Random time delay unmanned aerial vehicle formation control method and system based on event triggering mechanism

Abstract

The invention provides a random time delay unmanned aerial vehicle formation control method and system based on an event triggering mechanism, and relates to the technical field of unmanned aerial vehicle formation control. According to the method, a global inertial coordinate system is established, and an original motion state is converted into a standard motion state in a per-unit manner; defining a lumped interference truth value by using a heterogeneous dynamic constraint equation; designing a non-linear interference observer into which an auxiliary variable is introduced, and calculating a lumped interference estimated value approaching a truth value; constructing a full-state coupling dynamic event triggering mechanism, and calculating a dynamic triggering threshold value which comprises a minimum interval protection item and is in negative correlation with the interference mode length; correcting the neighbor nodes with the communication random time delay based on the lumped interference estimated value to obtain a predicted neighbor motion state; and inferring an optimal co-state vector by using a stable manifold neural network, and generating a physical control instruction by combining interference feed-forward compensation. According to the invention, the communication load is reduced while the formation anti-interference capability is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle formation control, in particular to a random delay unmanned aerial vehicle formation control method and system based on an event-triggered mechanism. BACKGROUND

[0002] With the wide application of unmanned aerial vehicle cluster technology in power inspection, emergency rescue and smart city, etc., multi-agent formation control has become the core direction of the integration of automation technology and aviation industry. In actual operation, the unmanned aerial vehicle formation not only needs to maintain high-precision cooperative formation, but also needs to cope with severe communication bandwidth and computing resource constraints. Therefore, applying advanced control algorithms to flight systems to realize the collaborative optimization of "communication-computation-control" is the key path to improve cluster efficiency.

[0003] In the prior art, there are certain achievements for the optimization control and communication load reduction of unmanned aerial vehicle formation. The paper "Optimal control for quadrotors UAV based on deep neural network approximations of stable manifold of HJB equation" published in the journal "Automatika" discloses a method of using deep neural network to approximate the stable manifold of HJB equation. This method replaces online partial differential equation solving by offline neural network training, realizes millisecond-level optimal control signal generation, and significantly improves control real-time performance. On the other hand, the publication number CN117991823A discloses a random delay unmanned aerial vehicle formation control method based on an event-triggered mechanism, which transmits data only at necessary time by setting trigger conditions, thereby reducing network load.

[0004] However, the above-mentioned technologies still have significant defects when facing complex and variable nonlinear physical disturbances in real flight environments. First, existing cooperative control schemes are usually based on ideal deterministic dynamic models. This kind of method often simplifies the gust wind, propeller aerodynamic disturbance and body gyroscopic effect as Gaussian white noise or completely ignores it, lacks active observation and feedforward compensation mechanism for lumped disturbance. When encountering sudden wind or load changes, this open-loop optimal control law is easy to fail, causing tracking error divergence. Second, in a strong disturbance environment, the traditional event-triggered mechanism faces serious challenges: external disturbance can cause the system state to fluctuate dramatically. If the triggering threshold cannot be dynamically coupled with the disturbance intensity, the triggering condition is likely to be met frequently, causing the Zeno phenomenon. This not only causes the control command update frequency to exceed the hardware processing capacity, but also generates a large number of redundant data packets, which worsens the communication load, making the originally used bandwidth-saving mechanism counterproductive. Therefore, there is an urgent need for a comprehensive control scheme that can not only retain adaptability to random time delays, but also solve the problems of model mismatch and communication trigger storm under strong disturbance through active anti-disturbance mechanism.

[0005] The above information disclosed in the background section is only intended to enhance the understanding of the background of the present disclosure, and thus it can include information that does not constitute the prior art known to those of ordinary skill in the art. SUMMARY

[0006] The purpose of the present application is to provide a random time delay UAV formation control method and system based on an event-triggered mechanism to solve the problems raised in the background technology.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solutions: The random time delay UAV formation control method based on an event-triggered mechanism comprises the following specific steps: A global inertial coordinate system is established to obtain the original motion state of each node of the formation cluster in real time and perform normalization processing to obtain the standard motion state, wherein the original motion state includes position, velocity, rotation angle and angular velocity; According to the standard motion state, a physical constraint is established using Newton-Euler equation to construct heterogeneous dynamic constraint equations of translational channel and rotational channel, and define the lumped disturbance true value; Combined with the heterogeneous dynamic constraint equation, auxiliary variables are introduced to convert the differential operation on the motion state into integral operation, a double-channel nonlinear disturbance observer is designed, and the lumped disturbance estimated value approximating the lumped disturbance true value is calculated through nonlinear iteration; A full-state coupled dynamic event triggering mechanism is constructed, and a dynamic triggering threshold is calculated. The dynamic triggering threshold is negatively correlated with the magnitude of the lumped interference estimate and includes a minimum interval protection term that decays exponentially with time. When the comprehensive state error exceeds the dynamic triggering threshold, communication is triggered, and the standard motion state is sent to the neighboring nodes and a request to obtain the standard motion state is sent to the neighboring nodes. Based on the lumped interference estimate, the motion state of neighboring nodes with random communication delays is corrected for interference to obtain the predicted neighbor motion state. A cascaded optimal disturbance rejection controller is constructed. The position and attitude coordination error of the nodes is calculated. The position and attitude coordination error of the nodes is input into an offline trained stable manifold neural network to directly infer the optimal costate vector. The optimal nominal control law is obtained by solving the problem. The lumped disturbance estimate is introduced for feedforward compensation. The final physical control command is generated and executed.

[0008] Furthermore, a three-dimensional rectangular coordinate system is established with the ground control base station as the origin, the horizontal plane as the XY axis plane, and the vertical plane as the XZ axis plane. Each device in the formation cluster is regarded as a different node. The original motion state of each node in the formation cluster is collected in real time through airborne sensors. The original motion state includes the original position vector, the original velocity vector, the original attitude angle vector, and the original angular velocity vector. The original motion states of each node are normalized according to the standard unit length, standard unit speed, standard unit attitude angle and standard angular velocity units preset by the relevant staff, so as to obtain the standard motion state of each node.

[0009] Furthermore, physical constraints are established using the Newton-Euler equations based on the standard motion states of each node. Based on Newton's second law, and combined with the standard velocity vector in a standard state of motion, we construct the... The equation relating the translational acceleration and force at each node, the first node... The product of the standard mass and standard acceleration vector of each node is equal to the sum of the standard thrust vector, the standard gravity vector, and the true value of the translational lumped disturbance, thus obtaining the dynamic constraint equation of the translational channel. Based on Euler's equations for rigid body dynamics, and combined with the standard angular velocity vector in the standard motion state, an equation is constructed between standard angular acceleration and torque. The product of the standard moment of inertia matrix and the standard angular acceleration vector is equal to the sum of the standard control torque vector minus the gyroscopic effect torque term and the true value of the rotational lumped disturbance. The gyroscopic effect torque term is obtained by performing a cross product operation on the standard angular velocity vector and the product of the standard moment of inertia matrix and the standard angular velocity vector. The dynamic constraint equations of the rotation channel are then obtained.

