High subsonic UAV cluster formation flight control system and method

Abstract

The present invention discloses a high-subsonic unmanned aerial vehicle (UAV) cluster formation flight control system and method thereof, which obtains a text description of task requirements; obtains real-time environmental information collected by a multi-sensor group; extracts task requirement semantic features of the text description of the task requirements; extracts real-time environmental semantic features of the real-time environmental information; constructs collaborative features between the task requirement semantic features and the real-time environmental semantic features; and determines a recommended formation based on the collaborative features. In this way, a suitable flight formation can be automatically recommended based on task requirements and real-time environmental information to dynamically adjust the flight formation and optimize the flight control strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent flight control, and in particular, to a high-subsonic unmanned aerial vehicle (UAV) cluster formation flight control system and method thereof. Background Art

[0002] A high-subsonic UAV cluster is a flight formation composed of multiple high-subsonic UAVs, which can perform various tasks, such as reconnaissance, strike, interference, etc., under the conditions of high speed, high altitude, and high maneuverability.

[0003] Generating a reasonable formation shape is of great necessity for a high-subsonic UAV cluster. A reasonable formation shape can enable the UAV cluster to obtain an optimal spatial distribution to achieve maximized detection range, strike coverage, interference resistance, and other indicators. In addition, a reasonable formation shape can also be timely transformed and adjusted according to the threats and interferences of the enemy to increase its own stealth and survivability. However, the existing formation shape generation algorithms lack consideration of uncertain factors, such as complex environmental information, resulting in the generated formation shape may not adapt to the actual situation, or even fail or malfunction.

[0004] Therefore, an optimized high-subsonic UAV cluster formation flight control scheme is expected. Summary of the Invention

[0005] An embodiment of the present invention provides a high-subsonic UAV cluster formation flight control system and method thereof, which obtain a text description of task requirements; obtain real-time environmental information collected by a multi-sensor group; extract task requirement semantic features of the text description of the task requirements; extract real-time environmental semantic features of the real-time environmental information; construct collaborative features between the task requirement semantic features and the real-time environmental semantic features; and determine a recommended formation based on the collaborative features. In this way, a suitable flight formation can be automatically recommended through task requirements and real-time environmental information to dynamically adjust the flight formation and optimize the flight control strategy.

[0006] An embodiment of the present invention also provides a high-subsonic UAV cluster formation flight control method, which includes:

[0007] Obtain a text description of task requirements;

[0008] Obtain real-time environmental information collected by a multi-sensor group;

[0009] Extract task requirement semantic features of the text description of the task requirements;

[0010] Extract real-time environmental semantic features of the real-time environmental information;

[0011] Construct a collaborative feature between the semantic features of the task requirements and the semantic features of the real-time environment;

[0012] Based on the collaborative feature, determine the recommended formation.

[0013] An embodiment of the present invention also provides a high-subsonic UAV swarm formation flight control system, which includes:

[0014] A text description acquisition module, configured to acquire a text description of task requirements;

[0015] A real-time environment information acquisition module, configured to acquire real-time environment information collected by a multi-sensor group;

[0016] A task requirement semantic feature extraction module, configured to extract task requirement semantic features of the text description of the task requirements;

[0017] A real-time environment semantic feature extraction module, configured to extract real-time environment semantic features of the real-time environment information;

[0018] A collaborative feature construction module, configured to construct a collaborative feature between the task requirement semantic features and the real-time environment semantic features;

[0019] A recommended formation determination module, configured to determine a recommended formation based on the collaborative feature. Description of the Drawings

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. In the drawings:

[0021] Figure 1 It is a flowchart of a high-subsonic UAV swarm formation flight control method provided in an embodiment of the present invention.

[0022] Figure 2 It is a schematic diagram of the system architecture of a high-subsonic UAV swarm formation flight control method provided in an embodiment of the present invention.

[0023] Figure 3 It is a block diagram of a high-subsonic UAV swarm formation flight control system provided in an embodiment of the present invention.

[0024] Figure 4 It is an application scenario diagram of a high-subsonic UAV swarm formation flight control method provided in an embodiment of the present invention. Detailed Embodiments

[0025] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer and more understandable, the following further describes the embodiments of the present invention in detail with reference to the accompanying drawings. Herein, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but not to limit the present invention.

[0026] As shown in the present invention and the claims, unless the context clearly indicates otherwise, words such as "a", "an", "one" and / or "the" are not specifically singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0027] Flowcharts are used in the present invention to illustrate the operations performed by the systems according to the embodiments of the present invention. It should be understood that the operations before or below are not necessarily executed precisely in order. On the contrary, various steps can be executed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more steps can be removed from these processes.

[0028] The following will detail various exemplary embodiments, features and aspects of the present disclosure with reference to the accompanying drawings. The same reference numerals in the drawings denote elements with the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings do not have to be drawn to scale unless otherwise specified.

[0029] The special term "exemplary" herein means "serving as an example, embodiment or illustration". Any embodiment described as "exemplary" here does not have to be construed as superior or better than other embodiments.

