Unmanned autonomous cooperative control method for road construction-oriented machine group

Abstract

The present application relates to the technical field of unmanned aerial vehicle control, in particular to a method for autonomous cooperative control of unmanned aerial vehicle group for road construction, comprising: recording unmanned aerial vehicles detecting construction abnormality of construction machines as frozen unmanned aerial vehicles, and recording unmanned aerial vehicles other than the frozen unmanned aerial vehicles as free unmanned aerial vehicles; obtaining a thawing coefficient of the frozen unmanned aerial vehicles according to the increase of the target value of the frozen unmanned aerial vehicles; obtaining a freezing coefficient of the free unmanned aerial vehicles according to the decrease of the target value of the free unmanned aerial vehicles, and taking the difference between the freezing coefficient and the thawing coefficient of the closest free unmanned aerial vehicle and the frozen unmanned aerial vehicle as a global compensation coefficient; correcting the target value of the unmanned aerial vehicles by using the thawing coefficient and the freezing coefficient, correcting the global target value by using the global compensation coefficient, and updating the position of each unmanned aerial vehicle by using a particle swarm algorithm, so that the corrected target value and the corrected global target value are maximized. The present application guarantees the efficiency and safety of highway pavement construction process.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, specifically to a method for autonomous and collaborative control of UAV swarms for road construction. Background Technology

[0002] Road construction fleets include various construction machines (such as paver fleets, roller fleets, transport fleets, etc.) as well as drone fleets used for inspection and monitoring of construction machines; when constructing highway pavements, coordinating the work of the road construction fleets can significantly improve construction efficiency, quality and safety.

[0003] Inspection and control of drone swarms is a crucial aspect of ensuring coordinated operation of drone swarms during road construction. Particle swarm optimization (PSO) is a commonly used inspection and control method. This method treats each drone as a particle and continuously updates the particle's position to ultimately update the drone's position, enabling each drone to reach its optimal value (e.g., optimizing the target value for assessing the distance between each drone's location and uninspected areas) and the drone swarm as a whole to reach its optimal value (e.g., optimizing the global target value for assessing the inspection range, power consumption, and energy savings of all drones).

[0004] However, when construction machines work collaboratively under the inspection of drone swarms, construction anomalies may occur. In order to eliminate these anomalies in a timely manner, some drones in the swarm may need to continuously track a particular anomaly, or they may no longer need to continue tracking after the anomaly is eliminated. This causes some drones in the swarm to have specific behaviors (tracking and non-tracking behaviors) and switch between different behaviors, making it impossible to directly use the aforementioned particle swarm algorithm to update the drone positions. In other words, when updating drone positions using the aforementioned particle swarm algorithm, it is unable to cope with sudden construction anomalies or mitigate or eliminate their impact, thus failing to guarantee the efficiency and safety of highway pavement construction. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a method for unmanned autonomous collaborative control of machine fleets for road construction.

[0006] The unmanned autonomous collaborative control method for road construction fleets of the present invention adopts the following technical solution: One embodiment of the present invention provides a method for unmanned autonomous collaborative control of a fleet of vehicles for road construction, the method comprising the following steps: The fleet comprises several drones used for inspecting construction machinery; the target value for each drone is obtained based on its location and distance from the uninspected area as well as its distance from the nearest drone; the global target value is obtained based on the monitoring field of view coverage and power saving indicators of all drones. Drones that detect abnormalities in construction machinery are classified as frozen drones, and drones other than frozen drones are classified as free drones. The difference between the increase in the target value of the frozen drone and the increase in the target value of the nearest free drone is denoted as the unfreezing coefficient of the frozen drone; the freezing coefficient of the free drone is obtained based on the suppression effect of the decrease in the target value of the free drone on the increase in the global target value; the freezing coefficient is negatively correlated with the suppression effect; the difference between the freezing coefficient and the unfreezing coefficient of the nearest free drone and the frozen drone is used as the global compensation coefficient. The target value of the frozen drone is corrected using the thawing coefficient, the target value of the free drone is corrected using the freezing coefficient, the global target value is corrected using the global compensation coefficient, and the position of each drone is updated using the particle swarm algorithm to maximize the corrected target value of each drone and the corrected global target value.

[0007] Preferably, the specific steps for obtaining the target value for each drone based on its location, distance from the uninspected area, and distance from the nearest drone are as follows: Each drone is considered a target object, and the position of the drone is recorded as the position of the target object. When a drone is used as a particle in the particle swarm algorithm, the particle corresponding to each drone is considered a target object, and the position of the particle is recorded as the position of the target object. For any target object, the Euclidean distance between the i-th position in the uninspected area and the target object is denoted as . The inspection intensity at the i-th position is denoted as The inspection intensity is positively correlated with the time taken for the drone to pass through the i-th position. Obtain the distance of attention in areas not inspected Where N1 represents the total number of N1 locations in the uninspected area; the target value of each target object is... It is negatively correlated with x2, which is positively correlated with x2, where x2 represents the distance between each target object and the nearest target object.

[0008] Preferably, the specific steps for obtaining the global target value based on the monitoring field of view coverage and power saving indicators of all UAVs are as follows: Each drone is considered a target object, and the position of the drone is recorded as the position of the target object. When a drone is used as a particle in the particle swarm algorithm, the particle corresponding to each drone is considered a target object, and the position of the particle is recorded as the position of the target object. For any target object and the inspection field of view area where the target object is located, obtain the sum of the inspection intensity of all locations within the inspection field of view area, denoted as y; the inspection necessity of the inspection field of view area is negatively correlated with y; obtain the mean of the inspection necessity of the inspection field of view area where all target objects are located, denoted as the monitoring field of view coverage rate. Before the particle swarm optimization algorithm is executed, the position of the drone is recorded as the initial position of the drone; when each drone is treated as a particle in the particle swarm optimization algorithm, the position of the particle is recorded as the target position of the drone; the Euclidean distance between the initial position and the target position is recorded as the travel distance; the power saving index is negatively correlated with the mean of the travel distance of all drones. The average of the monitoring field of view coverage and power saving indicators is used as the global target value for the drone swarm.

[0009] Preferably, the difference between the increase in the target value of the frozen drone and the increase in the target value of the nearest free drone is recorded as the thawing coefficient of the frozen drone, and the specific steps include the following: The current position of all drones is recorded as the current position; when the particle swarm algorithm is executed, all drones are treated as particles, and the particle positions corresponding to all drones are updated; After all the particles corresponding to the drones have their positions updated, the target value of each drone corresponding to the particle at the updated particle position is recorded as M1; the target value of each drone corresponding to the particle at the current position is recorded as M2, and (M1-M2) / M2 is recorded as the increase in the target value of each drone corresponding to the particle. The increase in the target value of any frozen drone is denoted as Q1. For the current position of the frozen drone, among all the particles corresponding to the free drones, the particle whose position is closest to the current position is obtained, and the free drone corresponding to this particle is taken as the nearest neighbor free drone. The increase in the target value of the nearest neighbor free drone is denoted as Q2. Let (Q1-Q2) / Q2 be denoted as the thawing coefficient of the frozen drone.

