A drone cluster monitoring system and method

Abstract

The application discloses an unmanned aerial vehicle cluster monitoring system and method, and relates to the technical field of unmanned aerial vehicle cluster cooperative operation.The method comprises the following steps: continuously sensing the state of a plurality of preset task areas, acquiring state values of the areas and constructing a time sequence; calculating the state change rate and acceleration of each area based on the sequence; combining the historical task execution frequency of the area to calculate a dynamic task priority value; generating an unmanned aerial vehicle cluster task allocation scheme by using a preset optimization model according to the priority value of all areas, the position of the task area, the emergency degree, the resource demand constraint condition and the target of load balancing; planning a flight path for each unmanned aerial vehicle, driving the cluster to execute the task and supporting dynamic reallocation according to real-time feedback.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) swarm collaborative operation technology, specifically a UAV swarm monitoring system and method. Background Technology

[0002] Currently, using drone swarms for automated inspection and monitoring of wide-area spaces has become a key means of improving operational efficiency. Existing mainstream task scheduling methods typically follow a fixed paradigm at the data processing level: after each detection of a specific area, the system, based on a preset threshold or a simple model, assigns a discrete state label to the detection result, such as "normal" or "abnormal," "with target" or "without target." This label is recorded and serves as the core, or even the sole, basis for subsequent task planning.

[0003] The fundamental flaw of this paradigm lies in its static, discrete decision-making logic being disconnected from the continuously evolving dynamic patterns in real-world application scenarios. The state of the task area (such as the probability of target appearance, the degree of abnormal development, and the intensity of environmental disturbances) is essentially a continuous, time-gradually changing physical process. However, existing methods attempt to describe and predict this continuous process using a series of isolated, instantaneous "snapshot" judgments, creating an inherent cognitive dilemma: no matter how many times "normal" discrete labels are collected for a certain area, it is impossible to effectively characterize the continuous trajectory and evolutionary trend of its state from quantitative to qualitative change.

[0004] This contradiction directly leads to the fact that while the scheduling strategy based on discrete labels aims to improve efficiency, it may have the opposite effect. Since the label system cannot distinguish the rate of change of state in different regions, the system will inevitably ignore those key regions that are deteriorating rapidly or whose target activity is rising sharply, but whose current label is still "normal". As a result, the limited drone swarm resources cannot achieve dynamic and precise focusing, and are continuously mismatched with real risk points or task hotspots, until emergencies occur or targets are lost, resulting in delayed response and reduced operational efficiency.

[0005] Therefore, the core problem of existing technologies can be summarized as follows: the static task data system that relies on discrete state determination labels is incompatible with the dynamic physical law of continuous and gradual changes in the state of the task area, which makes it impossible for the UAV swarm scheduling system to effectively perceive and respond to the continuous change process of the state, thereby causing resource misallocation and scheduling inaccuracy. Furthermore, even though some existing solutions attempt to introduce simple task sorting or polling mechanisms, their scheduling decisions fail to be deeply coupled with the aforementioned continuous and dynamic risk evolution trends, and they also fail to systematically consider the spatiotemporal constraints (such as endurance, flight distance, and operation time) of drone swarms in actual execution. As a result, although the scheduling schemes they generate are theoretically optimized to a certain extent, they still cannot achieve truly efficient load balancing and rapid dynamic response in complex actual scenarios.

[0006] To address the above problems, this invention proposes a drone swarm monitoring system and method. Summary of the Invention

[0007] The purpose of this invention is to provide a drone swarm monitoring system and method to solve the problems raised in the prior art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A method for monitoring drone swarms includes the following steps: S1. Perform continuous state perception on several preset task areas and obtain the task area state value of each area at the current time. S2. Based on the state values ​​obtained by each task region at different times, construct a state time sequence for each region; S3. For each task region, calculate its state change rate and change acceleration based on its state time sequence. S4. For each task region, calculate the task priority value of the task region by combining its state change rate, change acceleration and historical task execution frequency. S5. Based on the task priority values ​​of all task areas, determine the execution order of each area in the next cycle, generate a task allocation scheme, and drive the drone swarm to execute.

[0009] S1 further includes the following: The building surface is divided into several monitoring areas; For each task area state type that needs to be monitored, a physical quantity that can take continuous values ​​and changes continuously with the continuous change of the structural state is determined as the observed value M for that area; The task area state type refers to the manifestation of continuous changes in physical, chemical, and geometric properties in the task area caused by the appearance of targets, the occurrence of anomalies, environmental changes, and the dynamic evolution of task requirements. At each scheduling moment, a drone equipped with the corresponding sensor is dispatched to the monitoring area to collect raw sensing data. By performing spatial feature analysis on the raw sensing data, the dominant feature quantity is extracted, and the statistical feature value of the dominant feature quantity is obtained through mathematical statistical processing methods. The statistical feature value is used as the value of the observation M of the area at the current scheduling moment. The dominant feature quantity refers to the set of physical parameters that are identified and separated from the original sensing data through spatial feature analysis, directly representing the degree of continuous evolution of the state type in the task area; Record the M value, region identifier, and timestamp of each region at the current moment, and store them together.

[0010] S2 further includes the following: From the associated stored data records, extract all data entries with the same region identifier. Each data entry contains a timestamp of a sensing moment and its corresponding region status value. All the extracted data entries are sorted in ascending order according to their timestamps; and the regional state values ​​in the sorted data entries are arranged and connected according to their corresponding timestamps to form a state time sequence of the task region. This sequence, ordered by time, records the evolution of the region's state value over time.

