algorithm, while simultaneously verifying energy consumption and rate constraints, and the pairing result is output. Specifically, this includes the following steps: S101. Based on the horizontal coordinates of the UAV's real-time location, the horizontal distance between the tracking UAV and the relay UAV is calculated using the horizontal distance formula, laying the groundwork for the calculation of three-dimensional distance and channel gain. The horizontal distance calculation formula is as follows: ; In the formula, To track the horizontal distance between the drone and the relay drone; To track drones The x-coordinate of the horizontal coordinate; To track drones The horizontal coordinate and the vertical coordinate; For relay drones The x-coordinate of the horizontal coordinate; For relay drones The horizontal coordinate and the vertical coordinate; the set of tracking drones is denoted as (Total i), the relay drone set is denoted as (common (frame), and meet To ensure sufficient relay resources; S102. Combining horizontal distance and flight altitude, the three-dimensional distance between the tracking UAV and the relay UAV is calculated using a three-dimensional distance formula. This distance is a core factor affecting channel attenuation. The three-dimensional distance formula is as follows: ; In the formula, To track the three-dimensional distance between the drone and the relay drone; To track drones Flight altitude; For relay drones Flight altitude; and simultaneously introduce communication distance constraints: , The maximum communication range is 5km by default; S103. Based on the relative elevation angle between the tracking UAV and the relay UAV, calculate the LoS probability and NLoS probability using the LoS probability formula and NLoS probability formula respectively, determine the LoS or NLoS propagation path, distinguish the signal propagation path type, and ensure that the channel model fits the actual scenario: The LoS probability formula is: ; In the formula, This represents the Loss of Sorrow (LoS) probability. for and The angle of elevation between them; , Adapt environmental parameters for the scene (default) , Therefore, the greater the elevation angle, the higher the probability of Loss of Service (LOS). NLoS probability formula: ; In the formula, This represents the NLoS probability. This represents the Loss of Sorrow (LoS) probability. S104, By first introducing the path loss index This describes the rate attenuation of the signal as it travels further, and is then dynamically configured based on the monitoring environment, including unobstructed scenarios. Dense urban areas Calculate the average channel gain: The channel gain for LoS and NLoS scenarios is calculated using the following formulas respectively: ; ; In the formula, Reference channel gain per unit distance (default) (reference gain at time). To track the three-dimensional distance between the drone and the relay drone; This is the path loss index. , This introduces additional path loss; Then, the average channel gain is calculated using the average channel gain formula, where the average channel gain is the probability-weighted sum of the two scenarios. The specific formula is as follows: ; In the formula, This represents the Loss of Sorrow (LoS) probability. Channel gain in the LoS scenario; This represents the NLoS probability. Channel gain in NLoS scenarios; Let be the average channel gain, where A larger value indicates less signal attenuation and better communication quality, providing core input for subsequent transmission rate calculations; S105. Combining the signal-to-noise ratio and channel parameters, the single-pair communication capability is accurately quantified using the following formula, while simultaneously defining the rate constraint: ; In the formula, For relay drones To the receiving node The achievable data transmission rate; For communication bandwidth; To track drones The transmission power; The residual self-interference coefficient; The noise power spectral density (default -174dBm / Hz). The link interference power is measured based on actual environmental conditions; a minimum required transmission rate constraint is also introduced. (Default 8Mbps), requirements Only then can the subsequent matching and screening process begin; S106. Construct a weighted bipartite graph to screen feasible pairings. Specifically, based on transmission rate and communication distance constraints, screen pairing combinations that meet basic communication requirements to narrow down the optimization range: For each group Determine whether the following conditions are met simultaneously: and If the above two conditions are met, then a bipartite graph is established. and The edge weight is set to (i.e., single-link transmission rate); if any condition is not met, no edge is built; ultimately, a weighted bipartite graph is formed. ,in Let be the set of feasible edges. The set of edge weights; S107. Filter high-quality potential matches by setting a rate threshold; For each tracking drone Calculate the maximum transmission rate among all feasible pairings, and use it as the rate threshold: ; If a tracking drone is set If there is no associated edge, it is marked as a "node to be filled". The problem can be solved by lowering the threshold or scheduling a backup relay to avoid overall pairing failure. S108. Generate an unweighted subgraph, simplified matching space traversal, and weighted bipartite graph. All edges If the edge weights satisfy If the edge is unweighted, retain it and remove its weight attribute; otherwise, delete it. Here, the threshold is relaxed by 10% to avoid over-filtering, ultimately generating an unweighted binary graph. ,in This is the simplified set of feasible edges; S109. Clarify the calculation logic for energy consumption cost of a single pair. Calculate the total energy consumption using the following formula, which is the sum of the flight energy consumption of the tracking UAV, the flight energy consumption of the relay UAV, and the transmission energy consumption. The formula is as follows: ; In the formula, To track the drone's flight energy consumption coefficient (default 0.5J / m); for The three-dimensional distance to the monitored target; The energy consumption coefficient for relay drone flight (default 0.5J / m); for The three-dimensional distance to the base station; The transmission energy consumption coefficient (default 1J / (W・s)); for The transmission power; The communication time is set at 1 second by default; a maximum allowable total energy consumption per pair is also introduced. (Default 150J), Requirements ; For unweighted subgraphs Let there be an edge that does not exist in the middle. Marked as infeasible pairings, and finally constructed. Energy consumption cost matrix ; Then, the Hungarian algorithm is used for optimization. With the goal of "minimizing total energy consumption cost", the Hungarian algorithm is called to find the maximum match and output the preliminary matching results. S110. Filter and output the final pairing scheme. Specifically, perform energy consumption verification on the preliminary pairing results and eliminate combinations that exceed the standard: calculate the energy consumption of each combination in the preliminary pairing results. Total energy consumption ;like If, then keep the combination; if The suboptimal pairing is selected and re-verified until the constraints are met. Finally, a structured pairing relationship table is output. This table provides explicit input for the subsequent topology layer and power control layer, ensuring the system's collaborative optimization effect. The specific relationship table is as follows: ; S111, Energy Status Assessment and Charging Request Generation: After completing the pairing scheme output, the system immediately assesses the remaining energy of all drones. ; Charging request determination: If the drone The remaining energy is below the charging request threshold. ,Right now Then mark the drone as "charging request status", denoted as ; Charging urgency calculation: The charging urgency is calculated for each drone requesting charging using the following formula. : ; In the formula, This is the estimated remaining mission time for the drone; This is the total task duration; For weighting coefficients; where urgency is... The higher the value, the more priority the drone needs to be given to charging.

[0018] In the above embodiments, preferably, the optimization using the Hungarian algorithm, with the objective of "minimizing total energy consumption cost," involves calling the Hungarian algorithm to find the maximum match and outputting preliminary pairing results, specifically including the following steps: Input parameters: Energy cost matrix Total number of tracking drones Total number of relay drones Minimum transmission rate ; Matrix preprocessing: Preprocessing all data in the energy cost matrix... The value is replaced with 10 times the maximum finite value in the matrix to avoid algorithm calculation errors, while retaining the characteristic that "infeasible pairing" has the lowest priority. The standard Hungarian algorithm is invoked: based on the preprocessed matrix, the "minimum weight matching" logic is executed, and the initial pairing result is output, which includes... The value of is the relay unit corresponding to each tracking device; Rate verification: Check all pairs in the initial pairing results. The link ensures its transmission rate. If there is a link that does not meet the requirements, set the energy cost of that link to ∞, and re-execute the matrix preprocessing and call the standard Hungarian algorithm steps. Output feasible pairings: If all links meet the rate constraints, output the current pairing results. If there are still links that do not meet the constraints, proceed to subsequent iterations for adjustment. The objective function and constraints are as follows: Objective function (minimize total energy consumption): ; Constraints: ; in, The total energy cost for all pairs; 0-1 decision variables ( Indicates tracking drones With relay drones pair, (Indicates no pairing); Constraints This means that each tracking drone must be paired with one and only one relay drone; This means that each relay drone can be paired with a maximum of one tracking drone (to avoid resource conflicts). This indicates that the decision variable can only take two states: "paired" or "unpaired". A successfully paired link must meet the minimum transmission rate requirement (default 8Mbps) to ensure that the throughput meets the standard. If a complete match is not achieved, the unpaired tracking drone will be... Reduce by 10% ), regenerate the subgraph and repeat the matching, with the upper limit of iteration set to 5 times.

[0019] In the above embodiments, preferably, in step S2, the step of prioritizing the energy replenishment mode based on charging urgency, adaptively switching between single-hop and multi-hop transmission modes based on link quality, generating a smooth trajectory using multi-segment Dubins paths and rolling time-domain optimization, and achieving dynamic obstacle avoidance by fusing Kalman filter prediction of target position and velocity obstacle models to plan charging and return routes for low-energy drones specifically includes the following steps: S201. Priority Assessment of Energy Supply Needs: When the system detects a charging request, it must prioritize ensuring the energy supply to critical nodes before making decisions on the normal communication mode. The specific steps are as follows: The system obtains the charging request list in real time to trigger the charging schedule. and its urgency Define the charging scheduling trigger flag : ; in, This is the urgency threshold; exceeding this value indicates an urgent need for charging. This is the critical energy value; below this value, there is a risk of crash. Key node identification, if The system further identifies which of the drones requesting charging are key network nodes, including the set of key nodes. satisfy: ; in, For nodes Leaving aside the cost of impacting network connectivity, The threshold for the impact of network connectivity is used to determine whether a drone is a critical node in the network. Its value is adaptively set according to the mission scenario and network topology characteristics.

