An Event-Triggered Finite-Time Trajectory Tracking Control Method for Quadrotor UAVs Under DoS Attacks Based on Fully Connected Recurrent Neural Networks

By combining a fully connected recurrent neural network and a non-singular fast terminal sliding mode controller, a time-varying threshold event triggering mechanism was designed to solve the problem of high-precision trajectory tracking of UAVs under DoS attacks, achieving stable control effects with lightweight communication and computing.

CN123067544APending Publication Date: 2026-11-10TIANJIN POLYTECHNIC UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611199973.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-31
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing UAV trajectory tracking and control schemes suffer from problems such as communication disruption, control jitter, high computational overhead, and low approximation accuracy under DoS attacks, making it difficult to achieve high-precision, lightweight, and stable trajectory tracking.

Method used

By employing a fully connected recurrent neural network (FCRNN) combined with a practical non-singular fast terminal sliding mode controller, a time-varying threshold event triggering mechanism is designed, and a multi-feedback network structure is constructed to achieve finite-time trajectory tracking of a quadcopter UAV under DoS attacks. Combined with an attack-defense adaptive switching strategy, the communication and computation burden is reduced.

Benefits of technology

It achieves high-precision, low-jitter, and low-overhead trajectory tracking control for quadcopter drones under DoS attacks, improving the system's robustness and anti-attack capability, and ensuring the safe and stable flight of the drones.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123067544A_ABST
    Figure CN123067544A_ABST
Patent Text Reader

Abstract

This invention provides an event-triggered finite-time trajectory tracking control method for quadrotor UAVs under DoS attacks, based on a fully connected recurrent neural network, belonging to the field of networked motion control for UAVs. Addressing the challenges of strong coupling nonlinearity, external disturbances, communication disruptions, and sliding mode chattering in quadrotor systems, this invention establishes a cascaded dynamic model based on the Newton-Euler equations, designs an event-triggered mechanism with a time-varying threshold, and combines it with a non-singular fast terminal sliding mode controller to improve the system's anti-disturbance capability. A FCRNN network with a recursive feedback structure is constructed, updating parameters only at the triggering time to achieve dynamic compensation for unknowns. Furthermore, an adaptive switching strategy for attack and defense is designed to maintain stable control output during DoS attacks. Lyapunov theoretical analysis and simulation verification demonstrate that the method can reduce communication overhead, improve trajectory tracking accuracy under attack environments, and provide an effective control solution for the safe and autonomous flight of UAVs.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of networked anti-attack motion control technology for quadrotor UAVs, specifically relating to an event-triggered finite-time trajectory tracking control method for quadrotor UAVs under DoS attacks based on a fully connected recurrent neural network. Background Technology

[0002] Quadrotors possess strong coupling, multivariable, and underactuated nonlinear dynamic characteristics, making them susceptible to parameter perturbations caused by aerodynamic damping and load changes, as well as external time-varying unknown disturbances during flight. High-precision trajectory tracking has always been a core challenge in UAV motion control. Currently, most UAVs adopt a networked remote control architecture, with the controller and aircraft relying on wireless channels to transmit control commands. Communication bandwidth resources are limited, and continuous updates of control signals can easily lead to channel congestion and transmission delays, significantly reducing system real-time performance. Furthermore, wireless networks are vulnerable to denial-of-service (DoS) attacks. Attackers can overwhelm the channel with massive amounts of redundant data packets, causing command packet loss and communication interruptions, directly disrupting the stability of the closed-loop control system and seriously threatening flight safety. Therefore, robust trajectory tracking control for UAVs against DoS attacks has become a research hotspot in the UAV field.