[0010] Preferably, the specific steps for obtaining the suppression amplitude are as follows: After all the particles corresponding to the drones have their positions updated once, (M2-M1) / M1 is recorded as the decrease in the target value of each particle's corresponding drone; the global target value is recorded as F1, the maximum value of the global target value before all the particles corresponding to the drones have their positions updated once is recorded as F2, and (F1-F2) / F1 is recorded as the increase in the global target value, denoted as H1; Obtain the average reduction of the target value of all drones, M4. The reduction of any free drone is denoted as M5. The ratio of the absolute value of (M5-M4) to M4 is denoted as the global reduction of the free drone, and is represented as H2. The ratio of the absolute values ​​of (H2-H1) to H1 is denoted as the extent to which the decrease in the target value of the free UAV suppresses the increase in the global target value.

[0011] Preferably, the step of using the average of the freezing coefficient and thawing coefficient of the nearest free drone and frozen drone as the global compensation coefficient includes the following specific steps: After all the particles corresponding to drones are updated once, for any particle corresponding to a frozen drone, it is denoted as particle 1. The particle corresponding to the nearest free drone is obtained and denoted as particle 2. The drones corresponding to particle 2 and particle 1 are a pair of free drone and frozen drone. For all frozen drones, obtain all pairs of free drones and frozen drones; For any pair of free drones and frozen drones, the average of the freezing coefficient and thawing coefficient of the free drones and frozen drones is denoted as the coordination index of any pair of free drones and frozen drones; the average of all coordination indices of free drones and frozen drones is denoted as the global compensation coefficient.

[0012] Preferably, the specific steps of correcting the target value of the frozen UAV using the thawing coefficient, correcting the target value of the free UAV using the freezing coefficient, and correcting the global target value using the global compensation coefficient are as follows: The corrected target value of the frozen drone is positively correlated with the thawing coefficient; the corrected target value of the free drone is positively correlated with the freezing coefficient; and the corrected global target value is positively correlated with the global compensation coefficient.

[0013] Preferably, the particle swarm optimization algorithm includes the following specific steps: D1: Treat all drones as particles, with each drone's current position as its initial position, each drone's target value as its optimal target value, and the global target value as the optimal global target value for the particle swarm; randomly assign an initial velocity to each particle. D2: The particle position of each particle is updated based on its position and velocity, and the particle positions of all particles are updated accordingly. For the target value of all particles and the global target value, the target value of the frozen drone is corrected using the thawing coefficient, and the target value of the free drone is corrected using the freezing coefficient. After correcting the global target value using the global compensation coefficient, the maximum value between the corrected target value and the optimal target value of each particle is used as the optimal target value of each particle. The maximum value between the corrected global target value and the optimal global target value is used as the optimal global target value. The particle position corresponding to the optimal target value of each particle is recorded as the optimal position of each particle. The velocity of each particle is re-acquired based on the optimal target value and the optimal global target value of all particles. D3: Repeat D2 until the optimal position of each particle no longer changes, or when the number of times D2 is repeated is greater than or equal to the preset number of times, then stop repeating D2 and the particle swarm algorithm ends.

[0014] Preferably, after updating the position of each UAV using the particle swarm optimization algorithm, the specific steps include the following: Each drone moves to the updated position; however, for the frozen drone's updated position, if the Euclidean distance between the updated position and the original position is less than or equal to the preset response distance, the frozen drone will no longer move to the updated position, but will instead track the construction machine with the construction anomaly; when the free drone detects abnormal construction during its movement, it will stop moving to the updated position and instead track the construction machine with the construction anomaly.

[0015] Preferably, the specific steps for obtaining the inspection intensity are as follows: The location of each drone is recorded in real time, and the time when it leaves each location is abbreviated as the inspection time of each location; all inspection times corresponding to all locations are linearly normalized respectively. For all normalized inspection times at the same location, the average of all normalized inspection times at that location is used as the inspection index for all locations within the inspection field of view of that location. The inspection index with the highest value among all inspection indicators for each location is the inspection intensity for each location.

[0016] The beneficial effects of the technical solution of the present invention are: The drones in this invention operate in two modes: frozen drones that detect abnormal construction activity and free drones that do not. During the particle swarm optimization (PSO) process, the target value of the frozen drone is corrected using a thawing coefficient. This thawing coefficient describes whether construction anomaly detection can be performed solely by the frozen drone, or whether free drones can replace the frozen drone for anomaly detection. This allows the frozen drone to decide whether to track abnormal construction activity based on the distribution of free drones after obtaining the updated position using the corrected target value. Simultaneously, when updating the position of the free drones, the PSO algorithm corrects their target value using a freezing coefficient. This freezing coefficient describes whether the free drone can stop its inspection midway. This allows the drones to handle potential abnormal construction activity without affecting the overall drone's large global target value (i.e., high inspection coverage and power saving). Furthermore, the global target value is corrected using a global compensation coefficient, which describes the cooperation and coordination between the free drones and the frozen drones. This ensures that the updated position of each drone has the maximum corrected target value and the maximum corrected global target value. While ensuring that the drone swarm can promptly inspect uninspected areas and has a long endurance, the free drones and the frozen drones can cooperate and relay each other, enabling timely detection, tracking, and contact with abnormal construction situations. This ensures the normal collaborative work of the construction machine swarm and achieves high efficiency and safety in the highway pavement construction process. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the steps of a method for unmanned autonomous collaborative control of a fleet of vehicles for road construction, provided in one embodiment of the present invention. Detailed Implementation

[0019] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the unmanned autonomous collaborative control method for road construction based on the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] The following description, in conjunction with the accompanying drawings, details the specific scheme of the unmanned autonomous collaborative control method for road construction provided by this invention.

[0022] Example 1: The aircraft fleet provided in this embodiment for road construction includes: construction machines for road construction, and a fleet of drones for monitoring the construction machines.

[0023] Construction machinery includes a group of pavers, a group of rollers, and a fleet of transport vehicles, used for asphalt paving, compaction, and transportation, respectively. This embodiment includes two pavers, one facing the other, paving asphalt on two sections of the same highway (8 lanes in both directions). Other embodiments may include more pavers, for example, two pavers side-by-side on the same section (with a distance of at least 100mm between them) paving in the same direction, and two pavers paving in opposite directions. In this embodiment, four rollers are used to compact the paved asphalt on each section. Additionally, ten transport vehicles transport asphalt to all the pavers.