[0020] Pattern Decision Logic: like The system immediately enters "charging scheduling priority mode", giving priority to charging. The drone charging scheduling scheme in the system may temporarily switch to a degraded maintenance mode during this period. like but The system enters a hybrid mode, which processes the charging scheduling task in parallel with the decision-making of the regular communication mode (single hop / multi hop) or jointly optimizes them within an optimization window. That is, when planning the trajectory, the communication link quality and the scheduling needs of the charging drone are considered at the same time. like The system executes a single-hop or multi-hop decision-making process. S202, Single-hop or multi-hop mode decision process: Hybrid trajectory planning is used for mode decision, calculating the joint probability that all tracking UAVs and relay UAVs can communicate directly via single-hop mode, as shown in the following formula: ; ; In the formula, express The probability of single-hop network connectivity at any given time is the joint probability that all tracking drones and relay drones are connected in a single hop. For the first The single-point, single-hop connectivity probability between the tracking drone and the relay drone; The area of ​​the intersection of the two circles (radius of the relay UAV communication circle) , No. Radius of the circle of position uncertainty for tracking drones , (Probability multiplier) for Time of the first The straight-line distance between the tracking drone and the relay drone; For the first A tracking drone The horizontal coordinate in the two-dimensional plane at time; For relay drones in The x-coordinate in the two-dimensional plane at time; For the first A tracking drone The ordinate in the two-dimensional plane at that moment; Relay drones The x-coordinate in the two-dimensional plane at time t; if Furthermore, the single-link transmission rates of all tracking drones and relay drones meet the requirements. The single-link packet loss rate meets the requirements. Maintain single-hop mode; if Or there exists a single-link transmission rate between any tracking drone and relay drone. Or there may be packet loss rate on any single link. Switch to multi-hop mode; in, This is the threshold for the connectivity probability of a single-hop network. This is the maximum acceptable packet loss rate threshold for a single link; The single-link packet loss rate, i.e., the first... The packet loss rate of the single-hop link between the tracking drone and the relay drone. For the first The single-hop link transmission rate between the tracking drone and the relay drone. Minimum rate requirement for a single link; S203. In single-hop trajectory planning, a center positioning algorithm is used to calculate the optimal center position of the relay UAV, as shown in the following formula: ; Its analytical solution is to track the geometric center of the drone swarm: ; In the formula, for Optimal center position coordinates of the relay drone at any time; For the first Tracking drones Time and location coordinates To track the total number of drones; S204, the formula for guaranteeing transmission rate in single-hop mode is as follows: ; in, Communication redundancy coefficient ( Ensure that the single-hop link transmission rate of all tracking drones and relay drones meets the requirements. To avoid rate attenuation due to proximity to the communication range boundary; S205. In multi-hop mode path selection, filter the relay path with the highest throughput, using the following formula: ; Among them, total path throughput The calculation is as follows: ; In the formula, for Time-optimal multi-hop relay path; The set of all feasible multi-hop paths; For the first The transmission power of the link; For the first Channel gain of the link; Noise power; For the first Interference power of the link; S206, Multi-segment Dubins path and rolling temporal domain optimization, constructing a kinematic model for relay UAVs: ; The constraints are: ; In the formula, For relay drone continuous time The following position coordinates; For flight speed, For heading angle; For turning rate; These are the upper and lower limits of speed; Maximum steering ratio; This is the maximum acceleration; The rolling time-domain optimization window is The optimization objective is to maximize the average connectivity probability within the window. ; In the formula, For the connectivity probability of the corresponding mode (single hop / multi-hop), the path is composed of a combination of "curve-straight-curve" (CSC) or "curve-curve-curve" (CCC); This refers to the number of time steps, or the duration of a single step. ; This is the starting time of the current optimization window; S207, Predicting the future position of a tracking drone based on Kalman filtering. Tracking drone location sequence This allows relay drone trajectory planning to adapt to target movement in advance, reducing the risk of connection loss. The formula is as follows: Equations of state: ; State update equation: ; Kalman gain calculation: ; Error covariance calculation: ; In the formula, For the first Tracking drones State estimate at time step Here is the state transition matrix. To control the input matrix, To track the control input of the drone, These are sensor observations. The observation matrix; The state estimation error covariance, To observe the noise covariance, It is the identity matrix; S208: The speed obstacle model generates collision cones to achieve dynamic obstacle avoidance and associated trajectory optimization. The code is as follows: ; In the formula, for Real-time relay drone and obstacles The collision cone; Location of the relay drone; Location of the obstacle; For the speed of the obstacle, The candidate velocity vector for the relay UAV; S209, collision cone-triggered replanning and multi-aircraft synchronous arrival, i.e., "correction" after obstacle avoidance, if the relay drone candidate speed This triggers local replanning, adjusting the parameters of the Dubins path segments; synchronous arrival constraints: ; In the formula, For the first Segment path length; This represents the flight speed of the corresponding sub-path; This refers to the time it takes for multiple drones to arrive at the target area simultaneously. S210, Energy Sensing Optimization, the formula is as follows: ; In the formula, Energy consumption weighting coefficient ( (as a weighting coefficient), and flight speed Positive correlation; Energy consumption constraints are applied to high-speed maneuvers to reduce energy waste and prevent drones from prematurely leaving the network due to energy depletion. The formula is as follows: ; In the formula, For the energy consumption rate of relay drones, For acceleration, This refers to the air drag coefficient; For flight speed; S211, Dynamic scheduling and trajectory planning for energy replenishment to address the risk of network lifetime interruption caused by the limited energy of UAVs: The system receives charging requests from S111. and its urgency The system intelligently schedules and plans trajectories for these requests to ensure timely replenishment of energy for critical drones while maintaining overall network performance. The specific process is as follows: Candidate supply station reachability analysis: For Each drone in Calculate the reachable set based on its current position. Remaining energy Energy consumption per unit distance Calculate the set of all supply stations that can be safely reached: ; In the formula, For a fixed set of supply points; A collection of mobile charging stations; For three-dimensional distance; if If so, the emergency procedure will be triggered, directing the drone to fly to the nearest safe point at the most economical speed or dispatching a mobile charging station to provide assistance. Optimal Supply Station Cost Assessment and Assignment: For each reachable candidate supply station Calculate the overall cost of going to this station to charge. : ; In the formula, In terms of economic speed The time cost of flight; This is the estimated total time spent at the station. For mobile charging stations, this value is related to their current task queue and charging power. and its own energy Related; It is a drone Leaving the post affects the network's global objective function The estimated negative impact, which is calculated through rapid simulation. The network state after departure is obtained, and the difference in the objective function is calculated; the weight parameters are taken as shown in the table below: ; Optimal assignment: to the drone requesting charging. Choose the supply station with the lowest overall cost: ; This decision also applies to mobile charging stations. The next service target and destination were generated; Charging task trajectory planning for each assigned drone and its target supply station Plan the complete mission trajectory: The charging trajectory is generated using the Dubins path and rolling time-domain optimization method. arrive The flight path is based on economic cruising speed and integrates dynamic obstacle avoidance; Return-to-work route: After completing the planned charging, from... The trajectory returns to its optimal standby position, which depends on the drone's position. The network status at the expected return time is used to recalculate the optimal center position of the relay UAV in single-hop mode trajectory planning using a center positioning algorithm, or to re-select the relay path with the highest throughput in multi-hop mode path selection based on the network status at the expected return time. Simultaneously, a pre-calculated temporary network reconstruction scheme for maintaining network performance is executed, and corresponding transition trajectories are planned for other drones involved in the adjustment. Finally, an updated dynamic trajectory coordinate sequence of the relay drone is output, which integrates the charging scheduling task trajectory and the network transition trajectory.

[0021] S212, Connectivity test function, definition Indicator variables of network connectivity at any given time: ; :express The network must always meet connectivity requirements; :express The network connection is constantly interrupted; In the formula, the multi-hop connectivity probability ( Let be the link connectivity probability vector. (This is the connected state matrix); the connectivity of the network graph is verified by depth-first search (DFS): if there exists a path that makes all drone nodes reachable from each other, then it is considered connected; For the first Single-link packet loss rate of tracking drones and relay drones; S213, Countconnected, the effective time step count, is calculated from the initial time. Start, cumulatively and continuously satisfy The number of time steps until the first occurrence. To stop counting, use the following formula: ; If there are no interruptions throughout the entire process ,but ( (for the maximum simulation time step). S214. Maximize connectivity by optimizing connectivity probability, avoiding collisions, reducing energy consumption, and ensuring supply. The duration of time, thereby extending network lifespan. The derivation formula is as follows: ; In the formula, The duration of a single time step.

