Multi-uav cooperative crossing method and device

Abstract

The application belongs to the technical field of unmanned aerial vehicle cooperative crossing, and relates to a multi-unmanned aerial vehicle cooperative crossing method and device. The method comprises the following steps: acquiring scene information of multi-unmanned aerial vehicle cooperative crossing, obtaining a data set, a model of a crossing frame and a dynamic model of each unmanned aerial vehicle; constructing a model prediction control framework of each unmanned aerial vehicle according to the dynamic model of each unmanned aerial vehicle; acquiring a historical trajectory sequence of each unmanned aerial vehicle, obtaining a trained neural network and embedding the trained neural network into the model prediction control framework to form a distributed model prediction control framework of the multi-unmanned aerial vehicle; and performing path prediction planning on the current unmanned aerial vehicle and other unmanned aerial vehicles according to the distributed model prediction control framework of the current unmanned aerial vehicle, the model of the crossing frame and the dynamic model of the current unmanned aerial vehicle to obtain real-time positions of each unmanned aerial vehicle for multi-unmanned aerial vehicle cooperative crossing. The application can enable multi-unmanned aerial vehicles to realize cooperative crossing without communication with each other.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle cooperative crossing, in particular to a multi-unmanned aerial vehicle cooperative crossing method and device. BACKGROUND

[0002] With the development of science and technology and the progress of technology, the cooperative flight of multi-unmanned aerial vehicle systems has been researched day by day.

[0003] In the prior art, when the multi-unmanned aerial vehicle system performs a cooperative flight task, how to ensure that the unmanned aerial vehicles perform agile, accurate and orderly crossing is a very challenging research problem, especially in the presence of dynamic or static door frames. Multiple unmanned aerial vehicles not only need to efficiently plan the optimal path, but also need to avoid collisions with each other and complete the target task within a limited time.

[0004] However, this problem involves multiple core research directions such as unmanned aerial vehicle dynamics modeling, motion planning, path optimization, obstacle avoidance strategy, and needs to combine real-time optimization algorithms and predictive control methods to achieve robust and efficient trajectory planning. Currently, there is no related technology. SUMMARY

[0005] Therefore, it is necessary to provide a multi-unmanned aerial vehicle cooperative crossing method and device to enable multiple unmanned aerial vehicles to achieve cooperative crossing without communication.

[0006] A multi-unmanned aerial vehicle cooperative crossing method, comprising:

[0007] Obtaining scene information of multi-unmanned aerial vehicle cooperative crossing, collecting information, and generating a data set;

[0008] Modeling the crossing frame according to the scene information of the multi-unmanned aerial vehicle cooperative crossing to obtain a model of the crossing frame;

[0009] Modeling each unmanned aerial vehicle according to the scene information of the multi-unmanned aerial vehicle cooperative crossing to obtain a dynamics model of each unmanned aerial vehicle;

[0010] According to the dynamics model of each unmanned aerial vehicle, a model predictive control framework of each unmanned aerial vehicle is constructed by combining a quadratic programming optimization method;

[0011] The historical trajectory sequence of each unmanned aerial vehicle is acquired, a neural network is constructed, and the neural network is trained by taking a data set as an input of the neural network, so as to obtain a trained neural network; the trained neural network is embedded into a model predictive control framework, so as to form a distributed model predictive control framework of the multiple unmanned aerial vehicles; the current unmanned aerial vehicle and other unmanned aerial vehicles are path planning and prediction according to the distributed model predictive control framework of the current unmanned aerial vehicle, the model of the crossing frame and the dynamic model of the current unmanned aerial vehicle, and the action of the current unmanned aerial vehicle is output, and the real-time positions of all unmanned aerial vehicles are obtained by traversing all unmanned aerial vehicles;

[0012] According to the real-time positions of each unmanned aerial vehicle, the multiple unmanned aerial vehicles are cooperatively crossed.

[0013] In one embodiment, scene information of multiple unmanned aerial vehicles cooperatively crossing is acquired, information is collected, and a data set is generated, including:

[0014] The scene information of multiple unmanned aerial vehicles cooperatively crossing is acquired, and an initial environment state is obtained;

[0015] Initial observation information and initial state information are obtained according to the initial environment state; a corresponding Gaussian policy is constructed for each unmanned aerial vehicle according to the initial observation information and the initial state information; an initial action of each unmanned aerial vehicle is obtained by random sampling in the Gaussian policy of each unmanned aerial vehicle, and a set of candidate actions is formed; the candidate actions are interacted with the environment to obtain a reward index and a passing index; the sample weight is calculated according to the reward index and the passing index, and the Gaussian policy of each unmanned aerial vehicle is updated; new candidate actions are formed according to the updated policy, and the new candidate actions, the initial environment state, the initial observation information and the initial state information are taken as data of a current sampling time;

[0016] The environment state is updated, and the new candidate actions are sent to the environment to drive the system to update, so as to obtain a next environment state, a next observation information and a next state information; a next candidate action is obtained according to the next observation information and the next state information, and the next candidate action, the next environment state, the next observation information and the next state information are taken as data of a next sampling time;

[0017] The environment state is updated in a loop until the sampling period ends, and a data set is generated according to the data of all sampling times.

[0018] In one embodiment, according to the scene information of multiple unmanned aerial vehicles cooperatively crossing, a crossing frame is modeled to obtain a model of the crossing frame, including:

[0019] According to the scene information of multiple unmanned aerial vehicles cooperatively crossing, a state vector of the crossing frame is defined;

[0020] The crossing frame is subjected to force analysis, and the change amount of the swing angle of the crossing frame is obtained according to the state vector of the crossing frame.

[0021] According to the change amount of the swing angle of the crossing frame, an expression of the center position of the crossing frame is obtained;

[0022] According to the expression of the center position of the crossing frame, an expression of the corner position of the crossing frame is obtained;

[0023] According to the expression of the corner position of the crossing frame, the center position of the crossing frame is obtained, and is taken as a model of the crossing frame.

[0024] In an embodiment, according to the scene information of the cooperative crossing of the multiple unmanned aerial vehicles, each unmanned aerial vehicle is modeled to obtain a dynamic model of each unmanned aerial vehicle, including:

[0025] According to the scene information of the cooperative crossing of the multiple unmanned aerial vehicles, a state vector of each unmanned aerial vehicle is obtained;

[0026] According to the state vector of each unmanned aerial vehicle, a control vector of each unmanned aerial vehicle is obtained under the constraint of an objective function;

[0027] According to the control vector of each unmanned aerial vehicle, a linear velocity transformation rate and an attitude change rate of each unmanned aerial vehicle are obtained;

[0028] According to the linear velocity transformation rate and the attitude change rate of each unmanned aerial vehicle, a dynamic model of each unmanned aerial vehicle is obtained by dynamic modeling of each unmanned aerial vehicle.

