A Fast Calculation Method for Electrical Performance of Airborne Phased Array Antennas Based on CatBoost Proxy Model
By using the CatBoost proxy model and combining multiple algorithms, a load-electrical performance mapping model for airborne phased array antennas was constructed. This solved the problems of low computational efficiency and ambiguity of load-performance coupling mechanism in traditional methods, and achieved efficient and accurate electrical performance calculation and evaluation.
Patent Information
- Application Number
- CN202611214202.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-11
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies for airborne phased array antennas face feeding amplitude and phase errors and spatial phase distortions caused by environmental load disturbances during service. Traditional Monte Carlo methods have low computational efficiency and fail to effectively map the load-performance coupling mechanism.
The CatBoost surrogate model is adopted, combined with Latin hypercube sampling, Bayesian optimization, kernel density estimation and class gradient boosting tree algorithm, to construct surrogate models of thermal load-temperature field and mechanical load-structural deformation field, and the electrical performance is calculated by quasi-Monte Carlo method.
It enables efficient and accurate calculation of the electrical performance of airborne phased array antennas in complex service environments, significantly improving computational efficiency and accuracy, and providing scientific support for performance robustness assessment and optimization design.
Smart Images

Figure CN123069872A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of radar antenna technology, specifically relating to a method for rapid calculation of the electrical performance of airborne phased array antennas based on the CatBoost surrogate model. Background Technology
[0002] As the core of modern electronic equipment, active phased array antennas are rapidly developing towards higher frequencies, higher integration, and lighter weight. However, airborne radars face extremely harsh environmental load disturbances during service, including acceleration overload, half-sine shock, random vibration, and high heat dissipation. The structural deformation and temperature fields induced by these uncertain loads directly cause feed amplitude and phase errors and spatial phase distortion in the antenna, severely restricting the electrical performance robustness of high-performance phased array antennas in complex environments.
[0003] Although the Monte Carlo method is widely used for statistical analysis of antenna errors, traditional methods have two major bottlenecks: First, the efficiency of acquiring physical fields is extremely low. Traditional numerical simulation (finite element analysis) cannot withstand the computational overhead when faced with tens of thousands of random samples. Second, the load-performance coupling mechanism is ambiguous. Existing studies mostly model array element errors directly, ignoring the direct mapping relationship between the underlying environmental load and the top-level electrical performance, and do not fully consider the impact of uneven distribution of sampled data on model accuracy. Summary of the Invention
[0004] The purpose of this invention is to overcome the problem of low efficiency in statistical analysis of antenna errors using the Monte Carlo method. It proposes a fast calculation method for the electrical performance of airborne phased array antennas based on the CatBoost surrogate model. This method delves into the impact of load uncertainty on electrical performance, avoids a large amount of time-consuming and difficult simulation analysis, and greatly improves the calculation efficiency. It can quickly realize the calculation of the electrical performance of phased array antennas under the influence of load uncertainty in the service environment.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for fast calculation of the electrical performance of airborne phased array antennas based on the CatBoost surrogate model, comprising the following steps: The design parameters of the phased array antenna are determined, and a finite element model of the phased array antenna is established based on the design parameters. The range of values for thermal and mechanical loads under service conditions is determined, and multiple sets of load samples are generated within the range using the Latin hypercube sampling method. Each set of load samples is applied to the finite element model for simulation to obtain array surface temperature field data and structural deformation field data, and an initial physical field dataset is constructed. Adaptive downsampling based on the physical field spatial gradient is performed on the initial physical field dataset to obtain a representative dataset; the optimal hyperparameter combination of the class gradient boosting tree algorithm is determined by the Bayesian optimization algorithm; the weight of each sample in the representative dataset is calculated by the kernel density estimation method to construct a weighted loss function; based on the optimal hyperparameter combination and the weighted loss function, the class gradient boosting tree algorithm is used to train the thermal load-temperature field proxy model and the mechanical load-structural deformation field proxy model. An electromechanical coupling model of a phased array antenna is constructed. The temperature error output by the temperature field proxy model and the array element position error output by the structural deformation field proxy model are substituted into the electromechanical coupling model to obtain the load-electromagnetic coupling model. The quasi-Monte Carlo method is used to generate multiple sets of uncertain load samples within the value range. Each set of uncertain load samples is input into the load-electromagnetic coupling model, and the probabilistic characteristics of the phased array antenna's electrical performance are statistically obtained.
[0006] Furthermore, adaptive downsampling based on the physical field spatial gradient is performed on the initial physics dataset to obtain a representative dataset, including: Based on the physical field dataset, the node regions are divided according to the physical field spatial gradient magnitude. All nodes in the high gradient region are retained, and nodes in the low gradient region are sampled and simplified to obtain a representative dataset. The formula for calculating the magnitude of the spatial gradient of a physical field is shown below:
[0007] in, The partial derivatives are calculated by the physical field spatial gradient between nodes and their adjacent nodes.
[0008] Furthermore, the optimal combination of hyperparameters for the categorical gradient boosting tree algorithm is determined using a Bayesian optimization algorithm, including: The weighted root mean square error of the surrogate model on the cross-validation set is used as the objective function. The optimal combination of hyperparameters is obtained by iterative optimization through Gaussian process modeling and expected improvement criteria.
