A stochastic hybrid control method for helicopters in Markov jump wind fields

Abstract

The application discloses a kind of helicopter random compound control methods under Markov jump wind field, belong to flight control technical field.Aiming at the problem that integral dispersion and instability caused by physical actuator saturation of helicopter under complex jump wind field, the method first establishes the augmented random helicopter system model considering actuator saturation and three-dimensional random Markov jump wind field;Second, design disturbance observer based on adaptive radial basis function neural network, reconstruct wind field disturbance state on-line unbiased;Then, the inner and outer loop controller is designed by using command filter backstepping method, and the inner and outer loop anti-integral saturation compensator is updated synchronously;Finally, by introducing the bottom layer of non-negative discriminant truncation mechanism and control distribution, the actuator command of each channel that satisfies physical hard limiting is solved.The application effectively absorbs nonlinear truncation error from the bottom of control law, ensures the global consistent ultimate boundedness of 4th moment of closed-loop signal, and greatly improves flight safety and trajectory tracking accuracy.

Description

Technical Field

[0001] This invention relates to the field of helicopter flight control technology, and more specifically, to a stochastic composite control method for helicopters under Markov jump wind fields. Background Technology

[0002] Single-rotor helicopters are typical multivariable, strongly coupled, underactuated nonlinear systems. In real-world flight environments, flight control faces inherent challenges due to the complex aerodynamic interactions between the main rotor flapping dynamics and the rigid body motion of the fuselage. Furthermore, in harsh low-altitude environments such as urban areas, seas, and canyons, helicopters frequently encounter sudden wind speed changes, wind shear, and turbulence, with dramatic jumps in airflow intensity being particularly prominent. According to the US military flight quality standard MIL-HDBK-1797, typical wind field disturbances (including continuous turbulence, discrete gusts, and wind shear) can be categorized into modes of varying intensities, such as mild, moderate, and severe. Because the switching between these modes in real atmospheric environments exhibits significant stochastic characteristics and physical inertia, it cannot be adequately described by continuous differential equations; characterizing it as a Markov jump system is more consistent with physical reality.

[0003] To achieve accurate trajectory tracking in complex environments, backstepping has been widely used in cascaded helicopter dynamics. Meanwhile, to estimate and offset unstructured lumped uncertainties, researchers often combine disturbance observers with adaptive neural networks. However, a key paradox exists in engineering practice: to compensate for unknown, large-amplitude wind field changes, the flight control system inevitably outputs aggressive control commands, forcing the helicopter's physical actuators into the nonlinear saturation region. Ignoring these hard constraints typically leads to uncontrollable integral saturation effects, resulting in a severe deterioration in control performance or even closed-loop system instability.

[0004] Although the command filtering backstepping method combined with the anti-integral saturation compensator can effectively alleviate the above-mentioned computational burden and saturation effect, under the stochastic Markov transition framework, how to strictly and systematically integrate the observer for external unmeasurable wind field disturbances, command filtering backstepping control, anti-integral saturation compensator, and some unknown transfer rate matrices remains an unsolved technical gap in the field of helicopter flight control. Summary of the Invention

[0005] To overcome the instability problem of the control system caused by the coupling of physical actuator saturation and unknown abrupt wind field in the existing technology, this invention proposes a stochastic composite control method for helicopters under Markov abrupt wind field based on disturbance observer. It aims to achieve high-precision spatial trajectory tracking and stable flight of single-rotor helicopters under complex three-dimensional time-varying wind field disturbance and actuator hard constraints.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A stochastic composite control method for helicopters under Markov-type abrupt wind fields, characterized in that its complete execution cycle in the flight control hardware includes the following steps:

[0008] Step 1: Flight status and external wind field perception and augmented stochastic system model construction

[0009] A global aerodynamic model of a helicopter considering actuator saturation is established. The Jacobian matrix is ​​extracted using small perturbation theory, and the complex aerodynamics is decoupled into state-dependent nominal nonlinear terms and control gain matrices. A global affine 6-DOF equation is established. Combined with a three-dimensional stochastic wind field perturbation model driven by Markov jump process (covering continuous turbulence, discrete gusts and wind shear), an augmented stochastic helicopter system model under Markov jump wind field is constructed for control purposes.

[0010] Step 2: Online observation of unknown wind field disturbances and lumped uncertainty

[0011] To address the unmeasurable disturbances caused by external abrupt wind fields and the unmodeled dynamics of the system lumped array, a disturbance observer based on an adaptive radial basis function neural network is constructed using the helicopter's position and velocity states. An intermediate observation state with continuous compensation from the neural network is designed, and an adaptive update law for network weights is introduced to reconstruct the wind field disturbance state and the system lumped uncertainty online without bias.