[0029] In an embodiment, the objective function includes:

[0030] ;

[0031] wherein,

[0032] ;

[0033] ;

[0034] ;

[0035] ;

[0036] In the formula, is an objective function, is an objective proximity term, is a trajectory tracking term, is a control smoothing term, is a cooperative collision avoidance term, is a state at the i th moment, is an expected end state, is a positive definite matrix, is a control vector of the i th unmanned aerial vehicle, is a state vector of the i th unmanned aerial vehicle. time, is the predicted total step number, is the time weighting factor, is the predicted state of the first UAV at the step, is the reference state of the step, is the state error weight matrix, is the control input of the first UAV at the step, is the control input in hover state, is the control weight matrix, is the predicted step number of the other UAV, is the first UAV, is the first UAV, is the collision avoidance gain parameter, is the current position of the UAV, is the position of the first UAV at the step, is a positive number.

[0037] In one embodiment, according to the dynamic model of each UAV, a model predictive control framework of each UAV is constructed in combination with a quadratic programming optimization method, including:

[0038] ;

[0039] wherein, is the dynamic model of the UAV, is the state vector of the UAV, is the position coordinate of the UAV in the inertial coordinate system, is the attitude quaternion corresponding to the UAV, is the linear velocity component of the UAV in the inertial coordinate system, is the control vector of the UAV; is the total thrust in the body coordinate system, is the angular velocity component in the body coordinate system;

[0040] Euler discretization is performed:

[0041] ;

[0042] wherein, is the state of the UAV at time, is the state of the UAV at a state at a time t, for controlling a time step, for controlling an input.

[0043] In one embodiment, a historical trajectory sequence of each UAV is obtained, and a neural network is constructed, including:

[0044] The historical trajectory sequence is obtained, and is mapped to a high-dimensional feature space through a linear embedding layer to obtain trajectory data;

[0045] An encoder is designed, and a multi-head self-attention mechanism and a feedforward network are stacked to model the spatio-temporal dependency in the trajectory data, to obtain an output of the encoder;

[0046] The output of the encoder is taken as an input of a residual enhanced long short-term memory network, and time series modeling is performed;

[0047] A multi-scale causal convolution module is designed to process the result of the time series modeling to adjust fine-tuning movements of the UAV in a short time window, to obtain an initial neural network;

[0048] A multi-objective loss function of the initial neural network is constructed to obtain the neural network.

[0049] In one embodiment, a dataset is taken as an input of the neural network, and the neural network is trained to obtain a trained neural network, including:

[0050] ;

[0051] wherein, is an output of predicting a future step, is a feature fusion function, is a feature extraction function of a Transformer, is a feature extraction function of a long short-term memory network, is a feature extraction function of a CNN, is an input sequence from time 1 to .

[0052] In one embodiment, the trained neural network is embedded into a model predictive control framework to form a distributed model predictive control framework of multiple UAVs, including:

[0053] The trained neural network is embedded into a model predictive control framework to add states and quantities of other UAVs in the model predictive control framework of the current UAV, to obtain a distributed model predictive control framework of the current UAV;

[0054] According to the distributed model predictive control frameworks of all UAVs, a distributed model predictive control framework of multiple UAVs is formed.

[0055] A multi-UAV cooperative crossing device, comprising:

[0056] The first module is used to obtain scene information of multi-UAV collaborative traversal, collect information, and generate a data set;

[0057] The second module is used to model the crossing frame based on the scene information of the multi-UAV collaborative crossing, and obtain the crossing frame model;

[0058] The third module is used to model each UAV based on the scene information of multi-UAV collaborative crossing and obtain the dynamic model of each UAV;

[0059] The fourth module is used to build a model predictive control framework for each UAV based on its dynamic model and combined with quadratic programming optimization method;

[0060] The fifth module is used to obtain the historical trajectory sequence of each drone, construct a neural network, and use the dataset as the input to train the neural network to obtain a trained neural network. The trained neural network is embedded in the model predictive control framework to form a distributed model predictive control framework for multiple drones. Based on the distributed model predictive control framework of the current drone, the model of the traversal box, and the dynamic model of the current drone, the path prediction planning is performed for the current drone and other drones, the action of the current drone is output, and all drones are traversed to obtain the real-time position of each drone.

[0061] The sixth module is used to conduct multi-UAV collaborative crossing based on the real-time position of each UAV.

[0062] The above-mentioned multi-UAV collaborative crossing method and device is a trajectory optimization method based on model predictive control (MPC). Through spatiotemporal decoupling modeling (multi-head attention), dynamic gated fusion (Transformer-LSTM-CNN) and physical constraint injection, it achieves high precision and strong robustness in UAV trajectory prediction, while meeting the dynamic constraint requirements of the actual flight control system, providing a reliable state estimation basis for the autonomous navigation system, enabling multiple UAVs to complete the crossing task safely and efficiently under the constraints of dynamic and static obstacles. It is particularly suitable for the collaborative crossing scenario of multiple UAVs under the constraints of dynamic and static obstacles. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 Schematic diagram of a process of a multi-UAV collaborative traversal method according to an embodiment;

[0064] Figure 2 A network structure diagram of a multi-UAV cooperative traversal method in one embodiment;

[0065] Figure 3 A scene schematic diagram of a multi-unmanned aerial vehicle cooperative crossing method in an embodiment;

[0066] Figure 4 A structural block diagram of a multi-unmanned aerial vehicle cooperative crossing device in an embodiment. DETAILED DESCRIPTION

[0067] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0068] In addition, the descriptions such as "first", "second" and the like in the present application are only for the purpose of description and should not be understood as indicating or implying the relative importance of the technical features indicated or the number of technical features indicated. Therefore, the features limited by "first" and "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "multiple groups" is at least two groups, such as two groups, three groups, etc., unless otherwise explicitly specified.

[0069] In the present application, unless otherwise explicitly specified and limited, the terms "connection", "fixing" and the like should be understood in a broad sense, for example, "fixing" can be fixed connection, or detachable connection, or integral; can be mechanical connection, or electrical connection, or physical connection or wireless communication connection; can be directly connected, or indirectly connected through an intermediate medium, can be the internal communication of two elements or the interaction relationship between two elements, unless otherwise explicitly limited. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0070] In addition, the technical solutions of the various embodiments of the present application can be combined with each other, but it must be based on the fact that those of ordinary skill in the art can realize it, and when the combination of technical solutions appears to be contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, nor is it within the scope of protection claimed by the present application.

[0071] The present application provides a multi-unmanned aerial vehicle cooperative crossing method, as shown in the flowchart, in an embodiment, comprising: Figure 1

[0072] Step 101, acquiring the scene information of multi-unmanned aerial vehicle cooperative crossing, collecting information, and generating a data set.

