A Flight Lifesaving Vehicle Event-Triggered Attitude Control Method Based on Quantized Higher-Order Learning Extended State Observer

By quantizing a higher-order learning-based extended state observer to decompose lumped disturbances and combining it with an event-triggered mechanism, the problems of insufficient estimation of complex disturbances and improper error handling in the attitude control of flight life-saving vehicles were solved, thereby improving stability and resource efficiency.

CN123113976APending Publication Date: 2026-11-17DALIAN MARITIME UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611380702.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-07
Publication Date
2026-11-17

Smart Images

  • Figure CN123113976A_ABST
    Figure CN123113976A_ABST
Patent Text Reader

Abstract

This invention provides an event-triggered attitude control method for lifesaving vehicles based on a quantized high-order learning extended state observer, belonging to the field of aircraft control. Addressing the combined effects of continuous time-varying wind turbulence, nonlinear inertial coupling, state input quantization, and limited control resources on near-shore lifesaving vehicles, a three-axis attitude dynamics, state quantization, and input quantization model is established. The lumped disturbance is decomposed into structured learning terms and equivalent residual terms, constructing a quantized high-order learning extended state observer. An adaptive attitude controller is constructed based on the quantized attitude and quantized angular velocity. At the actuator end, quantized candidate control commands are updated according to a fixed threshold periodic event triggering rule and maintained at zero order. This invention improves disturbance reconstruction and attitude tracking performance under wind disturbances, reduces control command transmission and actuator update frequency, and is applicable to resource-constrained marine lifesaving vehicles.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle control and intelligent maritime rescue equipment technology, and more particularly to an event-triggered attitude control method for a flight life-saving device based on a quantized high-order learning extended state observer. Background Technology

[0002] With the increasing prevalence of maritime transportation, marine resource development, coastal tourism, and water operations, the rapid detection and emergency rescue of people falling overboard at sea has become a major concern. Traditional maritime rescue primarily relies on rescue vessels approaching the area where the person fell overboard, followed by rescuers deploying lifebuoys, flotation devices, or directly conducting surface rescue operations. However, the speed of rescue vessels is easily limited by sea conditions, channel conditions, vessel maneuverability, and the location of the person in the water. When the person in the water is far from a vessel or shore-based rescue station, or when the accident area contains reefs, shoals, narrow channels, or large waves, rescue vessels often struggle to reach the target location quickly. Furthermore, rescuers directly approaching the person in the water may face risks such as adverse sea conditions, low visibility, and surface obstacles. Therefore, how to quickly provide buoyancy support to people in the water after an accident, thus buying time for subsequent rescue efforts, has become a crucial issue that needs to be addressed in the field of maritime emergency rescue.

[0003] A flying rescue vehicle is a maritime emergency rescue device that combines a multi-rotor unmanned aerial vehicle platform with a buoyancy-based rescue mechanism. This equipment typically consists of a rotor propulsion system, airframe structure, buoyancy device, flight controller, communication module, positioning module, power module, and related sensors. Upon discovering a person has fallen into the water, the flying rescue vehicle can be rapidly deployed from a ship, shore-based station, or other rescue platform, using the lift generated by its rotors to reach the area where the person is in the water. Once above the target, the flying rescue vehicle can descend to the water's surface to provide timely buoyancy support to the person in the water, or, under the control of rescue personnel, guide the person to a safe area. Compared to relying entirely on rescue vessels to approach the target, flying rescue vehicles offer advantages such as faster response time, higher maneuverability, less susceptibility to surface navigation conditions, and reduced risk to rescue personnel. Therefore, they have promising applications in near-shore, port, island, ship deck, and coastal scenic areas.

[0004] Flight rescue vehicles need to complete multiple operational phases during flight, including takeoff, acceleration, cruise, deceleration, hovering, descent, and water contact. Among these, attitude control performance directly affects the flight stability, trajectory accuracy, and deployment safety of the flight rescue vehicle. Because flight rescue vehicles need to carry a rescue structure with a certain volume and buoyancy, their airframe shape, mass distribution, and aerodynamic characteristics typically differ from conventional small multi-rotor UAVs. The buoyancy structure may increase the frontal area and make the aircraft more sensitive to changes in external wind fields; the structural connection between the rescue device and the rotor platform may also cause a shift in the center of mass and changes in rotational inertia. Therefore, during roll, pitch, and yaw motions, there may be significant nonlinear inertial coupling between the various attitude channels, resulting in nonlinear, strongly coupled, and parameter-uncertain attitude dynamics of the flight rescue vehicle.

[0005] The wind field environment in the low-altitude region near the sea is quite complex. The exchange of heat and momentum between the sea surface and the air creates gusts and turbulence with randomness and time-varying characteristics. Ship superstructures, shore-based facilities, and coastal topography can also cause local airflow separation, wakes, and sudden changes in wind speed. When a flying rescue vehicle takes off from a ship or shore-based platform and approaches a surface target, it is easily and continuously affected by the combined effects of translational wind speed and angular turbulence rate. Especially during the low-altitude hovering, target alignment, and descent phases, wind turbulence can cause rapid changes in roll, pitch, and yaw angles, increasing attitude tracking errors and, in severe cases, causing the aircraft to deviate from the intended deployment area. In addition to external wind disturbances, changes in rotor aerodynamics, actuator response errors, airframe parameter perturbations, load changes, and unmodeled dynamics also affect attitude control performance. These factors usually coexist and can be coupled with the nonlinear inertia of the airframe to form a lumped disturbance with time-varying, randomness, and uncertainty.

[0006] Existing attitude control methods for multirotor aircraft mainly include proportional-integral-derivative (PID) control, sliding mode control, adaptive control, backstepping control, model predictive control, and disturbance observer-based control methods. PID is simple in structure and easy to implement in engineering, but its control parameters are usually tuned according to specific operating conditions. When external disturbances increase, system parameters change, or channel coupling becomes significant, control performance tends to degrade. Sliding mode control has strong robustness and can suppress parameter uncertainties and external disturbances to a certain extent. However, the discontinuous switching term in traditional sliding mode control may cause control input chattering, increasing the frequent operation of rotor motors and actuators, which is detrimental to the long-term stable operation of energy-constrained flight rescue vehicles. While using continuous approximation functions can reduce chattering, it may also introduce steady-state errors and weaken the control system's ability to compensate for rapidly changing disturbances.

[0007] Adaptive control can compensate for some parameter uncertainties by adjusting online parameters, but its control effect usually depends on the parameterized form of the uncertain terms. When the system is subjected to wind turbulence that is difficult to parameterize accurately, unmodeled dynamics, and nonlinear coupling, traditional adaptive methods may fail to achieve ideal compensation results. Extended state observers and other disturbance observers can treat external disturbances, parameter perturbations, and unmodeled terms as a unified lumped disturbance and estimate it online based on the measurable state. However, traditional extended state observers usually focus on the overall estimation of the lumped disturbance, mainly relying on high observer gain to improve the estimation speed. Under the combined effects of continuously changing wind turbulence and nonlinear inertial coupling, excessively high observer gain can easily amplify quantization noise and measurement noise, while low observer gain may produce significant estimation lag. Therefore, there is a certain contradiction between the disturbance representation capability, dynamic reconstruction accuracy, and noise suppression capability of traditional observers.

[0008] To improve the ability to approximate complex disturbances, some existing technologies introduce neural networks, fuzzy systems, or basis function learning mechanisms into the disturbance observation and control process, utilizing online learning weights to represent unknown nonlinear terms. Learning-based observers can extract structured components from disturbances to some extent, but their learning performance is typically affected by factors such as basis function selection, learning rate, measurement noise, and state accuracy. The attitude angle and angular velocity signals used by flight controllers are usually derived from sensor sampling and digital processing, and the actual signals obtained have limited resolution. After quantization, the state signals produce discontinuous jumps. Directly differentiating the quantized angular velocity can easily cause spikes in the differential signal, further leading to fluctuations in the learning residuals and rapid changes in weights. When the approximate basis function cannot fully describe the actual disturbance, basis function mismatch errors and unmodeled residuals also occur. Without proper handling of learning residuals and weight changes, the observer's estimated output may exhibit high-frequency fluctuations, thus affecting the smoothness of attitude control commands.