[0012] Step 3: Outer ring position-speed control and main rotor collective pitch command calculation

[0013] The position and velocity tracking errors, including the outer loop anti-integral saturation compensation signal, are defined. The outer loop virtual control signal with truncation compensation is generated based on the command filtering backstepping method, and the dynamic evolution equations of the outer loop anti-integral saturation compensator and the filter error compensator are updated synchronously. The theoretical main rotor collective pitch command is calculated by introducing a non-negative discriminant truncation mechanism, and the expected pitch and roll commands constrained by the safety envelope boundary are solved by combining the normalized expected thrust direction vector.

[0014] Step 4: Inner Loop Attitude-Angular Rate Control and Physical Actuator Command Allocation

[0015] The outer loop output is integrated to generate a complete inner loop attitude command vector. The attitude and angular rate tracking errors, including the inner loop anti-integral saturation compensation signal, are defined. The inner loop virtual control signal considering the nonlinear coupling of aerodynamic torque is calculated and the inner loop compensator evolution equation is updated synchronously. Finally, the flight control underlying control allocation module performs algebraic allocation to extract the longitudinal periodic pitch, lateral periodic pitch and tail rotor collective pitch commands. Together with the main rotor collective pitch command, the helicopter actuators are driven for closed-loop control after physical hardware saturation limiting.

[0016] 3. Beneficial effects

[0017] Compared with the prior art, the present invention has the following significant advantages:

[0018] Reconstructing the Mechanism of Complex Wind Fields: For abrupt atmospheric disturbances that are difficult to characterize by traditional models, this invention effectively reconstructs the unmeasurable Markov jump wind field state through an adaptive radial basis function neural network disturbance observer, which greatly improves the aircraft's ability to resist interference from external random natural environments.

[0019] Strictly avoid integral saturation: This invention deeply coordinates the instruction filtering backstepping mechanism with the anti-integral saturation mechanism, directly compensating for the nonlinear truncation constraint caused by physical actuator saturation from the bottom layer of the control law, thus avoiding actuator deadlock and divergence problems under aggressive maneuvers.

[0020] Rigorous stochastic stability guarantee: The stochastic composite control architecture proposed in this invention can systematically solve the theoretical coupling problem between the filter truncation error and the jump diffusion term under some unknown Markov transfer rates; in rigorous theoretical analysis, it can ensure that all closed-loop signals achieve the global uniformity of the fourth-order moment and the final bounded characteristic, thereby ensuring high-precision, high-safety, and stable tracking of the helicopter system throughout the entire flight envelope. Attached Figure Description

[0021] Figure 1 This is a data flow and system architecture diagram of a helicopter stochastic composite control method under Markov jump wind field in an embodiment of the present invention.

[0022] Figure 2 This is a flowchart of the overall aerodynamic modeling process of a helicopter considering actuator saturation in an embodiment of the present invention.

[0023] Figure 3 This is a schematic diagram of the three-dimensional spatial trajectory tracking of a helicopter under a complex Markov jump wind field, according to an embodiment of the present invention.

[0024] Figure 4 This is a three-dimensional position component tracking response curve of the helicopter in an embodiment of the present invention.

[0025] Figure 5 This is the Markov jump process in the embodiments of the present invention. The state transition curve.

[0026] Figure 6 This is a graph showing the variation of the three-dimensional wind speed component over time in an embodiment of the present invention.

[0027] Figure 7 This is a three-dimensional wind field estimation error response curve of the disturbance observer in an embodiment of the present invention.

[0028] Figure 8 This is a graph showing the change in the weight norm and diagonal elements of the neural network in this embodiment of the invention.

[0029] Figure 9 This is a tracking error response curve of the helicopter system status in an embodiment of the present invention.

[0030] Figure 10 This is a closed-loop response curve of the helicopter's three-dimensional attitude angle in an embodiment of the present invention.