[0073] Specifically:​

[0074] obtain scene information of the multi-unmanned aerial vehicle cooperative crossing, and obtain an initial environment state;

[0075] obtain initial observation information and initial state information according to the initial environment state; construct a corresponding Gaussian policy for each unmanned aerial vehicle according to the initial observation information and the initial state information; perform random sampling in the Gaussian policy of each unmanned aerial vehicle to obtain an initial action of each unmanned aerial vehicle, and form a set of candidate actions; interact the candidate actions with the environment to obtain a reward index and a passing index; calculate sample weights according to the reward index and the passing index, and update the Gaussian policy of each unmanned aerial vehicle; form new candidate actions according to the updated policy, and take the new candidate actions, the initial environment state, the initial observation information, and the initial state information as data at a current sampling time;

[0076] update the environment state, send the new candidate actions to the environment to drive the system to update, and obtain a next environment state, next observation information, and next state information; obtain next candidate actions according to the next observation information and the next state information, and take the next candidate actions, the next environment state, the next observation information, and the next state information as data at a next sampling time;

[0077] recursively update the environment state until the sampling period ends, and generate a data set according to data at all sampling times.

[0078] More specifically:

[0079] obtain scene information of the multi-unmanned aerial vehicle cooperative crossing, and call a reset function (which is prior art) of the environment to obtain an initial environment state (which can also be regarded as a state of a crossing frame);

[0080] initialize states of all unmanned aerial vehicles according to the initial environment state to obtain initial observation information and initial state information; the initial observation information is a ten-dimensional state vector of each unmanned aerial vehicle, including a position, an attitude (represented by a quaternion), and a linear velocity, and the initial state information is a four-dimensional action vector of each unmanned aerial vehicle, including a thrust and three angular velocities; in addition, by obtaining upper and lower bound information of an action space (i.e., a preset action boundary), the generated action can be clipped to ensure that the generated action is within a range allowed by physics and an actuator;

[0081] construct a corresponding Gaussian policy for each unmanned aerial vehicle according to the initial observation information and the initial state information, i.e., policy initialization; the action distribution of the policy is input with the initial observation information and the initial state information at a current time, and is clipped to be generated in combination with a lower bound and an upper bound of the action space, and is used to generate candidate actions for subsequent optimization;

[0082] For each UAV, randomly sample from the Gaussian policy within a specified maximum number of iterations (max_wml_iter) to obtain an initial action for each UAV and form a set of candidate actions; interact with the environment multiple sets (n_samples) of candidate actions and execute these actions in the environment to obtain corresponding reward feedback, resulting in a reward indicator and a pass indicator, and evaluate each action sample using the reward result to select the optimal action and update the policy in the subsequent;

[0083] For each UAV, calculate the sample weight (i.e., the reference factor) based on the candidate action and its corresponding reward indicator, normalize the sample weight using the reward to give higher weight to high-reward samples, and update the mean and covariance matrix of the policy using the Reward-Weighted Maximum Likelihood method to make the new policy more inclined to generate high-reward actions when sampling. Make a judgment. If the reward value of the updated policy increases, it means that the pass indicator is met, and the policy update is considered successful. Otherwise, do not update, still use the last optimal policy or heuristic backup solution as the current policy;

[0084] According to the updated policy, sample to form new candidate actions as the optimal action at the current time, and use the new candidate actions (including the action of each UAV), the initial environment state (the state of the crossing frame), the initial observation information (the state vector of the UAV), and the initial state information (the action vector of the UAV) as the data at the current sampling time;

[0085] Update the environment state, send the new candidate action to the environment to drive the system to update, and obtain the next environment state, the next observation information, and the next state information; according to the next observation information and the next state information, obtain the next candidate action, and use the next candidate action, the next environment state, the next observation information, and the next state information as the data at the next sampling time;

[0086] Perform environment stepping (i.e., call the stepping function of the environment, and the specific stepping function is the model of the crossing frame in step 102) to update the environment state (i.e., update the state of the center point of the crossing frame according to the model of the crossing frame) in a loop until the end of the sampling period. Count the number of samples to determine whether the upper limit of sampling has been reached, and generate a dataset based on the data at all sampling times .

[0087] In this step, the collected dataset will be stored in a dedicated dataset object for subsequent offline training and evaluation.

[0088] This step designs a data collection method based on environment interaction combined with weighted maximum likelihood, as shown in Table 1:

[0089] As shown in Table 1, by multiple sampling and online policy updating, the training data set is efficiently constructed to support the policy learning of multi-UAV cooperative control, the state, action and corresponding feedback reward of the UAV in the simulation environment are comprehensively collected, and data support is provided for subsequent policy learning and performance evaluation.

[0090] In step 102, according to the scene information of the multi-UAV cooperative crossing, the crossing frame is modeled to obtain the model of the crossing frame.

[0091] Specifically:

[0092] According to the scene information of the multi-UAV cooperative crossing, the state vector of the crossing frame is defined;

[0093] The force analysis of the crossing frame is performed, and the change amount of the swing angle of the crossing frame is obtained according to the state vector of the crossing frame;

[0094] According to the change amount of the swing angle of the crossing frame, the expression of the center position of the crossing frame is obtained;

[0095] According to the expression of the center position of the crossing frame, the expression of the corner position of the crossing frame is obtained;

[0096] According to the expression of the corner position of the crossing frame, the center position of the crossing frame is obtained, and is taken as the model of the crossing frame.

[0097] More specifically:

[0098] In order to accurately simulate the crossing behavior of the UAV in the dynamic environment, a moving frame structure based on a pendulum model is constructed as a crossing frame, which realizes dynamic change through the rotational motion of its fulcrum, and can effectively simulate the behavior of dynamic obstacles (such as swinging door frames) in the actual scene. The modeling of the crossing frame is described in detail below, including the kinematics of the single pendulum, the geometric structure and its dynamic characteristics;

[0099] According to the scene information of the multi-UAV cooperative crossing, the state vector of the crossing frame is defined:

[0100] ;

[0101] In the formula, is the swing angle of the frame (offset relative to the vertical direction), is the angular velocity of the frame;

[0102] The force analysis of the crossing frame is performed (because it is only subject to gravity, the motion of the frame is described by the single pendulum dynamics equation), and the change amount of the swing angle of the crossing frame is obtained according to the state vector of the crossing frame:

[0103] ;

[0104] wherein, g is the acceleration of gravity, , L is the length of the pendulum, i.e. the distance from the suspension point to the center of the frame, C is the damping coefficient, used to describe the energy loss due to friction or air resistance during the swinging of the frame, m is the mass of the frame;

[0105] According to the change in the swinging angle of the crossing frame, the expression of the center position of the crossing frame is obtained:

[0106] ;

[0107] ;

[0108] ;

[0109] wherein, is the fulcrum position;

[0110] According to the expression of the center position of the crossing frame, the expression of the corner position of the crossing frame is obtained:

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] wherein, is the position of the first corner, is the position of the second corner, is the position of the third corner, is the position of the fourth corner, is the inner ring width of the crossing frame, is the inner ring height of the crossing frame, is the center position;

[0116] According to the expression of the corner position of the crossing frame, the center position of the crossing frame is obtained, and is taken as the model of the crossing frame.

