Multi-structure aircraft attitude cooperative control system and method based on bionic learning

Abstract

The invention discloses a multi-variable-structure aircraft attitude cooperative control system and method based on bionic learning, and belongs to the technical field of multi-variable-structure aircrafts, and the system comprises a time-delay interference decoupling module, a multi-variable-structure aircraft bionic variable-structure module, and a multi-variable-structure aircraft attitude cooperative module. The time-delay interference decoupling module is used for learning and separating a time-delay and interference coupling mechanism of the multi-configuration aircraft and realizing time-delay and interference compensation; the multi-variable-structure aircraft bionic variable-structure module realizes the configuration control of the wingspan of the multi-variable-structure aircraft through bionic learning; the multi-configuration aircraft attitude coordination module is used for carrying out integrated modeling and attitude coordination control on the multi-configuration aircraft; and the time-delay interference decoupling module, the multi-variable-structure aircraft bionic variable-structure module and the multi-variable-structure aircraft attitude cooperation module are all realized by adopting a balanced adversarial network. According to the invention, the multi-variable-structure aircraft can have bionic intelligent behavior capabilities such as strong independence, strong adaptability and strong survival, and the intelligence, safety and reliability of the multi-variable-structure aircraft can be improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of multi-configuration aircraft, and particularly relates to a multi-configuration aircraft attitude cooperative control system and method based on bionic learning. BACKGROUND

[0002] In recent years, with the rapid development of logistics distribution, environmental protection, emergency rescue, urban air traffic and military operations, etc., the demand for unmanned aircraft with multiple performances such as high-low altitude / high-low speed, long cruise / short take-off and landing, etc., and multi-task execution ability is increasingly urgent. The previous fixed-wing single unmanned aircraft and multi-unmanned aircraft are difficult to meet the above requirements, and therefore the multi-configuration aircraft has become the most potential breakthrough means. However, the process of wing transformation and attitude coordination faces challenges such as strong nonlinearity, strong uncertainty, time delay and disturbance. In addition, the time delay of the transformation actuator leads to the uncertainty of the center of mass of the multi-configuration aircraft, and further leads to the uncertainty of the aerodynamic moment and the moment of inertia, which belongs to a kind of model uncertainty, indicating that there is a coupling relationship between the time delay of the transformation actuator and the model uncertainty. The single multi-configuration aircraft is affected by the external environment, resulting in wing vibration during the transformation process. After communication transmission, the wing vibration is transmitted to other aircraft in communication with it, and is affected by time delay, resulting in wing vibration lag during the transformation process of the corresponding aircraft in communication with it. The existence of these "coupling" relationships increases the difficulty of the attitude cooperative control of the multi-configuration aircraft. The above factors make the wing transformation and attitude coordination control become a bottleneck problem restricting the research and application of the multi-configuration aircraft.

[0003] In the prior art, only the attitude cooperative control of multi-aircraft or the control problem of single multi-configuration aircraft is usually studied, and the control problem of multi-configuration aircraft is not considered at the same time. In addition, the multi-source time delay problem caused by the communication time delay and the actuator time delay during the transformation process, and the cross-coupling problem caused by the time delay and disturbance are not studied. Therefore, it is particularly important to develop a multi-configuration aircraft attitude cooperative control system and method based on bionic learning. SUMMARY

[0004] The present application is aimed at the above problems, makes up for the deficiencies of the prior art, and provides a multi-configuration aircraft attitude cooperative control system and method based on bionic learning. The present application can effectively improve the reliability, safety and environmental adaptability of the multi-configuration aircraft operation.

[0005] To achieve the above purpose, the present application adopts the following technical solutions.

[0006] The multi-configuration aircraft attitude cooperative control system based on bionic learning provided by the present application comprises a time delay and disturbance decoupling module, a multi-configuration aircraft bionic transformation module and a multi-configuration aircraft attitude cooperative module. The time-delay interference decoupling module is used to learn and separate the time-delay and interference coupling mechanism of the multi-configuration aircraft, and to compensate for the time delay and interference. The biomimetic variable structure module of the variable structure aircraft is used to achieve configuration control of the wingspan of the variable structure aircraft through biomimetic learning. The variable-structure aircraft attitude coordination module is used for integrated modeling and attitude coordination control of the variable-structure aircraft. The time-delay interference decoupling module, the biomimetic variant module of the multi-variable aircraft, and the attitude coordination module of the multi-variable aircraft are all implemented using a balanced adversarial network.