[0024] The drone swarm in this embodiment comprises eight drones, each equipped with an industrial camera and a LiDAR. The camera's field of view is vertically downward, used to acquire RGB images of the construction process, with each image measuring 1024×1024 pixels. To ensure the accuracy of the camera and LiDAR's acquisition of the construction process and road surface (which contributes to the accuracy of subsequent identification and detection), all drones in this embodiment fly at a low altitude. The flight altitude of all drones in this embodiment is 50 meters. In other embodiments, the flight altitude of the drones can be set to other values, preferably 50 to 100 meters.

[0025] In this embodiment, 5G communication technology is used to assign task data to each construction machine. For example, the driving route, driving speed, and paving thickness are sent to the paver via the 5G network; the driving route and vibration frequency are sent to the road roller; and the driving route, driving speed, and load are sent to the transport vehicle. Each construction machine works according to the received task data.

[0026] The drone swarm inspects the entire construction site. The purpose of the inspection is twofold: firstly, to transmit images of the construction site to the monitoring platform in real time for online monitoring of the entire construction process; and secondly, to promptly identify and detect abnormal construction phenomena and issue early warning signals. After receiving the early warning signals, the monitoring platform reassigns task data to the construction machines to eliminate the abnormal construction phenomena.

[0027] In this embodiment, the normal and coordinated operation of various construction machines is maintained through the inspection and monitoring of a swarm of drones.

[0028] Comparative Example: This embodiment provides a method for autonomous collaborative control of unmanned aerial vehicle (UAV) swarms for road construction. This method controls UAV swarms, enabling them to perform efficient and sustained inspections, thereby maintaining collaborative operation among construction machines through these inspections. Specifically, it includes: First, each drone is remotely controlled to fly to a designated location (e.g., all drones are evenly spaced above the construction area). Then, in this embodiment, all drones are treated as a single particle, and the position of each drone is updated using a particle swarm optimization algorithm.

[0029] Specifically, in this embodiment, the target value of each drone is obtained based on the distance between each drone and the uninspected area and the distance of the nearest neighboring drone; the global target value is obtained based on the monitoring field coverage and power saving index of all drones.

[0030] A higher target value for each drone indicates a more suitable location for the drone, or that the location and target value of each drone are optimal. A lower target value indicates a less suitable location for the drone, or that the location and target value of each drone are not optimal. A higher global target value means that all drones can promptly inspect all locations within the construction area with low power consumption (strong endurance); a lower global target value means that all drones cannot promptly inspect all locations within the construction area with high power consumption (weak endurance).

[0031] The particle swarm algorithm in this embodiment is as follows: D1: Treat each drone as a particle, and the current position of each drone (referred to as the current position) as the initial particle position. The target value of each drone is taken as the optimal target value of the particle, and the global target value is taken as the optimal global target value of the particle swarm. Randomly assign an initial velocity to each particle.

[0032] D2: Obtain the next particle position for each particle based on its position and velocity (equivalent to updating the particle position of each particle; the next particle position is the updated particle position). Note that only the particle position is updated; the drone's position remains unchanged.

[0033] After updating the particle positions of all particles, obtain the target value and global target value for each particle. The maximum value between each particle's target value and the optimal target value is then used as the optimal target value for that particle. The maximum value between the global target value and the optimal global target value is then used as the optimal global target value. The particle position corresponding to each particle's optimal target value is recorded as the optimal position for that particle. Finally, the velocity of each particle is re-obtained based on the optimal target values ​​of all particles and the optimal global target value.

[0034] D3: Repeat D2. In this embodiment, continue until the optimal position of each particle no longer changes, or when the number of times D2 is repeated is greater than or equal to 30, then stop repeating D2 and the particle swarm algorithm ends.

[0035] After the particle swarm optimization algorithm finishes execution, the optimal position of each particle is used as the updated position of each drone (i.e., the position the drone will move to in the next moment). Each drone moves at a constant speed from its current position to its updated position.

[0036] Next, in this embodiment, every n seconds is considered a time period. At each time period, based on the drone's position (i.e., its current position), the particle swarm algorithm described above is used to obtain the updated inspection position of each drone, and each drone moves from its current position to the updated position. This process achieves real-time updates of the inspection positions of all drones, thereby enabling the inspection of the construction area, ensuring that drones do not interfere with each other, providing a wide inspection field of view, and demonstrating the drone swarm's strong endurance. This facilitates the monitoring of the collaborative process of the construction machine group, enabling the construction machine group to operate normally.

[0037] This embodiment uses n=1 as an example. In some embodiments, to reduce the computational load and save computational resources when running this embodiment, n can be set to 3. In other embodiments, to enable more reliable control of the drone (e.g., to update the drone's inspection position in a timely manner), n can be set to 0.2.

[0038] This concludes the example.

[0039] However, the problem with this comparative embodiment is that during the drone inspection process, when there is a construction machine with construction anomalies, the drone needs to continuously track and monitor the construction machine. Before the construction anomaly is eliminated, the drone cannot perform inspection according to the updated position obtained by the particle swarm optimization algorithm, or in other words, it cannot update its position as a free particle in the particle swarm optimization algorithm (or the drone is in a frozen state and cannot update its position as a particle in the particle swarm optimization algorithm according to the above method); only after the construction anomaly is eliminated can the drone update its position as a free particle in the particle swarm optimization algorithm (or the drone is in a free state and can update its position as a particle in the particle swarm optimization algorithm according to the above method).

[0040] In summary, when construction machines work collaboratively under the inspection of drone swarms, construction anomalies may occur. In order to eliminate these anomalies in a timely manner, some drones in the swarm may need to continuously track a particular anomaly, or they may no longer need to continue tracking once the anomaly is eliminated. This causes some drones in the swarm to exhibit specific behaviors (tracking and non-tracking behaviors) and switch between different behaviors, making it impossible to directly use the aforementioned particle swarm optimization algorithm to update drone positions. In other words, updating drone positions using the aforementioned particle swarm optimization algorithm cannot cope with sudden construction anomalies or mitigate or eliminate their impact.

[0041] Example 2: The unmanned autonomous collaborative control method for road construction fleets provided in this embodiment can further solve the problems of the above-mentioned comparative embodiments, such as... Figure 1 As shown, it specifically includes: Step S201: Obtain the target value for each drone based on its location and distance from the uninspected area and the nearest drone; obtain the global target value based on the monitoring field of view coverage and power saving indicators of all drones.

[0042] (1) Using images collected by UAVs, image processing algorithms are used to fuse and stitch the collected images to construct a panoramic top-down view of the construction area. Each UAV at each moment corresponds to a pixel position in the panoramic top-down view, referred to as the position.

[0043] In the panoramic overhead view, the location of each drone is recorded in real time (e.g., once every 0.1 seconds), and the time when it leaves each location is obtained (simply referred to as the inspection time for each location). In particular, when there is a location that has not been inspected by any drone, the inspection time of that location is recorded as the time when the drone swarm first begins inspection, such as the time when all drones take off and reach the specified altitude (i.e., 50 meters).