[0117] In this step, the crossing frame is modeled.

[0118] Step 103, according to the scene information of the cooperative crossing of multiple unmanned aerial vehicles, each unmanned aerial vehicle is modeled to obtain the dynamic model of each unmanned aerial vehicle.

[0119] Specifically:

[0120] According to the scene information of the cooperative crossing of the multiple unmanned aerial vehicles, a state vector of each unmanned aerial vehicle is obtained;

[0121] According to the state vector of each unmanned aerial vehicle, a control vector of each unmanned aerial vehicle is obtained under the constraint of a target function;

[0122] According to the control vector of each unmanned aerial vehicle, a linear velocity transformation rate and an attitude change rate of each unmanned aerial vehicle are obtained;

[0123] According to the linear velocity transformation rate and the attitude change rate of each unmanned aerial vehicle, a dynamic model of each unmanned aerial vehicle is obtained by dynamic modeling.

[0124] More specifically:

[0125] To model the unmanned aerial vehicle, the following basic assumptions are made:

[0126] 1) rigid body assumption: the unmanned aerial vehicle is regarded as an ideal rigid body, and its deformation and flexible vibration are ignored;

[0127] 2) uniform mass distribution: the body is symmetrical, and the mass is uniformly distributed at the center;

[0128] 3) air resistance is ignored: the influence of aerodynamic resistance on the unmanned aerial vehicle is not considered (the aerodynamic effect can be added in subsequent modeling);

[0129] 4) instantaneous response of motor: the motor dynamics model is not explicitly modeled, and it is assumed that the motor can respond to the angular velocity control instruction instantaneously.

[0130] To describe the motion of the unmanned aerial vehicle, two coordinate systems are used, one is an inertial coordinate system (ground fixed reference frame): , which is used to describe the absolute position and velocity of the unmanned aerial vehicle, wherein is the axis in the world coordinate system, is the axis in the world coordinate system, is the axis in the world coordinate system; the other is the body coordinate system: , wherein is the axis in the body coordinate system, pointing to the head of the unmanned aerial vehicle, is the axis in the body coordinate system, is the axis in the body coordinate system. The conversion between the two coordinate systems is represented by the rotation matrix or the quaternion .

[0131] According to the scene information of multi-UAV collaborative traversal, the state vector of each UAV is obtained:

[0132] ;

[0133] Where, is the position coordinate of the UAV in the inertial coordinate system, (where, The UAV is in the inertial coordinate system Location coordinates, The UAV is in the inertial coordinate system Location coordinates, The UAV is in the inertial coordinate system Position coordinates), is the attitude quaternion corresponding to the drone (where, The real part of the UAV attitude quaternion is used to represent the body attitude. is the imaginary part of the drone attitude quaternion Quantity, is the imaginary part of the drone attitude quaternion Quantity, is the imaginary part of the drone attitude quaternion Components) (Compared to Euler angles, quaternions can avoid the universal lock problem and have better numerical stability in optimization calculations), satisfying the unit constraint , is the linear velocity component of the UAV in the inertial coordinate system (where, The UAV is in the inertial coordinate system Directional linear velocity component, The UAV is in the inertial coordinate system Directional linear velocity component, The UAV is in the inertial coordinate system Directional linear velocity component);

[0134] According to the state vector of each UAV, under the constraint of the objective function, the control vector of each UAV is obtained:

[0135] ;

[0136] Where, is the total thrust in the body coordinate system (the translational motion of the drone is determined by the external force, among which the total thrust Provided by four rotors, acting on the body axis direction, controlling the lift and vertical acceleration of the drone), is the angular velocity component in the body coordinate system (where In the body coordinate system Directional angular velocity component, is the direction angular velocity component in the body coordinate system, is the direction angular velocity component in the body coordinate system, is the direction angular velocity component in the body coordinate system, is the direction angular velocity component in the body coordinate system), three angular velocity components form angular velocity , used to adjust the attitude of the unmanned aerial vehicle;

[0137] According to the control vector of each unmanned aerial vehicle, the linear velocity transformation rate and the attitude change rate of each unmanned aerial vehicle are obtained;

[0138] Wherein, the translational motion of the unmanned aerial vehicle is determined by external force, wherein the total thrust acts on the body coordinate system, the gravity acts on the inertial coordinate system, then the acceleration of the unmanned aerial vehicle in the inertial coordinate system is:

[0139] ;

[0140] Wherein,

[0141] ;

[0142] The linear velocity transformation rate of the unmanned aerial vehicle is:

[0143] ;

[0144] In the formula, is the rotation matrix, which is calculated by the quaternion, is the total thrust in the body coordinate system, is the gravity, which acts on the inertial coordinate system;

[0145] The attitude change of the unmanned aerial vehicle is controlled by angular velocity , and the derivative of the quaternion is:

[0146] ;

[0147] In the formula, is the derivative of the quaternion;

[0148] The attitude change rate of the unmanned aerial vehicle is obtained from the derivative of the quaternion (How to obtain it belongs to the prior art);

[0149] According to the linear velocity transformation rate and the attitude change rate of each unmanned aerial vehicle, the dynamic model of each unmanned aerial vehicle is obtained:

[0150] ;

[0151] In the formula, is the dynamic model of the unmanned aerial vehicle.

[0152] In the traditional MPC model, the controller solves the optimal control input based on the state prediction of a single system. For the cooperative task of multiple UAVs, in order to realize decentralized and real-time cooperative control, the application proposes a data-driven distributed model predictive control framework. The framework independently constructs a local optimization problem in each UAV, and realizes information coupling and cooperative constraints between multiple UAVs by introducing the state prediction information of neighbor UAVs, so as to simultaneously meet multiple control objectives such as target tracking, input smoothing and collision avoidance. The objective function includes: under the distributed MPC framework, each UAV independently constructs a local optimization problem, and the decision variables include three parts: 1) the state trajectory of itself : describes the state evolution at future time , starting from the current state, satisfying the discrete state transition equation; 2) the control input sequence : contains the control command at future time , and must satisfy the upper and lower limits of the physical input and the actuator, such as the extreme value limit of thrust and angular velocity; 3) neighbor state prediction parameters: in order to realize cooperation and collision avoidance between multiple UAVs, each UAV receives the state information of other UAVs within a limited prediction step (denoted as ) in the local optimization problem. After rearranging and splicing, these states are transmitted as optimization parameters, so that each local problem can consider the influence of the dynamic state of the adjacent UAV on its own planning;

[0153] On the basis of the above variable composition, considering the three indicators of approaching the target, trajectory tracking, control smoothing and cooperative collision avoidance, the objective function of the local optimization problem of each UAV is established as:

[0154] ;

[0155] Among them,

[0156] ;

[0157] ;

[0158] ;

[0159] ;