[0007] As a preferred embodiment of the present invention, the time-delay interference decoupling module includes a coupling mechanism learning and separation unit and a time-delay interference compensation unit; The coupling mechanism learning and separation unit is used to learn the coupling characteristics between the deformation actuator delay and the model uncertainty, as well as the coupling characteristics between external interference and communication delay, through the balanced adversarial network, and output the interference amount and the delay amount. The time delay interference compensation unit is used to synchronously compensate for and suppress time delay and interference through a multi-source compensation and suppression framework.

[0008] As another preferred embodiment of the present invention, the multi-source compensation and suppression framework includes a time delay observer, a time delay compensator, and an interference observer; the time delay observer is used to learn the multi-source time delays online through an equalized adversarial network to identify flight actuator time delays, deformable actuator time delays, and communication time delays; the time delay compensator is used to compensate for the multi-source time delays by recursively calculating k steps through the equalized adversarial network; the interference observer is used to learn the interference online through the equalized adversarial network and output the interference value.

[0009] As another preferred embodiment of the present invention, the biomimetic variable structure module of the multi-variable aircraft learns the control decision mechanism of the goose when changing its wingspan through an equal adversarial network. The input is environmental information and goose configuration parameters, and the output is the deformation decision mechanism of the multi-variable aircraft, so as to achieve precise control of the wingspan of the multi-variable aircraft.

[0010] As another preferred embodiment of the present invention, the attitude coordination module of the multi-configuration aircraft adopts a hybrid driving modeling method that combines mechanism modeling and data-driven modeling to perform integrated modeling of the multi-configuration aircraft; wherein, the data-driven modeling describes the uncertainty of the model through an equilibrium adversarial network, the input of the data-driven model is the angular velocity state value, and the output is the inertia matrix uncertainty value.

[0011] As another preferred embodiment of the present invention, the mechanism modeling adopts the following state vector representation of the first... i The status of each aircraft: , ; in, V i For flight speed, gamma i The inclination angle of the flight path. omega i The vector of the body's angular velocity. lambda i These are configuration parameters; T i Powered by the engine, Let α be the rate of change of velocity. i For the angle of attack, D i For aerodynamic drag, m ( lambda ( ) represents the mass of the variable-configuration aircraft, and g represents the acceleration due to gravity. The rate of change of the track angle. L i For aerodynamic lift, J i ( lambda i ) is the variable configuration inertia matrix. tau aero，i For aerodynamic torque, tau morph，i Additional torque for the variable configuration, Δ J i Due to the uncertainty of the inertia matrix, d ω，i To control surface jitter interference, mu λ,i For deformation control input, k λ The natural decay coefficient, tau λ,i For the time delay of the deformation actuator, tau α,i For the attitude control actuator time delay, tau c,ij For from the first j Machine to the i Communication delay of the machine.

[0012] As another preferred embodiment of the present invention, the multi-configuration aircraft attitude coordination module establishes an equilibrium-adversarial composite controller through an equilibrium-adversarial network, and adopts a composite anti-interference control strategy to achieve attitude coordination control; the specific expression of the equilibrium-adversarial composite controller is as follows: , , , , ; wherein two constraint conditions are included, the first constraint condition is a deformation rate limit condition, and the second constraint condition is a time delay safety constraint condition; is an angular velocity of a first k step prediction, omega ref is a reference angular velocity, lambda opt is an optimal configuration parameter, is a maximum deformation rate, N p is a prediction time domain step length, Q is an angular velocity error weight matrix; is a configuration parameter of a first k step prediction, which is an output of the balanced adversarial network for establishing a first k step prediction model; w is a configuration error weight matrix, tau λ,i is a deformation actuator time delay, △ lambda safe is a safety deformation margin; is a control surface jitter disturbance observation, which is an output value of a disturbance observer.

[0013] As another preferred scheme of the present application, the balanced adversarial network comprises an input layer, an intermediate layer and an output layer; wherein the intermediate layer comprises: a feature extractor for extracting shared features and private features; a competitive adversarial, for selecting a winning neuron through a competitive mechanism; a strategy generator for generating a final strategy according to the shared features and the private features.

