An artificial intelligence-based unmanned aerial vehicle cluster control method and system

Abstract

The application discloses an unmanned aerial vehicle cluster control method and system based on artificial intelligence, relates to the technical field of unmanned aerial vehicle cluster control, and comprises the following steps: acquiring global cooperative situation data through a sensor, and preprocessing the global cooperative situation data to obtain preprocessed global cooperative situation data; calculating path cost indexes of each unmanned aerial vehicle through a fuzzy C-means clustering algorithm according to the preprocessed global cooperative situation data, generating a preliminary flight path, and dynamically adjusting the preliminary flight path according to real-time environmental feedback through a reinforcement learning algorithm; and sharing the adjusted preliminary flight path and global environmental information of the unmanned aerial vehicle cluster through a 5G network, analyzing the speed, heading angle, height, acceleration gradient and height change rate of each unmanned aerial vehicle, and generating cooperative control parameters; the application realizes efficient cooperative control and reliable execution of the unmanned aerial vehicle cluster in a dynamic environment, and improves overall cooperativeness and task completion capability.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) swarm control technology, and in particular to an artificial intelligence-based UAV swarm control method and system. Background Technology

[0002] With the rapid development of unmanned aerial vehicle (UAV) technology, UAV swarm control has become an important research area in modern military, civilian, and commercial applications. The core of UAV swarm control lies in enabling multiple UAVs to operate collaboratively in complex environments, including path planning, dynamic obstacle avoidance, and task allocation. In recent years, the introduction of artificial intelligence (AI) technology has provided new solutions for UAV swarm control. Furthermore, intelligent optimization algorithms such as reinforcement learning and genetic algorithms are also widely used in UAV swarm control to address the needs of real-time path optimization and collaborative decision-making in dynamic environments.

[0003] Traditional path planning algorithms (such as A* and Dijkstra's algorithm) struggle to handle the dynamic path adjustments of large-scale UAV swarms in real-time in complex environments, resulting in low efficiency and a tendency to get trapped in local optima. Existing technologies often rely on high-bandwidth, low-latency communication networks to handle real-time environmental feedback. However, in practical applications, the stability and bandwidth limitations of these networks can lead to information transmission delays or losses, thus affecting the real-time collaborative control effectiveness of UAV swarms. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an artificial intelligence-based drone swarm control method to solve the problem that traditional path planning algorithms are difficult to handle in real time the dynamic path adjustment of large-scale drone swarms in complex environments.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an artificial intelligence-based method for controlling a swarm of unmanned aerial vehicles (UAVs), which includes acquiring global collaborative situational data through sensors, performing preprocessing, and acquiring preprocessed global collaborative situational data. Based on the preprocessed global collaborative situational awareness data, the path cost index of each UAV is calculated using the fuzzy C-means clustering algorithm to generate a preliminary flight path. The preliminary flight path is then dynamically adjusted based on real-time environmental feedback using a reinforcement learning algorithm. The drone swarm shares the adjusted preliminary flight path and global environmental information through the 5G network, analyzes the speed, heading angle, altitude, acceleration gradient and altitude change rate of each drone, and generates cooperative control parameters; The optimal flight path is obtained by optimizing the cooperative control parameters through a genetic algorithm, and cooperative control commands are generated and distributed to the UAV swarm in real time through edge computing.

[0007] As a preferred embodiment of the AI-based UAV swarm control method of the present invention, the steps of acquiring global collaborative situational awareness data through sensors and preprocessing it to obtain preprocessed global collaborative situational awareness data are as follows. By using sensors to collect environmental perception data, UAV status data, and task and collaboration data, a comprehensive collaborative situational awareness data can be formed. The global collaborative situational awareness data is unified to a global coordinate system for time synchronization, and high-frequency noise in the global collaborative situational awareness data is removed. Nearest neighbor interpolation is used to fill missing values, and normalization processing is performed.

[0008] As a preferred embodiment of the AI-based UAV swarm control method of the present invention, the specific steps for calculating the path cost index of each UAV using a fuzzy C-means clustering algorithm based on preprocessed global collaborative situational awareness data are as follows: Based on the preprocessed global collaborative situational data, path association features are extracted by dimensionality reduction through principal component analysis and combined with mutual information evaluation. The number of cluster centers is set by the elbow rule, and a fuzzy factor is set based on empirical values. Combined with path association features, the membership degree of each UAV to each cluster center is calculated. The cluster centers are recalculated based on the membership degree, and iterative optimization is performed until the change in the cluster centers is less than the convergence threshold, thus obtaining the cluster region to which each drone belongs. Based on the cluster region to which each drone belongs, candidate flight paths to each cluster region are generated, and the path cost index of each candidate flight path is calculated under the preset evaluation criteria.

[0009] As a preferred embodiment of the artificial intelligence-based UAV swarm control method of the present invention, the generation of the preliminary flight path refers to performing multi-criteria decision analysis on each candidate flight path based on the path cost vector and using an ideal solution approximation algorithm, calculating the relative proximity between each candidate flight path and the positive ideal solution and the negative ideal solution, and selecting the candidate flight path with the optimal relative proximity.

