Flight environment parameter inversion method and system based on energy and load closed-loop feedback
By using a closed-loop feedback method for energy and load, the error and distortion problems in the extrapolation of environmental parameters under unsteady flight conditions in existing technologies are solved, and high-precision environmental parameter inversion is achieved, ensuring the physical interpretability and robustness of the system.
Patent Information
- Application Number
- CN202611618057.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to accurately reproduce real flight environment parameters under complex flight conditions involving unsteady states and sensor distortion. Traditional methods are prone to calculation errors and model mismatches, leading to distorted parameter extrapolation results.
A closed-loop feedback method based on energy and load is adopted. Vertical wind speed estimates are generated by energy decoupling. A nonlinear correction factor is introduced, and iterative calculation is performed by combining a physical constraint network model and a dynamic model until the convergence condition is met and the environmental parameters are output.
It enables high-precision reverse engineering of environmental parameters under complex boundary conditions, ensuring the physical traceability and system robustness of the inference process, and improving the system's convergence ability and stability in harsh environments.
Smart Images

Figure CN123116883A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of flight simulation technology, and relates to data processing technology for flight simulators or complex simulation systems, and particularly to a method and system for inverting flight environment parameters based on energy and load closed-loop feedback. Background Technology
[0002] In civil aviation flight operations, the Quick Access Recorder (QAR) records sensor observation data during aircraft flight, serving as a crucial data source for flight operation quality monitoring and accident investigation. However, the QAR does not directly record three-dimensional wind fields, especially vertical downbursts caused by severe weather such as thunderstorms. Furthermore, under extreme unsteady conditions such as thunderstorms, rapid dives, or stalls, external sensors experience nonlinear distortion due to drastic attitude changes such as large angles of attack, leading to the loss or inaccurate measurement of core aerodynamic parameters such as the true angle of attack and sideslip angle.
[0003] To reconstruct the aforementioned missing environmental parameters, existing technologies mainly employ methods such as physical model analysis, energy conservation calculations, or parameter identification based on deep learning. In implementing the embodiments of this application, the inventors discovered that the existing technologies have at least the following problems: Traditional wind field estimation methods based on energy conservation rules heavily rely on actual airspeed parameters. In complex flow fields such as strong wind shear, the high coupling between parameters can easily lead to cyclic errors in the solution process, making it difficult to achieve parameter decoupling under boundary conditions. Secondly, methods based on six-degree-of-freedom dynamic analysis or state estimation (such as Kalman filtering) heavily rely on aerodynamic prior models. When the aircraft enters a high angle of attack or stall region, flow field separation can cause aerodynamic derivatives to exhibit strong nonlinearity. Conventional static or linearized assumption models may experience model mismatch, leading to physical distortions in the extrapolation results.
[0004] Furthermore, joint physical constraint data identification methods developed in recent years are typically limited to residual optimization within a single physical domain (such as the pure mechanical equation domain). When the baseline model itself has systematic biases, single-dimensional solutions often fail to effectively detect and converge their own model errors. Moreover, existing closed-loop correction logic generally tends to directly modify underlying aerodynamic parameters (such as directly overwriting the basic lift-drag coefficients) to forcibly achieve local mathematical fitting. This fitting method, which confuses external physical environment disturbances with internal aircraft aerodynamic characteristics, may obscure the actual fundamental physical meaning of the state equations, resulting in insufficient decoupling and separation of the system in terms of separating the parameter terms causing the anomalies.
[0005] In summary, there is an urgent need for an environmental parameter extrapolation technology that can objectively and accurately reproduce real flight environment parameters under complex flight conditions that are unsteady and where sensors are distorted. Summary of the Invention
[0006] To address the aforementioned problems in the prior art, namely the physical distortion that easily occurs when existing environmental parameter estimation methods derive environmental parameters under unsteady conditions, the first aspect of this application proposes a flight environmental parameter inversion method based on energy and load closed-loop feedback, comprising the following steps: Acquire aircraft observation data and raw aerodynamic database, construct power balance equations based on mechanical energy conservation, and generate raw vertical wind speed estimates through energy decoupling; A nonlinear correction factor is introduced and used to correct the original vertical wind speed estimate to obtain the energy domain vertical wind speed estimate. The estimated vertical wind speed in the energy domain and the state variables in the aircraft observation data are input into a preset physical constraint network model to deduce the corrected aerodynamic angle parameters. Substitute the corrected aerodynamic angle parameters into the dynamic model, call the basic aerodynamic coefficients in the original aerodynamic database to synthesize the theoretical acceleration, and compare it with the measured acceleration in the aircraft observation data to generate normal overload residuals and comprehensive residuals. If the comprehensive residual does not meet the preset convergence condition, the nonlinear correction factor is updated by combining the flow direction scalar characteristics of the normal overload residual and the original vertical wind speed estimate, and fed back to the decoupled mapping end to trigger closed-loop iterative solution. Until the comprehensive residual satisfies the convergence condition or reaches the maximum number of iterations, the current energy domain vertical wind speed estimate and the corrected aerodynamic angle parameter are output as environmental simulation parameters.
[0007] In some preferred embodiments, after acquiring the aircraft observation data, data preprocessing is performed, specifically as follows: The aircraft observation data in the body coordinate system is converted to the aerodynamic wind axis coordinate system, and magnetic heading adaptive correction is performed based on the preset geomagnetic model and positioning quality index. At the same time, the vertical velocity is calculated by fusing barometric altitude characteristics and inertial navigation data based on complementary filtering algorithm, and the preprocessed state variables are output.
[0008] In some preferred embodiments, a power balance equation is constructed based on the conservation of mechanical energy, and the original vertical wind speed estimate is generated through energy decoupling, including: The preprocessed state variables are analyzed to obtain the aircraft's altitude parameters, mass parameters, current thrust, and current flight drag. The thrust power is determined based on the thrust and airspeed, the drag power is determined based on the flight drag and airspeed, the gravitational potential energy rate of change is determined based on the mass parameter and the vertical velocity, and the kinetic energy rate of change is determined based on the mass parameter and airspeed. Based on the thrust power, drag dissipation power, gravitational potential energy rate of change and kinetic energy rate of change, calculate the overall power disturbance residual of the aircraft. Based on the original aerodynamic database, the current reference lift of the aircraft is determined, and based on the mapping ratio of the overall power disturbance residual to the current reference lift, the original vertical wind speed estimate is generated by decoupling.
[0009] In some preferred embodiments, the original vertical wind speed estimate is corrected to obtain an energy domain vertical wind speed estimate, specifically including: Based on the aircraft's pitch and track tilt angles, the original vertical wind speed estimate, and airspeed, the effective angle of attack is calculated. When the absolute value of the effective angle of attack is less than the preset angle threshold, and the absolute value of the ratio of the original vertical wind speed estimate to the air speed is less than the preset ratio limit, it is determined that the small angle interval condition is met, and the original vertical wind speed estimate is used as the energy domain vertical wind speed estimate. If the small angle interval condition is not met, the nonlinear decoupling mode is activated, and the nonlinear correction factor is used as a gain coefficient to compensate the original vertical wind speed estimate, thereby calculating the energy domain vertical wind speed estimate.