[0003] There are many technical shortcomings in the existing UAV trajectory tracking control schemes that need to be addressed: (1) Traditional sliding mode control has obvious high-frequency jitter, ordinary terminal sliding mode is prone to control singularity, and most can only guarantee the asymptotic convergence of error, and cannot achieve high-precision tracking within a limited time; at the same time, most sliding mode algorithms do not have a lightweight communication scheduling strategy, and continuous control updates bring huge communication overhead. (2) Conventional event-triggered control uses a single exponential decay threshold. As the tracking error approaches zero, the threshold shrinks infinitely, which will cause Zeno to trigger infinitely frequently, increasing the channel burden; and the existing triggering mechanism is not adapted to the DoS attack interruption scenario. After communication is blocked, there is no reliable update scheduling strategy, and the system stability is difficult to guarantee. (3) Traditional radial basis neural networks and ordinary recurrent neural networks can only iterate continuously online. The weights need to be calculated and updated in real time throughout the process, which consumes a lot of airborne computing power; and the network structure feedback dimension is single, and the approximation accuracy of the completely unknown coupled dynamics of the quadrotor is limited. It is difficult to synchronously adapt to the discrete update sequence of event triggers, and parameter drift is prone to occur during the attack. (4) Most existing research on anti-DoS UAV control is limited to single-input single-output linear systems, and very few are designed for quadrotor multi-input multi-output strongly coupled nonlinear objects. It fails to integrate event-triggered scheduling, recursive neural network dynamic approximation, finite-time sliding mode, and network attack fault tolerance into a unified and collaborative design, and cannot simultaneously meet the multiple requirements of anti-interference, lightweight communication, low computational overhead, and stable tracking under attack conditions. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, this invention proposes an event-triggered finite-time trajectory tracking control method for quadrotor UAVs under DoS attacks based on a fully connected recurrent neural network. It establishes a cascaded dynamic model integrating multiple disturbances and network attacks, designs a time-varying threshold event triggering mechanism with a constant lower bound constraint to eliminate Zeno behavior, constructs a practical non-singular fast terminal sliding mode controller to achieve jitter-free finite-time tracking, builds a multi-feedback FCRNN to approximate unknown dynamics online and provides an attack-freezing segmented weight update rule, and combines an attack-defense adaptive switching strategy to ensure continuous and stable tracking under communication interruption. This provides a novel solution for the attack-resistant, lightweight, and high-precision flight control of networked UAVs.

[0005] The technical solution adopted in this invention is as follows:

[0006] A method for event-triggered finite-time trajectory tracking and control of a quadcopter UAV under a DoS attack based on a fully connected recurrent neural network includes the following steps:

[0007] A. Establish a position and attitude dynamics model for a quadcopter UAV, and construct a UAV control system that considers unknown nonlinear dynamics, model parameter uncertainty, external disturbances and the impact of denial-of-service (DoS) attacks, so as to provide a system model for the design of a trajectory tracking controller;

[0008] B. To address the issues of limited communication network bandwidth and high communication load caused by continuous updates of control signals, an event triggering mechanism with a time-varying threshold is designed. A triggering lower bound constraint is introduced to adaptively determine the update time of control signals and neural network parameters, thereby reducing communication resource consumption while avoiding the Zeno phenomenon.

[0009] C. Construct a practical non-singular fast terminal sliding mode controller (NFTSMC), design the sliding surface, approach law and control law, and combine it with online disturbance estimation to realize finite-time trajectory tracking control of the position and attitude of the quadcopter UAV, so as to improve control accuracy and suppress sliding mode control chattering.

[0010] D. Construct a fully connected recurrent neural network (FCRNN) and utilize a multi-feedback network structure in which the input layer receives one-step delayed feedback from the output layer, hidden layer neurons provide self-feedback and full interconnection feedback within the layer, and the output layer sends one-step delayed feedback to the hidden layer. This network can be used to approximate the unknown nonlinear dynamics of UAVs online. Furthermore, an event-triggered mechanism is combined to design an adaptive update law for the neural network weights, which reduces the computational load and communication overhead of the network while ensuring the approximation accuracy.

[0011] E. Based on the DoS attack state switching control strategy, event-triggered control and neural network online learning are executed during the non-attack phase, the control input and neural network parameters are frozen during the attack phase, and trajectory tracking control is continued after communication is restored, so as to achieve stable trajectory tracking of the quadcopter UAV for a limited time in the DoS attack environment.