[0022] In the above embodiments, preferably, in step S3, the step of combining the pairing results with the trajectory dynamics to construct a full-dimensional calculation model that integrates flight energy consumption and transmission energy consumption, and adaptively adjusting the transmit power based on link rate, channel gain, and remaining energy through a time slot iteration mechanism to achieve total energy consumption optimization under the premise of meeting communication quality constraints, includes the following steps: S301. Derivation of the single-pair energy consumption calculation model: The total energy consumption of a single pair is the sum of the flight energy consumption of the tracking UAV, the single-pair flight energy consumption of the relay UAV, and the single-pair transmission energy consumption. Based on the energy consumption calculation logic and combined with the characteristic of power dynamically adjusting with mission duration, the derivation is as follows in modules: Single-pair flight energy consumption: ; In the formula, 0-1 decision variables (1 represents tracking drones) With relay drones (Pairing, 0 indicates no pairing). To track the energy consumption coefficient of drone flights; To track drones The three-dimensional distance to the monitored target; The energy consumption coefficient for relay drone flight; For relay drones The three-dimensional distance to the base station; Power consumption for single-pair transmission: ; In the formula, Transmission energy consumption coefficient; Total task duration; For pairing At any moment The transmission power; Total energy consumption per pair: ; In the formula, For pairing The total energy consumption formula integrates the fixed characteristics of flight energy consumption and the dynamic characteristics of transmission energy consumption, fully reflecting the energy consumption composition of a single pair. S302. Construct power constraints to ensure communication quality: Power adjustment must prioritize communication quality to avoid link disconnection or insufficient speed due to excessively low power. Combine speed constraints with link connectivity requirements to form a set of hard constraints. ; ; ; In the formula, For pairing At any moment The transmission power; Minimum transmission rate constraint; For communication bandwidth; The noise power spectral density; This refers to the link interference power. For a moment pair The average channel gain; The minimum connectivity threshold; Minimum transmission power; To achieve the maximum transmission power, all of the above constraints must be met simultaneously. ; S303. Design an adaptive power adjustment strategy to solve energy consumption optimization: Employ a time-slot iterative adjustment mechanism to dynamically optimize the transmit power based on the link real-time rate, channel gain, and the UAV's remaining energy state, minimizing transmission energy consumption while satisfying constraints. (1) Initial power allocation: ; In the formula, For pairing The initial transmit power; Pairing at the initial time The average channel gain; this formula ensures that the initial power meets the minimum rate requirement and avoids initial link disconnection; (2) Perform real-time iterative adjustments, including the time slot duration. Specifically: Total task duration Divided into Each time slot, for each time slot Power is dynamically adjusted based on the real-time link rate. like (Rate redundancy ≥20%, sufficient communication quality): (Reduce power consumption by 5% to save energy); like (Insufficient speed affects data transmission): (Increase power by 5% to improve speed), and must meet the following requirements. (Not exceeding the maximum transmission power); like (Speed ​​meets standards and has no redundancy; communication quality is balanced): (Maintain stable power); If drone If it has been dispatched to the charging station, then its transmission power... Further conservative settings can be implemented, such as using a lower power limit while still meeting the minimum rate constraint. To conserve energy during the journey to the maximum extent possible and ensure a safe arrival at the charging station; If drone It is charging at the charging station. ; If drone If a device has just returned from a charging station and has sufficient energy, the energy-saving restrictions on its power adjustment can be temporarily relaxed to prioritize its rapid reconstruction of a high-quality link and give full play to its "full-power" advantage. In the formula, For a moment pair The transmission rate; The previous time slot was used; power oscillations were avoided through small iterative adjustments to balance energy saving and communication stability. (3) Energy perception correction: ; In the formula, in the formula, Time of the first Tracking drone and the first The original transmit power (in W) of the relay UAV pairing link; for Time of the first The remaining energy of the relay drone, For the first The maximum energy of the relay drone is only when (When in a low-energy state, prioritizing energy preservation) this correction will be implemented, further reducing power by 10% while meeting the lower power limit, thereby extending the working time of the relay drone; This refers to the transmission power. S304. Derivation of the formula for calculating the total system energy consumption: The total system energy consumption is the sum of the flight energy consumption and transmission energy consumption of all pairs. Combining the pairing decision variables and the dynamic power adjustment results, the formula for the total system energy consumption is obtained as follows: ; In the formula, This represents the total energy consumption of the system. To track the total number of drones; This represents the total number of relay drones; To track the flight energy consumption coefficient of drones; This is the flight energy consumption coefficient for relay drones; the formula fully integrates the energy consumption influencing factors of three layers: route pairing, topology trajectory, and power control, to achieve global energy consumption quantification; S305. Total Energy Consumption Numerical Calculation and Output: Since the integral term is difficult to calculate directly, based on the time-slot iterative adjustment strategy, the integral term is discretized into a time-slot summation form, resulting in a numerical formula that can be directly substituted into the parameters for calculation. ; In the formula, Total number of time slots ; The duration of a single time slot; For a moment pair By substituting the transmission power, pairing results, trajectory dynamic parameters, and power adjustment results, the total energy consumption of the system can be directly calculated, providing core data support for system energy consumption assessment and optimization decisions.

[0023] In the above embodiments, preferably, in step S4, the step of defining the comprehensive objective function using the weighted sum method, dynamically allocating the weights of each objective according to the remaining energy ratio of the UAV, and establishing a linkage mechanism between the weights and the dual-objective optimization precise pairing algorithm, the throughput-aware hybrid trajectory planning algorithm, and the multi-constraint adaptive power optimization algorithm, to ensure that the optimization actions at each layer conform to the overall optimal requirements of the system, specifically includes the following steps: S401. Definition of Global Synthesis Objective Function: The global synthesis objective function of the system is constructed using the weighted sum method to quantify the synergistic optimization relationship of the three objectives. The specific formula is as follows: ; In the formula, This refers to the overall performance indicators of the system. Total network throughput; For the network's lifespan; This represents the total energy consumption of the system. For throughput weight, For the sake of survival options, As energy consumption weight, satisfy This is to ensure the rationality and binding nature of the weight allocation; S402. Based on the remaining energy percentage of the UAV, dynamically adjust the weight percentage of the three objectives to achieve adaptive switching of optimization focus at different stages. The weight values ​​can be flexibly adjusted according to the actual monitoring task requirements. The specific allocation rules are shown in the table below: ; S403. Establish global weights to ensure that weight rules are implemented: For route pairing: When making pairing decisions, calculate the comprehensive score of "throughput-energy consumption" based on the current weights, and prioritize the pairing combination with the best comprehensive score; For trajectory planning and charging scheduling: The dynamic weights of throughput, lifetime, and energy consumption directly regulate the topology and trajectory layer decisions. When the weights are biased towards lifetime, trajectory planning focuses on energy conservation and sustainability, adopts smooth and economical trajectories, and actively triggers charging scheduling to plan charging paths for key low-energy nodes, prioritizing energy replenishment to extend network lifetime. For power control: When allocating power, the power threshold is dynamically adjusted according to the energy consumption weight. The higher the weight, the more conservative the power allocation, giving priority to ensuring low energy consumption.

[0024] In the above embodiments, preferably, in step S5, the process of extracting task, UAV state, and environmental features through a multi-head heterogeneous attention strategy network, training pairing parameters for the routing and pairing layer, trajectory planning and charging scheduling parameters for the topology and trajectory layer, and power control parameters for the power control layer using the REINFORCE algorithm, deploying the trained model on a cloud platform, collecting data in real time, and calling a dual-objective optimization precise pairing algorithm, a throughput-aware hybrid trajectory planning algorithm, and a multi-constraint adaptive power optimization algorithm for collaborative calculation, outputting a structural optimization scheme, monitoring the system state based on a quantized threshold, and dynamically triggering adjustments, forming a closed-loop execution mechanism of "real-time perception - strategy reasoning - dynamic deployment," specifically includes the following steps: S501. Clarify the specific dimensions and meanings of the three types of heterogeneous input features to provide a data foundation for network modeling, specifically: The task features are 16 dimensions in total: target quantity (1 dimension), target average movement speed (1 dimension), minimum throughput requirement (1 dimension), monitoring range (1 dimension), and target distribution density (12 dimensions), divided according to the monitoring area grid. The drone's state characteristics consist of 24 dimensions: remaining energy. 1D, real-time location 3D, flight speed Dimensions, Current Transmission Power The system comprises 15 dimensions: communication range (1 dimension), charging request status (0 = no request, 1 = request), charging urgency (1 dimension), and historical average energy consumption (15 dimensions, divided into 5-second time windows). Environmental characteristics comprise 24 dimensions: channel gain 1D, noise power spectral density 1D, obstacle density 1D, wind speed 1D, LosS probability 1D, NLoS probability 1D, environmental interference intensity 18D, divided into segments according to communication frequency; All heterogeneous features are integrated into a 64-dimensional feature vector to ensure that the network input dimension is consistent; S502. Through standardization and filtering, the differences in dimensions and noise interference are eliminated, specifically as follows: First, for each feature Normalization is performed to eliminate the influence of dimensions: ; In the formula, Features The statistical mean on the training dataset. Features The statistical standard deviation, These are the standardized eigenvalues; Then, outlier handling is performed: a reasonable range for feature values ​​is set, and outliers outside the range are replaced using interpolation. ; In the formula, This represents the eigenvalues ​​after correction or interpolation; The feature value of the previous time step. This serves as a feature value for the next time step, ensuring data continuity. S503. Constructing a multi-head heterogeneous attention network architecture: An encoder-decoder structure is adopted, and eight heterogeneous attention heads are designed to achieve accurate feature extraction. Specifically: (1) Network hierarchy division: The encoder consists of three layers: a feature mapping layer, which linearly projects the input features to enhance feature representation and lay the foundation for subsequent attention computation; a multi-head heterogeneous attention layer, which uses eight attention heads with different functional preferences to work in parallel to extract and focus key information from three heterogeneous dimensions: task, UAV state, and environment; and a feedforward network layer, which performs nonlinear transformations and deep processing on the fused features output by the attention layer to further enhance the model's representation capabilities. The decoder consists of two layers: an attention fusion layer, which performs secondary attention fusion between the high-level features output by the encoder and the current decision state to ensure that the generated instructions focus on the most relevant contextual information; and an action output layer, which maps the fused features to specific, interpretable action parameters, including pairing weights, trajectory adjustment amounts, and power adjustment coefficients. (2) Heterogeneous attention head design (8, divided into 3 functional categories): The first two task priorities are: weight matrix Focus on the core characteristics of the task, such as the target quantity and throughput requirements; The first three elements of state awareness: weight matrix It focuses on the state characteristics of the drone, such as its remaining energy and location; The first three elements of environment adaptation are: weight matrix. Focusing on environmental characteristics such as channel gain and obstacle distribution; (3) Perform encoder calculations: Feature mapping: For standardized 64-dimensional features Perform linear projection, the formula is: ,in For the projection matrix, For bias terms; Individual attention head calculation: For each attention head, the calculation is performed according to the following formula: ; In the formula, the dimension of a single attention head , , , 、( It is the first The size of the Query projection matrix, It is the key projection matrix. (This is the Value projection matrix). Multi-head fusion: The outputs of eight attention heads are concatenated and then projected, using the following formula: ; in, To fuse the projection matrix, the fused features are obtained. ; Feed-Forward Transform: Performs a non-linear transformation on the fused features, with the following formula: ; In the formula, , This is the weight matrix; , The bias term is the output encoder's final feature. ( = ); (4) Perform decoder calculations: Attention fusion: converting encoder output With the current decision-making status Perform secondary fusion, the formula is as follows Among them, the decision state dimension ; Action output: fused features The mapping is divided into three types of action parameters, with the following formulas: Pairing weight coefficients: Normalized to by the Sigmoid function ( Let be the projection matrix. (for bias terms) Trajectory adjustment coefficient: Constrained by the Clip function ( Let be the projection matrix. (for bias terms) Power adjustment coefficient: Constrained by the Clip function ( Let be the projection matrix. (for bias terms) S504. Set uniform parameter initialization rules to ensure training convergence efficiency: all weight matrices follow a normal distribution. All bias terms ( , All parameters (e.g., 0.01) are initialized to 0.01 to avoid neuron initialization saturation; a dropout layer is added after the encoder's Feed-Forward network layer with a dropout rate of 0.1 to prevent overfitting during training; the initialized parameters are stored uniformly for easy retrieval and updating during training. S505. The REINFORCE algorithm is used to complete parameter training, with the system synthesis objective function as the basis. As a reward signal: (1) Construct training dataset: Generate trajectory data for 5 independent scenarios, with 1000 trajectories in each scenario and 100 time steps in each trajectory. The scenarios cover different numbers of targets (5-20) and environmental complexity. (2) Define the reward signal: Incorporate the S4 dynamic weights, the formula is: ; In the formula, , , These are the real-time dynamic weights for throughput, lifetime, and energy consumption, respectively. , , ; The reward signal is a connectivity indicator variable. It is a comprehensive evaluation function that quantifies the core optimization objectives of the system—total throughput, total energy consumption, and network connectivity—into a single scalar value. This signal serves as feedback for the reinforcement learning algorithm, guiding the policy network to autonomously explore and learn collaborative optimization strategies that can simultaneously improve transmission efficiency, extend network lifetime, and reduce energy consumption during the training process. This directly supports the core objective of this invention: "balancing 'fast transmission, long service life, and low energy consumption'."

