A method and system for simulating and analyzing performance of a phased array antenna

Abstract

The application discloses a kind of phased array antenna performance simulation analysis method and system, including the following steps: acquisition phased array antenna parameter and sparse reference electromagnetic simulation data, construct array topological function representation and array element mutual coupling matrix;Array topological function representation and frequency space query condition are input into improved DeepONet model, obtain initial electromagnetic field operator representation and initial port response representation;According to electromagnetic physical residual and array element mutual coupling matrix, physical correction and mutual coupling compensation are carried out, to obtain corrected electromagnetic field result and corrected port response result;Analysis performance index vector and calibrate model;Call calibration model to execute batch inference, generate parameter sensitivity analysis result, abnormal performance positioning result and analysis report.The application belongs to the field of phased array antenna simulation and artificial intelligence technology, can reduce repeated full-wave simulation calculation, improve simulation efficiency and prediction accuracy.

Description

Technical Field

[0001] This invention relates to the fields of phased array antenna simulation and artificial intelligence technology, specifically to a method and system for simulating and analyzing the performance of phased array antennas. Background Technology

[0002] Phased array antennas control the feed amplitude and phase of multiple array elements to achieve beam pointing adjustment, beamforming, and spatial scanning. They are widely used in radar detection, wireless communication, satellite communication, and electronic countermeasures. The performance of a phased array antenna is affected by a variety of factors, including element structure, element spacing, material properties, feed state, operating frequency, scanning direction, and element mutual coupling. Therefore, during antenna design and parameter optimization, simulation analysis of the radiation pattern, gain, beamwidth, sidelobe level, scanning loss, and port characteristics is necessary.

[0003] Current phased array antenna performance analysis typically employs full-wave electromagnetic simulation methods such as the finite element method, finite difference time-domain method, or method of moments. While these methods can obtain highly accurate electromagnetic field results, they require repeated mesh generation, matrix solving, and field quantity calculations when dealing with a large number of array elements, complex structural parameters, or when traversing a large number of operating frequencies and scanning directions. This results in significant computational resource consumption and makes it difficult to meet the needs of rapid simulation and batch analysis under multiple parameter combinations.

[0004] To improve simulation efficiency, some existing technologies use neural networks to establish the mapping relationship between antenna parameters and performance indicators. However, ordinary neural networks mainly fit data for fixed-dimensional inputs and outputs, making it difficult to effectively describe the operator mapping relationship between array structure functions and continuous electromagnetic field functions. At the same time, the models usually do not fully consider the topological connections of array elements, feeding states, and mutual coupling propagation between array elements, which can easily lead to increased prediction errors under conditions of large scan angles, high mutual coupling, and non-uniform arrays.

[0005] Furthermore, existing data-driven simulation models primarily rely on a large number of high-precision electromagnetic simulation samples for training, failing to fully utilize electromagnetic physical constraints such as Maxwell's equations, boundary continuity, and energy conservation. When the training samples are sparse or the parameters to be analyzed exceed the existing sample distribution, the prediction results may not satisfy the laws of electromagnetic physics. Existing methods also struggle to adaptively calibrate the model output based on a limited amount of reference simulation data, and further complete parameter sensitivity analysis and anomaly element localization.

[0006] Therefore, how to provide a method and system for simulating and analyzing the performance of phased array antennas is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] One objective of this invention is to propose a method and system for simulating and analyzing the performance of phased array antennas. This invention constructs an array topology function representation and an element mutual coupling matrix, inputs the array topology function representation and frequency-space query conditions into an improved DeepONet model, and introduces topology-aware operator mapping, physical residual correction, element mutual coupling propagation compensation, and multi-performance index decoding. This allows for joint simulation of the electromagnetic field distribution, radiation performance, and port performance of the phased array antenna. Furthermore, based on sparse reference electromagnetic simulation data, model calibration, parameter sensitivity analysis, and anomaly performance localization are completed. Compared with existing technologies, this invention can establish operator mappings between array structure parameters, feed states, frequency-space conditions, and continuous electromagnetic field results, reducing the computational overhead of repetitive full-wave simulations under multiple parameter combinations, and improving the performance prediction accuracy under large scan angles, high mutual coupling, and sparse sample conditions. It possesses advantages such as high simulation efficiency, strong physical consistency, accurate mutual coupling compensation, and a high degree of intelligent performance analysis.

[0008] A method and system for simulating and analyzing the performance of a phased array antenna according to an embodiment of the present invention includes the following steps: S1. Collect phased array antenna parameters and sparse reference electromagnetic simulation data, and obtain the antenna simulation sample set after correlation alignment and standardization processing; S2. Extract the spatial relationships, attributes, feeding states and electromagnetic response relationships of array elements from the antenna simulation sample set, and construct the array topology function representation and array element mutual coupling matrix; S3. Input the array topology function representation and frequency-space query conditions into the improved DeepONet model, perform topology-aware operator mapping, and obtain the initial electromagnetic field operator representation and the initial port response representation. S4. Based on the electromagnetic physical residual and the array element mutual coupling matrix, physical correction and mutual coupling compensation are performed on the two types of initial characterizations to obtain the corrected electromagnetic field results and the corrected port response results. S5. Perform multi-performance index analysis on the corrected electromagnetic field results and the corrected port response results to obtain the phased array antenna performance index vector. S6. Perform error verification between the performance index vector and the reference performance index vector, and update the model calibration parameters to obtain the calibrated improved DeepONet model and performance simulation results. S7. Call the calibrated and improved DeepONet model to perform batch operator inference, and obtain parameter sensitivity analysis results, abnormal performance location results and performance simulation analysis report.