[0012] Step A specifically includes: establishing the six-degree-of-freedom position and attitude coupling equations of the quadrotor based on the Newton-Euler equations; and defining the body's center of mass position vector in the inertial coordinate system as Π = [x, y, z]. T The roll, pitch, and yaw attitude angle vectors are defined as D = [φ, θ]. T The system's independent physical inputs include total lift u1 and three-axis control torques u2, u3, and u4. The complete set of dynamic equations is constructed as follows:

[0013]

[0014] In the formula, m is the total mass of the UAV, g is the acceleration due to gravity; J x J y J z For the three-axis rotational inertia, I r Characterizing rotor-gyroscope coupling and anti-torque effect; I x I y I z I is the translational aerodynamic damping coefficient. φ I θ I ψ Angular motion aerodynamic damping coefficient; total rotor speed ωi represents the angular velocity of the i-th motor rotor.

[0015] To address the underactuated characteristics of a quadcopter with six degrees of freedom and four inputs, three sets of virtual control variables are introduced to decouple the position channels:

[0016] u x =(cosφsinθcosψ+sinφsinψ)u1

[0017] u y =(cosφsinθsinψ-sinφcosψ)u1

[0018] u z =(cosφcosθ)u1

[0019] Define position control input u p =[u x u y u z ] T Attitude control input ua =[u2, u3, u4] T The system is divided into a position subsystem and an attitude subsystem, and the unified mathematical expression of the subsystems is as follows:

[0020]

[0021] Among them, the dynamic inherent term A p A a Positive definite control gain matrix Define the location set uncertainty term d p attitude lumped uncertainty term d a The unified mapping rule for uncertain sets is:

[0022] d p =d p1 +d p2 +d p3

[0023] d a =d α1 +d a2 +d a3

[0024] d p1 d a1 d represents the completely unknown nonlinear dynamics of the system. p2 d a2 The perturbation terms representing model parameters such as mass and moment of inertia, d p3 d a3 This represents bounded external disturbances such as time-varying aerodynamics and wind disturbances; assuming d p d α Bounded and upper bound Completely unknown.

[0025] Construct a mathematical model of the timing of intermittent DoS attacks, assuming {h} n}, n∈N is the sequence of times the nth attack is initiated, τ n >0 represents the duration of a single attack, and the time interval of the nth attack segment is defined. The total global attack duration set Θ(t) = U n H n ∩[0, t], the normal communication interval without attack is γ(t)=[0, t]\Θ(t).

[0026] Introduce two attack bounded constraint assumption operators:

[0027] Attack frequency constraint operator: n(t) represents the total number of attacks within the interval [0, t], η is the burst attack tolerance parameter, and τ d This is the lower bound of the average attack interval;

[0028] Attack duration constraint operator: |·| is a Lebesgue measure of time. For short-term attacks, i>1 represents the attack duration decay coefficient.

[0029] By integrating the dynamic equations, lumped uncertainty terms, and DoS attack timing constraints, a complete networked quadrotor MIMO nonlinear control model is obtained, which serves as a unified basic model for the subsequent integrated design of event triggering, sliding mode, and neural network controller.

[0030] The specific operations of step B include: (1) addressing the engineering pain points of limited wireless network bandwidth, continuous and uninterrupted issuance of control commands causing channel congestion and increased transmission delay, constructing a time-varying threshold event triggering mechanism with a constant positive lower bound constraint, and defining a discrete event triggering time sequence {t}. k}, adjacent trigger interval T k =t k+1 -t k Construct a comprehensive composite error evaluation function Ξ e (t) serves as the triggering criterion, and the mathematical expression of the composite error function is:

[0031]

[0032] In the formula, E(·) represents the event sampling error term, e, These are the system tracking error and its first derivative, and the nonlinear composite term, respectively. sig η(·) is a nonlinear power sign function.

[0033] (2) Design a segmented dynamic triggering threshold max{κexp(-δt), σ}, where κ is the initial amplitude coefficient of exponential decay, δ is the exponential decay rate, and σ>0 is a globally constant positive lower bound constant; within the normal communication interval without DoS attacks, the rule for solving the event triggering time is: t k+1 =inf{t>t k |Ξ e (t)>max{κe -δt Only when the composite error exceeds the dynamic threshold are the quadrotor position, attitude control commands, and all weight parameters of the FCRNN neural network synchronously updated. k , t k+1 Within the interval, the control quantity and network parameters remain constant, and data transmission is completed only at discrete moments, which greatly reduces the frequency of wireless data packet transmission and alleviates the pressure on communication bandwidth.