[0010] Furthermore, for translational disturbance observers; Based on the auxiliary variable transformation method in the design criteria for nonlinear disturbance observers in control theory, the first... The translational auxiliary variable of the nth node in the translational disturbance estimation calculation is defined as the nth node. The translational auxiliary variable of the nth node and the nth node The estimated translational lumped disturbance of the nth node satisfies a linear transformation relationship, specifically the nth node. The translational auxiliary variable of the nth node is equal to the nth node. The translational lumped disturbance estimate of the nth node minus the translational correction term, where the translational correction term is the preset translational observer gain matrix, the nth node, and the nth node, respectively. The product of the standard mass and standard velocity vector of each node; The first The translational auxiliary variable of the nth node and the nth node The linear relationship between the estimated translational lumped disturbance values ​​of the nodes is differentiated with respect to a preset unit time, and the nth node is set according to the negative feedback mechanism. The rate of change of the estimated translational lumped disturbance of the nth node and the 1st node The true value of the translational lumped disturbance of the nth node and the nth node The difference in the estimated translational lumped disturbance of the nth node is proportional to the difference in the estimated values ​​of the nth node. Combined with the dynamic constraint equations of the translational channel, the nth node is calculated. The sum of the standard thrust vector, standard gravity vector, translational auxiliary variable from the previous moment, and translational correction term for each node is then multiplied by the negative value of a preset translational observer gain matrix to obtain the result. The rate of change of the translational auxiliary variable of each node within a preset unit of time; Combined with the The rate of change of the translational auxiliary variable at the nth node within a preset unit time is obtained by numerical integration iteration to obtain the nth node at the current time. The translational auxiliary variable of the nth node is substituted into the current time step of the nth node. The translational auxiliary variable of the nth node and the nth node The linear relationship between the estimated translational lumped disturbance values ​​of the n nodes is used to calculate the current time n. Estimates of the translational lumped disturbance of each node; The principle for the rotational disturbance observer is the same as that for the translational disturbance observer. Construct the first The rotational auxiliary variable of the nth node in the translational disturbance estimation calculation is defined as the nth node. The rotation auxiliary variable of the nth node and the nth node The rotation lumped disturbance estimate of the nth node satisfies a linear transformation relationship, specifically the nth node. The rotation auxiliary variable of the nth node is equal to the nth node. The rotation lumped disturbance estimate of the nth node minus the rotation correction term, where the rotation correction term is the preset rotation observer gain matrix, the th node's rotation lumped disturbance estimate, the lumped disturbance estimate of the nth node, the rotation correction term is the preset rotation observer gain matrix ... The product of the standard moment of inertia matrix and the standard angular velocity vector of each node; The first The rotation auxiliary variable of the nth node and the nth node The linear relationship between the estimated lumped disturbance of the rotation at the nth node and the preset unit time is differentiated, and combined with the dynamic constraint equations of the rotation channel, the nth node is calculated. The sum of the standard control torque vector of each node, the rotation auxiliary variable of the previous moment, and the rotation correction term, minus the gyroscopic effect torque term, and then multiplying the final difference vector by the negative value of the preset rotation observer gain matrix, yields the result. The rate of change of the rotation auxiliary variable of each node within a preset unit of time; Combined with the The rate of change of the rotation auxiliary variable of the nth node within a preset unit time is obtained by numerical integration iteration to obtain the nth node at the current time. The rotation auxiliary variable of the nth node is substituted into the nth node. The rotation auxiliary variable of the nth node and the nth node The linear relationship between the estimated lumped disturbance of the rotation of the nth node is used to calculate the current time. Estimated lumped disturbance of rotation at each node.

[0011] Further, calculate the first The magnitudes of the translational lumped interference estimates and rotational lumped interference estimates of each node are used to quantify the external interference intensity at the current moment, and the difference between the current time and the last triggered communication time is calculated as the elapsed time. A fully coupled dynamic event triggering mechanism is constructed, and the dynamic triggering threshold at the current time is calculated. The specific calculation logic for the dynamic triggering threshold is as follows: First, the interference adaptation component is calculated. The interference adaptation component is used to adaptively adjust the threshold according to environmental changes. The value is negatively correlated with the intensity of external interference. Specifically, it is obtained by dividing the preset basic error tolerance constant by the weighted sum of interference intensity. The weighted sum of interference intensity is composed of the weighted value of the magnitude of the translational lumped interference estimate, the weighted value of the magnitude of the rotational lumped interference estimate, and the sum of preset constant terms. Secondly, the minimum interval protection term is calculated. The minimum interval protection term is used to prevent continuous triggering in a short period of time. Its value decays exponentially with the passage of time. Specifically, it is obtained by multiplying the preset protection term amplitude by the time decay factor. The time decay factor is calculated by performing a negative exponential operation based on the passage of time, so that the component reaches its maximum value at the moment after the communication is triggered and gradually approaches zero over time. Finally, a threshold truncation judgment is performed, the disturbance adaptive component is added to the interval protection component to obtain a preliminary calculation threshold, the preliminary calculation threshold is compared with a preset dynamic trigger threshold lower limit, and a larger value between the two is selected as the dynamic trigger threshold of the current time.

[0012] Further, a Euclidean distance is used to calculate a length of a deviation of the standard motion state of the current time from an expected motion state to obtain a comprehensive state error of the current time, the comprehensive state error of the current time is compared with the dynamic trigger threshold of the current time, if the comprehensive state error of the current time is less than the dynamic trigger threshold of the current time, communication is not triggered, if the comprehensive state error of the current time is greater than or equal to the dynamic trigger threshold of the current time, communication is triggered, the standard motion state of the current time is sent to a neighbor node and a request for obtaining the standard motion state is sent to the neighbor node, wherein the expected motion state is a current time theoretical state calculated based on a preset space-time trajectory function or a current time theoretical state deduced from a reference state sent by a set leader node to other nodes.

[0013] Further, the first node receives the state data packet sent by the neighbor node, a random communication time delay is obtained by calculating a difference between a time stamp in the data packet and the current time, and a standard random communication time delay is obtained by normalizing the random communication time delay; For the translational state time delay compensation prediction of the first node, a predicted neighbor standard position vector after disturbance compensation is calculated based on the kinematic Taylor expansion principle, combined with a translational lumped disturbance estimation value sent by the neighbor node and a dynamic constraint equation of the translational channel, and the specific calculation logic is as follows: taking the standard position vector in the state data packet sent by the neighbor node as a reference, a linear displacement generated by a standard velocity vector of the neighbor node within the standard random communication time delay is first superimposed, and then a second-order dynamic correction displacement generated by a resultant force is superimposed, the calculation method of the second-order dynamic correction displacement is as follows: a vector sum of the standard thrust vector of the neighbor node and the translational lumped disturbance estimation value is calculated, the vector sum is divided by the standard mass of the neighbor node to obtain an equivalent acceleration, and the equivalent acceleration is multiplied by half of the square of the standard random communication time delay to obtain the predicted neighbor standard position vector. For the translational state time delay compensation prediction of the first The rotation state time delay compensation prediction of the neighbor node of the node, combined with the rotation set interference estimation value sent by the neighbor node and the dynamic constraint equation of the rotation channel, calculates the predicted neighbor standard attitude angle vector after interference compensation based on the rotation dynamics Taylor expansion principle. The specific calculation logic is to take the standard attitude angle vector in the state data packet sent by the neighbor node as the reference. First, the angular displacement generated by the standard angular velocity vector of the neighbor node within the standard random communication time delay is superimposed, and then the second-order angular dynamics correction displacement generated by the resultant moment is superimposed. The calculation method of the second-order angular dynamics correction displacement is as follows: the vector sum of the standard control moment vector of the neighbor node and the rotation set interference estimation value is calculated, the inverse matrix of the standard moment of inertia matrix of the neighbor node is used to transform the vector sum to obtain the equivalent angular acceleration, and then the equivalent angular acceleration is multiplied by half of the square of the standard random communication time delay to obtain the predicted neighbor standard attitude angle vector.

[0014] Further, the comprehensive position error of the current node is calculated in a three-dimensional rectangular coordinate system. The comprehensive position error of the current node is composed of the cooperative error with the neighbor node and the tracking error of the expected motion state at the current time, wherein the cooperative error is obtained by calculating the difference between the standard position vector of the first node, the predicted standard position vector of the neighbor node and the expected distance vector, and then weighting the calculated difference; and the tracking error is obtained by calculating the difference between the standard position vector of the first node and the expected motion state at the current time and then weighting the calculated difference. The comprehensive position error of the first node is input into the position loop stable manifold neural network trained in advance to obtain the position loop optimal cooperative state vector of the standard motion state of the first node in the HJB equation. Based on the position loop optimal cooperative state vector, the difference between the optimal feedback control rate formula and the translational set interference estimation value is calculated to obtain a three-dimensional anti-interference virtual resultant force vector. The module length of the three-dimensional anti-interference virtual resultant force vector is calculated to obtain the final total thrust command of the unmanned aerial vehicle rotor. The expected attitude angle vector of the first node is obtained by combining backstepping control based on the three-dimensional anti-interference virtual resultant force vector and the preset reference yaw angle. The difference between the standard attitude vector in the standard motion state of the first node and the expected attitude angle vector is calculated to obtain an attitude tracking deviation vector. The comprehensive position error of the first node is input into the attitude loop stable manifold neural network trained in advance to obtain the attitude loop optimal cooperative state vector of the standard motion state of the first The nodes adjust motion states according to the final total thrust command and the final anti-disturbance torque command.

[0015] The random delay UAV formation control system based on the event triggering mechanism is used for realizing the UAV formation control method, and comprises: The data acquisition module is used for establishing a global inertial coordinate system, acquiring original motion states of each node of the formation cluster in real time and performing standardization processing to obtain standard motion states, wherein the original motion states include positions, velocities, rotation angles and angular velocities. The interference definition module is used for establishing physical constraints by using Newton-Euler equations according to the standard motion states, constructing heterogeneous dynamics constraint equations of the translational channel and the rotational channel, and defining a collective interference true value. The interference analysis module is used for combining the heterogeneous dynamics constraint equations, designing a double-channel nonlinear finite-time interference observer, introducing auxiliary variables to convert differential operations on the motion states into integral operations, and calculating a collective interference estimated value approximating the collective interference true value in real time through nonlinear iteration. The event determination module is used for constructing a full-state coupled dynamic event triggering mechanism, calculating a dynamic triggering threshold, and triggering communication when a state error exceeds the dynamic triggering threshold. The communication module is used for performing interference correction on motion states of neighbor nodes with communication random delay based on the collective interference estimated value to obtain predicted neighbor motion states. The formation correction module is used for constructing a cascaded optimal anti-disturbance controller, calculating position and attitude cooperative errors of the nodes, inputting the position and attitude cooperative errors of the nodes into a stable manifold neural network trained offline to directly infer an optimal cooperative state vector, solving an optimal nominal control law, introducing the collective interference estimated value for feedforward compensation, generating a final physical control command and executing the final physical control command.