[0010] In some preferred embodiments, the corrected aerodynamic angle parameters include corrected angle of attack and corrected sideslip angle; the physical constraint network model is constructed based on a physical information neural network, and obtaining the corrected aerodynamic angle parameters includes: The current manipulation sequence characteristics, the estimated vertical wind speed in the energy domain, and the corrected aerodynamic angle parameters output in the previous round are input into the physical information neural network to deduce the aerodynamic angle correction amount. The original apparent aerodynamic angles in the aircraft observation data are superimposed with the aerodynamic angle correction amount to obtain the corrected angle of attack and the corrected sideslip angle, which are used as the corrected aerodynamic angle parameters. The loss function inside the physical information neural network is composed of the fitting deviation term of the data observation channel and the constraint residual term that characterizes the nonlinear physical logic at large angles of attack. The constraint residual term covers the lateral force balance physical constraint, the normal force balance physical constraint, and the vertical velocity consistency constraint, which include the projection of the complete spatial nonlinear trigonometric function.
[0011] In some preferred embodiments, the nonlinear correction factor is updated by combining the flow direction scalar characteristics of the normal overload residual and the original vertical wind speed estimate, including: Extract the nonlinear correction factor from the previous iteration; Determine the direction and magnitude of the floating deviation of the measured normal component in the measured acceleration relative to the theoretical normal component in the theoretical acceleration; The flow direction scalar feature is used as a direction compensator to eliminate the negative feedback sign change interference caused by the opposite direction of the upward or downward airflow. The adjustment step size is generated by combining the deviation amplitude, the preset feedback gain ratio and the compensated adjustment direction. The adjustment step size is superimposed on the nonlinear correction factor of the previous iteration to obtain the updated nonlinear correction factor. The updated nonlinear correction factor is truncated using a preset boundary limit function.
[0012] A second aspect of this application proposes a flight environment parameter inversion system based on energy and load closed-loop feedback, comprising: The data acquisition module is used to acquire aircraft observation data and raw aerodynamic database; The kinematics preprocessing module is used to perform coordinate system datum unification and heading adaptive calibration on the acquired aircraft observation data, and output the preprocessed state quantities under the unified datum. The energy domain solution module is used to construct a power balance equation based on the conservation of mechanical energy, generate the original vertical wind speed estimate through energy decoupling, introduce a nonlinear correction factor, and use it to correct the original vertical wind speed estimate to obtain the energy domain vertical wind speed estimate. The neural inference module is used to input the estimated vertical wind speed in the energy domain and the state variables in the aircraft observation data into a preset physical constraint network model to deduce the corrected aerodynamic angle parameters. The cross-domain verification module is used to substitute the modified aerodynamic angle parameters into the dynamic model, call the basic aerodynamic coefficients in the original aerodynamic database to synthesize the theoretical acceleration, and compare it with the measured acceleration in the aircraft observation data to generate normal overload residuals and comprehensive residuals. The closed-loop feedback module is used to update the nonlinear correction factor by combining the flow direction scalar characteristics of the normal overload residual and the original vertical wind speed estimate when the comprehensive residual does not meet the preset convergence condition, and feed it back to the decoupled mapping end to trigger closed-loop iterative solution; until the comprehensive residual meets the convergence condition or reaches the preset maximum number of iterations, the current energy domain vertical wind speed estimate and the corrected aerodynamic angle parameter are output as environmental extrapolation parameters.
[0013] Compared with the prior art, the technical solution provided in this application has at least the following beneficial effects: 1) By introducing closed-loop feedback between the energy domain and the force domain, when the force domain verification residual does not meet the convergence condition, the nonlinear correction factor in the energy domain solution is updated by normal overload residual feedback, so as to build a cross-physical domain mutual inspection mechanism. This enables high-precision reverse inference of missing environmental parameters and cross-domain error self-correction under complex boundary conditions, ensuring that the entire inference process has aerospace physical traceability and system boundary robustness.
[0014] 2) Apply the nonlinear correction factor to the vertical wind speed mapping end to ensure that the aerodynamic characteristics called in during the force domain feedback verification process always maintain the original aerodynamic database mapping in an unmodified state within the normal flight envelope. This avoids the behavior of traditional methods that destroy the ontological state equations during fitting, thereby ensuring the independence and physical interpretability of the force domain verification data.
[0015] 3) When updating the correction factor, the flow direction scalar feature is introduced as compensation, so that no matter whether the system faces updraft or downburst, the adjustment feedback between the normal overload error and the correction factor follows a stable negative feedback gradient descent direction, which improves the system's convergence ability and stability under harsh boundary conditions.
[0016] 4) The vertical overload change rate is used as an early warning feature of global environmental deterioration. When the aircraft enters an unsteady state (such as stall or wind shear), the multi-module collaborative blocking is implemented to enhance the tolerance of the filter coefficient, double the physical constraint penalty, and relax the convergence condition, thereby improving the algorithm's obstacle-crossing self-healing ability. Attached Figure Description
[0017] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart of a flight environment parameter inversion method based on energy and load closed-loop feedback provided in one embodiment of this application; Figure 2 This is a logical flowchart of a flight environment parameter inversion method based on energy and load closed-loop feedback provided in one embodiment of this application; Figure 3 This is a schematic diagram of the structure of a computer system used to implement the methods, apparatus, and electronic devices of this application. Detailed Implementation
[0018] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.
[0019] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0020] To address the problem that existing single-physical-domain parameter derivation methods for extreme unsteady conditions such as thunderstorms and stalls cannot accurately and decoupledly reconstruct real flight environment parameters due to the loss of physical traceability of the body aerodynamic model, this application provides a flight environment parameter inversion method based on energy and load closed-loop feedback. This method includes: Acquire aircraft observation data and raw aerodynamic database, construct power balance equations based on mechanical energy conservation, and generate raw vertical wind speed estimates through energy decoupling; A nonlinear correction factor is introduced and used to correct the original vertical wind speed estimate to obtain the energy domain vertical wind speed estimate. The estimated vertical wind speed in the energy domain and the state variables in the aircraft observation data are input into a preset physical constraint network model to deduce the corrected aerodynamic angle parameters. Substitute the corrected aerodynamic angle parameters into the dynamic model, call the basic aerodynamic coefficients in the original aerodynamic database to synthesize the theoretical acceleration, and compare it with the measured acceleration in the aircraft observation data to generate normal overload residuals and comprehensive residuals. If the composite residual does not meet the preset convergence condition, the nonlinear correction factor is updated by combining the flow direction scalar characteristics of the normal overload residual and the original vertical wind speed estimate, and fed back to the decoupled mapping end to trigger closed-loop iterative solution; until the composite residual meets the convergence condition or reaches the maximum number of iterations, the current energy domain vertical wind speed estimate and the corrected aerodynamic angle parameter are output as environmental extrapolation parameters.
