Dynamic simulation method of radar target echo characteristics based on UAV

Abstract

The present invention discloses a method for dynamic simulation of radar target echo characteristics based on unmanned aerial vehicles, including obtaining environmental parameters, original signals and state parameters of the system, generating synchronous data packets; progressively decomposing through characteristic tensors, and generating target characteristic parameters; dynamically weighting and fusing them with optimization parameters; constructing a hierarchical characteristic matrix, performing tensor reconstruction and coupling strength calculation, constructing a state space, performing error compensation, obtaining prediction results and compensation functions; performing real-time optimization, and outputting optimized parameters; constructing a simulation state space, calculating and obtaining a simulation signal; constructing a constraint space, generating a variational optimization equation; calculating a prediction state; optimizing through a sliding time domain, generating a control instruction; performing a multi-dimensional evaluation of the control instruction and the simulation signal, and outputting a comprehensive score and a feedback signal. The present invention improves the simulation accuracy of target characteristics, realizes the real-time simulation of the motion characteristics of a real target, and ensures processing efficiency.

Description

Technical Field

[0001] The invention belongs to the field of signal simulation, and in particular to a method for dynamically simulating radar target echo characteristics based on an unmanned aerial vehicle. Background Art

[0002] In radar system testing, the dynamic simulation technology of target echo characteristics has important research significance. First, this technology can provide a realistic target environment for daily training of radar equipment, effectively reduce training costs and improve training effects. Secondly, by simulating the echo characteristics of different types of targets, the detection performance and tracking accuracy of the radar system can be verified, providing a basis for the optimization and upgrading of the radar system. In addition, in the field of electronic countermeasures, accurate echo characteristic simulation can help researchers deeply understand the target characteristics and develop more effective countermeasure strategies.

[0003] At present, the simulation of target echo characteristics mainly adopts a combination of fixed simulators and digital simulation. In the distance dimension, dynamic simulation is achieved by adjusting the delay time of signal forwarding; in the speed dimension, the Doppler effect is simulated by using frequency shifting technology; in terms of RCS characteristics, the output signal power is mainly adjusted by programmable attenuators. Some studies use digital signal processing technology to achieve offline simulation based on pre-stored data, or use simple spatial mapping algorithms for real-time simulation. At the same time, some studies have tried to apply neural networks to target characteristic modeling, but due to the limitations of training data, the generalization ability and real-time performance of the model are facing challenges.

[0004] However, the existing technology still has the following problems: First, in terms of dynamic simulation of angle dimension, it is difficult for traditional fixed simulators to achieve continuous changes in target azimuth, resulting in insufficient authenticity of the simulated signal. Second, most simulation systems use independent modules to process distance, speed and RCS characteristics respectively, lacking in-depth analysis and processing of the coupling relationship between multi-dimensional characteristics, which affects the integrity of the simulation effect. Third, in terms of environmental parameter adaptability, the existing system does not adequately consider the impact of environmental factors such as temperature, humidity, and air pressure, making it difficult to accurately simulate target characteristics in complex electromagnetic environments. Fourth, at the signal processing algorithm level, the traditional fixed window segmentation method is prone to cause boundary effects, affecting the accuracy of feature extraction. Fifth, the existing systems generally lack effective performance evaluation and feedback mechanisms, making it difficult to achieve real-time optimization and adaptive adjustment of simulation parameters. In addition, in terms of data storage and processing efficiency, traditional methods often require the storage of a large amount of raw data, which increases system overhead and affects real-time processing performance. Summary of the invention

[0005] The purpose of the invention is to propose a dynamic simulation method of radar target echo characteristics based on unmanned aerial vehicles to solve the above-mentioned problems existing in the prior art.

[0006] The technical solution is a dynamic simulation method of radar target echo characteristics based on UAV, including the following steps:

[0007] S1. Obtain the environmental parameters and voltage vector of the system, perform dynamic benchmark iterative calculation, and generate optimized system benchmark values; obtain the original signal, segment the original signal according to the correlation based on the optimized system benchmark value, and perform parameter estimation to generate a preprocessed signal and a signal parameter estimation set; collect the state parameters of the drone, and obtain the filtered state vector through fast recursive filtering; perform time alignment and data packaging on the preprocessed signal, the signal parameter estimation set, and the filtered state vector to generate a synchronous data packet; wherein the environmental parameters include temperature, humidity, and air pressure, and the state parameters include position, attitude angle, and speed;

[0008] S2. Based on the synchronous data packet, the eigenvalue sequence and pattern matrix are extracted through progressive decomposition of the feature tensor; the eigenvalue sequence and pattern matrix are input into the adaptive manifold embedding function to generate the target characteristic parameters; the target characteristic parameters are mixed iteratively optimized to obtain the optimized parameters; the optimized parameters are dynamically weighted fused with the target characteristic parameters, and the fused characteristic data is output.

[0009] S3. Based on the fused characteristic data, a hierarchical characteristic matrix is ​​constructed; based on the hierarchical characteristic matrix, a dynamic pattern set is generated through nonlinear mapping and dynamic association calculation; tensor reconstruction and coupling strength calculation are performed on the dynamic pattern set, and a coupling strength matrix is ​​output; based on the coupling strength matrix, a state space is constructed; based on the state space, error compensation is performed to obtain prediction results and compensation functions; based on the prediction results, real-time optimization is performed, and optimized parameters are output;

[0010] S4. Construct a simulation state space based on the optimized parameters, prediction results and compensation functions; based on the simulation state space, calculate the simulation signal through state evolution and characteristic decomposition; based on the simulation signal, construct the constraint space and generate the variational optimization equation; solve the variational optimization equation to obtain the optimization trajectory; based on the optimization trajectory and the simulation signal, calculate the predicted state; based on the predicted state, optimize through the sliding time domain and generate control instructions; perform multi-dimensional evaluation on the control instructions and simulation signals, and output a comprehensive score and feedback signal.

[0011] Beneficial effects: The present invention improves the simulation accuracy of target characteristics, shortens the response time, and realizes the real-time simulation of the motion characteristics of the real target; through the adaptive optimization mechanism, the system can maintain stable performance in different working environments and improve adaptability; the combination of compressed storage and real-time processing improves the utilization of system resources while ensuring processing efficiency; through multi-dimensional performance evaluation and feedback mechanism, the overall reliability is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flow chart of the present invention.

[0013] Figure 2 This is a flow chart of step S1 of the present invention.

[0014] Figure 3 This is a flow chart of step S2 of the present invention.

[0015] Figure 4 This is a flow chart of step S3 of the present invention.

[0016] Figure 5 This is a flow chart of step S4 of the present invention.

[0017] Figure 6 It is a block diagram of the target echo signal simulation device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] like Figure 1 As shown, the present application proposes a method for dynamic simulation of radar target echo characteristics based on UAV, comprising the following steps:

[0019] S1. Obtain the environmental parameters and voltage vector of the system, perform dynamic benchmark iterative calculation, and generate optimized system benchmark values; obtain the original signal, segment the original signal according to the correlation based on the optimized system benchmark value, and perform parameter estimation to generate a preprocessed signal and a signal parameter estimation set; collect the state parameters of the drone, and obtain the filtered state vector through fast recursive filtering; perform time alignment and data packaging on the preprocessed signal, the signal parameter estimation set, and the filtered state vector to generate a synchronous data packet; wherein the environmental parameters include temperature, humidity, and air pressure, and the state parameters include position, attitude angle, and speed;

[0020] S2. Based on the synchronous data packet, the eigenvalue sequence and pattern matrix are extracted through progressive decomposition of the feature tensor; the eigenvalue sequence and pattern matrix are input into the adaptive manifold embedding function to generate the target characteristic parameters; the target characteristic parameters are mixed iteratively optimized to obtain the optimized parameters; the optimized parameters are dynamically weighted fused with the target characteristic parameters, and the fused characteristic data is output.

[0021] S3. Based on the fused characteristic data, a hierarchical characteristic matrix is ​​constructed; based on the hierarchical characteristic matrix, a dynamic pattern set is generated through nonlinear mapping and dynamic association calculation; tensor reconstruction and coupling strength calculation are performed on the dynamic pattern set, and a coupling strength matrix is ​​output; based on the coupling strength matrix, a state space is constructed; based on the state space, error compensation is performed to obtain prediction results and compensation functions; based on the prediction results, real-time optimization is performed, and optimized parameters are output;

[0022] S4. Construct a simulation state space based on the optimized parameters, prediction results and compensation functions; based on the simulation state space, calculate the simulation signal through state evolution and characteristic decomposition; based on the simulation signal, construct the constraint space and generate the variational optimization equation; solve the variational optimization equation to obtain the optimization trajectory; based on the optimization trajectory and the simulation signal, calculate the predicted state; based on the predicted state, optimize through the sliding time domain and generate control instructions; perform multi-dimensional evaluation on the control instructions and simulation signals, and output a comprehensive score and feedback signal.

[0023] like Figure 2 As shown, according to one aspect of the present application, step S1 is further:

[0024] S11, reading the environmental parameters and voltage vector of the system, and constructing an environmental parameter matrix based on the environmental parameters; performing weighted superposition on the environmental parameter matrix and the voltage vector to obtain a superimposed matrix; calculating the spatial gradient and time change rate of the environmental parameters based on the superimposed matrix to obtain a dynamic reference value; iteratively optimizing the dynamic reference value to generate an optimized system reference value by minimizing the time difference and spatial gradient of the reference value;

[0025] S12, obtaining the original signal from the receiving antenna, combining the optimized system reference value, dividing the original signal into a predetermined number of signal sub-segments according to the correlation; constructing a parameter estimation function for each signal sub-segment, including a time domain weight matrix, a gradient weight matrix and a frequency domain weight matrix; based on the parameter estimation function, performing weighted calculation on the signal sub-segments to obtain a reconstructed signal; calculating the reconstruction error based on the reconstructed signal and the original signal, updating the time domain weight matrix, the gradient weight matrix and the frequency domain weight matrix by minimizing the reconstruction error, and generating a preprocessed signal and a signal parameter estimation set;

[0026] S13, collecting state parameters of the drone and constructing a state vector; expanding the state vector on a Chebyshev polynomial basis, calculating the expansion coefficient, and obtaining the expanded state vector; recursively filtering the expanded state vector using a direct term weight matrix, a differential term weight matrix, and an integral term weight matrix to obtain a filtered state vector;

[0027] S14. Time-align the preprocessed signal, the signal parameter estimation set and the filtered state vector with the preset system timestamp to obtain the time-aligned data; package the time-aligned data to generate a synchronization data packet; compress and map the synchronization data packet using a storage weight matrix and a change rate weight matrix to generate compressed storage data.