[0034] (3) Derive the minimum event interval by using the global maximum rate of change of the system error L and the lower bound of the threshold σ. The interval between any two event triggers is strictly greater than the lower limit of a normal number, thus overcoming the defect of traditional pure exponential thresholds that converge infinitely close to 0 with error, causing Zeno to trigger infinitely frequently; when the system falls into the DoS attack range... When switching to the fixed-period trigger mode, set upper and lower bound constraints for the trigger interval. The attack period also enforces a strict positive lower limit for the event interval, completely avoiding the Zeno phenomenon across the entire domain, and ensuring the basic frequency of control updates while reducing communication resource consumption.

[0035] The specific operations of step C include: (1) constructing practical non-singular fast terminal sliding surfaces for the position subsystem and attitude subsystem respectively, and defining the position tracking error e1 = Π d -Π、Attitude tracking error e2=D d -D, the position sliding surface is constructed as follows:

[0036]

[0037] The orientation sliding surface is constructed as follows:

[0038]

[0039] In the formula, α, β, α1, and β1 are all positive weighting coefficients, and the parameter constraints satisfy η1>η2, η3>η4 and 1<η2, η4<2, sig η (·) is a nonlinear power sign function. This sliding surface avoids the control singularity problem of traditional terminal sliding while taking into account fast convergence characteristics.

[0040] (2) Design a hierarchical composite control law. Both the position controller and the attitude controller are composed of three parts superimposed: an equivalent approximation term from an FCRNN, a power-law robust switching term, and an adaptive disturbance compensation term. The position control input is: Attitude control input: The robust switching control term is defined as follows: It is a diagonal positive definite matrix, 0<ε1, ε2<1. The power-approaching law is used to replace the standard sign function to reduce the mechanical chattering caused by high-frequency switching of sliding mode control.

[0041] (3) Construct an online adaptive disturbance estimation and compensation unit, with a position adaptive compensation term. Attitude adaptive compensation term The corresponding adaptive update rule is H p H α For positive definite feedforward gain, To compensate for the learning rate, this compensation term absorbs lumped residual disturbances in real time, such as the approximation residual of the FCRNN network, event-triggered discrete sampling error, and position and attitude cascade coupling error.

[0042] (4) Based on the finite-time stability lemma, the Lyapunov function is constructed and the convergence is proved. It can be deduced that the position and attitude tracking errors can converge to the minimum neighborhood of the origin in a finite time, realizing high-precision finite-time trajectory tracking of the quadrotor, while also having the control effect of no singularity and low jitter.

[0043] The specific operations of step D include: (1) constructing a multi-feedback fully connected recurrent neural network (FCRNN) with 9 input layer neurons, 7 hidden layer neurons, and 3 output layer neurons. The network has a built-in three-layer recurrent feedback connection structure, namely, the input layer receives one-step delayed feedback from the output layer, the hidden layer neurons self-feedback and fully interconnected feedback within the layer, and the output layer sends one-step delayed feedback to the hidden layer; independent weight matrices are configured between the input layer, the hidden layer, and the output layer. The position subsystem FCRNN and the pose subsystem FCRNN adopt the same network topology structure, and only the input vector and learning rate parameters are distinguished from each other.

[0044] (2) Define the formula for calculating the aggregated input of each layer in FCRNN: The aggregated input of the i-th neuron in the input layer is... (x i ) represents the external input vector component, w oI For the output to input feedback weights, m -1 (·) represents a one-step temporal delay operator; external input vector of the location network. External input vector of the pose network E p E a These are the position and attitude system event sampling error vectors, respectively.

[0045] (3) The integrated input structure of hidden layer neurons is as follows: In the formula w I Let w be the input-hidden layer weight matrix. D Let w be the self-recursive weight matrix of the hidden layer. oH The output is the hidden feedback weight matrix; the hidden layer activation function is the Sigmoid function. The output layer uses a linear mapping W is the hidden-output layer weight matrix, and the network output is used to fit the ideal equivalent control quantity u required for NFTSMC. pNN u aNN .

