Aircraft trajectory tracking method and device based on data-driven model predictive control

Abstract

The application provides an aircraft trajectory tracking method and device based on data-driven model predictive control. Firstly, the latest aircraft state response data is captured in real time during the flight of the aircraft, and the expected roll angle reference signal of the next moment obtained by optimization and solution of a data-driven model predictive controller is used as the input of the flight control of the aircraft to realize the planar lateral trajectory tracking of the aircraft. Meanwhile, the model parameters used by the data-driven model predictive controller are identified and dynamically adjusted on-line according to the real-time captured and stored aircraft state response data. The application can capture the latest response data in real time during the flight of the aircraft, and identify and dynamically adjust the model parameters used by the model predictive controller on-line to improve the flight robustness of the aircraft under different environmental conditions. The data-driven model predictive control method can make full use of the input and output data of the controlled system and enhance the adaptive ability of the controller to the system.

Description

Technical Field

[0001] This invention relates to the field of aircraft trajectory tracking, and in particular to an aircraft trajectory tracking method and apparatus based on data-driven model predictive control. Background Technology

[0002] Thanks to the improved computing power of terminal devices, model predictive control has become one of the more common control strategies in industry. Its principle lies in using a mathematical model to predict future outputs based on the current system state and control inputs, and solving for the optimal control sequence for the future. This process is repeated continuously for rolling optimization. Its characteristics include: handling multivariable control problems, handling input-output physical constraints, and adapting to structural changes. As a widely used control algorithm in the field of real-time control, model predictive control achieves near-optimal control performance in many situations.

[0003] With the increasing application of unmanned aerial vehicles (UAVs) in both military and civilian fields, which possess advantages such as low cost, small size, and easy maneuverability, higher demands are being placed on the autonomous control capabilities of flight controllers. The frequency of model predictive control (MMC) technology in flight control system design, both domestically and internationally, is rising year by year. However, existing research often involves only a single system parameter identification process for the controlled system, treating the resulting model as an inherent property of the system without modification. During flight missions, however, UAV systems may face variable flight environments: under different operating conditions, the aerodynamic response parameters of the UAV will exhibit unpredictable changes, causing discrepancies between the actual aerodynamic characteristics and aerodynamic calculations or ground wind tunnel measurements. This leads to performance degradation of control systems designed based on nominal dynamic models. Therefore, to achieve stable and adaptive control algorithms, a framework similar to incremental training in machine learning is needed for traditional MMCs, providing a framework for dynamic model updates. Summary of the Invention

[0004] The purpose of this invention is to solve the aforementioned technical problems and provide a method and apparatus for aircraft trajectory tracking based on data-driven model predictive control. By acquiring the aerodynamic response characteristics of the aircraft through a pre-planned flight mission, a controlled sparse identification method is used to identify the dynamic model. Furthermore, the model parameters of the model predictive controller are identified and dynamically adjusted online based on real-time flight data, enabling the aircraft to better complete the target trajectory tracking task. This invention can fully utilize the input and output data of the controlled system, enhancing the controller's adaptability to the system.

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

[0006] A data-driven model predictive control-based aircraft trajectory tracking method is proposed, which involves capturing and storing the latest aircraft state response data in real time during the aircraft's flight. The aircraft state response data includes the aircraft's northward position coordinates n, eastward position coordinates e, and yaw angle ψ. g Roll angle φ, roll rate p, lateral tracking error l e With heading tracking error ψ e ;

[0007] The current aircraft state response data is used as input to the data-driven model predictive controller. The data-driven model predictive controller continuously predicts the aircraft state response data in real time for a period of time after the current moment based on the aircraft's planar lateral dynamic equation, roll dimension response equation, lateral tracking error and heading tracking error equation. Based on the predicted aircraft state response data for a period of time after the current moment, an objective function is established for optimization to obtain the optimal expected roll angle reference signal for the next moment. The aircraft performs flight tasks according to the predicted expected roll angle reference signal, thereby realizing the aircraft's planar lateral trajectory tracking.

