Method and system for joint estimation of state and feature parameters based on maneuver strength perception

Abstract

This invention discloses a joint estimation method and system for state and characteristic parameters based on maneuver intensity perception. It primarily addresses the problems of filter divergence and low parameter estimation accuracy caused by sudden changes in target maneuvering when estimating parameters for hypersonic targets in existing technologies. The implementation scheme is as follows: receiving observation data and calculating innovation; introducing a forgetting factor to adaptively update the process noise covariance, and using the innovation norm to construct a gain factor to adjust the observation weights, completing the posterior state update; calculating the sliding window length based on the adaptive process noise covariance, and constructing the arrival cost using the output of the posterior state update; transforming physical constraints into a penalty function, and combining the arrival cost to establish a rolling time-domain estimation objective function and solve for the optimal estimate. This invention can perceive maneuver intensity in real time, dynamically adjust the estimation window and observation weights, significantly improving the accuracy and robustness of the joint estimation, and can be applied to guidance information extraction and trajectory prediction for near-space hypersonic gliders.

Description

Technical Field

[0001] This invention belongs to the field of target tracking and state estimation technology, and specifically relates to a joint estimation method and system for rolling time domain estimation, which can be applied to the extraction of guidance information and trajectory prediction of near-space hypersonic gliders. Background Technology

[0002] Near-space hypersonic glide vehicles are characterized by high speed, strong maneuverability, and wide flight envelope. During the gliding phase, they perform large-amplitude maneuvers by frequently adjusting aerodynamic forces, making target state estimation and identification of maneuver characteristic parameters a key technical challenge in air defense and missile defense systems. Accurately estimating the vehicle's motion state and maneuver characteristic parameters, including lift and drag parameters, as well as roll angles, is crucial for target trajectory prediction and guided interception.

[0003] Existing target state estimation methods are mainly based on nonlinear filtering techniques such as Extended Kalman Filter (EKF) and Capacitive Kalman Filter (CKF). The standard CKF uses a third-order spherical-radial volume criterion to numerically approximate the mean and covariance of the nonlinear function, avoiding the linearization error introduced by Taylor expansion in the EKF, and exhibits high estimation accuracy when dealing with strongly nonlinear systems. However, the standard CKF has two shortcomings when dealing with frequent maneuvers of hypersonic targets: firstly, the process noise covariance matrix... The filtering process remains unchanged, typically relying on prior experience for offline selection. However, when the aircraft's maneuvering intensity changes abruptly, the fixed... The inability to accurately reflect real-time changes in system dynamics leads to deviations in the estimation of prediction error covariance, causing a decrease in filtering accuracy or even divergence; secondly, in the state update of the standard CKF, the Kalman gain... When a fixed weight is applied to the innovation, the innovation residual increases sharply when the aircraft performs strong maneuvers. The fixed gain does not have the ability to adaptively adjust to this, and cannot make full use of the observation information to quickly correct the state when the innovation is large.

[0004] Furthermore, simple filtering methods fail to explicitly utilize the physical constraints of hypersonic vehicles, such as dynamic pressure constraints, overload constraints, and heat flux density constraints. Rolling time-domain estimation (MHE) can directly express the system's constraints in the optimization problem, dynamically satisfying the constraints through online rolling optimization. However, standard MHE has two core problems: one is the model confidence matrix... Typically, the selection relies on experience, making it difficult to accurately reflect the dynamic uncertainty of the system when process noise is unknown. Secondly, the MHE assumption makes it constant for the parameters to be estimated within the sliding window, and a fixed window length makes it difficult to balance computational complexity and estimation accuracy. Existing studies have attempted to use piecewise functions to construct gain factors to alleviate the problem of fixed observation weights, but piecewise functions have weight jumps at the segmentation points, and improper threshold selection can introduce additional estimation jitter, making it difficult to guarantee robustness.

[0005] Patent application CN201611222049.7 discloses an adaptive filtering method and system for NSHV tracking filtering. Based on a dynamic model and radar measurement model with an unknown angle of attack, it uses a suboptimal unbiased noise statistical estimator to recursively estimate the covariance of process noise and observation noise. It indirectly adjusts the Kalman gain by updating the noise statistics online to improve filtering stability and tracking accuracy. However, because this method relies solely on a single noise statistics recursive mechanism, it lags in response to sudden maneuvers and lacks the ability to adjust observation weights in real time at the current time step. Furthermore, since the Kalman gain still applies a fixed linear weight to the innovation, it cannot dynamically adjust the contribution intensity of observation corrections based on the innovation amplitude. In addition, because this method is limited to the filter level and does not explicitly incorporate flight physical constraints such as dynamic pressure, overload, and heat flux density into the estimation framework, the estimation accuracy and robustness are poor. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of the prior art by proposing a joint estimation method and system for state and feature parameters based on maneuver intensity perception, so as to improve the response speed and tracking sensitivity of the filter to maneuver changes, enhance the adaptive capability to maneuver intensity, and improve the accuracy and robustness of joint estimation.