[0030] A high-subsonic unmanned aerial vehicle (UAV) swarm is a flight formation composed of multiple high-subsonic UAVs, featuring high speed, high altitude, and high maneuverability. Such a UAV swarm can perform various tasks, including reconnaissance, strike, and interference. The high-subsonic UAV swarm has excellent high-speed performance and can fly at supersonic or near-supersonic speeds, enabling it to quickly reach the target area and complete tasks in a short time. The high-subsonic UAV swarm can fly at high altitudes, usually performing tasks in the higher regions of the atmosphere, which allows it to avoid detection and attack by ground air defense systems, enhancing its survivability and mission execution efficiency. The high-subsonic UAV swarm has excellent maneuverability and can rapidly change its speed, altitude, and angle, enabling it to flexibly respond to threats and interference from the enemy and having a high survivability. The high-subsonic UAV swarm can perform multiple tasks, including reconnaissance, target search and identification, striking enemy targets, and interfering with enemy communications. It can carry various sensors and weaponry and has strong information collection and combat capabilities. The UAVs in the high-subsonic UAV swarm can achieve cooperative operations through communication and coordinated control, form formations, conduct formation flight and task division to achieve higher combat effectiveness and mission completion rate.

[0031] The high-subsonic UAV swarm plays an important role in modern warfare. It can perform tasks in complex combat environments, providing capabilities such as battlefield situation awareness, target strike, and interference, and offering important support for command and decision-making. At the same time, the high-subsonic UAV swarm also faces technical challenges and the complexity of flight control, and it is necessary to comprehensively consider mission requirements, environmental information, and the optimization of the flight formation to improve mission execution efficiency and survivability.

[0032] For the high-subsonic UAV swarm, generating a reasonable formation shape is of great necessity. A reasonable formation shape can help the UAV swarm obtain an optimal spatial distribution to achieve indicators such as maximizing the detection range, strike coverage, and interference resistance. At the same time, a reasonable formation shape can also be promptly changed and adjusted according to the threats and interference from the enemy to increase its stealth and survivability.

[0033] A reasonable formation can evenly distribute the UAV swarm in space, maximizing the coverage of the target area. By optimizing the formation, conflicts and overlaps between UAVs can be avoided, and the detection and strike efficiency can be improved. A reasonable formation can maximize the sensor coverage of the UAV swarm. By arranging positions reasonably and optimizing the sensor angles, the reconnaissance and identification capabilities of targets can be improved, and the accuracy and timeliness of intelligence acquisition can be increased. A reasonable formation can optimize the strike range and density of the UAV swarm. By arranging positions reasonably and dividing tasks, all-round strikes on targets can be achieved, and the strike effect and combat capabilities can be improved. A reasonable formation can increase the resistance of the UAV swarm to enemy interference. By arranging positions reasonably and optimizing the formation structure, the impact of enemy interference on the UAV swarm can be reduced, and the reliability of communication and command can be maintained. A reasonable formation should have the ability to dynamically adjust and transform. According to changes in enemy threats, interference, and mission requirements, the UAV swarm can timely adjust the formation to increase stealth and survivability and maintain combat superiority.

[0034] However, existing formation generation algorithms have some limitations in considering uncertain factors. Complex environmental information includes factors such as enemy threats, interference, terrain, and meteorology, which may cause the generated formation to not adapt to the actual situation, and even result in failure or invalidation.

[0035] Therefore, in this application, an optimized high-subsonic UAV swarm formation flight control scheme is provided.

[0036] In an embodiment of the present invention, Figure 1 It is a flowchart of a high-subsonic UAV swarm formation flight control method provided in an embodiment of the present invention. Figure 2 It is a schematic diagram of the system architecture of a high-subsonic UAV swarm formation flight control method provided in an embodiment of the present invention. As Figure 1 and Figure 2 shown, according to the high-subsonic UAV swarm formation flight control method of the embodiment of the present invention, it includes: 110, obtaining a text description of mission requirements; 120, obtaining real-time environmental information collected by a multi-sensor group; 130, extracting the mission requirement semantic features of the text description of the mission requirements; 140, extracting the real-time environmental semantic features of the real-time environmental information; 150, constructing a collaborative feature between the mission requirement semantic features and the real-time environmental semantic features; 160, determining a recommended formation based on the collaborative feature.

[0037] In step 110, obtain the text description of the task requirements, ensuring accurate acquisition of the text description of the task requirements, including task objectives, execution conditions, constraint requirements, etc. The text description should be specific and clear for subsequent processing and analysis. Obtaining the text description of the task requirements can provide key information for subsequent formation shape generation. The task requirements description contains expectations and requirements for the formation shape, which can guide the processing and decision-making of subsequent steps.

[0038] In step 120, obtain the real-time environment information collected by the multi-sensor group. Select a suitable sensor combination to ensure the acquisition of diverse environment information and the accuracy and reliability of the sensors to obtain accurate real-time environment information. The real-time environment information provides the key data required for formation shape generation. Through the environment information collected by the multi-sensor group, factors such as enemy threats, interference, terrain, and meteorology can be understood, providing a basis for generating a reasonable formation shape.