[0010] As a preferred embodiment of the artificial intelligence-based drone swarm control method of the present invention, the specific steps of dynamically adjusting the initial flight path based on real-time environmental feedback using a reinforcement learning algorithm are as follows: Collect terrain change data and meteorological data to obtain real-time environmental feedback; Based on environmental perception data, UAV status data, and task and collaboration data, a state space and action space are defined, and a reward function is designed through a multi-objective optimization method to form a reinforcement learning framework. Based on real-time environmental feedback, the initial flight path is dynamically adjusted by combining a reinforcement learning framework with a near-end policy optimization algorithm to obtain the adjusted initial flight path.

[0011] As a preferred embodiment of the artificial intelligence-based drone swarm control method of the present invention, the drone swarm shares the adjusted preliminary flight path and global environmental information through a 5G network. The specific steps are as follows. Data fusion is performed on the preprocessed global collaborative situational awareness data, and principal component analysis and DS evidence theory are combined to generate global environmental information; The adjusted preliminary flight path and global environmental information are encapsulated into structured data packets, binary encoded using the Protobuf protocol, and compressed into terrain raster data using Huffman coding.

[0012] As a preferred embodiment of the artificial intelligence-based UAV swarm control method of the present invention, the specific steps for analyzing the speed, heading angle, altitude, acceleration gradient, and altitude change rate of each UAV to generate cooperative control parameters are as follows. Based on terrain raster data, transmission time slots are dynamically allocated according to the type of UAV service through TSN proxy nodes, and microsecond-level clock synchronization is performed using the NTP-Precise protocol. By combining visual odometry with Kalman filtering, the speed, heading angle, altitude, acceleration gradient, and altitude change rate of each UAV are extracted to construct a six-dimensional state vector; The six-dimensional state vectors of adjacent UAVs within a radius R are obtained through 5G D2D communication, and the cooperative control parameters of each UAV are output through distributed solution.

[0013] As a preferred embodiment of the artificial intelligence-based UAV swarm control method of the present invention, the step of optimizing the cooperative control parameters through a genetic algorithm to obtain the optimal flight path includes the following specific steps. The cooperative control parameters were encoded and segmented into time slices, mapped to multidimensional real-valued gene sequences, and an initial population was generated using the Latin hypercube sampling method. The initial population is iteratively optimized through tournament selection, simulated binary crossover, and directed polynomial mutation until it converges to the maximum number of iterations to obtain the optimal chromosome. The optimal chromosome is analyzed by time slice to obtain time series path points. The feasibility is verified by combining dynamic model and reinforcement learning, and the optimal flight path is output.

[0014] As a preferred embodiment of the artificial intelligence-based drone swarm control method of the present invention, the specific steps of generating collaborative control commands through edge computing and distributing them to the drone swarm in real time are as follows. A nested JSON structure is constructed, and Python is used to encapsulate and serialize the optimal flight path and cooperative control parameters into JSON format instructions. The AES-128 algorithm is used to generate a byte key to encrypt the JSON format instructions in CBC mode. By combining 5G network slicing with TSN proxy nodes to dynamically allocate transmission time slots, encrypted JSON format commands are distributed to the drone cluster in real time.

[0015] Secondly, the present invention provides an artificial intelligence-based unmanned aerial vehicle (UAV) swarm control system, including a data acquisition module, a path planning module, a control parameter module, and an instruction distribution module; The data acquisition module is used to acquire global collaborative situational awareness data through sensors, and to preprocess the data to obtain preprocessed global collaborative situational awareness data. The path planning module is used to calculate the path cost index of each UAV based on the preprocessed global collaborative situational data, using the fuzzy C-means clustering algorithm to generate a preliminary flight path, and to dynamically adjust the preliminary flight path based on real-time environmental feedback using a reinforcement learning algorithm. The control parameter module is used for the drone swarm to share the adjusted preliminary flight path and global environmental information through the 5G network, analyze the speed, heading angle, altitude, acceleration gradient and altitude change rate of each drone, and generate cooperative control parameters. The command distribution module is used to optimize the cooperative control parameters through a genetic algorithm to obtain the optimal flight path, and to generate cooperative control commands through edge computing methods and distribute them to the UAV cluster in real time.

[0016] The beneficial effects of this invention are as follows: Through global situational awareness and intelligent preprocessing, a unified understanding of complex environments and mission requirements is achieved by the UAV swarm; preliminary flight paths are generated using fuzzy clustering and multi-criteria decision-making, and real-time adaptive adjustments are made through reinforcement learning, making the path planning both stable and flexible; furthermore, control parameters are formed through swarm information sharing and distributed collaborative solving, and commands are issued in real time by combining global optimization and edge computing, realizing efficient collaborative control and reliable execution of the UAV swarm in dynamic environments, thereby improving overall collaboration and mission completion capabilities. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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 of an artificial intelligence-based drone swarm control method.

[0019] Figure 2 This is a schematic diagram of an artificial intelligence-based drone swarm control system.