[0031] Figure 11 This is a graph showing the variation of the four-channel control input of the helicopter in an embodiment of the present invention. Detailed Implementation

[0032] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0033] This invention proposes a stochastic composite control method for helicopters under Markov jump wind fields, such as... Figure 1 As shown, the control system designed in this invention consists of a cascaded sensing layer, a disturbance observer, an outer loop control module, an inner loop control module, and a bottom-level control allocation module. The sensing layer acquires the helicopter's physical state, which includes complex aerodynamic coupling, through sensors. Subsequently, the disturbance observer reconstructs the unmeasurable state of the Markov jump wind field and feeds it forward to the control loop. The outer and inner loop controllers sequentially combine command filtering and anti-integral saturation mechanisms to calculate virtual control commands, which are finally mapped to drive signals for the physical actuators by the bottom-level control allocation module. Its complete execution cycle in the flight control hardware includes the following steps:

[0034] Step 1: Flight status and external wind field perception acquisition

[0035] In the perception layer of the flight control system, the current system state of the helicopter is acquired through sensors. To implement the control law design, a physically constrained helicopter kinematic model, a wind field model, and an augmented stochastic system model are sequentially constructed in the flight control memory. The specific process is as follows:

[0036] like Figure 2 As shown, to obtain the accurate aerodynamic relationships required for control calculation, a global aerodynamic model of the helicopter considering actuator saturation is first established. Based on this modeling process, the Jacobian matrix is ​​extracted at the trim point using small perturbation theory, decoupling the complex aerodynamics of the helicopter into state-dependent nominal nonlinear terms and a control gain matrix. Considering the unmodeled dynamics and physical actuator saturation constraints, a global affine 6-DOF equation is established, with the specific expression as follows:

[0037]

[0038] In the formula, Let be the position vector of the helicopter in the northeast inertial frame; The Euler angle attitude vector; These represent the helicopter's roll angle, pitch angle, and yaw angle, respectively. This is the absolute velocity vector in the inertial frame; For the airspeed vector in the machine system; The wind speed vector relative to the ground within the machine system; The angular velocity vector of the machine system; Represents three-dimensional real number (Euclidean) space; Angular velocity vector The corresponding cross product matrix; The coordinate transformation matrix from the machine system to the inertial system; Here is the Euler kinematic transformation matrix; and These are the mass and inertia tensor matrices of the helicopter, respectively. It is the acceleration due to gravity; Let be the orthonormal basis vectors of the three-dimensional Euclidean space, represented here as The unit vector along the axial direction, i.e. ; The resultant aerodynamic force and resultant torque generated by each component at the balance point; Main rotor collective pitch control gain matrix; This is the lumped control gain matrix for the cyclic pitch control and the tail rotor. Main rotor collective pitch control input; The control vector includes longitudinal and lateral periodic pitch and tail rotor collective pitch. These are the parameters for the balancing points; This represents the lumped-out unmodeled dynamic error. Actual physical control input. Constrained by the upper and lower limit saturation functions of the actuator Constraints.

[0039] Furthermore, a continuous-time Markov jump process is employed. This characterizes the random switching of atmospheric conditions between light, moderate, and heavy patterns. These represent light, moderate, and heavy atmospheric environmental conditions, respectively. The total wind field disturbance under the system. From continuous turbulence Dispersed gusts Japanese wind shear It is formed by superposition, and its expression is:

[0040]

[0041] in, and These are the transformation matrices from the stable frame and the inertial frame to the mechanical frame, respectively. Driven by the Markov jump process, each wind field component satisfies the evolution laws of first-order low-pass filtering and stochastic differential equations.

[0042] Furthermore, define the system state vector. .in For location, For the speed of the machine body, For Euler angle orientation, Let ω be the angular velocity of the aircraft. Define the external wind field disturbance as... Its internal evolutionary state is defined as An augmented stochastic helicopter system model under Markov jump wind fields is constructed in the flight control memory for closed-loop control calculation:

[0043]

[0044] In the formula, This is the position kinematics mapping matrix; For the intrinsic nonlinear functions of the velocity loop system; This is the angular velocity cross-coupling matrix; This is the attitude kinematics mapping matrix; The eigenfunctions of the angular velocity loop system are nonlinear functions. Assign functions to the lumped control loop of the speed loop; Assign functions to the lumped control loop of the angular velocity ring; , These are the lumped unmodeled dynamic errors of the force and torque channels in the mechanical system, respectively. For the unknown nonlinear drift term function of the wind field associated with the Markov jump mode; The unknown diffusion coefficient matrix of the wind field; These are mutually independent three-dimensional standard Brownian motion processes.