[0008] Simultaneously, based on the real-time captured and stored aircraft status response data, the model parameters used by the data-driven model prediction controller are identified online and dynamically adjusted.

[0009] Furthermore, the aircraft is a fixed-wing aircraft, and the planar lateral dynamic equation of the fixed-wing aircraft is expressed as:

[0010] n k+1 =n k +V g cosψ gk δt

[0011] e k+1 =e k +V g sinψ gk δt

[0012] ψ gk+1 =ψ gk +(gtanφ k / V)δt

[0013] φ k+1 =φ k +p k δt

[0014] The roll-dimensional response equation is expressed as:

[0015] p k+1 =p k +(b0φ rk -a1p k -a0φk )δt

[0016] The lateral tracking error equation is expressed as:

[0017] l ek+1 =l ek +V g sinψ g δt

[0018] The heading tracking error equation is expressed as follows:

[0019] ψ ek+1 =ψ ek +(gtanφ k / V)δt

[0020] In the formula, n and e represent the northward and eastward positions of the aircraft, respectively, both with the takeoff coordinates as the origin; V g Represents the ground velocity vector; ψ g The angle between the ground velocity vector and the northward plane is represented by φ; g is the local gravitational acceleration; φ is the roll angle; V is the airspeed vector; p is the roll angular velocity; φ r is the reference signal for the desired roll angle of the input aircraft; a0, a1, and b0 are constant parameters of the model; k is the index of the sampling time, and δt represents the time difference of the sampling interval.

[0021] Furthermore, the model parameters used by the data-driven model prediction controller are predetermined by an optimizer obtained from the collected flight data using a controlled sparse identification method.

[0022] Furthermore, flight data is acquired through the following method: continuous control excitation is applied to the aircraft in a windless environment. The input control data includes multiple "2-1-1" cascaded maneuvers with different amplitudes. After each input control reaches a steady state, flight data is recorded, including a three-dimensional vector sequence of roll angle, roll rate, and the input aircraft's desired roll angle reference signal.

[0023] Furthermore, the desired roll angle reference signal output by the data-driven model predictive controller is subject to hard constraints, ensuring that the roll angle output by the controller is within a safe range.

[0024] Furthermore, the objective function includes the costs of lateral tracking error and heading tracking error, a roll angle penalty term, and a roll maneuver penalty term.

[0025] Furthermore, it also includes using a PID controller for longitudinal altitude tracking of the aircraft.

[0026] An aircraft trajectory tracking device based on data-driven model predictive control includes:

[0027] The data acquisition and storage unit is used to capture and store the latest aircraft status response data in real time during the flight of the aircraft; the aircraft status response data includes the aircraft's northward position coordinates n, eastward position coordinates e, and yaw angle ψ. g Roll angle φ, roll rate p, lateral tracking error l e With heading tracking error ψ e ;

[0028] The data-driven model predictive controller takes the aircraft's current state response data as input and continuously predicts the aircraft's state response data for a period of time after the current moment based on the aircraft's planar lateral dynamic equation, roll dimension response equation, lateral tracking error, and heading tracking error equation. It establishes an objective function based on the predicted aircraft state response data for a period of time after the current moment for optimization, and obtains the optimal expected roll angle reference signal for the next moment. The aircraft performs flight tasks according to the predicted expected roll angle reference signal, realizing the aircraft's planar lateral trajectory tracking.

[0029] The model parameter dynamic adjustment unit is used to identify and dynamically adjust the model parameters used by the data-driven model predictive controller based on the aircraft state response data stored in the data acquisition and storage unit.

[0030] Furthermore, it also includes a PID controller for longitudinal altitude tracking of the aircraft.

[0031] Furthermore, in the PID controller for longitudinal altitude control, the target pitch angle output by the controller can be limited so that both the aircraft's climb angle and glide angle are less than 10 degrees.

[0032] The beneficial effects of this invention are as follows:

[0033] 1. This invention targets the dynamic model of an aircraft and employs a "2-1-1" maneuvering combined with sparse identification method to quickly identify model parameters from data, with high accuracy and strong universality.