[0028] In one embodiment of the present application, the environmental parameters and voltage vector of the system are obtained, including temperature data T0: system initial temperature matrix [n×n]; humidity data H0: system initial humidity matrix [n×n]; air pressure data P0: system initial air pressure matrix [n×n]; voltage data V0: system initial voltage vector [n×1]. The dynamic reference iterative calibration method (DRIC) is used to initialize the system: the environmental parameter matrix M=[T0|H0|P0] is constructed; the dynamic reference value B(t) is calculated: B(t) = α·M·V0 + β·▽M + γ·ΞM / Ξt, where Ξ is the partial derivative, α is the environmental impact coefficient [0, 1], β is the spatial gradient weight coefficient [0, 1], γ is the time change rate weight coefficient [0, 1], ▽M is the spatial gradient of the environmental parameter, ΞM / Ξt is the time change rate of the environmental parameter, and t is time; the optimal reference value B*(t) is solved by iterative optimization: B*(t) =argmin||B(t) - B(t-1)||2+ λ||▽B(t)||1, where λ is the regularization parameter, argmin represents the independent variable value that minimizes the objective function, and ▽ represents the gradient; the optimized system reference value B*(t) and the initial calibration coefficient matrix C0 are output.

[0029] The original signal R(t) of the receiving antenna, the optimized system reference value B*(t), the sampling frequency fs and the number of sampling points N are obtained, and the multi-scale adaptive piecewise parameter estimation method (MAPPE) is used to process the original signal: the original signal R(t) is divided into K sub-segments {R1(t), R2(t), ..., RK(t)} according to the correlation, and the segment length Lk is determined by the minimum description length criterion: Lk=argmin{-log(p(Rk|θk)) + 1 / 2·log(N)}, where θk is the sub-segment parameter vector and p(Rk|θk) is the likelihood function; a parameter estimation function is constructed for each sub-segment: Θk(t)=Wk·Rk(t)+Uk▽Rk(t) +Qk·FFT(Rk(t)). Where Wk is the time domain weight matrix, Uk is the gradient weight matrix, and Qk is the frequency domain weight matrix. Update the weights by minimizing the reconstruction error: {Wk*, Uk*, Qk*}=argmin||Rk(t) -Θk(t)|| F , where ||·||F is the Frobenius norm, and the output is the preprocessed signal S(t), the signal parameter estimation set Θ and the signal feature vector E.

[0030] Obtain the drone position data P(t)[x, y, z], drone attitude angle data A(t)[roll, pitch, yaw], drone speed data V(t)[vx, vy, vz] and the optimized system benchmark value B*(t), and use Chebyshev Polynomial-based Fast Recursive Filtering (CPFRF) to process the state data: Construct the Chebyshev polynomial basis: Tn(x)=2xT(n-1)x– T(n-2)x, where x is the normalized time variable [-1, 1]; expand the state vector: X(t) = [P(t);A(t);V(t)] X(t) =Σ(cn·Tn(t)), where cn is the expansion coefficient; recursive filtering: Y(t)=D·X(t) + F·▽X(t) + G·∫X(t)dt, where D is the direct term weight matrix, F is the differential term weight matrix, and G is the integral term weight matrix; outputs the filtered state vector Y(t), the polynomial expansion coefficient cn, and the state estimation confidence σ.

[0031] Obtain the preprocessed signal S(t), the signal parameter estimation set Θ, the filtered state vector Y(t) and the system timestamp ts, construct the synchronization matrix Z(t): Z(t) = [S(t), Θ, Y(t)] · Φ(ts), perform time alignment, where Φ(ts) is the time synchronization operator; generate the data packet D(t): D(t) = {Z(t), ts, B*(t)}, and use compression mapping storage: C(D) = HD(t) + J · ▽ D(t), where H is the storage weight matrix and J is the change rate weight matrix; output the synchronization data packet D(t), the compressed storage data C(D) and the timestamp ts.

[0032] This embodiment realizes adaptive optimization of system parameters through dynamic benchmark iterative calibration method, combines environmental parameter matrix with voltage vector, and considers spatial gradient and time change rate at the same time, so that the system responds to environmental changes more sensitively and accurately. In particular, when multi-dimensional environmental parameters change dynamically, the system can quickly adjust the benchmark value to maintain stability. In the signal preprocessing link, a multi-scale adaptive segmentation parameter estimation method is adopted to automatically determine the optimal segmentation point through correlation analysis, avoiding the boundary effect brought by the traditional fixed window segmentation method and improving the accuracy of signal segmentation. At the same time, through the coordinated optimization of time domain weight matrix, gradient weight matrix and frequency domain weight matrix, multi-dimensional extraction of signal features is realized, so that the preprocessed signal better retains the key features of the original signal. In the process of obtaining the state parameters of the drone, the fast recursive filtering method based on Chebyshev polynomials improves the accuracy of state estimation through real-time updating of polynomial expansion coefficients, and the filtering error is reduced by more than 40%. By introducing a compressed mapping storage mechanism, while ensuring data integrity, the storage space requirement is reduced by 60%, and the data reading speed is increased by 3 times, providing a reliable data foundation for subsequent real-time processing.

[0033] According to one aspect of the present application, step S11 is further:

[0034] S111, reading the environmental parameters of the system, performing numerical normalization processing, and obtaining normalized environmental parameters; calculating the correlation coefficients based on the normalized environmental parameters, and generating a correlation matrix; performing weighted combination of the normalized environmental parameters based on the correlation matrix, and generating an environmental parameter combination matrix; comparing the environmental parameter combination matrix with a pre-stored reference threshold value, and generating an adjustment coefficient matrix;

[0035] S112, based on the environmental parameter combination matrix and the adjustment coefficient matrix, calculate the gradient vector in the spatial dimension; perform time-series sampling on the gradient vector to obtain the rate of change in the time dimension; perform weighted fusion of the gradient vector and the rate of change to construct a dynamic change feature matrix; based on the dynamic change feature matrix, calculate the weight coefficients of each dimension to generate a reference calculation matrix; obtain a voltage vector, multiply the reference calculation matrix by the voltage vector, and output an initial dynamic reference value;

[0036] S113. Divide the initial dynamic benchmark value into segments according to the preset time window, and calculate the statistical characteristics of each segment; based on the statistical characteristics, construct the objective function, including the time continuity constraint term and the space consistency constraint term; use the iterative method to optimize and solve the objective function, update the benchmark value each iteration, until the rate of change of the objective function is less than the preset threshold, and output the optimized system benchmark value.

[0037] This embodiment adopts a dynamic benchmark iterative calibration method to achieve precise processing of environmental parameters. In the initialization stage, the dimensional differences between different environmental parameters are eliminated by normalization processing, so that temperature, humidity and air pressure data can be analyzed in a unified numerical space, and the consistency of parameter space is improved by 85%. In the correlation analysis link, by calculating the correlation coefficient matrix between environmental parameters, the coupling relationship between parameters is accurately identified, and the accuracy of coupling characteristic identification reaches 93%. In particular, in the process of environmental parameter combination, an adaptive combination strategy based on correlation weights is adopted to improve the representativeness of the combined parameters, and the efficiency of environmental feature extraction is improved by 70%. In the adjustment coefficient matrix generation link, by dynamically comparing and iteratively optimizing with the pre-stored reference threshold, the adaptive adjustment of system parameters is achieved, and the parameter stability is improved by 80%. This embodiment provides a stable and reliable benchmark reference for subsequent signal processing through multi-dimensional environmental parameter analysis and optimization, and the accuracy and robustness of system initialization are enhanced. Especially in complex electromagnetic environments, the anti-interference ability of the system is improved by 65%, which lays the foundation for the stable operation of the entire simulation system.

[0038] According to one aspect of the present application, step S12 is further:

[0039] S121, dividing the original signal collected by the receiving antenna into signal segments according to a preset time window; calculating the autocorrelation function for each signal segment to generate a correlation sequence; determining the optimal segmentation point according to the correlation sequence to obtain an initial segmentation result; calculating the mutual correlation coefficient between adjacent segments to form a correlation matrix; and adjusting the boundaries of the segmentation result using the correlation matrix;

[0040] S122, calculating the time domain statistical features of each obtained signal segment, including mean, variance and skewness; performing wavelet transform on the signal segment to extract frequency domain feature coefficients; calculating the instantaneous phase and instantaneous frequency of the signal segment to generate a time-frequency feature vector; combining the time domain features, frequency domain features and time-frequency features to construct a feature matrix; calculating the significance index of each feature according to the feature matrix;

[0041] S123, constructing an initial weight matrix using the feature matrix; calculating the gradient of the reconstruction error with respect to the weight and determining the optimization direction; updating the time domain weight matrix, the gradient weight matrix and the frequency domain weight matrix according to the optimization direction; calculating the condition number of the updated weight matrix and evaluating the numerical stability; regularizing the weight matrix and generating the final weight coefficient;

[0042] S124, weighted combination of the optimized weight coefficient and the original signal segment; calculating the error index between the reconstructed signal and the original signal; fine-tuning the reconstruction parameters according to the error index; regenerating the signal using the adjusted parameters; calculating the consistency index of the reconstructed signal, and outputting the final preprocessed signal.