[0009] Furthermore, the kernel density estimation method is used to calculate the weights of each sample in the representative dataset, and a weighted loss function is constructed, including: The probability density of physical field output features in representative datasets is obtained by kernel density estimation. Smaller weights are assigned to high-density samples and larger weights are assigned to low-density samples. A weighted loss function is constructed based on the weights of each sample. The weighted loss function is shown in the following formula:
[0010]
[0011]
[0012] in, Automatic bandwidth parameters, The total number of samples, For density, These are weighting coefficients. It is a regulating factor.
[0013] Furthermore, mechanical loads include acceleration overload, half-sine impact, and random vibration; The mechanical load-structural deformation field proxy model receives multiple types of mechanical load parameters as input and outputs surface structural deformation field data.
[0014] Furthermore, an electromechanical coupling model of the phased array antenna is constructed, including: A triangular grating array radiation pattern model of a phased array antenna is constructed. The temperature error is converted into the amplitude and phase error of the array element excitation. The array element position error is substituted into the triangular grating array radiation pattern model to obtain the electromechanical coupling model.
[0015] Furthermore, the probabilistic characteristics of the electrical performance of the control array antenna include the mean, variance, and upper and lower bounds of the electrical performance indicators.
[0016] Secondly, this invention provides a system for fast calculation of the electrical performance of airborne phased array antennas based on the CatBoost surrogate model, comprising: The initial physics dataset construction module is used to determine the design parameters of the phased array antenna and establish a finite element model of the phased array antenna based on the design parameters; determine the value range of thermal load and mechanical load under service environment, and generate multiple sets of load samples within the value range using the Latin hypercube sampling method; apply each set of load samples to the finite element model for simulation to obtain array surface temperature field data and structural deformation field data, and construct the initial physics dataset. The surrogate model training module is used to perform adaptive downsampling based on the physical field spatial gradient on the initial physical field dataset to obtain a representative dataset; the optimal hyperparameter combination of the class gradient boosting tree algorithm is determined by the Bayesian optimization algorithm; the weight of each sample in the representative dataset is calculated by the kernel density estimation method to construct a weighted loss function; based on the optimal hyperparameter combination and the weighted loss function, the class gradient boosting tree algorithm is used to train the thermal load-temperature field surrogate model and the mechanical load-structural deformation field surrogate model. The coupling model construction module is used to construct the electromechanical coupling model of the phased array antenna. The temperature error output by the temperature field proxy model and the array element position error output by the structural deformation field proxy model are substituted into the electromechanical coupling model to obtain the load-electromagnetic coupling model. The probability characteristic statistics module is used to generate multiple sets of uncertain load samples within the value range using the quasi-Monte Carlo method. Each set of uncertain load samples is input into the load-electromagnetic coupling model to statistically obtain the probabilistic characteristics of the phased array antenna's electrical performance.
[0017] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a method for fast calculation of the electrical performance of an airborne phased array antenna based on the CatBoost proxy model.
[0018] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for rapidly calculating the electrical performance of an airborne phased array antenna based on a CatBoost proxy model.
[0019] Compared with the prior art, the present invention has the following beneficial technical effects: This invention proposes a rapid calculation method for the electrical performance of airborne phased array antennas based on the CatBoost surrogate model, enabling efficient and accurate calculation of the electrical performance of airborne phased array antennas under loads in complex service environments. Through adaptive downsampling based on the physical field spatial gradient, the method significantly reduces the dataset size while fully preserving key physical features of sensitive areas, effectively reducing the computational cost of training the surrogate model. The surrogate model is constructed using a category gradient boosting tree combining Bayesian hyperparameter optimization and a kernel density weighted loss function, which improves the fitting bias problem caused by uneven data distribution and enhances the reconstruction accuracy and model generalization ability under extreme conditions. By constructing a load-electromagnetic coupling model, a complete mapping link from thermal loads and mechanical loads to antenna electrical performance is established, accurately reflecting the antenna electrical performance response under multi-load coupling. Uncertainty analysis using a quasi-Monte Carlo method significantly reduces computation time while ensuring statistical accuracy, providing efficient technical support for the performance robustness assessment and optimization design of airborne phased array antennas.
