An unmanned aerial vehicle (UAV) dynamic redeployment method based on predictive perception and joint triggering

Abstract

The present application relates to a kind of unmanned aerial vehicle dynamic redeployment method based on prediction perception and joint triggering, belong to unmanned aerial vehicle dynamic redeployment technical field of integration.The method includes: real-time state estimation is carried out to each ground user and extrapolation is obtained the predicted position of next time;Based on trend deterioration, fluctuation is severe, early warning and periodic forced fine-tuning are evaluated whether triggering unmanned aerial vehicle redeployment;After triggering, current user posteriori position and predicted position are combined into joint point set, construct the complex weight containing prediction weight, distance bias coefficient, sensing collaborative penalty coefficient and remote enhancement factor and as sample replication multiple, weighted clustering is executed, the new hovering plane coordinate of unmanned aerial vehicle is obtained and radial projection processing is applied and distance sorting competition access mechanism is executed, complete dynamic redeployment.The present application aims to solve the technical problems that the existing technology has single triggering mechanism, lack of prediction ability, sensing synergy is insufficient and edge coverage fairness is poor.

Description

Technical Field

[0001] This invention relates to a dynamic redeployment method for unmanned aerial vehicles (UAVs) based on predictive perception and joint triggering, belonging to the field of dynamic redeployment technology for UAVs. Background Technology

[0002] Following natural disasters such as earthquakes and floods, ground communication infrastructure often suffers devastating damage, leaving disaster areas isolated and isolated. Unmanned aerial vehicles (UAVs), with their advantages of rapid takeoff, maneuverability, and lack of ground-based limitations, have become a key technological tool in post-disaster emergency communication. Modern UAV platforms can simultaneously carry communication and radar sensing payloads, achieving Integrated Information and Communication Capability (ISAC). While providing communication connectivity, they also acquire crucial situational information such as the location of people and building damage through radar echoes.

[0003] However, the locations of disaster survivors change over time—approximately 20% of users relocate to the command center, while about 80% wander randomly within local areas—causing static deployment schemes based on the initial distribution to quickly become ineffective. Simulations show that if the static scheme is used directly without adjustments, the overall network communication coverage will drop from approximately 95% to approximately 61% within 50 time steps. Therefore, the system must have dynamic redeployment capabilities.

[0004] Current research on dynamic deployment of drones suffers from three main shortcomings. First, triggering mechanisms are generally based on simple rules or single thresholds, such as triggering redeployment when communication coverage falls below a certain fixed threshold, or triggering redeployment at fixed intervals. These methods lack a multi-dimensional comprehensive assessment of coverage degradation trends, fluctuation levels, and future predicted states, and typically only respond passively after coverage has deteriorated, exhibiting inherent lag. Research by Wang et al. (IEEE Transactions on Wireless Communications, 2019) points out that existing methods lack predictive mechanisms and cannot proactively adjust positions before coverage deteriorates. Second, there is a lack of mechanisms to prevent frequent redeployment. A single threshold may lead to continuous triggering due to occasional fluctuations, wasting the drone's limited flight energy and causing link jitter. Third, the redeployment phase often focuses solely on optimizing communication coverage, neglecting sensing and detection performance and coverage fairness for edge users, resulting in unilateral degradation of both sensing and communication functions.

[0005] Therefore, there is an urgent need for a dynamic redeployment method for drones that can integrate location prediction, multi-dimensional joint triggering, sensory collaborative optimization and edge coverage assurance, and has anti-vibration capabilities. Summary of the Invention

[0006] The purpose of this invention is to provide a dynamic redeployment method for UAVs based on predictive perception and joint triggering, which aims to solve the technical problems of existing UAV dynamic deployment triggering mechanisms lacking multi-dimensional comprehensive early warning and prediction capabilities, passive response leading to delayed coverage recovery, and inability to avoid oscillations caused by frequent redeployment, while ignoring the technical issues of sensory collaboration and edge coverage fairness.

[0007] To achieve the above objectives, the technical solution of the present invention is: a dynamic redeployment method for unmanned aerial vehicles (UAVs) based on predictive perception and joint triggering, comprising the following steps:

[0008] S1: Construct a system model; wherein the system model includes a ground user hybrid mobility model, an air-to-ground communication coverage model, and a radar sensing and detection model;

[0009] S2: Construct a Kalman filter based on a constant velocity motion model to perform real-time state estimation for each ground user, obtain posterior position estimation, and extrapolate to obtain the predicted position for the next moment;

[0010] S3: The system maintains the historical sequence of coverage and evaluates whether to trigger UAV redeployment based on four-dimensional joint judgment conditions; wherein, the four-dimensional judgment conditions include trend deterioration judgment, drastic fluctuation judgment, forward coverage early warning judgment calculated using predicted location, and periodic forced fine-tuning judgment. When any of the four judgment conditions is met and the minimum redeployment interval constraint is satisfied, the redeployment state machine is activated.