[0007] The technical approach to achieve the objective of this invention is as follows: By online smoothing update of the process noise covariance based on the Sage-Husa principle and constructing a gain factor based on the Sigmoid function, the response speed and tracking sensitivity of the filter to sudden maneuvers are improved. By utilizing the Frobenius norm change rate of the adaptive process noise covariance to sense maneuver intensity, the dynamic adjustment of the sliding window length for rolling time-domain estimation is driven, and the adaptive arrival cost is constructed by reusing the real-time output of the adaptive filter, enhancing the adaptability to different maneuver intensities. By explicitly introducing flight physics constraints such as dynamic pressure, overload, and heat flux density into the rolling time-domain estimation optimization framework using the penalty function method, and employing a particle swarm optimization algorithm for gradient-free optimization, the accuracy and robustness of the joint estimation of hypersonic target motion state, aerodynamic parameters, and roll angle are improved.

[0008] Based on the above ideas, the technical solution of the present invention includes:

[0009] 1. A method for jointly estimating state and characteristic parameters based on maneuver intensity perception, characterized in that it includes:

[0010] (1) Receive discrete observation data and generate observation spatial volume points. The predicted state value is calculated based on the observed spatial volume point. Observational prediction values Kalman gain and new information vector ;

[0011] (2) Through the forgetting factor process noise covariance Perform adaptive updates and utilize the innovation vector. Calculating the gain factor using the Frobenius norm ;

[0012] (3) Utilizing Kalman gain Perform state updates to obtain posterior state estimates. and posterior covariance matrix Then return to (1) above to perform the next time step filtering recursion;

[0013] (4) Based on the noise covariance of the adaptive process Calculate the adaptive sliding window length Based on this length, the posterior state at the start time of the corresponding window is used. and posterior covariance , construct arrival cost ;

[0014] (5) Construct a penalty function based on the constraints of dynamic pressure, overload and heat flux density of the aircraft. Combined with this penalty function Cost of arrival and adaptive process noise covariance Establish a rolling time-domain estimation objective function and calculate the state vector estimate.

[0015] Furthermore, through the forgetting factor process noise covariance Adaptive updates are achieved by introducing a forgetting factor. Calculate the dynamic weighting coefficients The process noise covariance is calculated based on the Sage-Husa adaptive estimation principle. Adaptive update; and utilizing the new information vector Calculating the gain factor using the Frobenius norm It uses the continuous sigmoid function to calculate the gain factor. :

[0016] Furthermore, in the above (4), based on the adaptive process noise covariance Calculate the adaptive sliding window length , construct arrival cost Its implementation includes:

[0017] (4a) Define the normalized maneuver intensity index :

[0018] (4b) will Mapped to normalized maneuver strength via the Sigmoid function. :

[0019] (4c) Based on Calculate the target window length at the current time. :

[0020] (4d) Based on window length Utilizing the window start time Posterior state estimation and posterior covariance Calculate the arrival cost .

[0021] 2. A joint estimation system for state and characteristic parameters based on maneuver intensity perception, characterized in that it comprises:

[0022] The observation data receiving and information calculation module is used to receive discrete observation data and generate observation spatial volume points. The predicted state value is calculated based on the observed spatial volume point. Observational prediction values Kalman gain and new information vector ;

[0023] The adaptive noise and gain factor update module is used to update noise and gain factors. process noise covariance Perform adaptive updates and utilize the innovation vector. Calculating the gain factor using the Frobenius norm ;

[0024] The state and covariance posterior update module is used to utilize Kalman gain. Predicted state values Perform state updates to obtain posterior state estimates. and posterior covariance matrix Then it is returned to the observation data receiving and information calculation module for filtering and recursion at the next time step;

[0025] Adaptive window and arrival cost building blocks are used to build upon the adaptive process noise covariance. Calculate the adaptive sliding window length Based on this length, the posterior state at the start time of the corresponding window is used. and posterior covariance , construct arrival cost ;

[0026] The constraint MHE objective function construction and solution module is used to construct penalty functions based on the constraints of dynamic pressure, overload, and heat flux density of the aircraft. Combined with this penalty function Cost of arrival and adaptive process noise covariance Establish a rolling time-domain estimation objective function and calculate the state vector estimate.