[0020] Figure 3 This is a flowchart for calculating the path cost index based on fuzzy C-means clustering.

[0021] Figure 4 This is a flowchart of the initial dynamic adjustment of the flight path based on reinforcement learning. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is the first embodiment of the present invention, which provides an artificial intelligence-based unmanned aerial vehicle (UAV) swarm control method, including the following steps: S1. Acquire global collaborative situational awareness data through sensors and preprocess it to obtain preprocessed global collaborative situational awareness data.

[0026] By using sensors to collect environmental perception data, UAV status data, and mission and collaboration data, a comprehensive collaborative situational awareness data can be formed.

[0027] Unify the global collaborative situational awareness data into a global coordinate system for time synchronization.

[0028] It should be noted that coordinate transformation algorithms (such as Euler angles or quaternion transformation) are used to transform the global collaborative situation data to a unified global coordinate system; time synchronization methods (such as interpolation or nearest neighbor matching) are used to align the global collaborative situation data in time, ensuring that all global collaborative situation data are consistent under the same time reference, thus achieving spatiotemporal unification of global collaborative situation data in the global coordinate system.

[0029] High-frequency noise in the global collaborative situational awareness data is removed, missing values ​​are filled using nearest neighbor interpolation, and normalization is performed.

[0030] It should be noted that appropriate wavelet basis functions and decomposition levels are selected to perform wavelet decomposition on the global cooperative situation data; thresholding is used to filter out noise in high-frequency components while retaining effective low-frequency information; nearest neighbor interpolation is used to fill in missing values ​​in the global cooperative situation data to ensure its continuity; and Z-score normalization is used to normalize the global cooperative situation data, converting it into a distribution with a mean of 0 and a standard deviation of 1, ensuring analysis at the same scale.

[0031] S2. Based on the preprocessed global collaborative situational awareness data, the path cost index of each UAV is calculated using the fuzzy C-means clustering algorithm to generate a preliminary flight path.

[0032] Based on the preprocessed global collaborative situational data, path association features are extracted by dimensionality reduction through principal component analysis and combined with mutual information evaluation.

[0033] It should be noted that Principal Component Analysis (PCA) is used to reduce the dimensionality of the high-dimensional preprocessed global collaborative situational data, retaining the principal components to reduce computational complexity; mutual information is used to evaluate the correlation between each principal component and the path planning objective, and low-dimensional features that are strongly correlated with path planning are screened out to obtain path-related features.

[0034] The number of cluster centers is set by the elbow rule, and a fuzzy factor is set based on empirical values. Combined with path association features, the membership degree of each UAV to each cluster center is calculated.

[0035] Specifically as follows: Based on path association characteristics, the sum of squared clustering errors (SSE) is calculated for different numbers of cluster centers (from 1 to a preset maximum value), and a curve of SSE versus number of cluster centers is plotted. The inflection point (i.e., the elbow) of the curve is observed, and the number of cluster centers corresponding to the inflection point is selected as the optimal value. A fuzzy factor is set based on empirical values ​​(usually 1.5 to 2.5) to control the degree of fuzziness in clustering. Combining path association characteristics, the membership degree of each UAV to each cluster center is calculated using the fuzzy C-means clustering algorithm. The membership degree of each drone to each cluster center is calculated using the following expression: ; in, It is the first The drone belongs to the first Membership degree of each cluster center It is the first Path association characteristics of drones It is the first Cluster centers, It is the first Cluster centers, It is a fuzzy factor. It is the index variable of the cluster center. It is the number of cluster centers; It should be noted that Indicates the current cluster center index (the index of the current cluster center). (cluster centers) This represents the index for traversing all cluster centers (from 1 to 1). ); The path association features of the UAV are obtained by dimensionality reduction of the preprocessed global collaborative situational data through principal component analysis, and by screening the low-dimensional features with the greatest correlation to the path planning objectives through mutual information evaluation. The path association features include environmental constraint features, flight cost features, safety and risk features, collaborative features and mission-related features. Through normalization processing, features with different physical dimensions are mapped to a unified scale space with a mean of 0 and a standard deviation of 1, thereby ensuring the correctness of the normalization processing and the effectiveness of subsequent clustering calculations. The fuzzy factor is predetermined based on the empirical range of values ​​for the fuzzy C-means clustering algorithm, combined with the requirements for clustering stability and fuzziness.

[0036] The cluster centers are recalculated based on membership and iteratively optimized until the change in cluster centers is less than the convergence threshold, thus obtaining the cluster region to which each drone belongs.

[0037] It should be noted that, based on the current membership degree and path association characteristics, the change in cluster centers is checked to see if it is less than the preset convergence threshold (e.g., 10⁻⁵). If the condition is not met, the above process of calculating the membership degree of each drone to each cluster center is repeated until the change in cluster centers is less than the convergence threshold. Finally, based on the final membership degree distribution, each drone is assigned to the cluster region with the highest membership degree. It should also be noted that a relatively small convergence threshold (such as 10⁻⁵ or 10⁻⁶) should be initially set to ensure that the changes in cluster centers are small enough. Through verification, the convergence speed and clustering effect of the fuzzy C-means clustering algorithm under different convergence thresholds should be observed to find a balance between accuracy and computational cost. Finally, combined with empirical values, a convergence threshold that can both guarantee clustering quality and avoid excessive iteration should be determined.