[0009] Optionally, step S2 includes the following steps: S21. Establish a set of array element nodes according to the array element number, array element spatial coordinates and array element type, and write the array element size, material properties and polarization state into the corresponding array element node. S22. Establish the array element topology edges based on the array element spacing, arrangement relationship and power supply connection relationship to obtain the array topology connection structure; S23. Encode the feeding amplitude, feeding phase and array element working state into array element excitation features, and fuse them with array element node attributes to obtain array element state features. S24. Jointly encode the array element state characteristics and array topology connection structure to form an array topology function representation, and form an array element mutual coupling matrix based on the array element electromagnetic response relationship under the current operating conditions.

[0010] Optionally, step S3 includes the following steps: S31. The array topology function is used to represent the input array topology-aware branch network. The array element state features are propagated according to the array topology connection structure to obtain the local topology features of the array elements. S32. Perform effective state-weighted pooling and gated fusion on the local topological features of the pair elements to obtain fixed-dimensional branch operator features; S33. Extract the operating frequency, spatial coordinates and scanning direction from the antenna simulation sample set, construct the frequency-space query conditions, and input the frequency-space query conditions into the frequency-space joint backbone network to obtain the backbone query features; S34. Using the adaptive operator fusion unit, the branch operator features and the main query features are fused in a state-related manner to obtain the initial electromagnetic field operator characterization and the initial port response characterization.

[0011] Optionally, step S34 includes the following steps: S341. Determine the operator query status corresponding to the current working frequency, spatial location, and scanning direction based on the trunk query characteristics; S342. Generate candidate operator features and corresponding state-related weights based on the operator query state, branch operator features, and local topological features of the array elements. S343. Use state-related weights to perform weighted aggregation of candidate operator features, and combine the aggregation results with the main query features dimension by dimension. S344. Perform electromagnetic field component mapping and port response mapping on the dimension-by-dimensional combination results to obtain the initial electromagnetic field operator characterization and the initial port response characterization.

[0012] Optionally, step S4 includes the following steps: S41. Analyze the electric field components and magnetic field components from the initial electromagnetic field operator characterization, and extract the port response characteristics from the initial port response characterization. S42. Determine the field equation residuals, boundary continuity residuals, and energy deviations based on the electric field components, magnetic field components, medium properties, and boundary conditions. S43. Jointly encode the field equation residuals, boundary continuity residuals, and energy deviations to generate electromagnetic field component correction quantities and obtain the physical correction electromagnetic field results. S44. Based on the propagation of local electromagnetic response and port response of the array element mutual coupling matrix, the corrected electromagnetic field result and the corrected port response result are obtained.

[0013] Optionally, S44 includes the following steps: S441. Establish a response mapping index based on the spatial coordinates of the array elements and the port positions, and extract the local electromagnetic response characteristics and port response characteristics of each array element. S442. Generate the mutual coupling propagation coefficient based on the mutual coupling strength, mutual coupling direction, array element distance, and feed phase difference in the array element mutual coupling matrix; S443. By aggregating the two types of response characteristics of adjacent array elements using the mutual coupling propagation coefficient, the spatial field mutual coupling compensation amount and the port mutual coupling compensation amount are obtained. S444. The two types of mutual coupling compensation quantities are respectively fused with the physical correction electromagnetic field results and the initial port response characterization to obtain the correction electromagnetic field results and the correction port response results.

[0014] Optionally, step S5 includes the following steps: S51. Perform near-field to far-field conversion on the corrected electromagnetic field results to obtain far-field radiation data corresponding to different frequencies and scanning directions. S52. Extract the main beam, side lobes, nulls and polarization features from the far-field radiation data to obtain a set of radiation pattern features; S53. Based on the set of radiation pattern features, analyze the gain, beamwidth, sidelobe level, scanning loss, and front-to-back ratio to obtain the radiation performance indicators; S54. Analyze the active reflection coefficient, return loss, and active VSWR based on the correction port response results, and combine them with radiation performance indicators to form a performance index vector.

[0015] Optionally, step S6 includes the following steps: S61. Extract reference radiation performance indicators and reference port performance indicators from sparse reference electromagnetic simulation data to form a reference performance indicator vector. S62. Associate and match the performance index vector with the reference performance index vector to obtain the multi-index error set and the corresponding index calibration weights. S63. Based on the indicator calibration weight, fuse the multi-indicator error set to generate operator output correction parameters and performance indicator deviation compensation parameters; S64. Update the corresponding calibration unit based on the two types of calibration parameters and re-infer to obtain the calibrated improved DeepONet model and performance simulation results.

[0016] Optionally, step S7 includes the following steps: S71. Generate multiple antenna parameter combinations according to the element spacing, element size, feed amplitude and phase of each element, element operating status, operating frequency and scanning direction. S72. Call the calibrated and improved DeepONet model to calculate the performance index vector corresponding to each antenna parameter combination and obtain the batch performance simulation results. S73. Based on the correlation between the changes in antenna parameters and the changes in performance indicators, generate parameter sensitivity analysis results and parameter-performance correlation relationships; S74. Based on the degree of performance index anomaly and the parameter-performance correlation, determine the candidate anomaly array elements and candidate anomaly parameters, use them as the anomaly performance localization results, and generate a performance simulation analysis report.

[0017] A phased array antenna performance simulation and analysis system according to an embodiment of the present invention includes the following modules: The data processing module is used to collect phased array antenna parameters and sparse reference electromagnetic simulation data, perform correlation alignment and standardization processing, and obtain an antenna simulation sample set. The topology construction module is used to extract the spatial relationships, attributes, feeding states and electromagnetic response relationships of array elements, and obtain the array topology function representation and array element mutual coupling matrix. The operator mapping module includes an array topology sensing branch network, a frequency-space joint backbone network, and an adaptive operator fusion unit, which are used to generate initial electromagnetic field operator representations and initial port response representations. The electromagnetic field correction module includes a physical residual correction unit and a mutual coupling propagation unit, which are used to perform physical consistency correction and mutual coupling compensation to obtain the corrected electromagnetic field results and the corrected port response results. The performance analysis module includes a multi-performance index decoding unit, which is used to analyze the correction electromagnetic field results and correction port response results to obtain the phased array antenna performance index vector. The model calibration module includes an operator output calibration unit, which is used to update the model calibration parameters based on the error between the performance index vector and the reference performance index vector, so as to obtain the calibrated improved DeepONet model and performance simulation results. The simulation analysis module is used to call the calibrated and improved DeepONet model to perform batch operator inference and obtain parameter sensitivity analysis results and anomaly performance localization results. The report output module is used to associate antenna parameters, performance indicators, parameter sensitivity analysis results, and abnormal performance location results to generate a performance simulation analysis report.