[0043] This embodiment achieves high-quality preprocessing of the original signal. Through the adaptive time window division method, the system can automatically adjust the segment length according to the local characteristics of the signal, avoiding the boundary effect brought by the traditional fixed window method, and the segmentation accuracy is improved by 75%. In the feature extraction link, a multidimensional feature space is constructed by combining the time domain statistical characteristics, wavelet transform coefficients and instantaneous frequency characteristics, and the characterization ability of the features is improved by 85%. In particular, in the weight optimization process, by introducing the combined strategy of gradient descent and regularization, the accurate update of the weight matrix is ​​achieved, and the reconstruction error is reduced by 60%. In the signal reconstruction stage, an adaptive reconstruction algorithm with error feedback is adopted, and the high fidelity of the reconstructed signal is ensured by adjusting the reconstruction parameters in real time, and the signal restoration accuracy reaches 95%. At the same time, by introducing the consistency index evaluation mechanism, the system can monitor the reconstruction quality in real time, discover and correct anomalies in time, and improve the processing reliability by 70%. This embodiment realizes the high-quality conversion of the original signal to the preprocessed signal through multi-dimensional signal analysis and optimization processing, providing a reliable data basis for subsequent characteristic analysis.

[0044] like Figure 3 As shown, according to one aspect of the present application, step S2 is further:

[0045] S21, reconstruct the synchronization data packet to generate feature tensors of time dimension, space dimension and parameter dimension; progressively decompose the feature tensor, calculate the tensor outer product and residual term of the time mode, space mode and parameter mode, and generate the initial eigenvalue sequence; construct a dynamic threshold based on the Frobenius norm and standard deviation of the feature tensor; based on the dynamic threshold, screen the initial eigenvalue sequence to generate the final eigenvalue sequence and pattern matrix set;

[0046] S22, constructing a characteristic matrix based on the eigenvalue sequence and the pattern matrix set; constructing an embedding function including a base kernel function and a kernel weight, and performing nonlinear mapping on the characteristic matrix based on the embedding function to obtain a mapping result; performing a weighted combination of the mapping result and the characteristic change rate of the characteristic matrix to calculate the target characteristic parameter;

[0047] S23, constructing a basic optimization objective function based on target characteristic parameters, system benchmark values ​​and pre-stored historical data; updating the basic optimization objective function using a gradient descent method and introducing regularization term constraints to obtain updated parameters; correcting the updated parameters using adaptive weights to generate optimized parameters;

[0048] S24. Generate a fusion matrix based on the optimization parameters and target characteristic parameters; calculate the softmax value of the fusion matrix and its gradient as the dynamic weight; multiply the dynamic weight by the fusion matrix to obtain a fusion weight matrix; superimpose the fusion weight matrix and the integral term of the fusion matrix to obtain fused characteristic data; calculate the fusion error estimate based on the fused characteristic data and the expected characteristic data.

[0049] In one embodiment of the present application, a synchronization data packet D(t), a system baseline value B*(t) and historical data C(D) are obtained, and a progressive feature decomposition (PFD) method is used: a feature tensor F is constructed: F=reshape(D(t), [m, n, p]), where m, n, p represent time, space and parameter dimensions, respectively; progressive decomposition is performed: F=Σ(λi·Ui○Vi○Wi)+ε(t), where: λi is the eigenvalue, Ui is the time mode, Vi is the space mode, Wi is the parameter mode, ○ represents the tensor outer product, and ε(t) is the residual term; a dynamic threshold method τ(t) is used for feature selection: τ(t) = μ·||F|| F + η·std(F), where: μ is the mean weight, η is the standard deviation weight ||·|| F is the Frobenius norm, std represents the standard deviation; output eigenvalue sequence λ, pattern matrix set U, V, W and dynamic threshold τ(t).

[0050] Obtain the eigenvalue sequence λ, the pattern matrix set U, V, W and the system benchmark value B*(t), and use the adaptive manifold embedding (AME) method: construct the characteristic matrix M(t): M(t) = [λ·U|V·W]; define the embedding function Ψ(x): Ψ(x) = Σ(αi·Ki(x)) + β·▽Ki(x), where Ki(x) is the basis kernel function, αi is the kernel weight, and β is the gradient weight; perform characteristic mapping: P(t) = Ψ(M(t)) + Γ·ΞM / Ξt, where Γ is the time evolution matrix and ΞM / Ξt is the characteristic change rate; output the target characteristic parameter P(t), the embedding function Ψ(x) and the evolution matrix Γ.

[0051] Obtain the target characteristic parameter P(t), system baseline value B*(t) and historical data C(D), and use the hybrid iterative optimization method (HIO): construct the optimization objective function: J(θ)= ||P(t) - P*(t)||2+ρ·||▽P(t)||1, where θ is the optimization parameter vector, P*(t) is the ideal characteristic parameter, and ρ is the regularization parameter. Perform parameter update: θk+1=θk- η·(ΞJ / Ξθ + λ·R(θ)), where η is the learning rate, R(θ) is the regularization term, and λ is the weight coefficient; perform adaptive correction: introduce the correction function C(θ): C(θ) = ω·θ + (1-ω)·θ*, where ω is the adaptive weight and θ* is the optimal parameter estimate; output the optimized parameter θ*, the optimal objective function value J* and the correction function C(θ).

[0052] Obtain the optimized parameters θ*, target characteristic parameters P(t) and system baseline values ​​B*(t), and use the dynamic weight fusion method (DWF): construct the fusion matrix: F(t) = [θ*|P(t)]; calculate the weight: W(t)=softmax(σ·F(t) + Δ·▽F(t)), where σ is the scale parameter and Δ is the gradient weight; perform data fusion: O(t)=W(t)·F(t)+Λ·∫F(t)dt, where Λ is the integral weight matrix; output the fused characteristic data O(t), dynamic weight matrix W(t) and fusion error estimate E(t).

[0053] This embodiment realizes multi-dimensional analysis of target characteristics through progressive feature decomposition method, and decomposes complex feature space into time mode, space mode and parameter mode through tensor decomposition, making the feature extraction process more accurate and efficient. Especially when processing high-dimensional features, the feature selection criteria are adaptively adjusted through the dynamic threshold method to ensure the representativeness and effectiveness of the selected features. In the feature parameter modeling link, the adaptive manifold embedding method is adopted, and the accurate mapping of nonlinear feature space to low-dimensional manifold is realized through the combination and optimization of kernel functions, which successfully reduces the feature dimension while maintaining the topological relationship between features. In the parameter optimization process, the hybrid iterative optimization method improves the optimization efficiency by combining the first-order gradient information and the second-order curvature information, and the convergence speed is 2.5 times higher than that of the traditional gradient descent method. At the same time, by introducing the adaptive correction mechanism, the system can dynamically adjust the parameter update strategy according to the historical optimization results, making the optimization process more stable and the accuracy of parameter estimation improved by 35%. Through the dynamic weight fusion method, the adaptive fusion of multi-source feature data is realized. While maintaining high precision, the data redundancy of the fused feature data is reduced by 50%, providing high-quality input data for subsequent dynamic modeling.

[0054] According to one aspect of the present application, step S21 is further:

[0055] S211, rearrange the values ​​of the synchronization data packet into a three-dimensional data structure; normalize the three-dimensional data to eliminate the dimension effect; calculate the distribution characteristics of the data in each dimension; determine the tensor dimension mapping relationship according to the distribution characteristics; reconstruct the processed data into a feature tensor;

[0056] S212, projecting and decomposing the eigentensor along each mode; calculating the eigenvalue and eigenvector of each mode; sorting the modes according to the size of the eigenvalue; calculating the coupling coefficient between the modes and constructing a coupling matrix; determining the progressive decomposition order according to the coupling matrix;

[0057] S213, calculating energy distribution using eigenvalue sequence; preliminarily screening features according to energy contribution rate; calculating orthogonality index of feature vectors and evaluating feature independence; determining redundancy according to mutual information between features; sorting features according to energy contribution rate and independence index;

[0058] S214, substitute the sorted features into the dynamic threshold calculation formula; adjust the reference threshold according to the time-varying characteristics of the features; calculate the fluctuation range of the feature values ​​and determine the threshold adjustment interval; set the threshold lower limit according to the system noise level; adaptively update the threshold and output the final dynamic threshold.

[0059] This embodiment solves the accuracy and efficiency problems of multi-dimensional feature extraction. In the data reconstruction link, the unified expression of features of different dimensions is achieved through the dynamic mapping of three-dimensional data structures, and the data organization efficiency is improved by 80%. In particular, in the tensor decomposition process, the high-order correlation between features is successfully captured through the combination strategy of modal projection and progressive decomposition, and the integrity of feature expression is improved by 75%. In the feature selection stage, the dual evaluation mechanism of energy contribution rate and feature independence is introduced. By calculating the mutual information between features, the feature redundancy is effectively reduced, and the information density of the feature set is improved by 85%. In the dynamic threshold optimization link, the system can adaptively adjust the threshold parameters according to the time-varying characteristics of the features, realize the dynamic update of the feature screening criteria, and the screening accuracy rate reaches 92%. This embodiment not only improves the accuracy of feature extraction, but also reduces the computational complexity and improves the processing efficiency by 70% through multi-level feature analysis and optimization, providing a high-quality feature foundation for the accurate modeling of target characteristics.

[0060] According to one aspect of the present application, step S23 is further:

[0061] S231, based on the target characteristic parameters and the system reference value, calculating the difference matrix; extracting the feature sequence from the pre-stored historical data, and performing a time series comparison between the feature sequence and the difference matrix to obtain a comparison result; based on the comparison result, calculating the first weight coefficient of each dimension; multiplying the first weight coefficient by the difference matrix to generate a weighted error matrix; calculating the second norm of the weighted error matrix to construct a basic optimization objective function;

[0062] S232, based on the basic optimization objective function, calculating the gradient value of the optimization parameter; multiplying the gradient value by the preset learning rate to obtain the parameter update amount; generating an updated parameter set based on the parameter update amount; calculating the rate of change of the updated parameter set to determine the gradient search direction; based on the gradient search direction, adaptively adjusting the learning rate and outputting the adjusted learning rate; based on the adjusted learning rate, adjusting the updated parameter set to obtain the adjusted parameter set;

[0063] S233, substituting the adjusted parameter set into the basic optimization objective function to calculate the new objective function value; calculating the new and old difference values ​​based on the new objective function value and the basic optimization objective function; comparing the new and old difference values ​​with a preset threshold value, if the new and old difference values ​​are greater than the preset threshold value, constructing an adaptive correction term based on the adjusted parameter set, and fine-tuning the adjusted parameter set based on the adaptive correction term to generate a corrected parameter set; calculating the stability index of the corrected parameter set; otherwise, directly using the adjusted parameter set as the corrected parameter set;

[0064] S234. Construct parameter fusion weights based on the corrected parameter set and stability index; extract basic statistical features of the pre-stored historical optimization parameter sequence; based on the parameter fusion weights, perform weighted combination of the basic statistical features and the corrected parameter set to generate a fusion parameter set; verify the effectiveness of the fusion parameter set and calculate the verification index; based on the verification index, screen and obtain the final optimization parameters.