[0034] 2. Based on the traditional model predictive controller, this invention proposes a concept similar to incremental training in the field of machine learning and provides a technical framework for dynamic model updates, which can make full use of the input and output data of the controlled system to achieve more robust adaptive control behavior.

[0035] 3. The tracking control framework proposed in this invention combines a data-driven model predictive controller with a PID controller, which can effectively complete the three-dimensional trajectory tracking task of a fixed-wing aircraft in three-dimensional space, realize the spatial trajectory tracking task of the aircraft, and enable the aircraft's flight trajectory to maintain good convergence with the target path, thus exhibiting excellent control performance.

[0036] 4. The aircraft trajectory tracking method based on data-driven model predictive control of the present invention can be extended to more control strategies. Attached Figure Description

[0037] Figure 1 This is a control block diagram of the data-driven model predictive control method for fixed-wing trajectory tracking tasks according to the present invention.

[0038] Figure 2 This is a schematic diagram of the coordinate system and parameters used to establish the aircraft dynamics model;

[0039] Figure 3 This is a diagram showing the relationships between various software components in the simulation environment.

[0040] Figure 4 The flight data acquisition process includes some of the cascaded "2-1-1" maneuver inputs and corresponding outputs for aircraft roll.

[0041] Figure 5 A schematic diagram illustrating the relationship between the aircraft and waypoints, as well as the error term;

[0042] Figure 6 A comparison chart showing the data reconstructed using the identified model and the actual flight data;

[0043] Figure 7 A comparison chart of data predicted using the model obtained from identification and actual flight data;

[0044] Figure 8 This is a diagram showing the results of a fixed-wing UAV box-tracking flight test.

[0045] Figure 9 This is a diagram showing the results of a fixed-wing unmanned aerial vehicle (UAV) space trajectory tracking flight test. Detailed Implementation

[0046] This invention provides an aircraft trajectory tracking method based on a data-driven model predictive control method, wherein the method is to capture and store the latest aircraft state response data in real time during the flight of the aircraft;

[0047] The current aircraft state response data is used as input to the data-driven model predictive controller. The data-driven model predictive controller continuously predicts the aircraft state response data in real time for a period of time after the current moment based on the aircraft's planar lateral dynamic equation, roll dimension response equation, lateral tracking error and heading tracking error equation. Based on the predicted aircraft state response data for a period of time after the current moment, an objective function is established for optimization to obtain the optimal expected roll angle reference signal for the next moment. The aircraft performs flight tasks according to the predicted expected roll angle reference signal, thereby realizing the aircraft's planar lateral trajectory tracking.

[0048] Simultaneously, based on the stored aircraft status response data, the model parameters used by the data-driven model predictive controller are identified online and dynamically adjusted.

[0049] The tracking method proposed in this invention constructs a data-driven model predictive controller based on the aircraft's planar lateral dynamics equations, roll dimension response equations, and lateral tracking error and heading tracking error equations. This enables the aircraft to perform spatial trajectory tracking, maintaining good convergence between the aircraft's flight trajectory and the target path, and exhibiting excellent control performance. This method is applicable to various aircraft; the following description, using a fixed-wing aircraft as an example, is detailed with reference to the accompanying drawings.

[0050] Figure 1 The right side is a control block diagram of a data-driven model predictive control method for trajectory tracking of fixed-wing aircraft provided by this invention. The implementation of the block diagram includes the following steps:

[0051] Step S1: Establish the basic model of the model predictive controller. First, establish the planar lateral dynamic equations of the aircraft. Figure 2 This diagram illustrates the coordinate system and parameters used to establish the aircraft's dynamics model. The planar lateral dynamic equations of a fixed-wing aircraft can be derived using a local inertial coordinate system composed of north and east axes. and the body coordinate system The parameters are defined together, and the planar transverse dynamic equations are expressed as follows:

[0052]

[0053]

[0054]

[0055]