[0044] Before obtaining the current time (including the current time), all inspection times corresponding to all locations in the panoramic top-down view are linearly normalized to remove the dimensions and orders of magnitude of the inspection times. Specifically, if all inspection times at all locations are the same, then the result of the normalization process for all inspection times is set to 0.01. In some embodiments, it can also be set to a random value within the interval [0, 0.01].

[0045] For all normalized inspection times at the same location, a rectangular area of ​​K1×K1 centered at that location is defined as the inspection field of view (PDVAR) when the UAV is at that location. The average of all normalized inspection times at that location is used as the inspection index for all pixels within that DPVAR. A larger inspection index indicates that the DPVAR centered at that location has been inspected more recently; a smaller inspection index indicates that the DPVAR centered at that location has not been inspected for a longer period of time. Here, K1 represents the width of the defined DPVAR.

[0046] Traversing all locations in the panoramic top-down view, for all locations within the inspection field of view centered on each location, each location corresponds to an inspection index. Since the inspection field of view areas of different locations may overlap, each location in the panoramic top-down view may correspond to multiple inspection indices. In this embodiment, the inspection index with the largest value is marked as the inspection intensity for each location.

[0047] (2) The area formed by the locations where the inspection intensity is less than or equal to the first preset threshold th1 is recorded as the uninspected area. In this embodiment, th1 is the average of the inspection intensity of all locations; in other embodiments, the value of th1 can also be calculated using the Otsu threshold segmentation algorithm.

[0048] (3) As an optional example, the target value of each drone is obtained based on the distance between each drone and the uninspected area and the distance to the nearest neighbor drone, including the following methods: In the panoramic top-down view, each drone is regarded as a target object, and the position of the drone is recorded as the position of the target object. When the drone is used as a particle in the particle swarm algorithm, the particle corresponding to each drone is regarded as the target object, and the particle position of the particle is recorded as the position of the target object.

[0049] For any target object, the nearest distance between the target object and the uninspected area is recorded as x1. The distance between the target object and the nearest other target object is recorded as x2.

[0050] The target value for each target object is negatively correlated with x1 and positively correlated with x2. The closer the target object is to the uninspected area, the larger the target value; the farther the target object is from the nearest target object, the safer it is, and the larger the target value will be. The specific formula for calculating the target value for each target object is as follows: .

[0051] in, The target value for each target object is represented by exp(), which represents an exponential function with the natural constant as the base; W represents a preset normalization coefficient, which in this embodiment is equal to the maximum value of the width and height of the panoramic top view.

[0052] Furthermore, this implementation predetermines a safety distance Lx. When x2 is less than or equal to Lx, the target value for each drone is set to 0. This embodiment uses Lx=K1 as an example. The purpose is twofold: firstly, to avoid drone collisions, and secondly, to prevent different drones from simultaneously inspecting the same construction area, ensuring that drones in the drone swarm can be dispersed for inspection.

[0053] As a preferred example, the target value for each drone is obtained based on the distance between each drone and the uninspected area and the distance to its nearest neighbor drone, including the following methods: In the panoramic top-down view, each drone is regarded as a target object, and the position of the drone is recorded as the position of the target object; when the drone is used as a particle in the particle swarm algorithm, the particle corresponding to each drone is regarded as the target object, and the particle position of the particle is recorded as the position of the target object.

[0054] For any target object, the Euclidean distance between the i-th position in the uninspected area and the target object is denoted as . The inspection intensity at the i-th position is denoted as .

[0055] Obtain the distance of attention in areas not inspected Where N1 represents the total number of locations within the uninspected area. Distance of Interest It is obtained by weighting and summing each distance using the inspection intensity, where the lower the inspection intensity, the greater the weight in the weighted summation, making the distance of interest more significant. It can describe the distance between locations within an uninspected area that have not been inspected for a long time and the location of the drone.

[0056] The target value of each target object and It shows a negative correlation with x2 and a positive correlation with x2.

[0057] The calculation formula is as follows: .

[0058] Furthermore, a safety distance Lx is preset, and when x2 is less than or equal to Lx, the target value of each drone is set to 0.

[0059] Thus, the target value for each drone was obtained based on its position or its position as a particle when it is considered a particle.

[0060] (4) As an example, the global target value is obtained based on the monitoring field of view coverage and power saving indicators of all UAVs, including the following methods: In the panoramic top-down view, each drone is regarded as a target object, and the position of the drone is recorded as the position of the target object; when the drone is used as a particle in the particle swarm algorithm, the particle corresponding to each drone is regarded as the target object, and the particle position of the particle is recorded as the position of the target object.

[0061] For any target object and the inspection field of view area where the target object is located, obtain the sum of the inspection intensity of all locations within the inspection field of view area, denoted as y. Let exp(-y) represent the inspection necessity of the inspection field of view area. A larger y indicates a higher inspection intensity for all locations within the inspection field of view area, meaning that all locations within the inspection field of view area have recently been inspected and there is no need for further inspection (i.e., lower inspection necessity). A smaller y indicates a lower inspection intensity for all locations within the inspection field of view area, meaning that all locations within the inspection field of view area have not been inspected for a long time and need to be inspected promptly (i.e., higher inspection necessity).

[0062] The average value of the inspection necessity of the inspection field of view area of ​​all target objects is obtained and recorded as the monitoring field of view coverage rate.

[0063] In the panoramic top-down view, the initial position of each drone before it is used as a particle in the particle swarm optimization (PSO) algorithm (i.e., before the PSO algorithm begins execution) is recorded. When each drone is used as a particle in the PSO algorithm, its position is recorded as its target position. The Euclidean distance between the initial and target positions is recorded as the distance each drone travels as a particle in the PSO algorithm. The ratio of the average travel distance of all drones to W is recorded as g. exp(-g) is used as a power saving index to evaluate the power consumption during drone swarm inspections; the smaller g is, the greater the power saving index and the stronger the endurance.

[0064] The average of the monitoring field of view coverage and power saving indicators is used as the global target value for the drone swarm.

[0065] Specifically, when each drone is not a particle in the particle swarm algorithm (i.e., before the particle swarm algorithm is executed), the power saving index is directly set to 1 / W.

[0066] Thus, the global target value of the drone swarm was obtained based on the positions of all drones or the particle positions when all drones are considered as particles.

[0067] Step S202: Record the drones that are detected to have abnormal construction behavior as frozen drones, and record the drones other than frozen drones as free drones.

[0068] Each drone performs image acquisition and lidar scanning every 0.1 seconds during inspection.

[0069] The system detects construction anomalies based on collected images and laser-accelerated point cloud scans. If any drone detects a construction anomaly at any given time, it is designated as a "frozen drone"; other drones are designated as "free drones." In other words, if no construction anomaly is detected, the drone is designated as a "free drone."