[0020] Furthermore, this invention provides a method for rapid calculation of the electrical performance of airborne phased array antennas based on the CatBoost surrogate model. The core technical path is as follows: Simplification of physical field features and high-quality sampling: For the complex load response space, an adaptive downsampling strategy based on physical space gradients is first adopted to retain the complete information of large gradients and key feature points in the original dataset with a very small data scale (5%~10%). Construction of a high-precision, robust surrogate model: A Bayesian optimization algorithm is introduced to perform global hyperparameter optimization of the CatBoost class gradient boosting tree, locking in the optimal model architecture; to address the problem of uneven distribution of sampled data, an innovative kernel density estimation (KDE) is used to design an inverse density-weighted loss function, giving higher training weights to extreme and sparse conditions, and constructing a high-precision load-structure deformation / temperature mapping surrogate model. Reconstruction of the electromechanical coupling model: For the triangular grid arrangement characteristics, an electromechanical coupling mathematical model is established based on the basis vector analysis method. This invention integrates the spatial displacement deviation of array elements induced by three mechanical conditions (acceleration, impact, and random vibration) with the feed amplitude and phase drift induced by one thermal condition into a coupled model in real time, achieving deep fusion from the bottom-level load to the top-level electromagnetic characteristics. Efficient uncertainty quantification analysis: Based on a trained fast proxy model, a quasi-Monte Carlo method is used to simulate large-scale random load samples. By avoiding the time-consuming traditional finite element simulation, this invention can calculate the electrical performance response of all samples within milliseconds. Through statistical mean, variance, and performance bounds, the performance robustness of the antenna under service conditions is evaluated. Through the above algorithm system, this invention achieves cross-domain mapping and second-level evaluation from service environment loads to electrical performance indicators, providing scientific support for the rapid optimization design of high-performance airborne phased array antennas under uncertain service environments. Attached Figure Description
[0021] The accompanying drawings described herein are for illustrative purposes only and are not intended to limit the scope of the invention in any way. Furthermore, the shapes and proportions of the components in the drawings are merely schematic to aid in understanding the invention and do not specifically limit the shapes and proportions of the components. In the drawings: Figure 1 This is a flowchart of the fast calculation method for the electrical performance of airborne phased array antennas based on the CatBoost surrogate model of this invention.
[0022] Figure 2 This is a simplified structural diagram of the airborne phased array antenna electrical performance fast calculation system based on the CatBoost surrogate model of this invention.
[0023] Figure 3 This is an electronic device diagram of the fast calculation method for the electrical performance of airborne phased array antennas based on the CatBoost surrogate model of this invention.
[0024] Figure 4This is a flowchart of a method for fast calculation of the electrical performance of a phased array antenna based on the CatBoost surrogate model in an embodiment of the present invention.
[0025] Figure 5 This is a schematic diagram of the array element arrangement of the triangular grid planar array in an embodiment of the present invention.
[0026] Figure 6 This is a temperature field reconstruction cloud map in an embodiment of the present invention.
[0027] Figure 7 This is a reconstruction cloud map of displacement deformation in an embodiment of the present invention.
[0028] Figure 8 As described in the embodiments of the present invention The radiation pattern of the phased array antenna.
[0029] Figure 9 As described in the embodiments of the present invention The radiation pattern of the phased array antenna. Detailed Implementation
[0030] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0031] Example 1 See Figure 1 A fast calculation method for the electrical performance of airborne phased array antennas based on the CatBoost surrogate model includes the following steps: The design parameters of the phased array antenna are determined, and a finite element model of the phased array antenna is established based on the design parameters. The range of values for thermal and mechanical loads under service conditions is determined, and multiple sets of load samples are generated within the range using the Latin hypercube sampling method. Each set of load samples is applied to the finite element model for simulation to obtain array surface temperature field data and structural deformation field data, and an initial physical field dataset is constructed. Adaptive downsampling based on the physical field spatial gradient is performed on the initial physical field dataset to obtain a representative dataset; the optimal hyperparameter combination of the class gradient boosting tree algorithm is determined by the Bayesian optimization algorithm; the weight of each sample in the representative dataset is calculated by the kernel density estimation method to construct a weighted loss function; based on the optimal hyperparameter combination and the weighted loss function, the class gradient boosting tree algorithm is used to train the thermal load-temperature field proxy model and the mechanical load-structural deformation field proxy model. An electromechanical coupling model of a phased array antenna is constructed. The temperature error output by the temperature field proxy model and the array element position error output by the structural deformation field proxy model are substituted into the electromechanical coupling model to obtain the load-electromagnetic coupling model. The quasi-Monte Carlo method is used to generate multiple sets of uncertain load samples within the value range. Each set of uncertain load samples is input into the load-electromagnetic coupling model, and the probabilistic characteristics of the phased array antenna's electrical performance are statistically obtained.
[0032] This invention determines the overall dimensions, row spacing, column spacing, number of antenna elements, antenna frequency, and material parameters of a phased array antenna based on its overall functional requirements and performance indicators. A three-dimensional solid model and corresponding finite element model of the phased array antenna are established based on these design parameters. The upper and lower bounds of thermal conditions, acceleration overload, half-sine impact load, and random vibration load under service conditions are determined according to standard environmental testing methods and engineering experience. Latin hypercube sampling is used to uniformly sample within the intervals. A heat source and equivalent fixing bolts are placed on the phased array antenna, and different loads are applied to perform finite element thermal simulation and random vibration simulation to calculate the array surface temperature and structural deformation distribution. The spatial gradient of the physical field at the antenna array nodes is calculated, and an adaptive downsampling strategy is used to extract deformation and temperature-sensitive feature points, constructing a representative dataset to achieve data simplification and key feature preservation. Based on a Bayesian optimization algorithm, the fitting accuracy of the surrogate model on the validation set is used as the objective function. Gaussian process modeling and iteration are used to automatically optimize and process the hyperparameter space (including learning rate, tree depth, regularization coefficient, and number of leaf nodes) of the class gradient boosting tree algorithm globally to determine the optimal combination of hyperparameters. The probability density of temperature and deformation is calculated based on kernel density estimation, assigning smaller weights to high-density samples and higher weights to low-density samples in the dataset, and designing a weighted loss function for the surrogate model. Based on the weighted loss function from the class gradient boosting tree algorithm, a thermal power-temperature surrogate model and an array deformation field surrogate model integrating multi-dimensional loads such as acceleration overload, half-sine impact, and power spectral density are trained. The temperature error and structural deformation error calculated by the surrogate model are incorporated into the electromechanical coupling model of the active phased array antenna for electrical performance analysis, resulting in a load-electromagnetic coupling model and establishing a mapping relationship between load and electrical performance. Using a quasi-Monte Carlo method, a large number of uncertain input loads are randomly sampled, and the probabilistic characteristics of the phased array antenna's electrical performance are statistically analyzed based on the load-electromagnetic coupling model. This invention improves the imbalance problem of temperature and deformation data based on the category gradient boosting tree surrogate model and kernel density estimation method, establishes the mapping relationship between load and phased array antenna electrical performance, and realizes the rapid calculation of phased array antenna electrical performance considering the uncertainty of load in service environment.