[0045] Step 2: Online observation of unknown wind field disturbances and lumped uncertainty

[0046] Since the true external abrupt wind field state cannot be directly measured by airborne sensors, and its internal dynamic equations contain unknown nonlinear drift terms, a disturbance observer based on a radial basis function neural network is designed in the feedforward phase of the flight control calculation cycle to reconstruct the wind field disturbance state in real time. The specific process is as follows:

[0047] First, in order to make an unbiased estimate of the wind field state, the wind field disturbance state estimate is designed. and and intermediate observation states with continuous compensation via neural networks and Its equation is:

[0048]

[0049] In the formula, This is the adaptive estimation weight matrix for a radial basis function neural network; The estimated input vector for the neural network is composed of: ; In order to estimate the state The input is the radial basis function activation vector; and Choose a matrix for the structure of the objective function that satisfies and ,in and These are a third-order identity matrix and a third-order zero matrix, respectively. The time derivative of the transpose of the position kinematic mapping matrix satisfies ; Design the gain matrix for the pre-defined positive definite observer.

[0050] Subsequently, using the helicopter's position With speed The fundamental dynamic equations for constructing the observer are expressed as follows:

[0051]

[0052] In the formula, These are the observed estimates of the positional state; This is the error in position observation estimation; Design the gain matrix for a positive definite observer; This is an estimate of the wind field disturbance state.

[0053] Furthermore, in order to reasonably drive the neural network to approximate the lumped unknown objective function without direct measurement data of the internal wind field, the location observation estimation error is considered. Set the network weight adaptive update law as follows:

[0054]

[0055] In the formula, Assign a gain matrix to the observation error, satisfying ,in and The preset positive scalar learning weight parameters for the system; The learning rate matrix of the positive definite neural network; Design the gain matrix for the pre-defined positive definite observer.

[0056] Step 3: Outer ring position-speed control and main rotor collective pitch command calculation

[0057] Within the outer loop of helicopter trajectory tracking, the flight control computer generates an outer-loop virtual control law based on the position reference trajectory and calculates the main rotor collective pitch command and desired attitude angles through control allocation. The specific calculation process is as follows:

[0058] First, define the original tracking error of the outer loop position. and the original tracking error of speed Based on this, a signal containing anti-integral saturation compensation is defined. Outer ring position and velocity tracking error :

[0059]

[0060] In the formula, For the given desired location reference trajectory; For speed virtual control law The speed virtual control signal processed by the first-order low-pass command filter satisfies the following dynamic characteristics: ,in The time constant of the filter, This is the virtual speed control law before filtering.

[0061] Subsequently, based on the aforementioned tracking error and the observer state output in step 2, a virtual speed control law is designed. Virtual control signal after outer ring truncation compensation and outer ring anti-integral saturation compensation signal Its expression is:

[0062]

[0063] In the formula, The time derivative of the reference trajectory at the desired position; All are preset positive definite controller gain matrices; These are preset positive scalar robust feedback parameters; Let be a known nonnegative smooth continuous function, used to define the nonlinear growth boundary of the unmodeled dynamics, which satisfies ; Let be the Euclidean norm of a vector, satisfying the condition for any vector. Its norm ; This refers to the outer loop control deviation.

[0064] Further, outer loop control allocation is performed. The virtual control signal after offset is defined. To prevent actuator thrust limitation from causing unsolvable algebraic loops, a nonnegativity discriminative truncation mechanism is introduced at the system's underlying level. Its expression is:

[0065]

[0066] In the formula, This is the function for finding the maximum value.

[0067] Furthermore, the theoretical main rotor collective pitch command was calculated. :

[0068]

[0069] Finally, considering the physical upper and lower limits of the main rotor collective pitch actuator... The actual control input is then obtained. ,in The upper and lower limits of the saturation limiting function are, i.e. Satisfy: When Time output ,when Time output Otherwise output It itself. Calculate the normalized expected thrust direction vector. :

[0070]

[0071] Furthermore, ensure safe operation within the preset upper and lower limits of the pitch angle. and preset safe operating upper and lower limits for roll angle. In this case, the desired pitch angle command required for the inner loop can be derived from the inverse solution. And the desired roll angle command :

[0072]

[0073] In the formula, These are the unit orthonormal basis vectors in the X and Y directions of three-dimensional space, respectively. , ;

[0074] Step 4: Inner Loop Attitude-Angular Rate Control and Physical Actuator Command Allocation

[0075] Within the helicopter's inner loop, the flight control computer receives the desired attitude command output from step 3, generates a virtual angular rate control law, and completes the control allocation for cyclic pitch and tail rotor through algebraic calculation. The specific calculation process is as follows:

[0076] First, integrate the desired pitch angle command calculated in step 3. Expected roll angle command and externally given yaw angle commands Construct a complete inner loop attitude command vector Subsequently, the original tracking error of the inner loop attitude is defined. and the original tracking error of angular rate Furthermore, a signal containing inner-loop anti-integral saturation compensation is defined. Attitude and angular rate tracking error :

[0077]

[0078] In the formula, and These are the attitude and angular rate virtual control signals processed by a first-order low-pass command filter, respectively, and their dynamic characteristics satisfy... as well as ;in The filter time constant is This is the virtual control law for the angular rate before filtering.