[0014] As another preferred scheme of the present application, the learning algorithm of the balanced adversarial network adopts any one of a gradient optimization method, a meta-learning algorithm, a recursive least squares algorithm or a wake-sleep algorithm.

[0015] In addition, the present application provides a multi-variable configuration aircraft attitude cooperative control method based on bionic learning, which is implemented by using the multi-variable configuration aircraft attitude cooperative control system based on bionic learning, and comprises the following steps: Step 1: learning and separating the coupling mechanism of time delay and disturbance by using the balanced adversarial network, and realizing compensation of the time delay and the disturbance; Step 2: The control decision mechanism of the wing span of the wild goose is learned through the balanced antagonistic network, and the bionic control of the variable wing span of the multi-variable aircraft is realized; Step 3: The multi-variable aircraft is integrally modeled by adopting a hybrid driving modeling method combining mechanism modeling and data-driven modeling; Step 4: The composite controller is constructed through the balanced antagonistic network, and the attitude cooperative control of the multi-variable aircraft is realized.

[0016] The present application has the following beneficial effects: The multi-variable aircraft attitude cooperative control system and method based on bionic learning provided by the present application can make the multi-variable aircraft have strong autonomous, strong adaptive and strong survival bionic intelligent behavior capabilities by combining the time delay disturbance decoupling module, the multi-variable aircraft bionic variable configuration module and the multi-variable aircraft attitude cooperative module, and has the beneficial effects of improving the intelligence, safety and reliability of the multi-variable aircraft. The present application can provide a set of performance robust modeling and control scheme for the multi-variable aircraft attitude cooperative control, improve the task adaptability, flight safety and energy economy of the multi-variable aircraft, and help the development of important planning fields such as ecological environment, modern agriculture and smart city. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The mapping relationship between the input layer, the intermediate layer and the output layer of the balanced antagonistic network of the multi-variable aircraft attitude cooperative control system and method based on bionic learning of the present application.

[0018] In the figure, 1 is the input layer, 2 is the intermediate layer, and 3 is the output layer; 201 is a feature extractor, 202 is a competitive antagonist, and 203 is a strategy generator. DETAILED DESCRIPTION

[0019] In order to make the technical problems, technical solutions and beneficial effects of the present application clearer, the present application will be further described in detail below in combination with the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0020] Inspired by flying creatures in nature, such as wild geese, which can fly in formation and change their wing span during flight, they can flexibly change parts such as wing span according to tasks, environment and their own state to obtain high-low altitude and high-low speed optimized flight capability. Based on this, the wild goose is abstracted into a multi-variable aircraft for research, thereby breaking through the application limitations of fixed-wing single unmanned aerial vehicles and multiple unmanned aerial vehicles. Therefore, by implementing the multi-variable aircraft attitude cooperative control system and method based on bionic learning involved in the present application, the cooperative control of the attitude of the multi-variable aircraft can be completed, and the reliability, safety and environmental adaptability of the multi-variable aircraft operation can be improved.

[0021] In combination Figure 1 As shown in the embodiment of the present application, the bionic learning-based multi-configuration aircraft attitude cooperative control system comprises a time delay interference decoupling module, a multi-configuration aircraft bionic variable configuration module, and a multi-configuration aircraft attitude cooperative module. The time delay interference decoupling module is used to learn and separate the time delay and interference coupling mechanism of the multi-configuration aircraft, and to realize the compensation of the time delay and interference. The multi-configuration aircraft bionic variable configuration module is used to realize the configuration control of the wing span of the multi-configuration aircraft through bionic learning. The multi-configuration aircraft attitude cooperative module is used to model and control the attitude of the multi-configuration aircraft in an integrated manner. The time delay interference decoupling module, the multi-configuration aircraft bionic variable configuration module, and the multi-configuration aircraft attitude cooperative module are all realized by using balanced adversarial networks. In this way, the multi-configuration aircraft can be endowed with strong autonomous, adaptive, and survival biological intelligent behavior capabilities, and the intelligence, safety, and reliability of the multi-configuration aircraft can be effectively improved.