[0038] Based on the cluster region to which each drone belongs, candidate flight paths to each cluster region are generated, and the path cost index of each candidate flight path is calculated under the preset evaluation criteria.

[0039] It should be noted that, based on the results of UAV clustering region division, the clustering region to which each UAV belongs is determined, and the corresponding cluster center position is used as the target position of the clustering region; under the unified global coordinate system and time synchronization constraints, starting from the current position of each UAV, combined with environmental perception data, UAV status data and task and collaboration data in the global collaborative situational data, candidate flight paths to each clustering region are generated. The candidate flight paths consist of a sequence of waypoints arranged in chronological order, and the positions of the no-fly zone boundary and obstacle boundary are checked and adjusted point by point during the generation process to ensure that the candidate flight paths meet the spatial constraints; After the candidate flight paths are generated, the path length cost, energy consumption cost, safety interval cost and mission deviation cost are calculated for each candidate flight path in sequence according to the preset evaluation criteria. The results of each cost are summarized to form the corresponding path cost index, which is used for subsequent candidate flight path optimization.

[0040] It should also be noted that the evaluation criteria are used to comprehensively evaluate candidate flight paths leading to each cluster region. The evaluation criteria consist of evaluation items such as path length cost, energy consumption cost, safety interval cost, and mission deviation cost, and are used to reflect the merits of candidate flight paths from multiple aspects such as flight efficiency, safety, and mission adaptability.

[0041] Based on the path cost vector, an ideal solution approximation algorithm is used to perform multi-criteria decision analysis on each candidate flight path, calculate the relative proximity between each candidate flight path and the positive ideal solution and the negative ideal solution, and select the candidate flight path with the best relative proximity as the initial flight path.

[0042] It should be noted that, based on the path cost index, the path cost index of each candidate flight path is arranged into a path cost vector according to the same evaluation item order in accordance with the preset evaluation criteria. The path cost vector is then scaled to eliminate the difference in magnitude between path length cost, energy consumption cost, safety interval cost, and mission deviation cost. The evaluation item attributes in the preset evaluation criteria are used to distinguish between cost-type evaluation items and benefit-type evaluation items. The minimum value of the cost-type evaluation item and the maximum value of the benefit-type evaluation item are selected from the path cost vectors of all candidate flight paths to form a positive ideal solution. The maximum value of the cost-type evaluation item and the minimum value of the benefit-type evaluation item are selected from the path cost vectors of all candidate flight paths to form a negative ideal solution. For each candidate flight path, the distance from the path cost vector to the positive ideal solution and the distance from the path cost vector to the negative ideal solution are calculated. The distance calculation is performed by summing the squared differences of each item and then taking the square root. Based on the distance values, the relative proximity between each candidate flight path and the positive and negative ideal solutions is calculated. The relative proximity calculation is performed by using the ratio of the distance value of the negative ideal solution to the sum of the distance values ​​of the positive ideal solution to form a normalized proximity, which reflects the comprehensive degree of the candidate flight path's proximity to the positive ideal solution and its distance from the negative ideal solution. Finally, the candidate flight paths are sorted from largest to smallest relative proximity, and the candidate flight path with the best relative proximity is selected as the initial flight path.

[0043] It should also be noted that the ideal solution approximation algorithm is an evaluation method based on multi-criteria decision analysis. It forms an evaluation vector by forming multiple candidate solutions under a unified evaluation criterion, and constructs a positive ideal solution composed of the optimal values ​​of each evaluation item and a negative ideal solution composed of the worst values ​​of each evaluation item in the evaluation vector space. Then, it calculates the distance relationship between each candidate solution and the positive ideal solution and the negative ideal solution respectively. The degree to which the candidate solution is close to the positive ideal solution and far away from the negative ideal solution is used as the comprehensive evaluation basis, thereby realizing the ranking and selection of candidate solutions under multi-criteria conditions.

[0044] S3. The initial flight path is dynamically adjusted based on real-time environmental feedback using a reinforcement learning algorithm.

[0045] Collect terrain change data and meteorological data to obtain real-time environmental feedback.

[0046] Based on environmental perception data, UAV status data, and task and collaboration data, a state space and action space are defined, and a reward function is designed through a multi-objective optimization method to form a reinforcement learning framework.

[0047] It should be noted that environmental perception data (such as terrain changes and meteorological information), UAV state data (such as battery level, speed, and payload), and task and collaboration data (such as task objectives and collaboration strategies) are integrated into a multi-dimensional state vector to define the state space; the possible actions of the UAV (such as acceleration, turning, and altitude adjustment) are defined as the action space; and a reward function is defined through a multi-objective optimization method, comprehensively considering multiple objectives such as task completion, energy consumption, and obstacle avoidance effect, assigning corresponding weights to different objectives to ensure that the reward function can comprehensively reflect the quality of UAV actions. By combining the state space, action space, and reward function, a complete reinforcement learning framework is constructed.