[0011] S4: After triggering redeployment, merge the current user's posterior location and predicted location into a joint point set, and construct a composite weight containing prediction weight, distance bias coefficient, synesthetic collaboration penalty coefficient and remoteness enhancement factor;

[0012] S5: Using the composite weight as the sample replication factor, perform weighted clustering to obtain the new hovering plane coordinates of the UAV. Apply radial projection processing to the new hovering plane coordinates according to the maximum flight radius constraint defined in the system model, and execute the distance sorting competition access mechanism to complete the dynamic redeployment.

[0013] Optionally, S1 specifically includes:

[0014] Ground user hybrid mobility models include probabilistic The behavior of moving in a directed manner toward the command base and the probability The behavior of randomly walking in a local area;

[0015] The air-to-ground communication coverage model uses the equivalent coverage radius. ,user Communication coverage is provided if and only if a drone is present. This makes the horizontal Euclidean distance between the user and the drone... ;

[0016] The radar sensing and detection model uses the equivalent detection radius. ,user Being sensed and covered if and only if a drone is present This enables the user to achieve a three-dimensional slant distance with the drone. ;

[0017] The combined sensor coverage rate is defined as the percentage of users covered by air-to-ground communication or radar sensing / detection out of the total number of users, expressed as:

[0018]

[0019] in, To improve the combined coverage of sensory input, For users Communication coverage indicator variable, For users Perception coverage indicator variable, Total number of users This is an indicator function.

[0020] Optionally, S2 specifically includes:

[0021] For the For each user, construct a four-dimensional state vector, expressed as follows:

[0022]

[0023] in, Indicates user At any moment The state vector consists of planar position coordinates and velocity components. and Represents the user at discrete time steps The plane projection coordinates, and On behalf of users shaft and Instantaneous velocity along the axial direction, Represented as a transpose matrix;

[0024] Based on the assumption of constant velocity motion, a discrete-time state transition equation is established. and observation equations ,in, Indicates user In the next discrete time step The predicted state vector is derived from the current state vector. via state transition matrix Extrapolate and superimpose system process noise get, Indicates at time step For users The actual noisy observation vector of the planar position is determined by the observation configuration matrix. Acting on the current state vector And superimposed measurement noise get;

[0025] At every moment The Kalman prediction-update recursion is performed in parallel for all users, and the predicted user position for the next time step is finally obtained by extrapolating the state transition matrix from the posterior state estimate. .

[0026] Optionally, the four-dimensional determination condition is specifically as follows:

[0027] Dimension 1: Deterioration trend determination, maintenance length is The coverage history sequence, take the most recent one. Step is the observation window, front Step is the reference window, when the mean of the reference window is... =and the mean of the observation window The difference exceeds the adaptive threshold Triggered at time;

[0028] Dimension Two: Determining Dramatic Fluctuations, Calculating the Standard Deviation of Coverage Within the Observation Window ,when Exceeding the preset fluctuation threshold Triggered at time;

[0029] Dimension 3: Forward-looking coverage warning judgment, using the user's predicted location at the next moment, calculates the predicted communication coverage at the next moment under the current deployment. ,when Below the warning threshold Triggered at time;

[0030] Dimension 4: Periodic forced fine-tuning determination, when the time interval since the last redeployment... Exceeding the set period It is forcibly triggered when the minimum redeployment interval constraint is met;

[0031] The four-dimensional judgment conditions, after logical synthesis, are all true and At that time, redeployment is activated, in which, This is the minimum redeployment interval.

[0032] Optionally, S4 specifically includes:

[0033] Set the current user's Kalman posterior estimate location set With the predicted location set Merge into a joint point set ;

[0034] right Assign a basic weight to any point in the middle. In this case, the base weight of the predicted position is higher than that of the current position;

[0035] Introducing a distance bias term ,in, The Euclidean distance from the point to the command base. The maximum possible distance in the region. This is the distance bias coefficient;

[0036] Introducing a synesthetic penalty coefficient When the difference between communication coverage and perceived coverage exceeds a set threshold, ,otherwise ;

[0037] The Euclidean distance from the command base exceeds the preset far-end distance threshold. Points were identified as belonging to remote hotspot areas, and for points belonging to remote hotspot areas, a remoteness enhancement factor was applied. ;

[0038] The final composite weights of each point are obtained, expressed as follows:

[0039]

[0040] in, For the first The composite weight of each point For the first The basic weight of each point.