[0033] See Figure 4 A method for rapid calculation of the electrical performance of a phased array antenna, with the following specific steps: Step 1: Determine the overall dimensions of the phased array antenna, the row spacing between array elements, the column spacing, the number of rows and columns of antenna elements, the antenna frequency, and the material parameters; The length of the phased array antenna is determined based on the overall functional requirements and performance specifications of the phased array antenna. L ,Width W ,thickness H Array element row spacing dx Column spacing dy Number of antenna element rows M Column number N Antenna frequency f The elastic modulus, Poisson's ratio, density, and thermal conductivity of aluminum alloy materials are as follows:
[0034] Step 2: Based on the design parameters, establish a three-dimensional solid model and a corresponding finite element model of the phased array antenna; A three-dimensional solid model of the phased array antenna was created in SOLIDWORK based on the design parameters. The model was then imported into Hypermesh for mesh generation. The elastic modulus, Poisson's ratio, density, and thermal conductivity of the aluminum alloy were defined. The solid model was then converted to finite element method using SOLID185 elements.
[0035] Step 3: Determine the upper and lower bounds of the service environment load uncertainties and implement Latin hypercube sampling loads; based on the equipment's standard environmental testing methods and engineering experience, determine the upper and lower bounds of four loads: acceleration overload, half-sine shock, random vibration, and thermal power, and construct a multi-initial training sample space using Latin hypercube sampling. The specific process is as follows: 3.1. Definition of Steady-State Acceleration Overload Uncertainty: Based on the vehicle maneuver envelope curve, the steady-state acceleration overload range is set to 4g to 12g. To ensure the subsequent surrogate model's sensitivity to load fluctuations, the load identification step size is set to 0.2g.
[0036] Sampling Implementation: The parameter space [4g, 12g] is divided into 40 levels. A sampling point is randomly selected in each subspace, ensuring that the sampled values meet the resolution requirement of 0.2g after processing, and a uniformly distributed acceleration sampling sequence is generated.
[0037] 3.2. Definition of uncertainty in half-sine impact load: The pulse duration of the half-sine impact condition is fixed at 11ms. The uncertainty range of the peak impact acceleration is set to 20g to 30g, with a step size of 0.2g.
[0038] Sampling Implementation: Latin hypercube sampling was performed within the peak range of [20g, 30g]. The range was divided into 50 non-overlapping sub-ranges, and a peak sample was extracted from each sub-range to simulate the magnitude fluctuation of the impact load under extreme service conditions.
[0039] 3.3 Multidimensional PSD spatial modeling of random vibration conditions: According to the random vibration test specifications, seven key frequency center points were determined: 20Hz, 80Hz, 200Hz, 350Hz, 500Hz, 1000Hz, and 2000Hz.
[0040] Uncertainty definition: A 7-dimensional random vibration parameter space is constructed with the standard amplitude of the power spectral density (PSD) corresponding to each frequency point as the center and the allowable deviation range set to ±20%.
[0041] Sampling implementation: For the amplitude range under each frequency point, the system is independently divided into 50 equal parts. Multi-dimensional combination sampling is performed using the LHS criterion to ensure that the generated vibration samples can cover various random combinations of energy distribution in each frequency band.
[0042] 3.4 Definition of thermal operating condition uncertainty: The range of uncertainty of antenna array thermal power is set to 6W to 10W.
[0043] Sampling Implementation: The power consumption space [6W, 10W] is divided into 50 subspaces, each with a subspace length of 0.08W. A random sampling point is generated within each subspace.
[0044] Step 4: Thermal simulation and random vibration simulation to calculate temperature and deformation distribution; Based on the established high-fidelity finite element model of the phased array antenna constrained by the heat source and equivalent fixed bolts, samples of various working conditions obtained in step 3 were applied in batches (covering steady-state acceleration of 4g-12g, half-sine impact with peak value of 20g-30g and lasting for 11ms, random vibration at 7-dimensional frequency points, and thermal power of 6W-10W). By performing steady-state thermal simulation, static displacement simulation, transient dynamic analysis, and spectrum analysis in parallel, the equilibrium temperature distribution and three-dimensional structural displacement deviation of the array nodes were calculated and extracted. The data were then summarized to form the antenna array temperature dataset and the structural deformation dataset, thereby establishing a high-fidelity physical field response sample library, which provides data support for subsequent adaptive sampling and surrogate model training based on physical space gradient.