[0048] Based on real-time environmental feedback, the initial flight path is dynamically adjusted by combining a reinforcement learning framework with a near-end policy optimization algorithm to obtain the adjusted initial flight path.

[0049] It should be noted that environmental perception data (such as terrain changes and meteorological information) is acquired through multi-source sensors and integrated with UAV status data (such as battery level, speed, and payload) and task data (such as target location and task priority) to form the current state, which is then input into the reinforcement learning framework. Using the Proximal Policy Optimization (PPO) algorithm, an optimized action strategy (such as adjusting speed, heading angle, and altitude) is generated based on the current state and the reward function (comprehensively considering objectives such as task completion, energy consumption, and obstacle avoidance). The optimized strategy is executed to update the initial flight path of the UAVs, obtain the adjusted initial flight path, and continuously collect new environmental feedback data to iteratively optimize the strategy until the task is completed or the preset conditions are met, ultimately achieving dynamic adjustment of the flight path of the UAV swarm.

[0050] S4. The drone swarm shares the adjusted preliminary flight path and global environmental information through the 5G network, analyzes the speed, heading angle, altitude, acceleration gradient and altitude change rate of each drone, and generates cooperative control parameters.

[0051] Data fusion is performed on the preprocessed global collaborative situational awareness data, and global environmental information is generated by combining principal component analysis and DS evidence theory.

[0052] It should be noted that the system state is predicted through nonlinear state equations and observation equations. The predicted state value and error covariance are calculated using cubature points. The state estimate and covariance matrix are then updated by combining the observed values, achieving spatiotemporal unification and noise reduction of multi-source data. The covariance matrix of the data is calculated, eigenvalues ​​and eigenvectors are extracted, and the principal component with the largest contribution rate is selected. The original data is projected into a low-dimensional space, retaining the main features and reducing redundant information. A Frame of Discernment is defined, and a Basic Probability Assignment (BPA) function is assigned to each evidence source. The BPA functions of multi-source evidence are fused using the Dempster combination rule. The confidence intervals and uncertainty values ​​of each hypothesis are calculated, and an uncertainty model is constructed based on these results. The fused state estimate, dimensionality reduction features, and uncertainty model are integrated to generate high-precision, low-uncertainty global environmental information. It should also be noted that the uncertainty model is constructed using DS evidence theory to quantify the confidence intervals and uncertainties of each hypothesis during the fusion of multi-source data. It is calculated using the Basic Probability Assignment (BPA) function and Dempster's combination rule to ultimately generate high-precision, low-uncertainty global environmental information.

[0053] The adjusted flight path and global environmental information are encapsulated into structured data packets, binary encoded using the Protobuf protocol, and compressed into terrain raster data using Huffman coding.

[0054] It should be noted that the adjusted flight path and global environment information are encapsulated in a predefined structured format (such as JSON or a custom data structure) to ensure the integrity and logical consistency of the fields. Secondly, the structured data packets are serialized into binary format using the Protobuf protocol. The data structure is described by a predefined proto file, and the efficient encoding mechanism of Protobuf is used to convert the fields into a compact binary stream, significantly reducing the volume. Huffman coding is performed on the binary data, and an optimal prefix code table is constructed based on the frequency distribution of statistical data. High-frequency data is represented by short codes, and low-frequency data is represented by long codes, further compressing the data volume and generating compressed terrain raster data.

[0055] Based on terrain raster data, transmission time slots are dynamically allocated according to the type of UAV service through TSN proxy nodes, and microsecond-level clock synchronization is performed using the NTP-Precise protocol.

[0056] It should be noted that the TSN agent node dynamically allocates transmission time slots based on the priority and bandwidth requirements of the UAV service type (such as real-time control, data transmission, and environmental monitoring). A time-aware shaper allocates high-priority services (such as real-time control) to fixed time slots and low-priority services (such as data transmission) to remaining time slots, ensuring efficient service transmission. The NTP-Precise protocol is used for microsecond-level clock synchronization. Through a master-slave clock architecture, the master node sends timestamp messages, the slave node receives and calculates the clock deviation, uses filters to smooth clock errors, and achieves microsecond-level precision synchronization through frequency adjustment and phase compensation.

[0057] By combining visual odometry with Kalman filtering, the speed, heading angle, altitude, acceleration gradient, and altitude change rate of each UAV are extracted to construct a six-dimensional state vector.

[0058] It should be noted that continuous frame images are extracted from the camera or visual sensor on the UAV using visual odometry. The relative displacement and attitude changes of the UAV are calculated through feature point matching and motion estimation to initially obtain velocity, heading angle and altitude information. These observations are then input into a Kalman filter. Through two steps, state prediction (based on the motion model) and state update (combined with the observations), the estimated velocity, heading angle, altitude, acceleration gradient and altitude change rate of the UAV are optimized to eliminate noise and improve accuracy. These optimized state variables are then integrated into a six-dimensional state vector (containing velocity, heading angle, altitude, acceleration gradient, altitude change rate and timestamp).