[0046] (4) Design a segmented network weight adaptive update law that matches the timing of event triggering and DoS attack, and output weights using a positional FCRNN. For example, the update rule is defined as:

[0047]

[0048] Hidden layer input weights Hidden self-feedback weights Output input feedback weights Output hidden feedback weights All use the same update logic; To correspond to the positive learning rate, This is the Jacobian matrix corresponding to each weight; the weight update logic of the pose FCRNN is completely consistent with that of the position network, only replacing the sliding mode variable S2 corresponding to the pose and the pose-specific learning parameters.

[0049] (4) Network update execution constraint rules: Only when there is no DoS attack and the event trigger time t is reached. k All weight gradients are updated iteratively at regular intervals. When the interval between two triggers is reached, or when the DoS attack range Θ(t) is entered, all weight update derivatives are set to zero, all network parameters are frozen, and online learning is stopped. High-precision online approximation of the complete unknown coupled nonlinearity, parameter perturbation, and external disturbance of the quadrotor is achieved through discrete-time local iteration. No continuous onboard real-time calculation is required, and the frequency of wireless transmission of network weight parameters is reduced simultaneously. Under the premise of ensuring dynamic approximation accuracy, the onboard computing load and communication overhead are greatly reduced.

[0050] The specific operations of step E include: (1) Real-time monitoring of the current system communication status to determine whether the current time falls within the DoS attack range. There are two operating modes: Mode 1 is... Normal communication without attacks, mode two is (2) When the system is in attack-free mode, the time-varying threshold event triggering criterion designed in step B is activated, and when the composite error Ξ e (t) The trigger time t is generated when the dynamic threshold max{κexp(-δt), σ} is exceeded. k At the trigger time, the NFTSM composite control quantity constructed in step C is sent out synchronously, and the FCRNN segmented weight update rule given in step D is executed simultaneously to complete the gradient iteration of all weights of the input layer, hidden layer, and output layer, continuously approximating the unknown dynamics of the system online and compensating for lumped disturbances. (3) When the system is in DoS attack mode, the fixed periodic event update strategy is switched, and the trigger interval constraint is satisfied. The zero-order hold (ZOH) mechanism is executed to lock the control signal that was successfully transmitted during the last normal communication. Continuous output to the actuator, with no zeroing of control input throughout the process, to avoid loss of lift and attitude instability of the UAV; at the same time, update the derivatives of all weights of FCRNN to zero, freeze all network parameters, and suspend the online iterative learning of the neural network to prevent invalid computation and infinite weight drift during the attack period. (4) When the DoS attack ends and the time exits the attack interval After the communication link is restored to normal, the time-varying threshold event trigger judgment logic is immediately switched back. Based on the current real-time tracking error and sampling error, the FCRNN weight adaptive update process is restarted to continue finite-time trajectory tracking. (5) The Lyapunov function piecewise stability proof method is adopted to derive the Lyapunov function change law under normal communication interval and DoS attack interval respectively; the system divergence correlation coefficient w3 and convergence correlation coefficient w1 are defined. If the DoS attack meets the duration and frequency constraints, The overall convergence rate of the system is greater than the divergence rate caused by the attack. The closed-loop quadcopter UAV system can maintain stability for a limited time under intermittent DoS interference, and achieve high-precision continuous trajectory tracking.

[0051] The beneficial effects of the technical solution provided by this invention are:

[0052] This invention employs a fully connected recurrent neural network (FCRNN) to approximate the unknown nonlinear dynamics and external disturbances of a quadrotor UAV online. It combines a time-varying threshold event triggering mechanism with practical non-singular fast terminal sliding mode control to achieve synchronous event-triggered updates of neural network weights and control inputs. This effectively reduces communication frequency and computational burden while maintaining finite-time trajectory tracking accuracy. Compared to existing methods, this invention not only improves the UAV system's adaptability to unknown models and external disturbances but also effectively resists the impact of denial-of-service (DoS) attacks on the network control system. It balances communication resource utilization efficiency, control accuracy, and system robustness, providing an efficient and reliable trajectory tracking control scheme for the safe and stable flight of networked quadrotor UAVs. Attached Figure Description

[0053] To more clearly illustrate the technical solution of the present invention, the accompanying drawings involved in the implementation of the present invention will be briefly described below.