[0009] On the other hand, flight rescue vehicles are typically powered by onboard batteries and embedded flight controllers, which limit their communication bandwidth, computing power, and the number of permissible updates to the rotor actuators. Attitude angles and angular velocities, after being sampled by sensors and converted from analog to digital or digitally encoded, form signals with limited resolution. Control commands may also undergo input quantization during transmission and execution. State quantization errors directly affect tracking errors, sliding mode variables, and disturbance estimation results; input quantization causes deviations between the actual control quantities acting on the aircraft and the ideal control quantities calculated by the continuous controller. If the control system sends new control commands to the actuators at every sampling moment, it increases the communication burden, the controller's computational burden, and the frequent adjustments to the rotor motors, hindering efforts to reduce the flight rescue vehicle's energy consumption and actuator wear.

[0010] Event-triggered control can determine whether to update actuator instructions based on current control requirements, thereby reducing unnecessary communication and control updates. Existing continuous-time event-triggered methods typically require continuous monitoring of system state or triggering conditions, placing high demands on the processor's real-time computing power and potentially leading to a continuously shortening trigger interval under certain conditions. Methods based on preset sampling times for event judgment are easier to implement in digital control systems, but still require careful handling of the relationship between continuous control instructions, quantized candidate instructions, and the actual actuator input after zero-order hold. If stability analysis only considers the continuous controller output while ignoring errors caused by state quantization, input quantization, sampling mismatch, event thresholds, and zero-order hold, the theoretical model may not match the actual execution process, making it difficult to accurately describe the stability of the closed-loop system and the range of attitude tracking errors.

[0011] Furthermore, disturbance observation, state quantization, input quantization, and event triggering in the flight life-saving device control system are not independent of each other. Quantizing the state affects the observer's state correction process and learning basis function; disturbance estimation errors further affect continuous control commands; input quantization changes the candidate control variables used for event judgment; and event triggering and zero-order hold can lead to a mismatch between the actual actuator input and the current candidate command. These errors are coupled within the closed-loop system. If the observer, controller, and event triggering mechanism are designed separately without unified modeling and stability analysis, problems such as unreasonable control gain selection, insufficient disturbance compensation, or frequent control input updates may occur.

[0012] Therefore, it is necessary to provide an event-triggered attitude tracking control method suitable for flight rescue vehicles, which can effectively estimate lumped disturbances caused by continuous wind turbulence, nonlinear inertial coupling, parameter uncertainties, and unmodeled dynamics, while handling state quantization, input quantization, sampling mismatch, and zero-order hold error. In addition to improving the lumped disturbance reconstruction capability and attitude tracking performance, it is also necessary to reduce the update frequency of control commands, alleviate the burden on communication and actuator actions, and theoretically explain the boundedness of observation error, learning weight error, and attitude tracking error, as well as eliminate the possibility of an infinite number of triggers within a finite time, thereby improving the stability, reliability, and engineering feasibility of flight rescue vehicles operating in complex near-shore wind environments. Summary of the Invention

[0013] To address the aforementioned technical problems in existing technologies, such as insufficient accuracy in estimating lumped disturbances caused by continuous time-varying wind turbulence and nonlinear inertial coupling, lack of explicit handling of state and input quantization errors, and excessive burden on communication and actuator resources due to frequent control command updates, this invention provides an event-triggered attitude control method for flight rescue vehicles based on a quantized high-order learning extended state observer. This invention decomposes the lumped disturbance into structured learning terms and equivalent residual terms, achieving both rapid adaptation and weight smoothing through a residual-driven two-stage learning structure; it constructs an adaptive attitude controller using quantized states; and at the actuator end, it employs fixed-threshold periodic event triggering and zero-order hold, updating only when the deviation between the quantized candidate command and the current hold command reaches a threshold, thereby reducing the number of command updates while ensuring bounded closed-loop signals and attitude tracking performance.

[0014] The technical means employed in this invention are as follows: A flight life preserver event-triggered attitude control method based on a quantized high-order learning extended state observer includes: S1. Establish a three-axis attitude dynamics model based on the roll, pitch and yaw motions of the flight life-saving vehicle, and uniformly quantize the attitude angle, angular velocity and continuous control commands to obtain quantized state signals and quantized candidate control commands. S2. The angular turbulence rate of the flight rescue vehicle during near-sea surface flight is generated using the continuous Dryden wind turbulence model. The angular turbulence rate is converted into a three-axis equivalent angular acceleration disturbance and together with the nonlinear inertial coupling term to form a lumped disturbance. S3. Represent the lumped perturbation as the sum of the structured learning term and the equivalent residual term, and establish a quantized high-order learning extended state observer; the observer includes an attitude state estimation channel, an angular velocity state estimation channel, a residual estimation channel, a filtering differential channel, an intermediate learning weight channel, and a final smoothing learning weight channel. S4. Construct sliding mode variables and an adaptive quantized state feedback controller based on the quantized attitude angle, quantized angular velocity, desired attitude trajectory and the lumped disturbance estimate, generate continuous control commands, and limit and quantize the continuous control commands. S5. At the preset periodic evaluation time, compare the quantized candidate control command with the current hold command of the actuator. When the absolute value of the difference between the two is not less than the fixed trigger threshold, update the actuator command; otherwise, maintain the original command and perform zero-order hold on the actuator command between adjacent evaluation times. S6. Construct the Lyapunov functions for the observation learning subsystem and the attitude tracking subsystem, determine the control parameter conditions that make the closed-loop signal consistent and eventually bounded, and eliminate Zeno behavior according to the preset periodic evaluation interval.

[0015] Further, step S1 includes: S11. The attitude motion of the flight rescue vehicle is described as a quadrotor rigid body rotational dynamics model. Based on the quadrotor rigid body rotational dynamics model, the rotational motion of the flight rescue vehicle system is expressed as:

[0016] in, For roll angle, For the roll angular velocity, For roll acceleration, The pitch angle, Pitch angular velocity, For pitch acceleration, Yaw angle Yaw angular velocity, Yaw acceleration, The moment of inertia of the rolling channel. For the moment of inertia of the pitch channel, For the moment of inertia of the yaw channel, , , For three-axis control input; S12. Write the rotational dynamics model of the quadrotor rigid body in state-space form as follows:

[0017] in, The attitude angle vector, It is the angular velocity vector. It is a diagonal gain matrix. To control the input vector, This is the lumped disturbance vector; S13. Perform uniform quantization on the attitude angle and angular velocity respectively to obtain the uniform quantized signals of the attitude angle and angular velocity as follows:

[0018] in, For the first Channel quantization attitude angle, The step size for attitude angle quantization. For the first Actual attitude angle of the channel For the first Channel quantization angular velocity, Quantize the step size for angular velocity. For the first Actual angular velocity of the channel This is for rounding up to the nearest integer. S14. Set the quantization error to meet the following conditions:

[0019] S15, Continuous control command Input quantization is performed, and the following quantization candidate instructions are obtained:

[0020] in, For the first Channel quantization candidate control instructions, For input quantization operator, To input the quantization step size, For the first Channel continuous control commands, The switching threshold for quantizing equivalent representation; S16. The quantization candidate instruction is equivalently represented as:

[0021] in, For the first Channel time-varying gain, , It is an unknown and bounded time-varying gain, and its sign remains unchanged during quantization. For the first Bounded time-varying offset of the channel ; S17, due to ,when hour, It is also bounded, from which we can deduce... It is bounded, making ,in, For the first Upper bound of channel offset; S18, due to The lower bound of is uncertain and changes over time; therefore, an adaptive estimation method is used to estimate it. To avoid singular difficulties arising from estimating zero, therefore... The lower limit is used with time-varying gain It means that, among them, For time-varying gain The lower boundary.