[0027] Compared with the prior art, the present invention has the following advantages:

[0028] Firstly, this invention employs online updating of the process noise covariance based on Sage-Husa, and uses the Frobenius norm change rate of the process noise covariance to sense the maneuver intensity and the length of the driving sliding window in real time. Adaptive change can adjust the time-domain estimation window in real time according to the strength of the aircraft's maneuvering, thus getting rid of the problem of delayed response to sudden maneuvering caused by traditional methods that rely solely on single noise recursion. It has the ability to adjust the observation weights in real time at the current time step.

[0029] Secondly, this invention constructs an adaptive gain factor based on the Sigmoid function. It can dynamically adjust the contribution intensity of observation correction based on the nonlinearity of the innovation vector amplitude, replacing the traditional Kalman gain fixed linear weight mechanism, making the filter more sensitive to strong maneuvers and more stable in the stationary segment.

[0030] Thirdly, this invention explicitly introduces physical constraints such as dynamic pressure, overload, and heat flux density into the rolling time-domain estimation framework through a penalty function, and combines adaptive arrival cost and adaptive window. Achieving deep integration of filtering and optimization, the joint estimation of state, aerodynamic parameters and roll angle strictly conforms to physical boundaries and maneuver characteristics, significantly improving the accuracy and robustness of the estimation results. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the implementation of the joint estimation method for state and characteristic parameters based on maneuver intensity perception in this invention.

[0032] Figure 2 This is a block diagram of a joint estimation system for state and characteristic parameters based on maneuver intensity perception;

[0033] Figure 3 This is a flight trajectory diagram under the simulated maneuvering scenario of this invention;

[0034] Figure 4 This is a graph showing the actual values ​​of aerodynamic parameters under the simulation settings of this invention in a motorized scenario;

[0035] Figure 5 This invention simulates the tilt angle in a maneuvering scenario. True value curve;

[0036] Figure 6 This is a process constraint curve diagram of the aircraft under maneuvering scenarios according to the present invention;

[0037] Figure 7 This invention relates to the gain factor in mobile scenarios. The response curve;

[0038] Figure 8 This is the window length of the present invention in a mobile scenario. Resulting image;

[0039] Figure 9 This invention, compared to existing technologies, addresses the differences in aerodynamic parameters and roll angles during maneuvering scenarios. Comparison chart of estimation results. Detailed Implementation

[0040] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative effort should all fall within the protection scope of the present invention.

[0041] It should be noted that the step numbers in the specification and claims of this invention are only for the purpose of clearly describing the embodiments of this invention and facilitating understanding, and their order is not limited.

[0042] Example 1: A joint estimation method of state and characteristic parameters based on maneuver intensity perception.

[0043] Reference Figure 1 The implementation steps of this example include:

[0044] Step 1: Receive discrete observation data and calculate filter prediction and update parameters.

[0045] This step involves acquiring discrete observation data using radar sensors, combining this data with the nonlinear dynamic characteristics of the hypersonic vehicle, constructing an augmented state system model, and generating volume points using the third-order spherical-radial volume criterion. Nonlinear propagation of state and observation is performed to calculate state prediction values. Observational prediction values Kalman gain and new information vector The implementation steps include:

[0046] 1.1) Acquire observation data through radar sensors and establish a system based on geocentric distance. ,longitude ,latitude ,speed Track inclination Track deflection The six-dimensional motion state vector represents the basic motion state, and the aerodynamic parameters to be estimated are... , and tilt angle The augmented state vector is then constructed into a nine-dimensional augmented state vector. :

[0047] , Indicates transpose;

[0048] 1.2) Based on the above-constructed nine-dimensional augmented state vector A continuous-time dynamic model of a hypersonic vehicle was established and numerically discretized to obtain the state equations of the nonlinear discrete-time system:

[0049]

[0050] in, It is a nonlinear state transition function. For the first Process noise at any given moment;

[0051] 1.3) Based on radar observation data, establish the observation equations for the nonlinear system:

[0052]

[0053] in, For the observation function, For the first Observation noise at each moment;

[0054] 1.4) During the time update phase, the posterior covariance matrix at the initial time step Based on the manually set sensor accuracy and prior model error, the subsequent posterior covariance matrix is ​​recursively calculated using an adaptive gain factor capacitive Kalman filter, based on the posterior covariance matrix of the previous time step. Perform matrix decomposition to obtain its square root matrix.

[0055]

[0056] ,

[0057] in, and These are the orthogonal matrix and the diagonal matrix obtained by performing SVD decomposition on the posterior covariance matrix, respectively.

[0058] 1.5) Based on the square root matrix Generate state space volume points :

[0059]

[0060] in, , It is the dimension of the state vector. It is the first One standard volume point, It is the first of the identity matrix. List;

[0061] 1.6) Convert the state space volume point The predicted volume point at the current moment is obtained by propagation through a nonlinear state equation. :

[0062] ;

[0063] 1.7) By predicting volume points Calculate state prediction value and prediction error covariance :

[0064] ;

[0065] ;

[0066] in, , It is transpose. It is the first The process noise covariance matrix at each time step;

[0067] 1.8) During the observation update phase, the prediction error covariance is... SVD decomposition is performed as follows: To obtain its square root matrix: Based on the square root matrix and state prediction value Generate observation space volume points :

[0068] .