[0054] Figure 1 This is a diagram of the overall architecture of the present invention.

[0055] Figure 2 This is a flowchart of the event triggering mechanism in this invention.

[0056] Figure 3 This is the control law for the quadcopter unmanned aerial vehicle system in this invention.

[0057] Figure 4 This invention relates to trajectory tracking for the position subsystem of a quadcopter UAV.

[0058] Figure 5 This refers to the position tracking error of the quadcopter UAV in this invention.

[0059] Figure 6 This refers to the event triggering interval of the quadcopter UAV position subsystem in this invention.

[0060] Figure 7 This invention relates to trajectory tracking for the attitude subsystem of a quadcopter drone.

[0061] Figure 8 This refers to the attitude tracking error of the quadcopter UAV in this invention.

[0062] Figure 9 This refers to the event triggering interval of the quadcopter UAV attitude subsystem in this invention. Detailed Implementation

[0063] To more clearly illustrate the technical solution of the present invention, the present invention will be further described in detail below with reference to specific embodiments.

[0064] Example 1

[0065] This embodiment takes a quadcopter drone trajectory tracking control system as the object and conducts trajectory tracking control verification under denial-of-service (DoS) attacks and unknown external disturbances.

[0066] First, an adaptive event-triggered sliding mode control method based on a fully connected recurrent neural network (FCRNN) is implemented using MATLAB programming. A practical non-singular fast terminal sliding mode controller is constructed based on the quadrotor UAV dynamics model, and a time-varying threshold event triggering mechanism is designed to update the control input only when the triggering conditions are met, thereby reducing communication overhead and avoiding the Zeno phenomenon. Then, the FCRNN is used to approximate the unknown nonlinear dynamics of the UAV and external disturbances online, and the neural network weight updates are executed synchronously with the event triggering mechanism, achieving a synergistic improvement in control accuracy and computational efficiency. Finally, UAV trajectory tracking control is implemented under a DoS attack scenario, outputting the position tracking curve, tracking error, and event triggering results.

[0067] The results show that the present invention can achieve finite-time stable trajectory tracking of quadcopter UAVs under DoS attacks and unknown disturbances, effectively reducing the control update frequency and communication resource consumption, while improving the system's robustness, anti-attack capability and trajectory tracking accuracy.

Claims

1. A method for event-triggered finite-time trajectory tracking and control of a quadcopter UAV under a DoS attack based on a fully connected recurrent neural network, characterized in that, Includes the following parts: A. Establish a position and attitude dynamics model for a quadcopter UAV, and construct a UAV control system that considers unknown nonlinear dynamics, model parameter uncertainty, external disturbances and the impact of denial-of-service (DoS) attacks, so as to provide a system model for the design of a trajectory tracking controller; B. To address the issues of limited communication network bandwidth and high communication load caused by continuous updates of control signals, an event triggering mechanism with a time-varying threshold is designed. A triggering lower bound constraint is introduced to adaptively determine the update time of control signals and neural network parameters, thereby reducing communication resource consumption while avoiding the Zeno phenomenon. C. Construct a practical non-singular fast terminal sliding mode controller (NFTSMC), design the sliding surface, approach law and control law, and combine it with online disturbance estimation to realize finite-time trajectory tracking control of the position and attitude of the quadrotor UAV, so as to improve control accuracy and suppress sliding mode control chattering. D. Construct a fully connected recurrent neural network (FCRNN) and utilize a multi-feedback network structure in which the input layer receives one-step delayed feedback from the output layer, hidden layer neurons provide self-feedback and full interconnection feedback within the layer, and the output layer sends one-step delayed feedback to the hidden layer. This network can be used to approximate the unknown nonlinear dynamics of UAVs online. Furthermore, an event-triggered mechanism is combined to design an adaptive update law for the neural network weights, which reduces the computational load and communication overhead of the network while ensuring the approximation accuracy. E. Based on the DoS attack state switching control strategy, event-triggered control and neural network online learning are executed during the non-attack phase, the control input and neural network parameters are frozen during the attack phase, and trajectory tracking control is continued after communication is restored, so as to achieve stable trajectory tracking of the quadcopter UAV for a limited time in the DoS attack environment.