[0021] To more clearly explain the flight environment parameter inversion method based on energy and load closed-loop feedback in this application, the following will combine... Figure 1 The steps in the embodiments of this application are described in detail.
[0022] The first embodiment of this application provides a method for inverting flight environment parameters based on energy and load closed-loop feedback, including steps S1 to S5, as follows: Figure 1 As shown, each step is described in detail below: S1. Acquire aircraft observation data and raw aerodynamic database, construct power balance equations based on mechanical energy conservation, and generate raw vertical wind speed estimates through energy decoupling.
[0023] Preferably, after acquiring the aircraft observation data, data preprocessing is performed, specifically as follows: The aircraft observation data in the body coordinate system is converted to the aerodynamic wind axis coordinate system, and magnetic heading adaptive correction is performed based on the preset geomagnetic model and positioning quality index. At the same time, the vertical velocity is calculated by fusing barometric altitude characteristics and inertial navigation data based on complementary filtering algorithm, and the preprocessed state variables are output.
[0024] Preferably, a power balance equation is constructed based on the conservation of mechanical energy, and the original vertical wind speed estimate is generated through energy decoupling, including: The preprocessed state variables are analyzed to obtain the aircraft's altitude parameters, mass parameters, current thrust, and current flight drag. The thrust power is determined based on the thrust and the airspeed; the drag power is determined based on the flight drag and the airspeed; the gravitational potential energy rate of change is determined based on the mass parameter and the vertical velocity; and the kinetic energy rate of change is determined based on the mass parameter and the airspeed. Based on the thrust power, drag dissipation power, gravitational potential energy rate of change and kinetic energy rate of change, calculate the overall power disturbance residual of the aircraft. Based on the original aerodynamic database, the current reference lift of the aircraft is determined, and based on the mapping ratio of the overall power disturbance residual to the current reference lift, the original vertical wind speed estimate is generated by decoupling.
[0025] Preferably, aircraft observation data and raw aerodynamic database are acquired. The observation data is obtained from a Fast Access Recorder (QAR) at an 8Hz sampling frequency and includes at least: Airspeed (Vtas) and Mach number (from atmospheric data systems) ), atmospheric pressure altitude (ALT), total temperature ( ); Actual triaxial overload data derived from accelerometers , , ); derived from angle of attack sensor readings ( ); Attitude angles and angular rates (roll rate) derived from the inertial navigation system (IRS) Pitch rate yaw rate ); Control angle (elevator) originating from the flight control system ,aileron rudder ), input lever position (STICK), and engine low-pressure rotor speed ( ).
[0026] For specific acquisition methods, please refer to Table 1.
[0027] Table 1
[0028] More preferably, the raw QAR data to be processed is first subjected to time alignment and outlier removal using the Savitzky-Golay smoothing filter algorithm. Then, coordinate system standardization is performed, using a standard Euler angle rotation matrix to uniformly transform the body coordinate system data to the aerodynamic wind axis coordinate system. Finally, adaptive heading calibration is performed, utilizing the horizontal accuracy factor (HDOP) quality index from the WMM world geomagnetic model and GPS data to adaptively correct the magnetic heading, which is beneficial to the reference accuracy of subsequent wind field calculations.
[0029] More preferably, a power balance equation is constructed based on the law of conservation of mechanical energy, and then the overall power disturbance residual is calculated: ; in, The engine thrust is obtained by looking up the engine performance three-dimensional lookup table using the engine low-pressure rotor speed N1, total temperature TAT, and air pressure altitude ALT recorded by QAR. The drag coefficient is obtained by looking up a table in the aerodynamic database of the Aircraft Flight Manual (AFM). For the real-time quality of the aircraft; It is the acceleration due to gravity; The true vertical velocity is obtained by fusing barometric altitude and inertial navigation system (IRS) data through a first-order complementary filter. This represents the rate of change of airspeed.
[0030] More preferably, the current reference lift of the aircraft is determined based on the original aerodynamic database, and the original vertical wind speed estimate is generated by decoupling based on the mapping ratio of the overall power disturbance residual to the current reference lift, specifically as follows: ; in, This is the current baseline lift for the aircraft. This is the original vertical wind speed estimate.
[0031] It should be noted that the law of conservation of mechanical energy is common knowledge in this field and will not be elaborated upon here.
[0032] S2. Introduce a nonlinear correction factor and use it to correct the original vertical wind speed estimate to obtain the energy domain vertical wind speed estimate.
[0033] Preferably, it is used to correct the original vertical wind speed estimate to obtain an energy domain vertical wind speed estimate, specifically including: Based on the aircraft's pitch and track tilt angles, the original vertical wind speed estimate, and airspeed, the effective angle of attack is calculated. When the absolute value of the effective angle of attack is less than the preset angle threshold, and the absolute value of the ratio of the original vertical wind speed estimate to the air speed is less than the preset ratio limit, it is determined that the small angle interval condition is met, and the original vertical wind speed estimate is used as the energy domain vertical wind speed estimate. If the small angle interval condition is not met, the nonlinear decoupling mode is activated, and the nonlinear correction factor is used as a gain coefficient to compensate the original vertical wind speed estimate, thereby calculating the energy domain vertical wind speed estimate. The nonlinear correction factor keeps the basic aerodynamic coefficients in the original aerodynamic database in an unmodified state.
[0034] In one specific embodiment, a power balance equation is constructed based on the law of conservation of total mechanical energy of the aircraft, the power residual is calculated, and the energy domain vertical wind speed estimate is obtained by decoupling. Specifically, it includes the following steps: 1) Calculate the effective aerodynamic angle; Vertical wind speed The effective angle of attack is defined as follows: This is achieved by altering the relative airflow direction. ; in, For an effective angle of attack, The pitch angle is recorded by the QAR inertial navigation system, γ is the inclination angle of the track relative to the ground, and the true vertical velocity is corrected by complementary filtering. With airspeed determined, , This is airspeed. This constitutes an aerodynamic angle of attack calculation benchmark excluding vertical wind. It should be noted that the QAR angle of attack anemometer reading itself already includes the apparent aerodynamic angle due to relative airflow deflection. When the anemometer is functioning normally, its reading is already close to the effective angle of attack. If this reading is used as a benchmark and a wind-induced deflection correction is added, the vertical wind effect will be repeatedly subtracted. Furthermore, under target operating conditions such as thunderstorms and stall, the anemometer reading itself is distorted and is therefore unsuitable as a decoupling benchmark. Therefore, this application uses an inertial navigation system pitch angle and track inclination angle to construct a windless benchmark. The effective angle of attack is calculated, and the QAR angle of attack reading is used only as an apparent observation for cross-validation of the data channel. Based on this, the standard physical relationship between vertical wind speed and effective angle of attack is defined as follows: A positive value indicates a downwash (downburst), which reduces the aircraft's angle of attack and consequently affects the lift coefficient. Decrease; A negative value indicates an upwash (rising airflow), which increases the aircraft's angle of attack and consequently affects the lift coefficient. Increase.