[0011] S3 further includes the following: For each region, the data points corresponding to the three most recent consecutive sensing moments are selected from the state time sequence of that region as the basis for calculation; these three moments are defined as t in chronological order. k-2 t k-1 t k ; where t k Representing the latest perception moment, the corresponding region state values ​​for these three moments are M, respectively. k-2 M k-1 M k ; Calculate the time interval between adjacent sensing moments, denoted as Δt, where Δt1 = t k-1 -t k-2 Δt2=t k -t k-1 When the sensing period is fixed, Δt1 and Δt2 are equal. The rate of change of state V k This represents the rate at which the regional state value changes within the latest time interval, and is calculated using the following formula: V k =(M k -M k-1) / Δt2; Among them, V k For in t k Rate of change of state at time t; The changing acceleration A k It represents the trend of the rate of change of state from the previous time interval to the current time interval, and the calculation formula is as follows: A k =(V k -V k-1 ) / ((Δt1+Δt2) / 2); Among them, V k-1 For in t k-1 Rate of change of state at time t; The calculated rate of change of state V at the current moment k With changing acceleration A k The output is associated with the task area identifier.

[0012] S4 further includes the following: Obtain the historical task execution frequency F for each task region; the historical task execution frequency F is the cumulative value of the number of times the region status value has been successfully collected in all executed tasks for that task region; For each task region, when the state change rate V k When >0, its positive change component P is defined. v =V k When V k When ≤0, its positive variation component P is defined. v =0; When the change in acceleration A k When >0, its positive acceleration component P is defined. A =A k When the change in acceleration A k When ≤0, its positive acceleration component P is defined. A =0; According to the positive change component P v The positive acceleration component P A And based on the historical task execution frequency F, the task priority value R for this task region is calculated using the following formula: R=(P v (1+P A )) / (1+F); The calculated task priority value R is associated with the task area identifier and output.

[0013] S5 further includes the following: Get all task regions and their corresponding task priority values ​​R; sort all task regions in descending order of their task priority values ​​R to generate a task region priority list. Based on the task area priority list, the first N task areas are selected sequentially from the list to form the target area set for the current task cycle; The target area set is integrated, and a task priority scheduling scheme is generated according to the priority order. The scheduling scheme is then sent to the control system of the UAV cluster to drive the cluster to execute the current round of monitoring tasks.

[0014] The process of integrating the target area set and generating a task priority scheduling scheme based on priority order, and then distributing the scheduling scheme to the control system of the UAV swarm to drive the swarm to execute the current monitoring task, further includes the following: For each task region j, collect the geographical cost parameter Tj, the task urgency parameter Rj, and the regional operation time parameter Wj; The geographical cost parameter Tj represents the estimated flight time required for the UAV swarm to fly from the current assembly point to the mission area j; The task urgency parameter Rj represents the priority value of the task region; The regional operation time parameter Wj represents the estimated time required for the UAV to complete the status perception operation in the mission area; With the maximum endurance Ci of each drone i in the drone swarm as a constraint, and minimizing the maximum total time of all drone tasks in the swarm as the optimization objective, a task allocation model is constructed. The task allocation model takes as input a set of all task regions to be allocated, parameter groups Tj, Rj, Wj for each region, a set of drone numbers and their endurance constraints Ci. The task allocation model sorts all regions in descending order based on the task urgency parameter Rj. Each time, it selects the region with the highest urgency from the top of the allocation pool and assigns it to the drone with the shortest estimated total time Wj and the endurance constraint Ci. After allocation, the task sequence and total time of the drone are updated. The iterative process continues until all regions are allocated. The task allocation model outputs an optimized task allocation scheme, which clarifies the sequence of task areas that each UAV i is responsible for. Based on the mission sequence of each UAV i, with the goal of minimizing the total flight distance of the UAV to visit all its assigned areas, the system uses a preset optimization algorithm to perform route planning and generate the optimal cruise path for the UAV. Integrate the flight paths of all UAVs, generate a cluster collaborative task instruction set, and drive the UAV cluster to perform monitoring tasks along the planned path; During mission execution, the system receives newly collected sensing data transmitted back by the UAV in real time. When a new batch of data is acquired, it is used as input to re-execute S1 to S4, calculate the updated mission priority value Rj for all mission areas, and based on the updated Rj, the real-time position of the UAV and the remaining flight time, output a new mission allocation scheme for the mission areas that have not yet been executed through the mission allocation model, thereby realizing the online dynamic reallocation of mission schemes.

[0015] A drone swarm monitoring system includes the following modules: The building environment monitoring system includes a regional status perception module, a status time sequence construction module, a dynamic trend calculation module, a task priority calculation module, and a cluster task scheduling module.