[0041] Optionally, S5 specifically includes:

[0042] Using the composite weight of each point as the sample replication factor, the weighted K-means clustering algorithm is called and multiple independent initializations are performed to obtain the optimal value, thus obtaining the new hovering plane coordinates of each UAV.

[0043] Apply maximum flight radius to the new hovering plane coordinates Constraint: If the distance from the coordinates to the command base exceeds... Then, projecting radially to the boundary, the expression is:

[0044]

[0045] in, For planar projection coordinates, Center coordinates;

[0046] For each drone's communication coverage area, candidate users are sorted in ascending order of their Euclidean distance to the current drone's planar coordinates, and only the closest users are connected. A number of users complete dynamic redeployment under capacity constraints.

[0047] The beneficial effects of this invention are as follows: This invention achieves a shift from passive response to proactive prevention through Kalman prediction and four-dimensional joint triggering, which can trigger redeployment earlier than a single threshold scheme in extreme scenarios; through the minimum redeployment interval suppression mechanism, it effectively avoids energy waste and link jitter caused by frequent redeployment; through a composite weight function containing a synesthetic collaborative penalty term and a remote enhancement factor, it collaboratively optimizes communication and sensing coverage during redeployment and ensures that the coverage rate of remote hotspots reaches 92.12%, which can balance overall efficiency, edge fairness and synesthetic balance in extreme scenarios. Attached Figure Description

[0048] Figure 1 This is a diagram of the overall system architecture of the present invention;

[0049] Figure 2 This is a flowchart illustrating the complete process of the UAV network adaptive closed-loop control of the present invention.

[0050] Figure 3 This is a single-step recursive flowchart of the Kalman filter of the present invention;

[0051] Figure 4 This is a flowchart of the four-dimensional judgment condition evaluation process of the present invention;

[0052] Figure 5 This is a comparison chart of the overall coverage of the three strategies of this invention;

[0053] Figure 6 This is a comparison chart of the coverage performance of different deployment schemes of the present invention. Detailed Implementation

[0054] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0055] Example 1: As Figure 1 As shown, a dynamic redeployment method for unmanned aerial vehicles (UAVs) based on predictive perception and joint triggering is proposed. This method predicts the future location of users through Kalman filtering and utilizes a four-dimensional joint triggering and sensory collaborative redeployment mechanism to enable the UAV network to proactively and evenly adapt to the dynamic post-disaster environment. The method includes the following steps:

[0056] S1: Construct a system model; wherein the system model includes a ground user hybrid mobility model, an air-to-ground communication coverage model, and a radar sensing and detection model;

[0057] Optionally, the ground user hybrid mobility model includes probabilistic... The behavior of moving in a directed manner toward the command base and the probability The behavior of randomly walking in a local area;

[0058] The air-to-ground communication coverage model uses the equivalent coverage radius. ,user Communication coverage is provided if and only if a drone is present. This makes the horizontal Euclidean distance between the user and the drone... ;

[0059] The radar sensing and detection model uses the equivalent detection radius. ,user Being sensed and covered if and only if a drone is present This enables the user to achieve a three-dimensional slant distance with the drone. ;

[0060] The combined sensor coverage rate is defined as the percentage of users covered by air-to-ground communication or radar sensing and detection out of the total number of users. It is used as a quantitative indicator of system status, and its expression is:

[0061]

[0062] in, To improve the combined coverage of sensory input, For users Communication coverage indicator variable, For users Perception coverage indicator variable, Total number of users This is an indicator function.

[0063] Specifically, in this embodiment, the post-disaster emergency communication network adopts a two-tier "air-ground" architecture. The entire system operates in... rectangular disaster area The ground level consists of the emergency command base (central coordinates). ), The air layer consists of survivor users with mobile devices and ground scattering objects. It consists of an aerial base station for rotary-wing drones, operating at a fixed safe altitude. Hovering at the point, planar projection coordinates are marked. , Limited by range and control distance, the maximum flight radius of each drone is... ,satisfy:

[0064]

[0065] Survivors in the disaster area are non-uniformly clustered. Five hotspot clusters were generated using a Gaussian mixture model, and the parameters are shown in Table 1.

[0066] Table 1. Parameters of each survivor hotspot

[0067]

[0068] Survivors move according to a hybrid movement model: at each discrete time step (in, This represents the time interval between adjacent discrete time steps, which is taken in this embodiment. Step, reference step size (m / step), approximately 20% of users move directionally in the direction pointing to the base and are superimposed with Gaussian perturbations, while the remaining 80% engage in local random walks.