[0018] The beneficial effects of this invention are as follows: By constructing array topology function representations and element mutual coupling matrices, and utilizing an improved DeepONet model to establish operator mapping relationships between array topology functions, frequency-space query conditions, continuous electromagnetic field functions, and port response functions, this invention can generate initial electromagnetic field operator representations and initial port response representations under different element numbers, array configurations, operating frequencies, and scanning directions. Compared to repeatedly performing mesh generation and full-wave electromagnetic solution for each antenna parameter combination, this invention can reuse the branch operator features generated by the array topology-aware branch network, updating the backbone query features only according to the changing frequency-space query conditions, thereby reducing the repetitive computation load of multi-parameter combination simulations and improving the efficiency of batch performance analysis of phased array antennas.

[0019] This invention introduces an array topology-aware branch network, an adaptive operator fusion unit, a physical residual correction unit, and a mutual coupling propagation unit into the improved DeepONet model. The array topology-aware branch network preserves element positions, material properties, feed states, and element topological relationships, and converts the array topology function corresponding to different numbers of elements into fixed-dimensional branch operator features. The adaptive operator fusion unit adjusts the state-related weights of candidate operator features based on operating frequency, spatial location, and scanning direction, improving the adaptability of operator mapping under near-field, far-field, and large scan angle conditions. The physical residual correction unit constrains the electromagnetic field prediction results through field equation residuals, boundary continuity residuals, and energy deviations. The mutual coupling propagation unit compensates for local electromagnetic responses and port responses based on the element mutual coupling matrix corresponding to the current operating condition, thereby reducing non-physical prediction biases under conditions of high mutual coupling, non-uniform arrays, and sparse samples, and improving the consistency between electromagnetic field results and port performance results.

[0020] This invention can analyze gain, beam direction, beamwidth, sidelobe level, scanning loss, and front-to-back ratio from the corrected electromagnetic field results, and analyze active reflection coefficient, return loss, and active VSWR from the corrected port response results, achieving joint simulation of radiation performance and port performance. By performing error verification between the performance index vector and the reference performance index vector corresponding to the sparse reference electromagnetic simulation data, and updating the operator output calibration unit and multi-performance index decoding unit according to the normalized index error and index calibration weight, it can adaptively calibrate the model output while reducing the amount of high-precision reference simulation data for the target antenna, thereby improving the performance prediction accuracy in different frequency and scanning regions.

[0021] This invention can also invoke a calibrated and improved DeepONet model to perform batch operator inference on multiple antenna parameter combinations, calculate parameter sensitivity based on normalized performance changes and normalized parameter changes, and form parameter sensitivity analysis results and parameter-performance correlations. Furthermore, through candidate parameter recovery inference, it compares the overall anomaly degree before and after recovery, calculates the anomaly confidence of candidate array elements and candidate parameters, and obtains anomaly performance localization results. This processing method can screen candidate anomalous array elements and candidate anomalous parameters that lead to performance degradation, providing a basis for phased array antenna structure design, feed configuration optimization, scanning performance evaluation, and subsequent measured fault confirmation. Attached Figure Description

[0022] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a phased array antenna performance simulation and analysis method proposed in this invention; Figure 2 This is a schematic diagram of the structure of the improved DeepONet model proposed in this invention; Figure 3 This is a schematic diagram of the electromagnetic field physical correction and array element mutual coupling compensation process proposed in this invention. Detailed Implementation

[0023] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0024] refer to Figures 1-3 A method and system for simulating and analyzing the performance of a phased array antenna, comprising the following steps: S1. Collect phased array antenna parameters and sparse reference electromagnetic simulation data, and obtain the antenna simulation sample set after correlation alignment and standardization processing; S2. Extract the spatial relationships, attributes, feeding states and electromagnetic response relationships of array elements from the antenna simulation sample set, and construct the array topology function representation and array element mutual coupling matrix; S3. Input the array topology function representation and frequency-space query conditions into the improved DeepONet model, perform topology-aware operator mapping, and obtain the initial electromagnetic field operator representation and the initial port response representation. S4. The physical residual correction unit is used to physically correct the initial electromagnetic field operator characterization, and the physical correction electromagnetic field result and the initial port response characterization are mutually compensated according to the array element mutual coupling matrix to obtain the corrected electromagnetic field result and the corrected port response result. S5. Perform multi-performance index analysis on the corrected electromagnetic field results and the corrected port response results to obtain the phased array antenna performance index vector. S6. Perform error verification between the performance index vector and the reference performance index vector, and update the model calibration parameters to obtain the calibrated improved DeepONet model and performance simulation results. S7. Call the calibrated and improved DeepONet model to perform batch operator inference, and obtain parameter sensitivity analysis results, abnormal performance location results and performance simulation analysis report.

[0025] In this embodiment, the improved DeepONet model refers to a neural operator model that introduces array topology awareness, frequency-space joint query, adaptive operator fusion, physical residual correction, array element mutual coupling propagation, and operator output calibration mechanisms on the basis of the DeepONet operator learning structure.

[0026] The improved DeepONet model includes an array topology-aware branch network, a frequency-space joint backbone network, an adaptive operator fusion unit, a physical residual correction unit, a mutually coupled propagation unit, a multi-performance index decoding unit, and an operator output calibration unit.