[0070] Step S203: The difference between the increase in the target value of the frozen drone and the increase in the target value of the nearest free drone is recorded as the thawing coefficient of the frozen drone.

[0071] Retrieve all frozen drones and all free drones at the current moment. Record the positions of all frozen drones and all free drones as the current position.

[0072] All frozen drones and all free drones are treated as particles and implemented according to the comparative examples described above.

[0073] In the particle algorithm of the above comparative embodiments, the particle position of any given particle is continuously updated. That is, the updated particle position can be obtained each time D2 is executed. Note that when the particle swarm algorithm is executed, although each drone is treated as a particle, it is the particle position that is updated, and the position of the drone does not change (the position of the drone is changed only after the particle swarm algorithm ends, see subsequent step S206 for details).

[0074] The difference between this embodiment and the comparative embodiment is that: After all the particles corresponding to the drones have undergone a position update (that is, after executing D2 in the comparative example), the target value of the drone corresponding to the particle is recorded as M1 at the updated particle position; the target value of the drone corresponding to the particle at the current position is recorded as M2, and (M1-M2) / M2 is recorded as the increase in the target value of the drone corresponding to the particle.

[0075] The increase in the target value of any frozen drone is denoted as Q1. For the current position of the frozen drone, among all the particles corresponding to the free drones, the particle whose position is closest to the current position is obtained, and the free drone corresponding to this particle is taken as the nearest neighbor free drone. The increase in the target value of this free drone is denoted as Q2. In particular, when there are multiple nearest neighbor free drones, only the free drone with the largest increase in target value is considered.

[0076] Let (Q1-Q2) / Q2 be denoted as the thawing coefficient of the frozen drone. Specifically, when the Euclidean distance between the position of the nearest neighbor free drone and the current position of the frozen drone is less than the safety distance Lx, the thawing coefficient is set to 0. Additionally, when the thawing coefficient is less than 0, it is set to 0.

[0077] A higher thawing coefficient indicates that, compared to the nearest-neighbor free drone, the frozen drone's current position, after being updated by the particle swarm optimization algorithm, has a larger target value, and the nearest-neighbor free drone is more likely to replace the frozen drone for construction anomaly detection. Therefore, the frozen drone is more suitable as a particle for optimizing and updating its position using the particle swarm optimization algorithm.

[0078] The smaller the thawing coefficient, the smaller the target value of the frozen drone's current position after being updated by the particle swarm optimization algorithm, compared to the nearest neighbor free drone. Furthermore, the frozen drone is the only reliable source for construction anomaly detection. Therefore, the frozen drone is less suitable as a particle for optimizing its position using the particle swarm optimization algorithm.

[0079] Step S204: Obtain the freeze coefficient of the free drone based on the suppression effect of the decrease in the target value of the free drone on the increase in the global target value.

[0080] After all the particles corresponding to the drones undergo a position update (i.e., after executing D2 in the comparative example), (M2-M1) / M1 is recorded as the decrease in the target value of each drone corresponding to a particle. The obtained global target value is denoted as F1, the optimal global target value before all the particles corresponding to the drones undergo a position update (i.e., before executing D2 in the comparative example) is denoted as F2, and (F1-F2) / F1 is recorded as the increase in the global target value, denoted as H1.

[0081] Obtain the average reduction in target value for all drones, M4. Let M5 be the reduction for any single drone. The ratio of the absolute value of (M5-M4) to M4 is denoted as the global reduction magnitude of the single drone, expressed as H2. A larger global reduction magnitude H2 indicates that the reduction in target value for the single drone is greater compared to the reduction in target value for all drones.

[0082] The ratio of the absolute values ​​of (H2-H1) to H1 is denoted as the suppression effect of the decrease in the target value of the free UAV on the increase in the global target value, and is represented by y. Specifically, when y is less than or equal to 0, let y=0.

[0083] The greater the suppression magnitude, the more the target value of the free drone decreases, while the global target value increases only slightly, indicating that the reduction in the free drone's target value significantly suppresses the increase in the global target value. Conversely, the smaller the suppression magnitude, the less the target value of the free drone decreases, while the global target value increases only slightly, indicating that the reduction in the free drone's target value does not affect the increase in the global target value.

[0084] Let exp(-y) be the freezing coefficient of the free drone. The larger the freezing coefficient (i.e., the smaller y), the more likely the free drone will not affect the global target value from tending to its maximum value, even if it does not follow the updated position obtained by the particle swarm algorithm. In this case, the free drone can be used to track construction anomalies during inspections without having to continue moving to the updated position when encountering anomalies, while not affecting the increase of the global target value of the drone swarm (i.e., it will not lead to an excessively small inspection range and power saving).

[0085] Step S205: Use the difference between the freezing coefficient and thawing coefficient of the nearest free drone and the frozen drone as the global compensation coefficient.

[0086] After all the particles corresponding to drones are updated once (that is, after executing D2 in the comparative example), for any particle corresponding to a frozen drone, obtain the particle corresponding to the nearest free drone. The drones corresponding to these two particles are regarded as a pair of free drones and frozen drones; for all frozen drones, obtain multiple pairs of free drones and frozen drones.

[0087] For any pair of free drones and frozen drones, let the freezing coefficient and thawing coefficient of the free drone and frozen drone be denoted as q1 and q2, respectively.

[0088] As an example, the mean of q1 and q2 is denoted as the coordination index of any pair of free drones and frozen drones.

[0089] As another example, let c denote the absolute value of the difference between q1 and q2, and let c+1 denote the ratio of the mean of q1 and q2 to the matching index. The purpose of using c+1 as the denominator is to avoid the denominator being equal to 0.

[0090] The larger the matching index, the larger the freezing coefficient and thawing coefficient of the free drone and the frozen drone have. This means that the free drone is more capable of replacing the frozen drone in tracking construction anomalies, while the frozen drone can promptly inspect uninspected areas.

[0091] The average of all coordination indicators for free-roaming and frozen drones is recorded as the global compensation coefficient. The larger the global compensation coefficient, the better the drone swarm can cooperate with each other after the construction process is completed, ensuring high coverage and long endurance while promptly detecting, tracking, and resolving construction anomalies.

[0092] Step S206: Correct the target value of the frozen drone using the thawing coefficient, correct the target value of the free drone using the freezing coefficient, correct the global target value using the global compensation coefficient, and update the position of each drone using the particle swarm algorithm to maximize the corrected target value of each drone and the corrected global target value.

[0093] After all the particles corresponding to the drones are updated once (that is, after executing D2 in the comparative embodiment), the target value of the frozen drones is corrected using the thawing coefficient. The corrected target value is (1+w1)×g1×S1, where w1 represents the thawing coefficient, S1 represents the target value of the frozen drones before correction (that is, the target value obtained in step S201), and g1 represents the first scaling coefficient. In this embodiment, g1=1, but in other embodiments, it can be set to other values, preferably within the range (0, 2).