[0069] 1.9) Based on the observed spatial volume points Calculate state prediction value Observational prediction values Kalman gain and new information vector :

[0070] 1.9.1) Observe the spatial volume point Propagate to the observation equation and calculate the observed prediction value. Observation covariance State-observation cross-covariance , No. Kalman gain at time step :

[0071] ;

[0072] ;

[0073] ;

[0074] ;

[0075] in, , It is transpose. It is the first The process noise covariance matrix at each time step.

[0076] 1.9.2) Based on observed predicted values Calculate the innovation vector :

[0077] ,

[0078] in, It is the first The actual observations at each moment.

[0079] Step 2, Adaptive update of process noise covariance Calculate the gain factor .

[0080] The process noise covariance It is a key statistical parameter in the Kalman filter framework that describes the uncertainty of the dynamic model of the system. It reflects the degree of deviation of the target state prediction value caused by modeling errors, random environmental disturbances such as atmospheric disturbances, and small physical quantities ignored in the state prediction equation.

[0081] This step is based on the Sage-Husa adaptive estimation principle, introducing a forgetting factor. Constructing dynamic weight coefficients Using the new information vector and Kalman gain right Perform smooth updates using the innovation vector. The gain factor is constructed by combining the Frobenius norm with a continuous Sigmoid function. The implementation steps include:

[0082] 2.1) Introducing a forgetting factor Calculate the dynamic weighting coefficients The process noise covariance is calculated based on the Sage-Husa adaptive estimation principle. Adaptive updates:

[0083] ;

[0084] ,

[0085] in, It is the first The information vector at each moment, It is the first Kalman gain at each time step;

[0086] 2.2) Calculate the gain factor using the continuous Sigmoid function :

[0087] ,

[0088] in, Based on the bias, The amplitude coefficient, It is a continuous sigmoid function. The slope coefficient, for The information threshold point where significant changes occur.

[0089] Step 3, using Kalman gain Calculate posterior state estimates and posterior covariance matrix .

[0090] This step utilizes Kalman gain. By completing the posterior update of the prediction error covariance, a high-precision and high-stability posterior state estimate for the current time step is obtained. and posterior covariance matrix The implementation steps include:

[0091] 3.1) Utilizing Kalman gain Perform a state update to obtain a posterior state estimate. and posterior covariance matrix The formulas are as follows:

[0092] ,

[0093] ,

[0094] in, It is a state prediction value. It is the first The actual observations at each moment These are observed and predicted values. Gain factor It is the prediction error covariance. It is the observation covariance.

[0095] Step 4: Calculate the adaptive sliding window length , construct arrival cost .

[0096] This step is based on the adaptive process noise covariance. Calculate the adaptive sliding window length Based on this length, the posterior state at the start time of the corresponding window is used. and posterior covariance , construct arrival cost The implementation steps include:

[0097] 4.1) Based on the adaptive process noise covariance Define the normalized maneuver intensity index :

[0098] ,

[0099] in, It is the process noise covariance of the previous time step. It is the Frobenius norm. To prevent small positive numbers from degenerating into denominators, take ;

[0100] 4.2) Mapped to normalized maneuver strength via the Sigmoid function. :

[0101] ,

[0102] in, The slope coefficient, The center point of the mobility intensity threshold;

[0103] 4.3) Based on Calculate the target window length at the current time. :

[0104] ;

[0105] in, and These are the minimum window length and the maximum window length, respectively.

[0106] 4.4) Based on window length Utilizing the window start time Posterior state estimation and posterior covariance Calculate the arrival cost :

[0107] ,

[0108] in, It is the window start time The true state.

[0109] Step 5: Construct the penalty function With the rolling time-domain estimation objective function.

[0110] Hypersonic vehicles must strictly adhere to physical boundary conditions such as dynamic pressure, overload, and heat flux density during the gliding phase. By constructing a penalty function, these physical constraints can be transformed into additional penalty terms in the objective function. The aerodynamic parameters and states of hypersonic targets change drastically, making traditional single-step filtering insufficient to guarantee the stability of parameter estimates. By establishing a rolling time-domain estimation objective function and globally minimizing it, the system can dynamically balance historical priors, current measurements, and physical constraints within a finite sliding window, thereby calculating the most accurate and robust state vector, aerodynamic parameters, and roll angle estimates.