[0027] The array topology-aware branch network is used to process the array topology function representations corresponding to different numbers of array elements, different array sizes, and different arrangements, and converts the local topology features of a variable number of array elements into fixed-dimensional branch operator features. The frequency-space joint backbone network is used to encode the operating frequency, spatial coordinates, and scanning direction to generate backbone query features. The adaptive operator fusion unit is used to establish the operator mapping between the array topology function and the continuous electromagnetic field function and port response function, and at the same time generate the initial electromagnetic field operator representation and the initial port response representation.

[0028] The physical residual correction unit is used to correct the initial electromagnetic field operator characterization based on the field equation residuals, boundary continuity residuals, and energy deviations. The mutual coupling propagation unit is used to generate spatial field mutual coupling compensation and port mutual coupling compensation quantities based on the array element mutual coupling matrices corresponding to the current operating condition. The multi-performance index decoding unit is used to parse radiation performance indices and port performance indices from the corrected electromagnetic field results and corrected port response results. The operator output calibration unit is used to update the model calibration parameters based on sparse reference electromagnetic simulation data.

[0029] In this embodiment, S1 includes: Collect phased array antenna parameters. Phased array antenna parameters include array structure parameters, array element attribute parameters, feed control parameters, active circuit parameters, and operating condition parameters.

[0030] The array operating status parameters are used to characterize the actual operating status of the array elements and their corresponding RF channels under the current operating conditions. They are jointly determined by the power supply control parameters and active circuit parameters, including array element excitation status, transmit / receive status, power amplifier gain, phase shifter phase status, channel transmission delay, channel amplitude error, channel phase error, channel switching status, power supply status, and channel operating temperature.

[0031] The array structure parameters include the number of array elements, array type, spatial coordinates of each element, element spacing, element size, and array aperture. Array types include planar arrays, linear arrays, conformal arrays, non-uniform arrays, and combined arrays composed of multiple subarrays. A correspondence between the array structure parameters and each array element is established according to the element numbering.

[0032] Element attribute parameters include element type, conductor properties, dielectric properties, substrate thickness, polarization state, port location, and element mounting orientation. Element attribute parameters belonging to the same element number are grouped together to form an element attribute record.

[0033] The feed control parameters include the target feed amplitude, target feed phase, channel switch status, beam control code, and feed connection relationship for each array element. These feed control parameters represent the target control status sent by the beam control system to each RF channel.

[0034] Active circuit parameters are used to describe the actual transmission state of the RF channel corresponding to each array element, including at least one of the following: power amplifier gain, gain compression state, amplitude nonlinearity parameter, phase nonlinearity parameter, low noise amplifier gain and noise state, phase shifter insertion loss, phase shift accuracy and phase quantization state, attenuator insertion loss and attenuation accuracy, T / R switch insertion loss and isolation, channel transmission delay, channel amplitude error, channel phase error, power supply state, channel operating temperature, channel operating state, and channel calibration parameters.

[0035] Active circuit parameters can be derived from RF device design data, RF channel test data, channel calibration data, array operation monitoring data, or analog data generated based on the device's allowable operating range. When using analog data, a simulation verification identifier should be set for the corresponding data to distinguish it from actual RF channel data.

[0036] The operating condition parameters include operating frequency, scanning azimuth angle, scanning elevation angle, observation direction, ambient temperature, environmental medium state, and simulation boundary conditions. Different operating conditions are distinguished according to the simulation condition number.

[0037] The actual complex gain of the RF channel is determined based on the feed control parameters and active circuit parameters corresponding to each array element. The actual complex gain includes the actual amplitude transmission state and the actual phase transmission state.

[0038] The actual amplitude transmission state is determined by the target feed amplitude, power amplifier gain, power amplifier gain compression state, phase shifter insertion loss, attenuator insertion loss, T / R switch insertion loss, and channel amplitude error.

[0039] The actual phase transmission state is determined by the target feed phase, the phase quantization state of the phase shifter, the phase shifter phase error, the channel phase error, the phase nonlinearity parameter, and the channel transmission delay.

[0040] When the power amplifier is operating in the linear region, the actual amplitude transmission state is determined based on the power amplifier gain and the channel insertion loss; when the power amplifier enters the gain compression region, the actual output amplitude is corrected based on the gain compression state and the amplitude nonlinearity parameter.

[0041] When the phase shifter has a phase quantization error, the target feed phase is mapped to the corresponding discrete phase state, and the phase offset caused by the channel phase error and the channel transmission delay is superimposed to determine the actual phase transmission state.

[0042] The T / R switch status determines whether the corresponding RF channel is in transmit, receive, or inactive state. When the T / R switch is inactive, the RF channel power supply is abnormal, or the channel is in the off state, the actual complex gain of the corresponding RF channel is set to inactive.

[0043] By correcting the feed control parameters using the actual complex gain, the actual feed amplitude and actual feed phase corresponding to each array element are obtained. The target feed amplitude and target feed phase represent the target control state sent by the control terminal, while the actual feed amplitude and actual feed phase represent the actual excitation state that reaches the array element port after transmission through the active circuit. Both are recorded separately.

[0044] Based on the actual power supply amplitude, actual power supply phase, transmit / receive status, channel switch status, power supply status, power amplifier gain, phase shifter phase status, channel transmission delay, channel amplitude error, and channel phase error, the array operating status parameter record corresponding to each array element is formed, and a correspondence is established according to the array element number and the RF channel number.

[0045] A candidate parameter space is established based on the operating frequency range, scanning range, element spacing range, element size range, feed control range, and active circuit parameter range to be analyzed. Each element in the candidate parameter space corresponds to a set of phased array antenna parameter combinations.

[0046] The parameter combinations corresponding to the frequency boundary, scan angle boundary, large scan angle, high mutual coupling state, non-uniform array state, power amplifier gain compression state, channel phase offset state, channel transmission delay change state, and T / R channel stop working state are selected from the candidate parameter space as the basic reference parameter combinations.