[0045] Step 5: Simplification and construction of representative datasets based on the physical space gradient adaptive sampling strategy; 5.1 First, extract the physical field data (temperature field) of each node of the phased array antenna surface. and structural displacement field ), define the node set as Each node contains three-dimensional spatial coordinates and the corresponding physical response value.
[0046] 5.2 Calculate the spatial gradient magnitude of the physical field at each node to characterize the degree of drastic change in the physical characteristics of that region. For any node... Its spatial gradient The calculation is as follows
[0047] The partial derivatives are obtained by calculating the physical field difference between a node and its neighboring nodes.
[0048] 5.3 Introducing an adaptive downsampling function The retention probability or sampling weight of a node is determined based on the magnitude of the gradient. A gradient threshold is set. Constructing a criterion for sensitive feature extraction: like If the point is identified as a sensitive feature point (such as a hotspot edge or a sudden deformation point), it will be retained.
[0049] like If the point is found to be flat and redundant, it will be randomly removed based on the density of the local area.
[0050] 5.4 Constructing a representative dataset The threshold is adjusted through multiple rounds of iteration. until the extracted feature point set Able to achieve a very small number of secretaries (of the source dataset) The characteristic distribution of the original physical field is reconstructed, and the dataset is finally simplified.
[0051] Step 6: Global optimization and configuration of hyperparameters for the surrogate model based on Bayesian optimization algorithm; 6.1 First, determine the set of hyperparameters to be optimized. In this invention, the core optimization parameters include: learning rate. tree depth L2 regularization coefficient Random intensity and bagging sampling rate .
[0052] Set the objective function The weighted root mean square error (W-RMSE) of the surrogate model on the K-fold cross-validation machine:
[0053] in Let be the weight coefficient of the k-th fold. The ultimate goal of optimization is to find the optimal combination. Make the scalar function reach its global minimum:
[0054] 6.2 Establishing a Gaussian process prior model: This invention uses a Gaussian process (GP) as the response surface model to approximate the objective function. Assume the prior distribution of the wooden plaque function follows:
[0055] in For predicting the mean function; Let be the covariance function.
[0056] For projects with predictions The prediction results are expressed as follows:
[0057] Its uncertainty estimate is expressed as:
[0058] in The covariance matrix between observation points, To observe the noise variance.
[0059] 6.3. Introducing the Expected Improvement (EI) criterion to balance search efficiency, the EI function automatically selects parameter regions with high performance potential or high uncertainty for testing:
[0060] The Gaussian process model is continuously updated through iterative iterations until the optimal combination of hyperparameters that best generalizes the model is found within a set number of iterations. .
[0061] Step 7: Design of sample weighted loss function based on kernel density estimation; 7.1. Sample corresponding interval density modeling: Using a Gaussian kernel function, perform probability density scanning on the output features (epoch node temperature or displacement) in the dataset to calculate the local density of each sample point in the feature space. :
[0062] in Automatic bandwidth parameters, This represents the total number of samples. The lower the density value, the rarer the working condition is, or the more volatile it is.
[0063] 7.2 Inverse Density Weight Calculation and Normalization: To enhance the model's learning of rare and critical operating conditions, an inverse density weight function is designed. First, the density... Mapped to The interval is then used to calculate the weighting coefficients. :
[0064] in This is an adjustment factor used to control the strength of weight allocation. This formula allows information-rich points with lower density to receive higher weights, while suppressing the weights of redundant points.
[0065] 7.3 Construct a weighted loss function, inject the calculated sample weight vector into the model training process, and redefine the loss function. This loss function will directly determine the direction of weak tree growth in the subsequent class gradient boosting tree algorithm.
[0066] In this way, the surrogate model can maintain extremely high fitting accuracy under all operating conditions (especially extreme conditions) and avoid being overwhelmed by a large amount of redundant data.
[0067] Step 8: Training and physical field reconstruction of the weighted proxy model based on CatBoost (Category Gradient Boosting Tree); Using a weighted loss function and a categorical gradient boosting tree algorithm as input, three mechanical surrogate models and one thermodynamic surrogate model are trained. The final surrogate model can achieve the following: Input: Load condition parameters (thermal power, acceleration load, pulse time, pulse peak value, vibration PSD, etc.) and node coordinates X, Y, and Z.
[0068] Output: Temperature field distribution and structural displacement deformation field .
[0069] like Figure 6 , Figure 7 The figures shown are the temperature field reconstruction contour map and the displacement deformation reconstruction contour map, respectively. Step 9: Construction of load-electromagnetic coupling model and electrical performance analysis based on surrogate model; 9.1. Array geometric model based on triangular grating: The array element arrangement type of the phased array antenna is a triangular grating planar array antenna. See [reference needed]. Figure 5 Based on step 2, the element row spacing of the phased array antenna is obtained. dx Column spacing dy Number of antenna element rows M Column number N Antenna frequency f ,.along x The array element in the direction is denoted as the first. m Array element, along yThe array element in the direction is denoted as the first. n The number of array elements is [number], and the total number of array elements is [number]. M × N There are [number] triangular grids, with an angle of inclination of [value]. Then the array vector is represented as:
[0070]
[0071] No. The center of each array element relative to the reference array element The ideal position vector is represented as:
[0072] Therefore, the first The coordinates of each array element are represented as follows:
[0073]
[0074] 9.2 The radiated signals of each element of the array antenna are vector-superimposed in the far-field region, and at the observation point... The direction forms the array field strength pattern. The array field strength pattern is represented as:
[0075] in For array element pattern functions; The amplitude of the excitation current of the array element. For the excitation phase of the array element, For free space wavenumber, ,in This is the operating wavelength.