[0079] Subsequently, based on the inner loop tracking error, a virtual control law for the angular rate was designed. Inner loop virtual control signal and inner loop anti-integral saturation compensation signal Its expression is:

[0080]

[0081] In the formula, All are system-preset positive definite controller gain matrices; These are preset positive scalar robust feedback parameters; This refers to the inner loop control deviation.

[0082] Furthermore, the control allocation of the inner-loop physical commands is performed. The low-level control allocation module of the flight control system executes algebraic allocation equations from the inner-loop virtual control signals. Extract the command vector containing longitudinal cyclic pitch, lateral cyclic pitch, and tail rotor collective pitch. Its expression is:

[0083]

[0084] In the formula, ,in For theoretical longitudinal periodic pitch, For theoretical transverse periodic pitch. This is the theoretical collective pitch of the tail rotor; This is the cyclic pitch and tail rotor lumped control gain matrix established in step 1.

[0085] After completing the closed-loop cycle from steps 1 to 4 above, analysis using a constructed stochastic Lyapunov functional shows that the designed Markov jump stochastic composite control method based on a disturbance observer can strictly guarantee that all signals in the closed-loop system satisfy the globally consistent final bounded characteristic of fourth-order moments. Even under the dual constraints of external unknown strong wind jumps and internal actuator physical saturation, the system can still effectively absorb nonlinear truncation errors, achieving high-precision three-dimensional spatial trajectory tracking and attitude stabilization under complex wind fields.

[0086] To verify the effectiveness and robustness of the above technical solution, this invention is based on the parameters of the UH-60 single-rotor helicopter and employs... Figure 2 The overall aerodynamic modeling framework shown is used to construct a high-fidelity simulation of the controlled object, and three-dimensional spatial trajectory simulation tests are carried out under a complex Markov jump wind field environment. To capture the stochastic characteristics of the real wind field, a continuous-time Markov jump process is introduced. Describes the transition between three intensity modes: ambient wind field at mild ( ), moderate ( ) and severe ( The flight modes randomly switch between three modes and encompass complex disturbances such as continuous turbulence, discrete gusts, and wind shear. Referencing the US military rotorcraft flight quality design standard ADS-33E-PRF, the flight commands... The simulation was set to low-altitude hovering: the helicopter performs a lateral circular translation with a radius of 30 meters at a height of 3 meters, while keeping the nose pointing towards the center of the circle. This mission aims to evaluate the aircraft's multi-axis cooperative control, precise trajectory tracking, and disturbance rejection capabilities under constantly changing wind direction relative to the aircraft. Simulation results are as follows... Figures 3 to 11 As shown.

[0087] Depend on Figure 3 (Diagram of three-dimensional spatial trajectory tracking) Figure 4 (The three-dimensional position component tracking response curve) shows that even when encountering strong winds and sudden changes in wind direction, this method can still achieve accurate tracking of complex three-dimensional spatial trajectories by helicopters. From... Figure 5 (State transition curve of Markov transition process) Figure 6 (The graph showing the variation of three-dimensional wind speed components over time) indicates that the Markov process... This caused drastic changes in the intensity of the composite wind field. Specifically, the peak velocity of the composite wind field reached 14.63 m / s, the average velocity was 7.11 m / s, and the maximum 1-second wind speed vector change reached 10.39 m / s. These random pulses severely tested the robustness of the closed-loop system and the wind disturbance resistance of the actuator loop.

[0088] Despite the aforementioned unpredictable random fluctuations, the adaptive neural network disturbance observer can still effectively estimate lumped disturbances. Figure 7(The error response curve of the three-dimensional wind field estimation of the perturbation observer) and Figure 8 (The graph showing the variation of the weight norm and diagonal elements of the neural network) reveals that the designed perturbation observer ensures that the estimation error converges in a very short time and effectively resists the influence of Markov mode switching. Thanks to accurate perturbation estimation, the radial basis function neural network weights exhibit stable convergence characteristics. From... Figure 9 (The tracking error response curve of the helicopter system state) shows that the errors of each loop of the system can quickly converge to the minimum neighborhood near the origin, which rigorously verifies the global uniformity and eventual boundedness of the fourth moment of the control system.