[0039] In step 130, extract the semantic features of the task requirements from the text description of the task requirements. Use natural language processing techniques such as text parsing and semantic analysis to extract the key semantic features of the task requirements, ensuring accurate capture of the important elements and constraint conditions of the task requirements. The extraction of the semantic features of the task requirements can transform the task requirements into a form that can be understood by a computer, providing a basis for subsequent processing and analysis. The semantic features of the task requirements can help understand the task objectives and constraint requirements, guiding the decision-making process of formation shape generation.

[0040] In step 140, extract the semantic features of the real-time environment from the real-time environment information. Use data processing and analysis techniques to extract the key semantic features of the real-time environment information, including processing steps such as parsing of sensor data, feature extraction, and data fusion. The extraction of the semantic features of the real-time environment can transform the real-time environment information into a form that can be understood by a computer. By extracting the semantic features of the real-time environment, factors such as threats, interference, and terrain in the environment can be accurately captured, providing a basis for formation shape generation.

[0041] In step 150, construct the collaborative features between the semantic features of the task requirements and the semantic features of the real-time environment. Conduct collaborative analysis and modeling on the semantic features of the task requirements and the semantic features of the real-time environment to capture the association and influence between them. Machine learning or knowledge representation methods can be used for feature fusion and model construction. The construction of the collaborative features can comprehensively consider the task requirements and the real-time environment information, providing a comprehensive decision-making basis for generating the recommended formation shape. The collaborative features can reflect the interaction between the task requirements and the environment information, helping to generate a formation shape that is more adaptable to the actual situation.

[0042] In step 160, based on the collaborative features, determine the recommended formation. When using collaborative features for formation generation decision-making, optimization algorithms, planning methods, deep learning models, etc. can be used for decision-making and optimization. Considering the real-time requirements, it is necessary to ensure the efficiency and real-time nature of the decision-making process. Determining the recommended formation based on collaborative features can generate the optimal formation pattern according to the task requirements and real-time environmental information. The recommended formation can comprehensively consider task requirements, environmental information, and formation characteristics to maximize the execution efficiency and survivability.

[0043] To address the above technical problems, the technical concept of this application is to comprehensively utilize multi-modal data of task requirements and environmental information, and combine deep learning algorithms and natural language processing technologies to automatically analyze the complex environmental information where the UAV swarm is currently located. At the same time, using the task requirements as induced information to generate an appropriate formation pattern. That is, through task requirements and real-time environmental information, automatically recommend a suitable flight formation to dynamically adjust the flight formation and optimize the flight control strategy.

[0044] Based on this, in the technical solution of this application, first obtain the text description of the task requirements; and obtain the real-time environmental information collected by the multi-sensor group. Then, extract the task requirement semantic features of the text description of the task requirements. It should be understood that the text description of the task requirements may contain semantic information such as the goal of the task, the constraints of the task, and the evaluation indicators of the task. For example, the goal of the task is to perform various tasks such as reconnaissance, strike, interference, etc.; the constraints of the task are conditions such as high speed, high altitude, and high maneuverability; the evaluation indicators of the task are detection range, strike coverage, interference resistance, etc. These information are of great reference significance for the generation of the formation.

[0045] In a specific example of this application, the implementation method of extracting the task requirement semantic features of the text description of the task requirements is to pass the text description of the task requirements through a semantic encoder including a word embedding layer to obtain a task requirement semantic understanding feature vector. Among them, the task requirement semantic understanding feature vector is used as the task requirement semantic feature.

[0046] In a specific embodiment of this application, passing the text description of the task requirements through a semantic encoder including a word embedding layer to obtain a task requirement semantic understanding feature vector includes: performing word segmentation on the text description of the task requirements to convert the text description of the task requirements into a word sequence composed of multiple words; using the embedding layer of the semantic encoder including the word embedding layer to map each word in the word sequence to a word vector to obtain a sequence of word vectors; and using the transformer of the semantic encoder including the word embedding layer to perform global context semantic encoding on the sequence of word vectors to obtain the task requirement semantic understanding feature vector.

[0047] Among them, using the transformer of the semantic encoder including the word embedding layer to perform global context semantic encoding on the sequence of the word vectors to obtain the task requirement semantic understanding feature vector includes: arranging the sequence of the word vectors in one dimension to obtain a word global feature vector; calculating the product between the word global feature vector and the transposed vectors of each word vector in the sequence of the word vectors to obtain a plurality of self-attention correlation matrices; respectively performing normalization processing on each self-attention correlation matrix in the plurality of self-attention correlation matrices to obtain a plurality of normalized self-attention correlation matrices; passing each normalized self-attention correlation matrix in the plurality of normalized self-attention correlation matrices through the Softmax classification function to obtain a plurality of probability values; respectively using each probability value in the plurality of probability values as a weight to weight each word vector in the sequence of the word vectors to obtain the task requirement semantic understanding feature vector.