[0069] No. Individual users The planar position update equation at time t is:

[0070]

[0071] in, Indicates user In the next discrete time step Planar position coordinates, This represents the current position. This represents the displacement increment.

[0072] Air-to-ground communication coverage model defines the equivalent coverage radius of communication. .user With drones The horizontal Euclidean distance is .user Communication coverage indicator variable For if and only if make Otherwise .

[0073] Radar sensing and detection model defines the equivalent detection radius. Considering the three-dimensional slant distance, the user Perceptual coverage indicator variable For if and only if make Otherwise Maximum number of users served by a single drone When the capacity is exceeded, the previous entries are retained in ascending order of distance. name.

[0074] Furthermore, define the synesthetic joint coverage indicator variable. The expression is:

[0075]

[0076] This yields the synesthetic joint coverage rate, which serves as the unified basis for all subsequent trigger decisions and performance evaluations.

[0077] It is understood that the system model and synesthetic joint coverage rate constructed in this embodiment S1 provide the basis for calculating the historical sequence of coverage, the basis for calculating the synesthetic cooperative penalty coefficient, and the maximum flight radius constraint and capacity upper limit constraint.

[0078] S2: Construct a Kalman filter based on a constant velocity motion model to perform real-time state estimation for each ground user, obtain posterior position estimation, and extrapolate to obtain the predicted position for the next moment;

[0079] Optionally, for the first For each user, construct a four-dimensional state vector, expressed as follows:

[0080]

[0081] in, Indicates user At any moment The state vector consists of planar position coordinates and velocity components. and Represents the user at discrete time steps The plane projection coordinates, and On behalf of users shaft and Instantaneous velocity along the axial direction, Represented as a transpose matrix;

[0082] Based on the assumption of constant velocity motion, a discrete-time state transition equation is established. and observation equations ,in, Indicates user In the next discrete time step The predicted state vector is derived from the current state vector. via state transition matrix Extrapolate and superimpose system process noise get, Indicates at time step For users The actual noisy observation vector of the planar position is determined by the observation configuration matrix. Acting on the current state vector And superimposed measurement noise Obtain; among them, It follows a pattern with a mean of 0 and a covariance matrix of . Gaussian white noise distribution Represents a 4th-order identity matrix; Representing the multipath measurement error level in complex disaster-stricken ruin scenarios, it follows a mean of 0 and a covariance matrix of... Gaussian white noise distribution Represents a second-order identity matrix;

[0083] Optionally, in this embodiment, the state transition matrix and observation configuration matrix Specifically:

[0084]

[0085] At every moment The Kalman prediction-update recursion is performed in parallel for all users, and the predicted user position for the next time step is finally obtained by extrapolating the state transition matrix from the posterior state estimate. This is used for subsequent forward-looking early warning and joint point set construction.

[0086] Optionally, such as Figure 3 As shown, at each time The parallel execution of Kalman prediction-update recursion for all users is as follows:

[0087] In the forecasting phase:

[0088] The prior state is estimated as follows:

[0089] The prior covariance matrix is:

[0090]

[0091] During the update phase:

[0092] The Kalman gain is:

[0093]

[0094] The posterior state estimate is:

[0095]

[0096] The posterior covariance matrix is:

[0097]

[0098] Perform prediction output:

[0099] The predicted state at the next time step is obtained by extrapolating the state transition equation from the posterior estimate:

[0100]

[0101] Pick The first two dimensions constitute the predicted location set. .

[0102] S3: The system maintains the historical sequence of coverage and evaluates whether to trigger UAV redeployment based on four-dimensional joint judgment conditions; wherein, the four-dimensional judgment conditions include trend deterioration judgment, drastic fluctuation judgment, forward coverage early warning judgment calculated using predicted location, and periodic forced fine-tuning judgment. When any of the four judgment conditions is met and the minimum redeployment interval constraint is satisfied, the redeployment state machine is activated.

[0103] Optionally, the four-dimensional determination condition is specifically as follows:

[0104] Dimension 1: Deterioration trend detection is triggered when the short-term decline in coverage exceeds the adaptive threshold: maintenance length is... The coverage history sequence (in percentage form), take the most recent one. Step is the observation window, front Step is the reference window, when the mean of the reference window is... =and the mean of the observation window The difference exceeds the adaptive threshold Triggered at time, i.e. , The expression is: (This is a formula for dynamically adjusting based on the user's average movement speed.) , The current average movement speed of the user. The reference step size; in this embodiment, Take 6. Take 3. Pick / step;