[0022] Specifically, the time delay interference decoupling module has the functions of time delay interference coupling mechanism learning and separation and time delay interference compensation. The time delay interference decoupling module comprises a coupling mechanism learning and separation unit and a time delay interference compensation unit. The coupling mechanism learning and separation unit is used to learn the coupling characteristics between the time delay of the deformation actuator and the model uncertainty and the coupling characteristics between the external interference and the communication time delay through the balanced adversarial network, and to output the interference and time delay. At this time, the input of the balanced adversarial network is the time delay interference coupling information. The time delay interference compensation unit is used to synchronize the compensation and suppression of the time delay and interference through the multi-source compensation and suppression framework.

[0023] The multi-source compensation and inhibition framework comprises a time delay observer, a time delay compensator, and a disturbance observer; the time delay observer is configured to learn multi-source time delay online through a balanced adversarial network, and identify flight actuator time delay, deformation actuator time delay, and communication time delay; the time delay compensator is configured to compensate multi-source time delay through a balanced adversarial network recursively for k steps; and the disturbance observer is configured to learn disturbance online through a balanced adversarial network and output a disturbance value. Specifically, first, a time delay observer is established using a balanced adversarial network to learn time delay online, and multi-morphing aircraft time delay sources are analyzed according to time delay properties. The multi-source time delay includes flight actuator time delay, deformation actuator time delay, and communication time delay. At this time, the input of the time delay observer is multi-source time delay, and the output is flight actuator time delay, deformation actuator time delay, and communication time delay. Then, a recursive k-step time delay compensator is established using a balanced adversarial network to finely compensate multi-source time delay. At this time, the input of the time delay compensator is multi-morphing aircraft state values and time delay values of the previous k-1 steps, and the output is a k-step time delay value. Finally, a disturbance observer is established using a balanced adversarial network to learn disturbance online. At this time, the input of the disturbance observer is multi-morphing aircraft state values, and the output is a disturbance value.

[0024] Specifically, the multi-morphing aircraft bionic morphing module can realize wing span learning and control functions similar to a wild goose. The multi-morphing aircraft bionic morphing module learns the control decision mechanism of the wild goose during wing span change through a balanced adversarial network. The input is environmental information and wild goose configuration parameters, and the output is a morphing decision mechanism of the multi-morphing aircraft. to achieve accurate control of the wing span of the multi-morphing aircraft.

[0025] Specifically, the multi-morphing aircraft attitude coordination module has multi-morphing aircraft integrated modeling and multi-morphing aircraft attitude coordination control functions. The integrated modeling function of the multi-morphing aircraft attitude coordination module uses a hybrid driving modeling method combining mechanism modeling and data-driven modeling to model the multi-morphing aircraft. The data-driven modeling describes model uncertainty through a balanced adversarial network. The input of the data-driven model is angular velocity state values, and the output is inertia matrix uncertainty values.

[0026] Specifically, there are N morphing aircrafts. The mechanism modeling uses the following state vector to represent the state of the Nth morphing aircraft: i , ; wherein, V i is the flight speed, gamma i is the flight path inclination angle, omega ​i is the body angular velocity vector, lambda i is the configuration parameter; T i is the engine power, is the velocity change rate, a i is the attack angle, D i is the aerodynamic drag, m lambda is the variable configuration aircraft mass, g is the gravity acceleration, is the track angle change rate, L i is the aerodynamic lift, J i lambda i is the variable configuration inertia matrix, tau aero，i is the aerodynamic moment, tau morph，i is the variable configuration additional moment, Δ J i is the inertia matrix uncertainty, d ω，i is the control surface jitter disturbance, mu λ,i is the variable deformation control input, k λ is the natural attenuation coefficient, tau λ,i is the variable deformation actuator time delay, tau α,i is the attitude control actuator time delay, tau c,ij is the communication time delay from the first j aircraft to the second i aircraft; The additional moment caused by configuration change is: ; The variable mass inertia matrix is: ; The variable configuration parameter vector is: ; wherein, lambda 1 is the wing span extension coefficient, lambda 2 is the rear sweep angle change coefficient, lambda 3 is the wing area change coefficient.

[0027] Specifically, the multi-configuration aircraft attitude coordination module establishes a balanced counter-composite controller through a balanced counter network, and adopts a composite anti-interference control strategy to realize attitude coordination control. The specific expression form of the balanced counter-composite controller is as follows:​​ , , , , ; This includes two constraints, the first constraint... For: Deformation rate constraint, the second constraint condition For: Time-delay safety constraints; For the first k The predicted angular velocity of the step. omega ref For reference angular velocity, lambda opt For optimal configuration parameters, For the maximum deformation rate, N p To predict the time-domain step size, Q This is the angular velocity error weighting matrix; For the first k The predicted configuration parameters are used to establish an equal adversarial network. k The output of the step prediction model; w This is the configuration error weight matrix. tau λ,i For the time delay of the deformable actuator, △ lambda safe For safety deformation margin; The control surface jitter interference observation is represented by the interference observer output value.