[0035] 2) Perform energy wind decoupling; Under the small-angle approximation condition (the absolute value of the effective angle of attack is less than a preset angle threshold, and the absolute value of the ratio of the original vertical wind speed estimate to the air speed is less than a preset proportional limit), and Effective angle of attack ; Due to the pseudo-linearity under this condition, a linear decoupling operator is used to generate the original vertical wind estimate, with a nonlinear correction factor of 1. The energy domain vertical wind speed estimate is: ; When the small-angle condition is not met, it automatically switches to the nonlinear decoupling mode, introducing a data-driven state nonlinearity correction factor. Calculate the estimated vertical wind speed in the energy domain: .
[0036] Preferably, when the closed-loop iterative solution is activated for the first time, the nonlinear correction factor is introduced as follows: A multilayer perceptron model that has converged and is pre-trained on a stall dataset is used to predict the initial compensation value as the nonlinear correction factor based on the current macroscopic flight state characteristics of the aircraft. If the multilayer perceptron model does not output a valid prediction result, then the neutral benchmark value representing no correction effect will be used as the nonlinear correction factor. The multilayer perceptron model was obtained by training on a historical test set.
[0037] More preferably, The value is determined in two stages, specifically: When the closed-loop iteration is first initiated, the initial value of the nonlinear correction factor is predicted by a pre-trained multilayer perceptron (MLP) regression model. This MLP model is trained on a historical stall dataset (containing stall training flights and actual stall events from over 500 flight quality monitoring records, covering mainstream civil aircraft models such as wide-body and narrow-body passenger aircraft). In cases where the MLP does not provide a valid prediction, a neutral value of 1.0 is used. In subsequent iterations, the nonlinear correction factor is updated based on the feedback.
[0038] Specifically, the MLP model uses feature vectors The input is the initial predicted value obtained during training, which is then constrained to the range [0.5, 2.0] by the Sigmoid function.
[0039] It is worth emphasizing that the nonlinear correction factor acts directly on the vertical wind speed mapping end, and does not modify the core parameters such as the aerodynamic coefficients CL and CD provided by AFM to the baseline before entering the polarization boundary. This decoupled mapping action site ensures the purity and traceability of the environmental feedback basis during force domain verification.
[0040] S3. Input the estimated vertical wind speed in the energy domain and the state variables in the aircraft observation data into the preset physical constraint network model to deduce the corrected aerodynamic angle parameters.
[0041] Preferably, the corrected aerodynamic angle parameters include corrected angle of attack and corrected sideslip angle; the physical constraint network model is constructed based on a physical information neural network, and obtaining the corrected aerodynamic angle parameters includes: The current manipulation sequence characteristics, the estimated vertical wind speed in the energy domain, and the corrected aerodynamic angle parameters output in the previous round are input into the physical information neural network to deduce the aerodynamic angle correction amount. The original apparent aerodynamic angles in the aircraft observation data are superimposed with the aerodynamic angle correction amount to obtain the corrected angle of attack and the corrected sideslip angle, which are used as the corrected aerodynamic angle parameters. The loss function inside the physical information neural network is composed of the fitting deviation term of the data observation channel and the constraint residual term that characterizes the nonlinear physical logic at large angles of attack. The constraint residual term covers the lateral force balance physical constraint, the normal force balance physical constraint, and the vertical velocity consistency constraint, which include the projection of the complete spatial nonlinear trigonometric function.
[0042] More preferably, a Physical Information Neural Network (PINN) architecture is employed, with the output of step S2... As historical data input, the corrected aerodynamic angle parameters are derived: The above As a historical dimension feature, the historical state quantity (the corrected angle of attack at the previous moment) is used to define the historical state quantity. Correcting the sideslip angle and current ) and PINN input control sequence (elevator deflection) The system inputs the side stick (STICK) and engine speed (N1) into a physical information neural network (PINN), which learns the nonlinear mapping relationship from historical states to effective angle of attack and sideslip angle, and outputs the derived correction amount. and The final result is: ; ; in, To correct the angle of attack, To correct the sideslip angle.
[0043] By influencing the input feature space of the PINN network, it indirectly participates in... and The deduction and calculation. and All are based on raw QAR sensor readings ( , ) superimposed with the learning from PINN, which depends on The correction amount is obtained.
[0044] It is not directly used as an input parameter for force domain verification; its influence on force domain verification is entirely through the PINN output. and Indirect transmission. The PINN network learns a complete nonlinear mapping from historical states to corrected aerodynamic angles. This mapping is not only constrained by relations, but also incorporates multidimensional information such as manipulation inputs, historical states, and physical equation residuals.
[0045] More preferably, the training of the Physical Information Neural Network (PINN) architecture adopts a two-stage strategy: the first stage is pure data-driven pre-training, and the second stage incorporates physical constraint fine-tuning. Relationship between energy domain estimate and PINN output: a) The energy domain estimate calculated based on the power balance equation is used as the PINN. Historical state input. The closed-loop feedback modifies step two. The value of the factor, and thus the modification Instead of any parameters in step three.
[0046] b) : PINN network's own deduction This value is used only for internal PINN state consistency verification and does not replace [the PINN value]. As the final output.
[0047] c) The final vertical wind speed adopted is the one obtained after closed-loop convergence. .
[0048] In one specific embodiment, the Physical Information Neural Network (PINN) architecture has the following network structure: Input layer: Control sequence (elevator deflection) Side stick input (STICK, engine speed N1) and historical state quantities (corrected angle of attack at the previous moment). Correcting the sideslip angle and current There are a total of 6 input nodes; The output layer is: corrected angle of attack. Correcting the sideslip angle Vertical wind speed for internal verification There are a total of 3 output nodes; Hidden layer: using As an activation function.
[0049] More preferably, the loss function of the Physical Information Neural Network (PINN) is: ; Among them, the data fitting term The average value is taken from the samples within the training batch; , These are the raw readings of the angle of attack sensor and the sideslip angle sensor recorded by QAR, respectively. , The derived corrected angle of attack and corrected sideslip angle for the output layer of the PINN network. Physical constraint terms. Let L2 be the norm formed by the residuals of the three equations of motion.
[0050] Specifically, the residuals of each equation of motion are defined as follows: Normal force balance residual: ; Lateral force balance residuals: ; Vertical velocity consistency residual: ; in, , These are the measured normal overload and measured lateral overload directly recorded by the QAR accelerometer, respectively. , These are lift and drag, respectively. ; This is the lateral force coefficient; For dynamic pressure; This refers to the wing reference area. For engine thrust, Its lateral component; For aircraft mass; It is the acceleration due to gravity; Pitch angle recorded by the QAR inertial navigation system; The true vertical velocity after complementary filtering correction; Vertical wind speed auxiliary output derived for the output layer of the PINN network.