[0111] This step involves constructing a penalty function based on constraints related to the aircraft's dynamic pressure, overload, and heat flux density. These are the lower bounds of the height corresponding to the constraints of heat flux density, dynamic pressure, and overload, respectively. , and The penalty function is obtained by using a piecewise weighted approach. Combined with this penalty function Cost of arrival and adaptive process noise covariance The objective function for rolling time-domain estimation is established, and its implementation steps include:

[0112] 5.1) Construct a penalty function based on the constraints of dynamic pressure, overload, and heat flux density of the aircraft. These are the lower bounds of the height corresponding to the constraints of heat flux density, dynamic pressure, and overload, respectively. , and The penalty function is obtained by using a piecewise weighted approach. :

[0113] ,

[0114] in, , , As a penalty factor, for Flight altitude at any given moment for The speed of flight at any given moment;

[0115] 5.2) Combining this penalty function Cost of arrival and adaptive process noise covariance The objective function for rolling time-domain estimation is established as follows:

[0116] ,

[0117] in, It is the window start time The true state For the first The observation noise at each moment, For the first process noise at any given moment This is the observation noise matrix.

[0118] It should be noted that the step numbers in the specification and claims of this invention are only for the purpose of clearly describing the embodiments of this invention and facilitating understanding, and their order is not limited.

[0119] Example 2: Joint estimation system of state and characteristic parameters based on maneuver intensity perception.

[0120] Reference Figure 2This example includes: observation data receiving and innovation calculation module 1, adaptive noise and gain factor update module 2, state and covariance posterior update module 3, adaptive window and arrival cost construction module 4, and constrained MHE objective function construction and solution module 5. Among them, observation data receiving and innovation calculation module 1 includes: observation data receiving submodule 11, volume point generation submodule 12, and innovation calculation submodule 13.

[0121] The working principle of the entire system is as follows:

[0122] The observation data receiving and information calculation module 1 is used to receive discrete observation data and generate observation spatial volume points. The predicted state value is calculated based on the observed spatial volume point. Observational prediction values Kalman gain and new information vector Among them: the observation data receiving submodule 11 is used to receive discrete observation data and establish a system based on geocentric distance. ,longitude ,latitude ,speed Track inclination Track deflection The six-dimensional motion state vector represents the basic motion state, and the aerodynamic parameters to be estimated are... , and tilt angle The augmented state vector is then constructed into a nine-dimensional augmented state vector. The state equation and observation equation of the nonlinear discrete-time system are established, and the state equation and observation equation are provided to the volume point generation submodule 12 and the innovation calculation submodule 13, respectively. The volume point generation submodule 12 is used to calculate the posterior covariance matrix of the previous time step. Perform SVD decomposition to obtain its square root matrix. Based on this square root matrix, state space volume points are generated. The state-space volume point is propagated through the nonlinear state equation from the observation data receiving submodule 11 to obtain the predicted volume point at the current time. The predicted state value is calculated using this predicted volume point. and prediction error covariance The prediction error covariance is decomposed using SVD to obtain its square root matrix. Based on this square root matrix, the observation space volume points are generated. And the observed spatial volume points and state prediction values The output is sent to the innovation calculation submodule 13; this innovation calculation submodule 13 is used to propagate the observation spatial volume point to the observation equation from the observation data receiving submodule 11, and calculate the observation prediction value. Observation covariance State-observation cross-covariance , No. Kalman gain at time step Based on observed predicted values Calculate the innovation vector and the new information vector The output is sent to the adaptive noise and gain factor update module 2, which then updates the Kalman gain. The state and covariance posterior update module 3 is passed on.

[0123] The adaptive noise and gain factor update module 2 is used to update the noise and gain factor through the forgetting factor. process noise covariance Perform adaptive updates and utilize the innovation vector. Calculating the gain factor using the Frobenius norm The gain factor The output is given to the state and covariance posterior update module 3, and the adaptively updated process noise covariance is used as the input. The results are respectively output to the adaptive window and arrival cost construction module 4 and the constraint MHE objective function construction and solution module 5;

[0124] The state and covariance posterior update module 3 is used to utilize Kalman gain and gain factor Predicted state values Perform a state update to obtain a posterior state estimate. and posterior covariance matrix And estimate the posterior state. and posterior covariance matrix The output is sent to the adaptive window and arrival cost construction module 4, and then returned to the observation data receiving and information calculation module 1 for the next time step filtering recursion;

[0125] The adaptive window and arrival cost construction module 4 is used to update the adaptive process noise covariance transmitted by the adaptive noise and gain factor update module 2. Calculate the adaptive sliding window length Based on the length and state and covariance posterior update module 3, the corresponding window start time posterior state is updated. and posterior covariance , construct arrival cost and the cost of reaching it. The output is sent to module 5, which constructs and solves the constraint MHE objective function.