[0056] Where n and e represent the aircraft's northward and eastward positions, respectively, both with the takeoff coordinates as the origin; the superscript · indicates the derivative of the corresponding physical quantity. and V represents the aircraft's northbound and eastbound speeds, respectively. g Represents the ground velocity vector; ψ g It represents the angle between the ground velocity vector and the north-facing plane, and is defined as positive when rotating clockwise with north as the reference. Its value ranges from -π to π. ω represents the angular velocity of the ground velocity vector and the northward plane, g is the local gravitational acceleration; φ represents the roll angle, defined as positive with right roll; V represents the airspeed vector. p represents the roll angular velocity. Additionally, W in the diagram represents the wind speed. Furthermore, the following roll-dimensional response equation is added for the roll path of the fixed-wing aircraft:

[0057]

[0058] in, It is the roll acceleration; φ r It is the reference signal for the desired roll angle of the input aircraft; a0, a1, and b0 are constant parameters of the model.

[0059] When an aircraft performs a trajectory tracking task, the target trajectory specified by the user is given in the form of a discrete waypoint sequence. In the lateral trajectory tracking task, the waypoint information can be degenerated into two-dimensional plane coordinates and partitioned into KD tree space. Then, multiple waypoint information with the nearest target point as the starting point is extracted from it, projected onto the aircraft's plane track coordinate system, and high-order polynomial fitting is performed under this axis to obtain the curve F(x). Figure 5 This diagram illustrates the relationship between the aircraft and waypoints, as well as the error term. Sub-figure (a) shows the extraction of multiple waypoints starting from the nearest target point after constructing a spatially partitioned KD tree for the target trajectory waypoints. Sub-figure (b) presents the curve F(x) obtained by high-order polynomial fitting of the selected waypoints in the UAV's planar trajectory coordinate system, and gives the lateral tracking error term l in the controller state variables. e and heading tracking error term ψ e The geometric representation of .

[0060] Finally, based on the tracking objective, specific expressions for the equations of lateral tracking error and heading tracking error are established:

[0061] l e =F(x)-y

[0062] Ψ e =arctan(F′(x))-Ψ g

[0063] In the formula, x and y are the spatial position parameters of the waypoint in the track coordinate system. F′(x) is the derivative of F(x).

[0064]

[0065] The aircraft's northward position coordinates n, eastward position coordinates e, and the collected yaw angle ψ are included. g Roll angle φ, roll rate p, lateral tracking error l e With heading tracking error ψ e As a state vector, the control quantity is determined as φ, which is the input to the flight controller. r value.

[0066] Step S3: In order to set the objective function in the computer and perform optimization, the previously given plane lateral dynamics equations, roll dimension response equations, lateral tracking error equations, and heading tracking error equations of the aircraft need to be discretized into difference forms:

[0067] n k+1 =n k +V g cosψ gk δt

[0068] e k+1 =e k +V g sinψ gk δt

[0069] ψ gk+1 =ψ gk +(gtanφ k / V)δt

[0070] φ k+1 =φ k +p k δt

[0071] p k+1 =p k +(b0φ rk -a1p k -a0φ k )δt

[0072] l ek+1 =l ek +V g sinψ g δt

[0073] ψ ek+1 =ψ ek +(gtanφ k / V)δt

[0074] Where k is the index of the sampling time, and δt represents the time difference of the sampling interval.

[0075] Based on the above differential form, the data-driven model predictive controller can continuously predict the aircraft state response data in a time period after the current moment based on the input aircraft state response data at the current moment. Based on the predicted aircraft state response data in a time period after the current moment, an objective function is established for optimization to obtain the optimal expected roll angle reference signal for the next moment. The aircraft performs flight tasks based on the predicted expected roll angle reference signal, thereby realizing the aircraft's planar lateral trajectory tracking.