[0051] All three equations above are preserved in their complete form. The nonlinear terms were not linearized at small angles to accommodate high angle of attack / stall conditions. The weighting coefficient for the physical constraint term relative to the data fitting term is 1.0 under stable flight conditions, and is used when the vertical overload rate of change... The value is increased to 2.0 when the non-steady-state trigger threshold is exceeded, in order to strengthen the constraint of physical priors on the network output.
[0052] The PINN constraint term only serves as loss regularization to shape the network mapping during the training phase. Step four, force domain verification, involves independently verifying the synthesized triaxial acceleration using the original aerodynamic coefficients from the AFM database and the measured values after inference. These two steps have different stages of action and purposes, and both drive the network within the closed loop. The adjustment only applies to the force domain verification residual. This does not constitute a self-consistent cycle.
[0053] S4. Substitute the corrected aerodynamic angle parameters into the dynamic model, call the basic aerodynamic coefficients in the original aerodynamic database to synthesize the theoretical acceleration, and compare it with the measured acceleration in the aircraft observation data to generate the normal overload residual and the comprehensive residual.
[0054] Preferably, during the synthesis of theoretical acceleration: Real-time monitoring of the aircraft’s current corrected angle of attack; when the current corrected angle of attack is assessed to have crossed the preset stall critical point and entered the deep stall region, the aerodynamic coefficient algorithm based on database lookup table is disabled. The semi-empirical aerodynamic model with built-in nonlinear decay factor and maximum stall lift characteristics is invoked to take over the calculation of lift coefficient, and the lift coefficient in this state is used to correct the evaluation of flight drag in the power balance equation in order to calculate drag dissipation power.
[0055] More preferably, in the deep stall region where the small angle approximation condition is not met ( To more accurately calculate drag D and lift L, a semi-empirical model is used to characterize the nonlinear decay characteristics of the lift coefficient: ; in, The reference lift coefficient at the nozzle point in the deep stall model, and the nozzle angle of attack. Pick , Anchoring at this point ensures continuity during model switching. Stall angle of attack; The lift decay rate after stall is taken as [value missing] in a specific embodiment. .
[0056] More preferably, the corrected aerodynamic angle parameters are substituted into the dynamic model, the basic aerodynamic coefficients in the original aerodynamic database are called to synthesize the theoretical acceleration, and compared with the measured acceleration in the aircraft observation data to generate the normal overload residual and the comprehensive residual, specifically including: A. Theoretical Load Synthesis: The modified angle of attack obtained from the derivation is then combined. Correcting the sideslip angle Substituting the extended nonlinear stability derivative model, the theoretical triaxial acceleration is resynthesized ( , , ).
[0057] A1. Linear Synthesis Theoretical Loading: Calling the static structural constants of the aircraft (wing reference area) ,span Mean aerodynamic chord length ) and dynamic pressure Based on the AFM aerodynamic database that has not been modified by the nonlinear correction factor, all aerodynamic moments and fundamental mechanical derivatives are retrieved using linear interpolation, and then the overall aerodynamic coefficients (including but not limited to) are synthesized according to the standard Taylor expansion form: Lift coefficient: ; Drag coefficient: ; Lateral force coefficient: .
[0058] Each aerodynamic derivative was obtained from a multidimensional lookup table in the AFM aerodynamic database, with the lookup dimension being [ , [Configuration (flaps / landing gear status)], the interpolation strategy uses linear interpolation.
[0059] because No modification and All aerodynamic coefficients used here are original values from the AFM database. The mechanical composition boundary is compatible with the full envelope aerodynamic derivatives (including those derived from roll / yaw) due to asymmetry. 、 、 、 、 、 、 、 、 、 、 、 、 wait).
[0060] Table 2
[0061] A2. Consider the component of thrust along the fuselage axis. , , The trigonometric projection of the aerodynamic resultant force from the wind axis system to the body axis system, and the dimensionless sum of damping terms. The derivation of triaxial acceleration is as follows: ; ; ; in: For dynamic pressure, For wing reference area, air density, determined by the standard atmospheric model. and calculate; , indicating lift; , indicating resistance; , , Engine thrust in the airframe coordinate system , , Components on the shaft: Magnitude of thrust Based on engine performance model , , The direction is determined by referring to a table and is along the engine mounting axis; since the mounting angle is usually small, this is important. , , Approximately 0; , The dimensionless coefficient is the sum of the damping term and the control surface correction term; , which is the standard gravitational acceleration.
[0062] B. Calculate the composite residual: Compare the combined load with the measured overload to calculate the composite residual. ; Wherein, the normal overload residual is .
[0063] The Euler norm is used to evaluate the load obtained by "inverse construction" of the equation at this time and the measured load fed back by the actual QAR gravity sensor. The deviation is used to generate a comprehensive residual index. And the partial axis residuals in each direction (such as normal overload residuals) ).
[0064] S5. Perform convergence determination: If the comprehensive residual does not meet the preset convergence condition, the nonlinear correction factor is updated by combining the flow direction scalar characteristics of the normal overload residual and the original vertical wind speed estimate, and fed back to the decoupled mapping end to trigger closed-loop iterative solution. When the comprehensive residual satisfies the convergence condition or reaches the maximum number of iterations, the iteration stops, and the current energy domain vertical wind speed estimate and the corrected aerodynamic angle parameter are output as environmental simulation parameters.
[0065] Preferably, the nonlinear correction factor is updated by combining the normal overload residual and the flow direction scalar characteristics of the original vertical wind speed estimate, including: Extract the nonlinear correction factor from the previous iteration; Determine the direction and magnitude of the floating deviation of the measured normal component in the measured acceleration relative to the theoretical normal component in the theoretical acceleration; The flow direction scalar feature is used to characterize the direction of the original vertical wind speed estimate. The flow direction scalar feature is used as a direction compensator to eliminate the negative feedback sign interference caused by the opposite direction of the upward or downward airflow. The adjustment step size is generated by combining the deviation amplitude, the preset feedback gain ratio and the compensated adjustment direction. The adjustment step size is superimposed on the nonlinear correction factor of the previous iteration to obtain the updated nonlinear correction factor. The updated nonlinear correction factor is truncated using a preset boundary limit function.
[0066] More preferably, the convergence condition is that the comprehensive residual index is less than a preset convergence threshold. In a specific embodiment, the convergence threshold is set to 0.05g, which is determined based on the typical measurement accuracy of an airborne accelerometer (e.g., 0.01-0.02g) and taking into account model error tolerance. The convergence threshold is relaxed to 0.08g after triggering under non-steady-state conditions. Convergence is determined when the value is less than the current threshold.