[0076] The unit vector for the direction of observation is represented as: , , They are respectively with The angle along the positive direction of the axis. They are The element vectors of the direction, and their direction cosines satisfy:
[0077] After substituting the element coordinates, the ideal array pattern is represented as:
[0078] 9.3 Temperature Load-Induced Electrical Performance Drift: Temperature changes affect the amplitude and phase of the excitation current of the T / R module. Considering the effect of temperature on the amplitude of the excitation current of the array elements, the corrected excitation current is expressed as follows:
[0079] For array element temperature; This represents the normalized amplitude drift.
[0080] Based on the T / R component test data, the effect of temperature on amplitude drift is expressed as follows.
[0081]
[0082] Temperature changes simultaneously introduce element excitation phase drift, the corrected phase of which is expressed as:
[0083] Temperature-induced phase shift is expressed as:
[0084] 9.4 Structural displacement modeling under multiple mechanical load conditions, considering the structural displacement response under the combined action of acceleration overload, half-sine impact, and random vibration. The surrogate model of the multi-load condition structural response obtained in step S8 is represented as follows:
[0085] in To accelerate overload; The peak value is half-sine wave. The pulse width is half-sine. The power spectral density of random vibrations; It represents the random vibration frequency.
[0086] No. The actual position of each element under load is represented as follows:
[0087] 9.5 Electromechanical-thermal coupling pattern modeling: The thermal power-temperature proxy model trained in step 8 is represented as follows:
[0088] in This represents the thermal load power.
[0089] Define the environmental load vector as follows:
[0090] The electromechanical-thermal coupling pattern considering the combined effects of temperature drift and structural deformation is represented as follows:
[0091] in Reference array elements The corresponding three-phase displacements, These are the predicted results of the three-dimensional displacement of the array elements under environmental loads.
[0092] The electromechanical-thermal multi-field coupling model driven by the above proxy model enables rapid prediction and evaluation of the main lobe gain, side lobe level, and beam pointing accuracy of active phased array antennas under complex service environment loads.
[0093] Step 10: Uncertainty statistical analysis and performance evaluation based on the quasi-Monte Carlo method; Large-scale, low-bias random sample points are generated within the multidimensional uncertain load space (covering acceleration overload, half-sine shock, random vibration, and thermal power) determined in step 3 using the quasi-Monte Carlo method. These sample points are then batch-substituted into the CatBoost fast proxy model constructed in step 8 and the electromechanical coupling equations in step 9, completing the full sample response calculation within milliseconds. Statistical analysis of the output results extracts the mean, variance, and probability density distribution of key electrical performance indicators such as antenna gain, sidelobe level, and pointing accuracy. Figure 8 , Figure 9 As shown, this allows for the accurate acquisition of the upper and lower bounds of the electrical performance of the phased array antenna under complex service load disturbances, providing a quantitative basis for antenna reliability design and performance robustness assessment.
[0094] This invention achieves deep coupling prediction of the entire link between load, physical field, and electromagnetics under complex service environments. It delves into the impact mechanism of uncertainties in four typical load types—acceleration overload, half-sine shock, random vibration, and thermal power consumption—on electrical performance. For the characteristics of triangular grating arrangements, a high-precision electromechanical coupling model is established based on the basis vector analysis method, transforming the non-uniform deformation field and temperature field reconstructed by the surrogate model into feed amplitude-phase drift and spatial phase distortion in real time. Compared to traditional methods that directly model array element errors, this invention achieves a direct mapping from the underlying environmental load to the top-level electromagnetic characteristics, providing more realistic physical support for the performance evaluation of phased array antennas under extreme service environments.
[0095] This invention constructs a highly efficient surrogate model system based on data simplification and model refinement, significantly improving computational efficiency and accuracy. Addressing the high cost of traditional numerical simulation, this invention proposes a fast computational framework integrating multiple algorithms: At the sampling level: Through an adaptive downsampling strategy based on physical space gradients, key feature information is preserved with a very small data scale, reducing the computational load; At the modeling level: A Bayesian-optimized, automatically locked optimal CatBoost architecture is employed, and a kernel density estimation (KDE) weighted loss function is innovatively introduced to overcome the fitting bias caused by uneven distribution of sampled data. Compared with traditional full-scale finite element simulation, this invention reduces the computation time per cycle from hours to milliseconds while maintaining the same prediction accuracy, significantly improving design iteration efficiency.