[0025] (3) Calculate cumulative return: for each trajectory According to the formula calculate( (Discount factor) In the formula, From time The initial discount accumulates into rewards. To calculate the current starting time of the return, This represents the termination time of the trajectory. For traversal from arrive Time step index, It is the first Instant rewards earned at any time For a moment and The time interval is used to discount the rewards over time.

[0026] (4) Calculate the policy gradient: according to the formula Estimate the gradient. For the number of trajectories, For the action probability distribution, For the policy objective function Regarding strategy parameters The gradient is used to update the policy network. For trajectory indexing, traverse Sampling trajectory, For the first The total number of time steps for the trajectory. For time step indexes within a single trajectory, For the first In the trajectory Accumulated returns from discounts over time For the first Trajectory Actions performed at all times For the first Trajectory The environmental state at any given moment.

[0027] (5) Parameter iterative update: Learning rate Weight decay coefficient The network parameters are updated iteratively based on a batch size of 32 trajectories, with an iteration limit of 1000 times. (6) Convergence criterion: When the comprehensive objective function of 10 consecutive iterations converges... If the fluctuation is ≤1% and the average reward value of the 5 scenarios increases by ≥30% compared to the initial value, the training is considered to have converged. S506. Deploy the trained three-layer collaborative optimization model on a cloud platform, and collect three types of key information in real time at a sampling frequency of 1Hz through the IoT module (location coordinates are uniformly in latitude and longitude format): Surveillance mission information: target location (latitude, longitude, altitude), movement speed, required surveillance range, minimum throughput requirement. ; Drone status information: Real-time location, flight speed, and remaining energy of the drone being tracked or relayed. Communication range, current transmission power ; Environmental dynamics information: Channel gain Noise power spectral density Obstacle distribution and wind speed (affect flight energy consumption and channel stability); After data acquisition, a hybrid filtering preprocessing method is used to reduce noise interference. The specific formula is as follows: ; In the formula, This is the original data; This is the filtered data; S507. Convert the preprocessed data into a 64-dimensional feature vector and input it into the deployed model. The model outputs three types of action parameters: paired weight coefficients. 3D: Corresponds to the pairing priority of S1, with a value range of [0,1]; Trajectory adjustment coefficient (3D): Corresponds to the speed increment, steering angle increment, and altitude increment of S2, with a value range of [-0.2,0.2]; Power adjustment coefficient (1D): Corresponds to the transmit power adjustment ratio of S3, with a value range of [0.8,1.2]; Then, the algorithms are invoked: the dual-objective optimization precise pairing algorithm generates the optimal pairing relationship table based on the pairing weight coefficient; the throughput-aware hybrid trajectory planning algorithm updates the relay UAV trajectory sequence based on the trajectory adjustment coefficient; and the multi-constraint adaptive power optimization algorithm calculates the transmission power of each link based on the power adjustment coefficient. Final output scheme: Encapsulated into four types of structured results, transmitted wirelessly, the result is as follows: Pairing scheme: includes tracking / relay drone number, transmission rate, and total power consumption; Tracking plan: Includes latitude, longitude, altitude, speed, and turning angle every 1 second (including charging / returning trajectory); Power scheme: Includes transmit power for each link at each time step; Auxiliary information: Network lifetime, total energy consumption, and predicted throughput compliance rate; S508, Cloud-based continuous monitoring core indicator: Link connectivity Actual throughput Remaining energy Change in target position Packet loss rate Dynamic adjustment is triggered when any of the following conditions are met: Single change of target position ; Actual throughput ; Remaining energy of any drone ,in, To carry the maximum energy; Link connectivity Where 0.9 is the minimum connectivity threshold; The priority is adjusted as follows: insufficient energy > connectivity imbalance > insufficient throughput > target location change, with priority given to ensuring network lifetime; after triggering, based on the dynamic weight of the current moment according to the throughput-lifetime-energy consumption collaborative balance mechanism, the dual-objective optimization precise pairing algorithm, throughput-aware hybrid trajectory planning algorithm and multi-constraint adaptive power optimization algorithm are re-invoked for collaborative calculation, and the scheme update and instruction issuance are completed within 0.5s to ensure that the monitoring task is continuous and uninterrupted.

[0028] This invention also provides a device for cooperative trajectory planning and energy efficiency optimization of unmanned aerial vehicle (UAV) networks for multi-target surveillance, comprising: The first processing unit is used to perform feasible pairing of tracking drones and relay drones at the routing and pairing layer based on the three-dimensional distance between the tracking drone and the relay drone, LOS or NLOS path determination, channel gain quantization and rate constraint screening, find the globally optimal pairing combination through the Hungarian algorithm, and at the same time verify energy consumption and rate constraints, and output the pairing result. The pairing result includes a pairing relationship table and a charging request marked for the energy-crisis drone. The second processing unit is used to prioritize energy replenishment mode based on charging urgency at the topology and trajectory layers, adaptively switch between single-hop and multi-hop transmission modes based on link quality, generate smooth trajectories using multi-segment Dubins paths and rolling time-domain optimization, and achieve dynamic obstacle avoidance by integrating Kalman filter prediction of target position and velocity obstacle models. It plans charging and return paths for low-energy drones and outputs trajectory dynamics. The third processing unit is used to combine the pairing results with the trajectory dynamics, and in the power control layer, to construct a full-dimensional calculation model that integrates flight energy consumption and transmission energy consumption. Through a time slot iteration mechanism, the transmission power is adaptively adjusted based on the link rate, channel gain and remaining energy to achieve total energy consumption optimization under the premise of meeting communication quality constraints. The fourth processing unit is used to define a comprehensive objective function based on the total energy consumption optimization using a weighted sum method, dynamically allocate the weights of each objective according to the proportion of the UAV's remaining energy, and establish a linkage mechanism between the weights and the dual-objective optimization precise pairing algorithm, the throughput-aware hybrid trajectory planning algorithm, and the multi-constraint adaptive power optimization algorithm to ensure that the optimization actions at each level conform to the overall optimal requirements of the system. The fifth processing unit, based on the linkage mechanism, extracts task, UAV status, and environmental features through a multi-head heterogeneous attention strategy network. It uses the REINFORCE algorithm to train the pairing parameters of the routing and pairing layer, the trajectory planning and charging scheduling parameters of the topology and trajectory layer, and the power control parameters of the power control layer. The trained model is then deployed to the cloud platform, where data is collected in real time. The unit calls upon the dual-objective optimization precise pairing algorithm, the throughput-aware hybrid trajectory planning algorithm, and the multi-constraint adaptive power optimization algorithm for collaborative computation, outputting a structural optimization scheme. Based on a quantized threshold, the system status is monitored and dynamically adjusted, forming a closed-loop execution mechanism of "real-time perception - strategy reasoning - dynamic deployment".

[0029] like Figure 2 As shown, the system architecture of this invention includes: Tracking drones are used to collect target information and transmit the collected information to relay drones; A relay drone, which is telecommunication connected to the tracking drone, is used to receive the collected signals from the tracking drone and transmit them to the base station; The base station is telecommunication-connected to the relay drone and is used to receive signals from the relay drone and perform data aggregation. A refueling point for charging tracking and relay drones; The cloud platform is used to issue dispatch instructions to relay drones and tracking drones based on the aggregated data from the base stations.