[0067] More preferably, convergence determination is performed, and closed-loop iterative solution is executed, specifically including: Round 0 (Initial Solution): An initial value is taken, which is given by the MLP model based on the current flight state during the first iteration. If the MLP does not provide a valid prediction, a value of 1.0 is used as a neutral correction. The first round of energy domain calculation and force domain verification is completed in the order of steps two, three, and four. and residuals of each axis , , .
[0068] The subsequent nth round ( ): when If the value is less than the convergence threshold, the iteration is considered converged, and the current round's result is output. , , As the final result.
[0069] like If the convergence condition is not met, then the normal overload residual is used. For the core driving signal, update the nonlinear correction factor: ; ; in, This is the nonlinear correction factor from the previous iteration. This is the feedback gain coefficient; The scalar characteristic of the original vertical wind speed estimate is the flow direction, and the updraft is taken as... The burst is incremented by 1, which serves as a compensator for the direction of the scalar characteristic flow. The aerodynamic polarity compensation factor characterizes the current lift line slope. The symbol is +1 for the attached flow region and +1 for the deep stall region. , used to synchronously flip the adjustment direction of σ when the aerodynamic polarity is reversed; The function restricts the value of σ to a certain range. Inside.
[0070] The design of the item is based on: different airflows (rising / falling). The increase or decrease of the angle of attack has an opposite effect on the direction of the airflow direction reversal, and this compensation term is used to eliminate this airflow direction reversal effect. The design of the project is based on the fact that in the deep stall region, the slope of the lift line is... As the aerodynamic polarity shifts from positive to negative, the gain polarity of σ, transmitted through the angle of attack to lift, also reverses. This compensation term is used to synchronously adjust the direction when the aerodynamic polarity reverses. The two terms combined ensure that regardless of the airflow type, whether in the attached flow region or the deep stall region, The adjustment direction of σ always forms a stable negative feedback relationship.
[0071] For example, under the condition of rising airflow (i.e., the original decoupling is in Right now For example, the aerodynamic polarity compensation factor Taking +1, the complete self-healing derivation chain is as follows: Scenario A: When the estimated combined lift from the environment is too high, i.e. ( The deductive chain is as follows: → Decrease → Reduce → After PINN deduction Decrease → Decrease → Lift decreases → Descending → Approaching 0.
[0072] Scenario B: When the estimated combined lift from the environment is too small, i.e. ( The deductive chain is as follows: → Increase → Increase → After PINN deduction Increase → Increase → Increase lift → Rise → Approaching 0.
[0073] Under the condition of hitting the storm ( , Substituting the values below and verifying, we can obtain the same formula. The adjustment of direction conforms to the logic of negative feedback, which will not be elaborated here.
[0074] Furthermore, when it is determined that the region has entered a deep stall zone ( When the stall critical point is crossed and the lift coefficient enters the negative slope region, the aerodynamic polarity compensation factor... Pick 1, with The combined effect causes the adjustment direction of σ to be reversed overall. This is still under the condition of updraft (…). For example, when the environmental composite lift estimate is too high ( When ), the deduction chain is as follows: → Increase → Increase → After PINN deduction Increase → Reduce (negative slope in the stall region) → Decrease → The descent approaches 0, and the negative feedback relationship still holds after the aerodynamic polarity reverses; the same logic applies when substituting the downburst condition for verification.
[0075] Updated Feedback is sent to step S2, triggering a new round of energy domain calculation, neural deduction, and force domain verification, until... The process continues until the convergence condition is met or the maximum number of iterations (e.g., 50) is reached.
[0076] Add sign and After the double-sign compensation feature, regardless of the initial airflow direction or whether the current flow is in the attached flow region or the deep stall region, this update term can always ensure that when the load synthesis prediction distortion is too large, σ is adjusted in the direction of reducing the theoretical normal overload (in the attached flow region: σ is reduced under updraft conditions and increased under downburst conditions; in the deep stall region, the direction of σ's increase or decrease is reversed due to the reversed sign of the lift line slope); when the synthesis is too small, it is adjusted in the opposite direction to backfill the lift manifold, forming a robust negative gradient descent.
[0077] The second embodiment of this application provides a flight environment parameter inversion method based on energy and load closed-loop feedback. Building upon the first embodiment, the method further includes a cross-stage linkage correction mechanism for dealing with unsteady operating conditions.
[0078] The cross-stage linkage correction mechanism is specifically as follows: The normal measured component of the measured acceleration is extracted in real time, and the first derivative feature of the normal measured component with respect to time is calculated and used as the vertical overload change rate. When the vertical overload rate of change exceeds the preset polarization characteristic reference boundary, the following global correction strategy is triggered: In the flight data preprocessing stage, the gain coefficient of the complementary filter used to calculate the vertical velocity is increased to enhance the relaxation tolerance for extrapolating the long-period characteristics of the tracking. During the deduction phase, the penalty consideration weight of the constraint residual term in the physical information neural network is doubled to compensate for data distortion under extreme conditions; During the closed-loop feedback phase, the convergence determination threshold used for threshold blocking is amplified to a preset value to improve the system's obstacle-crossing convergence capability under non-stationary flight response mode.
[0079] In this embodiment, as Figure 2 As shown, the system reads the measured normal overload in real time according to the QAR sampling period (8Hz). Taking the time derivative yields the rate of change of vertical overload. ;when When the preset unsteady-state trigger threshold (nominal value 0.3 g / s) is exceeded, it indicates that the aircraft may enter an unsteady-state critical condition such as a thunderstorm microburst or deep stall. The system then triggers a one-to-many cascaded cooperative operation: (1) Preprocessing stage: Start the sliding data window re-run mechanism to merge the first-order complementary filter used in the true vertical velocity IVV solution. The coefficient was increased from 0.85 to 0.95, and the increased coefficient was used to perform a secondary calculation and overlay on the QAR buffer data within the current time window to enhance the filter's ability to track unsteady long-period disturbances. The application location and effect propagation path of this coefficient are as follows: The fusion weights of the barometric altitude channel and the IRS inertial channel in the complementary filter are determined, and their adjustment directly changes the true vertical velocity. The solution results; As input to the power balance equation in step two: ; Then through the linear decoupling operator This is transmitted to the estimated vertical wind speed. Therefore, The improvement is not an isolated parameter adjustment; its effect is ultimately reflected in the improved accuracy of vertical wind calculation in the energy domain through the power balance equation.
[0080] (2) Derivation stage: The weight λ of the physical constraint term relative to the data fitting term in the PINN loss function is increased from 1.0 to 2.0 to enhance the constraint of the physical prior of the motion equation on the network training, make up for the data defects caused by the distortion and scattering of some sensor channel data under the sudden change of wind field, and prevent the network fitting from being biased by the distorted data points.