2. The event-triggered finite-time trajectory tracking and control method for a quadcopter UAV based on a fully connected recurrent neural network under a DoS attack, as described in claim 1, is characterized in that... In part A, the method for establishing the position and attitude dynamics model of a quadcopter UAV includes: (1) A six-degree-of-freedom dynamic model of a quadcopter UAV is established based on Newton-Euler dynamics. The position coordinates of the UAV are represented as (x, y, z), and the attitude angles are represented as (φ, θ, ψ). The control inputs include the total lift u1 and the roll, pitch, and yaw moments u2, u3, and u4. The position motion equations and attitude motion equations of the UAV are established: Where m represents the mass of the UAV, g represents the acceleration due to gravity, and J x J y J z Let I represent the moments of inertia about the three coordinate axes respectively. r I represents the rotor gyroscope effect. x I y I z I represents the linear damping coefficient of air resistance. φ I θ I ψ The attitude air damping coefficient is given, and the total rotor angular velocity satisfies: (2) To address the underactuated characteristics of quadcopter UAVs, a virtual control input is introduced: Define the position state vector Π = [x, y, z] T and the attitude state vector D = [φ, θ, ψ] T The quadrotor dynamics model is decoupled into a position subsystem and an attitude subsystem: Among them, u p =[u x u y, u z ] T u a =[u2, u3, u4] T A p and A a B represents the known dynamics of the system. p and B a These represent the control gain matrix. (3) The unknown aerodynamics of the UAV, model parameter perturbations and external disturbances are uniformly represented as the lumped uncertainty term d of the position subsystem. p and the lumped uncertainty term d of the attitude subsystem a We construct a control model for a quadrotor UAV that simultaneously considers unknown nonlinear dynamics, parameter uncertainties, external disturbances, and the effects of DoS attacks, providing a unified mathematical model for subsequent online approximation of event-triggered controllers, sliding mode controllers, and fully connected recurrent neural networks.

3. The method for finite-time trajectory tracking and control of a quadrotor UAV under DoS attack based on a fully connected recurrent neural network according to claim 1, characterized in that, The event triggering mechanism in Part B specifically includes: (1) To address the transmission congestion caused by limited network bandwidth and continuous updates of control commands, a time-varying threshold event triggering mechanism with a constant positive lower bound is designed, defining the event triggering time sequence (t). k }, adjacent trigger interval T k =t k+1 -t k Construct the comprehensive error evaluation function: In the formula E(·) represents the event sampling error, e, This represents the tracking error and its derivative. (2) Set the dynamic trigger threshold max{κexp(-δt), σ}, where κ is the exponential decay amplitude, δ is the decay rate, and σ>0 is a fixed lower bound constant; the trigger criterion within the normal communication interval without DoS is: t k+1 =inf{t>t k |X e (t)>max{κe -δt ,s}} Only on discrete t that the triggering condition is met k Control commands are issued synchronously, and all weights of the neural network are updated. During other periods, the control quantity and network parameters remain constant, which greatly reduces the frequency of data packet transmission and reduces communication overhead. (3) Derive the minimum triggering interval T by using the global upper bound L of error growth and the lower bound σ of the threshold. min =σ / L>0, the interval between any two triggers is strictly greater than the lower limit of a positive number, theoretically eliminating the Zeno infinitely frequent triggering defect caused by the traditional exponential decay threshold approaching 0; the fixed-period triggering mode is switched during the DoS attack period, and the triggering interval is constrained. The event interval is guaranteed to have a strict positive lower bound throughout the process, completely avoiding the Zeno phenomenon.