[0048] Among them, word segmentation processing transforms the text description of the task requirement into a word sequence, providing a finer-grained semantic unit and making semantic encoding more accurate. Using the word embedding layer to map each word in the word sequence to a word vector captures the semantic relationship between words and enriches the representation ability of the task requirement. Global context semantic encoding can utilize the information of the entire word vector sequence to perform a deeper semantic understanding of the task requirement and extract the task requirement semantic understanding feature vector. The task requirement semantic understanding feature vector can provide a more comprehensive and accurate representation of the task requirement, which helps the subsequent formation shape generation and optimization process and improves the execution efficiency and survivability of the formation. At the same time, extract the real-time environment semantic features of the real-time environment information. It should be understood that the influence of the environment information on the formation shape generation is significant. The reason is that the environment information to a certain extent determines the flight performance, detection ability, strike effect and interference resistance of the UAV swarm. For example, terrain information affects the flight altitude, speed and heading of the UAV swarm, as well as the radar detection range and stealth effect of the enemy; meteorological information affects the flight stability, communication quality and navigation accuracy of the UAV swarm; electromagnetic information affects the communication reliability, interference intensity and anti-interference ability of the UAV swarm. Thus, the generation of the formation shape of the UAV swarm needs to consider the influence of the environment information in order to select the optimal formation shape according to different environmental conditions.

[0049] In a specific example of the present application, the encoding process of extracting the real-time environment semantic features of the real-time environment information includes: first encoding the real-time environment information to obtain a real-time environment encoding vector; then passing the real-time environment encoding vector through a semantic encoder based on a Bi-LSTM model to obtain a real-time environment semantic understanding feature vector. Among them, in an embodiment of the present application, the implementation process of encoding the real-time environment information to obtain the real-time environment encoding vector can be: after encoding each data item in the real-time environment information (for example, one-hot encoding), the encoding vectors of each data item are spliced ​​to obtain the real-time environment encoding vector.

[0050] Perform necessary preprocessing on real-time environmental information, such as data cleaning and normalization, to ensure data consistency and availability. Encoding real-time environmental information can convert it into a vector representation that can be processed by a computer, which is convenient for subsequent calculations and processing. One-hot encoding is a method of converting discrete features into binary vector representations. For each discrete feature, it can be converted into a vector in which only one element is 1 and the rest are 0, indicating the value of the feature. Using one-hot encoding to convert discrete features into vector representations retains the semantic information of the features and can process multiple discrete features. The Bi-LSTM model is a bidirectional long short-term memory network that can capture contextual information in sequence data. The semantic encoder based on the Bi-LSTM model can capture the contextual relationship and semantic information in real-time environmental information and extract the semantic understanding feature vector of the real-time environment.

[0051] Then, the collaborative features between the semantic features of the task requirements and the semantic features of the real-time environment are constructed. That is, the correlation and complementarity of the task requirements and the real-time environment are comprehensively considered to make the generation of the formation morphology more reliable. In a specific example of the present application, the collaborative features between the semantic features of the task requirements and the semantic features of the real-time environment are constructed by associating and encoding the task requirement semantic understanding feature vector and the real-time environment semantic understanding feature vector into a task requirement-real-time environment semantic collaborative input matrix, and then passing through a collaborative feature extractor based on a convolutional neural network model to obtain a task requirement-real-time environment semantic collaborative feature map. Here, the powerful expressive power of the deep learning model is used to explore the potential connections and laws between the task requirements and the real-time environment to characterize the collaborative features between the two.

[0052] In a specific embodiment of the present application, constructing the collaborative feature between the task requirement semantic feature and the real-time environment semantic feature includes: associatively encoding the task requirement semantic understanding feature vector and the real-time environment semantic understanding feature vector into a task requirement-real-time environment semantic collaborative input matrix, and then obtaining a task requirement-real-time environment semantic collaborative feature map through a collaborative feature extractor based on a convolutional neural network model; and using the task requirement-real-time environment semantic collaborative feature map as the collaborative feature.

[0053] Among them, the collaborative feature extractor based on the convolutional neural network model includes: an input layer, a convolutional layer, a pooling layer, an activation layer, and an output layer.

[0054] By associatively encoding the task requirement semantic understanding feature vector and the real-time environment semantic understanding feature vector, the task requirement-real-time environment semantic collaborative input matrix can provide more comprehensive and integrated information. Such integrated information can better reflect the association between the task requirement and the real-time environment, providing more accurate input for formation flight control. The collaborative feature extractor based on the convolutional neural network model can extract semantic association features from the task requirement-real-time environment semantic collaborative input matrix. The CNN model performs excellently in image processing and feature extraction, can learn the association between the task requirement and the real-time environment, and extract meaningful features from it. These features can capture information at different scales and levels, providing a more representative feature representation for formation flight control. The task requirement-real-time environment semantic collaborative feature map can provide more accurate and comprehensive information, helping the formation flight control system make more precise decisions. By utilizing the semantic association between the task requirement and the real-time environment, the feature map can provide better decision support in aspects such as formation shape selection, path planning, obstacle avoidance, and cooperative control, thereby improving the execution efficiency and survivability of the formation. The task requirement-real-time environment semantic collaborative feature map can be adaptively adjusted according to different task requirements and real-time environment conditions. Through the learning ability of the convolutional neural network model, the feature map can adapt to different task requirements and real-time environment changes, providing a robust feature representation. This adaptability and robustness enable the formation flight control system to have better adaptability and reliability in different scenarios and tasks.