[0160] In the formula, is the objective function, is the target approach term, which defines the terminal cost, so that the UAV approaches the expected target at the predicted end, is the trajectory tracking term, which defines the trajectory tracking cost, to make the UAV smoothly and accurately pass through the moving door frame, To control the smoothness term (to avoid the instability of execution caused by the dramatic change of control command, the part deviating from the preset reference of control input is punished twice, thus the control smoothness term is introduced), define the control smoothness cost, aiming to suppress the dramatic fluctuation of control input, To cooperative collision avoidance term (to ensure the safe distance between UAVs when running cooperatively, based on the potential field function theory, construct the cooperative collision avoidance term, i.e. collision avoidance cost term, when UAVs are close to each other, the cost rises rapidly, thus automatically driving the system to choose the strategy of keeping safe distance in the optimization process), define the collision avoidance cost, by using the state prediction of nearby UAVs Realize the safe coordination between multiple machines, For the state of the first moment, For the desired end state, For the positive definite matrix, give sufficient punishment to the state error, For the first moment, For the total number of steps predicted, For the time weighting factor, , wherein, Control decay speed, Indicates a key moment when higher precision control is expected to be obtained; For the first UAV, The predicted state of the first step, The reference state of the first step, The state error weight matrix, The control input of the first UAV at the first step, The control input in hovering state (such as ), The control weight matrix to balance tracking performance and control smoothness, , indicates the number of steps considering the influence of neighboring machines in the prediction interval; The first UAV, The first UAV, The collision avoidance gain parameter, The current position of the first UAV, The position of the first UAV at the first step, A positive number to prevent the denominator from being zero.

[0161] In this step, the dynamic model of each UAV is established.

[0162] Step 104 : Based on the dynamic model of each UAV and in combination with the quadratic programming optimization method, a model predictive control framework for each UAV is constructed.

[0163] Specifically:

[0164] A linearized UAV dynamics model is used in conjunction with a quadratic programming (QP) optimization method to construct a Model Predictive Control (MPC) framework. This framework optimizes control inputs within a finite time domain, guiding the UAV along a predetermined trajectory in an optimal manner. This allows for optimal trajectory planning and real-time trajectory tracking.

[0165] In the MPC framework, the discrete-time dynamics model is used for prediction, and the continuous-time dynamics equation of the UAV is:

[0166] ;

[0167] Where, is the dynamic model of the UAV, is the state vector of the UAV, is the control vector of the UAV (as input);

[0168] In the MPC framework, it is discretized using Euler:

[0169] ;

[0170] in, For drones at all times status, For drones at all times status, To control the time step, For control input.

[0171] In this step, the model predictive control framework for each UAV is constructed.

[0172] Step 105: Obtain the historical trajectory sequence of each UAV, construct a neural network, and use the data set as the input of the neural network to train the neural network to obtain a trained neural network; embed the trained neural network into the model predictive control framework to form a distributed model predictive control framework for multiple UAVs; based on the distributed model predictive control framework of the current UAV, the model of the crossing box, and the dynamic model of the current UAV, perform path prediction planning for the current UAV and other UAVs, output the action of the current UAV, and traverse all UAVs to obtain the real-time position of each UAV.

[0173] In particular,

[0174] obtain a historical trajectory sequence, and map the historical trajectory sequence to a high-dimensional feature space through a linear embedding layer to obtain trajectory data; design an encoder, and stack a multi-head self-attention mechanism and a feedforward network to model a spatio-temporal dependency relationship in the trajectory data to obtain an output of the encoder; input the output of the encoder into a residual enhanced long short-term memory network to perform time series modeling; design a multi-scale causal convolution module to process a result of the time series modeling to adjust fine-tuning movements of the unmanned aerial vehicle in a short time window to obtain an initial neural network; construct a multi-objective loss function of the initial neural network to obtain a neural network;

[0175] input a dataset into the neural network to train the neural network to obtain a trained neural network;

[0176] embed the trained neural network into a model predictive control framework to add states and quantities of other unmanned aerial vehicles into a model predictive control framework of a current unmanned aerial vehicle to obtain a distributed model predictive control framework of the current unmanned aerial vehicle; form a distributed model predictive control framework of multiple unmanned aerial vehicles according to distributed model predictive control frameworks of all unmanned aerial vehicles;

[0177] perform path prediction planning on the current unmanned aerial vehicle and other unmanned aerial vehicles according to the distributed model predictive control framework of the current unmanned aerial vehicle, a model of a crossing frame, and a dynamics model of the current unmanned aerial vehicle, output an action of the current unmanned aerial vehicle, and traverse all unmanned aerial vehicles to obtain real-time positions of each unmanned aerial vehicle.

[0178] More specifically,

[0179] obtain a historical trajectory sequence of each unmanned aerial vehicle wherein, is a batch size, is a sequence length, 10 is a dimension, including a position vector , an attitude quaternion and a velocity vector , and the historical trajectory sequence is mapped to a high-dimensional feature space through a linear embedding layer to improve representation ability of original state quantities to obtain trajectory data:

[0180] ;

[0181] wherein, is a learning weight matrix, and the feature is mapped to a hidden dimension , is a bias term;

[0182] After obtaining the high-dimensional embedding, a Transformer encoder is designed to model the spatio-temporal dependencies in the trajectory data, and a multi-head self-attention mechanism and a feedforward neural network (FFN) are stacked in the encoder layer to capture the complex spatio-temporal interactions in the UAV trajectory and model the spatio-temporal dependencies in the trajectory data, so as to obtain the output of the encoder.

[0183] The multi-head self-attention mechanism can enhance the modeling capability of the time dependencies between the UAV states, and is represented as follows:

[0184] ;

[0185] wherein the query matrix Q, the key matrix K, and the value matrix V are obtained by linear transformation of the input state;

[0186] The feedforward network is used for feature transformation and enhancement of the nonlinear representation capability of the model, and is represented as follows:

[0187] ;

[0188] wherein x represents the input of the network structure; represents a nonlinear activation function; represents different weight matrices; represents different bias matrices; The output of the encoder is taken as the input of a residual enhanced long short-term memory network (residual enhanced LSTM), the residual enhanced LSTM is used to process the time sequence inertia characteristics of the UAV trajectory, so as to perform collaborative modeling of the global features and the local memory and realize time sequence modeling; the core calculation formula is as follows:

[0189]

[0190] ; wherein

[0191] respectively represent the activation values of the forget gate, the input gate, and the output gate; is a candidate memory cell, the current context information of the Transformer is introduced to enhance the adaptability to the complex environment; is an updated memory cell, which explicitly records the movement history of the UAV; represents different weight matrices; represents different bias terms; is an activation function; tan h represents a hyperbolic tangent function; is a logical operator; is the i-th element of the input sequence x; is the i-th element of the output sequence y.​​​ The hidden state of the moment; is the output of transformer at time t; For the The final output at the moment represents the current state;

[0192] Because LSTM has a response lag when dealing with local mutations and short-term dynamic changes, it is difficult to accurately model small disturbances. Therefore, to enhance robustness to rapidly changing states, a multi-scale causal convolution module is designed based on time series modeling. The results of time series modeling are processed to explore local patterns and adjust the drone's fine-tuning within a short time window. ReLU activation and batch normalization are used to enhance the response to emergency acceleration and deceleration features, resulting in the following initial neural network:

[0193] ;

[0194] Where, is the convolution kernel width, is the number of input channels, is the convolution kernel is the time offset, is the output channel, is the input channel, Bias term;

[0195] After the structure of the initial neural network is determined, a multi-objective loss function of the initial neural network is constructed to improve the prediction accuracy and dynamic smoothness, and the neural network is obtained;

[0196] Here, the total loss is composed of the prediction error and the physical constraint:

[0197] ;

[0198] Where, is the total loss, is the weight coefficient of the trajectory prediction loss term, is the predicted future time step, is the index of the current time step in the future sequence, For the The true target value of time steps, For the The predicted value of the time step, is the weight coefficient of the smoothing loss term, is the smoothness loss, which is used to constrain the continuity and smoothness of the predicted trajectory;

[0199] The data set is used as the input of the neural network, that is, based on a large amount of historical data collected in advance, the neural network is trained offline to realize efficient online decision-making in the cooperative task of multiple unmanned aerial vehicles. Specifically, an independent strategy network (Actor) is built for each unmanned aerial vehicle, which is used to map high-dimensional observation input to control action output. The training process adopts a supervised learning paradigm, which continuously updates network parameters by minimizing the difference between the predicted action of the strategy network and the actual expert action, thereby improving the generalization performance and robustness of the strategy. In order to facilitate network training, the training data of each unmanned aerial vehicle is loaded in batches. Specifically, for each unmanned aerial vehicle, a custom data set is constructed using preprocessed observation data and action data to ensure that the data dimension meets the training requirements. Subsequently, the DataLoader function provided by PyTorch is used to randomly shuffle (shuffle) and batch the data set. The batch size is specified by the hyperparameter batch size. This operation not only improves data utilization efficiency, but also helps to reduce noise caused by data correlation in the gradient descent process. Each unmanned aerial vehicle is equipped with an independent Actor network, which takes the observation vector (position, attitude, velocity) as input and outputs the control instruction (thrust, angular velocity). The number of hidden layer units, activation function, and learning rate in the network structure are all hyperparameters. The training goal is to make the network output as close as possible to the optimal action collected in the historical record, so that the strategy network can accurately predict the optimal control input of the unmanned aerial vehicle in the current state. During network training, the Mean Squared Error (MSE) is used as the loss function, and the network parameters are updated through the backpropagation algorithm and gradient descent method. In each training batch, the train batch method of the Actor network is used to calculate the loss of the current batch, and the loss is accumulated and recorded. During the entire training process, the system periodically outputs the average loss value of each epoch and saves the network parameters at the current training round, so as to facilitate subsequent strategy initialization and online control.

[0200] The trained neural network is obtained:

[0201] ;

[0202] wherein, is the output of the predicted future step, is a feature fusion function, is a feature extraction function of the Transformer, is a feature extraction function of the Long Short-Term Memory network, is a feature extraction function of the CNN, is the input sequence from time 1 to .

[0203] The trained neural network is embedded into the model predictive control framework to add the states and quantities of other unmanned aerial vehicles to the model predictive control framework of the current unmanned aerial vehicle, to obtain a distributed model predictive control framework of the current unmanned aerial vehicle; and the distributed model predictive control frameworks of all unmanned aerial vehicles are formed to obtain a distributed model predictive control framework of multiple unmanned aerial vehicles;

[0204] According to the distributed model predictive control framework of the current unmanned aerial vehicle, the model of the crossing frame and the dynamic model of the current unmanned aerial vehicle, the path prediction planning is performed on the current unmanned aerial vehicle and other unmanned aerial vehicles, and the action of the current unmanned aerial vehicle is output; the action is performed according to the current time state of the current unmanned aerial vehicle, to obtain the next time state of the current unmanned aerial vehicle; all unmanned aerial vehicles are traversed to obtain the next time state of all unmanned aerial vehicles; in the next time, the action and state of the current unmanned aerial vehicle are still output according to the states of other unmanned aerial vehicles, and the states of all unmanned aerial vehicles are obtained by traversal; all time instants are traversed to obtain the real-time positions of each unmanned aerial vehicle.

[0205] In this step, a dual-mode trajectory predictor is designed to further improve the accuracy and safety of cooperative control. Specifically, when there is insufficient historical state data, a constant speed model is used for simplified kinematic prediction; while in the case of sufficient data, a pre-trained Transformer-LSTM-CNN hybrid model is used to predict the future state of the neighboring unmanned aerial vehicle; this prediction module not only improves the state estimation accuracy of the neighboring unmanned aerial vehicle, but also provides more accurate cooperative constraint parameters for local optimization problems, thereby effectively reducing the potential collision risk; the architecture of the proposed hybrid model is shown in Figure 2 , which includes an input end, a Transformer encoder, an LSTM module, a CNN module, and an output end connected in sequence. The Transformer encoder includes a Multi-Head Attention module, an Add&Norm module, a FeedForward module, and an Add&Norm module connected in sequence. The architecture is composed of three core modules: 1) the Transformer encoder captures global spatiotemporal dependencies; 2) the LSTM network models temporal continuity; 3) the CNN extracts local motion patterns; finally, multi-modal feature integration is achieved through an attention weighted fusion layer, and multi-step prediction results are output; this design optimizes the nonlinear and strongly coupled characteristics of the unmanned aerial vehicle trajectory, and the expression is:

[0206] ;

[0207] wherein, is the output of the prediction of the future step, is a feature fusion function, The feature extraction function for the Transformer, The feature extraction function for the LSTM, The feature extraction function for the CNN, The input sequence from time 1 to is the historical state information.

[0208] In step 106, according to the real-time position of each UAV, the multi-UAV cooperative crossing is performed.

[0209] In this step, how to perform the multi-UAV cooperative crossing according to the real-time position of each UAV belongs to the prior art, and will not be repeated here.

[0210] In this embodiment, as shown in the crossing scene in Figure 3 , the circle represents the starting point, the triangle represents the ending point, represents the length of the pendulum, that is, the distance from the hanging point to the center of the crossing frame, is the inner ring width of the crossing frame, is the inner ring height of the crossing frame.

[0211] The learning-based distributed model predictive control realizes real-time solving and adaptive adjustment of complex dynamic constraints in multi-UAV cooperative tasks by combining distributed MPC with data-driven policy training. This framework not only meets the requirements of real-time performance and safety, but also continuously optimizes online decision-making strategies through offline data learning, providing a practical solution for multi-UAV system control in complex environments.