[0047] The full-wave electromagnetic simulation program is called to solve the combination of basic reference parameters, and the corresponding spatial electric field components, spatial magnetic field components, far-field radiation data, port incident response, port reflection response, inter-port transmission response and phased array antenna performance indicators are obtained, forming basic reference electromagnetic simulation data.

[0048] By using multiple improved DeepONet models pre-trained with different initial parameters based on historical phased array antenna samples, joint inference is performed on the remaining parameter combinations in the candidate parameter space to obtain multiple performance index prediction vectors.

[0049] For each performance index, the dispersion of multiple prediction results relative to the prediction average is calculated, and the dispersion is divided by the allowable range of variation of the corresponding performance index to obtain the normalized prediction dispersion.

[0050] The normalized prediction dispersion of each item is weighted to obtain the discrete evaluation value of the prediction result. The field equation residuals, boundary continuity residuals, and energy deviations corresponding to the candidate parameter combinations are normalized and weighted to obtain the physical residual evaluation value.

[0051] Calculate the normalized performance change and normalized parameter change between the current parameter combination and its adjacent parameter combinations. Use the ratio between the normalized performance change and the normalized parameter change as the local performance change evaluation value.

[0052] The prediction uncertainty is obtained by weighting the discrete evaluation values, physical residual evaluation values, and local performance change evaluation values ​​of the prediction results. All weights are non-negative and their sum is one. A higher prediction uncertainty indicates lower prediction stability of the improved DeepONet model under the corresponding parameter combination.

[0053] Candidate parameter combinations are sorted from largest to smallest based on their prediction uncertainty. Parameter combinations whose prediction uncertainty is higher than a preset uncertainty threshold and whose normalized parameter distance from existing reference parameter combinations is higher than a preset distance threshold are selected as supplementary reference parameter combinations.

[0054] The full-wave electromagnetic simulation program is called to solve for the supplementary reference parameter combination, and the electromagnetic field data, port response data and performance indicators corresponding to the supplementary reference parameter combination are added to the basic reference electromagnetic simulation data to form sparse reference electromagnetic simulation data.

[0055] Based on the antenna scheme number, array element number, RF channel number, frequency point corresponding to the operating frequency, scanning direction and spatial sampling position, the array structure parameters, array element attribute parameters, feed control parameters, active circuit parameters, array operating status parameters, operating condition parameters and sparse reference electromagnetic simulation data are correlated and aligned.

[0056] The array spatial coordinates, array size, array spacing, target feed amplitude, actual feed amplitude, power amplifier gain, channel transmission delay, operating frequency, and electromagnetic field amplitude are standardized.

[0057] The target feed phase, actual feed phase, phase shifter phase state, channel phase error, scanning azimuth angle and scanning elevation angle are encoded according to a periodic continuous relationship to avoid characteristic jumps at phase boundaries and angle boundaries.

[0058] The sparse reference electromagnetic simulation data is subjected to quality checks to identify simulation results that do not meet the solution convergence conditions, missing spatial sampling points, abnormal port responses, and abrupt data that cannot be explained by parameter changes, and the corresponding data validity status is set.

[0059] The phased array antenna parameters, array operating status parameters, and sparse reference electromagnetic simulation data that have undergone correlation alignment, standardization, and quality checks are uniformly stored to obtain an antenna simulation sample set.

[0060] In this embodiment, S2 includes: Extract the array element number, array element spatial coordinates, and array element type from the antenna simulation sample set. Establish an array element node set based on the array element number. Each array element node in the array element node set corresponds to an actual array element in the phased array antenna.

[0061] The array element size, conductor properties, dielectric properties, substrate thickness, polarization state, port properties, and array element mounting orientation are written into the corresponding array element node to form array element node properties.

[0062] The array element topology edges are established based on the element spacing, arrangement, and feeding connection relationships. For array elements with spatial adjacency, feeding correlation, or significant electromagnetic influence, corresponding topological connections are established to obtain the array topology connection structure.

[0063] Record the distance between array elements, relative direction, power supply branch relationship and basic mutual coupling state for each array element topology edge to form topology edge attributes.

[0064] The actual power supply amplitude, actual power supply phase, channel operating status, power supply status, operating temperature, power amplifier gain status, phase shifter status, and channel transmission delay are encoded as active channel status features.

[0065] The active channel state characteristics are fused with the corresponding array element node attributes to obtain the array element state characteristics.

[0066] When the power amplifier enters the gain compression state, the array element status characteristics record the actual output amplitude after compression and the corresponding degree of gain compression; when there is a phase error or transmission delay change in the RF channel, the array element status characteristics record the corrected actual feed phase; when the channel is closed or the power supply is abnormal, the array element status characteristics record the inactive state of the corresponding array element.

[0067] Multi-layer topology propagation is performed on the array element state features according to the array topology connection structure, so that each array element node receives the state features and topology edge attributes of the adjacent array element nodes, and obtains the local topology features of each array element.

[0068] The first layer of topology propagation is used to extract the structural relationships, feeding relationships, and active channel state differences between directly adjacent array elements. The second layer of topology propagation is used to aggregate the array arrangement state, feeding amplitude and phase change state, and local mutual coupling state within a larger neighborhood. Subsequent topology propagation is used to fuse the local array element state and the overall array state.

[0069] During each layer of topology propagation, the original node attributes of the current array element are preserved, and gating parameters are generated based on the array element working state, channel working state, and node effective state. The gating parameters are used to control the fusion ratio between the original node attributes and the neighborhood propagation features.

[0070] Based on the effective state of the array element node, the local topological features of the array element are weighted and pooled to extract the average state, peak state and discrete state of the array element features respectively, and the three types of states are fused to form the global topological features of the array with fixed dimensions.