[0055] Further, the task requirement-real-time environment semantic collaborative feature map is passed through a feature autocorrelation association reinforcement module to obtain an autocorrelation reinforcement task requirement-real-time environment semantic collaborative feature map. Among them, the feature autocorrelation association reinforcement module aggregates the complete information of the model for formation generation by constructing the similarity association between each element. That is to say, the feature autocorrelation association reinforcement module induces the network to pay more attention to the important regions in the feature distribution, making the feature expression of the autocorrelation reinforcement task requirement-real-time environment semantic collaborative feature map closer to the ultimate goal of generating a suitable formation.

[0056] In a specific embodiment of the present application, based on the collaborative feature, determining the recommended formation includes: passing the task requirement-real-time environment semantic collaborative feature map through a feature autocorrelation association reinforcement module to obtain an autocorrelation reinforcement task requirement-real-time environment semantic collaborative feature map; calculating the global mean of each feature matrix of the task requirement-real-time environment semantic collaborative feature map to obtain a task requirement-real-time environment semantic collaborative feature vector; calculating the global mean of each feature matrix of the autocorrelation reinforcement task requirement-real-time environment semantic collaborative feature map to obtain an autocorrelation reinforcement task requirement-real-time environment semantic collaborative feature vector; performing dot product correction on the autocorrelation reinforcement task requirement-real-time environment semantic collaborative feature vector with the task requirement-real-time environment semantic collaborative feature vector to obtain a corrected autocorrelation reinforcement task requirement-real-time environment semantic collaborative feature vector; optimizing the corrected autocorrelation reinforcement task requirement-real-time environment semantic collaborative feature vector to obtain an optimized corrected autocorrelation reinforcement task requirement-real-time environment semantic collaborative feature vector; weighting the autocorrelation reinforcement task requirement-real-time environment semantic collaborative feature map along the channels with the optimized corrected autocorrelation reinforcement task requirement-real-time environment semantic collaborative feature vector to obtain an optimized autocorrelation reinforcement task requirement-real-time environment semantic collaborative feature map; and passing the optimized autocorrelation reinforcement task requirement-real-time environment semantic collaborative feature map through a classifier to obtain a classification result, and the classification result is used to represent the recommended formation label.

[0057] Among them, obtaining the self - correlation enhanced task requirement - real - time environment semantic collaboration feature map by passing the task requirement - real - time environment semantic collaboration feature map through the feature self - correlation association enhancement module includes: passing the task requirement - real - time environment semantic collaboration feature map through a first convolutional layer to obtain a dimensionality - reduced feature map; passing the dimensionality - reduced feature map through a second convolutional layer to obtain an efficient association construction map; calculating the relationship matrix of the efficient association construction map using a cosine similarity operation; normalizing the relationship matrix using a Softmax function to obtain a normalized relationship matrix; using an element - wise multiplication operation to complete the modeling of the relationship between any two eigenvalues in the dimensionality - reduced feature map by the normalized relationship matrix to obtain an association feature map; performing a de - convolution operation on the association feature map to obtain a de - convolved association feature map; adding the de - convolved association feature map and the dimensionality - reduced feature map element - wise to obtain a preliminary result feature map; and after performing channel expansion on the preliminary result feature map to obtain an expanded preliminary result feature map, performing a residual connection between the expanded preliminary result feature map and the task requirement - real - time environment semantic collaboration feature map to obtain the self - correlation enhanced task requirement - real - time environment semantic collaboration feature map.

[0058] In the technical solution of this application, when obtaining the self - correlation enhanced task requirement - real - time environment semantic collaboration feature map by passing the task requirement - real - time environment semantic collaboration feature map through the feature self - correlation association enhancement module, considering that each feature matrix of the task requirement - real - time environment semantic collaboration feature map represents the high - order association features of the text description of the task requirement and the full - semantic - domain association of the real - time environment information, and the channel distribution of the convolutional neural network model is followed among its respective feature matrices. Through the feature self - correlation association enhancement module, taking the channel vector of the task requirement - real - time environment semantic collaboration feature map as a unit, based on the high - order association feature distribution of the full - semantic - domain of the feature matrix, feature self - correlation enhancement in the feature matrix distribution dimension can be performed. While this improves the overall expression consistency of the self - correlation enhanced task requirement - real - time environment semantic collaboration feature map, it will also cause the channel distribution expression of the self - correlation enhanced task requirement - real - time environment semantic collaboration feature map to deviate from the channel distribution expression of the task requirement - real - time environment semantic collaboration feature map, affecting its target distribution expression consistency with respect to the classification result, thereby affecting the accuracy of the classification result obtained by the self - correlation enhanced task requirement - real - time environment semantic collaboration feature map through the classifier.