4. The event-triggered finite-time trajectory tracking and control method for a quadcopter UAV based on a fully connected recurrent neural network under a DoS attack, as described in claim 1, is characterized in that... The practical non-singular fast terminal sliding mode controller in part C specifically includes: (1) Construct practical non-singular fast terminal sliding surfaces for the position and attitude subsystems respectively. Position sliding: Attitude sliding mode: Where (e1=∏) d -Π) represents the position tracking error, (e2=D) d -D) represents the attitude tracking error; the parameters satisfy η1>η2, η3>η4 and 1<η2, η4<2, thus avoiding the singularity problem of traditional terminal sliding mode control. (2) Design a three-layer composite control law, namely, an FCRNN dynamic approximation term, a power-law robust switching term, and an adaptive disturbance compensation term; Position control: Attitude control: Among the switching items 0 < ε1, ε2 < 1, weaken the high-frequency switching of the sign function and suppress mechanical chattering. (3) Adaptive compensation term: Adaptive update law It absorbs residuals from neural network approximation, discrete errors from event sampling, and cascaded errors from position-attitude coupling in real time. Based on the finite-time stability lemma, it proves through Lyapunov that the tracking error can converge to the zero minimum neighborhood in finite time, thereby improving trajectory tracking accuracy.

5. The event-triggered finite-time trajectory tracking control method for a quadcopter UAV based on a fully connected recurrent neural network under a DoS attack, as described in claim 1, is characterized in that... The fully connected recurrent neural network in part D specifically includes: (1) The multi-feedback fully connected recurrent neural network (FCRNN) employs 9 input layer neurons, 7 hidden layer neurons, and 3 output layer neurons. Its three-layer feedback structure is as follows: the input layer receives one-step delayed feedback from the output layer, hidden layer neurons provide self-feedback and mutual feedback through full interconnection within the layer, and the output layer sends one-step delayed feedback to the hidden layer. The location network input vector uses: The input to the attitude network is: E p E a The event sampling error vector is used; the hidden layer activation function is Sigmoid, and the output layer uses linear activation to output the equivalent control quantity. (2) Hidden layer input aggregation expression: w I w D w oH These are the feedback weight matrices from the input layer to the hidden layer, the hidden self-feedback, and the output layer to the hidden layer, respectively. -1 (·) represents a one-step time delay signal; network output is the hidden layer activation vector, and W is the output layer weight. (3) Design an event-triggered segmented weight adaptive update law to output weights from the position subsystem neural network. For example: The update rules for various weights in the hidden layer are consistent with the segmentation logic described above; network parameters are iterated only at the time of normal communication triggering, and all weights are frozen during the interval between attacks and events, continuously approximating all unknown nonlinearities of the system online, reducing the overhead of airborne real-time calculation and parameter transmission while ensuring approximation accuracy.

6. The event-triggered finite-time trajectory tracking control method for a quadcopter UAV based on a fully connected recurrent neural network under a DoS attack, as described in claim 1, is characterized in that... The switching strategy under DoS attack in Part E specifically includes: (1) Determine in real time whether the current time is within the DoS attack range. It has two operating modes that can be switched adaptively; Mode 1: Normal communication without attacks; Mode 2: Communication disruption attack scenario. (2) No-attack mode: Enable time-varying threshold event triggering criteria, and synchronously update the sliding mode control quantity after the triggering threshold is reached, and perform adaptive online learning of all weight gradients of FCRNN to continuously compensate for the total uncertainty of the system. (3) DoS attack mode: Switching to a fixed-period triggering strategy with trigger interval constraints. Employing a zero-hold (ZOH) mechanism, the actuator latches the most recently successfully transmitted control signal. The control input is not set to zero to prevent the drone from losing lift and becoming unstable in attitude; at the same time, the derivatives of all FCRNN weights are updated to zero to freeze the network learning and avoid invalid iterations and parameter drift during the attack period. (4) After the DoS attack ends and the communication link is restored, immediately switch back to the time-varying threshold event triggering logic, and restart the FCRNN online approximation and weight update based on the current real-time tracking error; use Lyapunov function interval analysis to determine when the attack duration and frequency constraints are met. At that time, the overall decay rate of the Lyapunov function of the system is greater than the divergence rate caused by the attack, ensuring that the closed-loop system still has finite-time trajectory tracking stability under continuous DoS interference.