[0105] It should be noted that the adaptive threshold The coefficient 0.8 in the expression was determined after repeated adjustments in simulation experiments, within the baseline step size. Pick Under the condition of / step, this embodiment tested different values ​​of the coefficient from 0.5 to 1.0. When the coefficient is too small (such as 0.5, 0.6), even a slight fluctuation in coverage will trigger redeployment. The drone has not yet had time to fly to the new location and has to take off again, causing unnecessary energy consumption and link jitter. When the coefficient is too large (such as 0.9, 1.0), the response is slow, and adjustments are only made after the coverage has dropped significantly. In this embodiment, the coefficient is 0.8, which can trigger when the coverage shows an early downward trend (generally a decrease of about 1% to 3%). This not only intervenes in advance to prevent further deterioration, but also avoids false triggers caused by normal random fluctuations, and has a relatively balanced sensitivity and anti-interference ability.

[0106] Dimension Two: Judgment of Dramatic Fluctuations. Triggered when the standard deviation of coverage within the observation window exceeds a preset threshold: Calculate the standard deviation of coverage within the observation window. ,when Exceeding the preset fluctuation threshold Triggered at time, i.e. In this embodiment, Take 1.2; the determination of dimension two is used to capture severe coverage fluctuations caused by sudden user gatherings or partial drone failures;

[0107] It should be noted that the preset fluctuation threshold The value of 1.2 was determined after repeated adjustments in simulation experiments: During simulation, when the system is running smoothly, the standard deviation of the coverage over the most recent three steps is generally stable between 0.5 and 0.8. When shocks such as drone malfunctions or a sudden increase in remote users occur, the standard deviation will jump instantly to over 1.5. In this embodiment, the value of 1.2 is significantly higher than the normal fluctuation range, yet it can effectively capture sudden anomalies and avoid false alarms during stable periods.

[0108] Dimension 3: Forward-looking coverage warning judgment, using the user's predicted location at the next moment, calculates the predicted communication coverage at the next moment under the current deployment. ,when Below the warning threshold Triggered at time, i.e. In this embodiment, Taking 88%; the judgment of dimension three enables the system to respond proactively before the coverage deteriorates, which is the key to realizing the transformation from passive response to proactive prevention.

[0109] It should be noted that the warning threshold The 88% threshold was determined after repeated adjustments in simulation experiments: In post-disaster emergency scenarios, communication coverage is generally required to be no less than 85%. In this embodiment, both the static and dynamic solutions maintain a stable coverage rate between 93% and 97% under normal conditions. This embodiment uses 88% as the forward warning threshold, leaving a 3-percentage-point buffer compared to the minimum requirement of 85%. Therefore, when the predicted coverage drops to 88%, the system can proactively adjust in advance, rather than passively responding only after the coverage falls below 85%, thus ensuring greater security.

[0110] Dimension 4: Periodic forced fine-tuning determination, when the time interval since the last redeployment... Exceeding the set period And it is forcibly triggered when the minimum redeployment interval constraint is met, that is and Forced triggering is implemented to prevent the system from falling into deep suboptimal conditions due to slow drift and prolonged lack of adjustment; in this embodiment... Take 10 steps;

[0111] It should be noted that the cycle The chosen interval of 10 steps was determined after repeated adjustments in simulation experiments. The total simulation duration was 50 steps. This embodiment tested different intervals. If the interval was too short (e.g., 5 steps), the drone would be flying almost continuously, resulting in high energy consumption and negatively impacting system stability. If the interval was too long (e.g., 15 or 20 steps), and other triggering conditions were not met for an extended period, the coverage might slowly drift to a suboptimal position without being corrected. In this embodiment, the 10-step interval is equivalent to approximately 5 fine-tuning adjustments throughout the process. This provides a safety net without excessively frequent flight, achieving an acceptable balance between energy consumption and coverage quality.

[0112] like Figure 4 As shown, the four-dimensional decision conditions, after being synthesized using OR logic, are all true and At that time, redeployment is activated, in which, To minimize the redeployment interval, 6 steps are used in this embodiment;

[0113] Furthermore, if This suppresses any triggering and delays it to the next evaluation cycle, effectively avoiding energy waste and link jitter caused by frequent redeployment.

[0114] It should be noted that the minimum redeployment interval The six steps were determined after repeated debugging in simulation experiments: minimum redeployment interval The value of this interval is related to the drone's flight speed. It takes approximately 4-5 steps for the drone to fly from the central base to the farthest boundary (500m). If the interval between two redeployments is less than this time, the drone will be pulled to a new location before the link has stabilized after completing its flight maneuver, causing severe service jitter and wasting a lot of onboard energy. This embodiment sets a minimum interval of 6 steps to ensure that the system can operate stably at the new location for at least one cycle after each deployment before evaluating the next triggering requirement, effectively suppressing continuous false triggers.