[0096] This invention utilizes an efficient uncertainty quantification mechanism to achieve rapid and accurate assessment of electrical performance robustness. Addressing the slow convergence of the Monte Carlo method, this invention employs a quasi-Monte Carlo method (QMC) for low-bias sampling within a multidimensional load boundary space. Based on a pre-trained, highly robust surrogate model, this invention can quickly complete batch evaluations of large-scale samples, accurately calculating the mean, variance, and performance boundaries of indicators such as antenna gain, sidelobe level, and pointing error. This provides an efficient quantitative means for identifying environmental load-sensitive points and assessing the reliability of phased array antennas under uncertain service environments.
[0097] Example 2 See Figure 2 A fast calculation system for the electrical performance of airborne phased array antennas based on the CatBoost surrogate model, including: The initial physics dataset construction module is used to determine the design parameters of the phased array antenna and establish a finite element model of the phased array antenna based on the design parameters; determine the value range of thermal load and mechanical load under service environment, and generate multiple sets of load samples within the value range using the Latin hypercube sampling method; apply each set of load samples to the finite element model for simulation to obtain array surface temperature field data and structural deformation field data, and construct the initial physics dataset. The surrogate model training module is used to perform adaptive downsampling based on the physical field spatial gradient on the initial physical field dataset to obtain a representative dataset; the optimal hyperparameter combination of the class gradient boosting tree algorithm is determined by the Bayesian optimization algorithm; the weight of each sample in the representative dataset is calculated by the kernel density estimation method to construct a weighted loss function; based on the optimal hyperparameter combination and the weighted loss function, the class gradient boosting tree algorithm is used to train the thermal load-temperature field surrogate model and the mechanical load-structural deformation field surrogate model. The coupling model construction module is used to construct the electromechanical coupling model of the phased array antenna. The temperature error output by the temperature field proxy model and the array element position error output by the structural deformation field proxy model are substituted into the electromechanical coupling model to obtain the load-electromagnetic coupling model. The probability characteristic statistics module is used to generate multiple sets of uncertain load samples within the value range using the quasi-Monte Carlo method. Each set of uncertain load samples is input into the load-electromagnetic coupling model to statistically obtain the probabilistic characteristics of the phased array antenna's electrical performance.
[0098] This invention determines the overall size, row spacing, column spacing, number of antenna elements, antenna frequency, and material parameters of a phased array antenna based on its overall functional requirements and performance indicators. A three-dimensional solid model and corresponding finite element model of the phased array antenna are established based on these design parameters. The upper and lower bounds of thermal conditions, acceleration overload, half-sine impact, and random vibration loads under service conditions are determined according to standard environmental testing methods and engineering experience, and Latin hypercube sampling is used for uniform sampling within the intervals. Heat sources and equivalent fixing bolts are placed on the phased array antenna, and finite element thermal simulation and random vibration simulation are performed to calculate the temperature and structural deformation distribution of the phased array antenna surface. Based on the spatial gradient distribution of the physical field of the antenna surface, an adaptive downsampling strategy is adopted to accurately capture sensitive feature points of deformation and temperature, significantly reducing the size of the antenna. While maintaining the scale of the dataset, key physical response information is preserved. A kernel density estimation method is used to calculate the probability density of temperature and deformation, assigning smaller weights to samples with high density and higher weights to samples with low density, and designing a surrogate model weighted loss function. A thermal power-temperature surrogate model is trained by integrating Bayesian hyperparameter optimization and the class gradient boosting tree algorithm, combined with a reweighted loss function, as well as an array deformation field surrogate model integrating multi-dimensional loads such as acceleration overload, half-sine impact, and power spectral density. Based on the electromechanical coupling model of the active phased array antenna, temperature and structural errors are incorporated into the electrical performance analysis, establishing a load-electromagnetic coupling model. A quasi-Monte Carlo method is used iteratively to randomly sample uncertain input loads, and based on the load-electromagnetic coupling model, the statistical characteristics of the phased array antenna's electrical performance are obtained. This invention improves the imbalance problem of temperature and deformation data based on the class gradient boosting tree surrogate model and kernel density estimation method, establishes a mapping relationship between load and phased array antenna electrical performance, and realizes rapid calculation of the phased array antenna's electrical performance considering the uncertainties of loads in the service environment.
[0099] Example 3 See Figure 3 An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a method for fast calculation of the electrical performance of an airborne phased array antenna based on the CatBoost proxy model.
[0100] Example 4 A computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for rapid calculation of the electrical performance of an airborne phased array antenna based on a CatBoost proxy model.
[0101] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, read-only optical discs, optical storage, etc.) containing computer-usable program code.
[0102] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0103] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0104] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.
Claims
1. A method for fast calculation of the electrical performance of airborne phased array antennas based on the CatBoost surrogate model, characterized in that, Includes the following steps: Determine the design parameters of the phased array antenna, and establish a finite element model of the phased array antenna based on the design parameters; The range of values for thermal and mechanical loads under service conditions is determined, and multiple load samples are generated within the range using the Latin hypercube sampling method. The load samples of each group are applied to the finite element model for simulation to obtain surface temperature field data and structural deformation field data, and to construct an initial physical field dataset. Adaptive downsampling based on the physical field spatial gradient is performed on the initial physical field dataset to obtain a representative dataset; the optimal hyperparameter combination of the class gradient boosting tree algorithm is determined using the Bayesian optimization algorithm; the weight of each sample in the representative dataset is calculated using the kernel density estimation method to construct a weighted loss function; based on the optimal hyperparameter combination and the weighted loss function, the class gradient boosting tree algorithm is used to train the thermal load-temperature field proxy model and the mechanical load-structural deformation field proxy model. An electromechanical coupling model of the phased array antenna is constructed. The temperature error output by the temperature field proxy model and the array element position error output by the structural deformation field proxy model are substituted into the electromechanical coupling model to obtain the load-electromagnetic coupling model. Multiple sets of uncertain load samples are generated within the value range using the quasi-Monte Carlo method. Each set of uncertain load samples is input into the load-electromagnetic coupling model, and the probabilistic characteristics of the phased array antenna's electrical performance are statistically obtained.