[0016] The area state perception module is used to continuously perceive the state of several preset task areas, and to acquire and associate the area state value and its identifier of each area at different times. The state-time sequence construction module is used to extract historical state values ​​for each task area from the stored data and construct a state-time sequence ordered by time. The dynamic trend calculation module is used to calculate the rate of change and acceleration of change of each task area based on the state time sequence of each task area. The task priority calculation module is used to calculate the task priority value of each task region by comprehensively considering the state change rate, change acceleration and historical task execution frequency of each task region. The cluster task scheduling module receives the task priority value R of each task region output by the task priority calculation module, and obtains the preset parameters of each task region. With the maximum endurance time of each drone in the drone swarm as a constraint and minimizing the maximum total time of all drone tasks in the swarm as the optimization objective, a load balancing task allocation model is constructed and solved to generate the task area sequence for each drone. Based on the mission area sequence of each UAV, a path planning algorithm is used to generate the optimal cruise path for it; and a cluster collaborative task instruction set is generated to drive the UAV cluster to perform monitoring tasks. During mission execution, the mission priority is recalculated in real time based on newly collected perception data, and unexecuted missions are dynamically reallocated based on the updated priority and the real-time status of the UAV.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention continuously acquires the state values ​​of each task region and constructs a time series. It not only calculates the rate of change of the state, but also calculates the acceleration of change, thereby capturing the rate of state deterioration and whether it is accelerating. This overcomes the shortcomings of traditional methods that rely only on single discrete judgments and cannot predict the evolution trend. 2. This invention proposes combining the rate of state change, acceleration of change, and historical task execution frequency to calculate task priority values ​​using a formula. This model considers both the severity and acceleration of the current deterioration, and automatically adjusts the level of attention through a historical frequency factor, allowing areas that have not been monitored for a long time but show signs of deterioration to receive higher priority, thus ensuring the objectivity and comprehensiveness of scheduling decisions. 3. This invention, based on dynamically calculated task priority values, further integrates multiple constraints such as the geographical location of the task area and the operation time, constructing an optimization model with load balancing as the goal for task allocation and flight path planning. This enables the system to not only focus on risks at the decision-making level but also achieve scientific and efficient allocation of UAV swarm resources at the execution level. By minimizing the maximum task time, it ensures the balance of the overall workload of the swarm, avoiding individual overload or idleness, thereby maximizing the monitoring coverage and efficiency of a single task under given endurance constraints. Attached Figure Description

[0018] Figure 1 This is a flowchart of a method for monitoring a drone swarm according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example: Figure 1 As shown, the present invention provides a technical solution. A method for monitoring drone swarms includes the following steps: S1. Perform continuous state perception on several preset task areas and obtain the task area state value of each area at the current time. S2. Based on the state values ​​obtained by each task region at different times, construct a state time sequence for each region; S3. For each task region, calculate its state change rate and change acceleration based on its state time sequence. S4. For each task region, calculate the task priority value of the task region by combining its state change rate, change acceleration and historical task execution frequency. S5. Based on the task priority values ​​of all task areas, determine the execution order of each area in the next cycle, generate a task allocation scheme, and drive the drone swarm to execute.

[0021] S1 further includes the following: The building surface is divided into several monitoring areas; For each task area state type that needs to be monitored, a physical quantity that can take continuous values ​​and changes continuously with the continuous change of the structural state is determined as the observed value M for that area; The task area state type refers to the manifestation of continuous changes in physical, chemical, and geometric properties in the task area caused by the appearance of targets, the occurrence of anomalies, environmental changes, and the dynamic evolution of task requirements. At each scheduling moment, a drone equipped with the corresponding sensor is dispatched to the monitoring area to collect raw sensing data. By performing spatial feature analysis on the raw sensing data, the dominant feature quantity is extracted, and the statistical feature value of the dominant feature quantity is obtained through mathematical statistical processing methods. The statistical feature value is used as the value of the observation M of the area at the current scheduling moment. The dominant feature quantity refers to the set of physical parameters that are identified and separated from the original sensing data through spatial feature analysis, directly representing the degree of continuous evolution of the state type in the task area; Record the M value, region identifier, and timestamp of each region at the current moment, and store them together.

[0022] S2 further includes the following: From the associated stored data records, extract all data entries with the same region identifier. Each data entry contains a timestamp of a sensing moment and its corresponding region status value. All the extracted data entries are sorted in ascending order according to their timestamps; and the regional state values ​​in the sorted data entries are arranged and connected according to their corresponding timestamps to form a state time sequence of the task region. This sequence, ordered by time, records the evolution of the region's state value over time.

[0023] S3 further includes the following: For each region, the data points corresponding to the three most recent consecutive sensing moments are selected from the state time sequence of that region as the basis for calculation; these three moments are defined as t in chronological order. k-2t k-1 t k ; where t k Representing the latest perception moment, the corresponding region state values ​​for these three moments are M, respectively. k-2 M k-1 M k ; Calculate the time interval between adjacent sensing moments, denoted as Δt, where Δt1 = t k-1 -t k-2 Δt2=t k -t k-1 When the sensing period is fixed, Δt1 and Δt2 are equal. The rate of change of state V k This represents the rate at which the regional state value changes within the latest time interval, and is calculated using the following formula: V k =(M k -M k-1 ) / Δt2; Among them, V k For in t k Rate of change of state at time t; The changing acceleration A k It represents the trend of the rate of change of state from the previous time interval to the current time interval, and the calculation formula is as follows: A k =(V k -V k-1 ) / ((Δt1+Δt2) / 2); Among them, V k-1 For in t k-1 Rate of change of state at time t; The calculated rate of change of state V at the current moment k With changing acceleration A k The output is associated with the task area identifier.

[0024] The core function of this step is to achieve quantitative perception and forward-looking judgment of the dynamic trend of the state evolution of the task area; by analyzing the state time series composed of continuous monitoring data, it not only calculates the "state change rate" that represents the speed of the current change, but also further calculates the "change acceleration" that represents whether the trend of change is intensifying.