[0126] The constraint MHE objective function construction and solution module 5 is used to construct a penalty function based on the constraints of dynamic pressure, overload, and heat flux density of the aircraft. Combined with this penalty function Cost of arrival and adaptive process noise covariance Establish a rolling time-domain estimation objective function and solve it to calculate the state vector estimate.

[0127] It should be noted that the above functional modules can be implemented, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, they can be implemented, in whole or in part, as program instruction products. A program instruction product includes one or a set of program instructions. When the program instructions are loaded and executed on a computer, the described process or function is generated, in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The program instructions can be stored in a computer-readable and writable storage medium, or transferred from one computer's readable and writable storage medium to another.

[0128] In this embodiment, the direct coupling or communication connection between the modules can be achieved through indirect coupling or communication connection via interfaces, devices, or modules. The functional modules and sub-modules in this embodiment can dynamically reside within a single processing unit, or each module can exist physically independently, or two or more modules can dynamically reside within a single processing unit. When these dynamic components are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable and writable storage medium. This storage medium can be a memory, disk, or optical disc, etc.

[0129] The effectiveness of this invention can be further illustrated by the following simulation results:

[0130] I. Simulation Conditions

[0131] The hardware environment is as follows: GPU (RTX 4090, 24 GB VRAM), CPU (64 cores), and 90 GB RAM.

[0132] The software environment is: MATLAB R2023b

[0133] The target's trajectory is set to always remain within the detection range, and its own performance is constrained to be... , , The initial position of the aircraft is its initial longitude. initial latitude initial height initial velocity Using existing dynamic models to generate integrals such as Figure 3 The flight trajectory shown represents the maneuver scenario. Among them: Figure 3 a) A three-dimensional diagram of the flight trajectory; Figure 3 b) is a horizontal plane diagram of the flight trajectory; Figure 3 c) is the flight trajectory altitude-time diagram; the actual aerodynamic parameter curves corresponding to this maneuver scenario are as follows: Figure 4 As shown, where, Figure 4 a) are aerodynamic parameters True curve graph; Figure 4 b) are aerodynamic parameters The actual curve of the roll angle corresponding to this maneuver scenario is shown below. Figure 5 As shown; the process constraint curve corresponding to this maneuver scenario is as follows: Figure 6 As shown, its heat flux, overload, and dynamic pressure constraints are all less than its own performance constraints, indicating that the simulated trajectory meets the requirements of the actual trajectory.

[0134] II. Simulation Content

[0135] Simulation 1: Under the above simulation conditions, the present invention is used to estimate the parameters of the flight trajectory and calculate the gain factor. Response curve, results as follows Figure 7 .from Figure 7 As can be seen, when performing parameter estimation in this invention, the gain factor... The system exhibits significant adaptive dynamic adjustment characteristics based on the maneuvering state of the hypersonic vehicle. The response curve demonstrates that this invention effectively overcomes the filtering divergence problem caused by fixed observation weights in traditional filtering, significantly improving the system's response sensitivity to sudden maneuvering changes and the robustness of joint estimation.

[0136] Simulation 2: Under the above simulation conditions, the present invention is used to estimate the parameters of the flight trajectory, and the calculated window length is... The result is as follows Figure 8 .from Figure 8 As can be seen, when performing parameter estimation in this invention, the adaptive sliding window length... It can make precise dynamic adjustments based on the maneuver intensity characteristics of the aircraft. When the aircraft undergoes sudden maneuvering changes at times such as 100s, 200s, and 300s, the aerodynamic parameters change drastically. The system senses the increase in maneuver intensity through the rate of change of process noise covariance, effectively reducing the mismatch error caused by the constant assumption of the model during the strong maneuvering phase.

[0137] Simulation 3: Under the above simulation conditions, the flight trajectory parameters are estimated using the present invention and existing parameter estimation methods. The estimation results are then compared with... Figure 4 The results are as follows: The actual curves of the shown aerodynamic parameters are compared. Figure 9 .in:

[0138] Figure 9 a) Aerodynamic lift parameters estimated by the present invention and existing parameter estimation methods and Figure 4 a) Aerodynamic lift parameters A comparison chart of the actual curves; from Figure 9 a) As can be seen, the aerodynamic lift parameters are estimated during the parameter estimation process of this invention. Compared to existing parameter estimation methods, it more closely approximates aerodynamic lift parameters. Actual curve results.

[0139] Figure 9 b) Aerodynamic lift parameters estimated by the present invention and existing parameter estimation methods The comparison results of mean square error; from Figure 9 b) As can be seen, the aerodynamic lift parameters estimated by this invention The mean squared error is smaller than that of existing parameter estimation methods.