[0022] Further, step S2 includes: S21. Establish a continuous Dryden wind turbulence model, and use the continuous Dryden wind turbulence model to generate the angular turbulence rate of the flight rescue vehicle during near-sea surface flight. The input-output relationship of the continuous Dryden wind turbulence model is as follows:

[0023] in, The translational wind velocity vector, The angular turbulence rate vector, For the Dryden wind turbulence model operator, For flight altitude, For airspeed, The direction cosine matrix from the machine system to the navigation system, Gaussian random driving noise; S22. Convert the angular turbulence rate into a triaxial equivalent angular acceleration disturbance:

[0024] in, This is the three-axis equivalent angular acceleration perturbation vector. This is the mapping matrix from angular turbulence rate to angular acceleration perturbation; S23. The nonlinear inertial coupling term and the triaxial equivalent angular acceleration disturbance are combined to form a lumped disturbance, as follows:

[0025] in, This is the lumped perturbation vector. This is a nonlinear inertial coupling term. This is the angular velocity vector.

[0026] Further, step S3 includes: S31. Decompose the lumped perturbation into the sum of the structured learning term and the equivalent residual term, and establish the quantized high-order learning extended state observer as follows:

[0027] in, For the basis function matrix, For an unknown ideal weight vector, This is the residual approximation error; S32. For any attitude channel, define The sixteen-dimensional basis function vector is used as follows:

[0028] S33. Establish attitude state estimation channels, angular velocity state estimation channels, and residual estimation channels, and quantize the high-order learning extended state observer to satisfy:

[0029] in, This is the attitude state estimation vector. This is the actual attitude angle vector. This is the angular velocity state estimation vector. This is the actual angular velocity vector. It is a diagonal gain matrix. To control the input vector, For the basis function matrix, To finally smooth the learned weight vector, This is the residual estimation vector. Here is the gain matrix of the positive definite observer. The residual leakage gain matrix; S34. Establish a filter differential channel. To avoid discontinuous quantization of angular velocity from direct differentiation, the following filter differential state is introduced:

[0030] in, This is the filtered angular velocity state vector. The filtering time constant matrix, The filtered differential output vector serves as a substitute signal for angular acceleration. S35. Establish intermediate learning weight channels and define the first... The original learning residuals for the channel are as follows:

[0031] in, For the first Channel original learning residual, For the first Channel filter differential output, For the first Channel control gain, for matrix The diagonal elements, For the first Channel actual actuator input, For the first Channel basis function row vectors For the first Learning weight vectors in the middle of the channel; S36. Implement dead zone processing on the original learning residuals as follows:

[0032] in, For the first Residual after dead zone handling in the channel For the first Channel dead zone threshold, It is a symbolic function; S37. Apply a smooth bounded mapping to the residuals after dead zone processing, as follows:

[0033] in, For the first Smooth, bounded learning residuals For the first Channel smoothing mapping limit; S38. Establish the final smoothed learning weight channel, wherein the intermediate learning weights and the final smoothed learning weights are updated according to the following formulas:

[0034]

[0035] in, For the first Channel intermediate weights learning gain matrix, For the first Channel intermediate weight leakage coefficient, For the first Channel smoothing weight gain matrix, For the first Channel smoothing weight leakage coefficient; S39. Establish the perturbation estimation output channel. The observer's original perturbation estimate is:

[0036] in, For the first Original channel disturbance estimate; S310. Apply amplitude limiting to the original disturbance estimate to obtain the actual output as follows:

[0037] in, For the first Perturbation estimate after channel limiting The limit value is saturation function; S311, Order When the limit is not activated This allows the basis function to learn the structured components of the perturbation online, the residual state to compensate for the remaining approximation error and unmodeled terms, and the second-level smoothing weights to suppress the rapid parameter fluctuations caused by quantization switching.

[0038] Further, step S4 includes: S41. Set the desired attitude trajectory and define the actual tracking error and quantized measurable error as follows:

[0039]

[0040]

[0041] in, Let be the desired attitude trajectory vector. For the desired roll angle, For the desired pitch angle, For the desired yaw angle, This is the actual attitude tracking error vector. This represents the vector of the actual attitude tracking error rate of change. The actual attitude angle vector, This is the actual angular velocity vector. To transform the measurable attitude error vector, To quantize the vector of measurable attitude error change rate, To quantize the attitude angle vector, To quantize the angular velocity vector, This is the attitude angle quantization error vector. This is the angular velocity quantization error vector; S42. Define the actual sliding mode variables as follows:

[0042] in, This is the diagonal matrix of sliding mode gain; S43. Define the quantization sliding mode variable as follows:

[0043] in, Quantize the deviation vector for the sliding mode variable. ; S44. Determine the sliding mode variable quantization error boundary. And construct measurable auxiliary terms as follows:

[0044] in, For the first Channel sliding mode variable quantization deviation, Quantize the step size for angular velocity. Sliding mode gain matrix The diagonal elements, The step size for attitude angle quantization. For the first Channel measurable auxiliary items, For the first Channel quantization measures the rate of change of attitude error. For the first Desired angular acceleration of the channel; S45. Define the quantized effective input gain as follows:

[0045] in, For the first The effective input gain after channel quantization, and the unknown positive lower bound of the effective input gain are: ,make And introduce estimates ,in, For the first Channel adaptive parameters, For the first Channel adaptive parameter estimates; S46. Based on the quantized sliding mode variable, measurable auxiliary term, and lumped disturbance estimate, construct the continuous control command as follows:

[0046] in, The limit value is The saturation function, For the first Channel robust gain, It is the hyperbolic tangent function. For the first Channel quantization sliding mode variable, For the first Channel smoothing parameters, For the first Channel adaptive parameter estimates, For the first Channel control auxiliary variables, For the first Channel denominator regularization parameter; S47. Establish an adaptive parameter update law and construct an adaptive quantization state feedback controller, as follows:

[0047] in, For the first Rate of change of channel adaptive parameter estimates To project the estimated value to the interval The projection operator guarantees , For the first Channel adaptive learning rate, For the first Channel adaptive leakage coefficient, For adaptive parameter estimation lower bound, This provides an upper bound for adaptive parameter estimation.

[0048] Furthermore, in step S5, a fixed threshold periodic event triggering mechanism is designed, including: S51, The preset event evaluation time is... In each At this point, the input quantizer is used to calculate the quantization candidate control command, as follows:

[0049] in, For the input quantization operator, For the first Channel at time Continuous control commands; S52. Compare the quantized candidate control instructions with the actuator's currently held instructions, and update the actuator instructions according to a fixed threshold rule, as follows:

[0050] in, For the first Channel at time Updated executor instructions, For the first Channel at time The executor retains the instructions before the update. For the first Fixed trigger threshold for the channel; S53. Set the initial executor instructions as follows:

[0051] in, For the first Executor instructions at the initial moment of the channel. For the first Quantization candidate control commands at the initial moment of the channel; S54. Hold the actuator instruction at zero order between adjacent evaluation times, as follows:

[0052] in, For the first Channel at time The actual actuator input; S55, Define sampling mismatch And keep the input equivalent as:

[0053] in, Indicates the first Channel sampling mismatch For the first Channel at time Continuous control commands, For the first Channel and the Bounded coefficients related to each evaluation moment.