[0059] The six-dimensional state vectors of adjacent UAVs within a radius R are obtained through 5G D2D communication, and the cooperative control parameters of each UAV are output through distributed solution.

[0060] It should be noted that each UAV exchanges six-dimensional state vectors (including velocity, heading angle, altitude, acceleration gradient, altitude change rate, and timestamp) with neighboring UAVs in real time within a radius R via 5G D2D communication, constructing local network state information. The cooperative control problem is modeled as a distributed optimization problem, with the goal of minimizing the overall energy consumption and path deviation of the UAV swarm while satisfying constraints such as obstacle avoidance and task completion. The Alternating Direction Multiplier Method (ADMM) is used to decompose the global optimization problem into multiple subproblems. Each UAV solves the subproblems independently based on its local state information. Constraints are linearized by introducing Lagrange multipliers and penalty functions. The local solution and Lagrange multipliers are updated in each iteration, and intermediate results are exchanged with neighboring UAVs through communication to gradually approach the global optimal solution. Each UAV generates cooperative control parameters (such as velocity, heading angle, and altitude adjustment) based on the solution results.

[0061] S5. Optimize the cooperative control parameters using a genetic algorithm to obtain the optimal flight path.

[0062] The collaborative control parameters were encoded and segmented by time slices, mapped to multidimensional real-valued gene sequences, and an initial population was generated using the Latin hypercube sampling method.

[0063] It should be noted that the cooperative control parameters (such as speed, heading angle, altitude adjustment, etc.) are divided into time slices, with each time slice corresponding to a control parameter vector. Arrange them in chronological order into a multidimensional real-valued gene sequence. ,in The first step is to determine the total number of time slices. The second step is to generate an initial population using the Latin hypercube sampling method. The intervals are uniformly divided within the domain of each dimension parameter to ensure that each interval is sampled only once, generating uniformly distributed sample points. These sample points are then randomly combined into multidimensional gene sequences to form the initial population.

[0064] The initial population is iteratively optimized through tournament selection, simulated binary crossover, and directed polynomial mutation until it converges to the maximum number of iterations to obtain the optimal chromosome.

[0065] Specifically as follows: A tournament selection method is used to select high-quality individuals from the initial population. Several individuals (e.g., 2 or 3) are randomly selected for comparison, and the individual with the highest fitness is chosen as the parent to enter the next generation. Simulated binary crossover (SBX) is performed on the selected parent individuals, and offspring individuals are generated by controlling the crossover probability. The expression is: ; ; in, It is the gene value of the first offspring individual. It is the gene value of the second offspring individual. It is the genetic value of the first parent individual. It is the gene value of the second parent individual, which is obtained by the tournament selection method in the initial population or the current generation population, based on the fitness comparison results, from the gene value of the corresponding position in the gene sequence of another parent individual; The crossover distribution index is used as a control parameter for the simulated binary crossover method. It is pre-set before the crossover operation based on the requirements for the ability to explore the search space and the convergence stability. Directed polynomial mutation is performed on offspring individuals, and mutant individuals are generated by controlling the mutation probability. The expression is as follows: ; in, It is the new gene value of the offspring after the mutation. It is the upper limit of gene values, determined by the physical constraint range corresponding to the collaborative control parameters, and is given in advance as the maximum allowable value of each gene dimension during the gene encoding stage; The lower limit of gene values ​​is determined by the physical constraint range corresponding to the collaborative control parameters, and is given in advance as the minimum allowable value for each dimension of gene during the gene encoding stage; The mutation distribution index is used as a control parameter for the directional multinomial mutation method. It is preset before the mutation operation based on the mutation amplitude and the local search capability requirements. The offspring's gene value is obtained by performing cross-calculation on the parent's gene value using a simulated binary cross-calculation method, and then taking the value of the corresponding gene position in the offspring. Repeat the tournament selection, simulated binary crossover, and directed polynomial mutation processes until the maximum number of iterations is reached (e.g., 100 in the example). After all iterations are completed, select the chromosome with the highest fitness from the current population as the optimal solution.

[0066] The optimal chromosome is analyzed by time slice to obtain time series path points. The feasibility is verified by combining dynamic model and reinforcement learning, and the optimal flight path is output.

[0067] Specifically as follows: The optimal chromosome is analyzed by dividing it into time slices, with each time slice corresponding to a control parameter vector. The control parameters are then mapped to time series path points through integration. The feasibility of the path points is verified by combining the UAV dynamics model (such as the six-degree-of-freedom model) to ensure that the path meets the speed, acceleration and attitude constraints, such as checking whether the speed is within the maximum speed range and whether the heading angle is within the turning angle limit. Then, the path points are further optimized using a reinforcement learning framework to dynamically adjust the path points to improve performance. Finally, the verified and optimized optimal flight path is output.