[0016] Compared with the prior art, the beneficial effects of the present application are: The present application unifies the huge difference in physical dimension between heterogeneous nodes according to the established global inertial coordinate system and the normalization processing of the original motion state. On this basis, the heterogeneous dynamics constraint equation is constructed by using Newton-Euler equation and the lumped disturbance true value is defined, and then the differential operation on the motion state is converted into integral operation by introducing auxiliary variables to design a double-channel nonlinear disturbance observer. This processing method can obtain the lumped disturbance estimation value approximating the lumped disturbance true value in real time through nonlinear iteration without relying on acceleration sensor and avoiding differential noise amplification, so as to accurately capture the nonlinear physical disturbances such as airflow and friction. At the same time, the stable manifold neural network trained offline is combined to directly infer the optimal co-state vector, and the optimal nominal control law is obtained. This method replaces the online complex solving process of the traditional HJB equation, and improves the generation speed of the control command. By introducing the lumped disturbance estimation value into the optimal nominal control law for feedforward compensation, the final physical control command not only has global convergence based on neural network optimization, but also has active cancellation ability for environmental disturbances, realizing the dual improvement of control accuracy and calculation efficiency. The present application also designs a dynamic triggering threshold containing a minimum interval protection term exponentially decaying with time and negatively correlated with the modulus of the lumped disturbance estimation value by constructing a full-state coupled dynamic event triggering mechanism. This design enables the communication triggering condition to adaptively adjust the sensitivity according to the intensity of external disturbance, and triggers the minimum interval protection term in the initial stage of just triggering communication to improve the dynamic triggering threshold, fundamentally avoiding the phenomenon of infinite triggering in a short time, ensuring the timeliness of state updating while significantly reducing the communication load. In addition, for the random time delay problem existing in communication, the motion state of the neighbor node is corrected based on the dynamics model by using the lumped disturbance estimation value to obtain the predicted neighbor motion state. This prediction compensation method based on physical force analysis corrects the trajectory deviation caused by simple linear extrapolation based on speed, ensuring that the position and attitude coordination error calculation between nodes is still accurate and reliable in the random communication delay environment, maintaining the dynamic stability of the formation shape in complex communication environment. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The present application is a whole method flowchart; Figure 2 The present application is a whole system structure schematic diagram. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical scheme and advantages of the present application more clear and obvious, the present application is further described in detail below combined with specific embodiments.

[0019] It should be noted that, unless otherwise defined, technical terms or scientific terms used in the present application shall be understood as having the usual meaning as understood by a person with ordinary skill in the art to which the present application pertains. The terms "first", "second", and the like used in the present application do not indicate any order, number, or importance, but are only used to distinguish different components. The terms "include", "contain", and the like mean that the elements or objects before the terms encompass the elements or objects listed after the terms and their equivalents, and do not exclude other elements or objects. The terms "connect" or "connected" and the like do not mean physical or mechanical connection, but can include electrical connection, whether direct or indirect. The terms "upper", "lower", "left", "right", and the like are only used to indicate relative positional relationships, and when the absolute positions of the described objects change, the relative positional relationships can also change accordingly.

[0020] Embodiment: Please refer to Figure 1 The present application provides a technical solution: The random time delay UAV formation control method based on the event triggering mechanism includes the following specific steps: Step 1: Establish a global inertial coordinate system, and obtain the original motion state of each node in the formation cluster in real time and perform normalization processing to obtain the standard motion state, wherein the original motion state includes position, velocity, rotation angle, and angular velocity; In this embodiment, the ground control base station is taken as the coordinate origin, the horizontal plane is taken as the X-Y axis plane, and the vertical plane is taken as the X-Z axis plane to establish a three-dimensional rectangular coordinate system. Each device in the formation cluster is regarded as a different node, and the original motion state of each node in the formation cluster is collected in real time by an on-board sensor. The original motion state includes an original position vector, an original velocity vector, an original attitude angle vector, and an original angular velocity vector. The original motion state of each node collected is normalized according to the standard unit length, the standard unit velocity, the standard unit attitude angle, and the standard angular velocity unit preset by the relevant staff to obtain the standard motion state of each node. Position and velocity data are the fundamental feedback quantities for trajectory tracking and formation maintenance, determining the accuracy of outer-loop position coordination control. Attitude angles and angular velocities are crucial for underactuated systems like rotary-wing UAVs, as horizontal displacement must be achieved through attitude tilting, and angular velocity directly affects the stability of inner-loop flight, serving as a key input for subsequent disturbance rejection control. Heterogeneous nodes exhibit orders of magnitude differences in their physical properties. For example, two rotary-wing UAVs of different weights may perform different tasks and collaborate on a formation mission; the moments of inertia of nodes in such a heterogeneous formation cluster will differ. Directly using the collected raw data for joint calculations can lead to ill-conditioned matrix operations in the control algorithm, causing imbalances in weight updates or computational divergence. By introducing standard unit length, standard unit velocity and other benchmark values ​​for standardization, the influence of dimensions can be eliminated, and the state of each node in the formation cluster of different magnitudes can be mapped to a unified dimensionless mathematical space. This makes the same set of dynamic models and control parameters compatible with different types of hardware platforms, thereby ensuring the numerical stability and generalization ability of the system calculation. Specifically, standardization means expressing physical quantities as proportional values ​​based on benchmark values ​​preset by relevant personnel.

[0021] Step 2: Based on the standard motion state, establish physical constraints using the Newton-Euler equations, construct heterogeneous dynamic constraint equations for the translational and rotational channels, and define the true value of lumped disturbances; In this embodiment, physical constraints are established using the Newton-Euler equations based on the standard motion states of each node. Based on Newton's second law, and combined with the standard velocity vector in a standard state of motion, we construct the... The equation relating the translational acceleration and force at each node, the first node... The product of the standard mass and standard acceleration vector of each node is equal to the sum of the standard thrust vector, the standard gravity vector, and the true value of the translational lumped disturbance, thus obtaining the dynamic constraint equation of the translational channel. The dynamic constraint equations for the translational channel are expressed as follows: In the formula, Indicates the node number. Indicates the preset first Standard quality of each node Indicates the first Standard velocity vector of each node The first derivative with respect to time, i.e., the standard acceleration vector, Indicates the first The standard thrust vector of each node is obtained from the physical control command output in the previous control cycle. Indicates the preset first The standard gravity vector of each node Indicates the first The true value of the translational lumped disturbance of each node; The true value of translational lumped disturbance comprehensively reflects all unknown resultant forces acting on the UAV's position channel. Its magnitude directly characterizes the strength of disturbance factors such as gust drag, model error force, and unmodeled dynamic friction in the actual flight environment. The introduction of this parameter allows this method to quantify environmental disturbances that cannot be directly measured into explicit variables in the mathematical model, thus providing subsequent observers with accurate target approximations. The calculation results of the translational lumped disturbance true value have a close physical coupling relationship with the standard acceleration vector, standard thrust vector, and standard gravity vector. This is because, according to Newton's second law, the change in an object's state of motion is determined by the vector sum of all acting forces. Given the thrust and gravity, the deviation between the actual acceleration and the theoretical acceleration must originate from external disturbances. Specifically, the true value of translational lumped disturbance is positively correlated with the standard acceleration vector and negatively correlated with the standard thrust vector. This means that if the standard acceleration actually generated by the UAV increases under the same standard thrust, it indicates the presence of a positive disturbance force, and the true value of the disturbance increases. Conversely, if the thrust is large but the acceleration is small, it indicates the presence of a reverse drag force, and the true value of the disturbance decreases. Based on Euler's equations of rigid body dynamics, and combined with the standard angular velocity vector in the standard motion state, an equation is constructed between standard angular acceleration and torque. The product of the standard moment of inertia matrix and the standard angular acceleration vector is equal to the sum of the standard control torque vector minus the gyroscopic effect torque term and the true value of the rotational lumped disturbance. The gyroscopic effect torque term is obtained by performing a cross product operation on the product of the standard angular velocity vector and the standard moment of inertia matrix and the standard angular velocity vector. The dynamic constraint equations of the rotation channel are then obtained. The dynamic constraint equations for the rotating channel are expressed as follows: In the formula, Indicates the first The standard moment of inertia matrix of each node is obtained by recording motor torque commands and angular velocity response data, and then fitting it based on the rigid body dynamics equations using a frequency domain identification algorithm. Indicates the first Standard angular velocity of each node The first derivative with respect to time, i.e., the standard angular acceleration vector, Indicates the first The standard control torque vector for each node is obtained from the torque command output in the previous cycle. This represents the vector cross product operation. Indicates the first The true value of the lumped disturbance of the rotation of each node.