[0115] S4: After triggering redeployment, the current user's posterior location and predicted location are merged into a joint point set, and a composite weight is constructed containing prediction weight, distance bias coefficient, sensory coordination penalty coefficient and remoteness enhancement factor; wherein, the sensory coordination penalty coefficient is calculated based on the real-time difference between communication coverage and perception coverage;

[0116] Optionally, the current user's Kalman posterior estimated location set With the predicted location set Merge into a joint point set ,common One point;

[0117] right any point in the middle Assign basic weights The predicted location has a higher base weight than the current location, which makes it more attractive to the centroid in the clustering, guiding the drone closer to the user's future location. In this embodiment, the predicted location has a base weight of 1.5 and the current location has a base weight of 1.

[0118] Introducing a distance bias term ,in, For point Euclidean distance to the command base This represents the maximum possible distance within the region (distance from the corner to the center). This is the distance bias coefficient; in this embodiment, Pick , The distance bias term is set to 0.5; it gives higher weight to edge points far from the base, which in turn pulls the drones to spread outwards to reduce coverage redundancy in the central area.

[0119] Introducing a synesthetic penalty coefficient When communication coverage With perceived coverage When the difference exceeds the set threshold, that is hour, (In this embodiment, take) ), compressing all weights as a whole to weaken the pure density-driven effect, otherwise The synergistic penalty coefficient prompts redeployment schemes to adjust in the direction of narrowing the synergistic gap, avoiding over-optimization of single functional indicators.

[0120] The Euclidean distance from the command base exceeds the preset far-end distance threshold. Points were identified as belonging to remote hotspot areas, and for points belonging to remote hotspot areas, a remoteness enhancement factor was applied. ;

[0121] Specifically, in this embodiment, if point If it belongs to the remote hotspot areas D and E or their predicted points, then ;otherwise The remoteness enhancement factor increases the clustering influence of individual remote hotspot users by 5 times, fundamentally preventing deployment bias that prioritizes the center over the edge. The preset remote distance threshold... The specific value can be determined according to the scale of the actual scenario. In this embodiment, the coordinates of the emergency command base are (500, 500) m, and the centers of remote hotspots D and E are located at (250, 250) m and (750, 250) m respectively. The Euclidean distance from both to the command base is approximately 354 m. Accordingly, It can be set to 350m (or 400m depending on the area's edge) to accurately include edge hotspots in the remote enhancement range.

[0122] The final composite weights of each point are obtained, expressed as follows:

[0123]

[0124] in, For the first The composite weight of each point For the first The basic weight of each point.

[0125] S5: Combining weights As the sample replication factor (weighted by) The point is considered (with overlapping samples), perform weighted clustering to obtain the new hovering plane coordinates of the UAV. Apply radial projection processing to the new hovering plane coordinates according to the maximum flight radius constraint defined in the system model, and perform a distance sorting competition access mechanism according to the UAV capacity limit to complete dynamic redeployment.

[0126] Optionally, using the composite weight of each point as the sample replication factor, a weighted K-means clustering algorithm is called and multiple independent initializations are performed to obtain the optimal result, thus obtaining the new hovering plane coordinates of each UAV. In this embodiment, to suppress initial value sensitivity, 20 independent initializations are performed to obtain the optimal result (i.e., clustering is repeated 20 times, with different initial centroids randomly selected each time), and the result with the minimum sum of squared weighted distances within the cluster is selected. A new set of hovering plane coordinates for drones ;

[0127] Apply maximum flight radius to the new hovering plane coordinates Constraints: If coordinates The distance to the command base exceeds ,Right now Then, projecting radially to the boundary, the expression is:

[0128]

[0129] It is understood that this embodiment ensures that all drones are within the effective control range through projection;

[0130] A distance-ordered contention access mechanism under capacity constraints is implemented: For candidate users within the communication coverage area of ​​each drone, they are sorted in ascending order of their Euclidean distance to the current drone's planar coordinates, and only the closest users are accessed. One user is selected, and the remaining users are set to an inaccessible state, thus completing the dynamic redeployment under capacity constraints; in this embodiment, [the following is taken]. , Take 100.

[0131] Furthermore, the entire adaptive closed-loop control process is as follows: Figure 2 As shown: In each time step, the system sequentially executes user location update → coverage calculation → Kalman prediction → four-dimensional trigger evaluation → (if triggered) composite weight calculation → weighted K-means redeployment → constraint processing → next cycle, forming a seamless closed loop of "perception-prediction-decision-execution".