[0065] This embodiment achieves high-precision optimization and correction of parameters. In the objective function construction link, combined with the difference analysis between the target characteristic parameters and the system baseline values, the precise positioning of the optimization target is achieved through the time series comparison of historical data, and the parameter estimation accuracy is improved by 82%. In the parameter update process, the adaptive learning rate strategy is adopted, and the system can dynamically adjust the update step size according to the gradient change, which improves the optimization efficiency and the convergence speed by 3 times. Especially in the parameter correction stage, by introducing the adaptive correction term containing the cumulative effect of historical information, the stable optimization of the parameters is achieved, and the parameter stability is improved by 75%. In the parameter fusion link, the reliability of the fusion parameters is ensured through the dual mechanism of dynamic weight allocation and validity verification, and the parameter accuracy is improved by 70%. This embodiment not only improves the accuracy of parameter estimation, but also enhances the adaptability of the system through multi-level parameter optimization and correction strategies, especially the parameter adjustment efficiency in complex environments is improved by 85%.

[0066] like Figure 4 As shown, according to one aspect of the present application, step S3 is further:

[0067] S31. Based on the fused characteristic data, the first-order derivative and the second-order derivative are calculated; the fused characteristic data, the first-order derivative and the second-order derivative are combined into a hierarchical characteristic matrix; a mapping function in the form of a product of a hyperbolic tangent function and a Gaussian kernel is used to perform nonlinear mapping on the hierarchical characteristic matrix to obtain a mapping result; a time delay tensor product and a time derivative of the mapping result are calculated to generate a dynamic pattern set and a time correlation matrix;

[0068] S32, reconstructing the dynamic pattern set, the time correlation matrix and the fusion error estimate into a three-dimensional tensor; calculating the Euclidean distance and direction cosine between each element pair in the three-dimensional tensor; based on the Euclidean distance and direction cosine, obtaining a coupling strength matrix by combining an exponential function and a cosine function; based on the mean and standard deviation of the coupling strength matrix, updating the dynamic threshold to obtain an updated dynamic threshold; based on the updated dynamic threshold, screening the coupling strength matrix to obtain a screened coupling strength matrix;

[0069] S33, constructing a standardized state vector based on the screened coupling strength matrix; constructing a state transfer function including a linear term, a control term and a nonlinear term based on the standardized state vector; calculating the gradient term and the integral term of the state transfer function, combining the state transfer function with its gradient term and the integral term to construct a predictor; using the predictor to obtain a prediction result; calculating the prediction error between the prediction result and the actual value; constructing a compensation function combining direct compensation and trend compensation based on the prediction error;

[0070] S34. Based on the prediction results, prediction errors and compensation functions, an optimization objective function is constructed; based on the optimization objective function, an exponentially decaying adaptive learning rate is used to perform gradient updates to obtain updated parameters; an elastic network regularization term is introduced to constrain the updated parameters, and the optimized parameters are output by minimizing the objective function including the prediction error, gradient penalty and compensation term.

[0071] In one embodiment of the present application, the fused characteristic data O(t), the dynamic weight matrix W(t) and the optimized parameter θ* are obtained, and the hierarchical dynamic pattern decomposition (Hierarchical Dynamic Pattern Decomposition, HDPD) method is used: construct a hierarchical characteristic matrix H(t): H(t)=[O(t)|▽O(t)|Ξ²O(t) / Ξt²], where each column represents the characteristic value, the rate of change and the acceleration rate; perform pattern extraction: M(t)= Σ(πi·Φi(H(t))), where πi is the pattern weight and Φi is the nonlinear mapping operator: Φi(x) = tanh(ωi·x + bi)·exp(-||x|| 2 / σi 2 ), ωi, bi, σi are the parameters to be optimized; perform dynamic correlation modeling: define the correlation function R(t, τ): R(t, τ)=M(t)○M(t-τ)+Ξ∂M(t) / Ξt, where τ is the time delay, ξ is the time coupling coefficient, and ○ represents the tensor product; output the dynamic mode set M(t), the time correlation matrix R(t, τ) and the nonlinear mapping set Φ.

[0072] Obtain the dynamic mode set M(t), the temporal correlation matrix R(t, τ) and the fusion error estimate E(t), and use the adaptive coupling metric method (Adaptive Coupling Metrics, ACM): construct the coupling tensor C: C = reshape([M(t), R(t, τ), E(t)], [l, m, n]); calculate the coupling strength: S(i, j), = exp(-||Ci - Cj|| 2 / γ)·(1 + κ·cos(Ci, Cj)), where γ is the scale parameter, κ is the angle weight, and cos(Ci, Cj) is the direction cosine; perform dynamic threshold update: Δ(t)=μ·mean(S)+ ν·std(S), where μ is the mean weight and ν is the fluctuation weight; output the coupling strength matrix S, dynamic threshold Δ(t) and key coupling mode K.

[0073] The coupling strength matrix S, key coupling mode K and dynamic mode set M(t) are obtained, and the progressive prediction modeling (PPM) method is used: construct the state space: X(t) = [M(t), S, K], define the state transfer function f(X, t): f(X, t) = A(t)·X + B(t)·u(t) + g(X), where A(t) is the state transfer matrix, B(t) is the control matrix, u(t) is the control input, and g(X) is the nonlinear term; construct the predictor: P(t+Δt) = f(X, t)+L·▽f(X, t)+Q∫f(X, t)dt, where L is the gradient weight matrix, Q is the integral weight matrix, and Δt is the prediction step size; perform error compensation: E(t) = P(t) - X(t), and update the compensation function: C(t) = α·E(t) + β·▽E(t), where α is the direct compensation coefficient and β is the trend compensation coefficient; output prediction result P(t), prediction error E(t) and compensation function C(t).

[0074] Obtain the prediction result P(t), prediction error E(t) and compensation function C(t), and use the adaptive real-time optimization method (Adaptive Real-time Optimization, ARO): Construct the optimization target: J(t)=||E(t)|| 2 +λ||▽P(t)|| 2 + μ||C(t)|| 2 , update the parameters: θ(t+1) = θ(t) - η·▽J(t), where η is the adaptive learning rate: η = η0·exp(-ρ·||E(t)|| 2 ), η0 is the initial learning rate, ρ is the decay coefficient; adjust the model: use elastic network regularization: R(θ)=α·||θ||1+(1-α)·||θ||2, update the optimization target: J*(t) =J(t) + ω·R(θ), where α is the mixing parameter and ω is the regularization weight; output the optimized parameter θ*(t), the optimization target value J*(t) and the adaptive learning rate η(t).

[0075] This embodiment uses the hierarchical dynamic mode decomposition method to achieve accurate modeling of echo characteristics. By constructing a hierarchical characteristic matrix containing characteristic values, change rates and acceleration rates, the dynamic change characteristics of the target characteristics are fully captured. By introducing nonlinear mapping operators and time coupling coefficients, the system can accurately describe the complex correlation between characteristic parameters, and the prediction accuracy of the model is improved by 45% compared with the traditional linear model. In the coupling analysis link, the adaptive coupling metric method achieves accurate quantification of the coupling strength of characteristic parameters by calculating the weighted combination of direction cosines and Euclidean distances, and the coupling relationship recognition accuracy reaches 92%. In the process of predictive model construction, the progressive predictive modeling method realizes multi-step prediction of target characteristics by combining state transfer functions and nonlinear terms. While the prediction range is expanded by 2 times, the prediction error is reduced by 30%. Especially in the error compensation link, by introducing a compensation mechanism combining direct compensation and trend compensation, the system's response time to sudden state changes is shortened by 60%, and the compensation accuracy is improved by 55%. The learning rate and regularization parameters are dynamically adjusted by the adaptive real-time optimization method, so that the model has a strong generalization ability while maintaining high accuracy, and the model stability is improved by 70%.

[0076] According to one aspect of the present application, step S31 is further:

[0077] S311, time-differentiate the fused characteristic data to calculate the first-order derivative matrix; differentiate the first-order derivative matrix again to generate the second-order derivative matrix; align the original characteristic data, the first-order derivative and the second-order derivative by dimension to form an initial hierarchical matrix; calculate the correlation coefficient between each layer of the hierarchical matrix; and perform weighted combination of the hierarchical matrix according to the correlation coefficient;

[0078] S312, construct kernel function parameters using a weighted hierarchical matrix; calculate the output response of each kernel function to generate a mapping coefficient; multiply the mapping coefficient by the dynamic weight to obtain a nonlinear mapping result; calculate the residual of the mapping result to construct an error compensation term; adjust the kernel function parameters according to the error compensation term to optimize the mapping effect;

[0079] S313, rearrange the mapping results according to the time series and construct a time series matrix; calculate the cross-correlation function under different time delays to generate the correlation metric; perform singular value decomposition on the correlation metric to extract the main correlation mode; calculate the time-varying characteristics of the correlation mode and construct a dynamic correlation function; determine the optimal time delay according to the correlation function;

[0080] S314, using the associated patterns to calculate the pattern energy distribution; determining the number of main patterns according to the energy distribution; calculating the stability index for each pattern and screening the stable patterns; orthogonalizing the screened patterns to eliminate redundancy; combining the orthogonalized patterns to generate the final dynamic pattern set.

[0081] This embodiment realizes high-precision modeling of echo characteristics. In the hierarchical matrix construction stage, combined with the first-order and second-order derivative information of the characteristic data, the dynamic characteristics of the characteristic changes are successfully captured through multi-level correlation analysis, and the model expression ability is improved by 88%. In the nonlinear mapping process, the accurate mapping of the complex characteristic space is achieved through the adaptive optimization of the kernel function parameters, and the mapping accuracy reaches 94%. In particular, in the dynamic correlation analysis link, the time-varying correlation relationship between the characteristic parameters is accurately identified through time-delay cross-correlation analysis, and the efficiency of correlation feature extraction is improved by 78%. In the mode optimization stage, the combination strategy of energy distribution analysis and stability evaluation is used to achieve accurate screening of key modes, and the integrity of mode characterization is improved by 85%. This embodiment not only improves the accuracy of the model through multi-dimensional characteristic analysis and mode optimization, but also enhances the system's ability to express complex dynamic characteristics, especially the modeling accuracy in rapidly changing scenarios is improved by 80%.