[0022] The rotation lumped disturbance true value specifically reflects unknown moments acting on the attitude channel of the unmanned aerial vehicle, and covers factors such as aerodynamic yawing moment, gravity center offset moment and propeller airflow disturbance moment. By introducing a gyroscopic effect term in the equation , the dynamic constraint equation of the rotation channel can separate the nonlinear physical characteristics specific to the rotating rigid body from the disturbance, ensure that the calculated disturbance true value purely represents the torsional action of the external environment on the body attitude, and avoid misjudging the rotational inertia of the body as disturbance, thereby greatly improving the accuracy of attitude control. The rotation lumped disturbance true value is closely related to the standard angular acceleration, the standard control moment and the gyroscopic effect term, and this relationship is based on the Euler equation of rigid body dynamics, which describes how the moment changes the angular momentum of the object. Numerically, the rotation lumped disturbance true value is positively correlated with the standard angular acceleration and negatively correlated with the standard control moment, that is, under the action of the same control moment, if the actual angular acceleration of the unmanned aerial vehicle abnormally changes and cannot be explained by the gyroscopic effect, the disturbance true value will increase significantly in value, indicating that there is an external airflow that is exerting additional torsional moment on the body; Since the rotor unmanned aerial vehicle belongs to an under-actuated system, there is a strong coupling relationship between its translation and rotation, and it is easily disturbed by external airflow. In order to achieve high-precision anti-disturbance control, based on the Newton-Euler equation, the translation channel constraint describing the relationship between force and linear motion, and the rotation channel constraint describing the relationship between moment and angular motion are established in the normalized space. Through these two dynamic constraint equations, the difference between the actual motion state and the theoretical model is explicitly defined as the lumped disturbance true value, thereby providing a physical benchmark for the subsequent observer design. In particular, in the rotation channel, by explicitly introducing and separating the gyroscopic effect term, the coupling between external moment disturbance and the rotational inertia of the body is effectively decoupled, ensuring that the defined lumped disturbance true value purely reflects the torsional action of the external environment on the body. It should be noted that although the dynamic equation provides an equation relationship for directly calculating the disturbance true value based on acceleration, mass and thrust in mathematical form, in the actual engineering application of the embodiment, the true value is not calculated by directly obtaining the acceleration through differentiation operation on the speed signal. This is because the differentiation operation has a significant high-pass filtering characteristic in signal processing, which greatly amplifies the high-frequency noise in the sensor signal, resulting in a significant fluctuation in the directly calculated disturbance value and making it impossible to be used for control feedback. Therefore, the core significance of the method is to define the approximation target of the observer, and the specific numerical acquisition will be realized by the subsequent nonlinear disturbance observer with integral filtering characteristics, thereby ensuring the accuracy of the physical model while effectively avoiding the influence of differential noise on the stability of the control system.

[0023] Step 3: Combining the heterogeneous dynamic constraint equations, introduce auxiliary variables to transform the differential operation on the motion state into an integral operation, design a dual-channel nonlinear disturbance observer, and calculate the lumped disturbance estimate that approximates the true value of the lumped disturbance in real time through nonlinear iteration; In this embodiment, for the translational interference observer; Based on the auxiliary variable transformation method in the design criteria for nonlinear disturbance observers in control theory, the first... The translational auxiliary variable of the nth node in the translational disturbance estimation calculation is defined as the nth node. The translational auxiliary variable of the nth node and the nth node The estimated translational lumped disturbance of the nth node satisfies a linear transformation relationship, specifically the nth node. The translational auxiliary variable of the nth node is equal to the nth node. The translational lumped disturbance estimate of the nth node minus the translational correction term, where the translational correction term is the preset translational observer gain matrix, the nth node, and the nth node, respectively. The product of the standard mass and standard velocity vector of each node: No. The translational auxiliary variable of the nth node and the nth node The linear relationship between the estimated translational lumped disturbance values ​​of each node is as follows: In the formula, The first one represents the construction Translational auxiliary variables of each node, Indicates the first Estimates of translational lumped disturbance at each node This represents the preset translational observer gain matrix; The first The translational auxiliary variable of the nth node and the nth node The linear relationship between the estimated translational lumped disturbance values ​​of the nodes is differentiated with respect to a preset unit time, and the nth node is set according to the negative feedback mechanism. The rate of change of the estimated translational lumped disturbance of the nth node and the 1st node The true value of the translational lumped disturbance of the nth node and the nth node The difference in the estimated translational lumped disturbance of the nth node is proportional to the difference in the estimated values ​​of the nth node. Combined with the dynamic constraint equations of the translational channel, the nth node is calculated. The sum of the standard thrust vector, standard gravity vector, translational auxiliary variable from the previous moment, and translational correction term for each node is then multiplied by the negative value of a preset translational observer gain matrix to obtain the result. The rate of change of the translational auxiliary variable of each node within a preset unit of time; No. The expression for calculating the rate of change of the translational auxiliary variable of each node within a preset unit time is as follows: wherein, represents the rate of change of the translational auxiliary variable of the i-th node in a preset unit time, represents the translational auxiliary variable of the i-th node at the previous time of the observer output; represents the translational auxiliary variable of the i-th node at the previous time of the observer output; combining the rate of change of the translational auxiliary variable of the i-th node in a preset unit time, the translational auxiliary variable of the i-th node at the current time is obtained through numerical integration iteration, and substituted into the linear relationship between the translational auxiliary variable of the i-th node at the current time and the translational lumped disturbance estimation value of the i-th node to obtain the translational lumped disturbance estimation value of the i-th node at the current time; By time derivation on both ends of the above-defined translational auxiliary variable linear relationship and by combining the translational channel dynamics constraint equation constructed in step two, the known terms in the equation, i.e., the standard thrust vector, the standard gravity vector and the disturbance term to be estimated, are used to replace and algebraically eliminate the original standard acceleration vector, thereby deriving the new formula which only depends on the standard thrust vector, the standard gravity vector, the standard velocity vector and the translational auxiliary variable itself. Only the numerical integration of the rate of change of the translational auxiliary variable is needed to update the translational auxiliary variable in real time, and then the translational lumped disturbance estimation value is inversely solved in combination with the standard velocity vector. This method mathematically converts the differential observation sensitive to high-frequency noise into the integral observation with low-pass filtering characteristics, thereby realizing high signal-to-noise ratio disturbance estimation under the condition of no acceleration sensor.

[0024] For the rotational disturbance observer, the principle is the same as that of the translational disturbance observer; constructing the rotational auxiliary variable of the i-th node in the translational disturbance estimation operation, defining that the rotational auxiliary variable of the i-th node and the rotational lumped disturbance estimation value of the i-th node satisfy a linear transformation relationship, specifically that the rotational auxiliary variable of the i-th node is equal to the rotational lumped disturbance estimation value of the i-th node minus a rotational correction term, wherein the rotational correction term is the product of a preset rotational observer gain matrix, the standard rotational inertia matrix of the i-th node and the standard angular velocity vector; The linear relationship between the rotational auxiliary variable of the i-th node and the rotational lumped disturbance estimation value of the i-th node is as follows: ​​​​​​​​​​​​​​ wherein, represents a rotation auxiliary variable of a first node of the structure, represents a rotation lumped disturbance estimation value of a first node, represents a preset rotation observer gain matrix; deriving the linear relationship between the rotation auxiliary variable of the first node and the rotation lumped disturbance estimation value of the first node with respect to a preset unit time, and combining a dynamic constraint equation of a rotation channel, a standard control torque vector of the first node, the rotation auxiliary variable of the last time, and a vector sum of the rotation correction term are calculated, and a gyro effect torque term is subtracted, and a final difference vector is multiplied by a negative value of the preset rotation observer gain matrix to obtain a change rate of the rotation auxiliary variable of the first node in the preset unit time; The change rate of the rotation auxiliary variable of the first node in the preset unit time is calculated as follows: wherein, represents a change rate of the rotation auxiliary variable of the first node in the preset unit time, represents the rotation auxiliary variable of the first node of the last time of the observer output; combining the change rate of the rotation auxiliary variable of the first node in the preset unit time, the rotation auxiliary variable of the first node of the current time is obtained through numerical integration iteration, and the rotation auxiliary variable of the first node is substituted into the linear relationship between the rotation auxiliary variable of the first node and the rotation lumped disturbance estimation value of the first node to obtain the rotation lumped disturbance estimation value of the first