[0089] Furthermore, by Figure 10 (Closed-loop response curve of helicopter three-dimensional attitude angle) and Figure 11 As can be seen from the graph of the change in the four-channel control input of the helicopter, the anti-integral saturation compensator designed in this invention not only strictly constrains the attitude angle and control input within the preset safety boundary, but also effectively avoids integral saturation and system instability caused by command saturation truncation.

[0090] In summary, this method overcomes the engineering bottleneck of traditional helicopter control, which struggles to balance strong external disturbances from unknown Markov jumps with hard saturation of internal actuators. Through a composite design of adaptive observation and anti-saturation cascade, it significantly improves the trajectory control accuracy and physical safety of the aircraft in extreme and harsh environments.

[0091] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A stochastic composite control method for helicopters under Markov jump wind fields, characterized in that, Its complete execution cycle in the flight control hardware includes the following steps: Step 1: Flight status and external wind field perception and augmented stochastic system model construction; A global aerodynamic model of a helicopter considering actuator saturation is established. The Jacobian matrix is ​​extracted using small perturbation theory, and the complex aerodynamics is decoupled into state-dependent nominal nonlinear terms and control gain matrix. A global affine 6-DOF equation is established. Combined with a three-dimensional stochastic wind field perturbation model driven by Markov jump process, an augmented stochastic helicopter system model under Markov jump wind field is constructed for control. Step 2: Online observation of unknown wind field disturbances and lumped uncertainties; To address the unmeasurable disturbances caused by external abrupt wind fields and the unmodeled dynamics of the system lumped array, a disturbance observer based on an adaptive radial basis function neural network is constructed using the helicopter's position and velocity states. An intermediate observation state with continuous compensation from the neural network is designed, and an adaptive update law for network weights is introduced to reconstruct the wind field disturbance state and the system lumped uncertainty online without bias. Step 3: Calculation of outer ring position-speed control and main rotor collective pitch commands; Define the position and velocity tracking error including the outer loop anti-integral saturation compensation signal, generate the truncated and compensated outer loop virtual control signal based on the command filtering backstepping method, and update the dynamic evolution equations of the outer loop anti-integral saturation compensator and the filter error compensator simultaneously. The theoretical main rotor collective pitch command is calculated by introducing a non-negative discriminative truncation mechanism, and the expected pitch and roll commands constrained by the safety envelope boundary are obtained by combining the normalized expected thrust direction vector. Step 4: Inner loop attitude-angular rate control and physical actuator command allocation; The outer loop output is integrated to generate a complete inner loop attitude command vector. The attitude and angular rate tracking errors, including the inner loop anti-integral saturation compensation signal, are defined. The inner loop virtual control signal considering the nonlinear coupling of aerodynamic torque is calculated and the inner loop compensator evolution equation is updated synchronously. Finally, the flight control underlying control allocation module performs algebraic allocation to extract the longitudinal periodic pitch, lateral periodic pitch and tail rotor collective pitch commands. Together with the main rotor collective pitch command, the helicopter actuators are driven for closed-loop control after physical hardware saturation limiting.