[0028] Specifically, the equilibrium adversarial network mimics the brain's decision-making mechanism, completing the final feature classification and description through game-like competition between neurons; such as Figure 1 As shown, the balanced adversarial network includes an input layer 1, an intermediate layer 2, and an output layer 3. The intermediate layer 2 includes: a feature extractor 201 for extracting shared and private features; a competitive adversarial network 202 for selecting the winning neuron through a competitive mechanism; and a policy generator 203 for generating the final policy based on the shared and private features. The output of the feature extractor 201 serves as the input to the competitive adversarial network, and the output of the competitive adversarial network 202 serves as the input to the policy generator 203. The policy generator 203 is fully connected, as detailed below: (1) The feature extractor 201 includes shared features f s With private characteristics The specific expression is as follows: ; wherein, s is the current state, W s is the weight matrix, b s is the bias, and ReLU is an activation function; ; wherein, s is the current state, represents the action at the last moment, and LSTM is a long short-term memory network.

[0029] (2) The competitive adversary 202: when the input signal is transmitted, the competitive layer neuron compares the similarity of the input vector and its own weight, and finally only one neuron wins (the activation value is the highest), representing the characteristics or category of the input; the specific process is as follows: The similarity of the neuron and the input is calculated, and the Euclidean distance is usually used to measure the similarity of the input x and the weight of the neuron w j The calculation formula is: The winning neuron c , c The expression is: .

[0030] (3) The policy generator 203 is represented as follows: ; wherein, The shared feature f s (global state information) and the private feature (individual information of the competitor) are spliced into a joint feature vector; is the unique weight matrix of the competitor, is the corresponding bias term; and softmax is normalization.

[0031] Specifically, the learning algorithm of the balanced adversarial network adopts any one of a gradient optimization method, a meta-learning algorithm, a recursive least square algorithm or a wake-sleep algorithm; the learning algorithm can also adopt other algorithms that can meet the optimization function, and is not limited to the above several learning algorithms.

[0032] In addition, the bionic learning-based multi-configuration aircraft attitude cooperative control method provided by the application is realized by using the bionic learning-based multi-configuration aircraft attitude cooperative control system, and includes the following steps: Step 1: learning and separating the coupling mechanism of time delay and disturbance by using the balanced adversarial network, and realizing compensation of the time delay and the disturbance; Step 2: learn the control decision mechanism of the wing span of the wild goose through the balanced adversarial network, so as to realize the bionic control of the variable-configuration aircraft wing span; Step 3: the mixed driving modeling method combining mechanism modeling and data driven modeling is used for integrated modeling of the variable-configuration aircraft; Step 4: the composite controller is constructed through the balanced adversarial network, and the attitude cooperative control of the variable-configuration aircraft is realized.

[0033] In summary, through the embodiments provided by the present application, the following advantages of the present application can be summarized: (1) The present application combines bionic technology, anti-interference technology, automatic control technology and artificial intelligence technology to design a variable-configuration aircraft attitude cooperative control system and method based on bionic learning, which learns and separates the time delay interference coupling mechanism, compensates the time delay interference, learns and controls the wing span of the wild goose, and realizes the integrated modeling of the variable-configuration aircraft and the attitude cooperative control of the variable-configuration aircraft, so as to endow the variable-configuration aircraft with strong autonomous, adaptive and survival bionic intelligent behavior capabilities, and effectively improve the intelligence, safety and reliability of the variable-configuration aircraft.

[0034] (2) The present application combines bionic technology, automatic control technology and artificial intelligence technology, and uses the balanced adversarial network as a coupling mechanism learning model to learn and separate the time delay interference coupling mechanism of the variable-configuration aircraft, which can effectively and accurately separate the time delay and interference.

[0035] (3) The present application combines bionic technology, automatic control technology and artificial intelligence technology, and uses the balanced adversarial network to construct a time delay observer, a time delay compensator and an interference observer, which can estimate and compensate the time delay and interference of the variable-configuration aircraft.