[0140] Figure 9 c) Aerodynamic drag parameters estimated by the present invention and existing parameter estimation methods and Figure 4 b) Aerodynamic drag parameters A comparison chart of the actual curves; from Figure 9 c) As can be seen, the aerodynamic drag parameters are estimated during the parameter estimation process of this invention. Compared to existing parameter estimation methods, it more closely approximates aerodynamic lift parameters. Actual curve results.

[0141] Figure 9 d) Aerodynamic drag parameters estimated by the present invention and existing parameter estimation methods. The comparison results of mean square error; from Figure 9 d) As can be seen, the aerodynamic drag parameters estimated by this invention... The mean squared error is smaller than that of existing parameter estimation methods.

[0142] Figure 9 e) The tilt angle estimated by the present invention and existing parameter estimation methods and Figure 5 The image shows a comparison of the actual tilt angle curves; from Figure 9 e) As can be seen, the tilt angle estimated by the present invention Compared to existing parameter estimation methods, it more closely approximates the true curve of the tilt angle.

[0143] Figure 9 f) is the tilt angle estimated by the present invention and existing parameter estimation methods. The comparison results of mean square error; from Figure 9d) As can be seen, the tilt angle estimated by this invention The mean squared error is smaller than that of existing parameter estimation methods.

[0144] Simulation results show that when performing parameter estimation in this invention, the gain factor... It can exhibit significant adaptive dynamic adjustment characteristics based on the maneuvering state of the hypersonic vehicle, while also adapting the sliding window length. It can make precise dynamic adjustments based on the maneuverability characteristics of the aircraft, and its performance indicators in parameter estimation are superior to existing parameter estimation methods.

Claims

1. A method for joint estimation of state and characteristic parameters based on maneuver intensity perception, characterized in that, include: (1) Receive discrete observation data and generate observation spatial volume points. The predicted state value is calculated based on the observed spatial volume point. Observational prediction values Kalman gain and new information vector ; (2) Through the forgetting factor process noise covariance Perform adaptive updates and utilize the innovation vector. Calculating the gain factor using the Frobenius norm ; (3) Utilizing Kalman gain Perform state updates to obtain posterior state estimates. and posterior covariance matrix Then return to (1) above to perform the next time step filtering recursion; (4) Based on the noise covariance of the adaptive process Calculate the adaptive sliding window length Based on this length, the posterior state at the start time of the corresponding window is used. and posterior covariance , construct arrival cost ; (5) Construct a penalty function based on the constraints of dynamic pressure, overload and heat flux density of the aircraft. Combined with this penalty function Cost of arrival and adaptive process noise covariance Establish a rolling time-domain estimation objective function and calculate the state vector estimate.

2. The method according to claim 1, characterized in that, In step (1), discrete observation data is received, and observation spatial volume points are generated. Its implementation includes: (1a) Establishing a distance from the Earth's center ,longitude ,latitude ,speed Track inclination Track deflection The six-dimensional motion state vector represents the basic motion state, and the aerodynamic parameters to be estimated are... , and tilt angle The augmented state vector is then constructed into a nine-dimensional augmented state vector. : , Indicates transpose; (1b) Establish the state equation and observation equation of the nonlinear discrete-time system: ； ； in, It is a nonlinear state transition function. For the observation function, For the first Process noise at any given moment For the first Observation noise at each moment; (1c) The posterior covariance matrix of the previous time step SVD decomposition is performed as follows: Obtain its square root matrix : ， in, and These are the orthogonal matrix and the diagonal matrix obtained by performing SVD decomposition on the posterior covariance matrix, respectively. (1d) Based on the square root matrix Generate state space volume points : ， in, , It is the dimension of the state vector. It is the first One standard volume point, It is the first of the identity matrix. List; (1e) Convert the state space volume point The predicted volume point at the current moment is obtained by propagation through a nonlinear state equation. : ； (1f) By predicting the volume point Calculate state prediction value and prediction error covariance : ； ； in, , It is transpose. It is the first The process noise covariance matrix at each time step; (1g) The prediction error covariance SVD decomposition is performed as follows: Obtain its square root matrix : ， (1h) Based on the square root matrix and state prediction value Generate observation space volume points : 。 3. The method according to claim 1, characterized in that, The state prediction value is calculated based on the observed spatial volume point in (1). Observational prediction values Kalman gain and new information vector ,include: (1i) Observe the spatial volume point Propagate to the observation equation and calculate the observed prediction value. Observation covariance State-observation cross-covariance , No. Kalman gain at time step : ； ； ； ； in, , It is transpose. It is the first The process noise covariance matrix at each time step; (1j) Based on observed predicted values Calculate the innovation vector : ， in, It is the first The actual observations at each moment.