[0071] The global topological features of the array are fused with the selectively aggregated local topological features of the array elements to form fixed-dimensional branch operator features, enabling different numbers of array elements, different array sizes, and different arrangements to be input into the same improved DeepONet model.

[0072] The array element node attributes, array topology connection structure, active channel state characteristics, array element local topology characteristics, and array global topology characteristics are uniformly encapsulated to form an array topology function representation.

[0073] The port responses and local electromagnetic responses of array elements are extracted from sparse reference electromagnetic simulation data. Based on the influence of source element excitation changes on the port responses and local electromagnetic fields of target array elements, the basic mutual coupling strength and direction between array elements are determined.

[0074] A basic mutual coupling relationship is established based on the array element type, operating frequency, array element spatial distance, relative direction, and port response.

[0075] Based on the current operating frequency, scanning direction, actual feed amplitude of each array element, actual feed phase of each array element, active channel gain, channel transmission delay, operating temperature, and array element operating status, the basic mutual coupling relationship is dynamically corrected to form the array element mutual coupling matrix corresponding to the current operating condition.

[0076] When the gain of the active channel decreases, the corresponding mutual coupling propagation strength is reduced according to the actual output amplitude of the source array element; when the power amplifier enters the gain compression state, the active coupling contribution is corrected according to the actual output amplitude after compression.

[0077] When the channel phase error or channel transmission delay changes, the mutual coupling propagation direction and mutual coupling propagation state are corrected according to the actual phase difference between the source array element and the target array element.

[0078] When changes in operating temperature cause changes in the gain or phase state of the active channel, the mutual coupling propagation relationship is updated using the actual complex gain at the corresponding temperature.

[0079] When the T / R channel is inactive, closed, or has an abnormal power supply, the active mutual coupling propagation connection of the corresponding array element is set to inactive, but the passive coupling relationship of other array elements propagating to this array element is retained.

[0080] When the spatial distance between array elements exceeds the preset mutual coupling propagation range, the corresponding mutual coupling connection will be set to an inactive state.

[0081] In this embodiment, S3 includes: The array topology function is input into the array topology-aware branch network of the improved DeepONet model. The array element state features are propagated according to the array topology connection structure to obtain the local topology features of the array elements.

[0082] Effective state weighted pooling and gated fusion are performed on the local topological features of the pairwise elements to obtain fixed-dimensional branch operator features.

[0083] The operating frequency, spatial coordinates, and scanning direction are extracted from the antenna simulation sample set to construct frequency-space query conditions, which are then input into the frequency-space joint backbone network.

[0084] The frequency-space joint backbone network encodes the operating frequency, spatial location, and scanning direction as features, respectively.

[0085] The spatial positional relationship between the point to be predicted and the array aperture, array element positions, and observation area is determined based on the spatial coordinates, thus forming the spatial positional state.

[0086] The relationship between the current operating frequency and the operating region and resonant region of the array element is determined based on the operating frequency, thus forming the frequency state.

[0087] The relationship between the target beam direction and the observation direction is determined based on the scanning direction, thus forming the scanning state.

[0088] The spatial location state, frequency state, and scan state are jointly encoded to obtain the backbone query features. The backbone query features and the branch operator features have corresponding operator base dimensions.

[0089] An adaptive operator fusion unit is used to perform state-related fusion of branch operator features and trunk query features.

[0090] The operator query status corresponding to the current working frequency, spatial location, and scanning direction is determined based on the trunk query characteristics.

[0091] Multiple candidate operator features and corresponding state-related weights are generated based on the operator query state, branch operator features, and local topological features of the array elements.

[0092] In near-field query mode, increase the feature weights of candidate operators corresponding to array elements near the query position. In far-field query mode, increase the feature weights of candidate operators corresponding to the global array state and the overall actual feed phase distribution. In large scan angle mode, increase the feature weights of candidate operators corresponding to the array edge region, the actual feed phase gradient, and the highly coupled region.

[0093] When some active channels are in a state of decreased gain, phase shift, or inactivity, the state-related weights are adjusted according to the local topological characteristics of the corresponding array elements so that the candidate operator features can reflect the changes in local electromagnetic response caused by the changes in the state of the active channels.

[0094] The candidate operator features are weighted and aggregated using state-related weights, and the aggregation results are combined with the main query features dimension by dimension.

[0095] Perform electric field component mapping, magnetic field component mapping, and port response mapping on the dimension-by-dimensional combination results respectively.

[0096] The mapping results of the electric field component and the magnetic field component are combined to obtain the initial electromagnetic field operator characterization.

[0097] The port response mapping results are arranged according to the port number to obtain the initial port response characterization. The initial port response characterization includes the incident response, reflection response, and inter-port transmission response corresponding to each array element port.

[0098] In this embodiment, S4 includes: The electric and magnetic field components corresponding to different spatial sampling points are analyzed from the initial electromagnetic field operator characterization to obtain the initial electromagnetic field prediction results.

[0099] The incident response, reflection response, and inter-port transmission response corresponding to each array element port are extracted from the initial port response characterization to form port response features.

[0100] Obtain the material dielectric properties, material magnetic properties, conductor boundary conditions, dielectric interface conditions, and radiation boundary conditions corresponding to the current operating conditions, and correlate the above data with the initial electromagnetic field prediction results.

[0101] The improved DeepONet model outputs automatic differentiation results relative to spatial coordinates, calculates the spatial partial derivatives corresponding to each electric field component and magnetic field component, and forms the electric field curl and magnetic field curl based on the combination of spatial partial derivatives.

[0102] The difference between the electric field curl and the target term jointly determined by the current operating frequency, material permeability, and magnetic field components is calculated to form the residual of the first field equation.

[0103] The difference between the magnetic field curl and the target term jointly determined by the material conductivity, material dielectric constant, current operating frequency, and electric field components is calculated to form the residual of the second field equation.