[0094] The target value of the free drone is corrected using the freeze coefficient. The corrected target value is (1+w2)×g2×S2, where w2 represents the freeze coefficient, S2 represents the target value of the free drone before correction (i.e. the target value obtained in step S201), and g2 represents the first scaling coefficient. In this embodiment, g2=1. In other embodiments, it can be set to other values, preferably within the range (0, 2).

[0095] The global target value is corrected using a global compensation coefficient. The corrected global target value is (1+w3)×g3×S3, where w3 represents the global compensation coefficient, S3 represents the global target value before correction (i.e., the global target value obtained in step S201), and g3 represents the first scaling coefficient. In this embodiment, g3=0.5. In other embodiments, it can be set to other values, preferably within the range (0, 2).

[0096] Next, the maximum value between the corrected target value of the drone corresponding to each particle and the optimal target value of each particle is used as the optimal target value for each particle. The maximum value between the corrected global target value and the optimal global target value is used as the optimal global target value; the particle position corresponding to the optimal target value of each particle is recorded as the optimal position of each particle. Then, the velocity of each particle is re-acquired based on the optimal target values ​​of all particles and the optimal global target value.

[0097] Then, the particle positions of all drones are updated again according to D2. After updating the particle positions of all drones, the target value and global target value of each drone are obtained according to step S201. The thawing coefficient and freezing coefficient of all frozen drones and all free drones are obtained again according to steps S203-S205 of this embodiment. Then, the target values ​​of frozen drones and free drones, as well as the global target value, are corrected again according to step S206 of this embodiment.

[0098] Next, the maximum value between the corrected target value of the drone corresponding to each particle and the optimal target value of each particle is used as the optimal target value for each particle. The maximum value between the corrected global target value and the optimal global target value is used as the optimal global target value; the particle position corresponding to the optimal target value of each particle is recorded as the optimal position of each particle. Then, the velocity of each particle is re-acquired based on the optimal target values ​​of all particles and the optimal global target value.

[0099] This process continues until the particle swarm optimization algorithm in the comparative example ends. The optimal position of each particle is then used as the updated position for all frozen drones and all free drones. Drones at the updated positions have the maximum corrected target value and the corrected global target value.

[0100] Each drone moves from its current position to its updated position at a constant speed (e.g., 5 meters per second). It should be noted that when each drone moves, the inspection intensity and uninspected area of ​​each position need to be reacquired according to steps (1) and (2) in S201.

[0101] Specifically, for the frozen drone's updated position, if the Euclidean distance between the updated position and the previous position (i.e., the current position) is less than or equal to a preset response distance (e.g., less than or equal to 0.7×K1), then the frozen drone will no longer move to the updated position, but will instead track the construction machine exhibiting abnormal construction conditions. When the free drone detects abnormal construction conditions during its movement, it will stop moving to the updated position and instead track the abnormal construction machine.

[0102] This concludes the example.

[0103] As time progresses, the above implementation is repeated in real-time when the next moment arrives. By continuously implementing the above implementation, the position of each UAV is updated. During this process, UAVs fall into two categories: frozen UAVs that have detected abnormal construction activity and free UAVs that have not detected any construction activity. When updating the position of frozen UAVs, the particle swarm optimization algorithm corrects the target value of the frozen UAVs using a thawing coefficient. This thawing coefficient describes whether construction anomaly detection can be performed solely by frozen UAVs, or whether free UAVs can replace frozen UAVs for construction anomaly detection. This allows frozen UAVs, after obtaining the updated position using the corrected target value, to decide whether to track abnormal construction activity based on the distribution of free UAVs. This ensures that frozen UAVs continue to participate in inspections when abnormal construction activity is eliminated, and continue tracking abnormal construction activity even if it persists. Meanwhile, when updating the position of the free drones, the aforementioned particle swarm algorithm corrects the target value of the free drones using a freezing coefficient. This freezing coefficient describes whether the free drones can stop their inspection midway (i.e., stop moving towards the updated position). This allows drones to handle potential abnormal construction situations after obtaining the updated position using the corrected target value, without affecting the overall drone fleet's large global target value (i.e., large inspection coverage and power saving). Furthermore, a global compensation coefficient is used to correct the global target value, describing the cooperation and coordination between the free and frozen drones. Finally, when the particle swarm algorithm ends, each drone's updated position has the maximum corrected target value and the maximum corrected global target value. This ensures that the drone swarm can promptly inspect uninspected areas and has a large endurance, while enabling the free and frozen drones to cooperate and relay each other, allowing for timely detection, tracking, and contact with abnormal construction situations, thus ensuring the normal collaborative operation of the construction machine swarm.

[0104] Example 3: In step S201 of Example 2, images collected by the UAV are fused and stitched together using image processing algorithms to construct a panoramic top-down view of the construction area. Each UAV at any given time corresponds to a pixel position in the panoramic top-down view, referred to simply as its position. The specific methods include: Before construction begins, a drone (any drone in a swarm) is remotely controlled to fly over the construction area. The industrial camera on the drone faces downwards, capturing images of the construction area. In this embodiment, 4 frames are captured per second, each frame being 1024×1024 pixels. Adjacent frames have overlapping fields of view. The SIFT corner detection algorithm is used to obtain the corners of each frame. One frame is randomly selected as the reference image (in some embodiments, an image captured at the center of the construction area can also be selected as the reference image). Images with overlapping fields of view with the reference image (e.g., images from adjacent frames) are then obtained and recorded as the images to be fused. The corner points in the reference image and the image to be fused are matched (e.g., by using a normalized cross-correlation matching algorithm) to obtain all corner point pairs. Based on all corner point pairs, the homography matrix is ​​obtained using the RANSAC algorithm. The homography matrix is ​​then used to perform an affine transformation on the image to be fused. The affine-transformed image to be fused is aligned with the reference image (i.e., the pixels of the repeated fields of view are aligned). Then, the Gaussian pyramid fusion algorithm is used to fuse the affine-transformed image to be fused with the reference image. The resulting fused image is then used as the reference image again.

[0105] Then, images with overlapping fields of view with the reference image (e.g., adjacent frames of the image to be fused) are acquired and re-designated as the image to be fused. The reference image and the image to be fused are fused again using the method described above, and the resulting fused image is again used as the reference image. This process is repeated until all images have been fused into the reference image, at which point the reference image is used as the panoramic top-down view.

[0106] In addition, several (e.g., 5) positioning markers are placed in the construction area. The positioning coordinates of each positioning marker are obtained by using a positioning device (e.g., a Beidou positioning device). In this embodiment, the Mercator projection method is used to map the spherical coordinates obtained by the positioning device into two-dimensional planar coordinates.