[0025] This allows the system to go beyond the static snapshot mode of traditional methods that judge "normal / abnormal" based on a single threshold, and instead capture the rate of state deterioration (or improvement) and the development trend of that rate itself; For example, even if a region's current state value is not exceeded, but its rate of change and acceleration are both positive, it indicates that the region is deteriorating rapidly and the risk is accumulating quickly. S3 provides the system with key dynamic indicators for evaluating the "evolution trajectory" of the state, rather than just the "instantaneous state." S4 further includes the following: Obtain the historical task execution frequency F for each task region; the historical task execution frequency F is the cumulative value of the number of times the region status value has been successfully collected in all executed tasks for that task region; For each task region, when the state change rate V k When >0, its positive change component P is defined. v =V k When V k When ≤0, its positive variation component P is defined. v =0; When the change in acceleration A k When >0, its positive acceleration component P is defined. A =A k When the change in acceleration A k When ≤0, its positive acceleration component P is defined. A =0; According to the positive change component P v The positive acceleration component P A And based on the historical task execution frequency F, the task priority value R for this task region is calculated using the following formula: R=(P v (1+P A )) / (1+F); The calculated task priority value R is associated with the task area identifier and output.

[0026] The core function of this step is to establish a comprehensive and objective decision-making model that integrates dynamic risk trends with historical monitoring resource allocation to generate quantifiable task priorities. The calculation rules defined in this step implement a triple decision-making logic: Focus on risk increments: Only when the rate of state change is positive (i.e., the state is deteriorating) is it included in the calculation as a positive change component, ensuring that the priority increase is directly related to the unfavorable state evolution.

[0027] Amplified acceleration signal: Introducing a positive component of changing acceleration, when the degradation process accelerates, it amplifies the priority value, enabling the system to pay more attention to areas where the risk is accelerating.

[0028] Optimize resource allocation: Use the historical task execution frequency F as the denominator factor so that areas that have been monitored less frequently in the past (small F value) will have a significantly higher calculated priority value R once they begin to deteriorate; This mechanism automatically balances the breadth and depth of monitoring coverage, effectively preventing risks from being overlooked in certain areas due to long-term lack of patrols, and guiding resources to dynamically tilt towards blind spots or weak points that are "under-monitored and deteriorating".

[0029] S3 and S4 are sequentially linked. The dynamic trend indicators generated by S3 are the direct input for S4's risk assessment. Together, they transform the continuous, raw monitoring data stream into a unified and comparable task priority value R. This value is no longer a simple "yes / no" judgment, but a comprehensive evaluation score that integrates the current degree of degradation, the trend of accelerated degradation, and historical monitoring attention.

[0030] This priority value serves as the sole basis for S5 scheduling decisions, ensuring that the task allocation scheme for the UAV swarm strictly adheres to the real-time risk urgency ranking of the region. This fundamentally transforms monitoring resources from periodic equal distribution or allocation based on fixed rules to dynamic adaptive flow according to state risk. This directly addresses the resource mismatch and response lag problems pointed out in the background technology, caused by the inability to perceive and respond to continuously changing processes.

[0031] S5 further includes the following: Get all task regions and their corresponding task priority values ​​R; sort all task regions in descending order of their task priority values ​​R to generate a task region priority list. Based on the task area priority list, the first N task areas are selected sequentially from the list to form the target area set for the current task cycle; The target area set is integrated, and a task priority scheduling scheme is generated according to the priority order. The scheduling scheme is then sent to the control system of the UAV cluster to drive the cluster to execute the current round of monitoring tasks.

[0032] The process of integrating the target area set and generating a task priority scheduling scheme based on priority order, and then distributing the scheduling scheme to the control system of the UAV swarm to drive the swarm to execute the current monitoring task, further includes the following: For each task region j, collect the geographical cost parameter Tj, the task urgency parameter Rj, and the regional operation time parameter Wj; The geographical cost parameter Tj represents the estimated flight time required for the UAV swarm to fly from the current assembly point to the mission area j; The task urgency parameter Rj represents the priority value of the task region; The regional operation time parameter Wj represents the estimated time required for the UAV to complete the status perception operation in the mission area; With the maximum endurance Ci of each drone i in the drone swarm as a constraint, and minimizing the maximum total time of all drone tasks in the swarm as the optimization objective, a task allocation model is constructed. The task allocation model takes as input a set of all task regions to be allocated, parameter groups Tj, Rj, Wj for each region, a set of drone numbers and their endurance constraints Ci. The task allocation model sorts all regions in descending order based on the task urgency parameter Rj. Each time, it selects the region with the highest urgency from the top of the allocation pool and assigns it to the drone with the shortest estimated total time Wj and the endurance constraint Ci. After allocation, the task sequence and total time of the drone are updated. The iterative process continues until all regions are allocated. The task allocation model outputs an optimized task allocation scheme, which clarifies the sequence of task areas that each UAV i is responsible for. Based on the mission sequence of each UAV i, with the goal of minimizing the total flight distance of the UAV to visit all its assigned areas, the system uses a preset optimization algorithm to perform route planning and generate the optimal cruise path for the UAV. Integrate the flight paths of all UAVs, generate a cluster collaborative task instruction set, and drive the UAV cluster to perform monitoring tasks along the planned path; During mission execution, the system receives newly collected sensing data transmitted back by the UAV in real time. When a new batch of data is acquired, it is used as input to re-execute S1 to S4, calculate the updated mission priority value Rj for all mission areas, and based on the updated Rj, the real-time position of the UAV and the remaining flight time, output a new mission allocation scheme for the mission areas that have not yet been executed through the mission allocation model, thereby realizing the online dynamic reallocation of mission schemes.

[0033] A drone swarm monitoring system includes the following modules: The building environment monitoring system includes a regional status perception module, a status time sequence construction module, a dynamic trend calculation module, a task priority calculation module, and a cluster task scheduling module.