[0059] Therefore, preferably, first calculate the global mean of each feature matrix of the task requirement - real - time environment semantic collaboration feature map to obtain a task requirement - real - time environment semantic collaboration feature vector, denoted as V for example. 1, then calculate the global mean of each feature matrix of the self - correlation enhanced task requirement - real - time environment semantic collaboration feature map to obtain the self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector, denoted as V for example 2 , then use the task requirement - real - time environment semantic collaboration feature vector V 1 to perform dot - product correction on the self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector V 2 to obtain the corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector, denoted as V' for example 2 , and then optimize the corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector V' 2 . Specifically, optimizing the corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector to obtain the optimized corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector includes the following steps: Add the eigenvalue at each position of the corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector V' 2 to the square root of the length of the corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector V' 2 and the reciprocal of the square root of the two - norm of the corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector V' 2 . Then, take the sum value as the exponent of the exponential function with the natural constant as the base. Then, multiply the eigenvalue at each position of the corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector V' 2 by the one - norm of the corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector V' 2 and a weighted hyperparameter, and add the product value to the above - mentioned exponential value to obtain the optimized corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector.

[0060] In this way, by using the structured - norm representation of the corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector as the local canonical coordinate for each eigenvalue, to determine the rotation offset of the overall vector distribution representation of the corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector relative to the eigenvalue for the offset prediction direction of each eigenvalue as the center, and using the bounding box of the vector eigenvalue distribution of the corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector for eigenvalue constraint, to enhance the consistency constraint between the eigenvalues of the corrected self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector, thereby enhancing the task requirement - real - time environment semantic collaboration feature vector V 1 and the self - correlation enhanced task requirement - real - time environment semantic collaboration feature vector V 2Consistency. In this way, with the optimized corrected autocorrelation to strengthen the task requirements - real-time environment semantic collaboration feature vector V 2 'Weight the self-correlation reinforcement task requirements - real-time environment semantic collaboration feature map along the channels, and the accuracy of the classification result obtained by the self-correlation reinforcement task requirements - real-time environment semantic collaboration feature map through the classifier can be improved.

[0061] Subsequently, input the optimized self-correlation reinforcement task requirements - real-time environment semantic collaboration feature map into the classifier to obtain a classification result, and the classification result is used to represent the recommended formation label. By inputting the optimized self-correlation reinforcement task requirements - real-time environment semantic collaboration feature map into the classifier, the corresponding classification result can be obtained. These classification results can be used to represent the recommended formation label. By combining the task requirements with the real-time environment information, the classifier can classify different task requirements and real-time environment conditions according to the semantic association features in the feature map and recommend suitable formations.

[0062] By using the classifier to classify the optimized self-correlation reinforcement task requirements - real-time environment semantic collaboration feature map, personalized formation recommendation can be achieved. Different task requirements and real-time environment conditions may correspond to different optimal formation selections. Through the classifier, according to the information in the feature map, the most suitable formation label can be recommended for each task requirement and real-time environment condition, providing a personalized formation flight control scheme. By inputting the task requirements - real-time environment semantic collaboration feature map into the classifier, the evaluation and recommendation degree for different formations can be obtained. These results can help decision-makers better understand the applicability, advantages and disadvantages of each formation, so as to make wise decisions in the process of generating flight control strategies. By recommending appropriate formation labels according to the classification results, the performance of the formation can be optimized. Different formations may correspond to different balances of flight performance and task requirements. Through the recommendation of the classifier, the formation that best meets the task requirements and real-time environment conditions can be selected, thereby improving the execution efficiency and survivability of the formation.

[0063] In summary, the high-subsonic UAV swarm formation flight control method based on the embodiments of the present invention is clarified. It comprehensively utilizes multi-modal data of task requirements and environmental information, combines deep learning algorithms and natural language processing technologies to automatically analyze the complex environmental information where the UAV swarm is currently located, and at the same time uses the task requirements as the induced information to generate appropriate formation shapes.

[0064] In an embodiment of the present invention, Figure 3 is a block diagram of a high-subsonic UAV swarm formation flight control system provided in the embodiments of the present invention. As Figure 3As shown, the high-subsonic UAV cluster formation flight control system 200 according to an embodiment of the present invention includes: a text description acquisition module 210 for acquiring a text description of a mission requirement; a real-time environment information acquisition module 220 for acquiring real-time environment information collected by a multi-sensor group; a mission requirement semantic feature extraction module 230 for extracting mission requirement semantic features of the text description of the mission requirement; a real-time environment semantic feature extraction module 240 for extracting real-time environment semantic features of the real-time environment information; a collaborative feature construction module 250 for constructing collaborative features between the mission requirement semantic features and the real-time environment semantic features; and a recommended formation determination module 260 for determining a recommended formation based on the collaborative features.

[0065] Here, those skilled in the art can understand that the specific functions and operations of each unit and module in the above high-subsonic UAV cluster formation flight control system have been introduced in detail in the description of the high-subsonic UAV cluster formation flight control method above with reference to Figures 1 to 2 and thus, the repeated description thereof will be omitted.

[0066] As described above, the high-subsonic UAV cluster formation flight control system 200 according to an embodiment of the present invention can be implemented in various terminal devices, such as a server for high-subsonic UAV cluster formation flight control. In one example, the high-subsonic UAV cluster formation flight control system 200 according to an embodiment of the present invention can be integrated into a terminal device as a software module and / or a hardware module. For example, the high-subsonic UAV cluster formation flight control system 200 can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the high-subsonic UAV cluster formation flight control system 200 can also be one of many hardware modules of the terminal device.