[0036] (4) The present application combines bionic technology, automatic control technology and artificial intelligence technology, and learns the configuration parameters of the wild goose when the wing span changes in different environments through the balanced adversarial network, so as to accurately obtain the control decision mechanism, and effectively control the wing span of the variable-configuration aircraft.

[0037] (5) The present application combines bionic technology, automatic control technology and artificial intelligence technology, and adopts a form combining mechanism modeling and data driven modeling to construct an integrated model of the variable-configuration aircraft, wherein the data driven modeling uses the balanced adversarial network to characterize the uncertainty of the variable-configuration aircraft, which can effectively improve the modeling accuracy.

[0038] (6) The present application combines bionic technology, automatic control technology and artificial intelligence technology, and establishes a balanced adversarial composite controller through the balanced adversarial network, and uses a composite anti-interference control strategy to control the variable-configuration aircraft, which effectively improves the control precision and reliability.

[0039] (7) The application combines artificial intelligence technology to establish a balanced confrontation network, including an input layer, an intermediate layer and an output layer, wherein the intermediate layer includes a feature extractor, a competitive antagonist and a strategy generator, which has strong non-linear approximation, fast convergence, strong generalization ability, self-learning ability, robustness and anti-interference performance, and can be well applied as an approximation model and a classification model.

[0040] In conclusion, the application combines bionic technology, anti-interference technology, automatic control technology and artificial intelligence technology to provide a multi-structure aircraft attitude cooperative control system and method based on bionic learning, which can enable the multi-structure aircraft to have strong autonomous, strong adaptive and strong survival bionic intelligent behavior capabilities, and has the beneficial effects of improving the intelligence, safety and reliability of the multi-structure aircraft.

[0041] It can be understood that the above specific description of the application is only used to illustrate the application and is not limited to the technical solutions described in the embodiments of the application, and those skilled in the art should understand that the application can still be modified or replaced equivalently to achieve the same technical effect, as long as it meets the use needs, it is within the protection scope of the application.

Claims

1. A multi-variable aircraft attitude cooperative control system based on biomimetic learning, characterized in that: The time delay interference decoupling module, the multi-configuration aircraft bionic variable configuration module, and the multi-configuration aircraft attitude coordination module are all implemented by using a balanced adversarial network. The time delay interference decoupling module is configured to learn and separate the time delay and interference coupling mechanism of the multi-configuration aircraft, and to compensate for the time delay and interference. The multi-configuration aircraft bionic variable configuration module is configured to learn the control decision mechanism of the wing span of a wild goose by using the balanced adversarial network, and to input environmental information and wild goose configuration parameters and output a variable configuration decision mechanism of the multi-configuration aircraft, so as to accurately control the wing span of the multi-configuration aircraft. The multi-configuration aircraft attitude coordination module is configured to model and control the attitude of the multi-configuration aircraft in an integrated manner. The time delay interference decoupling module, the multi-configuration aircraft bionic variable configuration module, and the multi-configuration aircraft attitude coordination module are all implemented by using a balanced adversarial network.

2. The biomimicry learning based multi-dof aerial vehicle cooperative attitude control system of claim 1, wherein: The time delay interference decoupling module includes a coupling mechanism learning and separation unit and a time delay interference compensation unit. The coupling mechanism learning and separation unit is configured to learn the coupling characteristics between the time delay of the deformation actuator and the model uncertainty and the coupling characteristics between the external interference and the communication time delay by using the balanced adversarial network, and to output the interference and the time delay. The time delay interference compensation unit is configured to compensate for and suppress the time delay and the interference by using a multi-source compensation and suppression framework.

3. The biomimicry learning based multi-dof aerial vehicle cooperative attitude control system of claim 2, wherein: The multi-source compensation and suppression framework includes a time delay observer, a time delay compensator, and an interference observer. The time delay observer is configured to learn the multi-source time delay online by using the balanced adversarial network, and to identify the flight actuator time delay, the deformation actuator time delay, and the communication time delay. The time delay compensator is configured to compensate for the multi-source time delay by using the balanced adversarial network recursively for k steps. The interference observer is configured to learn the interference online by using the balanced adversarial network and to output the interference value.