[0076] Minimizing the objective function is a key part of the model predictive controller, as this objective determines which control actions and flight states are desirable and which flight actions are best avoided in order to better track the reference. In the lateral trajectory tracking problem, this invention presents an optimization objective function:

[0077]

[0078] in ω φr , The weights represent the coefficients of each term, used to adjust the importance of each term in the optimization problem. N represents the prediction interval length of the model predictive controller. The first line describes the cost of lateral tracking error and heading tracking error, minimizing which allows the aircraft to track the reference trajectory more accurately. The second line describes the roll angle penalty term, reducing which allows the aircraft to fly as smoothly as possible while tracking the reference. The third line describes the roll maneuver penalty term, minimizing which ensures a smoother transition in the controller's output control quantity, preventing the aircraft from making frequent large-amplitude maneuvers during the mission, thereby increasing flight stability and reducing energy consumption.

[0079] As a preferred option, the objective function weights can be taken as follows:

[0080]

[0081] Tests have shown that the various metrics of the adjustable data-driven model predictive controller (MPC controller) can be achieved under this weighting coefficient. Minimizing the objective function enables the aircraft to achieve better planar trajectory tracking with fewer and smoother maneuvers. The optimization solution can be solved in real time using the Ipopt open-source solver.

[0082] Step S4: During real-time aircraft trajectory tracking, the model parameters used by the data-driven model predictive controller are identified and dynamically adjusted using the Sparse Identification of Nonlinear Dynamics with Control (SINDYc) method based on the real-time captured and stored aircraft state response data. Based on the foregoing description, the model parameters used by the data-driven model predictive controller are mainly the three constant parameters a0, a1, and b0 in the roll dimension response equation. For the roll channel, the stored aircraft state response data includes a three-dimensional vector sequence of roll angle, roll rate, and input roll angle reference value. Regarding the candidate library configuration, step S1 has already provided a definite second-order model form—the roll dimension response equation—so the corresponding basis functions can be set according to this equation. Therefore, the specific parameter values ​​corresponding to a0, a1, and b0 can be obtained through the optimizer, achieving online identification and dynamic adjustment.

[0083] Furthermore, before initial use, the model parameters used by the data-driven model prediction controller can be predetermined by an optimizer through a controlled sparse identification method based on the collected flight data.

[0084] Flight data is collected by performing programmed flight missions. Fixed-wing aircraft need to continuously input control excitations and record all flight data in a windless environment. This data should include multiple 2-1-1 dual-level maneuvers with different amplitudes, and the input control amplitude should be controlled within an appropriate range. Figure 4 This section describes the output results of the dual-cascaded "2-1-1" maneuver input and aircraft roll response used in the flight data acquisition step. The "2-1-1" maneuver is an alternating pulse signal input with the same amplitude and a duty cycle of 2:1:1. This invention uses an improved dual-cascaded "2-1-1" maneuver as a primary control input combination, as shown by the dashed line in the figure; the solid line in the figure represents the response of the fixed-wing UAV under this excitation input. It should be noted that this process can be performed using actual flight data acquisition or simulation data acquisition. In this embodiment, the simulation platform is built using mainstream open-source toolkits such as PX4, Gazebo, and ROS, enabling realistic simulation of multiple objects including the aircraft, flight control system, onboard computer, and atmospheric environment. The programmable control part is completed by ROS node programs; the software block diagrams are shown below. Figure 3This platform enables realistic simulation of multiple objects, including the aircraft, flight control system, onboard computer, and atmospheric environment. Gazebo is a robot simulation toolset with a built-in high-performance real-time physics engine, providing realistic 3D physics calculations. Virtual fixed-wing UAVs can be created in Gazebo, and environmental parameters, including wind speed and turbulence, can be configured. The simulator can then simulate the dynamic behavior of the aircraft and output onboard sensor data with simulated noise. PX4 is an open-source unmanned system control software solution that can connect to the sensors and actuators of the virtual aircraft to achieve low-level control. The data-driven model predictive controller proposed in this invention, as a high-level control loop, runs on top of the Robot Operating System (ROS), a general-purpose robot development toolset that can standardize and coordinate the operation of the controller and its various auxiliary processes. It is integrated into the PX4 flight control software using the MAVROS software package.

[0085] As one implementation scheme, the sampling frequency during the flight data acquisition process can be set to 20Hz, wherein the control input includes a “2-1-1” dual-level maneuver with different amplitudes ranging from 0.1 to 0.8 radians and spaced at 0.1 radians. After each control unit completes its input, a sufficiently long standby interval is arranged to reach a stable state.