2. The method for fast calculation of the electrical performance of airborne phased array antennas based on the CatBoost surrogate model according to claim 1, characterized in that, The process of performing adaptive downsampling based on the physical field spatial gradient on the initial physical field dataset to obtain a representative dataset includes: Based on the physical field spatial gradient magnitude of the initial physical field dataset, node regions are divided, all nodes in high gradient regions are retained, and nodes in low gradient regions are sampled and simplified to obtain the representative dataset. The formula for calculating the magnitude of the physical field spatial gradient is as follows: in, The partial derivatives are calculated by the physical field spatial gradient between nodes and their adjacent nodes.
3. The method for fast calculation of the electrical performance of airborne phased array antennas based on the CatBoost surrogate model according to claim 1, characterized in that, The process of determining the optimal hyperparameter combination for the categorical gradient boosting tree algorithm using Bayesian optimization includes: The weighted root mean square error of the surrogate model on the cross-validation set is used as the objective function, and the optimal hyperparameter combination is obtained by iterative optimization through Gaussian process modeling and expected improvement criteria.
4. The method for fast calculation of the electrical performance of airborne phased array antennas based on the CatBoost surrogate model according to claim 1, characterized in that, The step of calculating the weights of each sample in the representative dataset using kernel density estimation and constructing a weighted loss function includes: The probability density of the physical field output features in the representative dataset is obtained based on the kernel density estimation method. Smaller weights are assigned to high-density samples and larger weights are assigned to low-density samples. The weighted loss function is constructed based on the weights of each sample. The weighted loss function is shown in the following equation: in, Automatic bandwidth parameters, The total number of samples, For density, These are weighting coefficients. It is a regulating factor.
5. The method for fast calculation of the electrical performance of airborne phased array antennas based on the CatBoost surrogate model according to claim 1, characterized in that, The mechanical loads include acceleration overload, half-sine impact, and random vibration; The mechanical load-structural deformation field proxy model receives multiple types of mechanical load parameters as input and outputs surface structural deformation field data.
6. The method for fast calculation of the electrical performance of airborne phased array antennas based on the CatBoost surrogate model according to claim 1, characterized in that, The electromechanical coupling model for constructing the phased array antenna includes: A triangular grating array radiation pattern model of the phased array antenna is constructed. The temperature error is converted into the amplitude and phase error of the array element excitation. The array element position error is substituted into the triangular grating array radiation pattern model to obtain the electromechanical coupling model.
7. The method for fast calculation of the electrical performance of airborne phased array antennas based on the CatBoost surrogate model according to claim 1, characterized in that, The probabilistic characteristics of the electrical performance of the control array antenna include the mean, variance, and upper and lower bounds of the electrical performance indicators.
8. A system for fast calculation of the electrical performance of airborne phased array antennas based on the CatBoost surrogate model, characterized in that, include: An initial physics dataset construction module is used to determine the design parameters of the phased array antenna and to establish a finite element model of the phased array antenna based on the design parameters. The range of values for thermal and mechanical loads under service conditions is determined, and multiple load samples are generated within the range using the Latin hypercube sampling method. The load samples of each group are applied to the finite element model for simulation to obtain surface temperature field data and structural deformation field data, and to construct an initial physical field dataset. The surrogate model training module is used to perform adaptive downsampling based on the physical field spatial gradient on the initial physical field dataset to obtain a representative dataset; to determine the optimal hyperparameter combination of the class gradient boosting tree algorithm using a Bayesian optimization algorithm; to calculate the weight of each sample in the representative dataset using the kernel density estimation method and to construct a weighted loss function; and to train the thermal load-temperature field surrogate model and the mechanical load-structural deformation field surrogate model using the class gradient boosting tree algorithm based on the optimal hyperparameter combination and the weighted loss function. The coupling model construction module is used to construct the electromechanical coupling model of the phased array antenna. The temperature error output by the temperature field proxy model and the array element position error output by the structural deformation field proxy model are substituted into the electromechanical coupling model to obtain the load-electromagnetic coupling model. The probability characteristic statistics module is used to generate multiple sets of uncertain load samples within the value range using the quasi-Monte Carlo method, input each set of uncertain load samples into the load-electromagnetic coupling model, and statistically obtain the probability characteristics of the phased array antenna's electrical performance.
9. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method for fast calculation of the electrical performance of an airborne phased array antenna based on the CatBoost proxy model as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method for rapid calculation of the electrical performance of an airborne phased array antenna based on the CatBoost proxy model as described in any one of claims 1-7.