[0104] Calculate the magnitudes of the residuals of the first and second field equations respectively, and divide the residual magnitudes by the sum of the magnitudes of the corresponding electromagnetic field target terms and the preset stability constants to obtain the normalized field equation residuals.

[0105] The average of the normalized field equation residuals corresponding to all spatial sampling points is used to obtain the field equation residual evaluation value.

[0106] For sampling points on the conductor surface, calculate the deviation of the tangential electric field component relative to the target conductor boundary state. For sampling points on both sides of the dielectric interface, calculate the continuity deviation of the tangential electric and magnetic field components. For sampling points at the radiation boundary, calculate the deviation of the predicted electromagnetic field propagation state relative to the target radiation boundary relationship.

[0107] The boundary continuity residual evaluation value is obtained by dividing each type of boundary deviation by the sum of the amplitude of the corresponding boundary field component and the preset stability constant, and averaging the results based on the number of sampling points corresponding to the conductor surface, dielectric interface and radiation boundary.

[0108] The port input power is calculated based on the port incident response in the initial port response characterization, the reflected power is calculated based on the port reflection response, the radiated power is calculated based on the initial electromagnetic field prediction results, and the material loss power is determined based on the material properties and conductor properties.

[0109] Calculate the absolute difference between the port input power and the sum of reflected power, radiated power and material loss power, and divide this absolute difference by the sum of the port input power and the preset stability constant to obtain the energy deviation.

[0110] The residual evaluation values ​​of the field equations, the boundary continuity residual evaluation values, and the energy deviation are multiplied by their respective physical constraint weights, and the weighted results are summed to obtain the comprehensive physical residual. All physical constraint weights are non-negative and their sum is one.

[0111] The residuals of the field equations, the residuals of the boundary continuity, and the energy deviations corresponding to each spatial sampling point are spliced ​​together with the initial electromagnetic field characteristics to form the residual correction input.

[0112] The residual type encoding, spatial location encoding, and field component state encoding are performed on the residual correction input to form physical residual characteristics.

[0113] Based on the amplitude of the physical residual, the direction of residual change, and the spatial location of the residual, correction coefficients are generated for each electric field component and magnetic field component, so that the location with a larger physical residual obtains a higher correction intensity, and the location with a lower overall physical residual maintains a lower correction intensity.

[0114] The electromagnetic field component correction amount is generated based on the correction coefficients of each field component, and the electromagnetic field component correction amount is fused with the initial electromagnetic field prediction result by residual fusion to obtain the first physical correction electromagnetic field result.

[0115] Recalculate the comprehensive physical residual corresponding to the physical correction electromagnetic field result, and compare the current comprehensive physical residual with the previous comprehensive physical residual.

[0116] When the decrease in the combined physical residual between two consecutive iterations is lower than the preset residual decrease threshold, or when the preset maximum number of corrections is reached, the physical residual correction is stopped, and the physical correction electromagnetic field result is obtained.

[0117] A response mapping index is established based on the spatial coordinates and port positions of each array element. The physical correction electromagnetic field results are divided into the local electromagnetic response characteristics corresponding to each array element, and the initial port response characterization is mapped to the corresponding array element port.

[0118] The mutual coupling propagation coefficient is generated based on the mutual coupling strength, mutual coupling direction, spatial distance between array elements, actual feed phase difference, active channel gain state, and channel transmission delay in the mutual coupling matrix of array elements.

[0119] By aggregating the local electromagnetic response characteristics of adjacent array elements using the mutual coupling propagation coefficient, the mutual coupling compensation amount of the spatial field is obtained.

[0120] By aggregating the port response features corresponding to adjacent array elements using the mutual coupling propagation coefficients, the port mutual coupling compensation amount is obtained.

[0121] When the gain of the active channel decreases or enters a gain compression state, the spatial field mutual coupling compensation and port mutual coupling compensation generated by the source array element are reduced according to the corresponding mutual coupling propagation coefficient.

[0122] When there is a phase error or transmission delay change in the active channel, the phase state of the spatial field mutual coupling compensation and the port mutual coupling compensation is adjusted according to the corrected actual phase difference.

[0123] When the T / R channel is inactive, the active mutual coupling propagation contribution of the corresponding array element is canceled, while the passive mutual coupling compensation generated by other array elements for that array element is retained.

[0124] The spatial field mutual coupling compensation quantity and the physical correction electromagnetic field result are residually fused to obtain the correction electromagnetic field result.

[0125] The port mutual coupling compensation amount and the initial port response characterization are residually fused to obtain the corrected port response result.

[0126] In this embodiment, S5 includes: Based on the simulation region boundary, array aperture, and observation direction, near-field to far-field conversion is performed on the corrected electromagnetic field results to obtain far-field radiation data corresponding to different operating frequencies and scanning directions.

[0127] Identify the main beam region, side lobe region, null region, and back radiation region from far-field radiation data to form a set of radiation pattern features.

[0128] The antenna gain is determined based on the maximum radiation intensity within the main beam area.

[0129] The beamwidth is determined based on the two observation angles corresponding to the decrease in radiation intensity on both sides of the main beam peak to a preset ratio.

[0130] Local radiation peaks are searched outside the main beam region, and the amplitude difference between the highest sidelobe peak and the main beam peak is determined as the sidelobe level.

[0131] The peak gain in the current scanning direction is compared with the peak gain in the normal scan or reference scan state to determine the scan loss.

[0132] The maximum radiation intensity in the forward and backward observation areas is determined separately, and the front-to-back ratio is determined based on the amplitude difference between the two.

[0133] Based on the port incident response, port reflection response, and inter-port transmission response in the corrected port response results, and combined with the actual feed amplitude and actual feed phase of each array element, the active reflection response of each array element port is determined.

[0134] For each array element port, the reflection response of this port is superimposed with the responses transmitted to this port from other ports according to the actual feed amplitude and actual feed phase of each port to obtain the total reflection response under the current scanning state.