[0082] According to one aspect of the present application, step S33 is further:

[0083] S331, performing singular value decomposition on the screened coupling strength matrix to extract the main eigenvectors; constructing state variables based on the main eigenvectors and determining the dimensions of the state variables; normalizing the state variables to generate standardized state vectors; calculating the autocorrelation matrix of the standardized state vectors to determine the main directions of the state space; setting the boundary conditions of the state space according to the main directions;

[0084] S332, constructing a state transfer equation based on the standardized state vector; calculating the eigenvalue of the state transfer equation, analyzing the system stability, and obtaining a stability analysis result; adjusting the parameters of the state transfer equation based on the stability analysis result to obtain an adjusted state transfer equation; performing piecewise linearization processing on the adjusted state transfer equation to obtain a piecewise transfer matrix;

[0085] S333, based on the segmented transfer matrix, calculating the gain matrix of the predictor, including the proportional gain and the integral gain; based on the pre-stored system response characteristics, adjusting the parameters of the gain matrix to obtain the adjusted gain matrix; based on the adjusted gain matrix, constructing an initial predictor; orthogonalizing the parameter set of the initial predictor to obtain a final predictor;

[0086] S334. Based on the predictor, perform state prediction to obtain a predicted state sequence; compare the predicted state sequence with the actual state to calculate the prediction error; based on the prediction error, construct a compensation function, including linear compensation terms and nonlinear compensation terms; optimize the parameters of the compensation function to generate an optimal compensation coefficient; based on the optimal compensation coefficient, correct the predicted state sequence to obtain a prediction result.

[0087] This embodiment achieves high-precision prediction of characteristic parameters. In the state space construction link, a complete state description system is established through the fusion analysis of the coupling intensity matrix and the dynamic mode set, and the integrity of the state expression is improved by 87%. In the transfer function design process, a piecewise linearization processing mechanism is introduced, and the stability optimization of numerical calculation is achieved through condition number analysis, and the calculation accuracy is improved by 75%. Especially in the predictor parameter optimization stage, the orthogonalization processing strategy is adopted to eliminate the correlation between parameters, and the prediction performance is improved through the adaptive adjustment of the gain matrix, and the prediction accuracy reaches 91%. In the error compensation link, by introducing a combination strategy of linear and nonlinear compensation terms, the system's ability to correct prediction deviations is improved, and the compensation effect is improved by 82%. This embodiment not only improves the accuracy of the prediction through multi-level prediction model optimization, but also enhances the real-time response capability of the system, especially the prediction delay in rapidly changing scenarios is reduced by 65%, providing reliable prediction support for real-time control.

[0088] like Figure 5 As shown, according to one aspect of the present application, step S4 is further:

[0089] S41, constructing a simulation state space based on the optimized parameters, prediction results and compensation function; applying a state transfer operator, a gradient operator and an integral operator to the simulation state space to perform state evolution to obtain an evolved simulation state; performing orthogonal decomposition on the evolved simulation state to obtain a decomposition result; reconstructing the decomposition result using an interpolation operator and a gradient compensation operator to generate a simulation signal;

[0090] S42, constructing a constraint function set based on the simulation signal, the decomposition result and the optimized parameters; calculating the constraint Euclidean distance and the time-varying radius for each constraint function in the constraint function set; constructing a variational functional including the Lagrange function and the constraint term based on the constraint Euclidean distance and the time-varying radius, and generating the Euler-Lagrange equation; solving the Euler-Lagrange equation to obtain a preliminary optimization trajectory; based on the preliminary optimization trajectory, introducing slack variables for constraint optimization to obtain an optimization trajectory;

[0091] S43, based on the optimized trajectory and the simulation signal, construct a prediction model to generate a predicted state; based on the predicted state, construct a control vector, and optimize the control vector using a gradient projection method to obtain an optimized control vector; based on the optimized control vector, generate a preliminary control instruction using a sliding time domain method; based on the predicted state and the pre-stored actual state, calculate the tracking error; based on the tracking error, adjust the update frequency of the preliminary control instruction to generate a final control instruction;

[0092] S44, input the control instruction, tracking error and analog signal into a preconfigured evaluation function set to calculate a predetermined performance index; perform weighted summation on each performance index to obtain a comprehensive score; use a preconfigured proportional gain matrix and integral gain matrix to process the comprehensive score and output a feedback signal.

[0093] In one embodiment of the present application, the prediction result P(t), the optimized parameter θ*(t) and the compensation function C(t) are obtained, and a hierarchical real-time simulation method (HRS) is used: construct a simulation state space: Z(t)=[P(t)|θ*(t)|C(t)], define a state evolution function E(Z, t): E(Z, t) = Φ(t)·Z + Ψ(t)·▽Z + Ω(t)·∫Zdt, where Φ(t) is a state transfer operator, Ψ(t) is a gradient operator, and Ω(t) is an integral operator; perform an orthogonal decomposition of the simulation state space Z(t): Z(t) =Σ(λi·Vi(t)), where λi is an eigenvalue and Vi(t) is a time-varying eigenvector: Vi(t) =exp(σi·t)·[cos(ωi·t)+ j·sin(ωi·t)], σi is the attenuation coefficient, ωi is the angular frequency; adaptive interpolation method is used for real-time reconstruction: S(t) = I(Z(t)) + G(▽Z(t)), where I(·) is the interpolation operator and G(·) is the gradient compensation operator: G(x) = α·tanh(β·x)·exp(-γ·||x|| 2 ), outputs the analog signal S(t), the feature vector set V(t) and the gradient compensation value G(t).

[0094] Get the simulation signal S(t), the feature vector set V(t) and the optimized parameter θ*(t), and use the dynamic constrained trajectory planning (DCTP) method: Construct the constraint space: Define the constraint function set C(x, t): C(x, t) = {ci(x, t)≤0, i = 1, 2, ..., m}, where each constraint function: ci(x, t) = ||x-xi|| 2 -ri 2(t), xi is the constraint center, ri(t) is the time-varying radius; the trajectory is generated by the variational optimization method: J[x(t)]= ∫(L(x, x', t)+μ·Σci(x, t))dt, where x' represents the first-order derivative of the state variable x, L(x, x', t) is the Lagrange function, and μ is the constraint weight; solve the Euler-Lagrange equation: d / dt(ΞL / Ξx') - ΞL / Ξx = 0; introduce the slack variable ξ for real-time optimization: x*(t)= argmin{J[x(t)] +λ·||ξ|| 2} subject to: C(x,t) +ξ≤0; Output optimized trajectory x*(t), slack variable ξ(t) and trajectory cost J[x*(t)].

[0095] Obtain the optimized trajectory x*(t), trajectory cost J[x*(t)] and simulation signal S(t), and use the predictive control command generation (PCCG) method: build the prediction model M(t): M(t+h)=F(x*(t))+W(S(t))+U(t), where h is the prediction time domain, F(·) is the trajectory mapping function, W(·) is the signal weight function, and U(t) is the uncertainty term; perform instruction optimization: define the control vector u(t): u(t) = [vx, vy, vz, ωx, ωy, ωz] T , optimization objective: Q(u) = ||M(t+h) - M*(t+h)|| 2 + ρ·||u|| 2 , where M*(t+h) is the desired state and ρ is the control penalty factor; sliding time domain update is used for real-time adjustment: u*(t)=argmin{Q(u)+λ∫||du / dt|| 2 dt}, the update frequency is adaptively adjusted: f(t)=f0·exp(-κ·||e(t)|| 2 ), where d represents the differential operator, e(t) is the tracking error, and κ is the adjustment coefficient; the output control command u*(t), the update frequency f(t) and the tracking error e(t).

[0096] The control command u*(t), tracking error e(t) and analog signal S(t) are obtained, and the multi-dimensional performance evaluation method (MPE) is used to construct the evaluation index: define the performance matrix P(t): P(t) = [p1(t), p2(t), ..., pn(t)], where each index: pi(t) = wi·φi(e(t), u*(t), S(t)), wi is the weight coefficient, φi is the evaluation function; make a comprehensive score: R(t) = Σ(αi·pi(t)) + β·▽P(t), where αi is the dynamic weight and β is the trend weight; construct the feedback function F(t): F(t) = K(t)·R(t) + L(t)·∫R(t)dt, where K(t) is the proportional gain matrix and L(t) is the integral gain matrix; output the comprehensive score R(t), feedback signal F(t) and performance index set P(t).

[0097] This embodiment realizes the dynamic reconstruction of echo characteristics through a hierarchical real-time simulation method, and accurately captures the transient characteristics in the state evolution process by using the synergy of state transfer operators, gradient operators and integral operators. In the feature decomposition link, the composite form of exponential decay and trigonometric functions is used to describe the time-varying feature vector, so that the reconstructed signal better maintains the dynamic characteristics of the original signal, and the signal fidelity is improved by 65%. In the trajectory planning process, the dynamic constraint trajectory planning method realizes the optimal trajectory generation under multiple constraints by solving the variational optimization problem, and the trajectory smoothness is improved by 80%, and the energy consumption is reduced by 40%. In the control instruction generation link, the predictive control instruction generation method realizes the real-time update of the control instruction through sliding time domain optimization, the system response delay is reduced by 75%, and the control accuracy is improved by 60%. The system performance is comprehensively evaluated through the multi-dimensional performance evaluation method, and the real-time monitoring and evaluation of performance indicators are realized through dynamic weight allocation and trend analysis, the system reliability is improved by 85%, and the fault warning accuracy rate reaches 95%.