[0030] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for cooperative trajectory planning and energy efficiency optimization of unmanned aerial vehicle (UAV) networks for multi-target surveillance, characterized in that, Includes the following steps: At the routing and pairing layer, feasible pairing of tracking drones and relay drones is performed based on the three-dimensional distance between the tracking drone and the relay drone, LOS or NLOS path determination, channel gain quantization and rate constraint screening. The globally optimal pairing combination is found through the Hungarian algorithm, while energy consumption and rate constraints are verified, and the pairing results are output. The pairing results include a pairing relationship table and a charging request for energy-crisis drones. At the topology and trajectory layer, the energy replenishment mode is prioritized based on the charging urgency. The single-hop or multi-hop transmission mode is adaptively switched based on link quality. A smooth trajectory is generated by using multi-segment Dubins paths and rolling time-domain optimization. Kalman filter is fused to predict the target position and velocity obstacle model to achieve dynamic obstacle avoidance. The system plans charging and return paths for low-energy drones and outputs the trajectory dynamics. Combining the pairing results with the trajectory dynamics, a full-dimensional calculation model integrating flight energy consumption and transmission energy consumption is constructed at the power control layer. The transmit power is adaptively adjusted based on link rate, channel gain and remaining energy through a time slot iteration mechanism to achieve total energy consumption optimization while meeting communication quality constraints. Based on total energy consumption optimization, a weighted sum method is used to define the comprehensive objective function. The weights of each objective are dynamically allocated according to the proportion of the UAV's remaining energy. A linkage mechanism is established between the weights and the dual-objective optimization precise pairing algorithm, the throughput-aware hybrid trajectory planning algorithm, and the multi-constraint adaptive power optimization algorithm to ensure that the optimization actions at each level conform to the overall optimal requirements of the system. Based on the linkage mechanism, the system extracts task, UAV status and environmental features through a multi-head heterogeneous attention strategy network. The REINFORCE algorithm is used to train the pairing parameters of the routing and pairing layer, the trajectory planning and charging scheduling parameters of the topology and trajectory layer, and the power control parameters of the power control layer. The trained model is deployed on the cloud platform, and data is collected in real time. The system calls the dual-objective optimization precise pairing algorithm, throughput-aware hybrid trajectory planning and multi-constraint adaptive power optimization algorithm to perform collaborative calculations and output structural optimization schemes. The system status is monitored based on quantized thresholds and dynamically adjusted, forming a closed-loop execution mechanism of "real-time perception-policy reasoning-dynamic deployment".

2. The method for cooperative trajectory planning and energy efficiency optimization of UAV networks for multi-target surveillance according to claim 1, characterized in that, Based on the three-dimensional distance calculation of tracking UAVs and relay UAVs, LOS or NLOS path determination, channel gain quantization, and rate constraint screening, feasible pairing of tracking UAVs and relay UAVs is performed. The globally optimal pairing combination is found through the Hungarian algorithm, while energy consumption and rate constraints are verified, and the pairing results are output. The specific steps include the following: Based on the horizontal coordinates of the UAV's real-time location, the horizontal distance between the tracking UAV and the relay UAV is calculated using the horizontal distance formula, laying the groundwork for the calculation of three-dimensional distance and channel gain. The horizontal distance calculation formula is as follows: ； In the formula, To track drones With relay drones Horizontal distance; To track drones The x-coordinate of the horizontal coordinate; To track drones The horizontal coordinate and the vertical coordinate; For relay drones The x-coordinate of the horizontal coordinate; For relay drones The horizontal coordinate and the vertical coordinate; the set of tracking drones is denoted as The relay drone ensemble is denoted as And satisfy To ensure sufficient relay resources; Combining horizontal distance and flight altitude, the three-dimensional distance between the tracking UAV and the relay UAV is calculated using a three-dimensional distance formula. This distance is a core factor affecting channel attenuation. The three-dimensional distance formula is as follows: ； In the formula, To track the three-dimensional distance between the drone and the relay drone; To track drones Flight altitude; For relay drones Flight altitude; and simultaneously introduce communication distance constraints: , The maximum communication range is 5km by default; Based on the relative elevation angle between the tracking drone and the relay drone, the LoS probability and NLoS probability are calculated using the LoS probability formula and the NLoS probability formula, respectively. The LoS or NLoS propagation path is then determined to distinguish the signal propagation path type, ensuring that the channel model closely matches the actual scenario. The LoS probability formula is: ； In the formula, This represents the Loss of Sorrow (LoS) probability. for and The angle of elevation between them; , To adapt environmental parameters for the scene; therefore, the larger the elevation angle, the higher the probability of Loss of Horizon (LOS). NLoS probability formula: ； In the formula, This represents the NLoS probability. This represents the Loss of Sorrow (LoS) probability. By first introducing the path loss index This describes the rate attenuation of the signal as it travels further, and is then dynamically configured based on the monitoring environment, including unobstructed scenarios. Dense urban areas Calculate the average channel gain: The channel gain for LoS and NLoS scenarios is calculated using the following formulas respectively: ； ； In the formula, The reference channel gain per unit distance; The three-dimensional distance between the tracking drone and the relay drone; This is the path loss index. , This introduces additional path loss; Then, the average channel gain is calculated using the average channel gain formula, where the average channel gain is the probability-weighted sum of the two scenarios. The specific formula is as follows: ； In the formula, This represents the Loss of Sorrow (LoS) probability. Channel gain in the LoS scenario; This represents the NLoS probability. Channel gain in NLoS scenarios; Let be the average channel gain, where A larger value indicates less signal attenuation and better communication quality, providing core input for subsequent transmission rate calculations; Combining signal-to-noise ratio and channel parameters, the single-pair communication capability is accurately quantified using the following formula, while clearly defining the rate constraint: ； In the formula, For relay drones To the receiving node The achievable data transmission rate; For communication bandwidth; To track drones The transmission power; The residual self-interference coefficient; The noise power spectral density; The link interference power is considered; a minimum required transmission rate constraint is also introduced. ,Require Only then can the subsequent matching and screening process begin; A weighted bipartite graph is constructed to screen feasible pairings. Specifically, based on constraints of transmission rate and communication distance, pairing combinations that meet basic communication requirements are screened to narrow down the optimization range. For each group Determine whether the following conditions are met simultaneously: and If the above two conditions are met, then a bipartite graph is established. and The edge weight is set to If any condition is not met, no edge is built; ultimately, a weighted bipartite graph is formed. ,in Let be the set of feasible edges. The set of edge weights; By setting a rate threshold, high-quality potential matches can be screened. For each tracking drone Calculate the maximum transmission rate among all feasible pairings, and use it as the rate threshold. : ； If a tracking drone is set If there is no associated edge, it is marked as a node to be filled. The problem can be solved by lowering the threshold or scheduling a backup relay to avoid overall pairing failure. Generate unweighted subgraphs, simplified matching space traversal, weighted bipartite graphs. All edges If the edge weights satisfy If the edge is unweighted, retain it and remove its weight attribute; otherwise, delete it. Here, the threshold is relaxed by 10% to avoid over-filtering, ultimately generating an unweighted binary graph. ,in This is the simplified set of feasible edges; The calculation logic for energy consumption cost per pair is clearly defined. The total energy consumption is calculated using the following formula, which is the sum of the energy consumption of the tracking UAV, the energy consumption of the relay UAV, and the transmission energy consumption. The formula is as follows: ； In the formula, To track the energy consumption coefficient of drone flights; for The three-dimensional distance to the monitored target; The energy consumption coefficient for relay drone flight; for The three-dimensional distance to the base station; Transmission energy consumption coefficient; for The transmission power; The duration of a single communication session is considered; simultaneously, the maximum allowable total energy consumption for a single pair is introduced. ,Require ; For unweighted subgraphs Let there be an edge that does not exist in the middle. Marked as infeasible pairings, and finally constructed. Energy consumption cost matrix ; Then, the Hungarian algorithm is used for optimization. With the goal of "minimizing total energy consumption cost", the Hungarian algorithm is called to find the maximum match and output the preliminary matching results. The final pairing scheme is filtered out by performing energy consumption verification on the initial pairing results and eliminating combinations that exceed the limit. Specifically, the energy consumption of each combination in the initial pairing results is calculated. Total energy consumption ;like If, then keep the combination; if The suboptimal pairing is selected and re-verified until the constraints are met; finally, a structured pairing relationship table is output, which provides explicit input for the subsequent topology layer and power control layer, ensuring the system's collaborative optimization effect; Energy Status Assessment and Charging Request Generation: After completing the pairing scheme output, the system immediately assesses the remaining energy of all drones. ; Charging request determination: If the drone The remaining energy is below the charging request threshold. ,Right now If so, the drone is marked as being in a charging request state, denoted as ; Charging urgency calculation: The charging urgency is calculated for each drone requesting charging using the following formula. : ； In the formula, This is the estimated remaining mission time for the drone; This is the total task duration; These are the weighting coefficients; where urgency is... The higher the value, the more priority the drone needs to be given to charging.