[0025] The acquisition of the rotation lumped disturbance estimation value is also based on the above algebraic elimination logic, which avoids direct differentiation of the standard angular velocity vector. The formula fully reproduces the physical characteristics of the rotating rigid body in the observer dynamics model by explicitly introducing the gyroscopic effect torque term. The observer automatically approximates the external disturbance torque by monitoring the residual between the applied standard control torque vector and the actual standard angular velocity vector evolution trend in real time using integral feedback mechanism. Since the standard angular acceleration vector term has been replaced by the rotation channel dynamics constraint equation in the algebraic derivation process, the calculation process only requires the standard angular velocity vector and the standard control torque vector as inputs. This design not only solves the noise amplification problem caused by direct differentiation, but also ensures that when the unmanned aerial vehicle performs rapid maneuvers, the observer can automatically deduct the influence of its own inertial torque, outputting a pure external aerodynamic disturbance estimation value to provide accurate feedforward signals for subsequent attitude disturbance rejection control. The method ingeniously converts the differential operation of velocity and angular velocity in the dynamics equation into integral operation of auxiliary variables by introducing auxiliary variable technology. The technical effect of this processing method is that it is mathematically equivalent to low-pass filtering the signal, thereby effectively avoiding the noise amplification effect caused by directly differentiating the sensor velocity signal. Compared to calculating the translational lumped disturbance true value and the rotational lumped disturbance true value directly using Newton's second law and the rigid body dynamics Euler equation, the nonlinear disturbance observer designed by the method can greatly improve the signal-to-noise ratio and smoothness of the translational lumped disturbance estimation value and the rotational lumped disturbance estimation value while ensuring response speed, providing feedforward compensation signals for the stable operation of the subsequent controller. The initial values of the translational auxiliary variable and the rotational auxiliary variable are set by using the zero initial disturbance assumption method, specifically, when the unmanned aerial vehicle starts, the translational lumped disturbance estimation value and the rotational lumped disturbance estimation value are set to zero vectors, and the initial values of the translational auxiliary variable and the rotational auxiliary variable are calculated in reverse according to the set linear relationship.

[0026] Step 4: Construct a full-state coupled dynamic event triggering mechanism and calculate a dynamic triggering threshold, which is negatively related to the modulus of the lumped disturbance estimation value and contains a minimum interval protection term that decays exponentially over time. When the state error exceeds the dynamic triggering threshold, communication is triggered, and the standard motion state of the node is sent to the neighbor nodes and a request for obtaining the standard motion state is sent to the neighbor nodes; In this embodiment, the modulus of the translational lumped disturbance estimation value and the rotational lumped disturbance estimation value of the first node is calculated to quantify the external disturbance intensity at the current time, and the difference between the current time and the last triggering communication time is calculated as the elapsed time. A full-state coupled dynamic event triggering mechanism is constructed, and the dynamic triggering threshold at the current time is calculated. The calculation logic of the dynamic triggering threshold is as follows: First, the interference adaptive component is calculated, which is used to adaptively adjust the threshold according to the change of the environment, the value is negatively correlated with the intensity of external interference, and is specifically obtained by dividing the preset basic error tolerance constant by the interference intensity weighted sum, which is composed of the weighted value of the translational set total interference estimation value modulus, the weighted value of the rotational set total interference estimation value modulus, and the preset constant term accumulation; Secondly, the minimum interval protection term is calculated, which is used to prevent continuous triggering in a short time, the value is exponentially decayed with the elapse of time, and is specifically obtained by multiplying the preset protection term amplitude by the time decay factor, and the time decay factor is negatively exponentially operated based on the elapse of time, so that the component reaches the maximum value at the moment after the communication triggering and gradually tends to zero over time; Finally, the threshold truncation judgment is executed, the interference adaptive component and the interval protection component are added to obtain the preliminary calculation threshold, the preliminary calculation threshold is compared with the preset dynamic trigger threshold lower limit, and the larger value of the two is selected as the dynamic trigger threshold of the current time; The design principle of the dynamic trigger threshold of the current time is as follows: In the formula, represents the dynamic trigger threshold of the current time t, represents the preset basic error tolerance constant, represents the translational interference sensitive coefficient preset by the worker, represents the modulus of the translational set total interference estimation value, represents the rotational interference sensitive coefficient preset by the worker, represents the modulus of the rotational set total interference estimation value, represents the Zeno protection term amplitude preset by the worker, represents the natural constant, represents the time decay rate coefficient preset by the worker, represents the current time, represents the time of the last triggered communication, which is automatically updated after each triggered communication event, represents the dynamic trigger threshold lower limit; The dynamic trigger threshold, as a quantitative criterion for determining whether to perform a communication operation, represents the maximum state deviation limit that can be tolerated at the current time. The calculation result of the dynamic trigger threshold is directly determined by the module length of the translational and rotational collective interference estimates, the elapsed time since the last communication, and the basic error tolerance. This design achieves on-demand allocation of communication strategies: when external wind resistance or torque interference increases, causing the interference module length to increase, the dynamic trigger threshold is negatively related to it and automatically decreases, thereby increasing the sensitivity to counteract interference through high-frequency communication, and vice versa in smooth flight to reduce sensitivity to save bandwidth. At the same time, the minimum interval protection term included in the formula makes the dynamic trigger threshold reach a maximum value at the moment of each just triggered communication. This positive correlation decay with time characteristic constructs a forced minimum event interval mechanism, effectively avoiding the possibility of Zeno phenomenon from a mathematical logic point of view, avoiding communication congestion and exhausting of computing resources caused by calculation noise or instantaneous error fluctuations.

[0027] Based on the error truncation principle, the module length of the deviation between the current standard motion state and the expected motion state is calculated using the Euclidean distance, and the comprehensive state error at the current time is obtained. The comprehensive state error at the current time is compared with the dynamic trigger threshold at the current time. If the comprehensive state error at the current time is less than the dynamic trigger threshold at the current time, communication is not triggered. If the comprehensive state error at the current time is greater than or equal to the dynamic trigger threshold at the current time, communication is triggered, and the standard motion state of the node is sent to the neighbor node and a request for obtaining the standard motion state is sent to the neighbor node. The expected motion state is a theoretical state at the current time calculated based on a preset space-time trajectory function, or a theoretical state at the current time derived from a reference state sent by a leader node to other nodes. The comprehensive state error specifically reflects the degree to which the actual flight state of the unmanned aerial vehicle deviates from the predetermined trajectory or the leader following target, and its meaning covers the position tracking deviation and the attitude holding deviation. The value of the comprehensive state error is determined by the difference between the standard motion state and the expected motion state, and is positively related to the deviation degree of the two. When the unmanned aerial vehicle deviates from the flight route or the flight rhythm due to interference, the module length of the difference between the two increases, and once the error exceeds the dynamic threshold determined by the current interference intensity, it is determined to trigger the event, and data broadcast is immediately performed and updated to request the neighbor nodes to adjust cooperatively. This trigger mechanism based on trajectory tracking error rather than simple position change quantity effectively eliminates redundant communication caused by normal maneuvering in consistent formation flight, and only occupies the channel when unexpected deviation occurs.