[0132] Understandably, in this embodiment, the Kalman prediction location is the key link throughout the entire process: the predicted value of S2 is directly used for forward warning in S3, and merged with the current point in S4 to form a joint point set; by assigning higher weights to the predicted points, the drone clustering center is guided to move closer to the user's future location in advance; at the same time, distance bias, synesthesia penalty, and remoteness enhancement mechanisms work together to ensure that each redeployment maintains overall efficiency while proactively tilting resources towards the edges and areas with weak perception. This closed-loop chain is the fundamental reason why this invention overcomes the lag and local optima defects of existing single-threshold or pure density-driven schemes.

[0133] Furthermore, to verify the effectiveness of the present invention, this embodiment also conducted simulation experiments on the MATLAB platform, and the main parameters are summarized in Table 2.

[0134] Table 2 Summary of Dynamic Simulation Parameters

[0135]

[0136] Furthermore, this experiment also set up three comparison strategies: static deployment (no adjustment after initial deployment), dynamic triggering (only trend, fluctuation, and periodic triggering, without prediction and synergistic coordination), and predictive deployment (this invention). In step 25, a drone malfunction (the 4th drone failed) and a sudden increase in users in remote hotspots (approximately 12 more people each) were simultaneously injected to simulate an extreme combined impact.

[0137] Furthermore, Figure 5 Coverage curves for three schemes are presented. The scheme of this invention can proactively trigger redeployment using predictive early warning after an impact, rapidly suppressing coverage degradation. Table 3 and... Figure 6 The average overall performance and remote hotspot coverage details of the three deployment schemes were further compared.

[0138] Table 3 Performance Comparison of Three Schemes

[0139]

[0140] from Figure 6As shown in Table 3, the remote hotspot coverage rate of the present invention (predictive deployment) reached 92.12%, which is 1.60 percentage points higher than the dynamic triggering scheme and 19.39 percentage points higher than the static deployment. Although the average overall coverage rate of this scheme (83.33%) is slightly lower than that of the purely density-driven dynamic triggering scheme in order to tilt resources towards vulnerable edge areas, it greatly improves the communication and perception quality of remote hotspots while ensuring a high overall service level, fully verifying the effectiveness and fairness of the prediction mechanism, sensory collaboration, and remote enhancement strategy in complex environments.

[0141] In summary, this invention constructs a system model and defines the synesthetic joint coverage rate as a quantitative indicator of system state; it performs state estimation and prospective prediction of user location based on a Kalman filter; and it constructs a joint triggering mechanism with four dimensions—coverage degradation trend, fluctuation intensity, predicted coverage warning, and periodic forced fine-tuning—using historical coverage sequences and predicted locations. Redeployment is triggered when any dimension satisfies and passes the minimum redeployment interval constraint. After triggering, a composite weight function containing prediction weights, distance bias coefficients, synesthetic joint penalty coefficients, and remoteness enhancement factors is constructed by fusing the current and predicted user locations. A redeployment scheme is generated through weighted K-means clustering.

[0142] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A dynamic redeployment method for unmanned aerial vehicles (UAVs) based on predictive perception and joint triggering, characterized in that, The method includes the following steps: S1: Construct a system model; wherein the system model includes a ground user hybrid mobility model, an air-to-ground communication coverage model, and a radar sensing and detection model; S2: Construct a Kalman filter based on a constant velocity motion model to perform real-time state estimation for each ground user, obtain posterior position estimation, and extrapolate to obtain the predicted position for the next moment; S3: The system maintains the historical sequence of coverage and evaluates whether to trigger UAV redeployment based on four-dimensional joint judgment conditions; wherein, the four-dimensional judgment conditions include trend deterioration judgment, drastic fluctuation judgment, forward coverage early warning judgment calculated using predicted location, and periodic forced fine-tuning judgment. When any of the four judgment conditions is met and the minimum redeployment interval constraint is satisfied, the redeployment state machine is activated. S4: After triggering redeployment, merge the current user's posterior location and predicted location into a joint point set, and construct a composite weight containing prediction weight, distance bias coefficient, synesthetic collaboration penalty coefficient and remoteness enhancement factor; S5: Using the composite weight as the sample replication factor, perform weighted clustering to obtain the new hovering plane coordinates of the UAV. Apply radial projection processing to the new hovering plane coordinates according to the maximum flight radius constraint defined in the system model, and execute the distance sorting competition access mechanism to complete the dynamic redeployment.