3. The method for cooperative trajectory planning and energy efficiency optimization of UAV networks for multi-target surveillance according to claim 2, characterized in that, The optimization using the Hungarian algorithm, with the objective of minimizing total energy consumption cost, involves calling the Hungarian algorithm to find the maximum match and outputting preliminary pairing results. Specifically, this includes the following steps: Input parameters: Energy cost matrix Total number of tracking drones Total number of relay drones Minimum transmission rate constraint ; Matrix preprocessing: Preprocessing all data in the energy cost matrix... The value is replaced with 10 times the maximum finite value in the matrix to avoid algorithm calculation errors, while retaining the characteristic that "infeasible pairing" has the lowest priority. The standard Hungarian algorithm is invoked: based on the preprocessed matrix, minimum weight matching logic is executed, and an initial pairing result is output, which includes... The value of is the relay unit corresponding to each tracking device; Rate verification: Check all pairs in the initial pairing results. The link ensures its transmission rate. If there is a link that does not meet the requirements, set the energy cost of that link to ∞, and re-execute the matrix preprocessing and call the standard Hungarian algorithm steps. Output feasible pairings: If all links meet the rate constraints, output the current pairing results. If there are still links that do not meet the constraints, proceed to subsequent iterations for adjustment. The objective function and constraints are as follows: Objective function: ； Constraints: ； in, Total energy cost for all pairs; For 0-1 decision variables, Indicates tracking drones With relay drones pair, Indicates no pairing; Constraints This means that each tracking drone must be paired with one and only one relay drone; This means that each relay drone can be paired with a maximum of one tracking drone; This indicates that the decision variable can only take two states: "paired" or "unpaired". A successfully paired link must meet the minimum transmission rate requirement to ensure that the throughput meets the standard. If a complete match is not achieved, the unpaired tracking drone will be... Reduce by 10%, regenerate the subgraph and repeat the matching, with the iteration limit set to 5 times.

4. The method for cooperative trajectory planning and energy efficiency optimization of UAV networks for multi-target surveillance according to claim 1, characterized in that, The process of prioritizing energy replenishment modes based on charging urgency, adaptively switching between single-hop and multi-hop transmission modes based on link quality, generating smooth trajectories using multi-segment Dubins paths and rolling time-domain optimization, and achieving dynamic obstacle avoidance by fusing Kalman filter prediction of target position and velocity obstacle models to plan charging and return routes for low-energy drones includes the following steps: Energy replenishment demand priority assessment: When the system detects a charging request, it must prioritize ensuring the energy replenishment of critical nodes before making decisions on the normal communication mode. The specific steps are as follows: The system obtains the charging request list in real time to trigger the charging schedule. and its urgency Define the charging scheduling trigger flag : ； in, This is the urgency threshold; exceeding this value indicates an urgent need for charging. This is the critical energy value; below this value, there is a risk of crash. Key node identification, if The system further identifies which of the drones requesting charging are key network nodes, including the set of key nodes. satisfy: ； in, For nodes The cost of leaving network connectivity; Pattern Decision Logic: like The system immediately enters the charging priority scheduling mode, giving priority to charging for... The drone charging scheduling scheme in the system may temporarily switch to a degraded maintenance mode during this period. like but The system enters a hybrid mode, processing charging scheduling tasks and regular communication mode decisions in parallel or jointly optimizing them within an optimization window. That is, when planning the trajectory, the communication link quality and the scheduling needs of the charging drone are considered simultaneously. like The system executes a single-hop or multi-hop decision-making process. Single-hop or multi-hop mode decision process: Hybrid trajectory planning is used for mode decision, calculating the joint probability that all tracking UAVs and relay UAVs can communicate directly via single-hop mode, as shown in the following formula: ； ； In the formula, express The probability of single-hop network connectivity at any given time is the joint probability that all tracking drones and relay drones are connected in a single hop. For the first The single-point, single-hop connectivity probability between the tracking drone and the relay drone; R is the area of ​​the intersection of the two circles; R is the radius of the relay UAV communication circle. For the first The radius of the circle representing the positional uncertainty of the tracking drone; for Time of the first The straight-line distance between the tracking drone and the relay drone; For the first A tracking drone The horizontal coordinate in the two-dimensional plane at a given time; For relay drones in The x-coordinate in the two-dimensional plane at time t; For the first A tracking drone The ordinate in the two-dimensional plane at that moment; Relay drones The x-coordinate in the two-dimensional plane at time t; like Furthermore, the single-link transmission rates of all tracking drones and relay drones meet the requirements. The single-link packet loss rate meets the requirements. Maintain single-hop mode; if Or there exists a single-link transmission rate between any tracking drone and relay drone. Or there may be packet loss rate on any single link. Switch to multi-hop mode; in, This is the threshold for the connectivity probability of a single-hop network. This is the maximum acceptable packet loss rate threshold for a single link; The single-link packet loss rate, i.e., the first... The packet loss rate of the single-hop link between the tracking drone and the relay drone. For the first The single-hop link transmission rate between the tracking drone and the relay drone. Minimum rate requirement for a single link; In single-hop trajectory planning, a center-localization algorithm is used to calculate the optimal center position of the relay UAV, as shown in the following formula: ； Its analytical solution is to track the geometric center of the drone swarm: ； In the formula, for Optimal center position coordinates of the relay drone at any time; For the first Tracking drones Time and location coordinates To track the total number of drones; The formula for guaranteeing transmission rate in single-hop mode is as follows: ； in, To ensure communication redundancy, the single-hop link transmission rate of all tracking UAVs and relay UAVs meets the requirements. To avoid rate attenuation due to proximity to the communication range boundary; In multi-hop mode path selection, the relay path with the highest throughput is selected using the following formula: ； Among them, total path throughput The calculation is as follows: ； In the formula, for Time-optimal multi-hop relay path; The set of all feasible multi-hop paths; For the first The transmission power of the link; For the first Channel gain of the link; Noise power; For the first Interference power of the link; Multi-segment Dubins path and rolling temporal optimization to construct a kinematic model for a relay UAV: ； The constraints are: ; In the formula, For relay drone continuous time The following position coordinates; For flight speed, For heading angle; For turning rate; These are the upper and lower speed limits, respectively. Maximum steering ratio; This is the maximum acceleration; The rolling time-domain optimization window is The optimization objective is to maximize the average connectivity probability within the window. ； In the formula, The path is composed of a curve-straight line-curve or a curve-curve-curve combination, representing the connectivity probability of the corresponding pattern. This refers to the number of time steps, or the duration of a single step. ; This is the starting time of the current optimization window; Predicting the future position of a tracking drone based on Kalman filtering. Tracking drone location sequence This allows relay drone trajectory planning to adapt to target movement in advance, reducing the risk of connection loss. The formula is as follows: Equations of state: ; State update equation: ; Kalman gain calculation: ; Error covariance calculation: ; In the formula, For the first Tracking drones State estimate at time step Here is the state transition matrix. To control the input matrix, To track the control input of the drone, These are sensor observations. The observation matrix; The state estimation error covariance, To observe the noise covariance, It is the identity matrix; The speed obstacle model generates collision cones to achieve dynamic obstacle avoidance and associated trajectory optimization. The code is as follows: ； In the formula, for Real-time relay drone and obstacles The collision cone; Location of the relay drone; Location of the obstacle; For the speed of the obstacle, The candidate velocity vector for the relay UAV; Collision cone triggers replanning and multi-aircraft synchronous arrival, if the relay drone candidate speed This triggers local replanning, adjusting the parameters of the Dubins path segments; synchronous arrival constraints: ； In the formula, For the first Segment path length; This represents the flight speed of the corresponding sub-path; This refers to the time it takes for multiple drones to arrive at the target area simultaneously. Energy sensing optimization, the formula is as follows: ； In the formula, Energy consumption weighting coefficient The weighting coefficient is related to the flight speed. Positive correlation; Energy consumption constraints are applied to high-speed maneuvers to reduce energy waste and prevent drones from prematurely leaving the network due to energy depletion. The formula is as follows: ； In the formula, For the energy consumption rate of relay drones, For acceleration, This refers to the air drag coefficient; For flight speed; Dynamic scheduling and trajectory planning for energy replenishment to address the risk of network lifetime interruption caused by the limited energy of drones: The system receives charging requests. and its urgency The system intelligently schedules and plans trajectories for these requests to ensure timely replenishment of energy for critical drones while maintaining overall network performance. The specific process is as follows: Candidate supply station reachability analysis: For Each drone in Calculate the reachable set based on its current position. Remaining energy Energy consumption per unit distance Calculate the set of all supply stations that can be safely reached: ； In the formula, For a fixed set of supply points; A collection of mobile charging stations; For three-dimensional distance; if If so, the emergency procedure will be triggered, directing the drone to fly to the nearest safe point at the most economical speed or dispatching a mobile charging station to provide assistance. Optimal Supply Station Cost Assessment and Assignment: For each reachable candidate supply station Calculate the overall cost of going to this station to charge. : ； In the formula, In terms of economic speed The time cost of flight; This is the estimated total time spent at the station. For mobile charging stations, this value is related to their current task queue and charging power. and its own energy Related; It is a drone Leaving the post affects the network's global objective function The estimated negative impact, which is calculated through rapid simulation. The network state after departure is obtained, and the difference in the objective function is calculated. Optimal assignment: to the drone requesting charging. Choose the supply station with the lowest overall cost: ； This decision also applies to mobile charging stations. The next service target and destination were generated; Charging task trajectory planning for each assigned drone and its target supply station Plan the complete mission trajectory: The charging trajectory is generated using the Dubins path and rolling time-domain optimization method. arrive The flight path is based on economic cruising speed and integrates dynamic obstacle avoidance; Return-to-work route: After completing the planned charging, from... The trajectory returns to its optimal standby position, which depends on the drone's position. The network status at the expected return time is used to recalculate the optimal center position of the relay UAV in single-hop mode trajectory planning using a center positioning algorithm, or to re-select the relay path with the highest throughput in multi-hop mode path selection based on the network status at the expected return time. At the same time, the pre-calculated temporary network reconstruction scheme for maintaining network performance is executed, and corresponding transition trajectories are planned for other drones involved in the adjustment. Finally, the updated dynamic trajectory coordinate sequence of the relay drone is output, which has integrated the charging scheduling task trajectory and the network transition trajectory. Connectivity test function, defined Indicator variables for network connectivity at any given time: ； express The network always meets connectivity requirements; express The network connection is constantly interrupted; In the formula, the multi-hop connectivity probability , Let be the link connectivity probability vector. The network graph is a connected state matrix; the connectivity of the network graph is verified by depth-first search: if there exists a path that makes all drone nodes reachable from each other, then it is considered connected. For the first Single-link packet loss rate of tracking drones and relay drones; The effective time step count (Countconnected) is calculated from the initial time step. Start, cumulatively and continuously satisfy The number of time steps until the first occurrence. To stop counting, use the following formula: ； If there are no interruptions throughout the entire process ,but , This represents the maximum simulation time step. By optimizing connectivity probability, avoiding collisions, reducing energy consumption, and ensuring supply, the maximum The duration of time, thereby extending network lifespan. The derivation formula is as follows: ； In the formula, The duration of a single time step.