[0034] The area state perception module is used to continuously perceive the state of several preset task areas, and to acquire and associate the area state value and its identifier of each area at different times. The state-time sequence construction module is used to extract historical state values ​​for each task area from the stored data and construct a state-time sequence ordered by time. The dynamic trend calculation module is used to calculate the rate of change and acceleration of change of each task area based on the state time sequence of each task area. The task priority calculation module is used to calculate the task priority value of each task region by comprehensively considering the state change rate, change acceleration and historical task execution frequency of each task region. The cluster task scheduling module receives the task priority value R of each task region output by the task priority calculation module, and obtains the preset parameters of each task region. With the maximum endurance time of each drone in the drone swarm as a constraint and minimizing the maximum total time of all drone tasks in the swarm as the optimization objective, a load balancing task allocation model is constructed and solved to generate the task area sequence for each drone. Based on the mission area sequence of each UAV, a path planning algorithm is used to generate the optimal cruise path for it; and a cluster collaborative task instruction set is generated to drive the UAV cluster to perform monitoring tasks. During mission execution, the mission priority is recalculated in real time based on newly collected perception data, and unexecuted missions are dynamically reallocated based on the updated priority and the real-time status of the UAV.

[0035] Consider an application scenario for health monitoring of a large bridge structure. This bridge has a complex structure, comprising multiple key components such as piers, box girders, suspension cables, and bridge decks. Traditional manual inspections are inefficient, risky, and difficult to implement for high-frequency, comprehensive monitoring. To implement this invention, based on the bridge's design drawings, historical structural risk records, and maintenance priorities, its surface is first divided into 50 logically independent monitoring areas. Each area corresponds to a specific structural unit. For example, the north side of pier No. 3 is designated as area A-1, the lower flange of the east-side box girder at mid-span is designated as area B-3, and a specific section of the west-side bridge deck is designated as area C-5. Each area is assigned a unique geographical identifier and spatial coordinates.

[0036] For typical defects that may occur in different structural parts, corresponding continuously observed physical quantities are determined as state values ​​M. For the concrete pier area A-1, the focus is mainly on crack development; therefore, M is defined as the maximum crack width in this area extracted through high-resolution image analysis, in millimeters. For the steel box girder area B-3, the focus is mainly on corrosion; M is defined as the average surface temperature difference between the suspected corrosion area and the normal area in the image acquired by infrared thermal imager, in degrees Celsius. For the bridge deck area C-5, the focus is mainly on smoothness degradation; M is defined as the root mean square value of the deviation of the elevation of all measuring points in this area from the initial benchmark, obtained through lidar scanning, in millimeters. These physical quantities M all possess the characteristics of being continuous, measurable, and capable of characterizing the continuous evolution of the structural state.

[0037] The system schedules data on a fixed 24-hour perception cycle. At the first scheduling time T1 (e.g., 2023-10-25 10:00:00), the system commands the drone swarm to deploy. Drones equipped with visible light cameras fly to area A-1 to acquire images, drones equipped with infrared thermal imagers fly to area B-3, and drones equipped with lidar fly to area C-5. After the data is transmitted back, the system performs real-time analysis: edge detection is performed on the image of A-1, identifying a maximum crack width of 0.5 mm; the thermal image of B-3 is used to calculate the temperature difference between zones, obtaining an average temperature difference of 2.0℃; and the point cloud data of C-5 is processed to calculate the root mean square value of the elevation deviation, which is 0.1 mm. The system records these three values ​​as the state values ​​M of areas A-1, B-3, and C-5 at time T1, and stores them in association with the area identifier and timestamp.

[0038] 24 hours later, at time T2 (2023-10-26 10:00:00), the drone swarm performed its mission again. After this data collection and analysis, the M value for area A-1 was 0.6 mm, for area B-3 it was 2.2℃, and for area C-5 it remained at 0.1 mm. The data was stored again. At time T3 (2023-10-27 10:00:00), the third monitoring results were: the M value for area A-1 increased to 0.9 mm, area B-3 remained at 2.2℃, and area C-5 remained at 0.1 mm. At this point, the system had accumulated historical records of state values ​​for each area, arranged in chronological order.

[0039] Based on this historical data, the system constructs a time series sequence for each region. For region A-1, the sequence is [0.5(T1), 0.6(T2), 0.9(T3)], clearly showing the growth process of crack width. The sequence for region B-3 is [2.0, 2.2, 2.2], showing a trend of initial increase followed by stabilization in the corrosion temperature difference. The sequence for region C-5 is [0.1, 0.1, 0.1], indicating a stable state.

[0040] Next, the system performs dynamic trend analysis. Taking region A-1 as an example, the latest three consecutive data points (M~k-2~=0.5, M~k-1~=0.6, M~k~=0.9) are selected for calculation. The sensing period is fixed at 24 hours (Δt=86400 seconds). First, the rate of change of state V~k~ at the current time T3 is calculated: V~k~=(0.9-0.6) / 86400≈3.47×10 -6 mm / s. Next, calculate the rate of change of state Vk-1 at the previous time T2: Vk-1 = (0.6 - 0.5) / 86400 ≈ 1.16 × 102 -6 mm / s. Then calculate the changing acceleration A~k~: A~k~=(3.47×10 -6 -1.16×10-6 ) / 86400≈2.67×10 -11 millimeters per second 2 Calculations show that region A-1 is not only deteriorating (V~k~>0), but the rate of deterioration is also accelerating (A~k~>0). Similarly, for region B-3: V~k~=(2.2-2.2) / 86400=0, V~k-1~=(2.2-2.0) / 86400≈2.31×10 -6 ℃ / second, A~k~=(0-2.31×10 -6 ) / 86400≈-2.67×10 -11 ℃ / second 2 This indicates that the state has stabilized and the rate of change has slightly decreased. Both V and A in region C-5 are 0.