[0054] Further, step S6 includes: S61. Define the observation error as follows:

[0055] S62. Define the intermediate weight error as follows:

[0056] in, For the first The ideal weight vector for the channel is unknown; S63. Define the final weight error as follows:

[0057] in, For the first The channel ultimately learns a smooth weight vector; S64. When the ideal weights are bounded, and the equivalent residuals and their rates of change are bounded, the dynamic equation for the intermediate weight error is established as follows:

[0058] in, For the first Channel intermediate weight leakage coefficient, For the first The bounded perturbation term of the channel intermediate weight error. For the first Channel smoothing mapping limit; S65. Define the asymptotic upper bound of the intermediate weight error as follows:

[0059] in, For the first Upper bound of the channel intermediate weight error disturbance term; S66. Establish the final weight error dynamic equation as follows:

[0060] in, For the first The channel smoothing weight leakage coefficient; therefore, the intermediate weights, final smoothing weights, observer estimation error, and perturbation estimation error are all uniformly and ultimately bounded. S67, regarding the first The attitude channel constructs the Lyapunov function of the attitude tracking subsystem as follows:

[0061] in, For the first Channel adaptive parameter estimation error, ; S68. Determine the control parameter conditions that make the closed-loop signals consistent and ultimately bounded, as follows:

[0062] in, For the first Positive lower bound of the channel's effective input gain. For the first Channel control gain parameters For the first Upper bound of channel lumped disturbance and quantization error synthesis; S69. Considering the bounded realization mismatch caused by state quantization, there exists and make Therefore, we get:

[0063] in, For the first Convergence rate coefficients of the channel Lyapunov function. For the first Bounded residual coefficients of the channel Lyapunov function For the first Initial values ​​for the Lyapunov function of the channel; S610, Set the actual sliding mode variables and attitude tracking errors to satisfy the following:

[0064]

[0065] in, For the first Actual sliding mode variables of the channel For the first Channel attitude tracking error, For the first Channel sliding mode gain; S611, Since all triggering times belong to Based on the preset periodic evaluation interval, the minimum time interval between adjacent triggering moments is determined, meaning that any two adjacent triggering moments must satisfy the following:

[0066] Therefore, an infinite number of triggers is impossible within a finite timeframe, thus ruling out Zeno's behavior; among them, For the first Channel 1 Next trigger time For the first Channel 1 Next trigger time For the first Channel 1 The evaluation sequence number corresponding to the next trigger. For the first Channel 1 The evaluation sequence number corresponding to the next trigger. This is the event assessment cycle.

[0067] Compared with the prior art, the present invention has the following advantages: 1. The present invention provides a flight life-saving device event-triggered attitude control method based on a quantized high-order learning extended state observer. By decomposing the lumped disturbance into a structured learning term and an equivalent residual term, and constructing a quantized high-order learning extended state observer including an attitude state estimation channel, an angular velocity state estimation channel, a residual estimation channel, a filtering differential channel, an intermediate learning weight channel, and a final smoothing learning weight channel, the method achieves online identification of structured components and real-time compensation of residual approximation errors in the lumped disturbance caused by continuous wind turbulence and nonlinear inertial coupling.

[0068] 2. The present invention provides a flight life-saving device event-triggered attitude control method based on a quantized high-order learning extended state observer. Through a high-order two-stage learning process consisting of rapid learning of intermediate weights and smoothing of final weights, and combined with dead zone processing, smooth bounded mapping, leakage term correction and disturbance estimation output limiting of the learning residuals, the method effectively suppresses weight abrupt changes and parameter drift caused by quantization switching, filtering differentiation and basis function approximation errors.

[0069] 3. The present invention provides a flight life-saving vehicle event-triggered attitude control method based on a quantized high-order learning extended state observer. By explicitly introducing equivalent mathematical descriptions of state quantization error, input quantization error, sampling mismatch, fixed trigger threshold, and zero-order hold error in the observer state correction, controller construction, and event trigger discrimination, the consistency between the theoretical model and the digital control implementation is achieved.

[0070] 4. The present invention provides a flight life preserver event-triggered attitude control method based on a quantized high-order learning extended state observer. By comparing the deviation between the quantized candidate control command and the current hold command of the actuator at a preset periodic evaluation time, and selectively updating the actuator command according to a fixed threshold rule, the number of control command transmissions and actuator update times are significantly reduced, and Zeno behavior is structurally eliminated by the positive periodic evaluation interval.

[0071] 5. The present invention provides a flight life-saving vehicle event-triggered attitude control method based on a quantized high-order learning extended state observer. By constructing a Lyapunov function containing an observation learning subsystem and an attitude tracking subsystem, and determining the control parameter conditions that make the closed-loop signal consistent and ultimately bounded, the closed-loop stability is strictly guaranteed under the combined action of compound disturbances, state quantization, input quantization, and event-triggered zero-order hold.

[0072] In summary, existing extended state observers rely solely on a single extended state to reconstruct all complex disturbances, making it difficult to balance disturbance reconstruction accuracy and measurement noise suppression under conditions of continuous time-varying wind turbulence and nonlinear inertial coupling. Furthermore, they fail to explicitly distinguish and uniformly model the multiple errors caused by state quantization, input quantization, sampling mismatch, and event-triggered zero-order hold, leading to unreasonable closed-loop gain selection, insufficient disturbance compensation, and frequent control input updates when designing the observer, controller, and event triggering mechanism separately. The technical solution of this invention decomposes the lumped disturbance into structured learning terms and equivalent residual terms, establishing a high-order two-stage learning structure. Simultaneously, it performs unified modeling and closed-loop stability analysis of state quantization, input quantization, sampling mismatch, and event-triggered zero-order hold, ensuring that observation learning errors, attitude tracking errors, and internal signals remain consistently bounded under resource-constrained conditions. Therefore, the technical solution of this invention solves the problems of insufficient reconstruction capability for complex lumped disturbances, lack of explicit handling of state and input quantization and event triggering errors, and excessive resource burden on communication and actuators due to frequent control command updates in existing technologies.

[0073] Based on the above reasons, this invention can be widely applied in fields such as unmanned aerial vehicle control, intelligent maritime rescue equipment, and resource-constrained embedded control systems. Attached Figure Description

[0074] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0075] Figure 1 This is a flowchart of the method of the present invention; Figure 2 A schematic diagram of the structure and attitude coordinates of a flight life-saving device; Figure 3 The results of roll, pitch, and yaw attitude angle tracking are shown in the figure. Figure 4 Figures showing the tracking errors for roll, pitch, and yaw attitude angles; Figure 5The results of roll, pitch, and yaw rate tracking are shown in the figure. Figure 6 Figures showing the tracking errors for roll, pitch, and yaw angular velocities; Figure 7 A comparison chart of lumped perturbation estimation between Q-HOLESO and traditional ESO; Figure 8 A comparison chart of the perturbation estimation errors and root mean square errors of Q-HOLESO and traditional ESO; Figure 9 This is a graph showing the event intervals under a fixed threshold strategy. Figure 10 Maintain an input comparison map for continuous control commands, quantized candidate commands, and event triggers; Figure 11 The graph shows the online evolution of the final smoothed learning weight norm for the three channels. Detailed Implementation

[0076] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0077] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product or device.

[0078] like Figure 1 As shown, this invention provides a flight life preserver event-triggered attitude control method based on a quantized high-order learning extended state observer, comprising: S1. Establish a three-axis attitude dynamics model based on the roll, pitch and yaw motions of the flight life-saving vehicle, and uniformly quantize the attitude angle, angular velocity and continuous control commands to obtain quantized state signals and quantized candidate control commands. S2. The angular turbulence rate of the flight rescue vehicle during near-sea surface flight is generated using the continuous Dryden wind turbulence model. The angular turbulence rate is converted into a three-axis equivalent angular acceleration disturbance and together with the nonlinear inertial coupling term to form a lumped disturbance. S3. Represent the lumped perturbation as the sum of the structured learning term and the equivalent residual term, and establish a quantized high-order learning extended state observer; the observer includes an attitude state estimation channel, an angular velocity state estimation channel, a residual estimation channel, a filtering differential channel, an intermediate learning weight channel, and a final smoothing learning weight channel. S4. Construct sliding mode variables and an adaptive quantized state feedback controller based on the quantized attitude angle, quantized angular velocity, desired attitude trajectory and the lumped disturbance estimate, generate continuous control commands, and limit and quantize the continuous control commands. S5. At the preset periodic evaluation time, compare the quantized candidate control command with the current hold command of the actuator. When the absolute value of the difference between the two is not less than the fixed trigger threshold, update the actuator command; otherwise, maintain the original command and perform zero-order hold on the actuator command between adjacent evaluation times. S6. Construct the Lyapunov functions for the observation learning subsystem and the attitude tracking subsystem, determine the control parameter conditions that make the closed-loop signal consistent and eventually bounded, and eliminate Zeno behavior according to the preset periodic evaluation interval.