2. The stochastic composite control method for helicopters under Markov jump wind fields according to claim 1, characterized in that, In step 1, the three-dimensional stochastic wind field disturbance model is composed of continuous turbulence, discrete gusts, and wind shear superimposed; the switching characteristics of the atmospheric environment between light, moderate, and severe modes are characterized by a continuous-time Markov jump process, specifically as follows: A1: Establish the global affine 6-DOF equations, with the following specific expression: ； In the formula, This is the absolute velocity vector in the inertial frame; Let be the coordinate transformation matrix from the body coordinate system to the inertial coordinate system, where b represents the body coordinate system and i represents the inertial coordinate system. For the airspeed vector in the machine system; The wind speed vector relative to the ground within the machine system; and These are the mass and inertia tensor matrices of the helicopter, respectively. The time derivative of the airspeed vector in the machine system is represented. This represents the time derivative of the wind speed vector relative to the ground in the machine system. Angular velocity vector The corresponding cross product matrix; The angular velocity vector of the machine system. This represents the angular acceleration vector in the machine system. Let be the position vector of the helicopter in the northeast inertial frame; Let Euler angles be the attitude vector. Represents the Euler angular rate; These represent the helicopter's roll angle, pitch angle, and yaw angle, respectively. Represents three-dimensional real number (Euclidean) space; Here is the Euler kinematic transformation matrix; It is the acceleration due to gravity; Let be the orthonormal basis vectors of the three-dimensional Euclidean space, represented here as The unit vector along the axial direction, i.e. ; The resultant aerodynamic force and resultant torque generated by each component at the balance point; Main rotor collective pitch control gain matrix; This is the lumped control gain matrix for the cyclic pitch control and the tail rotor. Main rotor collective pitch control input; The control vector includes longitudinal and lateral periodic pitch and tail rotor collective pitch. These are the parameters for the balancing points; The lumped unmodeled dynamic error; actual physical control input Constrained by the upper and lower limit saturation functions of the actuator Constraints; A2: Employ a continuous-time Markov jump process. Characterizing the random switching of atmospheric conditions between light, moderate, and severe patterns, among which, These represent the light, moderate, and heavy atmospheric environmental conditions, respectively; the total wind field disturbance under the system. From continuous turbulence Dispersed gusts Japanese wind shear It is formed by superposition, and its expression is: ； in, and , respectively, are the transformation matrices from the stable frame and the inertial frame to the mechanical frame, where s represents the stable coordinate system; each wind field component, driven by the Markov jump process, satisfies the evolution law of first-order low-pass filtering and stochastic differential equations; A3: Define the system state vector ;in For location, For the speed of the machine body, For Euler angle orientation, Let be the angular velocity of the aircraft; define the external wind field disturbance as... Its internal evolutionary state is defined as An augmented stochastic helicopter system model under Markov jump wind field is constructed in the flight control memory for closed-loop control calculation: ； In the formula, This is the position kinematics mapping matrix; For the intrinsic nonlinear functions of the velocity loop system; This is the angular velocity cross-coupling matrix; This is the attitude kinematics mapping matrix; The eigenfunctions of the angular velocity loop system are nonlinear functions. Assign functions to the lumped control loop of the speed loop; Assign functions to the lumped control loop of the angular velocity ring; , These are the lumped unmodeled dynamic errors of the force and torque channels in the mechanical system, respectively. For the unknown nonlinear drift term function of the wind field associated with the Markov jump mode; The unknown diffusion coefficient matrix of the wind field; These are mutually independent three-dimensional standard Brownian motion processes.

3. The stochastic composite control method for helicopters under Markov jump wind fields according to claim 1, characterized in that, In step 2, the network weight adaptive update law combines the location observation estimation error, the location kinematic mapping matrix, and the preset observation error allocation gain matrix for dynamic updating, so as to drive the neural network to approximate the lumped unknown objective function under the condition of no direct measurement data of the internal wind field. The specific method is as follows: Step 2.1: Estimated values ​​of wind field disturbance state and and intermediate observation states with continuous compensation via neural networks and Its equation is: ； In the formula, This is the adaptive estimation weight matrix for a radial basis function neural network; The estimated input vector for the neural network is composed of: ; In order to estimate the state The input is the radial basis function activation vector; and Choose a matrix for the structure of the objective function that satisfies and ,in and These are a third-order identity matrix and a third-order zero matrix, respectively. The time derivative of the transpose of the position kinematic mapping matrix satisfies ; Design the gain matrix for the pre-defined positive definite observer; Step 2.2: Utilizing the helicopter's position With speed The fundamental dynamic equations for constructing the observer are expressed as follows: ； In the formula, These are the observed estimates of the positional state; This is the error in position observation estimation; Design the gain matrix for a positive definite observer; This is an estimate of the wind field disturbance state; Step 2.3: Estimate error by combining location observations Set the network weight adaptive update law as follows: ； In the formula, Assign a gain matrix to the observation error, satisfying ,in and The preset positive scalar learning weight parameters for the system; The learning rate matrix of the positive definite neural network; Design the gain matrix for the pre-defined positive definite observer.