[0107] Then, the pixel containing each positioning marker is manually located in the panoramic top-down view, and each pixel corresponds to a positioning coordinate. A linear interpolation algorithm is used to obtain the positioning coordinates of all pixels in the panoramic top-down view. In other embodiments, a keypoint detection network can also be used to obtain the pixel containing each positioning marker in the panoramic top-down view. The keypoint detection network is a well-known convolutional neural network, which will not be described in detail in this embodiment.

[0108] Each drone in the drone swarm is equipped with a positioning device. In Example 1, during drone inspection, the positioning coordinates of each drone are obtained, and the pixel with the closest Euclidean distance to the positioning coordinates of each drone is obtained in the panoramic top view. The position coordinates of this pixel are the pixel position of each drone in the panoramic top view, which is simply referred to as the position.

[0109] It should be noted that the position in this embodiment is dimensionless; during implementation, only the pixel position in the panoramic top-down view is considered, without regard to specific dimensions. Furthermore, pixel regions not belonging to the construction area are artificially demarcated in the panoramic top-down view; pixels within these regions are deleted, meaning that pixels within these regions do not participate in the implementation of any embodiments of this invention.

[0110] As an example, in Embodiment 2, the width K1 of the inspection field of view is 513. In other embodiments, K1 can also be set to other values, preferably 30% to 90% of the minimum width and height of the image captured by the UAV. When K1 is not an integer, K1 is rounded down. If the rounded result is not odd, the rounded result is incremented by 1 as the value of K1.

[0111] Example 4: In the above embodiment 2, the position of the frozen drone and the free drone is updated by using the particle swarm algorithm in the comparative embodiment. The implementation of the comparative embodiment includes: randomly assigning an initial velocity to each particle.

[0112] As an example, a specific method for randomly assigning an initial velocity to each particle is: A value, denoted as a, is randomly generated in the interval [0, 2π]. The vector [A×cos(a), A×sin(a)] is used as the initial velocity. In this embodiment, A=0.05×K1.

[0113] The comparative implementation includes: obtaining the next particle position of each particle based on the particle position and velocity of each particle.

[0114] As an example, the next particle position (denoted as p2) of each particle is obtained based on the particle position (denoted as p1) and velocity (denoted as v1) of each particle. The formula is: p2=p1+v1×t, where t is a preset scaling factor. In this embodiment, t=1 is used as an example for description.

[0115] The comparative implementation includes: re-acquiring the velocity of each particle based on the optimal target value of all particles and the optimal global target value.

[0116] As an example, the velocity of each particle is re-acquired based on the optimal target value of all particles and the optimal global target value, using the formula: v2 = z × v1 + c1 × r1 × L(pbest, p2) + c2 × r2 × gbest. Where v2 is the updated velocity of each particle (i.e., the re-acquired velocity of each particle), z represents the inertia weight (in this embodiment, z = 1.2 is used as an example); c1 and c2 represent the first and second learning factors (in this embodiment, c1 = 0.2 and c2 = 0.6 are used as examples); r1 and r2 represent random numbers between (0, 1); pbest represents the optimal position of each particle; gbest represents the two-dimensional global target vector (both dimensions of gbest are the optimal global target values); and L(pbest, p2) represents the displacement vector from p2 to pbest.

[0117] It should be noted that since the particle position and velocity do not have actual physical meaning, but are only algorithm parameters of the particle swarm algorithm, and the particle position and velocity are set and calculated in a panoramic top view, this embodiment does not consider the dimensions of the particle position and velocity. The particle position and velocity values ​​can be directly substituted into the above process for implementation.

[0118] This embodiment is an application of a well-known technique in the particle swarm optimization algorithm, and the specific principles of the above formulas will not be elaborated in this embodiment.

[0119] Example 5: In step S202 of Example 2, the method includes detecting whether there is a construction abnormality based on the acquired image and the point cloud obtained by laser scanning.

[0120] As an example, the specific method for detecting construction anomalies based on the acquired images and laser-accumulated point cloud scans is as follows: The YOLOv5 neural network is used to identify the construction machine in the image and obtain the bounding box of the construction machine. The point cloud corresponding to the center point of the bounding box is obtained. In this embodiment, the coordinates of the point cloud are three-dimensional coordinates (including horizontal plane coordinates and the depth value of the point cloud) with the UAV as the coordinate origin. Only the horizontal plane coordinates of the point cloud are considered. The positioning coordinates of the UAV are obtained by the positioning device. The positioning coordinates plus the horizontal plane coordinates of the point cloud are the positioning coordinates of the construction machine.

[0121] The system acquires the set travel trajectory of the construction machine, which represents a manually set path. It then acquires the closest distance between the construction machine's location coordinates and the travel trajectory. If the closest distance is greater than a preset distance (e.g., greater than 1.5 meters), the system determines that the drone has detected an anomaly and records the construction machine as an anomaly. If the closest distance is less than or equal to the preset distance, the system determines that the drone has not detected an anomaly.

[0122] When the drone tracks abnormal construction machinery, it uses the Kalman filter algorithm to track the machinery based on its location coordinates.

[0123] In this embodiment, when the drone's battery level is less than 20%, the drone will no longer perform inspections or participate in particle swarm optimization calculations. The drone will be remotely controlled to return to the charging station for recharging. If fewer than 6 drones are engaged in inspection tasks, 8 drones need to be added.

[0124] Example 6: In this embodiment, a server is installed outside the construction area. The server communicates with each drone via a 5G network. The images and point clouds collected by the drones, as well as positioning coordinates and other data, are sent to the server. The computer on the server runs the method described in the above embodiment, and the updated position obtained from the server is sent to each drone.

[0125] In addition, the server sends the images and panoramic top-down views collected by the drone, as well as the drone's position in the panoramic top-down view, to the monitoring platform. The monitoring platform has several monitors to display the images and panoramic top-down views collected by the drone. The monitoring personnel input the task data for each construction machine through the monitoring platform. After the task data is transmitted to the server, the server then transmits it to each construction machine.