[0079] In a specific implementation, as a preferred embodiment of the present invention, step S1 includes: S11, the flight life-saving vehicle includes multiple propulsion units arranged symmetrically, such as Figure 2 As shown, the airframe configuration, inertial coordinate system, and roll, pitch, and yaw attitude directions of the flight rescue vehicle are given. Each propulsion unit generates thrust and reaction forces in the roll, pitch, and yaw directions to control the torques. During attitude adjustment, the rigid body rotation of the flight rescue vehicle plays a dominant role. Therefore, the attitude motion of the flight rescue vehicle can be described as a quadrotor rigid body rotational dynamics model. Based on the quadrotor rigid body rotational dynamics model, the rotational motion of the flight rescue vehicle system is expressed as:

[0080] in, For roll angle, For the roll angular velocity, For roll acceleration, The pitch angle, Pitch angular velocity, For pitch acceleration, Yaw angle Yaw angular velocity, Yaw acceleration, The moment of inertia of the rolling channel. For the moment of inertia of the pitch channel, For the moment of inertia of the yaw channel, , , For three-axis control input; S12. Write the rotational dynamics model of the quadrotor rigid body in state-space form as follows:

[0081] in, The attitude angle vector, It is the angular velocity vector. It is a diagonal gain matrix. To control the input vector, This is the lumped disturbance vector; S13. Perform uniform quantization on the attitude angle and angular velocity respectively to obtain the uniform quantized signals of the attitude angle and angular velocity as follows:

[0082] in, For the first Channel quantization attitude angle, The step size for attitude angle quantization. For the first Actual attitude angle of the channel For the first Channel quantization angular velocity, Quantize the step size for angular velocity. For the first Actual angular velocity of the channel This is for rounding up to the nearest integer. S14. Set the quantization error to meet the following conditions:

[0083] S15, Continuous control command Input quantization is performed, and the following quantization candidate instructions are obtained:

[0084] in, For the first Channel quantization candidate control instructions, For input quantization operator, To input the quantization step size, For the first Channel continuous control commands, The switching threshold for quantizing equivalent representation; S16. The quantization candidate instruction is equivalently represented as:

[0085] in, For the first Channel time-varying gain, , It is an unknown and bounded time-varying gain, and its sign remains unchanged during quantization. For the first Bounded time-varying offset of the channel ; S17, due to ,when hour, It is also bounded, from which we can deduce... It is bounded, making ,in, For the first Upper bound of channel offset; S18, due to The lower bound of is uncertain and changes over time; therefore, an adaptive estimation method is used to estimate it. To avoid singular difficulties arising from estimating zero, therefore... The lower limit is used with time-varying gain It means that, among them, For time-varying gain The lower boundary.

[0086] In a specific implementation, as a preferred embodiment of the present invention, step S2 includes: S21. Establish a continuous Dryden wind turbulence model, and use the continuous Dryden wind turbulence model to generate the angular turbulence rate of the flight rescue vehicle during near-sea surface flight. The input-output relationship of the continuous Dryden wind turbulence model is as follows:

[0087] in, The translational wind velocity vector, The angular turbulence rate vector, For the Dryden wind turbulence model operator, For flight altitude, For airspeed, The direction cosine matrix from the machine system to the navigation system, Gaussian random driving noise; S22. Convert the angular turbulence rate into a triaxial equivalent angular acceleration disturbance:

[0088] in, This is the three-axis equivalent angular acceleration perturbation vector. This is the mapping matrix from angular turbulence rate to angular acceleration perturbation; S23. The nonlinear inertial coupling term and the triaxial equivalent angular acceleration disturbance are combined to form a lumped disturbance, as follows:

[0089] in, This is the lumped perturbation vector. This is a nonlinear inertial coupling term. This is the angular velocity vector.

[0090] In a specific implementation, as a preferred embodiment of the present invention, step S3 includes: S31. Decompose the lumped perturbation into the sum of the structured learning term and the equivalent residual term, and establish the quantized high-order learning extended state observer as follows:

[0091] in, For the basis function matrix, For an unknown ideal weight vector, This is the residual approximation error; S32. For any attitude channel, define The sixteen-dimensional basis function vector is used as follows:

[0092] S33. Establish attitude state estimation channels, angular velocity state estimation channels, and residual estimation channels, and quantize the high-order learning extended state observer to satisfy:

[0093] in, This is the attitude state estimation vector. This is the actual attitude angle vector. This is the angular velocity state estimation vector. This is the actual angular velocity vector. It is a diagonal gain matrix. To control the input vector, For the basis function matrix, To finally smooth the learned weight vector, This is the residual estimation vector. Here is the gain matrix of the positive definite observer. The residual leakage gain matrix; S34. Establish a filter differential channel. To avoid discontinuous quantization of angular velocity from direct differentiation, the following filter differential state is introduced:

[0094] in, This is the filtered angular velocity state vector. The filtering time constant matrix, The filtered differential output vector serves as a substitute signal for angular acceleration. S35. Establish intermediate learning weight channels and define the first... The original learning residuals for the channel are as follows:

[0095] in, For the first Channel original learning residual, For the first Channel filter differential output, For the first Channel control gain, for matrix The diagonal elements, For the first Channel actual actuator input, For the first Channel basis function row vectors For the first Learning weight vectors in the middle of the channel; S36. Implement dead zone processing on the original learning residuals as follows:

[0096] in, For the first Residual after dead zone handling in the channel For the first Channel dead zone threshold, It is a symbolic function; S37. Apply a smooth bounded mapping to the residuals after dead zone processing, as follows:

[0097] in, For the first Smooth, bounded learning residuals For the first Channel smoothing mapping limit; S38. Establish the final smoothed learning weight channel, wherein the intermediate learning weights and the final smoothed learning weights are updated according to the following formulas:

[0098]

[0099] in, For the first Channel intermediate weights learning gain matrix, For the first Channel intermediate weight leakage coefficient, For the first Channel smoothing weight gain matrix, For the first Channel smoothing weight leakage coefficient; S39. Establish the perturbation estimation output channel. The observer's original perturbation estimate is:

[0100] in, For the first Original channel disturbance estimate; S310. Apply amplitude limiting to the original disturbance estimate to obtain the actual output as follows:

[0101] in, For the first Perturbation estimate after channel limiting The limit value is saturation function; S311, Order When the limit is not activated This allows the basis function to learn the structured components of the perturbation online, the residual state to compensate for the remaining approximation error and unmodeled terms, and the second-level smoothing weights to suppress the rapid parameter fluctuations caused by quantization switching.