[0068] S6. Generate collaborative control commands through edge computing methods and distribute them to the drone swarm in real time.

[0069] A nested JSON structure is constructed, and Python is used to encapsulate and serialize the optimal flight path and cooperative control parameters into JSON format instructions. The AES-128 algorithm is used to generate a byte key to encrypt the JSON format instructions in CBC mode.

[0070] It should be noted that a nested JSON structure is constructed, encapsulating the optimal flight path and cooperative control parameters (such as speed, heading angle, altitude adjustment, etc.) into a nested dictionary. Python is used to serialize the nested dictionary into a JSON format string, generating a 16-byte random key. Python's cryptography library is used to encrypt the JSON string using AES-128, employing CBC (Cipher Block Chaining) mode to padded the JSON string to multiples of 16 bytes, generating a random initialization vector. The data is then encrypted in blocks using the key and the random initialization vector, and finally, the encrypted byte stream is output.

[0071] By combining 5G network slicing with TSN proxy nodes to dynamically allocate transmission time slots, encrypted JSON format commands are distributed to the drone cluster in real time.

[0072] It should be noted that 5G network slicing technology is used to create a dedicated slice for the drone swarm, allocating high-priority, low-latency network resources to ensure the real-time and reliable transmission of commands. The TSN agent node dynamically allocates transmission time slots according to the drone service type (such as real-time control and data transmission), and uses a time-aware shaper to allocate encrypted JSON format commands to fixed time slots to ensure that high-priority commands are transmitted first. Then, the encrypted commands are sent to the drone swarm through the 5G slice network.

[0073] This embodiment also provides an artificial intelligence-based unmanned aerial vehicle (UAV) swarm control system, including: a data acquisition module, a path planning module, a control parameter module, and a command distribution module; The data acquisition module is used to acquire global collaborative situational awareness data through sensors, and to preprocess the data to obtain preprocessed global collaborative situational awareness data. The path planning module is used to calculate the path cost index of each UAV based on the preprocessed global collaborative situational data, using the fuzzy C-means clustering algorithm to generate a preliminary flight path, and to dynamically adjust the preliminary flight path based on real-time environmental feedback using a reinforcement learning algorithm. The control parameter module is used for the drone swarm to share the adjusted preliminary flight path and global environmental information through the 5G network, analyze the speed, heading angle, altitude, acceleration gradient and altitude change rate of each drone, and generate cooperative control parameters. The command distribution module is used to optimize the cooperative control parameters through a genetic algorithm to obtain the optimal flight path, and to generate cooperative control commands through edge computing methods and distribute them to the UAV cluster in real time.

[0074] In summary, this invention achieves a unified understanding of complex environments and mission requirements for UAV swarms through: global situational awareness and intelligent preprocessing; it generates preliminary flight paths using fuzzy clustering and multi-criteria decision-making, and performs real-time adaptive adjustments through reinforcement learning, making path planning both stable and flexible; furthermore, it forms control parameters through swarm information sharing and distributed collaborative solving, and combines global optimization and edge computing to issue commands in real time, achieving efficient collaborative control and reliable execution of UAV swarms in dynamic environments, thus improving overall collaboration and mission completion capabilities.

[0075] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for controlling a swarm of unmanned aerial vehicles (UAVs) based on artificial intelligence, characterized in that: include, Global collaborative situational awareness data is acquired through sensors and preprocessed to obtain preprocessed global collaborative situational awareness data. Based on the preprocessed global collaborative situational awareness data, the path cost index of each UAV is calculated using the fuzzy C-means clustering algorithm to generate a preliminary flight path. The preliminary flight path is then dynamically adjusted based on real-time environmental feedback using a reinforcement learning algorithm. The drone swarm shares the adjusted preliminary flight path and global environmental information through the 5G network, analyzes the speed, heading angle, altitude, acceleration gradient and altitude change rate of each drone, and generates cooperative control parameters; The optimal flight path is obtained by optimizing the cooperative control parameters through a genetic algorithm, and cooperative control commands are generated and distributed to the UAV swarm in real time through edge computing.

2. The artificial intelligence-based drone swarm control method as described in claim 1, characterized in that: The steps for acquiring global collaborative situational awareness data through sensors and preprocessing it to obtain preprocessed global collaborative situational awareness data are as follows. By using sensors to collect environmental perception data, UAV status data, and task and collaboration data, a comprehensive collaborative situational awareness data can be formed. The global collaborative situational awareness data is unified to a global coordinate system for time synchronization, and high-frequency noise in the global collaborative situational awareness data is removed. Nearest neighbor interpolation is used to fill missing values, and normalization processing is performed.

3. The artificial intelligence-based drone swarm control method as described in claim 1, characterized in that: The specific steps for calculating the path cost index of each UAV based on the preprocessed global collaborative situational awareness data using the fuzzy C-means clustering algorithm are as follows: Based on the preprocessed global collaborative situational data, path association features are extracted by dimensionality reduction through principal component analysis and combined with mutual information evaluation. The number of cluster centers is set by the elbow rule, and a fuzzy factor is set based on empirical values. Combined with path association features, the membership degree of each UAV to each cluster center is calculated. The cluster centers are recalculated based on the membership degree, and iterative optimization is performed until the change in the cluster centers is less than the convergence threshold, thus obtaining the cluster region to which each drone belongs. Based on the cluster region to which each drone belongs, candidate flight paths to each cluster region are generated, and the path cost index of each candidate flight path is calculated under the preset evaluation criteria.