[0126] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for unmanned autonomous collaborative control of machine fleets for road construction, characterized in that, The method includes the following steps: The fleet comprises several drones used for inspecting construction machinery; target values ​​for each drone are obtained based on its location relative to uninspected areas and its distance from the nearest drone, including: For any target object, the Euclidean distance between the i-th position in the uninspected area and the target object is denoted as . The inspection intensity at the i-th position is denoted as The intensity of the inspection is positively correlated with the time it takes for the drone to pass through the i-th location. Obtain the distance of attention in areas not inspected Where N1 represents the total number of N1 locations in the uninspected area; the target value of each target object is... It is negatively correlated with x2, and positively correlated with x2, where x2 represents the distance between each target object and its nearest target object; The global target value is obtained based on the monitoring field of view coverage and power saving indicators of all drones, including: For any target object and the inspection field of view area where the target object is located, obtain the sum of the inspection intensity of all locations within the inspection field of view area, denoted as y; the inspection necessity of the inspection field of view area is negatively correlated with y; obtain the mean of the inspection necessity of the inspection field of view area where all target objects are located, denoted as the monitoring field of view coverage rate. The initial position of the drone is recorded as its location; the Euclidean distance between the initial position and the target position is recorded as the travel distance; the power saving index is negatively correlated with the mean travel distance of all drones. The average of the monitored field of view coverage and power saving indicators is used as the global target value for the drone swarm. Drones that detect abnormalities in construction machinery are classified as frozen drones, and drones other than frozen drones are classified as free drones. The difference between the increase in the target value of the frozen drone and the increase in the target value of the nearest free drone is denoted as the unfreezing coefficient of the frozen drone; the freezing coefficient of the free drone is obtained based on the suppression effect of the decrease in the target value of the free drone on the increase in the global target value; the freezing coefficient is negatively correlated with the suppression effect; the average of the coordination index between the nearest free drone and the frozen drone is used as the global compensation coefficient. The target value of the frozen drone is corrected using the thawing coefficient, the target value of the free drone is corrected using the freezing coefficient, the global target value is corrected using the global compensation coefficient, and the position of each drone is updated using the particle swarm algorithm to maximize the corrected target value of each drone and the corrected global target value.

2. The unmanned autonomous collaborative control method for road construction fleets according to claim 1, characterized in that, The difference between the increase in the target value of the frozen drone and the increase in the target value of the nearest free drone is denoted as the freezing coefficient of the frozen drone. The specific steps involved are as follows: The current position of all drones is recorded as the current position; when the particle swarm algorithm is executed, all drones are treated as particles, and the particle positions corresponding to all drones are updated; After all the particles corresponding to drones have their positions updated, the target value of each drone corresponding to each particle is recorded as M1 at the updated particle position; the target value of each drone corresponding to each particle at the current position is recorded as M2, and (M1-M2) / M2 is recorded as the increase in the target value of each drone corresponding to each particle. The increase in the target value of any frozen drone is denoted as Q1. For the current position of the frozen drone, among all the particles corresponding to the free drones, the particle whose position is closest to the current position is obtained, and the free drone corresponding to this particle is taken as the nearest neighbor free drone. The increase in the target value of the nearest neighbor free drone is denoted as Q2. Let (Q1-Q2) / Q2 be denoted as the thawing coefficient of the frozen drone.

3. The unmanned autonomous collaborative control method for road construction fleets according to claim 2, characterized in that, The specific steps for obtaining the suppression amplitude are as follows: After all the particles corresponding to the drones have their positions updated once, (M2-M1) / M1 is recorded as the decrease in the target value of each particle's corresponding drone; the global target value is recorded as F1, the maximum value of the global target value before all the particles corresponding to the drones have their positions updated once is recorded as F2, and (F1-F2) / F1 is recorded as the increase in the global target value, denoted as H1; Obtain the average reduction of the target value of all drones, M4. The reduction of any free drone is denoted as M5. The ratio of the absolute value of (M5-M4) to M4 is denoted as the global reduction of the free drone, and is represented as H2. The ratio of the absolute values ​​of (H2-H1) to H1 is denoted as the extent to which the decrease in the target value of the free UAV suppresses the increase in the global target value.

4. The unmanned autonomous collaborative control method for road construction fleets according to claim 2, characterized in that, The specific steps involved in using the average of the coordination indicators of the nearest free drone and the frozen drone as the global compensation coefficient are as follows: After all the particles corresponding to drones are updated once, for any particle corresponding to a frozen drone, it is denoted as particle 1. The particle corresponding to the nearest free drone is obtained and denoted as particle 2. The drones corresponding to particle 2 and particle 1 are a pair of free drone and frozen drone. For all frozen drones, obtain all pairs of free drones and frozen drones; For any pair of free drones and frozen drones, the average of the freezing coefficient and thawing coefficient of the free drones and frozen drones is denoted as the coordination index of any pair of free drones and frozen drones; the average of all coordination indices of free drones and frozen drones is denoted as the global compensation coefficient.

5. The unmanned autonomous collaborative control method for road construction fleets according to claim 1, characterized in that, The specific steps involved in correcting the target value of the frozen drone using a thawing coefficient, correcting the target value of the free drone using a freezing coefficient, and correcting the global target value using a global compensation coefficient are as follows: The corrected target value of the frozen drone is positively correlated with the thawing coefficient; the corrected target value of the free drone is positively correlated with the freezing coefficient; and the corrected global target value is positively correlated with the global compensation coefficient.

6. The unmanned autonomous collaborative control method for road construction fleets according to claim 1 or 2, characterized in that, The specific steps included in the particle swarm optimization algorithm are as follows: D1: Treat all drones as particles, with each drone's current position as its initial position, each drone's target value as its optimal target value, and the global target value as the optimal global target value for the particle swarm; randomly assign an initial velocity to each particle. D2: The particle position of each particle is updated based on its position and velocity, and the particle positions of all particles are updated accordingly. For the target value of all particles and the global target value, the target value of the frozen drone is corrected using the thawing coefficient, and the target value of the free drone is corrected using the freezing coefficient. After correcting the global target value using the global compensation coefficient, the maximum value between the corrected target value and the optimal target value of each particle is used as the optimal target value of each particle. The maximum value between the corrected global target value and the optimal global target value is used as the optimal global target value. The particle position corresponding to the optimal target value of each particle is recorded as the optimal position of each particle. The velocity of each particle is re-acquired based on the optimal target value and the optimal global target value of all particles. D3: Repeat D2 until the optimal position of each particle no longer changes, or when the number of times D2 is repeated is greater than or equal to the preset number of times, then stop repeating D2 and the particle swarm algorithm ends.

7. The unmanned autonomous collaborative control method for road construction fleets according to claim 1, characterized in that, After updating the position of each drone using the particle swarm optimization algorithm, the specific steps include the following: Each drone moves to the updated position; however, for the frozen drone's updated position, if the Euclidean distance between the updated position and the original position is less than or equal to the preset response distance, the frozen drone will no longer move to the updated position, but will instead track the construction machine with the construction anomaly; when the free drone detects abnormal construction during its movement, it will stop moving to the updated position and instead track the construction machine with the construction anomaly.

8. The unmanned autonomous collaborative control method for road construction fleets according to claim 1, characterized in that, The specific steps for obtaining the inspection intensity are as follows: The location of each drone is recorded in real time, and the time when it leaves each location is abbreviated as the inspection time of each location; all inspection times corresponding to all locations are linearly normalized respectively. For all normalized inspection times at the same location, the average of all normalized inspection times at that location is used as the inspection index for all locations within the inspection field of view of that location. The inspection index with the highest value among all inspection indicators for each location is the inspection intensity for each location.

Images