2. The method for dynamic redeployment of unmanned aerial vehicles based on predictive perception and joint triggering according to claim 1, characterized in that, Specifically, S1 is: Ground user hybrid mobility models include probabilistic The behavior of moving in a directed manner toward the command base and the probability The behavior of randomly walking in a local area; The air-to-ground communication coverage model uses the equivalent coverage radius. ,user Communication coverage is provided if and only if a drone is present. This makes the horizontal Euclidean distance between the user and the drone... ; The radar sensing and detection model uses the equivalent detection radius. ,user Being sensed and covered if and only if a drone is present This enables the user to achieve a three-dimensional slant distance with the drone. ; The combined sensor coverage rate is defined as the percentage of users covered by air-to-ground communication or radar sensing / detection out of the total number of users, expressed as: ； in, To improve the combined coverage of sensory input, For users Communication coverage indicator variable, For users Perception coverage indicator variable, Total number of users This is an indicator function.

3. The method for dynamic redeployment of unmanned aerial vehicles (UAVs) based on predictive perception and joint triggering, as described in claim 1, is characterized in that... Specifically, S2 is: For the For each user, construct a four-dimensional state vector, expressed as follows: ； in, Indicates user At any moment The state vector consists of planar position coordinates and velocity components. and Represents the user at discrete time steps The plane projection coordinates, and On behalf of users shaft and Instantaneous velocity along the axial direction, Represented as a transpose matrix; Based on the assumption of constant velocity motion, a discrete-time state transition equation is established. and observation equations ,in, Indicates user In the next discrete time step The predicted state vector is derived from the current state vector. via state transition matrix Extrapolate and superimpose system process noise get, Indicates at time step For users The actual noisy observation vector of the planar position is determined by the observation configuration matrix. Acting on the current state vector And superimposed measurement noise get; At every moment The Kalman prediction-update recursion is performed in parallel for all users, and the predicted user position for the next time step is finally obtained by extrapolating the state transition matrix from the posterior state estimate. .

4. The UAV sensor-integrated dynamic redeployment method based on predictive perception and joint triggering as described in claim 1, characterized in that, The specific four-dimensional determination conditions are as follows: Dimension 1: Deterioration trend determination, maintenance length is The coverage history sequence, take the most recent one. Step is the observation window, front Step is the reference window, when the mean of the reference window is... =and the mean of the observation window The difference exceeds the adaptive threshold Triggered at time; Dimension Two: Determining Dramatic Fluctuations, Calculating the Standard Deviation of Coverage Within the Observation Window ,when Exceeding the preset fluctuation threshold Triggered at time; Dimension 3: Forward-looking coverage warning judgment, using the user's predicted location at the next moment, calculates the predicted communication coverage at the next moment under the current deployment. ,when Below the warning threshold Triggered at time; Dimension 4: Periodic forced fine-tuning determination, when the time interval since the last redeployment... Exceeding the set period It is forcibly triggered when the minimum redeployment interval constraint is met; The four-dimensional judgment conditions, after logical synthesis, are all true and At that time, redeployment is activated, in which, This is the minimum redeployment interval.

5. The method for dynamic redeployment of unmanned aerial vehicles based on predictive perception and joint triggering according to claim 1, characterized in that, Specifically, S4 is: Set the current user's Kalman posterior estimate location set With the predicted location set Merge into a joint point set ; right Assign a basic weight to any point in the middle. In this case, the base weight of the predicted position is higher than that of the current position; Introducing a distance bias term ,in, The Euclidean distance from the point to the command base. The maximum possible distance in the region. This is the distance bias coefficient; Introducing a synesthetic penalty coefficient When the difference between communication coverage and perceived coverage exceeds a set threshold, ,otherwise ; The Euclidean distance from the command base exceeds the preset far-end distance threshold. Points were identified as belonging to remote hotspot areas, and for points belonging to remote hotspot areas, a remoteness enhancement factor was applied. ; The final composite weights of each point are obtained, expressed as follows: ； in, For the first The composite weight of each point For the first The basic weight of each point.

6. The method for dynamic redeployment of unmanned aerial vehicles based on predictive perception and joint triggering according to claim 1, characterized in that, Specifically, S5 is: Using the composite weight of each point as the sample replication factor, the weighted K-means clustering algorithm is called and multiple independent initializations are performed to obtain the optimal value, thus obtaining the new hovering plane coordinates of each UAV. Apply maximum flight radius to the new hovering plane coordinates Constraint: If the distance from the coordinates to the command base exceeds... Then, projecting radially to the boundary, the expression is: ； in, For planar projection coordinates, Center coordinates; For each drone's communication coverage area, candidate users are sorted in ascending order of their Euclidean distance to the current drone's planar coordinates, and only the closest users are connected. A number of users complete dynamic redeployment under capacity constraints.

Images