[0041] After obtaining dynamic trend indicators, the system calculates task priority by combining historical monitoring frequency. Assume historical records show that area A-1 has been successfully monitored 5 times (F=5), B-3 3 times (F=3), and C-5 10 times (F=10). The priority calculation formula is R=(P~v~×(1+P~A~)) / (1+F), where P~v~ and P~A~ only take positive values.

[0042] For A-1: ​​P~v~=V~k~≈3.47×10 -6 PA = Ak ≈ 2.67 × 102 -11 Then R_A1≈(3.47×10 -6 ×(1+2.67×10 -11 )) / (1+5)≈5.78×10 -7 .

[0043] For B-3: Since V~k~=0, P~v~=0, and regardless of the value of A~k~, R_B3=0.

[0044] For C-5: P~v~=0, therefore R_C5=0.

[0045] The calculation results show that region A-1, which is deteriorating at an accelerated rate, received the highest priority value.

[0046] After performing the above calculations on all 50 areas, the system sorts all areas in descending order of their task priority values ​​R, generating a global priority list. In this list, A-1 is ranked at the top due to its higher R value, while most stable areas have R values ​​close to 0 and are ranked at the bottom. Assume that in the current scheduling period, the cluster has 3 UAVs of the same model (UAV1, UAV2, UAV3) available, each with a maximum endurance C of 1.5 hours (5400 seconds), and that cluster resources allow for monitoring of the 15 highest priority areas within this period.

[0047] Next, the system enters the load balancing task allocation phase, which integrates constraints. First, pre-defined quantization parameters are integrated for these 15 target regions: Geographic cost parameter Tj: The flight time is calculated based on the coordinates of the UAV's current assembly point (or charging station) and the center coordinates of each area, combined with the preset average cruising speed (e.g., 5 m / s). For example, area A-1: ​​T_A1 = 600 seconds; area D-7: T_D7 = 900 seconds; area F-12: T_F12 = 750 seconds.

[0048] The time parameter Wj for regional operations is preset according to the sensor operation mode. For example, crack image acquisition (type A-1) requires W=120 seconds, thermal imaging scanning (type B-3) requires W=180 seconds, and laser scanning (type C-5) requires W=200 seconds.

[0049] Task urgency parameter Rj: the priority value calculated above.

[0050] The system aims to minimize the maximum total time (flight time + operation time) of all drone missions in the cluster, and executes a load balancing algorithm. The model input includes: 15 target areas and their parameter sets (T,R,W), 3 drones and their endurance constraint C=5400 seconds.

[0051] The allocation process is completed iteratively: Initialization: Each drone mission sequence is empty, and the current cumulative time TotalTime_i = 0.

[0052] Iterative Allocation: Sort the 15 regions in descending order of their Rj values. Take region A-1, which has the highest R value, and calculate the new total time (NewTotalTime_i) for each UAV after allocating it to each UAV: ​​NewTotalTime_i = TotalTime_i + T_A1_to_i + W_A1. Here, T_A1_to_i represents the flight time (T_A1) from the assembly point to A-1 for UAVs with empty task sequences; for non-empty UAVs, it represents the flight time from the last region in their sequence to A-1. The system selects the UAV with the smallest NewTotalTime_i value that satisfies NewTotalTime_i <= C. Assuming that UAV1 has the smallest estimated new total time after calculation (and satisfies the constraint), then A-1 is allocated to UAV1. Update UAV1's task sequence to [A-1], and update its TotalTime_1 = T_A1 + W_A1 = 720 seconds.

[0053] Loop: Take the next highest priority region (e.g., D-7) and repeat the calculation in step 2. For UAV1, calculate the flight time from A-1 to D-7, T_A1->D7 (which may differ from T_D7), and then calculate NewTotalTime_1 = 720 + T_A1->D7 + W_D7. Simultaneously calculate NewTotalTime for UAV2 and UAV3 (the sequence is still empty). Again, select the UAV with the smallest flight time that meets the endurance constraint for allocation.

[0054] Termination: Continue this process until all 15 areas have been allocated, or no drone can meet the endurance constraints for subsequent areas. The final output is an allocation scheme, for example: UAV1 mission sequence: [A-1, F-12, ...], total time 4980 seconds.

[0055] UAV2 mission sequence: [D-7,G-2,...], total time 4650 seconds.

[0056] UAV3 mission sequence: [H-5,...], total time 4520 seconds.

[0057] This approach ensures a high degree of workload balance among the three drones (with a maximum flight time of 4980 seconds, close to but not exceeding 5400 seconds), and prioritizes high-priority tasks.

[0058] Once the allocation scheme is determined, the system performs refined single-aircraft route planning for each UAV. Taking the mission sequence [A-1, F-12,...] of UAV1 as an example, the system models this problem as a Traveling Salesman Problem (TSP), with the goal of finding the shortest path to visit all regions in the sequence once and return to the rendezvous point. The system calls the built-in TSP optimization solver (such as one based on a genetic algorithm), inputs the geographical coordinates of these regions, and calculates an optimal cruising path, for example: rendezvous point → A-1 → F-12 → ... → last assigned region. This path ensures that UAV1 minimizes its flight distance when performing all assigned tasks, thereby maximizing its endurance.