[0081] (3) Closed-loop feedback stage: integrate residuals The convergence threshold is temporarily relaxed from 0.05g to 0.08g. The comprehensive residual is defined as follows: ; That is, the three-axis overload residual , , The equal-weighted Euclidean norm—the three-axis residuals originate from the same accelerometer component, are uniformly expressed in g, and have the same order of magnitude of measurement accuracy—is used. Therefore, the weight of each axis is set to 1. This equal-weighting process ensures that the convergence criterion remains physically neutral for the three axes, avoiding the introduction of subjective weighting parameters. It should be noted that... It is used only as a convergence criterion for iteration termination throughout the entire closed loop; the adjustment direction and magnitude of the σ factor are determined solely by the normal overload residual. The driving force (σ acts on the vertical wind mapping end, and its error force domain response is concentrated in the normal overload channel). The technical basis for relaxing the threshold is that sensor distortion and aerodynamic model mismatch in the stall region under unsteady conditions will increase the noise level of the triaxial residual. If the initial threshold of 0.05g is maintained, the closed-loop iteration may run to the upper limit of 50 times and still fail to meet the convergence condition. Relaxing the threshold to 0.08g matches the noise level of the residual under extreme conditions, ensuring that the closed-loop iteration terminates reliably within a reasonable accuracy range.
[0082] A third embodiment of this application proposes a flight environment parameter inversion system based on energy and load closed-loop feedback, comprising: The data acquisition module is used to acquire aircraft observation data and raw aerodynamic database; The kinematics preprocessing module is used to perform coordinate system datum unification and heading adaptive calibration on the acquired aircraft observation data, and output the preprocessed state quantities under the unified datum. The energy domain solution module is used to construct a power balance equation based on the conservation of mechanical energy, generate the original vertical wind speed estimate through energy decoupling, introduce a nonlinear correction factor, and use it to correct the original vertical wind speed estimate to obtain the energy domain vertical wind speed estimate. The neural inference module is used to input the estimated vertical wind speed in the energy domain and the state variables in the aircraft observation data into a preset physical constraint network model to deduce the corrected aerodynamic angle parameters. The cross-domain verification module is used to substitute the modified aerodynamic angle parameters into the dynamic model, call the basic aerodynamic coefficients in the original aerodynamic database to synthesize the theoretical acceleration, and compare it with the measured acceleration in the aircraft observation data to generate normal overload residuals and comprehensive residuals. The closed-loop feedback module is used to update the nonlinear correction factor by combining the flow direction scalar characteristics of the normal overload residual and the original vertical wind speed estimate when the comprehensive residual does not meet the preset convergence condition, and feed it back to the decoupled mapping end to trigger closed-loop iterative solution; until the comprehensive residual meets the convergence condition or reaches the maximum number of iterations, the current energy domain vertical wind speed estimate and the corrected aerodynamic angle parameter are output as environmental extrapolation parameters.
[0083] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the system described above can be found in the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0084] It should be noted that the flight environment parameter inversion system based on energy and load closed-loop feedback provided in the above embodiments is only an example of the division of the above functional units. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of this application can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of this application are only for distinguishing the various modules or steps and are not considered as an improper limitation of this application.
[0085] A device according to the fourth embodiment of this application includes: At least one processor; and memory that is communicatively connected to at least one processor; The memory stores instructions that can be executed by the processor to implement the aforementioned method for inverting flight environment parameters based on energy and load closed-loop feedback.
[0086] A computer-readable storage medium according to the fifth embodiment of this application stores computer instructions that are executed by a computer to implement the above-described method for inverting flight environment parameters based on energy and load closed-loop feedback.
[0087] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the storage device and processing device described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0088] The following is for reference. Figure 3 It shows a schematic diagram of the structure of a computer system for implementing the methods, apparatus, and electronic devices of this application. Figure 3 The server shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.
[0089] like Figure 3 As shown, the computer system includes a Central Processing Unit (CPU) 201, which can perform various appropriate actions and processes based on programs stored in Read Only Memory (ROM) 202 or programs loaded from storage into Random Access Memory (RAM) 203. The RAM 203 also stores various programs and data required for system operation. The CPU 201, ROM 202, and RAM 203 are interconnected via a bus 204. An Input / Output (I / O) interface 205 is also connected to the bus 204.
[0090] The following components are connected to I / O interface 205: an input section 206 including a keyboard, mouse, etc.; an output section 207 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 208 including a hard disk, etc.; and a communication section 209 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 209 performs communication processing via a network such as the Internet. A drive 210 is also connected to I / O interface 205 as needed. Removable media 211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 210 as needed so that computer programs read from them can be installed into storage section 208 as needed.
[0091] Specifically, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 209, and / or installed from removable medium 211. When the computer program is executed by central processing unit (CPU) 201, it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.
[0092] Computer program code for performing the operations of this application can be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, and C++, and conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0093] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0094] In the description of this application, it should be noted that the terms "first," "second," "primary," "secondary," etc., are used only to distinguish the objects being described and do not indicate or imply relative importance, nor should they be construed as limiting a specific order. The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to those processes, methods, articles, or apparatus / devices.
[0095] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. Without departing from the principles of this application, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of this application.
Claims
1. A method for inverting flight environment parameters based on closed-loop feedback of energy and load, characterized in that, Includes the following steps: Acquire aircraft observation data and raw aerodynamic database, construct power balance equations based on mechanical energy conservation, and generate raw vertical wind speed estimates through energy decoupling; A nonlinear correction factor is introduced and used to correct the original vertical wind speed estimate to obtain the energy domain vertical wind speed estimate. The estimated vertical wind speed in the energy domain and the state variables in the aircraft observation data are input into a preset physical constraint network model to deduce the corrected aerodynamic angle parameters. Substitute the corrected aerodynamic angle parameters into the dynamic model, call the basic aerodynamic coefficients in the original aerodynamic database to synthesize the theoretical acceleration, and compare it with the measured acceleration in the aircraft observation data to generate normal overload residuals and comprehensive residuals. If the comprehensive residual does not meet the preset convergence condition, the nonlinear correction factor is updated by combining the flow direction scalar characteristics of the normal overload residual and the original vertical wind speed estimate, and fed back to the decoupled mapping end to trigger closed-loop iterative solution. Until the comprehensive residual satisfies the convergence condition or reaches the maximum number of iterations, the current energy domain vertical wind speed estimate and the corrected aerodynamic angle parameter are output as environmental simulation parameters.
2. The method as described in claim 1, characterized in that, After acquiring the observation data from the aircraft, data preprocessing is performed, specifically as follows: The aircraft observation data in the body coordinate system is converted to the aerodynamic wind axis coordinate system, and magnetic heading adaptive correction is performed based on the preset geomagnetic model and positioning quality index. At the same time, the vertical velocity is calculated by fusing barometric altitude characteristics and inertial navigation data based on complementary filtering algorithm, and the preprocessed state variables are output.