[0086] Based on the collected flight data, the Sparse Identification of Nonlinear Dynamics with Control (SINDYc) method is also used to identify the specific parameter values ​​corresponding to a0, a1, and b0 after the optimizer is used to determine the initial values ​​of the model parameters used by the data-driven model predictive controller.

[0087] After determining the initial values, the basic model is used to fit and restore the aircraft's flight state under the same initial state and control sequence conditions during the data acquisition process. The error between the model and the actual flight record is evaluated. If the accuracy meets certain requirements, the verification is passed. Otherwise, the flight data acquisition and identification process is checked for improper operation, and the relevant steps are re-executed.

[0088] Figure 6This figure compares the data obtained by reconstructing the training set using the model identified from the collected flight data with the actual flight data from the training set. The figure shows the model's fitting results to the dataset under the same control sequence and initial state input as the training set, with the solid and dashed lines representing the original and fitted curves, respectively. It's evident that with smaller roll amplitudes, the identification model can accurately fit and reconstruct the roll angle φ term in the dataset. However, with larger roll amplitudes, a decrease in fitting accuracy occurs, which is due to the low-order model's inability to accurately reproduce high-frequency signals. Fortunately, the fitting results are perfectly suitable for the design of the model predictive controller. Furthermore, UAV flight aims for stability, and in actual flight, such high-frequency, large-amplitude maneuvers are rare. In most cases, the control input falls within a relatively small amplitude range, which is more conducive to predicting flight states.

[0089] Figure 7 This figure compares the data predicted by the model from the training set of collected flight data with the actual flight data in the test set. To verify the generalizability of the fitted parameters, it is beneficial to include some input forms other than "2-1-1" maneuvers in the test set, such as some additional mission guidance processes. The solid line in the figure represents the flight data during a segment of random guidance mission, while the dashed line represents the forward prediction results obtained after inputting the same control sequence as this segment of flight into the model. Observation shows that in the 100-second experimental test, the model prediction results fit the real flight data very well, and there is no significant prediction deviation even after the experimental segment. In model predictive control, it is usually only necessary to predict the state of the system a few seconds after the current moment, and the performance of this model has far exceeded the requirements of the controller. At the same time, this experiment also demonstrates the universality of model parameter identification based on the "2-1-1" two-level maneuver dataset.

[0090] In addition, the stored aircraft state response data can be continuously stored during actual flight by maintaining a queue whose length can be adjusted according to actual conditions. This data serves as an incremental dataset in the model identification process, and the weight of the incremental dataset in correcting the model's basic parameters can be controlled by adjusting the length of the queue. Furthermore, the incremental dataset can be used for sparse identification together with the previously acquired flight data collected by the "2-1-1" dual-cascade maneuver when necessary. This allows for the correction of model parameters based on the latest data, enabling dynamic model correction and achieving data-driven dynamic model identification.

[0091] As one implementation scheme, the data queue length for storing the real-time status and control inputs of the aircraft can be set to 10 seconds at a sampling rate of 20Hz. Adjusting the length of this queue can also control the correction weights of the incremental dataset on the model's basic parameters.

[0092] Furthermore, hard constraints can be specified for the model predictive controller. These constraints typically stem from the factory physical limitations of the system components and safety-related limitations. Since the aircraft's response lags behind the control input, constraining the control input will constrain the corresponding state input. Here, no hard constraints of any kind need to be added to the aircraft's state input; only roll constraints need to be set for the controller output to ensure stable flight and autopilot safety.

[0093] As one implementation, the hard constraint on the output of the model predictive controller can set the roll angle between -40 degrees and 40 degrees.