5. The method for cooperative trajectory planning and energy efficiency optimization of UAV networks for multi-target surveillance according to claim 1, characterized in that, The process involves combining the pairing results with the trajectory dynamics to construct a comprehensive calculation model that integrates flight energy consumption and transmission energy consumption. Through a time-slot iteration mechanism, the transmit power is adaptively adjusted based on link rate, channel gain, and remaining energy to optimize total energy consumption while meeting communication quality constraints. This includes the following steps: Derivation of the single-pair energy consumption calculation model: The total energy consumption of a single pair is the sum of the flight energy consumption of the tracking UAV, the single-pair flight energy consumption of the relay UAV, and the single-pair transmission energy consumption. Based on the energy consumption calculation logic and considering the characteristic of power dynamically adjusting with mission duration, the derivation is as follows: Single-pair flight energy consumption: ； In the formula, For 0-1 decision variables, a value of 1 indicates tracking the drone. With relay drones Pairing; a value of 0 indicates no pairing. To track the energy consumption coefficient of drone flights; To track drones The three-dimensional distance to the monitored target; The energy consumption coefficient for relay drone flight; For relay drones The three-dimensional distance to the base station; Power consumption for single-pair transmission: ； In the formula, Transmission energy consumption coefficient; Total task duration; For pairing At any moment The transmission power; Total energy consumption per pair: ； In the formula, For pairing Total energy consumption; Constructing power constraints to ensure communication quality: Power adjustment must prioritize communication quality, avoiding link disconnection or rate failure due to excessively low power. Combining rate constraints and link connectivity requirements, a set of hard constraints is formed. ； ； ； In the formula, For pairing At any moment The transmission power; Minimum transmission rate constraint; For communication bandwidth; The noise power spectral density; This refers to the link interference power. For a moment pair The average channel gain; The minimum connectivity threshold; Minimum transmission power; To achieve the maximum transmission power, all of the above constraints must be met simultaneously. ; An adaptive power adjustment strategy is designed to solve for energy consumption optimization: a time-slot iterative adjustment mechanism is adopted to dynamically optimize the transmit power based on the real-time link rate, channel gain, and the remaining energy state of the UAV, minimizing transmission energy consumption while satisfying constraints. Initial power allocation: ； In the formula, For pairing The initial transmit power; Pairing at the initial time The average channel gain; Real-time iterative adjustments are made, including the duration of time slots. Specifically: Total task duration Divided into Each time slot, for each time slot Power is dynamically adjusted based on the real-time link rate. like : ; like : And must meet ; like : ; If drone If it has been dispatched to the charging station, then its transmission power... Further conservative settings can be implemented, such as using a lower power limit while still meeting the minimum rate constraint. To conserve energy during the journey to the maximum extent possible and ensure a safe arrival at the charging station; If drone It is charging at the charging station. ; If drone Having just returned from a charging station with ample energy, the energy-saving restrictions on its power adjustment can be temporarily relaxed, prioritizing its rapid reconstruction of a high-quality link and leveraging its "full-power" advantage. In the formula, For a moment pair The transmission rate; This refers to the previous time slot. Energy perception correction: ； In the formula, Time of the first Tracking drone and the first The original transmit power (in W) of the relay UAV pairing link; for Time of the first The remaining energy of the relay drone, For the first The maximum energy of the relay drone is only when (When in a low-energy state, prioritizing energy preservation) this correction will be implemented, further reducing power by 10% while meeting the lower power limit, thereby extending the working time of the relay drone; This refers to the transmission power. Derivation of the formula for calculating the total system energy consumption: The total system energy consumption is the sum of the flight energy consumption and transmission energy consumption of all pairs. Combining the pairing decision variables and the dynamic power adjustment results, the formula for the total system energy consumption is obtained as follows: ； In the formula, This represents the total energy consumption of the system. To track the total number of drones; This represents the total number of relay drones; To track the flight energy consumption coefficient of drones; The flight energy consumption coefficient of the relay drone; Total energy consumption numerical calculation and output: Since the integral term is difficult to calculate directly, based on the time-slot iterative adjustment strategy, the integral term is discretized into a time-slot summation form, resulting in a numerical formula that can be directly substituted into the parameters for calculation. ； In the formula, Total number of time slots ; The duration of a single time slot; For a moment pair By substituting the transmission power, pairing results, trajectory dynamic parameters, and power adjustment results, the total energy consumption of the system can be directly calculated.

6. The method for cooperative trajectory planning and energy efficiency optimization of UAV networks for multi-target surveillance according to claim 1, characterized in that, The method employs a weighted summation to define the comprehensive objective function, dynamically allocates the weights of each objective based on the remaining energy percentage of the UAV, and establishes a linkage mechanism between the weights and the dual-objective optimization precise pairing algorithm, the throughput-aware hybrid trajectory planning algorithm, and the multi-constraint adaptive power optimization algorithm. This ensures that the optimization actions at each level align with the overall optimal requirements of the system. Specifically, the steps include the following: Definition of the global comprehensive objective function: The global comprehensive objective function of the system is constructed using the weighted sum method to quantify the synergistic optimization relationship of the three objectives. The specific formula is as follows: ； In the formula, This refers to the overall performance indicators of the system. Total network throughput; For the network's lifespan; This represents the total energy consumption of the system. For throughput weight, For the sake of survival options, As energy consumption weight, satisfy This is to ensure the rationality and binding nature of the weight allocation; Based on the remaining energy percentage of the drone, the weight percentage of the three objectives is dynamically adjusted to achieve adaptive switching of optimization focus at different stages. The weight values ​​can be flexibly adjusted according to the actual monitoring task requirements. Establish global weights to ensure that weight rules are implemented effectively: For route pairing: When making pairing decisions, calculate the comprehensive score of "throughput-energy consumption" based on the current weights, and prioritize the pairing combination with the best comprehensive score; For trajectory planning and charging scheduling: The dynamic weights of throughput, lifetime, and energy consumption directly regulate the topology and trajectory layer decisions. When the weights are biased towards lifetime, trajectory planning focuses on energy conservation and sustainability, adopts smooth and economical trajectories, and actively triggers charging scheduling to plan charging paths for key low-energy nodes, prioritizing energy replenishment to extend network lifetime. For power control: When allocating power, the power threshold is dynamically adjusted according to the energy consumption weight. The higher the weight, the more conservative the power allocation, giving priority to ensuring low energy consumption.