[0102] In a specific implementation, as a preferred embodiment of the present invention, step S4 includes: S41. Set the desired attitude trajectory and define the actual tracking error and quantized measurable error as follows:

[0103]

[0104]

[0105] in, Let be the desired attitude trajectory vector. For the desired roll angle, For the desired pitch angle, For the desired yaw angle, This is the actual attitude tracking error vector. This represents the vector of the actual attitude tracking error rate of change. The actual attitude angle vector, This is the actual angular velocity vector. To transform the measurable attitude error vector, To quantize the vector of measurable attitude error change rate, To quantize the attitude angle vector, To quantize the angular velocity vector, This is the attitude angle quantization error vector. This is the angular velocity quantization error vector; S42. Define the actual sliding mode variables as follows:

[0106] in, This is the diagonal matrix of sliding mode gain; S43. Define the quantization sliding mode variable as follows:

[0107] in, Quantize the deviation vector for the sliding mode variable. ; S44. Determine the sliding mode variable quantization error boundary. And construct measurable auxiliary terms as follows:

[0108] in, For the first Channel sliding mode variable quantization deviation, Quantize the step size for angular velocity. Sliding mode gain matrix The diagonal elements, The step size for attitude angle quantization. For the first Channel measurable auxiliary items, For the first Channel quantization measures the rate of change of attitude error. For the first Desired angular acceleration of the channel; S45. Define the quantized effective input gain as follows:

[0109] in, For the first The effective input gain after channel quantization, and the unknown positive lower bound of the effective input gain are: ,make And introduce estimates ,in, For the first Channel adaptive parameters, For the first Channel adaptive parameter estimates; S46. Based on the quantized sliding mode variable, measurable auxiliary term, and lumped disturbance estimate, construct the continuous control command as follows:

[0110] in, The limit value is The saturation function, For the first Channel robust gain, It is the hyperbolic tangent function. For the first Channel quantization sliding mode variable, For the first Channel smoothing parameters, For the first Channel adaptive parameter estimates, For the first Channel control auxiliary variables, For the first Channel denominator regularization parameter; S47. Establish an adaptive parameter update law and construct an adaptive quantization state feedback controller, as follows:

[0111] in, For the first Rate of change of channel adaptive parameter estimates To project the estimated value to the interval The projection operator guarantees , For the first Channel adaptive learning rate, For the first Channel adaptive leakage coefficient, For adaptive parameter estimation lower bound, This provides an upper bound for adaptive parameter estimation.

[0112] In a specific implementation, as a preferred embodiment of the present invention, step S5 includes a fixed threshold periodic event triggering mechanism, comprising: S51, The preset event evaluation time is... In each At this point, the input quantizer is used to calculate the quantization candidate control command, as follows:

[0113] in, For the input quantization operator, For the first Channel at time Continuous control commands; S52. Compare the quantized candidate control instructions with the actuator's currently held instructions, and update the actuator instructions according to a fixed threshold rule, as follows:

[0114] in, For the first Channel at time Updated executor instructions, For the first Channel at time The executor retains the instructions before the update. For the first Fixed trigger threshold for the channel; S53. Set the initial executor instructions as follows:

[0115] in, For the first Executor instructions at the initial moment of the channel. For the first Quantization candidate control commands at the initial moment of the channel; S54. Hold the actuator instruction at zero order between adjacent evaluation times, as follows:

[0116] in, For the first Channel at time The actual actuator input; S55, Define sampling mismatch And keep the input equivalent as:

[0117] in, Indicates the first Channel sampling mismatch For the first Channel at time Continuous control commands, For the first Channel and the Bounded coefficients related to each evaluation moment.

[0118] In this embodiment, the designed fixed threshold periodic event triggering mechanism clearly distinguishes between continuous control instructions, quantized candidate instructions, and the actual actuator input after zero-order hold.

[0119] In a specific implementation, as a preferred embodiment of the present invention, step S6 includes: S61. Define the observation error as follows:

[0120] S62. Define the intermediate weight error as follows:

[0121] in, For the first The ideal weight vector for the channel is unknown; S63. Define the final weight error as follows:

[0122] in, For the first The channel ultimately learns a smooth weight vector; S64. When the ideal weights are bounded, and the equivalent residuals and their rates of change are bounded, the dynamic equation for the intermediate weight error is established as follows:

[0123] in, For the first Channel intermediate weight leakage coefficient, For the first The bounded perturbation term of the channel intermediate weight error. For the first Channel smoothing mapping limit; S65. Define the asymptotic upper bound of the intermediate weight error as follows:

[0124] in, For the first Upper bound of the channel intermediate weight error disturbance term; S66. Establish the final weight error dynamic equation as follows:

[0125] in, For the first The channel smoothing weight leakage coefficient; therefore, the intermediate weights, final smoothing weights, observer estimation error, and perturbation estimation error are all uniformly and ultimately bounded. S67, regarding the first The attitude channel constructs the Lyapunov function of the attitude tracking subsystem as follows:

[0126] in, For the first Channel adaptive parameter estimation error, ; S68. Determine the control parameter conditions that make the closed-loop signals consistent and ultimately bounded, as follows:

[0127] in, For the first Positive lower bound of the channel's effective input gain. For the first Channel control gain parameters For the first Upper bound of channel lumped disturbance and quantization error synthesis; S69. Considering the bounded realization mismatch caused by state quantization, there exists and make Therefore, we get:

[0128] in, For the first Convergence rate coefficients of the channel Lyapunov function. For the first Bounded residual coefficients of the channel Lyapunov function For the first Initial values ​​for the Lyapunov function of the channel; S610, Set the actual sliding mode variables and attitude tracking errors to satisfy the following:

[0129]

[0130] in, For the first Actual sliding mode variables of the channel For the first Channel attitude tracking error, For the first Channel sliding mode gain; S611, Since all triggering times belong to Based on the preset periodic evaluation interval, the minimum time interval between adjacent triggering moments is determined, meaning that any two adjacent triggering moments must satisfy the following:

[0131] Therefore, an infinite number of triggers is impossible within a finite timeframe, thus ruling out Zeno's behavior; among them, For the first Channel 1 Next trigger time For the first Channel 1 Next trigger time For the first Channel 1 The evaluation sequence number corresponding to the next trigger. For the first Channel 1 The evaluation sequence number corresponding to the next trigger. This is the event assessment cycle.

[0132] In this embodiment, a Lyapunov function incorporating the sliding surface, adaptive parameter estimation error, and disturbance estimation error is constructed to perform stability analysis on the closed-loop system. This demonstrates that all state variables and intermediate signals in the closed-loop system are uniformly and eventually bounded, and that the attitude tracking error converges to a preset residual domain. Theoretical analysis rigorously proves the uniform and eventually bounded nature of all closed-loop signals and clarifies the boundedness of internal signals and quantization errors.

[0133] Example To verify the effectiveness of the present invention, this embodiment uses MATLAB for computer simulation research. The simulation object is the flight life-saving device of Liaoning Provincial Key Laboratory of Maritime Emergency Rescue - Dalian Maritime University, with the following parameter settings: The initial angle and angular velocity of the actual controlled object are both The designed controller is used to control the flight life-saving device. The ideal attitude angle of the flight life-saving device is set as follows: In this embodiment, the moment of inertia of the flight life-saving device is taken as... , Attitude angular quantization step size Angular velocity quantization step size Input quantization step size Sliding mode gain selection , Robust gain , , , , Adaptive law gain selection , and restrictions The amplitude of the control auxiliary variable is limited to... The continuous control command limit is Observer gain is taken , , Learning and smoothing gain selection , Leakage coefficient is taken as , , The filter time constant is taken as follows: The disturbance estimation limit is... The smooth learning residual limit is The flight altitude of the continuous Dryden wind turbulence model is set to... Flight speed set to The initial direction cosine matrix is ​​the identity matrix. Event evaluation period. Fixed trigger threshold The simulation duration is .