[0212] The Distributed Model Predictive Control (DMPC) method for multi-UAV cooperation proposed in this application introduces a deep learning-based neighboring UAV trajectory prediction network to effectively avoid collisions between UAVs while ensuring control performance. The overall framework can be summarized as follows: Each UAV needs to obtain or estimate the motion state of neighboring UAVs in the future steps in real time when performing distributed MPC, and these information are used as dynamic obstacles or coupled constraints in the local optimization problem. At the same time, a hybrid Transformer-LSTM-CNN prediction model is used to capture the spatio-temporal dependence between multiple UAVs, output more accurate neighboring UAV trajectory prediction results, and reduce collisions and coupling conflicts.

[0213] The main flow of the framework includes: 1) Multi-UAV state acquisition and local MPC optimization: Each UAV obtains the state observation of itself and the adjacent UAVs, and generates the control sequence of the next time or several times according to the rolling optimization idea of DMPC. DMPC updates the control decision through iterative information exchange, so that better global coordination and convergence can be ensured in the distributed scene. 2) Adjacent UAV future trajectory prediction: In order to accurately estimate the possible position and speed change of the adjacent machine in the next period of time, a deep network is used to predict the motion state of the adjacent machine for a short time. Unlike the traditional simple constant speed / constant acceleration assumption, the application utilizes Transformer to process global time sequence dependence, LSTM to focus on short-term dynamic characteristics, and CNN for local convolution extraction or multi-feature fusion, so that the prediction accuracy and robustness are improved. 3) Coupling constraint and collision avoidance mechanism: The predicted trajectory of the adjacent machine is regarded as a coupling term in the distributed MPC, and the safety distance limit and collision penalty term are added to the optimization objective or cost function to ensure that each UAV will not collide in the future time domain. In addition, to cope with model uncertainty or prediction error, a safety boundary (safety distance scaling) is introduced into the DMPC to further ensure system safety.

[0214] The above multi-UAV cooperative crossing method is a trajectory optimization method based on model predictive control (MPC). Through space-time decoupling modeling (multi-head attention), dynamic gate fusion (Transformer-LSTM-CNN) and physical constraint injection, high precision and strong robustness of UAV trajectory prediction are realized, while meeting the dynamics constraint requirements of the actual flight control system. It provides a reliable state estimation basis for the autonomous navigation system, so that the multi-UAV can safely and efficiently complete the crossing task under the constraints of dynamic and static obstacles. It is especially suitable for the cooperative crossing scene of multi-UAV under the constraints of dynamic and static obstacles.

[0215] To realize the collaborative modeling of global features and local memory, a trajectory prediction and optimization method suitable for multi-UAV dynamic crossing scenarios is designed. The overall framework is based on model predictive control (MPC), and the multi-head self-attention mechanism is introduced to realize the spatio-temporal decoupling modeling to capture the complex interaction relationship between multi-UAV trajectories. At the same time, the global representation ability of the Transformer is combined with the time memory ability of the LSTM, and a multi-scale convolution module is fused to enhance the perception ability of short-term trajectory disturbance, thereby constructing a deep fusion structure with dynamic gating capability. This structure not only can accurately predict the future state of other UAVs, but also can ensure that the prediction results meet the execution conditions of the actual flight control system by introducing physical feasibility constraints (such as kinetic limitations, control boundaries, etc.). Further, the loss function design considers the trajectory prediction error and smoothness requirements, and realizes the common improvement of trajectory accuracy and system stability through multi-objective optimization. Thanks to the above design, the proposed method can realize the cooperative state estimation and high-quality trajectory generation among multi-UAVs in complex environments where dynamic and static obstacles coexist, ensuring a good balance between safety and efficiency, thereby significantly improving the overall performance of the multi-UAV system in the crossing task.

[0216] It should be understood that, although Figure 1 the steps in the flowchart of the method are shown in sequence according to the arrows, these steps are not necessarily executed in sequence according to the arrows. Unless otherwise specified herein, there is no strict order limitation for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 at least part of the steps in the method can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be alternately executed with other steps or at least part of the sub-steps or stages of other steps.

[0217] The application also provides a multi-UAV cooperative crossing device, as shown in Figure 4 in one embodiment, comprising: a first module 401, a second module 402, a third module 403, a fourth module 404, a fifth module 405 and a sixth module 406, wherein:

[0218] The first module 401 is configured to acquire the scene information of the multi-UAV cooperative crossing, collect information, and generate a data set;

[0219] The second module 402 is configured to model the crossing frame according to the scene information of the multi-UAV cooperative crossing, and obtain a model of the crossing frame;

[0220] The third module 403 is configured to model each unmanned aerial vehicle according to scene information of cooperative crossing of the unmanned aerial vehicles, to obtain a dynamic model of each unmanned aerial vehicle.

[0221] The fourth module 404 is configured to construct a model predictive control framework of each unmanned aerial vehicle according to the dynamic model of each unmanned aerial vehicle, in combination with a quadratic programming optimization method.

[0222] The fifth module 405 is configured to obtain a historical trajectory sequence of each unmanned aerial vehicle, to construct a neural network, to train the neural network by taking a data set as an input of the neural network, to obtain a trained neural network, to embed the trained neural network into the model predictive control framework, to form a distributed model predictive control framework of the unmanned aerial vehicles, to perform path prediction and planning on the current unmanned aerial vehicle and other unmanned aerial vehicles according to the distributed model predictive control framework of the current unmanned aerial vehicle, the model of the crossing frame and the dynamic model of the current unmanned aerial vehicle, to output an action of the current unmanned aerial vehicle, and to obtain real-time positions of each unmanned aerial vehicle by traversing all the unmanned aerial vehicles.

[0223] The sixth module 406 is configured to perform cooperative crossing of the unmanned aerial vehicles according to the real-time positions of each unmanned aerial vehicle.

[0224] Specific limitations of the device for cooperative crossing of the unmanned aerial vehicles can be seen from the limitations of the method for cooperative crossing of the unmanned aerial vehicles, which will not be described herein. Each module in the device can be realized by software, hardware and combinations thereof. Each module can be embedded in or independent of a processor in a computer device in a hardware form, or can be stored in a memory in a computer device in a software form, so as to be called and executed by a processor to perform operations corresponding to each module.

[0225] The contents not described in detail in the specification belong to the prior art known to those skilled in the art.

[0226] Each technical feature of the above embodiments can be combined arbitrarily. To make the description concise, all possible combinations of each technical feature in the above embodiments are not described, however, as long as the combinations of the technical features do not exist, they should be considered as the scope of the present application.

[0227] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the present application. It should be noted that, for those skilled in the art, without departing from the concept of the present application, some modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended application documents.