4. The method according to claim 1, characterized in that, In (2), through the forgetting factor process noise Covariance Perform adaptive updates and utilize the innovation vector. Calculating the gain factor using the Frobenius norm Its implementation includes: (2a) Introducing a forgetting factor Calculate the dynamic weighting coefficients The process noise covariance is calculated based on the Sage-Husa adaptive estimation principle. Adaptive updates: ； ； in, It is the first The information vector at each moment, It is the first Kalman gain at each time step; (2b) Calculate the gain factor using the continuous Sigmoid function : ， in, Based on the bias, The amplitude coefficient, It is a continuous sigmoid function. The slope coefficient, for The information threshold point where significant changes occur.

5. The method according to claim 1, characterized in that, The use of Kalman gain in (3) Perform a state update to obtain a posterior state estimate. and posterior covariance matrix The formulas are as follows: ， ， in, It is a state prediction value. It is the first The actual observations at each moment These are observed and predicted values. Gain factor It is the prediction error covariance. It is the observation covariance.

6. The method according to claim 1, characterized in that, In the above (4), based on the adaptive process noise coordination variance Calculate the adaptive sliding window length , construct arrival cost Its implementation includes: (4a) Define the normalized maneuver intensity index : ， in, It is the process noise covariance of the previous time step. It is the Frobenius norm. To prevent small positive numbers from degenerating into denominators, take ; (4b) will Mapped to normalized maneuver strength via the Sigmoid function. : ； in, The slope coefficient, The center point of the mobility intensity threshold; (4c) Based on Calculate the target window length at the current time. : ； in, and These are the minimum window length and the maximum window length, respectively. (4d) Based on window length Utilizing the window start time Posterior state estimation and posterior covariance Calculate the arrival cost : ， in, It is the window start time The true state.

7. The method according to claim 1, characterized in that: In (5), based on the dynamic pressure and overload of the aircraft Constructing a penalty function based on heat flux density constraints These are the lower bounds of the height corresponding to the constraints of heat flux density, dynamic pressure, and overload, respectively. , and The penalty function is obtained by using a piecewise weighted approach. : ； in, , , As a penalty factor, for Flight altitude at any given moment for The speed of flight at any given moment.

8. The method according to claim 1, characterized in that: The penalty function mentioned in (5) Cost of arrival and adaptive process noise covariance The objective function for rolling time-domain estimation is established as follows: ； in, It is the window start time The true state For the first The observation noise at each moment, For the first Process noise at any given moment This is the observation noise matrix.

9. A joint estimation system for state and characteristic parameters based on maneuver intensity perception, characterized in that, include: The observation data receiving and information calculation module is used to receive discrete observation data and generate observation spatial volume points. The predicted state value is calculated based on the observed spatial volume point. Observational prediction values Kalman gain and new information vector ; The adaptive noise and gain factor update module is used to update noise and gain factors. process noise covariance Perform adaptive updates and utilize the innovation vector. Calculating the gain factor using the Frobenius norm ; The state and covariance posterior update module is used to utilize Kalman gain. State prediction values Perform state updates to obtain posterior state estimates. and posterior covariance matrix Then it is returned to the observation data receiving and information calculation module for filtering and recursion at the next time step; Adaptive window and arrival cost building blocks are used to build upon the adaptive process noise covariance. Calculate the adaptive sliding window length Based on this length, the posterior state at the start time of the corresponding window is used. and posterior covariance , construct arrival cost ; The constraint MHE objective function construction and solution module is used to construct penalty functions based on the constraints of dynamic pressure, overload, and heat flux density of the aircraft. Combined with this penalty function Cost of arrival and adaptive process noise covariance Establish a rolling time-domain estimation objective function and calculate the state vector estimate.

10. The system according to claim 9, characterized in that, The observation data receiving and information calculation module includes: The observation data receiving submodule is used to receive discrete observation data and establish a data structure based on geocentric distance. ,longitude ,latitude ,speed Track inclination Track deflection The six-dimensional motion state vector represents the basic motion state, and the aerodynamic parameters to be estimated are... , and tilt angle The augmented state vector is then constructed into a nine-dimensional augmented state vector. Establish the state equation and observation equation for the nonlinear discrete-time system; The volume point generation submodule is used to generate the posterior covariance matrix from the previous time step. Perform SVD decomposition to obtain its square root matrix. Based on this square root matrix, state space volume points are generated. The state-space volume point is propagated through a nonlinear state equation to obtain the predicted volume point at the current time. The predicted state value is calculated using this predicted volume point. and prediction error covariance The prediction error covariance is decomposed using SVD to obtain its square root matrix. Based on this square root matrix, the observation space volume points are generated. ; The innovation calculation submodule: calculates the spatial volume points of the observation space. Propagate to the observation equation and calculate the observed prediction value. Observation covariance State-observation cross-covariance , No. Kalman gain at time step Based on observed predicted values Calculate the innovation vector .

Images