[0134] Figure 3 The display shows that the actual attitude angles of the three channels track the expected trajectory after the initial transient, and the quantized attitude angles remain close to the actual attitude angles. The local magnification results show that the deviations between the expected, actual and quantized attitude signals are small in the steady state stage. Figure 4 Further, it is shown that the three attitude angle tracking errors converge to a bounded neighborhood near zero after the initial non-zero error. Figure 5 The roll, pitch, and yaw angular velocities are displayed to track the desired angular velocity after the initial transient, and the quantized angular velocity is kept close to the actual angular velocity. Figure 6 This indicates that the tracking errors of the three channels are all bounded and approach a small neighborhood near zero, verifying the dynamic tracking performance under the combined effects of continuous wind disturbance, state and input quantization, and event-triggered zero-order hold. Figure 7 Compare the actual lumped disturbance, the Q-HOLESO estimate, and the traditional ESO estimate. Both observers face the same controlled object, Dryden wind excitation, and controller and actuator inputs. Q-HOLESO identifies the structured components of the disturbance online through two-stage learning and compensates for residual approximation errors and unmodeled dynamics with residual states, thus more closely approximating the main trends of the actual lumped disturbance in the three channels. Figure 8 Give The disturbance estimation error within the interval and the root mean square error for each channel. The root mean square errors of Q-HOLESO in the roll, pitch, and yaw channels are respectively... , and Traditional ESOs are respectively , and Corresponding reduction , and . Figure 9 The event intervals generated by the fixed threshold event triggering mechanism are all... an integer multiple of and not less than This is consistent with the aforementioned Zeno behavior exclusion analysis. Figure 10 This displays the relationship between continuous control commands, quantization candidate commands, and event-triggered zero-order hold inputs. The quantization and hold signals exhibit step characteristics, and the hold input follows the candidate commands with a bounded deviation. Simulation and Under these conditions, each channel needs to be updated during periodic transmission. Next, three channels in total Next; the roll, pitch, and yaw channels in this embodiment are updated respectively. Second-rate, Subsequent The corresponding update reduction rate is [number] times. , and Total number of updates: The overall reduction rate was [number] times. . Figure 11 The results show that the final smoothed learning weight norms of the three channels remain bounded throughout the entire simulation period. In the initial stage, the weight norms increase as the structured components of the disturbances are identified during the intermediate learning process, and then gradually evolve according to continuous Dryden wind disturbances and inertial coupling. The curves do not show continuous high-frequency abrupt changes or parameter drift, indicating that the second-level smoothing mechanism effectively suppresses the rapid changes in intermediate weights.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A flight life preserver event-triggered attitude control method based on a quantized high-order learning extended state observer, characterized in that, include: S1. Establish a three-axis attitude dynamics model based on the roll, pitch and yaw motions of the flight life-saving vehicle, and uniformly quantize the attitude angle, angular velocity and continuous control commands to obtain quantized state signals and quantized candidate control commands. S2. The angular turbulence rate of the flight rescue vehicle during near-sea surface flight is generated using the continuous Dryden wind turbulence model. The angular turbulence rate is converted into a three-axis equivalent angular acceleration disturbance and together with the nonlinear inertial coupling term to form a lumped disturbance. S3. Represent the lumped perturbation as the sum of the structured learning term and the equivalent residual term, and establish a quantized high-order learning extended state observer; the observer includes an attitude state estimation channel, an angular velocity state estimation channel, a residual estimation channel, a filtering differential channel, an intermediate learning weight channel, and a final smoothing learning weight channel. S4. Construct sliding mode variables and an adaptive quantized state feedback controller based on the quantized attitude angle, quantized angular velocity, desired attitude trajectory and the lumped disturbance estimate, generate continuous control commands, and limit and quantize the continuous control commands. S5. At the preset periodic evaluation time, compare the quantized candidate control command with the current hold command of the actuator. When the absolute value of the difference between the two is not less than the fixed trigger threshold, update the actuator command; otherwise, maintain the original command and perform zero-order hold on the actuator command between adjacent evaluation times. S6. Construct the Lyapunov functions for the observation learning subsystem and the attitude tracking subsystem, determine the control parameter conditions that make the closed-loop signal consistent and eventually bounded, and eliminate Zeno behavior according to the preset periodic evaluation interval.

2. The flight life preserver event-triggered attitude control method based on a quantized high-order learning extended state observer according to claim 1, characterized in that, Step S1 includes: S11. The attitude motion of the flight rescue vehicle is described as a quadrotor rigid body rotational dynamics model. Based on the quadrotor rigid body rotational dynamics model, the rotational motion of the flight rescue vehicle system is expressed as: in, For roll angle, For the roll angular velocity, For roll acceleration, The pitch angle, Pitch angular velocity, For pitch acceleration, Yaw angle Yaw angular velocity, Yaw acceleration, The moment of inertia of the rolling channel. For the moment of inertia of the pitch channel, For the moment of inertia of the yaw channel, , , For three-axis control input; S12. Write the rotational dynamics model of the quadrotor rigid body in state-space form as follows: in, The attitude angle vector, It is the angular velocity vector. It is a diagonal gain matrix. To control the input vector, The lumped disturbance vector; S13. Perform uniform quantization on the attitude angle and angular velocity respectively to obtain the uniform quantized signals of the attitude angle and angular velocity as follows: in, For the first Channel quantization attitude angle, The step size for attitude angle quantization. For the first Actual attitude angle of the channel For the first Channel quantization angular velocity, Quantize the step size for angular velocity. For the first Actual angular velocity of the channel This is for rounding up to the nearest integer. S14. Set the quantization error to meet the following conditions: S15, Continuous control command Input quantization is performed, and the following quantization candidate instructions are obtained: in, For the first Channel quantization candidate control instructions, For input quantization operator, To input the quantization step size, For the first Channel continuous control commands, The switching threshold for quantizing equivalent representation; S16. The quantization candidate instruction is equivalently represented as: in, For the first Channel time-varying gain, , It is an unknown and bounded time-varying gain, and its sign remains unchanged during quantization. For the first Bounded time-varying offset of the channel ; S17, due to ,when hour, It is also bounded, from which we can deduce... It is bounded, making ,in, For the first Upper bound of channel offset; S18, due to The lower bound of is uncertain and changes over time; therefore, an adaptive estimation method is used to estimate it. To avoid singular difficulties arising from estimating zero, therefore... The lower limit is used with time-varying gain It means that, among them, For time-varying gain The lower boundary.

3. The flight life preserver event-triggered attitude control method based on a quantized high-order learning extended state observer according to claim 1, characterized in that, Step S2 includes: S21. Establish a continuous Dryden wind turbulence model, and use the continuous Dryden wind turbulence model to generate the angular turbulence rate of the flight rescue vehicle during near-sea surface flight. The input-output relationship of the continuous Dryden wind turbulence model is as follows: in, The translational wind velocity vector, The angular turbulence rate vector, For the Dryden wind turbulence model operator, For flight altitude, For airspeed, The direction cosine matrix from the machine system to the navigation system, Gaussian random driving noise; S22. Convert the angular turbulence rate into a triaxial equivalent angular acceleration disturbance: in, This is the three-axis equivalent angular acceleration perturbation vector. This is the mapping matrix from angular turbulence rate to angular acceleration perturbation; S23. The nonlinear inertial coupling term and the triaxial equivalent angular acceleration disturbance are combined to form a lumped disturbance, as follows: in, This is the lumped perturbation vector. This is a nonlinear inertial coupling term. This is the angular velocity vector.