[0028] Step 5: Based on the collective interference estimate, the motion state of the neighbor node with communication random delay is interfered and corrected to obtain a predicted neighbor motion state. In this embodiment, when the first When a node receives the state data packet sent by a neighbor node, the random communication time delay is obtained by calculating the difference between the time stamp in the data packet and the current time, and the standard random communication time delay is obtained by standardizing the random communication time delay; For the translational state time delay compensation prediction of the neighbor node of the first node, the predicted neighbor standard position vector after interference compensation is calculated based on the kinematic Taylor expansion principle, in combination with the translational collective interference estimation value sent by the neighbor node and the dynamic constraint equation of the translational channel. The specific calculation logic is as follows: taking the standard position vector in the state data packet sent by the neighbor node as the reference, a linear displacement generated by the standard velocity vector of the neighbor node within the standard random communication time delay is first superimposed, and then a second-order dynamic correction displacement generated by the resultant force is superimposed. The calculation method of the second-order dynamic correction displacement is as follows: the vector sum of the standard thrust vector of the neighbor node and the translational collective interference estimation value is calculated, and then divided by the standard mass of the neighbor node to obtain the equivalent acceleration, and then the equivalent acceleration is multiplied by half of the square of the standard random communication time delay to obtain the predicted neighbor standard position vector. The calculation principle of the predicted neighbor standard position vector is as follows: In the formula, represents the neighbor node, represents the predicted standard position vector of the neighbor node after interference compensation, represents the standard position vector of the neighbor node , represents the standard velocity vector of the neighbor node , represents the standard random communication time delay between the first node and the neighbor node , represents the standard mass of the neighbor node , represents the standard thrust vector of the neighbor node , represents the translational collective interference estimation value of the neighbor node ; The predicted neighbor standard position vector specifically reflects the true physical position of neighbor nodes at the current moment. It means extrapolating the received positions from past moments to the current moment based on physical laws. Obtaining this prediction eliminates time deviations caused by random delays, enabling the local controller to perform collaborative calculations based on the real-time state of neighbors, thereby avoiding formation collisions or formation divergence caused by information asynchrony. The predicted neighbor standard position vector mainly depends on the historical state, delay duration, and current force conditions of the neighbor nodes in the data packets they send. This is based on the Taylor expansion principle of kinematics: position updates depend not only on the linear extrapolation of velocity but also on the second-order correction of acceleration. Acceleration is determined by both thrust and disturbance force. In particular, the introduction of translational lumped disturbance estimates allows the prediction model to fully consider the nonlinear effects of environmental factors such as wind resistance on the neighbor's trajectory during the delay interval. Numerically, the predicted location vector is positively correlated with the delay duration and also positively correlated with the estimated translational lumped interference. That is, if the neighbor is in a downwind state, even within a small gap of communication interruption, its actual location will be farther than that calculated based solely on speed. This formula can accurately capture this dynamic deviation.

[0029] For the The rotation state delay compensation prediction of each node's neighbor nodes is combined with the rotation lumped interference estimate sent by the neighbor nodes and the dynamic constraint equation of the rotation channel. Based on the Taylor expansion principle of rotational dynamics, the predicted standard attitude angle vector of the neighbor nodes after interference compensation is calculated. The specific calculation logic is as follows: taking the standard attitude angle vector in the state data packet sent by the neighbor nodes as the reference, firstly, the angular displacement generated by the standard angular velocity vector of the neighbor nodes within the standard random communication delay is superimposed, and then the second-order angular dynamic correction displacement generated by the resultant torque is superimposed. The calculation method of the second-order angular dynamic correction displacement is as follows: calculate the vector sum of the standard control torque vector of the neighbor nodes and the rotation lumped interference estimate, use the inverse matrix of the standard rotational inertia matrix of the neighbor nodes to transform the vector sum to obtain the equivalent angular acceleration, and then multiply the equivalent angular acceleration by half of the square of the standard random communication delay to obtain the predicted standard attitude angle vector of the neighbor nodes. The principle behind calculating the predicted standard attitude angle vector of neighbors is as follows: In the formula, Indicates neighboring nodes after interference compensation The predicted neighbor standard pose angle vector, Representing neighboring nodes The standard attitude angle vector in the sent status data packet, Representing neighboring nodes The standard angular velocity vector, Representing neighboring nodes The standard moment of inertia matrix, a standard thrust vector of the neighbor node a standard thrust vector of the neighbor node a standard thrust vector of the neighbor node a standard thrust vector of the neighbor node The predicted neighbor standard attitude angle vector accurately restores the body attitude of the neighbor node at the current time, and its technical effect lies in that it provides a non-lagged reference benchmark for attitude cooperative control. The predicted value is also based on the second-order expansion of rotational dynamics, and the addition of the rotational lumped disturbance estimation value ensures that the torsion effect of air flow disturbance on the neighbor attitude under random time delay is included in the prediction calculation. The rotational lumped disturbance estimation value is positively correlated with the predicted attitude angle vector, which means that if the neighbor is disturbed by a sudden external moment, the prediction algorithm can calculate the attitude deviation caused by the disturbance accumulated in the time delay period at the moment when the data packet arrives, thereby guiding the host to make accurate avoidance or cooperative response.

[0030] Step 6: Construct a cascade optimal disturbance rejection controller, calculate the position and attitude cooperative error of the node, input the position and attitude cooperative error of the node into the offline trained stable manifold neural network to directly infer the optimal cooperative attitude vector, solve the optimal nominal control law, and introduce the lumped disturbance estimation value for feedforward compensation, generate the final physical control command and execute it; In this embodiment, the comprehensive position error of the current node is calculated in a three-dimensional rectangular coordinate system, and the comprehensive position error of the current node is composed of the cooperative error with the neighbor node and the tracking error of the expected motion state at the current time, wherein the cooperative error is obtained by calculating the difference between the standard position vector of the first node, the predicted standard position vector of the neighbor node and the expected distance vector, and the calculated difference is weighted, and the tracking error is obtained by calculating the difference between the standard position vector of the first node and the expected motion state at the current time and weighted calculation, the comprehensive position error of the first node is input into the pre-trained position loop stable manifold neural network to obtain the position loop optimal cooperative attitude vector of the first node in the HJB equation, based on the position loop optimal cooperative attitude vector, combining the optimal feedback control rate formula and calculating the difference with the translational set disturbance estimation value, a three-dimensional disturbance rejection virtual force vector is obtained. The core purpose of constructing the cascade optimal disturbance rejection controller is to solve the technical problem that the traditional control method is difficult to balance the calculation real-time, global optimality and disturbance rejection robustness. Although the traditional HJB equation provides a global optimal control solution for nonlinear systems in theory, as a high-dimensional nonlinear partial differential equation, its calculation load is too large to be solved online, which is difficult to be directly applied to high dynamic unmanned aerial vehicle on-board control. Therefore, based on the stable manifold theory of HJB equation, this method adopts a data-driven approximate solution strategy, which uses the off-line trained stable manifold neural network to fit the solution structure of HJB equation. In the online running stage, the neural network can replace the complex equation solving process, and directly map the optimal coordination state vector according to the input error state through forward reasoning. By calculating the comprehensive position error containing the neighbor node collaborative error and the tracking error of the expected motion state at the current time, it is input to the off-line trained position loop stable manifold neural network. The neural network outputs the position loop optimal coordination state vector that can minimize the system cost function within milliseconds, and then calculates the optimal nominal virtual force. On this basis, the controller introduces the translational collective disturbance estimation value output by step three for feedforward compensation, that is, the disturbance value is superimposed on the optimal nominal force in the opposite direction. The technical effect of this combination strategy is that the unmanned aerial vehicle can not only plan the flight trajectory with the least power consumption and the fastest convergence, but also maintain the rigidity of the trajectory when encountering sudden airflow or model parameter disturbance, realizing the fusion of optimal control and active disturbance rejection.

[0031] The module of the three-dimensional disturbance rejection virtual force vector is calculated to obtain the final total thrust command of the unmanned aerial vehicle rotor. According to the three-dimensional disturbance rejection virtual force vector and the preset reference yaw angle, the expected attitude angle vector of the first node is obtained by combining the backstepping method control. The difference between the standard attitude vector in the standard motion state of the first node and the expected attitude angle vector is calculated to obtain the attitude tracking deviation vector. The comprehensive position error of the first node is input into the pre-trained attitude loop stable manifold neural network to obtain the attitude loop optimal coordination state vector of the first node in the HJB equation. Based on the attitude loop optimal coordination state vector, the final disturbance torque command is obtained by combining the optimal feedback control rate formula and calculating the difference with the rotational collective disturbance estimation value. In view that the quadrotor unmanned aerial vehicle belongs to an under-actuated system and cannot directly generate horizontal thrust, this step decomposes the disturbance-resistant virtual resultant force calculated by the outer loop into total thrust instruction and expected attitude angle by using a vector mapping function. Subsequently, in the inner loop attitude control, a neural network is also used to infer the optimal moment and feedforward compensation of the rotational disturbance. By inputting the attitude tracking error into the neural network of the attitude loop to obtain the optimal moment and subtracting the estimated value of the rotational lumped disturbance, the system can resist the overturning risk caused by the aerodynamic yawing moment or the center of gravity offset in real time. Finally, the generated final physical control instruction containing the total thrust and three-axis moment directly drives the motor speed adjustment, so that the unmanned aerial vehicle formation can still maintain a close cooperative formation and accurately track the preset space-time trajectory in a complex high-dynamic environment, realizing closed-loop control from the decision algorithm to physical execution.

[0032] The first node adjusts the motion state according to the final total thrust instruction and the final disturbance-resistant moment instruction. The first node adjusts the motion state according to the final total thrust instruction and the final disturbance-resistant moment instruction.