[0059] By integrating the optimized flight paths of all UAVs, the system generates a detailed set of swarm collaboration task instructions, such as: "UAV1: Fly along path P1, sequentially performing operations in areas A-1, F-12..., with operation instructions as...; UAV2: ...". This instruction set is then sent to the UAV swarm control system, and the three UAVs take off autonomously to perform monitoring tasks as planned.

[0060] The system maintains real-time monitoring during mission execution. When a UAV (e.g., UAV2) completes monitoring of a certain area and transmits new perception data, this data is immediately fed into the data processing pipeline, triggering a dynamic feedback and reallocation mechanism. For example, after UAV2 completes monitoring of its first area D-7, the transmitted data analysis shows that another area E-9 (not among the initial 15 targets), which originally had a priority of 0, shows new signs of surface peeling, and the state value M begins to rise. The system immediately uses the new data as input to re-execute the state perception, time series construction, trend calculation, and priority calculation steps. The updated mission priority value R_E9 for area E-9 changes from a value close to 0 to a significantly positive value.

[0061] At this point, the system initiates dynamic reallocation: Status Update: Obtain the latest status of all drones, including real-time location, remaining flight time, and their respective unexecuted task sequences.

[0062] Problem Restructuring: The newly emerging risky area E-9, together with all other mission areas that have not yet been started by any drones, will form a new set to be assigned.

[0063] Re-solution: Based on the latest priority list, the real-time location of the UAV (recalculating the flight time Tj to each region), and the remaining endurance constraints, re-run the above load balancing task allocation model and TSP path planning algorithm.

[0064] Command Update: Generate new mission allocation schemes and routes adapted to the latest situation, and immediately issue adjustment commands. For example, the system may decide to insert the E-9 area into the remaining mission sequence of the nearest UAV3 with sufficient remaining range, re-plan the optimal route for it, and cancel or adjust some of the original low-priority missions of the UAV3.

[0065] Through the aforementioned coherent closed-loop process of "perception-analysis-decision-execution-redistribution," this embodiment fully demonstrates the working principle of the UAV swarm monitoring system and method. The system completely eliminates reliance on fixed inspection cycles and static thresholds. It proactively captures evolution trends through differential analysis of continuous state time sequences, objectively quantifies risk urgency through a priority algorithm incorporating historical execution frequencies, achieves optimal allocation of swarm resources through an integrated multi-constraint load balancing model, and ensures that the monitoring focus remains synchronized with the latest risk situation through a real-time data-driven dynamic redistribution mechanism. Ultimately, it achieves a fundamental shift from "post-event response" to "pre-event warning," from "equal coverage" to "precise focusing," and from "static planning" to "dynamic adaptation," significantly improving the initiative, efficiency, and intelligence level of health monitoring for large-scale building infrastructure.

[0066] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for monitoring unmanned aerial vehicle (UAV) swarms, characterized in that: Includes the following steps: S1. Perform continuous state perception on several preset task areas and obtain the task area state value of each area at the current time. S2. Based on the state values ​​obtained by each task region at different times, construct a state time sequence for each region; S3. For each task region, calculate its state change rate and change acceleration based on its state time sequence. S4. For each task region, calculate the task priority value of the task region by combining its state change rate, change acceleration and historical task execution frequency. S5. Based on the task priority values ​​of all task areas, determine the execution order of each area in the next cycle, generate a task allocation scheme, and drive the drone swarm to execute. The S3 further Includes the following: For each region, the data points corresponding to the three most recent consecutive sensing moments are selected from the state time sequence of that region as the basis for calculation; these three moments are defined as t in chronological order. k-2 t k-1 t k ; where t k Representing the latest perception moment, the corresponding region state values ​​for these three moments are M, respectively. k-2 M k-1 M k ; Calculate the time interval between adjacent sensing moments, denoted as Δt, where Δt1 = t k-1 -t k-2 Δt2=t k -t k-1 When the sensing period is fixed, Δt1 and Δt2 are equal. The rate of change of state V k This represents the rate at which the regional state value changes within the latest time interval, and is calculated using the following formula: V k =(M k -M k-1 ) / Δt2; Among them, V k For in t k Rate of change of state at time t; The changing acceleration A k It represents the trend of the rate of change of state from the previous time interval to the current time interval, and the calculation formula is as follows: A k =(V k -V k-1 ) / ((Δt1+Δt2) / 2); Among them, V k-1 For in t k-1 Rate of change of state at time t; The calculated rate of change of state V at the current moment k With changing acceleration A k The output is associated with the task area identifier; S4 further includes the following: Obtain the historical task execution frequency F for each task region; the historical task execution frequency F is the cumulative value of the number of times the region status value has been successfully collected in all executed tasks for that task region; For each task region, when the state change rate V k When >0, its positive change component P is defined. v =V k When V k When ≤0, its positive variation component P is defined. v =0; When the change in acceleration A k When >0, its positive acceleration component P is defined. A =A k When the change in acceleration A k When ≤0, its positive acceleration component P is defined. A =0; According to the positive change component P v The positive acceleration component P A And based on the historical task execution frequency F, the task priority value R for this task region is calculated using the following formula: R=(P v (1+P A )) / (1+F); The calculated task priority value R is associated with the task area identifier and output.