Claims

1. A multi-UAV collaborative crossing method, characterized in that: include: Obtain scene information of multi-UAV collaborative traversal, collect information, and generate data sets; According to the scene information of multi-UAV collaborative crossing, the crossing frame is modeled to obtain the crossing frame model; Based on the scene information of multi-UAV collaborative crossing, each UAV is modeled to obtain the dynamic model of each UAV; Based on the dynamic model of each UAV and combined with the quadratic programming optimization method, a model predictive control framework for each UAV is constructed; The historical trajectory sequence of each drone is obtained, and a neural network is constructed. The neural network is trained using the dataset as input to obtain a trained neural network. The trained neural network is embedded in the model predictive control framework to form a distributed model predictive control framework for multiple drones. Based on the distributed model predictive control framework of the current drone, the model of the traversal box, and the dynamic model of the current drone, path prediction and planning are performed for the current drone and other drones, the current drone's actions are output, and all drones are traversed to obtain the real-time position of each drone. Conduct multi-UAV collaborative crossing based on the real-time position of each UAV; Based on the scene information of multi-UAV collaborative crossing, each UAV is modeled to obtain the dynamic model of each UAV, including: According to the scene information of multi-UAV collaborative traversal, the state vector of each UAV is obtained; According to the state vector of each UAV, under the constraint of the objective function, the control vector of each UAV is obtained; According to the control vector of each UAV, the linear velocity change rate and attitude change rate of each UAV are obtained; According to the linear velocity change rate and attitude change rate of each UAV, dynamic modeling is performed on each UAV to obtain the dynamic model of each UAV; The objective function includes: ； in, ； ； ； ； Where, is the objective function, is the target approach term, is the trajectory tracking term, To control the smoothing term, is the collaborative collision avoidance term, For the The state of the moment, is the desired end state, is a positive definite matrix, For the time, To predict the total number of steps, is the time weighting factor, For the The first drone The predicted state of the step, For the The reference state of the step, is the state error weight matrix, For the The drone The control input of the step, is the control input in the hover state, is the control weight matrix, is the predicted number of steps of other drones, For the drones, For the drones, is the collision avoidance gain parameter, is the current position of the drone, For the The drone The position of the step, Is a positive number.

2. The multi-UAV cooperative crossing method according to claim 1, characterized in that: Obtain scene information of multi-UAV collaborative traversal, collect information, and generate a data set, including: Obtain scene information of multi-UAV collaborative traversal and obtain the initial environment state; Initial observation information and initial state information are obtained based on the initial environment state; a corresponding Gaussian strategy is constructed for each drone based on the initial observation information and initial state information; random sampling is performed in each drone's Gaussian strategy to obtain each drone's initial action and form a set of candidate actions; the candidate actions interact with the environment to obtain reward indicators and pass indicators; based on the reward indicators and pass indicators, the sample weights are calculated and the Gaussian strategy of each drone is updated; new candidate actions are formed based on the updated strategy, and the new candidate actions, initial environment state, initial observation information, and initial state information are used as data at the current sampling time; Update the environment state and send the new candidate action to the environment to drive the system update, obtain the next environment state, the next observation information, and the next state information; obtain the next candidate action based on the next observation information and the next state information, and use the next candidate action, the next environment state, the next observation information, and the next state information as the data for the next sampling time; The environmental status is updated cyclically until the end of the sampling period, and a data set is generated based on the data at all sampling moments.

3. A multi-UAV cooperative crossing method according to claim 1 or 2, characterized in that: According to the scene information of multi-UAV collaborative crossing, the crossing frame is modeled to obtain the crossing frame model, including: Based on the scene information of multi-UAV collaborative crossing, the state vector of the crossing box is defined; Perform force analysis on the crossing frame and obtain the change in the swing angle of the crossing frame based on the state vector of the crossing frame; According to the change of the swing angle of the crossing frame, the expression of the center position of the crossing frame is obtained; Based on the expression of the center position of the crossing frame, the expression of the corner point position of the crossing frame is obtained; According to the expression of the corner point position of the crossing frame, the center position of the crossing frame is obtained and used as the model of the crossing frame.

4. A multi-UAV cooperative crossing method according to claim 1 or 2, characterized in that: Based on the dynamic model of each UAV and combined with the quadratic programming optimization method, a model predictive control framework for each UAV is constructed, including: ； Where, is the dynamic model of the UAV, is the state vector of the UAV, is the position coordinate of the UAV in the inertial coordinate system, is the attitude quaternion corresponding to the drone, is the linear velocity component of the UAV in the inertial coordinate system, is the control vector of the UAV, is the total thrust in the body coordinate system, is the angular velocity component in the body coordinate system; Perform Euler discretization: ； Where, For drones The state of the moment, For drones The state of the moment, To control the time step, For control input.

5. A multi-UAV cooperative crossing method according to claim 1 or 2, characterized in that: Obtain the historical trajectory sequence of each drone and build a neural network, including: Obtain the historical trajectory sequence and map it to a high-dimensional feature space through a linear embedding layer to obtain trajectory data; Design an encoder and stack it with a multi-head self-attention mechanism and a feedforward network to model the spatiotemporal dependencies in the trajectory data and obtain the encoder output; The output of the encoder is used as the input of the residual enhanced long short-term memory network for temporal modeling; Design a multi-scale causal convolution module to process the results of time series modeling to adjust the drone's fine-tuning within a short time window and obtain the initial neural network; Construct a multi-objective loss function of the initial neural network and obtain the neural network.

6. A multi-UAV cooperative crossing method according to claim 1 or 2, characterized in that: The dataset is used as the input of the neural network to train the neural network and obtain the trained neural network, including: ； Where, To predict the future The output of the step, is the feature fusion function, is the feature extraction function of Transformer, is the feature extraction function of the long short-term memory network, is the feature extraction function of CNN, From time 1 to The input sequence.

7. A multi-UAV cooperative crossing method according to claim 1 or 2, characterized in that: The trained neural network is embedded into the model predictive control framework to form a distributed model predictive control framework for multiple drones, including: The trained neural network is embedded into the model predictive control framework to add the status and quantity of other drones to the model predictive control framework of the current drone, thus obtaining a distributed model predictive control framework for the current drone. Based on the distributed model predictive control framework of all drones, a distributed model predictive control framework for multiple drones is formed.

8. A multi-UAV cooperative crossing device, characterized in that: A multi-UAV cooperative crossing method according to any one of claims 1 to 7 is adopted, comprising: The first module is used to obtain scene information of multi-UAV collaborative traversal, collect information, and generate a data set; The second module is used to model the crossing frame based on the scene information of the multi-UAV collaborative crossing, and obtain the crossing frame model; The third module is used to model each UAV based on the scene information of multi-UAV collaborative crossing and obtain the dynamic model of each UAV; The fourth module is used to build a model predictive control framework for each UAV based on its dynamic model and combined with quadratic programming optimization method; The fifth module is used to obtain the historical trajectory sequence of each drone, construct a neural network, and use the dataset as the input to train the neural network to obtain a trained neural network. The trained neural network is embedded in the model predictive control framework to form a distributed model predictive control framework for multiple drones. Based on the distributed model predictive control framework of the current drone, the model of the traversal box, and the dynamic model of the current drone, the path prediction planning is performed for the current drone and other drones, the action of the current drone is output, and all drones are traversed to obtain the real-time position of each drone. The sixth module is used to conduct multi-UAV collaborative crossing based on the real-time position of each UAV.

Images