[0033] Referring to Figure 2 , the application further provides a random delay unmanned aerial vehicle formation control system based on an event triggering mechanism, which is used to implement the unmanned aerial vehicle formation control method and comprises: A data acquisition module is configured to establish a global inertial coordinate system, acquire original motion states of each node in the formation cluster in real time, and perform standardization processing to obtain standard motion states, wherein the original motion states include position, velocity, rotation angle and angular velocity. An interference definition module is configured to establish physical constraints by using Newton-Euler equations according to the standard motion states, construct heterogeneous dynamics constraint equations of the translational channel and the rotational channel, and define a lumped disturbance true value. An interference analysis module is configured to combine the heterogeneous dynamics constraint equations, design a double-channel nonlinear finite-time disturbance observer, introduce auxiliary variables to convert differential operation on the motion states into integral operation, and calculate a lumped disturbance estimated value approximating the lumped disturbance true value in real time through nonlinear iteration. An event determination module is configured to construct a full-state coupled dynamic event triggering mechanism and calculate a dynamic triggering threshold, wherein the dynamic triggering threshold is negatively correlated with the length of the lumped disturbance estimated value and contains a minimum interval protection term that exponentially decays over time, and communication is triggered when the state error exceeds the dynamic triggering threshold. A communication module is configured to correct the motion states of the neighbor nodes with communication random delay based on the lumped disturbance estimated value to obtain predicted neighbor motion states. The formation correction module is used for constructing a cascade optimal disturbance rejection controller, calculating the position and attitude cooperative error of the node, inputting the position and attitude cooperative error of the node into an offline trained stable manifold neural network to directly infer an optimal cooperative attitude vector, solving an optimal nominal control law, introducing a lumped disturbance estimation value for feedforward compensation, generating a final physical control instruction and executing the same.

[0034] The above formulas are dimensionless values calculated, the formulas are obtained by collecting a large amount of data to simulate a formula of the nearest real situation, and the preset parameters in the formulas are set by a person skilled in the art according to actual conditions.

[0035] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. Those skilled in the art can realize that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software methods depends on the specific application and design constraints of the technical solutions.

[0036] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, which can be located in one place or distributed on multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiments according to actual needs.

[0037] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A random time delay UAV formation control method based on an event trigger mechanism, characterized in that, The specific steps include: A global inertial coordinate system is established to acquire the original motion state of each node in the formation cluster in real time and perform per-unit processing to obtain the standard motion state. The original motion state includes position, velocity, rotation angle and angular velocity. Based on the standard motion state, physical constraints are established using the Newton-Euler equations, heterogeneous dynamic constraint equations for translational and rotational channels are constructed, and the true value of lumped disturbance is defined. By combining heterogeneous dynamic constraint equations and introducing auxiliary variables, the differential operation on the motion state is transformed into an integral operation. A dual-channel nonlinear disturbance observer is designed, and the lumped disturbance estimate approximates the true value of the lumped disturbance is calculated through nonlinear iteration. A full-state coupled dynamic event triggering mechanism is constructed, and a dynamic triggering threshold is calculated. The dynamic triggering threshold is negatively correlated with the magnitude of the lumped interference estimate and includes a minimum interval protection term that decays exponentially with time. When the comprehensive state error exceeds the dynamic triggering threshold, communication is triggered, and the standard motion state is sent to the neighboring nodes and a request to obtain the standard motion state is sent to the neighboring nodes. Based on the lumped interference estimate, the motion state of neighboring nodes with random communication delays is corrected for interference to obtain the predicted neighbor motion state. A cascaded optimal disturbance rejection controller is constructed. The position and attitude coordination error of the nodes is calculated. The position and attitude coordination error of the nodes is input into an offline trained stable manifold neural network to directly infer the optimal costate vector. The optimal nominal control law is obtained by solving the problem. The lumped disturbance estimate is introduced for feedforward compensation. The final physical control command is generated and executed.

2. The random time delay UAV formation control method based on event trigger mechanism according to claim 1, characterized in that: Using the ground control base station as the origin, the horizontal plane is set as the XY axis plane, and the vertical plane is set as the XZ axis plane to establish a three-dimensional rectangular coordinate system. Each device in the formation cluster is regarded as a different node. The original motion state of each node in the formation cluster is collected in real time through airborne sensors. The original motion state includes the original position vector, the original velocity vector, the original attitude angle vector, and the original angular velocity vector. The original motion states of each node are normalized according to the standard unit length, standard unit speed, standard unit attitude angle and standard angular velocity units preset by the relevant staff, so as to obtain the standard motion state of each node.

3. The random time delay UAV formation control method based on event trigger mechanism according to claim 2, characterized in that: Based on the standard motion state of each node, physical constraints are established using the Newton-Euler equations. Based on Newton's second law, combined with the standard velocity vector in the standard motion state, the equation relationship between the translational acceleration and the force of the first node is constructed, and the product of the standard mass and the standard acceleration vector of the first node is equal to the sum of the standard thrust vector, the standard gravity vector and the translational collective interference true value, to obtain the dynamic constraint equation of the translational channel. Based on Euler's equations for rigid body dynamics, and combined with the standard angular velocity vector in the standard motion state, an equation is constructed between standard angular acceleration and torque. The product of the standard moment of inertia matrix and the standard angular acceleration vector is equal to the sum of the standard control torque vector minus the gyroscopic effect torque term and the true value of the rotational lumped disturbance. The gyroscopic effect torque term is obtained by performing a cross product operation on the standard angular velocity vector and the product of the standard moment of inertia matrix and the standard angular velocity vector. The dynamic constraint equations of the rotation channel are then obtained.

4. The random time-delay UAV formation control method based on event triggering mechanism according to claim 3, characterized in that: For translational interference observers; Based on the auxiliary variable transformation method in the design criteria for nonlinear disturbance observers in control theory, the first... The translational auxiliary variable of the nth node in the translational disturbance estimation calculation is defined as the nth node. The translational auxiliary variable of the nth node and the nth node The estimated translational lumped disturbance of the nth node satisfies a linear transformation relationship, specifically the nth node. The translational auxiliary variable of the nth node is equal to the nth node. The translational lumped disturbance estimate of the nth node minus the translational correction term, where the translational correction term is the preset translational observer gain matrix, the nth node, and the nth node, respectively. The product of the standard mass and standard velocity vector of each node; The first The translational auxiliary variable of the nth node and the nth node The linear relationship between the estimated translational lumped disturbance values ​​of the nodes is differentiated with respect to a preset unit time, and the nth node is set according to the negative feedback mechanism. The rate of change of the estimated translational lumped disturbance of the nth node and the 1st node The true value of the translational lumped disturbance of the nth node and the nth node The difference in the estimated translational lumped disturbance of the nth node is proportional to the difference in the estimated values ​​of the nth node. Combined with the dynamic constraint equations of the translational channel, the nth node is calculated. The sum of the standard thrust vector, standard gravity vector, translational auxiliary variable from the previous moment, and translational correction term for each node is then multiplied by the negative value of a preset translational observer gain matrix to obtain the result. The rate of change of the translational auxiliary variable of each node within a preset unit of time; Combined with the The rate of change of the translational auxiliary variable at the nth node within a preset unit time is obtained by numerical integration iteration to obtain the nth node at the current time. The translational auxiliary variable of the nth node is substituted into the current time step of the nth node. The translational auxiliary variable of the nth node and the nth node The linear relationship between the estimated translational lumped disturbance values ​​of the n nodes is used to calculate the current time n. Estimates of the translational lumped disturbance of each node; The principle for the rotational disturbance observer is the same as that for the translational disturbance observer. Construct the first The rotational auxiliary variable of the nth node in the translational disturbance estimation calculation is defined as the nth node. The rotation auxiliary variable of the nth node and the nth node The rotation lumped disturbance estimate of the nth node satisfies a linear transformation relationship, specifically the nth node. The rotation auxiliary variable of the nth node is equal to the nth node. The rotation lumped disturbance estimate of the nth node minus the rotation correction term, where the rotation correction term is the preset rotation observer gain matrix, the th node's rotation lumped disturbance estimate, the lumped disturbance estimate of the nth node, the rotation correction term is the preset rotation observer gain matrix ... The product of the standard moment of inertia matrix and the standard angular velocity vector of each node; The first The rotation auxiliary variable of the nth node and the nth node The linear relationship between the estimated lumped disturbance of the rotation at the nth node and the preset unit time is differentiated, and combined with the dynamic constraint equations of the rotation channel, the nth node is calculated. The sum of the standard control torque vector of each node, the rotation auxiliary variable of the previous moment, and the rotation correction term, minus the gyroscopic effect torque term, and then multiplying the final difference vector by the negative value of the preset rotation observer gain matrix, yields the result. The rate of change of the rotation auxiliary variable of each node within a preset unit of time; Combined with the The rate of change of the rotation auxiliary variable of the nth node within a preset unit time is obtained by numerical integration iteration to obtain the nth node at the current time. The rotation auxiliary variable of the nth node is substituted into the nth node. The rotation auxiliary variable of the nth node and the nth node The linear relationship between the estimated lumped disturbance of the rotation of the nth node is used to calculate the current time. Estimated lumped disturbance of rotation at each node.