4. The biomimicry learning based multi-dof aerial vehicle cooperative attitude control system of claim 1, wherein: The multi-configuration aircraft bionic variable configuration module learns the control decision mechanism of the wing span of a wild goose by using the balanced adversarial network, inputs environmental information and wild goose configuration parameters, and outputs a variable configuration decision mechanism of the multi-configuration aircraft, so as to accurately control the wing span of the multi-configuration aircraft.

5. The biomimicry learning based multi-dof aerial vehicle cooperative attitude control system of claim 3, wherein: The multi-configuration aircraft attitude coordination module models the multi-configuration aircraft in an integrated manner by using a hybrid driving modeling method combining mechanism modeling and data-driven modeling. The data-driven modeling describes the model uncertainty by using the balanced adversarial network, and the input of the data-driven model is the angular velocity state value, and the output is the inertia matrix uncertainty value.

6. The bionic learning-based multi-variable aircraft cooperative control system according to claim 5, wherein: The mechanism modeling uses the following state vector to represent the first... i The state of a variant aircraft: ， ； where V i is the flight velocity, The multi-configuration aircraft attitude coordination module establishes a balanced adversarial composite controller by using the balanced adversarial network, and implements attitude coordination control by using a composite anti-interference control strategy. i is the flight path bank angle, The balanced adversarial composite controller has the following specific expression form: i is the body angular velocity vector, The balanced adversarial network includes an input layer, an intermediate layer, and an output layer. i is the configuration parameter; T i is the engine power, is the velocity rate of change, a i is the angle of attack, D i is the aerodynamic drag, m The intermediate layer includes a feature extractor configured to extract shared features and private features, a competitive adversarial configured to select a winning neuron by using a competitive mechanism, and a strategy generator configured to generate a final strategy based on the shared features and the private features. is the variable configuration aircraft mass, g is the gravitational acceleration, is the flight path angle rate of change, L i is the aerodynamic lift, J i The learning algorithm of the balanced adversarial network uses any one of a gradient optimization method, a meta-learning algorithm, a recursive least squares algorithm, or a wake-sleep algorithm. i is the variable configuration inertia matrix, ​ aero，i is the aerodynamic moment, ​ morph，i is the variable configuration additional moment, Δ J i is the inertia matrix uncertainty, d ω，i is the control surface buzz disturbance, ​ λ,i is the morphing control input, k λ is the natural decay coefficient, ​ λ,i is the morphing actuator time delay, ​ α,i is the attitude control actuator time delay, ​ c,ij is the communication time delay from the first j aircraft to the second i aircraft.​​ 7. The biomimicry learning based multi-dof aerial vehicle cooperative attitude control system of claim 3, wherein: ​ ， ， ， ， ； wherein two constraints are included, the first constraint is a deformation rate limit constraint, and the second constraint is a time delay safety constraint; is the angular velocity predicted in the first k step, ​ ref is the reference angular velocity, ​ opt is the optimal configuration parameter, is the maximum deformation rate, N p is the prediction time domain step size, Q is the angular velocity error weight matrix; is the configuration parameter predicted in the first k step, which is the output of the balanced adversarial network k establishing the first step prediction model; w is the configuration error weight matrix, ​ λ,i is the deformation actuator time delay, △ ​ safe is the safety deformation margin; is the control surface jitter disturbance observation, which is the output value of the disturbance observer.

8. The biomimicry learning based multi-dof aerial vehicle cooperative attitude control system of claim 1, wherein: ​ ​ ​ ​ ​ 9. The biomimicry learning based multi-dof aerial vehicle cooperative attitude control system of claim 1, wherein: ​ 10. A method for cooperative control of the attitude of a morphing aircraft based on biomimetic learning, characterized in that: The biomimetic learning-based multi-variable aircraft attitude cooperative control system of any one of claims 1-9 is implemented, comprising the following steps: Step 1: learning and separating the coupling mechanism of time delay and disturbance through the balanced adversarial network, and realizing the compensation of time delay and disturbance; Step 2: learning the control decision mechanism of the variable wingspan of the wild goose through the balanced adversarial network, and realizing the biomimetic control of the wingspan of the multi-variable aircraft; Step 3: using a hybrid driving modeling method combining mechanism modeling and data-driven modeling to model the multi-variable aircraft as a whole; Step 4: constructing a composite controller through the balanced adversarial network to realize the attitude cooperative control of the multi-variable aircraft.

Images