[0067] Alternatively, in another example, the high-subsonic UAV cluster formation flight control system 200 and the terminal device can also be separate devices, and the high-subsonic UAV cluster formation flight control system 200 can be connected to the terminal device through a wired and / or wireless network and transmit interaction information in accordance with a predefined data format.

[0068] Figure 4 This is an application scenario diagram of a high-subsonic UAV cluster formation flight control method provided in an embodiment of the present invention. As Figure 4 shown, in this application scenario, first, a text description of a mission requirement is acquired (for example, C1 as illustrated in Figure 4 ); and real-time environment information collected by a multi-sensor group is acquired (for example, as Figure 4as schematically shown in C2); then, input the text description of the obtained mission requirements and the real-time environment information into a server (e.g., as schematically shown in S) deployed with a high-subsonic UAV cluster formation flight control algorithm, where the server can process the text description of the mission requirements and the real-time environment information based on the high-subsonic UAV cluster formation flight control algorithm to determine a recommended formation. Figure 4 as schematically shown in S), wherein the server can process the text description of the mission requirements and the real-time environment information based on the high-subsonic UAV cluster formation flight control algorithm to determine a recommended formation.

[0069] In the specific embodiments described above, the object, technical solution and beneficial effects of the present invention have been further described in detail. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A high subsonic UAV swarm formation flight control method, characterized in that: include: Get a text description of the task requirements; Obtain real-time environmental information collected by a multi-sensor group; Extracting semantic features of task requirements from the text description of the task requirements; Extracting real-time environment semantic features of the real-time environment information; Constructing a collaborative feature between the task requirement semantic feature and the real-time environment semantic feature; Determining a recommended formation based on the collaborative features; Wherein, determining the recommended formation based on the collaborative feature includes: The task requirement-real-time environment semantic collaborative feature map is passed through a feature autocorrelation association enhancement module to obtain an autocorrelation enhanced task requirement-real-time environment semantic collaborative feature map; Calculating the global mean of each feature matrix of the task requirement-real-time environment semantics collaborative feature map to obtain a task requirement-real-time environment semantics collaborative feature vector; Calculating the global mean of each feature matrix of the self-correlation reinforcement task requirement-real-time environment semantics collaborative feature map to obtain the self-correlation reinforcement task requirement-real-time environment semantics collaborative feature vector; Performing point multiplication correction on the autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature vector with the task requirement-real-time environment semantics collaborative feature vector to obtain a corrected autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature vector; Optimizing the corrected autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature vector to obtain an optimized corrected autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature vector; The autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature map is weighted along the channel by the optimized corrected autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature vector to obtain an optimized autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature map; Passing the optimized autocorrelation enhanced task requirement-real-time environment semantic collaborative feature map through a classifier to obtain a classification result, wherein the classification result is used to represent a recommended formation label; The task requirement-real-time environment semantics collaborative feature graph is subjected to a feature autocorrelation association enhancement module to obtain an autocorrelation enhanced task requirement-real-time environment semantics collaborative feature graph, including: Pass the task requirement-real-time environment semantic collaborative feature map through the first convolution layer to obtain a reduced-dimensional feature map; Passing the dimension-reduced feature map through a second convolutional layer to obtain an efficient association structure map; Calculating the relationship matrix of the efficient association construction graph using a cosine similarity operation; Using a Softmax function to normalize the relationship matrix to obtain a normalized relationship matrix; The normalized relationship matrix is ​​used to model the relationship between any two eigenvalues ​​in the reduced-dimensional feature map by using an element-by-element multiplication operation to obtain a correlation feature map; Performing a deconvolution operation on the associated feature map to obtain a deconvolved associated feature map; Adding the deconvolution associated feature map and the dimension reduction feature map element by element to obtain a preliminary result feature map; and After channel expansion is performed on the preliminary result feature map to obtain an expanded preliminary result feature map, the expanded preliminary result feature map and the task requirement-real-time environment semantic collaborative feature map are residually connected to obtain the autocorrelation enhanced task requirement-real-time environment semantic collaborative feature map.

2. The high subsonic UAV swarm formation flight control method according to claim 1, characterized in that: Extracting semantic features of the task requirement from the text description of the task requirement includes: Passing the text description of the task requirement through a semantic encoder including a word embedding layer to obtain a task requirement semantic understanding feature vector; and The task requirement semantic understanding feature vector is used as the task requirement semantic feature.

3. The high subsonic UAV swarm formation flight control method according to claim 2 is characterized in that: The text description of the task requirement is passed through a semantic encoder including a word embedding layer to obtain a task requirement semantic understanding feature vector, including: Performing word segmentation processing on the text description of the task requirement to convert the text description of the task requirement into a word sequence consisting of multiple words; Mapping each word in the word sequence to a word vector using the embedding layer of the semantic encoder including the word embedding layer to obtain a sequence of word vectors; and The converter of the semantic encoder including the word embedding layer is used to perform global context semantic encoding on the sequence of word vectors to obtain the semantic understanding feature vector required by the task.