3. The method as described in claim 2, characterized in that, A power balance equation is constructed based on the conservation of mechanical energy. The original vertical wind speed estimate is generated through energy decoupling, including: The preprocessed state variables are analyzed to obtain the aircraft's altitude parameters, mass parameters, current thrust, and current flight drag. The thrust power is determined based on the thrust and airspeed, the drag power is determined based on the flight drag and airspeed, the gravitational potential energy rate of change is determined based on the mass parameter and the vertical velocity, and the kinetic energy rate of change is determined based on the mass parameter and airspeed. Based on the thrust power, drag dissipation power, gravitational potential energy rate of change and kinetic energy rate of change, calculate the overall power disturbance residual of the aircraft. Based on the original aerodynamic database, the current reference lift of the aircraft is determined, and based on the mapping ratio of the overall power disturbance residual to the current reference lift, the original vertical wind speed estimate is generated by decoupling.
4. The method as described in claim 3, characterized in that, The original vertical wind speed estimate is corrected to obtain the energy domain vertical wind speed estimate, specifically including: Based on the aircraft's pitch and track tilt angles, the original vertical wind speed estimate, and airspeed, the effective angle of attack is calculated. When the absolute value of the effective angle of attack is less than the preset angle threshold, and the absolute value of the ratio of the original vertical wind speed estimate to the air speed is less than the preset ratio limit, it is determined that the small angle interval condition is met, and the original vertical wind speed estimate is used as the energy domain vertical wind speed estimate. If the small angle interval condition is not met, the nonlinear decoupling mode is activated, and the nonlinear correction factor is used as a gain coefficient to compensate the original vertical wind speed estimate, thereby calculating the energy domain vertical wind speed estimate.
5. The method as described in claim 1, characterized in that, The corrected aerodynamic angle parameters include corrected angle of attack and corrected sideslip angle; the physical constraint network model is constructed based on a physical information neural network, and obtaining the corrected aerodynamic angle parameters includes: The current manipulation sequence characteristics, the estimated vertical wind speed in the energy domain, and the corrected aerodynamic angle parameters output in the previous round are input into the physical information neural network to deduce the aerodynamic angle correction amount. The original apparent aerodynamic angles in the aircraft observation data are superimposed with the aerodynamic angle correction amount to obtain the corrected angle of attack and the corrected sideslip angle, which are used as the corrected aerodynamic angle parameters. The loss function inside the physical information neural network is composed of the fitting deviation term of the data observation channel and the constraint residual term that characterizes the nonlinear physical logic at large angles of attack. The constraint residual term covers the lateral force balance physical constraint, the normal force balance physical constraint, and the vertical velocity consistency constraint, which include the projection of the complete spatial nonlinear trigonometric function.
6. The method as described in claim 2, characterized in that, By combining the normal overload residual and the flow scalar characteristics of the original vertical wind speed estimate, the nonlinear correction factor is updated, including: Extract the nonlinear correction factor from the previous iteration; Determine the direction and magnitude of the floating deviation of the measured normal component in the measured acceleration relative to the theoretical normal component in the theoretical acceleration; The flow direction scalar feature is used as a direction compensator to eliminate the negative feedback sign change interference caused by the opposite direction of the upward or downward airflow. The adjustment step size is generated by combining the deviation amplitude, the preset feedback gain ratio and the compensated adjustment direction. The adjustment step size is superimposed on the nonlinear correction factor of the previous iteration to obtain the updated nonlinear correction factor. The updated nonlinear correction factor is truncated using a preset boundary limit function.
7. The method as described in claim 6, characterized in that, When the closed-loop iterative solution is activated for the first time, the nonlinear correction factor is introduced as follows: A multilayer perceptron model that has converged and is pre-trained on a stall dataset is used to predict the initial compensation value as the nonlinear correction factor based on the current macroscopic flight state characteristics of the aircraft. If the multilayer perceptron model does not output a valid prediction result, then the neutral benchmark value representing no correction effect will be used as the nonlinear correction factor. The convergence condition is that the comprehensive residual index is less than a preset convergence threshold, and the multilayer perceptron model is obtained by training based on historical test sets.
8. The method as described in claim 7, characterized in that, The method also includes a cross-stage linkage correction mechanism for dealing with unsteady operating conditions, specifically: The normal measured component of the measured acceleration is extracted in real time, and the first derivative feature of the normal measured component with respect to time is calculated and used as the vertical overload change rate. When the vertical overload rate of change exceeds the preset polarization characteristic reference boundary, the following global correction strategy is triggered: In the flight data preprocessing stage, the gain coefficient of the complementary filter used to calculate the vertical velocity is increased to enhance the relaxation tolerance for extrapolating the long-period characteristics of the tracking. During the extrapolation phase, the penalty consideration weights of the constraint residual terms in the physical information neural network are doubled to compensate for data distortion under extreme conditions. During the closed-loop feedback phase, the convergence determination threshold used for threshold blocking is amplified to a preset value.
9. The method as described in claim 1, characterized in that, During the process of synthetic theoretical acceleration: Real-time monitoring of the aircraft’s current corrected angle of attack; when the current corrected angle of attack is assessed to have crossed the preset stall critical point and entered the deep stall region, the aerodynamic coefficient algorithm based on database lookup table is disabled. The semi-empirical aerodynamic model with built-in nonlinear decay factor and maximum stall lift characteristics is invoked to take over the calculation of the lift coefficient, and the lift coefficient in this state is used to correct the evaluation of flight drag in the power balance equation in order to calculate drag dissipation power.
10. A flight environment parameter inversion system based on energy and load closed-loop feedback, characterized in that, include: The data acquisition module is used to acquire aircraft observation data and raw aerodynamic database; The kinematics preprocessing module is used to perform coordinate system datum unification and heading adaptive calibration on the acquired aircraft observation data, and output the preprocessed state quantities under the unified datum. The energy domain solution module is used to construct a power balance equation based on the conservation of mechanical energy, generate the original vertical wind speed estimate through energy decoupling, introduce a nonlinear correction factor, and use it to correct the original vertical wind speed estimate to obtain the energy domain vertical wind speed estimate. The neural inference module is used to input the estimated vertical wind speed in the energy domain and the state variables in the aircraft observation data into a preset physical constraint network model to deduce the corrected aerodynamic angle parameters. The cross-domain verification module is used to substitute the modified aerodynamic angle parameters into the dynamic model, call the basic aerodynamic coefficients in the original aerodynamic database to synthesize the theoretical acceleration, and compare it with the measured acceleration in the aircraft observation data to generate normal overload residuals and comprehensive residuals. The closed-loop feedback module is used to update the nonlinear correction factor by combining the flow direction scalar characteristics of the normal overload residual and the original vertical wind speed estimate when the comprehensive residual does not meet the preset convergence conditions, and feed it back to the decoupled mapping end to trigger closed-loop iterative solution. Until the comprehensive residual satisfies the convergence condition or reaches the maximum number of iterations, the current energy domain vertical wind speed estimate and the corrected aerodynamic angle parameter are output as environmental simulation parameters.