4. The stochastic composite control method for helicopters under Markov jump wind fields according to claim 1, characterized in that, In step 3, the definition of the outer loop position and speed tracking error includes the speed virtual control signal after the speed virtual control law has been processed by a first-order low-pass command filter; the dynamic evolution equation of the outer loop anti-integral saturation compensator is updated by the outer loop control deviation, specifically as follows: Step 3.1: Define the original tracking error of the outer loop position. and the original tracking error of speed Based on this, a signal containing anti-integral saturation compensation is defined. Outer ring position and velocity tracking error : ； In the formula, For the given desired location reference trajectory; For speed virtual control law The speed virtual control signal processed by the first-order low-pass command filter satisfies the following dynamic characteristics: ,in The time constant of the filter, The virtual speed control law before filtering. This represents the filtered virtual speed control signal. The first derivative with respect to time; Step 3.2: Based on the tracking error described above and the observer state output in Step 2, design the virtual speed control law. Virtual control signal after outer ring truncation compensation and outer ring anti-integral saturation compensation signal Its expression is: ； In the formula, The time derivative of the reference trajectory at the desired position; All are preset positive definite controller gain matrices; These are preset positive scalar robust feedback parameters; Let be a known nonnegative smooth continuous function, used to define the nonlinear growth boundary of the unmodeled dynamics, which satisfies ; Let be the Euclidean norm of a vector, satisfying the condition for any vector. Its norm ; This refers to the outer loop control deviation. Step 3.3: Perform outer loop control allocation; define the virtual control signal after offset. m represents the total mass of the helicopter. The control gain matrix represents the collective pitch of the main rotor. The trim control input represents the collective pitch of the main rotor. To prevent the algebraic loop from becoming unsolvable due to actuator thrust limitation, a non-negative discriminative truncation mechanism is introduced at the system's underlying level. Its expression is: ； In the formula, To find the maximum value function, Step 3.4: Calculate the theoretical main rotor collective pitch command : 。 5. The stochastic composite control method for helicopters under Markov jump wind fields according to claim 4, characterized in that, In step 3, the nonnegative discriminant truncation mechanism is used to prevent the actuator thrust limitation from causing the solution algebraic loop to be unsolvable; after solving the theoretical main rotor collective pitch command, the amplitude is limited by the physical upper and lower limits of the main rotor collective pitch actuator, and then the normalized expected thrust direction vector is calculated. The specific method is as follows: B1: Physical upper and lower limits of the main rotor collective pitch actuator The actual control input is then obtained. ,in The upper and lower limits of the saturation limiting function are, i.e. Satisfy: When Time output ,when Time output Otherwise output itself, Represent the desired control input command for the main rotor collective pitch; calculate the normalized desired thrust direction vector. : ； B2: Ensure safe operation within the preset pitch angle upper and lower limits. and preset safe operating upper and lower limits for roll angle. In this case, the desired pitch angle command required for the inner loop can be derived from the inverse solution. And the desired roll angle command : ； In the formula, These are the unit orthonormal basis vectors in the X and Y directions of three-dimensional space, respectively. , .

6. The stochastic composite control method for helicopters under Markov jump wind fields according to claim 1, characterized in that, In step 4, the definition of the inner loop attitude and angular rate tracking error includes the virtual control signals of attitude and angular rate processed by a first-order low-pass command filter; the dynamic evolution equation of the inner loop anti-integral saturation compensator is updated by the inner loop control deviation; in the helicopter's inner loop, the flight control computer receives the desired attitude command output in step 3, generates the virtual angular rate control law, and completes the control allocation of cyclic pitch and tail rotor through algebraic calculation; the specific method is as follows: Step 4.1: Integrate the desired pitch angle command obtained in Step 3 Expected roll angle command and externally given yaw angle commands Construct a complete inner loop attitude command vector Subsequently, the original tracking error of the inner loop attitude is defined. and the original tracking error of angular rate ; Step 4.2: Define the signal containing the inner loop anti-integral saturation compensation. Attitude and angular rate tracking error : ； In the formula, and These are the attitude and angular rate virtual control signals processed by a first-order low-pass command filter, respectively, and their dynamic characteristics satisfy... as well as ;in The filter time constant is The virtual control law for angular rate before filtering; Step 4.3: Design a virtual control law for angular rate based on the inner loop tracking error. Inner loop virtual control signal and inner loop anti-integral saturation compensation signal Its expression is: ； In the formula, All are system-preset positive definite controller gain matrices; These are preset positive scalar robust feedback parameters; This refers to the inner loop control deviation.

7. The stochastic composite control method for helicopters under Markov jump wind fields according to claim 6, characterized in that, In step 4, the underlying control allocation module of the flight control system uses a pre-established main rotor cyclic pitch and tail rotor lumped control gain matrix to execute an algebraic allocation equation from the inner-loop virtual control signal to obtain the theoretical control command vector required by the physical actuators. The specific method is as follows: The flight control system's underlying control allocation module executes algebraic allocation equations to allocate control signals from the inner loop physical commands. Extract the command vector containing longitudinal cyclic pitch, lateral cyclic pitch, and tail rotor collective pitch. Its expression is: ； In the formula, ,in For theoretical longitudinal periodic pitch, For theoretical transverse periodic pitch. This is the theoretical collective pitch of the tail rotor; This is the cyclic pitch and tail rotor lumped control gain matrix established in step 1.

Images