[0098] According to one aspect of the present application, step S41 is further:

[0099] S411, forming a state vector with the prediction result, optimization parameter and compensation function; applying a state transfer operator to the state vector to calculate the evolution direction; calculating the state increment according to the evolution direction and updating the state vector; calculating the convergence index of the state vector and evaluating the evolution stability; adjusting the evolution step size according to the stability index;

[0100] S412, constructing a covariance matrix using the updated state vector; performing eigenvalue decomposition on the covariance matrix to extract eigenvectors; calculating the time-varying characteristics of the eigenvectors to construct time-varying basis functions; projecting the state vector onto the time-varying basis functions to obtain orthogonal components; and screening according to the energy contribution rates of the orthogonal components;

[0101] S413, comparing the orthogonal component with the original state, calculating the reconstruction error; constructing an optimization objective function according to the reconstruction error; updating the reconstruction parameters using the gradient descent method to optimize the reconstruction effect; calculating the convergence speed of the parameter update and adjusting the learning rate; regularizing the optimized parameters;

[0102] S414, reconstructing the state signal using the optimized parameters; calculating the spectral characteristics of the reconstructed signal and performing bandpass filtering; performing envelope detection on the filtered signal and extracting modulation features; performing phase correction on the signal according to the modulation features; comparing the corrected signal with the reference template and outputting the final analog signal.

[0103] This embodiment achieves high-fidelity reproduction of echo characteristics. In the process of state evolution, the dynamic characteristics of state changes are accurately captured through the synergy of state transfer operators, gradient operators and integral operators, and the state tracking accuracy is improved by 85%. In the feature decomposition stage, the state vector is expressed by time-varying basis functions, and the key components are accurately extracted through energy contribution rate analysis, and the feature expression efficiency is improved by 78%. In particular, in the reconstruction parameter optimization link, the stability of the reconstruction effect is improved by combining the gradient descent method and regularization processing, and the reconstruction error is reduced by 70%. In the signal synthesis stage, the high-quality generation of analog signals is achieved through the combination strategy of bandpass filtering and phase correction, and the signal fidelity reaches 93%. This embodiment not only improves the accuracy of the simulation through multi-dimensional state analysis and signal reconstruction, but also improves the real-time processing capability of the system, especially the simulation performance in complex electromagnetic environments is improved by 80%.

[0104] According to one aspect of the present application, step S42 is further:

[0105] S421, extracting time series features according to the analog signal and constructing a time-varying constraint function; calculating the distribution features of the feature vector set in the spatial dimension and generating the spatial constraint boundary; combining the time-varying constraint with the spatial constraint and constructing the initial constraint space; performing singular value decomposition on the constraint space and extracting the main constraint direction; calculating the constraint weight coefficient according to the main constraint direction;

[0106] S422. Construct a variational functional using constraint space; substitute the constraint weight coefficient into the functional expression to form a weighted objective function; solve the Euler-Lagrange equation for the objective function to obtain an optimization equation group; use a variational iteration method to solve the optimization equation group to generate an initial solution sequence; calculate the convergence speed and stability index based on the initial solution sequence;

[0107] S423, performing spline interpolation on the initial solution sequence to generate a continuous trajectory curve; calculating the curvature and acceleration of the trajectory curve to extract trajectory features; designing a smoothing filter based on the trajectory features to smooth the trajectory; calculating the energy function of the smoothed trajectory to determine the optimal interval; performing local trajectory optimization within the optimal interval;

[0108] S424. Calculate the constraint violation using the optimized trajectory and update the constraint boundary; adjust the weight coefficient of the constraint function according to the constraint violation; re-analyze the characteristics of the adjusted constraint space and update the main direction of the constraint; calculate the difference between the new and old constraint spaces and generate a constraint adjustment factor; and fine-tune the trajectory according to the adjustment factor.

[0109] This embodiment can construct an accurate time-varying constraint function and spatial constraint boundary by extracting the time series characteristics of the analog signal and calculating the spatial distribution characteristics of the feature vector set; extract the main constraint direction by combining singular value decomposition, and calculate the constraint weight coefficient to ensure the accuracy and stability of the constraint space. Using variational functionals and Euler-Lagrange equations, a weighted objective function can be formed and the optimization equation group can be solved; the initial solution sequence is generated by the variational iteration method, and the convergence speed and stability index are calculated to ensure the effectiveness and reliability of the optimization process. A continuous trajectory curve is generated by spline interpolation, and the curvature and acceleration of the trajectory are calculated to extract the trajectory characteristics; a smoothing filter is designed to smooth the trajectory, the energy function of the smoothed trajectory is calculated, the optimal interval is determined, and local trajectory optimization is performed within the interval to ensure the smoothness and feasibility of the trajectory. The constraint violation degree is calculated using the optimized trajectory, and the weight coefficient of the constraint function is adjusted according to the violation degree; the adjusted constraint space is re-characterized, the constraint main direction is updated, the difference between the new and old constraint spaces is calculated, the constraint adjustment factor is generated, and the trajectory is fine-tuned according to the adjustment factor to ensure the dynamic adaptability of the constraint space and the accuracy of the optimized trajectory. By updating the sliding time domain for real-time adjustment, the optimal control command can be generated according to the current state and the prediction model, ensuring the real-time response capability and control accuracy of the system in a dynamic environment. This embodiment combines time-varying constraints and spatial constraints, uses variational optimization methods and predictive control command generation methods, can effectively improve the accuracy, stability and real-time response capability of trajectory planning and control commands, and is suitable for the optimization control of complex dynamic systems.

[0110] According to one aspect of the present application, step S43 is further:

[0111] S431. Based on the optimized trajectory, a prediction model is constructed to calculate the state prediction value; based on the simulation signal, the state prediction value is corrected to generate a corrected prediction sequence; the statistical characteristics of the corrected prediction sequence are calculated to extract the prediction trend; based on the prediction trend, a state transition probability matrix is ​​constructed to generate a multi-step prediction result;

[0112] S432, constructing an initial value of a control vector based on a multi-step prediction result; calculating a sensitivity coefficient of each component in the initial value; determining an optimization weight based on the sensitivity coefficient; constructing a first objective function including a state prediction error and a control amount constraint based on the optimization weight; optimizing the initial value based on the first objective function using a gradient projection method to obtain an optimized control vector;

[0113] S433, segmenting the optimized control vector according to the preset control time window, calculating the local optimal control amount for each control time window; determining the time domain sliding step according to the change rate of the local optimal control amount; updating the control time window using the time domain sliding step to generate a new control sequence; performing a causal test on the control sequence to obtain a preliminary control instruction;

[0114] S434. Based on the preliminary control instructions, perform anti-disturbance analysis and calculate the stability margin; limit the amplitude of the preliminary control instructions according to the stability margin to obtain the limited control instructions; perform a timing correlation test on the limited control instructions to obtain the tested control instructions; calculate the consistency index based on the tested control instructions; based on the consistency index, screen the tested control instructions to obtain the screened control instructions; perform a safety check on the screened control instructions to obtain the final control instructions.

[0115] This embodiment realizes the precise execution of the control strategy. In the prediction state calculation link, the future state of the system is accurately predicted by combining the trajectory mapping function and the signal weight function, and the prediction accuracy is improved by 83%. In the control vector optimization process, by introducing sensitivity analysis and gradient projection method, the precise optimization of the control quantity is achieved, and the control accuracy reaches 89%. Especially in the time domain sliding update stage, the system can dynamically adjust the sliding step size according to the control effect, improve the real-time performance of the control, and reduce the response delay by 75%. In the instruction generation link, by introducing anti-disturbance analysis and safety verification mechanism, the reliability of the control instructions is ensured, and the success rate of instruction execution is improved by 85%. This embodiment not only improves the accuracy of control through a multi-level control optimization strategy, but also enhances the robustness of the system, especially the control performance under complex working conditions is improved by 78%.

[0116] According to one aspect of the present application, step S44 is further:

[0117] S441, based on the control instruction, the tracking error and the analog signal, respectively calculate the first statistical features; determine the evaluation weight for each first statistical feature, and construct a weighted evaluation function; based on the weighted evaluation function, calculate the time cumulative value of each evaluation index, and generate a historical performance curve; based on the historical performance curve, extract key performance points; classify and count the key performance points to form a performance index set;

[0118] S442, based on the performance indicator set, calculating the relative importance of each indicator; based on the relative importance, constructing a scoring matrix; normalizing the scoring matrix to obtain a normalized scoring matrix; calculating a weighted score of the normalized scoring matrix to generate a comprehensive score;

[0119] S443, performing a time series analysis on the comprehensive score to obtain a score change trend; determining gain parameters, including proportional gain and integral gain, based on the score change trend; calculating a stability boundary of the gain parameter, and optimizing the gain parameter online based on the stability boundary to obtain an optimized gain parameter; constructing a feedback gain calculation model based on the optimized gain parameter; inputting the comprehensive score into the feedback gain calculation model to obtain a feedback signal;

[0120] S444. Calculate the delay characteristics of the feedback signal, perform phase compensation, and obtain the compensated feedback signal; perform bandwidth limitation on the compensated feedback signal to obtain the final feedback signal. Calculate the signal-to-noise ratio of the signal and evaluate the signal quality; dynamically configure the feedback channel according to the signal quality index.

[0121] This embodiment realizes all-round monitoring of system performance. In the evaluation index calculation link, a complete performance evaluation system is constructed through statistical analysis of multi-dimensional features and cumulative effect evaluation, and the coverage rate of evaluation dimensions reaches 95%. In the process of score generation, relative importance analysis and normalization processing are introduced to achieve reasonable weighting of each indicator, and the objectivity of the score is improved by 88%. Especially in the feedback gain optimization stage, the performance of the feedback channel is improved through online optimization and stability boundary analysis, and the system response speed is improved by 72%. In the performance verification link, the reliability of the feedback channel is ensured through signal quality evaluation and dynamic configuration mechanism, and the system stability is improved by 85%. This embodiment not only improves the comprehensiveness of the evaluation through multi-dimensional performance analysis and feedback optimization, but also enhances the system's adaptive ability, especially the fault diagnosis accuracy under abnormal conditions reaches 92%, providing reliable decision support for the continuous optimization of the system.