5. The random time-delay UAV formation control method based on event triggering mechanism according to claim 4, characterized in that: Calculate the first The magnitudes of the translational lumped interference estimates and rotational lumped interference estimates of each node are used to quantify the external interference intensity at the current moment, and the difference between the current time and the last triggered communication time is calculated as the elapsed time. A fully coupled dynamic event triggering mechanism is constructed, and the dynamic triggering threshold at the current time is calculated. The specific calculation logic for the dynamic triggering threshold is as follows: First, the interference adaptation component is calculated. The interference adaptation component is used to adaptively adjust the threshold according to environmental changes. The value is negatively correlated with the intensity of external interference. Specifically, it is obtained by dividing the preset basic error tolerance constant by the weighted sum of interference intensity. The weighted sum of interference intensity is composed of the weighted value of the magnitude of the translational lumped interference estimate, the weighted value of the magnitude of the rotational lumped interference estimate, and the sum of preset constant terms. Secondly, the minimum interval protection term is calculated. The minimum interval protection term is used to prevent continuous triggering in a short period of time. Its value decays exponentially with the passage of time. Specifically, it is obtained by multiplying the preset protection term amplitude by the time decay factor. The time decay factor is calculated by performing a negative exponential operation based on the passage of time, so that the component reaches its maximum value at the moment after the communication is triggered and gradually approaches zero over time. Finally, a threshold truncation determination is performed. The interference adaptation component and the interval protection component are added together to obtain a preliminary calculated threshold. This preliminary calculated threshold is compared with a preset lower limit of the dynamic trigger threshold, and the larger of the two values ​​is selected as the dynamic trigger threshold for the current time.

6. The random time-delay UAV formation control method based on event triggering mechanism according to claim 5, characterized in that: The Euclidean distance is used to calculate the magnitude of the deviation between the standard motion state and the expected motion state at the current time, resulting in the comprehensive state error at the current time. This comprehensive state error is then compared with the dynamic trigger threshold at the current time. If the comprehensive state error at the current time is less than the dynamic trigger threshold, communication is not triggered. If the comprehensive state error at the current time is greater than or equal to the dynamic trigger threshold, communication is triggered, and the standard motion state is sent to neighboring nodes, along with a request to obtain the standard motion state from neighboring nodes. The expected motion state is either the theoretical state at the current time calculated based on a preset spatiotemporal trajectory function, or the theoretical state at the current time derived from the baseline state sent to other nodes by a designated leader node.

7. The random time-delay UAV formation control method based on event triggering mechanism according to claim 1, characterized in that: When the When a node receives a status data packet sent by a neighboring node, it calculates the difference between the timestamp in the data packet and the current time to obtain the random communication delay, and then normalizes it to obtain the standard random communication delay. For the The translational state delay compensation prediction of each node's neighbor nodes is combined with the translational lumped interference estimate sent by the neighbor nodes and the dynamic constraint equations of the translational channel. Based on the kinematic Taylor expansion principle, the predicted neighbor standard position vector after interference compensation is calculated. The specific calculation logic is as follows: taking the standard position vector in the state data packet sent by the neighbor nodes as the reference, firstly, the linear displacement generated by the standard velocity vector of the neighbor nodes within the standard random communication delay is superimposed, and then the second-order dynamic correction displacement generated by the resultant external force is superimposed. The calculation method of the second-order dynamic correction displacement is as follows: calculate the vector sum of the standard thrust vector of the neighbor nodes and the translational lumped interference estimate, divide it by the standard mass of the neighbor nodes to obtain the equivalent acceleration, and then multiply the equivalent acceleration by half of the square of the standard random communication delay to obtain the predicted neighbor standard position vector. For the The rotation state delay compensation prediction of each node's neighbor nodes is combined with the rotation lumped interference estimate sent by the neighbor nodes and the dynamic constraint equation of the rotation channel. Based on the Taylor expansion principle of rotational dynamics, the predicted standard attitude angle vector of the neighbor nodes after interference compensation is calculated. The specific calculation logic is as follows: taking the standard attitude angle vector in the state data packet sent by the neighbor nodes as the reference, firstly, the angular displacement generated by the standard angular velocity vector of the neighbor nodes within the standard random communication delay is superimposed, and then the second-order angular dynamic correction displacement generated by the resultant torque is superimposed. The calculation method of the second-order angular dynamic correction displacement is as follows: calculate the vector sum of the standard control torque vector of the neighbor nodes and the rotation lumped interference estimate, use the inverse matrix of the standard rotational inertia matrix of the neighbor nodes to transform the vector sum to obtain the equivalent angular acceleration, and then multiply the equivalent angular acceleration by half of the square of the standard random communication delay to obtain the predicted standard attitude angle vector of the neighbor nodes.

8. The random time-delay UAV formation control method based on an event-triggered mechanism according to claim 7, characterized in that: The comprehensive position error of the current node is calculated in a three-dimensional Cartesian coordinate system. This comprehensive position error is a weighted sum of the cooperative error with neighboring nodes and the tracking error for the expected motion state at the current moment. The cooperative error is calculated by... The tracking error is obtained by calculating the difference between the standard position vector of the nth node, the predicted standard position vector of the neighboring nodes, and the expected distance vector, and then weighting the calculated differences. The difference between the standard position vector of the nth node and the expected motion state at the current moment is used to calculate the weighted sum. The combined position error of the nth node is input into a pre-trained position loop-stabilized manifold neural network to obtain the nth node's position error. The optimal co-state vector of the position loop in the HJB equation is the standard motion state of each node. Based on the optimal co-state vector of the position loop, combined with the optimal feedback control law formula and the difference between the estimated value of the translational set disturbance, the three-dimensional disturbance-resistant virtual resultant force vector is obtained. The magnitude of the three-dimensional anti-disturbance virtual resultant force vector is calculated to obtain the final total thrust command of the UAV rotor. Based on the three-dimensional anti-disturbance virtual resultant force vector and the preset reference yaw angle, the first thrust is obtained by combining backstepping control. The expected attitude angle vector of the nth node is calculated. The difference between the standard attitude vector and the desired attitude angle vector in the standard motion state of the nth node is used to obtain the attitude tracking deviation vector. The combined position error of the nth node is input into the pre-trained attitude loop stable manifold neural network to obtain the nth node. The optimal co-state vector of the attitude loop in the HJB equation is the standard motion state of each node. Based on the optimal co-state vector of the attitude loop, combined with the optimal feedback control law formula and the difference between the estimated value of the rotation set disturbance, the final disturbance rejection torque command is obtained. No. Each node adjusts its motion state according to the final total thrust command and the final disturbance rejection torque command.

9. A random time-delay UAV formation control system based on an event-triggered mechanism, characterized in that: The drone formation control system is used to implement the drone formation control method according to any one of claims 1-8, including: The data acquisition module is used to establish a global inertial coordinate system, acquire the original motion state of each node in the formation cluster in real time, and perform per-unit processing to obtain the standard motion state. The original motion state includes position, velocity, rotation angle and angular velocity. The disturbance definition module is used to establish physical constraints based on the standard motion state using the Newton-Euler equations, construct heterogeneous dynamic constraint equations for translational and rotational channels, and define the true value of lumped disturbances. The disturbance analysis module is used to design a dual-channel nonlinear finite-time disturbance observer by combining heterogeneous dynamic constraint equations. It introduces auxiliary variables to transform the differential operation on the motion state into an integral operation, and calculates the lumped disturbance estimate that approximates the true value of the lumped disturbance in real time through nonlinear iteration. The event determination module is used to construct a fully coupled dynamic event triggering mechanism, calculate the dynamic triggering threshold, which is negatively correlated with the magnitude of the lumped disturbance estimate and includes a minimum interval protection term that decays exponentially over time. Communication is triggered when the state error exceeds the dynamic triggering threshold. The communication module is used to correct the motion state of neighboring nodes with random communication delays based on the lumped interference estimate, so as to obtain the predicted motion state of the neighbors. The formation correction module is used to construct a cascaded optimal disturbance rejection controller. It calculates the position and attitude coordination error of the nodes, inputs the position and attitude coordination error of the nodes into an offline trained stable manifold neural network to directly infer the optimal costate vector, solves to obtain the optimal nominal control law, and introduces the lumped disturbance estimate for feedforward compensation to generate the final physical control command and execute it.

Images