7. The method for cooperative trajectory planning and energy efficiency optimization of UAV networks for multi-target surveillance according to claim 1, characterized in that, The process involves extracting task, UAV state, and environmental features through a multi-head heterogeneous attention strategy network. The REINFORCE algorithm is used to train pairing parameters for the routing and pairing layer, trajectory planning and charging scheduling parameters for the topology and trajectory layer, and power control parameters for the power control layer. The trained model is deployed on a cloud platform, where data is collected in real time. A dual-objective optimization precise pairing algorithm, a throughput-aware hybrid trajectory planning algorithm, and a multi-constraint adaptive power optimization algorithm are used for collaborative computation to output a structural optimization scheme. Based on a quantized threshold, the system state is monitored and dynamically adjusted, forming a closed-loop execution mechanism of "real-time perception - policy reasoning - dynamic deployment." Specifically, this includes the following steps: The specific dimensions and meanings of the three types of heterogeneous input features are clearly defined to provide a data foundation for network modeling, specifically: The task features are 16 dimensions in total: target quantity (1 dimension), target average movement speed (1 dimension), minimum throughput requirement (1 dimension), monitoring range (1 dimension), and target distribution density (12 dimensions), divided according to the monitoring area grid. The drone's state characteristics consist of 24 dimensions: remaining energy. 1D, real-time location 3D, flight speed 1D, Current Transmission Power 1D, communication range 1D, charging request status 1D (0=no request, 1=request), charging urgency 1D, historical average energy consumption 15D, divided into 5-second time windows; Environmental characteristics comprise 24 dimensions: channel gain 1D, noise power spectral density 1D, obstacle density 1D, wind speed 1D, LosS propagation probability 1D, NLoS propagation probability 1D, environmental interference intensity 18D, divided into segments according to communication frequency; All heterogeneous features are integrated into a 64-dimensional feature vector to ensure that the network input dimension is consistent; Standardization and filtering processes are used to eliminate dimensional differences and noise interference, specifically: First, for each feature Normalization is performed to eliminate the influence of dimensions: ； In the formula, Features The statistical mean on the training dataset. Features The statistical standard deviation, These are the standardized eigenvalues; Then, outlier handling is performed: a reasonable range for feature values ​​is set, and outliers outside the range are replaced using interpolation. ； In the formula, This represents the eigenvalues ​​after correction or interpolation; The feature value of the previous time step. This serves as a feature value for the next time step, ensuring data continuity. Constructing a multi-head heterogeneous attention network architecture: An encoder-decoder structure is adopted, and eight heterogeneous attention heads are designed to achieve accurate feature extraction, specifically: Network hierarchy division: The encoder consists of three layers: a feature mapping layer, which linearly projects the input features to enhance feature representation and lay the foundation for subsequent attention calculation; a multi-head heterogeneous attention layer, which uses eight attention heads with different functional preferences to work in parallel to extract and focus key information from three heterogeneous dimensions: task, UAV state, and environment; and a feedforward network layer, which performs nonlinear transformation and deep processing on the fused features output by the attention layer to further enhance the model's representation ability. The decoder consists of two layers: an attention fusion layer, which performs secondary attention fusion between the high-level features output by the encoder and the current decision state to ensure that the generated instructions focus on the most relevant contextual information; and an action output layer, which maps the fused features to specific, interpretable action parameters, including pairing weights, trajectory adjustment amounts, and power adjustment coefficients. Heterogeneous attention head design: The first two task priorities are: weight matrix Focus on the core characteristics of the task, such as the target quantity and throughput requirements; The first three elements of state awareness: weight matrix It focuses on the state characteristics of the drone, such as its remaining energy and location; The first three elements of environment adaptation are: weight matrix. Focusing on environmental characteristics such as channel gain and obstacle distribution; Perform encoder calculations: Feature mapping: For standardized 64-dimensional features Perform linear projection, the formula is: ,in Let be the projection matrix. For bias terms; Individual attention head calculation: For each attention head, the calculation is performed according to the following formula: ； In the formula, the dimension of a single attention head , , , , It is the first The size of the Query projection matrix, It is the key projection matrix. It is the Value projection matrix; Multi-head fusion: The outputs of eight attention heads are concatenated and then projected, using the following formula: ； in, To fuse the projection matrix, the fused features are obtained. ; Feed-Forward Transform: Performs a non-linear transformation on the fused features, with the following formula: ； In the formula, , This is the weight matrix; , The bias term is the output encoder's final feature. ,in = ; Perform decoder calculations: Attention fusion: converting encoder output With the current decision-making status Perform secondary fusion, the formula is as follows Among them, the decision state dimension ; Action output: fused features The mapping is divided into three types of action parameters, with the following formulas: Pairing weight coefficients: Normalized to by the Sigmoid function ( Let be the projection matrix. (for bias terms); Trajectory adjustment factor: Constrained by the Clip function ( Let be the projection matrix. (for bias terms); Power adjustment coefficient: Constrained by the Clip function ( Let be the projection matrix. (for bias terms); Establish uniform parameter initialization rules to ensure training convergence efficiency: all weight matrices follow a normal distribution. All bias terms are initialized to 0.01 to avoid neuron initialization saturation; a dropout layer is added after the encoder's Feed-Forward network layer with a dropout rate of 0.1 to prevent overfitting; the initialized parameters are stored uniformly for easy retrieval and updating during training. The REINFORCE algorithm is used to train the parameters, and the system synthesis objective function is obtained. As a reward signal: Construct a training dataset: Generate trajectory data for 5 independent scenarios, with 1000 trajectories in each scenario and 100 time steps in each trajectory, covering different numbers of targets and environmental complexity; Define the reward signal: Incorporate dynamic weights, the formula is: ； In the formula, , , These are the real-time dynamic weights for throughput, lifetime, and energy consumption, respectively. , , ; The reward signal is a connectivity indicator variable. It is a comprehensive evaluation function that quantifies the core optimization objectives of the system—total throughput, total energy consumption, and network connectivity—into a single scalar value. Calculate cumulative reward: for each trajectory According to the formula calculate Discount factor; In the formula, From time The initial discount accumulates into rewards. To calculate the current starting time of the return, This represents the termination time of the trajectory. For traversal from arrive Time step index, It is the first Instant rewards earned at any time For a moment and The time interval is used to discount the rewards over time. Calculate the policy gradient: according to the formula Estimate the gradient. For the number of trajectories, For the action probability distribution, The policy objective function Regarding strategy parameters The gradient is used to update the policy network. For trajectory indexing, traverse Sampling trajectory, For the first The total number of time steps for the trajectory. For time step indexes within a single trajectory, For the first In the trajectory Accumulated returns from discounts over time For the first Trajectory Actions performed at all times For the first Trajectory The environmental state at any given time; Parameter iterative update: Learning rate Weight decay coefficient The network parameters are updated iteratively based on a batch size of 32 trajectories, with an iteration limit of 1000 times. Convergence criterion: When the comprehensive objective function of 10 consecutive iterations converges... If the fluctuation is ≤1% and the average reward value of the 5 scenarios increases by ≥30% compared to the initial value, the training is considered to have converged. The trained three-layer collaborative optimization model is deployed on a cloud platform, and three types of key information are collected in real time at a sampling frequency of 1Hz via an IoT module: Surveillance mission information: target location, movement speed, required surveillance range, minimum throughput requirements. ; Drone status information: Real-time location, flight speed, and remaining energy of the drone being tracked or relayed. Communication range, current transmission power ; Environmental dynamics information: Channel gain Noise power spectral density Obstacle distribution and wind speed; After data acquisition, a hybrid filtering preprocessing method is used to reduce noise interference. The specific formula is as follows: ； In the formula, This is the original data; This is the filtered data; The preprocessed data is converted into a 64-dimensional feature vector and input into the deployed model. The model outputs three types of action parameters: paired weight coefficients. 3D: Corresponds to the pairing priority of S1, with a value range of [0,1]; Trajectory adjustment coefficient (3D): Corresponds to the speed increment, steering angle increment, and altitude increment of S2, with a value range of [-0.2,0.2]; Power adjustment coefficient (1D): Corresponds to the transmit power adjustment ratio of S3, with a value range of [0.8,1.2]; Then, the algorithms are invoked: the dual-objective optimization precise pairing algorithm generates the optimal pairing relationship table based on the pairing weight coefficient; the throughput-aware hybrid trajectory planning algorithm updates the relay UAV trajectory sequence based on the trajectory adjustment coefficient; and the multi-constraint adaptive power optimization algorithm calculates the transmission power of each link based on the power adjustment coefficient. Final output scheme: Encapsulated into four types of structured results, transmitted wirelessly, the result is as follows: Pairing scheme: includes tracking / relay drone number, transmission rate, and total power consumption; Track scheme: includes latitude, longitude, altitude, speed, and turning angle every 1 second; Power scheme: Includes transmit power for each link at each time step; Auxiliary information: Network lifetime, total energy consumption, and predicted throughput compliance rate; Cloud-based continuous monitoring of key metrics: Link connectivity Actual throughput Remaining energy Change in target position Packet loss rate Dynamic adjustment is triggered when any of the following conditions are met: Single change of target position ; Actual throughput ; Remaining energy of any drone ,in, To carry the maximum energy; Link connectivity Where 0.9 is the minimum connectivity threshold; The priority is adjusted as follows: insufficient energy > connectivity imbalance > insufficient throughput > target location change, with priority given to ensuring network lifetime; after triggering, based on the dynamic weight of the current moment according to the throughput-lifetime-energy consumption collaborative balance mechanism, the dual-objective optimization precise pairing algorithm, throughput-aware hybrid trajectory planning algorithm and multi-constraint adaptive power optimization algorithm are re-invoked for collaborative calculation, and the scheme update and instruction issuance are completed within 0.5s to ensure that the monitoring task is continuous and uninterrupted.

8. A device for cooperative trajectory planning and energy efficiency optimization of unmanned aerial vehicle (UAV) networks for multi-target surveillance, comprising executing the method of any one of claims 1 to 7, characterized in that, include: The first processing unit is used to perform feasible pairing of tracking drones and relay drones at the routing and pairing layer based on the three-dimensional distance between the tracking drone and the relay drone, LOS or NLOS path determination, channel gain quantization and rate constraint screening, find the globally optimal pairing combination through the Hungarian algorithm, and at the same time verify energy consumption and rate constraints, and output the pairing result. The pairing result includes a pairing relationship table and a charging request marked for the energy-crisis drone. The second processing unit is used to prioritize energy replenishment mode based on charging urgency at the topology and trajectory layers, adaptively switch between single-hop and multi-hop transmission modes based on link quality, generate smooth trajectories using multi-segment Dubins paths and rolling time-domain optimization, and achieve dynamic obstacle avoidance by integrating Kalman filter prediction of target position and velocity obstacle models. It plans charging and return paths for low-energy drones and outputs trajectory dynamics. The third processing unit is used to combine the pairing results with the trajectory dynamics, and in the power control layer, to construct a full-dimensional calculation model that integrates flight energy consumption and transmission energy consumption. Through a time slot iteration mechanism, the transmission power is adaptively adjusted based on the link rate, channel gain and remaining energy to achieve total energy consumption optimization under the premise of meeting communication quality constraints. The fourth processing unit is used to define a comprehensive objective function based on the total energy consumption optimization using a weighted sum method, dynamically allocate the weights of each objective according to the proportion of the UAV's remaining energy, and establish a linkage mechanism between the weights and the dual-objective optimization precise pairing algorithm, the throughput-aware hybrid trajectory planning algorithm, and the multi-constraint adaptive power optimization algorithm to ensure that the optimization actions at each level conform to the overall optimal requirements of the system. The fifth processing unit, based on the linkage mechanism, extracts task, UAV status, and environmental features through a multi-head heterogeneous attention strategy network. It uses the REINFORCE algorithm to train the pairing parameters of the routing and pairing layer, the trajectory planning and charging scheduling parameters of the topology and trajectory layer, and the power control parameters of the power control layer. The trained model is then deployed on a cloud platform. Data is collected in real time, and the dual-objective optimization precise pairing algorithm, throughput-aware hybrid trajectory planning algorithm, and multi-constraint adaptive power optimization algorithm are called to perform collaborative calculations. The system outputs a structural optimization scheme, monitors the system status based on a quantized threshold, and dynamically triggers adjustments, forming a closed-loop execution mechanism of "real-time perception - strategy reasoning - dynamic deployment".

Images