[0094] Furthermore, a PID controller can be used to improve the longitudinal altitude control of the aircraft. Specifically, multiple waypoints obtained from the MPC controller, starting from the nearest point to the current aircraft position, are reused, and their altitudes are interpolated and fitted to calculate a smoothly changing target altitude reference value, which is then input to the controller during flight. Here, the output of the PID controller is set to the pitch angle reference value to approximate the target altitude. For safety, the target pitch angle output by the controller can be limited so that both the aircraft's climb and glide angles are less than 10 degrees. At the same time, a piecewise function curve is used to map an appropriate throttle output setpoint based on the pitch reference value, providing an energy management strategy for the system. The throttle output value and the pitch control quantity are sent together to the flight controller to achieve longitudinal altitude tracking.

[0095] A box trajectory tracking command is issued to the aircraft, with a side length of 100 meters, which is a relatively small value for a fixed-wing aircraft and quite challenging. The tracking method of this invention is used for tracking. During the tracking flight test, the model parameters used by the data-driven model predictive controller are identified and dynamically adjusted using the Sparse Identification of Nonlinear Dynamics with Control (SINDYc) method based on the stored aircraft state response data. The adjustment frequency is once every 10 seconds. Figure 8 The figures show the results of a box-tracking flight test of a fixed-wing UAV. Subfigure (a) shows the control effect achieved using the control method proposed in this invention under windless conditions; subfigure (b) shows the tracking effect of the aircraft after adding a southwest wind with an average wind speed of 4 m / s; and subfigure (c) shows the trajectory tracking results under a southwest gust environment with an average wind speed of 4 m / s and a maximum wind speed of 8 m / s. Overall, the results show that the controller used has excellent performance, can perform the planar trajectory tracking task well, and can achieve more robust adaptive control behavior.

[0096] Figure 9This is a diagram showing the results of a fixed-wing UAV space trajectory tracking flight test. The space trajectory includes both planar trajectory tracking commands and altitude tracking commands. The diagram shows the differences between the plane coordinates, altitude coordinates, roll angle, and actual reference values ​​during the aircraft's mission execution. The results indicate that the controller performed very well.

[0097] Corresponding to the aforementioned aircraft trajectory tracking method based on data-driven model predictive control, the present invention also provides an aircraft trajectory tracking device based on data-driven model predictive control.

[0098] This invention provides an aircraft trajectory tracking device based on data-driven model predictive control, comprising:

[0099] The data acquisition and storage unit is used to capture and store the latest aircraft status response data in real time during the flight of the aircraft; the aircraft status response data includes the aircraft's northward position coordinates n, eastward position coordinates e, and yaw angle ψ. g Roll angle φ, roll rate p, lateral tracking error l e With heading tracking error ψ e ;

[0100] The data-driven model predictive controller takes the aircraft's current state response data as input and continuously predicts the aircraft's state response data for a period of time after the current moment based on the aircraft's planar lateral dynamic equation, roll dimension response equation, lateral tracking error, and heading tracking error equation. It establishes an objective function based on the predicted aircraft state response data for a period of time after the current moment for optimization, and obtains the optimal expected roll angle reference signal for the next moment. The aircraft performs flight tasks according to the predicted expected roll angle reference signal, realizing the aircraft's planar lateral trajectory tracking.

[0101] The model parameter dynamic adjustment unit is used to identify and dynamically adjust the model parameters used by the data-driven model predictive controller based on the aircraft state response data stored in the data acquisition and storage unit.

[0102] Furthermore, it also includes a PID controller for longitudinal altitude tracking of the aircraft.

[0103] The device implementation can be achieved through software, hardware, or a combination of both.