4. The artificial intelligence-based drone swarm control method as described in claim 1, characterized in that: The generation of the preliminary flight path refers to performing multi-criteria decision analysis on each candidate flight path based on the path cost vector and using an ideal solution approximation algorithm, calculating the relative proximity between each candidate flight path and the positive and negative ideal solutions, and selecting the candidate flight path with the optimal relative proximity.

5. The artificial intelligence-based drone swarm control method as described in claim 4, characterized in that: The initial flight path is dynamically adjusted based on real-time environmental feedback using a reinforcement learning algorithm. The specific steps are as follows: Collect terrain change data and meteorological data to obtain real-time environmental feedback; Based on environmental perception data, UAV status data, and task and collaboration data, a state space and action space are defined, and a reward function is designed through a multi-objective optimization method to form a reinforcement learning framework. Based on real-time environmental feedback, the initial flight path is dynamically adjusted by combining a reinforcement learning framework with a near-end policy optimization algorithm to obtain the adjusted initial flight path.

6. The artificial intelligence-based drone swarm control method as described in claim 1, characterized in that: The drone swarm shares the adjusted preliminary flight path and global environmental information via a 5G network. The specific steps are as follows. Data fusion is performed on the preprocessed global collaborative situational awareness data, and principal component analysis and DS evidence theory are combined to generate global environmental information; The adjusted preliminary flight path and global environmental information are encapsulated into structured data packets, binary encoded using the Protobuf protocol, and compressed into terrain raster data using Huffman coding.

7. The artificial intelligence-based drone swarm control method as described in claim 1, characterized in that: The analysis of each UAV's speed, heading angle, altitude, acceleration gradient, and rate of change of altitude generates cooperative control parameters. The specific steps are as follows: Based on terrain raster data, transmission time slots are dynamically allocated according to the type of UAV service through TSN proxy nodes, and microsecond-level clock synchronization is performed using the NTP-Precise protocol. By combining visual odometry with Kalman filtering, the speed, heading angle, altitude, acceleration gradient, and altitude change rate of each UAV are extracted to construct a six-dimensional state vector; The six-dimensional state vectors of adjacent UAVs within a radius R are obtained through 5G D2D communication, and the cooperative control parameters of each UAV are output through distributed solution.

8. The artificial intelligence-based drone swarm control method as described in claim 7, characterized in that: The optimal flight path is obtained by optimizing the cooperative control parameters using a genetic algorithm. The specific steps are as follows: The cooperative control parameters were encoded and segmented into time slices, mapped to multidimensional real-valued gene sequences, and an initial population was generated using the Latin hypercube sampling method. The initial population is iteratively optimized through tournament selection, simulated binary crossover, and directed polynomial mutation until it converges to the maximum number of iterations to obtain the optimal chromosome. The optimal chromosome is analyzed by time slice to obtain time series path points. The feasibility is verified by combining dynamic model and reinforcement learning, and the optimal flight path is output.

9. The artificial intelligence-based drone swarm control method as described in claim 1, characterized in that: The specific steps for generating collaborative control commands through edge computing and distributing them to the drone swarm in real time are as follows: A nested JSON structure is constructed, and Python is used to encapsulate and serialize the optimal flight path and cooperative control parameters into JSON format instructions. The AES-128 algorithm is used to generate a byte key to encrypt the JSON format instructions in CBC mode. By combining 5G network slicing with TSN proxy nodes to dynamically allocate transmission time slots, encrypted JSON format commands are distributed to the drone cluster in real time.

10. An artificial intelligence-based unmanned aerial vehicle (UAV) swarm control system, based on the artificial intelligence-based UAV swarm control method according to any one of claims 1 to 9, characterized in that: It includes a data acquisition module, a path planning module, a control parameter module, and a command distribution module; The data acquisition module is used to acquire global collaborative situational awareness data through sensors, and to preprocess the data to obtain preprocessed global collaborative situational awareness data. The path planning module is used to calculate the path cost index of each UAV based on the preprocessed global collaborative situational data, using the fuzzy C-means clustering algorithm to generate a preliminary flight path, and to dynamically adjust the preliminary flight path based on real-time environmental feedback using a reinforcement learning algorithm. The control parameter module is used for the drone swarm to share the adjusted preliminary flight path and global environmental information through the 5G network, analyze the speed, heading angle, altitude, acceleration gradient and altitude change rate of each drone, and generate cooperative control parameters. The command distribution module is used to optimize the cooperative control parameters through a genetic algorithm to obtain the optimal flight path, and to generate cooperative control commands through edge computing methods and distribute them to the UAV cluster in real time.

Images