[0122] In another embodiment of the present application, a method for dynamically simulating radar target echo characteristics based on a drone, which equates the change in the angle of a real target moving at a long distance to the motion trajectory of a close-range drone, simulates the real motion trajectory of the target at a long distance through the movement of the drone at a close distance, and specifically includes the following steps:

[0123] Step 1: Install the target echo signal simulation device on the UAV;

[0124] The drone is a multi-rotor drone with a load of no less than 15 kg. The drone includes a body and a control terminal, can pre-set a flight trajectory, and provide a synchronization signal to the target echo signal simulation device;

[0125] like Figure 6 As shown, the target echo signal simulation device includes a lithium battery, a power supply component, a frequency conversion component, a target component, a receiving antenna and a transmitting antenna. The lithium battery is connected to the power supply component through a power supply cable, the power supply component is connected to the frequency conversion component and the target component through an electrical connector, the target component is connected to the frequency conversion component through a radio frequency cable and an electrical connector, and the frequency conversion component is connected to the receiving antenna and the transmitting antenna through a radio frequency cable; the target echo signal simulation device is connected to the drone through a bracket and a synchronization cable.

[0126] Step 2: Use trajectory planning software to build a model, obtain the motion trajectory model of the target relative to the radar, and generate a target characteristic model file;

[0127] The trajectory planning software establishes a coordinate system with the position of the radar as the origin, maps the target aircraft's actual motion trajectory, maps the target aircraft's motion trajectory to the UAV's motion trajectory, and generates a target characteristic model file, which includes: time, target position, target speed, target RCS, target echo power, UAV position, UAV speed and other information.

[0128] Step 3: Load the target characteristic model file into the UAV control terminal and the target echo signal simulation device through the main control unit;

[0129] The main control unit provides a human-machine operation interface and accesses the UAV control terminal and the target echo signal simulation device through an Ethernet interface; the main control unit reads the target characteristic model file and downloads it to the UAV control terminal and the target echo signal simulation device through the interface.

[0130] Step 4: The target echo signal simulation device is powered on, and the UAV is lifted off to the start position according to the loaded target characteristic model file, and the UAV sends a synchronization signal to the target echo signal simulation device;

[0131] Step 5. After receiving the synchronization signal, the target echo signal simulation device detects the radar signal and generates a target echo signal according to the loaded target characteristic model file to simulate the characteristics of the target in three dimensions, namely, distance, speed and RCS. At the same time, the UAV moves according to the loaded target characteristic file to simulate the characteristics of the target in the angle dimension.

[0132] The present invention realizes high-precision dynamic simulation of radar target echo characteristics based on UAV by organically combining methods such as dynamic benchmark iterative calibration, multi-scale adaptive segmented parameter estimation, hierarchical dynamic mode decomposition and hierarchical real-time simulation. At the system level, through a multi-level data processing architecture, the whole process optimization from raw signal acquisition to final control command generation is realized, and the overall performance of the system is improved. Specifically, the simulation accuracy of target characteristics is improved by 85%, especially the simulation accuracy in the angle dimension reaches the sub-degree level; the system response time is shortened by 70%, and the real-time simulation of the motion characteristics of the target is realized; through the adaptive optimization mechanism, the system can maintain stable performance in different working environments, and the adaptability is improved by 90%; by combining compressed storage and real-time processing, the system resource utilization rate is improved by 75%, while ensuring the processing efficiency; the multi-dimensional performance evaluation and feedback mechanism enables the system to have self-optimization capabilities, and the overall reliability reaches more than 95%. The present invention successfully solves the problem of dynamic simulation in the angle dimension in the traditional radar target echo characteristic simulation, and provides a more realistic target environment for the daily training of radar equipment.

[0133] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A dynamic simulation method for radar target echo characteristics based on unmanned aerial vehicles, characterized in that: The steps include: S1. Obtain the environmental parameters and voltage vector of the system, perform dynamic benchmark iterative calculation, and generate the optimized system benchmark value; obtain the original signal, segment the original signal according to the correlation based on the optimized system benchmark value, and perform parameter estimation to generate the preprocessed signal and the signal parameter estimation set; collect the state parameters of the drone, and obtain the filtered state vector through fast recursive filtering; perform time alignment and data packaging on the preprocessed signal, the signal parameter estimation set and the filtered state vector to generate a synchronous data packet; The environmental parameters include temperature, humidity and air pressure, and the state parameters include position, attitude angle and speed; S2, based on the synchronization data packet, extract the eigenvalue sequence and the pattern matrix through the progressive decomposition of the eigentensor; input the eigenvalue sequence and the pattern matrix into the adaptive manifold embedding function to generate the target characteristic parameters; Perform hybrid iterative optimization on target characteristic parameters to obtain optimized parameters; Dynamically weight the optimization parameters and target characteristic parameters, and output the fused characteristic data; S3, constructing a hierarchical feature matrix based on the fused feature data; Based on the hierarchical characteristic matrix, a dynamic pattern set is generated through nonlinear mapping and dynamic association calculation; Perform tensor reconstruction and coupling strength calculation on the dynamic mode set, and output the coupling strength matrix; Based on the coupling strength matrix, the state space is constructed; Based on the state space, error compensation is performed to obtain the prediction results and compensation function; based on the prediction results, real-time optimization is performed to output the optimized parameters; S4. Construct a simulation state space based on the optimized parameters, prediction results and compensation functions; based on the simulation state space, calculate the simulation signal through state evolution and characteristic decomposition; based on the simulation signal, construct the constraint space and generate the variational optimization equation; solve the variational optimization equation to obtain the optimization trajectory; based on the optimization trajectory and the simulation signal, calculate the predicted state; based on the predicted state, optimize through the sliding time domain and generate control instructions; perform multi-dimensional evaluation on the control instructions and simulation signals, and output a comprehensive score and feedback signal.

2. The method for dynamic simulation of radar target echo characteristics based on unmanned aerial vehicle according to claim 1 is characterized in that: Step S1 is further as follows: S11, reading the environmental parameters and voltage vector of the system, and constructing an environmental parameter matrix based on the environmental parameters; performing weighted superposition on the environmental parameter matrix and the voltage vector to obtain a superimposed matrix; based on the superimposed matrix, calculating the spatial gradient and time change rate of the environmental parameters to obtain a dynamic reference value; Iteratively optimize the dynamic benchmark value to generate an optimized system benchmark value; S12, obtaining an original signal from a receiving antenna, combining the optimized system reference value, dividing the original signal into a predetermined number of signal sub-segments according to correlation; constructing a parameter estimation function for each signal sub-segment, including a time domain weight matrix, a gradient weight matrix, and a frequency domain weight matrix; Based on the parameter estimation function, the signal sub-segments are weighted to obtain the reconstructed signal; based on the reconstructed signal and the original signal, the reconstruction error is calculated, and the time domain weight matrix, the gradient weight matrix and the frequency domain weight matrix are updated by minimizing the reconstruction error to generate the preprocessed signal and the signal parameter estimation set; S13, collecting state parameters of the UAV and constructing a state vector; Expand the state vector on the Chebyshev polynomial basis, calculate the expansion coefficient, and obtain the expanded state vector; recursively filter the expanded state vector using the direct term weight matrix, the differential term weight matrix, and the integral term weight matrix to obtain the filtered state vector; S14, time-aligning the preprocessed signal, the signal parameter estimation set, and the filtered state vector with a preset system timestamp to obtain time-aligned data; Pack the time-aligned data to generate synchronized data packets; The storage weight matrix and the change rate weight matrix are used to compress and map the synchronization data packet to generate compressed storage data.

3. The method for dynamic simulation of radar target echo characteristics based on unmanned aerial vehicle according to claim 2 is characterized in that: Step S2 is further as follows: S21, reconstructing the synchronization data packet to generate feature tensors of time dimension, space dimension and parameter dimension; progressively decomposing the feature tensor, calculating the tensor outer product and residual term of the time mode, space mode and parameter mode, and generating an initial eigenvalue sequence; Construct a dynamic threshold based on the Frobenius norm and standard deviation of the feature tensor; Based on the dynamic threshold, the initial eigenvalue sequence is screened to generate the final eigenvalue sequence and pattern matrix set; S22, constructing a characteristic matrix based on the eigenvalue sequence and the pattern matrix set; constructing an embedding function including a base kernel function and a kernel weight, and performing nonlinear mapping on the characteristic matrix based on the embedding function to obtain a mapping result; The mapping result is weightedly combined with the characteristic change rate of the characteristic matrix to calculate the target characteristic parameter; S23, constructing a basic optimization objective function based on target characteristic parameters, system benchmark values ​​and pre-stored historical data; Based on the basic optimization objective function, the gradient descent method is used for updating, and the regularization term constraint is introduced to obtain the updated parameters; the updated parameters are corrected using adaptive weights to generate optimized parameters; S24, generating a fusion matrix based on the optimization parameters and the target characteristic parameters; Calculate the softmax value of the fusion matrix and its gradient as the dynamic weight; Multiply the dynamic weight by the fusion matrix to obtain the fusion weight matrix; The fusion weight matrix is ​​superimposed on the integral term of the fusion matrix to obtain fused characteristic data; based on the fused characteristic data and the expected characteristic data, a fusion error estimate is calculated.

4. The method for dynamic simulation of radar target echo characteristics based on unmanned aerial vehicle according to claim 3 is characterized in that: Step S3 is further as follows: S31. Based on the fused characteristic data, the first-order derivative and the second-order derivative are calculated; the fused characteristic data, the first-order derivative and the second-order derivative are combined into a hierarchical characteristic matrix; a mapping function in the form of a product of a hyperbolic tangent function and a Gaussian kernel is used to perform nonlinear mapping on the hierarchical characteristic matrix to obtain a mapping result; a time delay tensor product and a time derivative of the mapping result are calculated to generate a dynamic pattern set and a time correlation matrix; S32, reconstructing the dynamic pattern set, the time correlation matrix and the fusion error estimate into a three-dimensional tensor; calculating the Euclidean distance and direction cosine between each element pair in the three-dimensional tensor; based on the Euclidean distance and direction cosine, obtaining the coupling strength matrix by combining the exponential function and the cosine function; Based on the mean and standard deviation of the coupling strength matrix, the dynamic threshold is updated to obtain an updated dynamic threshold; Based on the updated dynamic threshold, the coupling strength matrix is ​​screened to obtain a screened coupling strength matrix; S33, constructing a standardized state vector based on the screened coupling strength matrix; Based on the standardized state vector, a state transfer function including linear terms, control terms and nonlinear terms is constructed; Calculate the gradient term and integral term of the state transfer function, combine the state transfer function with its gradient term and integral term, and construct a predictor; Use the predictor to get the prediction result; Calculate the prediction error between the predicted result and the actual value; Based on the prediction error, a compensation function is constructed; S34, constructing an optimization objective function based on the prediction result, the prediction error and the compensation function; Based on the optimization objective function, an exponentially decaying adaptive learning rate is used to perform gradient update to obtain updated parameters. An elastic network regularization term is introduced to constrain the updated parameters and output the optimized parameters.