[0104] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0105] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for tracking aircraft trajectory based on data-driven model predictive control, characterized in that, The method involves capturing the latest aircraft status response data in real time during the aircraft's flight. The aircraft status response data includes the aircraft's northward position coordinates. Eastward position coordinates Yaw angle Roll angle Rollover rate Lateral tracking error With heading tracking error ; The current aircraft state response data is used as input to the data-driven model predictive controller. The data-driven model predictive controller continuously predicts the aircraft state response data in real time for a period of time after the current moment based on the aircraft's planar lateral dynamic equation, roll dimension response equation, lateral tracking error and heading tracking error equation. Based on the predicted aircraft state response data for a period of time after the current moment, an objective function is established for optimization to obtain the optimal expected roll angle reference signal for the next moment. The aircraft performs flight tasks according to the predicted expected roll angle reference signal, thereby realizing the aircraft's planar lateral trajectory tracking. Simultaneously, based on the real-time captured and stored aircraft status response data, the model parameters used by the data-driven model prediction controller are identified online and dynamically adjusted. The aircraft is a fixed-wing aircraft, and the planar lateral dynamic equation of the fixed-wing aircraft is expressed as follows: The roll-dimensional response equation is expressed as: The lateral tracking error equation is expressed as: The heading tracking error equation is expressed as follows: In the formula and These represent the aircraft's north and east positions, respectively, with the takeoff coordinates as the origin. Represents the ground velocity vector; This represents the angle between the ground velocity vector and the northward-facing plane; This refers to the local gravitational acceleration. Indicates the roll angle; Represents the airspeed vector. Indicates the roll angular velocity; It is the input reference signal for the desired roll angle of the aircraft; , , represents the constant parameters of the model; k is the index of the sampling time. This represents the time difference between sampling intervals.

2. The method according to claim 1, characterized in that, The model parameters used by the data-driven model prediction controller are predetermined by an optimizer obtained from the collected flight data using a controlled sparse identification method.

3. The method according to claim 2, characterized in that, Flight data is collected by continuously inputting control excitations to the aircraft in a windless environment. The input control data includes multiple "2-1-1" cascaded maneuvers with different amplitudes. After each input control reaches a steady state, flight data is recorded, including a three-dimensional vector sequence of roll angle, roll rate, and the desired roll angle reference signal of the input aircraft.

4. The method according to claim 1, characterized in that, The data-driven model predicts that the desired roll angle reference signal output by the controller is subject to hard constraints, ensuring that the roll angle output by the controller is within a safe range.

5. The method according to claim 1, characterized in that, The objective function includes the costs of lateral tracking error and heading tracking error, a roll angle penalty, and a roll maneuver penalty.

6. The method according to claim 1, characterized in that, It also includes using a PID controller for longitudinal altitude tracking of the aircraft.

7. An aircraft trajectory tracking device based on data-driven model predictive control, characterized in that, include: The data acquisition and storage unit is used to capture and store the latest aircraft status response data in real time during the flight of the aircraft; The aircraft status response data includes the aircraft's northward position coordinates. Eastward position coordinates Yaw angle Roll angle Rollover rate Lateral tracking error With heading tracking error ; The data-driven model predictive controller takes the aircraft's current state response data as input and continuously predicts the aircraft's state response data for a period of time after the current moment based on the aircraft's planar lateral dynamic equation, roll dimension response equation, lateral tracking error, and heading tracking error equation. It establishes an objective function based on the predicted aircraft state response data for a period of time after the current moment for optimization, and obtains the optimal expected roll angle reference signal for the next moment. The aircraft performs flight tasks according to the predicted expected roll angle reference signal, realizing the aircraft's planar lateral trajectory tracking. The model parameter dynamic adjustment unit is used to identify and dynamically adjust the model parameters used by the data-driven model prediction controller based on the aircraft state response data stored in the data acquisition and storage unit. The aircraft is a fixed-wing aircraft, and the planar lateral dynamic equation of the fixed-wing aircraft is expressed as follows: The roll-dimensional response equation is expressed as: The lateral tracking error equation is expressed as: The heading tracking error equation is expressed as follows: In the formula and These represent the aircraft's north and east positions, respectively, with the takeoff coordinates as the origin. Represents the ground velocity vector; This represents the angle between the ground velocity vector and the northward-facing plane; This refers to the local gravitational acceleration. Indicates the roll angle; Represents the airspeed vector. Indicates the roll angular velocity; It is the input reference signal for the desired roll angle of the aircraft; , , represents the constant parameters of the model; k is the index of the sampling time. This represents the time difference between sampling intervals.

8. The apparatus according to claim 7, characterized in that, It also includes a PID controller for tracking the aircraft's longitudinal altitude.

Images