4. The flight life preserver event-triggered attitude control method based on a quantized high-order learning extended state observer according to claim 1, characterized in that, Step S3 includes: S31. Decompose the lumped perturbation into the sum of the structured learning term and the equivalent residual term, and establish the quantized high-order learning extended state observer as follows: in, For the basis function matrix, For an unknown ideal weight vector, This is the residual approximation error; S32. For any attitude channel, define The sixteen-dimensional basis function vector is used as follows: S33. Establish attitude state estimation channels, angular velocity state estimation channels, and residual estimation channels, and quantize the high-order learning extended state observer to satisfy: in, This is the attitude state estimation vector. This is the actual attitude angle vector. This is the angular velocity state estimation vector. This is the actual angular velocity vector. It is a diagonal gain matrix. To control the input vector, For the basis function matrix, To finally smooth the learned weight vector, For the residual estimation vector, Here is the gain matrix of the positive definite observer. The residual leakage gain matrix; S34. Establish a filter differential channel. To avoid discontinuous quantization of angular velocity from direct differentiation, the following filter differential state is introduced: in, This is the filtered angular velocity state vector. The filtering time constant matrix, The filtered differential output vector serves as a substitute signal for angular acceleration. S35. Establish intermediate learning weight channels and define the first... The original learning residuals for the channel are as follows: in, For the first Channel original learning residual, For the first Channel filter differential output, For the first Channel control gain, for matrix The diagonal elements, For the first Channel actual actuator input, For the first Channel basis function row vectors For the first Learning weight vectors in the middle of the channel; S36. Implement dead zone processing on the original learning residuals as follows: in, For the first Residual after dead zone handling in the channel For the first Channel dead zone threshold, It is a symbolic function; S37. Apply a smooth bounded mapping to the residuals after dead zone processing, as follows: in, For the first Smooth, bounded learning residuals For the first Channel smoothing mapping limit; S38. Establish the final smoothed learning weight channel, wherein the intermediate learning weights and the final smoothed learning weights are updated according to the following formulas: in, For the first Channel intermediate weights learning gain matrix, For the first Channel intermediate weight leakage coefficient, For the first Channel smoothing weight gain matrix, For the first Channel smoothing weight leakage coefficient; S39. Establish the perturbation estimation output channel. The observer's original perturbation estimate is: in, For the first Original channel disturbance estimate; S310. Apply amplitude limiting to the original disturbance estimate to obtain the actual output as follows: in, For the first Perturbation estimate after channel limiting The limit value is saturation function; S311, Order When the limit is not activated This allows the basis function to learn the structured components of the perturbation online, the residual state to compensate for the remaining approximation error and unmodeled terms, and the second-level smoothing weights to suppress the rapid parameter fluctuations caused by quantization switching.

5. The flight life preserver event-triggered attitude control method based on a quantized high-order learning extended state observer according to claim 1, characterized in that, Step S4 includes: S41. Set the desired attitude trajectory and define the actual tracking error and quantized measurable error as follows: in, Let be the desired attitude trajectory vector. For the desired roll angle, For the desired pitch angle, For the desired yaw angle, This is the actual attitude tracking error vector. This represents the vector of the actual attitude tracking error rate of change. The actual attitude angle vector, This is the actual angular velocity vector. To transform the measurable attitude error vector, To quantize the vector of measurable attitude error change rate, To quantize the attitude angle vector, To quantize the angular velocity vector, This is the attitude angle quantization error vector. This is the angular velocity quantization error vector; S42. Define the actual sliding mode variables as follows: in, This is the diagonal matrix of sliding mode gain; S43. Define the quantization sliding mode variable as follows: in, Quantize the deviation vector for the sliding mode variable. ; S44. Determine the sliding mode variable quantization error boundary. And construct measurable auxiliary terms as follows: in, For the first Channel sliding mode variable quantization deviation, Quantize the step size for angular velocity. Sliding mode gain matrix The diagonal elements, The step size for attitude angle quantization. For the first Channel measurable auxiliary items, For the first Channel quantization measures the rate of change of attitude error. For the first Desired angular acceleration of the channel; S45. Define the quantized effective input gain as follows: in, For the first The effective input gain after channel quantization, and the unknown positive lower bound of the effective input gain are: ,make And introduce estimates ,in, For the first Channel adaptive parameters, For the first Channel adaptive parameter estimates; S46. Based on the quantized sliding mode variables, measurable auxiliary terms, and lumped disturbance estimates, construct the continuous control command as follows: in, The limit value is The saturation function, For the first Channel robust gain, It is the hyperbolic tangent function. For the first Channel quantization sliding mode variable, For the first Channel smoothing parameters, For the first Channel adaptive parameter estimates, For the first Channel control auxiliary variables, For the first Channel denominator regularization parameter; S47. Establish an adaptive parameter update law and construct an adaptive quantization state feedback controller, as follows: in, For the first Rate of change of channel adaptive parameter estimates To project the estimated value to the interval The projection operator guarantees , For the first Channel adaptive learning rate, For the first Channel adaptive leakage coefficient, For adaptive parameter estimation lower bound, This provides an upper bound for adaptive parameter estimation.

6. The flight life preserver event-triggered attitude control method based on a quantized high-order learning extended state observer according to claim 1, characterized in that, In step S5, a fixed threshold periodic event triggering mechanism was designed, including: S51, The preset event evaluation time is... In each At this point, the input quantizer is used to calculate the quantization candidate control command, as follows: in, For the input quantization operator, For the first Channel at time Continuous control commands; S52. Compare the quantized candidate control instructions with the actuator's currently held instructions, and update the actuator instructions according to a fixed threshold rule, as follows: in, For the first Channel at time Updated executor instructions, For the first Channel at time The executor retains the instructions before the update. For the first Fixed trigger threshold for the channel; S53. Set the initial executor instructions as follows: in, For the first Executor instructions at the initial moment of the channel. For the first Quantization candidate control commands at the initial moment of the channel; S54. Hold the actuator instruction at zero order between adjacent evaluation times, as follows: in, For the first Channel at time The actual actuator input; S55, Define sampling mismatch And keep the input equivalent as: in, Indicates the first Channel sampling mismatch For the first Channel at time Continuous control commands, For the first Channel and the Bounded coefficients related to each evaluation moment.

7. The flight life preserver event-triggered attitude control method based on a quantized high-order learning extended state observer according to claim 1, characterized in that, Step S6 includes: S61. Define the observation error as follows: S62. Define the intermediate weight error as follows: in, For the first The ideal weight vector for the channel is unknown; S63. Define the final weight error as follows: in, For the first The channel ultimately learns a smooth weight vector; S64. When the ideal weights are bounded, and the equivalent residuals and their rates of change are bounded, the dynamic equation for the intermediate weight error is established as follows: in, For the first Channel intermediate weight leakage coefficient, For the first The bounded perturbation term of the channel intermediate weight error. For the first Channel smoothing mapping limit; S65. Define the asymptotic upper bound of the intermediate weight error as follows: in, For the first Upper bound of the channel intermediate weight error disturbance term; S66. Establish the final weight error dynamic equation as follows: in, For the first The channel smoothing weight leakage coefficient; therefore, the intermediate weights, final smoothing weights, observer estimation error, and perturbation estimation error are all uniformly and ultimately bounded. S67, regarding the first The attitude channel constructs the Lyapunov function of the attitude tracking subsystem as follows: in, For the first Channel adaptive parameter estimation error, ; S68. Determine the control parameter conditions that make the closed-loop signals consistent and ultimately bounded, as follows: in, For the first Positive lower bound of the channel's effective input gain. For the first Channel control gain parameters, For the first Upper bound of channel lumped disturbance and quantization error synthesis; S69. Considering the bounded realization mismatch caused by state quantization, there exists and make Therefore, we get: in, For the first Convergence rate coefficients of the Lyapunov function in the channel. For the first Bounded residual coefficients of the channel Lyapunov function For the first Initial values ​​for the Lyapunov function of the channel; S610, Set the actual sliding mode variables and attitude tracking errors to satisfy the following: in, For the first Actual sliding mode variables of the channel For the first Channel attitude tracking error, For the first Channel sliding mode gain; S611, Since all triggering times belong to Based on the preset periodic evaluation interval, the minimum time interval between adjacent triggering moments is determined, meaning that any two adjacent triggering moments must satisfy the following: Therefore, an infinite number of triggers is impossible within a finite timeframe, thus ruling out Zeno's behavior; among them, For the first Channel 1 Next trigger moment For the first Channel 1 Next trigger moment For the first Channel 1 The evaluation sequence number corresponding to the next trigger. For the first Channel 1 The evaluation sequence number corresponding to the next trigger. This is the event assessment cycle.