4. The high subsonic UAV swarm formation flight control method according to claim 3 is characterized in that: Using the converter of the semantic encoder including the word embedding layer to perform global context semantic encoding on the sequence of word vectors to obtain the task requirement semantic understanding feature vector, including: Arranging the sequence of word vectors in one dimension to obtain a word global feature vector; Calculating the product of the word global feature vector and the transposed vector of each word vector in the sequence of word vectors to obtain multiple self-attention association matrices; Normalizing each of the plurality of self-attention association matrices to obtain a plurality of standardized self-attention association matrices; Passing each of the plurality of standardized self-attention association matrices through a Softmax classification function to obtain a plurality of probability values; Each word vector in the sequence of word vectors is weighted using each probability value in the multiple probability values ​​as a weight to obtain the semantic understanding feature vector required by the task.

5. The high subsonic UAV swarm formation flight control method according to claim 4 is characterized in that: Extracting the real-time environment semantic features of the real-time environment information includes: Encoding the real-time environment information to obtain a real-time environment encoding vector; Passing the real-time environment encoding vector through a semantic encoder based on a Bi-LSTM model to obtain a real-time environment semantic understanding feature vector; and The real-time environment semantic understanding feature vector is used as the real-time environment semantic feature.

6. The high subsonic UAV swarm formation flight control method according to claim 5, characterized in that: Constructing a collaborative feature between the task requirement semantic feature and the real-time environment semantic feature, including: The task requirement semantic understanding feature vector and the real-time environment semantic understanding feature vector are associated and encoded into a task requirement-real-time environment semantic collaborative input matrix, and then the matrix is ​​passed through a collaborative feature extractor based on a convolutional neural network model to obtain a task requirement-real-time environment semantic collaborative feature graph; and The task requirement-real-time environment semantic collaborative feature graph is used as the collaborative feature.

7. The high subsonic UAV swarm formation flight control method according to claim 6, characterized in that: The collaborative feature extractor based on the convolutional neural network model includes: an input layer, a convolution layer, a pooling layer, an activation layer and an output layer.

8. A high subsonic UAV cluster formation flight control system, characterized in that: include: A text description acquisition module is used to obtain the text description of the task requirements; A real-time environment information acquisition module is used to acquire real-time environment information collected by a multi-sensor group; A task requirement semantic feature extraction module, used to extract task requirement semantic features from the text description of the task requirement; A real-time environment semantic feature extraction module, used to extract the real-time environment semantic features of the real-time environment information; A collaborative feature construction module, used to construct collaborative features between the task requirement semantic features and the real-time environment semantic features; A recommended formation determination module, used to determine a recommended formation based on the collaborative features; Wherein, determining the recommended formation based on the collaborative feature includes: The task requirement-real-time environment semantic collaborative feature map is passed through a feature autocorrelation association enhancement module to obtain an autocorrelation enhanced task requirement-real-time environment semantic collaborative feature map; Calculating the global mean of each feature matrix of the task requirement-real-time environment semantics collaborative feature map to obtain a task requirement-real-time environment semantics collaborative feature vector; Calculating the global mean of each feature matrix of the self-correlation reinforcement task requirement-real-time environment semantics collaborative feature map to obtain the self-correlation reinforcement task requirement-real-time environment semantics collaborative feature vector; Performing point multiplication correction on the autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature vector with the task requirement-real-time environment semantics collaborative feature vector to obtain a corrected autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature vector; Optimizing the corrected autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature vector to obtain an optimized corrected autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature vector; The autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature map is weighted along the channel by using the optimized corrected autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature vector to obtain an optimized autocorrelation reinforcement task requirement-real-time environment semantics collaborative feature map; Passing the optimized autocorrelation enhanced task requirement-real-time environment semantic collaborative feature map through a classifier to obtain a classification result, wherein the classification result is used to represent a recommended formation label; The task requirement-real-time environment semantics collaborative feature graph is subjected to a feature autocorrelation association enhancement module to obtain an autocorrelation enhanced task requirement-real-time environment semantics collaborative feature graph, including: Pass the task requirement-real-time environment semantic collaborative feature map through the first convolution layer to obtain a reduced-dimensional feature map; Passing the dimension-reduced feature map through a second convolutional layer to obtain an efficient association structure map; Calculating the relationship matrix of the efficient association construction graph using a cosine similarity operation; Using a Softmax function to normalize the relationship matrix to obtain a normalized relationship matrix; The normalized relationship matrix is ​​used to model the relationship between any two eigenvalues ​​in the reduced-dimensional feature map by using an element-by-element multiplication operation to obtain a correlation feature map; Performing a deconvolution operation on the associated feature map to obtain a deconvolved associated feature map; Adding the deconvolution associated feature map and the dimension reduction feature map element by element to obtain a preliminary result feature map; and After channel expansion is performed on the preliminary result feature map to obtain an expanded preliminary result feature map, the expanded preliminary result feature map and the task requirement-real-time environment semantic collaborative feature map are residually connected to obtain the autocorrelation enhanced task requirement-real-time environment semantic collaborative feature map.

Images