2. The method for monitoring unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that: S1 further includes the following: The building surface is divided into several monitoring areas; For each task area state type that needs to be monitored, a physical quantity that can continuously take values ​​and continuously change with the continuous change of the task area state type is determined as the observed value M for that area; The task area state type refers to the manifestation of continuous changes in physical, chemical, and geometric properties in the task area caused by the appearance of targets, the occurrence of anomalies, environmental changes, and the dynamic evolution of task requirements. At each scheduling moment, a drone equipped with the corresponding sensor is dispatched to the monitoring area to collect raw sensing data. By performing spatial feature analysis on the raw sensing data, the dominant feature quantity is extracted, and the statistical feature value of the dominant feature quantity is obtained through mathematical statistical processing methods. The statistical feature value is used as the value of the observation M of the area at the current scheduling moment. The dominant feature quantity refers to the set of physical parameters that are identified and separated from the original sensing data through spatial feature analysis, directly representing the degree of continuous evolution of the state type in the task area; Record the observation value M, region identifier, and timestamp of each region at the current moment, and store them together.

3. The method for monitoring unmanned aerial vehicle (UAV) swarms according to claim 2, characterized in that: S2 further includes the following: From the associated data records containing area identifiers, timestamps, and area status values, extract all data entries with the same area identifier. Each data entry contains a timestamp of a sensing moment and its corresponding area status value. All the extracted data entries are sorted in ascending order according to their timestamps; and the regional state values ​​in the sorted data entries are arranged and connected according to their corresponding timestamps to form a state time sequence of the task region. This sequence, ordered by time, records the evolution of the region's state value over time.

4. The method for monitoring unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that: S5 further includes the following: Get all task regions and their corresponding task priority values ​​R; sort all task regions in descending order of their task priority values ​​R to generate a task region priority list. Based on the task area priority list, the first N task areas are selected sequentially from the list to form the target area set for the current task cycle; Integrate the target region set and generate a task priority scheduling scheme according to priority order; The scheduling scheme is sent to the control system of the UAV cluster, driving the cluster to execute the monitoring task for this round.

5. The method for monitoring unmanned aerial vehicle (UAV) swarms according to claim 4, characterized in that: The target region set is integrated, and a task priority scheduling scheme is generated according to the priority order. The scheduling scheme is sent to the control system of the UAV swarm to drive the swarm to execute the current monitoring task, and further includes the following: For each task region j, collect the geographical cost parameter Tj, the task urgency parameter Rj, and the regional operation time parameter Wj; The geographical cost parameter Tj represents the estimated flight time required for the UAV swarm to fly from the current assembly point to the mission area j; The task urgency parameter Rj represents the priority value of the task region; The regional operation time parameter Wj represents the estimated time required for the UAV to complete the status perception operation in the mission area; With the maximum endurance Ci of each drone i in the drone swarm as a constraint, and minimizing the maximum total time of all drone tasks in the swarm as the optimization objective, a task allocation model is constructed. The task allocation model takes as input a set of all task regions to be allocated, parameter groups Tj, Rj, Wj for each region, a set of drone numbers and their endurance constraints Ci. The task allocation model sorts all regions in descending order based on the task urgency parameter Rj. Each time, it selects the region with the highest urgency from the top of the allocation pool and assigns it to the drone with the shortest estimated total time Wj and the endurance constraint Ci. After allocation, it updates the task sequence and total time of the drone. The above selection and allocation steps are repeated until all regions are allocated. The task allocation model outputs an optimized task allocation scheme, which clarifies the sequence of task areas that each UAV i is responsible for. Based on the mission sequence of each UAV i, with the goal of minimizing the total flight distance of the UAV to visit all its assigned areas, the system uses a preset optimization algorithm to perform route planning and generate the optimal cruise path for the UAV. Integrate the flight paths of all UAVs, generate a cluster collaborative task instruction set, and drive the UAV cluster to perform monitoring tasks along the planned path; During mission execution, the system receives newly collected sensing data transmitted back by the UAV in real time. When a new batch of data is acquired, it is used as input to re-execute S1 to S4, calculate the updated mission priority value Rj for all mission areas, and based on the updated Rj, the real-time position of the UAV and the remaining flight time, output a new mission allocation scheme for the mission areas that have not yet been executed through the mission allocation model, thereby realizing the online dynamic reallocation of mission schemes.

6. A drone swarm monitoring system, applied to the drone swarm monitoring method according to any one of claims 1-5, characterized in that: Includes the following modules: The monitoring system includes a regional status perception module, a status time sequence construction module, a dynamic trend calculation module, a task priority calculation module, and a cluster task scheduling module.

7. The UAV swarm monitoring system according to claim 6, characterized in that: The area state perception module is used to continuously perceive the state of several preset task areas, and to acquire and associate the area state value and its identifier of each area at different times. The state-time sequence construction module is used to extract historical state values ​​for each task area from the stored data and construct a state-time sequence ordered by time. The dynamic trend calculation module is used to calculate the rate of change and acceleration of change of each task area based on the state time sequence of each task area. The task priority calculation module is used to calculate the task priority value of each task region by comprehensively considering the state change rate, change acceleration and historical task execution frequency of each task region. The cluster task scheduling module receives the task priority value R of each task region output by the task priority calculation module, and obtains the preset parameters of each task region. With the maximum endurance time of each drone in the drone swarm as a constraint and minimizing the maximum total time of all drone tasks in the swarm as the optimization objective, a load balancing task allocation model is constructed and solved to generate the task area sequence for each drone. Based on the mission area sequence of each UAV, a path planning algorithm is used to generate the optimal cruise path for it; and a cluster collaborative task instruction set is generated to drive the UAV cluster to perform monitoring tasks. During mission execution, the mission priority is recalculated in real time based on newly collected perception data, and unexecuted missions are dynamically reallocated based on the updated priority and the real-time status of the UAV.

Images