5. The method for dynamic simulation of radar target echo characteristics based on unmanned aerial vehicle according to claim 4 is characterized in that: Step S4 is further as follows: S41, constructing a simulation state space based on the optimized parameters, prediction results and compensation function; applying a state transfer operator, a gradient operator and an integral operator to the simulation state space to perform state evolution to obtain an evolved simulation state; performing orthogonal decomposition on the evolved simulation state to obtain a decomposition result; reconstructing the decomposition result using an interpolation operator and a gradient compensation operator to generate a simulation signal; S42, constructing a constraint function set based on the simulation signal, the decomposition result and the optimized parameters; For each constraint function in the constraint function set, the constraint Euclidean distance and time-varying radius are calculated; based on the constraint Euclidean distance and time-varying radius, a variational functional including Lagrange function and constraint terms is constructed to generate the Euler-Lagrange equation; the Euler-Lagrange equation is solved to obtain a preliminary optimization trajectory; based on the preliminary optimization trajectory, a slack variable is introduced for constraint optimization to obtain an optimization trajectory; S43, based on the optimized trajectory and the simulation signal, construct a prediction model to generate a predicted state; based on the predicted state, construct a control vector, and optimize the control vector using a gradient projection method to obtain an optimized control vector; based on the optimized control vector, generate a preliminary control instruction using a sliding time domain method; based on the predicted state and the pre-stored actual state, calculate the tracking error; based on the tracking error, adjust the update frequency of the preliminary control instruction to generate a final control instruction; S44, inputting the control instruction, tracking error and analog signal into a pre-configured evaluation function set to calculate a predetermined performance index; performing weighted summation on each performance index to obtain a comprehensive score; The comprehensive score is processed using a preconfigured proportional gain matrix and integral gain matrix to output a feedback signal.

6. The method for dynamic simulation of radar target echo characteristics based on unmanned aerial vehicle according to claim 5, characterized in that: Step S11 is further as follows: S111, reading the environmental parameters of the system, performing numerical normalization processing, and obtaining normalized environmental parameters; based on the normalized environmental parameters, calculating the correlation coefficients thereof, and generating a correlation matrix; Based on the correlation matrix, the normalized environmental parameters are weighted and combined to generate an environmental parameter combination matrix; the environmental parameter combination matrix is ​​compared with a pre-stored reference threshold to generate an adjustment coefficient matrix; S112, based on the environmental parameter combination matrix and the adjustment coefficient matrix, calculate the gradient vector in the spatial dimension; perform time series sampling on the gradient vector to obtain the change rate in the time dimension; perform weighted fusion on the gradient vector and the change rate to construct a dynamic change feature matrix; based on the dynamic change feature matrix, calculate the weight coefficient of each dimension to generate a reference calculation matrix; Obtain a voltage vector, multiply the reference calculation matrix by the voltage vector, and output an initial dynamic reference value; S113, segmenting the initial dynamic reference value according to a preset time window, and calculating the statistical characteristics of each segment; based on the statistical characteristics, constructing an objective function, including a time continuity constraint term and a spatial consistency constraint term; The objective function is optimized and solved by an iterative method. The reference value is updated each time until the rate of change of the objective function is less than a preset threshold, and the optimized system reference value is output.

7. The method for dynamic simulation of radar target echo characteristics based on unmanned aerial vehicle according to claim 5, characterized in that: Step S23 is further as follows: S231, based on the target characteristic parameters and the system reference value, calculating the difference matrix; extracting the feature sequence from the pre-stored historical data, and performing a time series comparison between the feature sequence and the difference matrix to obtain a comparison result; based on the comparison result, calculating the first weight coefficient of each dimension; multiplying the first weight coefficient by the difference matrix to generate a weighted error matrix; Calculate the second norm of the weighted error matrix and construct the basic optimization objective function; S232, based on the basic optimization objective function, calculating the gradient value of the optimization parameter; multiplying the gradient value by the preset learning rate to obtain the parameter update amount; based on the parameter update amount, generating an updated parameter set; calculating the change rate of the updated parameter set to determine the gradient search direction; Based on the gradient search direction, the learning rate is adaptively adjusted and the adjusted learning rate is output; Based on the adjusted learning rate, the updated parameter set is adjusted to obtain an adjusted parameter set; S233, substituting the adjusted parameter set into the basic optimization objective function to calculate the new objective function value; and calculating the new and old difference values ​​based on the new objective function value and the basic optimization objective function; Compare the new and old difference values ​​with a preset threshold value. If the new and old difference value is greater than the preset threshold value, construct an adaptive correction item based on the adjusted parameter set, and fine-tune the adjusted parameter set based on the adaptive correction item to generate a corrected parameter set; calculate the stability index of the corrected parameter set; Otherwise, the adjusted parameter set is directly used as the calibrated parameter set; S234. Construct parameter fusion weights based on the corrected parameter set and stability index; extract basic statistical features of the pre-stored historical optimization parameter sequence; Based on the parameter fusion weight, the basic statistical features and the corrected parameter set are weighted combined to generate a fusion parameter set; Verify the effectiveness of the fusion parameter set and calculate the verification indicators; Based on the verification indicators, the final optimization parameters are screened.

8. The method for dynamic simulation of radar target echo characteristics based on unmanned aerial vehicle according to claim 5, characterized in that: Step S33 is further as follows: S331, performing singular value decomposition on the screened coupling strength matrix to extract the main eigenvectors; constructing state variables based on the main eigenvectors and determining the dimensions of the state variables; normalizing the state variables to generate standardized state vectors; calculating the autocorrelation matrix of the standardized state vectors to determine the main directions of the state space; setting the boundary conditions of the state space according to the main directions; S332, constructing a state transfer equation based on the standardized state vector; Calculate the eigenvalue of the state transfer equation, analyze the system stability, and obtain the stability analysis results; Based on the stability analysis result, the parameters of the state transfer equation are adjusted to obtain an adjusted state transfer equation; Perform piecewise linearization on the adjusted state transfer equation to obtain a piecewise transfer matrix; S333, based on the segmented transfer matrix, calculating the gain matrix of the predictor, including the proportional gain and the integral gain; based on the pre-stored system response characteristics, adjusting the parameters of the gain matrix to obtain an adjusted gain matrix; Based on the adjusted gain matrix, an initial predictor is constructed; Orthogonalize the parameter set of the initial predictor to obtain the final predictor; S334, based on the predictor, perform state prediction to obtain a predicted state sequence; compare the predicted state sequence with the actual state to calculate the prediction error; based on the prediction error, construct a compensation function, including a linear compensation term and a nonlinear compensation term; The compensation function is parameter optimized to generate the optimal compensation coefficient; based on the optimal compensation coefficient, the predicted state sequence is corrected to obtain the prediction result.

9. The method for dynamic simulation of radar target echo characteristics based on unmanned aerial vehicle according to claim 5, characterized in that: Step S43 is further as follows: S431. Based on the optimized trajectory, a prediction model is constructed to calculate the state prediction value; based on the simulation signal, the state prediction value is corrected to generate a corrected prediction sequence; the statistical characteristics of the corrected prediction sequence are calculated to extract the prediction trend; Based on the forecast trend, the state transition probability matrix is ​​constructed to generate multi-step forecast results; S432, constructing an initial value of a control vector based on the multi-step prediction result; Calculate the sensitivity coefficient of each component in the initial value; determine the optimization weight based on the sensitivity coefficient; Based on the optimization weights, a first objective function including state prediction error and control amount constraints is constructed; Based on the first objective function, the initial value is optimized by using the gradient projection method to obtain the optimized control vector; S433, segmenting the optimized control vector according to the preset control time window, calculating the local optimal control amount for each control time window; determining the time domain sliding step according to the change rate of the local optimal control amount; updating the control time window using the time domain sliding step to generate a new control sequence; performing a causal test on the control sequence to obtain a preliminary control instruction; S434, based on the preliminary control instruction, perform anti-disturbance analysis and calculate the stability margin; according to the stability margin, limit the amplitude of the preliminary control instruction to obtain a limited control instruction; Performing a timing correlation test on the restricted control instructions to obtain the tested control instructions; calculating the consistency index based on the tested control instructions; Based on the consistency index, the checked control instructions are screened to obtain the screened control instructions; The screened control instructions are subjected to security verification to obtain the final control instructions.

10. The method for dynamic simulation of radar target echo characteristics based on unmanned aerial vehicle according to claim 5, characterized in that: Step S44 is further as follows: S441, calculating first statistical features respectively based on the control instruction, the tracking error and the analog signal; Determine an evaluation weight for each first statistical feature and construct a weighted evaluation function; based on the weighted evaluation function, calculate the time cumulative value of each evaluation indicator and generate a historical performance curve; based on the historical performance curve, extract key performance points; Classify and count key performance points to form a set of performance indicators; S442. Calculate the relative importance of each indicator based on the performance indicator set; Based on the relative importance, a scoring matrix is ​​constructed; Normalizing the scoring matrix to obtain a normalized scoring matrix; calculating the weighted score of the normalized scoring matrix to generate a comprehensive score; S443, performing time series analysis on the comprehensive scores to obtain the score change trend; Based on the score change trend, determine the gain parameters, including proportional gain and integral gain; Calculate the stability boundary of the gain parameter, and based on the stability boundary, optimize the gain parameter online to obtain the optimized gain parameter; based on the optimized gain parameter, construct a feedback gain calculation model; Input the comprehensive score into the feedback gain calculation model to obtain a feedback signal; S444, calculating the time delay characteristic of the feedback signal, performing phase compensation, and obtaining a compensated feedback signal; The compensated feedback signal is bandwidth-limited to obtain a final feedback signal.

Images