[0135] The active reflection coefficient is determined based on the amplitude relationship between the total reflection response and the incident response at the corresponding port.

[0136] The return loss is determined based on the amplitude of the active reflection coefficient, and the active standing wave ratio is determined based on the ratio between the maximum and minimum voltage amplitudes after the incident and reflected waves are superimposed at the port.

[0137] When the power amplifier gain, phase shifter phase error, channel transmission delay, or T / R channel state changes, the actual feed amplitude and actual feed phase change accordingly, and are reflected in the active reflection coefficient, return loss, and active VSWR through the port response superposition process.

[0138] By combining gain, beam direction, beamwidth, sidelobe level, scanning loss, front-to-back ratio, active reflection coefficient, return loss, and active VSWR, a vector of performance indicators for the phased array antenna is obtained.

[0139] In this embodiment, S6 includes: Based on the antenna parameter combination, operating frequency, and scanning direction, reference far-field radiation data and reference port response data are extracted from sparse reference electromagnetic simulation data.

[0140] Using the same performance index analysis process as S5, reference gain, reference beam direction, reference beamwidth, reference sidelobe level, reference scan loss, and reference front-to-back ratio are extracted from the reference far-field radiation data.

[0141] Extract the reference active reflection coefficient, reference return loss, and reference active VSWR from the reference port response data.

[0142] The reference radiation performance index and the reference port performance index are combined to form a reference performance index vector.

[0143] The performance index vector output by the improved DeepONet model is correlated and matched with the reference performance index vector according to the antenna parameter combination, operating frequency and scanning direction to obtain a multi-index error set.

[0144] For each performance index, the absolute difference between the model's predicted value and the corresponding reference value is calculated, and the absolute difference is divided by the sum of the allowable range of variation of the corresponding performance index and the preset stability constant to obtain the normalized index error.

[0145] For gain, beam direction, beamwidth, sidelobe level, scanning loss, and return loss, the allowable variation range is the difference between the maximum and minimum values ​​of the corresponding indicators in the antenna simulation sample set. For active reflection coefficient and active VSWR, the allowable variation range is the preset engineering evaluation range.

[0146] The reliability of the simulation is determined based on the convergence state of the solution from the reference electromagnetic simulation data. When the reference simulation results meet the preset convergence conditions, the reliability of the corresponding simulation is increased; when the reference simulation results are close to the convergence boundary or have some missing data, the reliability of the corresponding simulation is decreased.

[0147] The basic values ​​of the indicator weights are determined based on the importance of each performance indicator, the reliability of the simulation, and the normalized indicator error. The basic values ​​of each indicator weights are then divided by the sum of the basic values ​​of all indicator weights to obtain the calibration weights of each indicator, so that the sum of the calibration weights of all indicators is one.

[0148] The comprehensive performance error is obtained by weighting the corresponding normalized index error using the calibration weights of each index.

[0149] Electromagnetic field operator output correction parameters are generated based on the weighted error corresponding to the far-field radiation performance index, port response operator correction parameters are generated based on the weighted error corresponding to the port performance index, and performance index deviation compensation parameters are generated based on the remaining deviation of all performance indexes.

[0150] The operator output calibration unit is updated using the electromagnetic field operator output correction parameters and the port response operator correction parameters, so that the operator output calibration unit corrects the initial electromagnetic field operator characterization and the initial port response characterization respectively.

[0151] The performance index parsing status in the multi-performance index decoding unit is updated using performance index deviation compensation parameters.

[0152] After updating the model calibration parameters, the updated and improved DeepONet model is called to re-execute topology-aware operator mapping, physical correction, mutual coupling compensation, and multi-performance index analysis on the sparse reference parameter combination to obtain new performance simulation results.

[0153] Recalculate the overall performance error between the new performance simulation results and the reference performance index vector.

[0154] When the decrease in the overall performance error between two consecutive model calibrations is lower than the preset calibration termination threshold, or when the overall performance error is lower than the preset target error threshold, the model calibration is stopped, and the calibrated improved DeepONet model and performance simulation results are obtained.

[0155] In this embodiment, S7 includes: Multiple antenna parameter combinations are generated based on the task to be analyzed. The antenna parameter combination includes at least one of the following: element spacing, element size, material properties, target feed amplitude, target feed phase, power amplifier gain, gain compression state, phase shifter phase error, channel amplitude error, channel phase error, channel propagation delay, operating temperature, T / R channel state, operating frequency, and scanning direction.

[0156] Each antenna parameter combination is converted into a corresponding array topology function representation, array element mutual coupling matrix, and frequency-space query condition, and then uniformly encapsulated according to the parameter combination number to form a batch operator inference input set.

[0157] By calling the calibrated and improved DeepONet model, array topology coding, topology-aware operator mapping, physical residual correction, array element mutual coupling compensation, and multi-performance index analysis are performed on each antenna parameter combination to obtain batch performance simulation results.

[0158] For parameter combinations with the same array structure but changes only in operating frequency, spatial coordinates, scanning direction, or active channel state, the array structure encoding in the array topology sensing branch network is reused, and the array element state features and branch operator features are updated according to the changed active channel state.

[0159] For each parameter to be analyzed, under the condition that other antenna parameters remain unchanged, two antenna parameter combinations are generated, one with the preset perturbation amount increased and the other with the preset perturbation amount decreased.

[0160] For continuous parameters such as power amplifier gain, channel amplitude error, and operating temperature, positive and negative disturbances are set according to preset ratios of the allowable variation range of the corresponding parameters.

[0161] To address phase errors in the phase shifter and channel phase errors, positive and negative disturbance values ​​are set according to preset angles.

[0162] The state of the T / R channel and the power supply state switch between normal operation and inactive states, which creates state disturbances.

[0163] By calling the calibrated and improved DeepONet model, performance index vectors corresponding to the two parameter combinations are obtained respectively.

[0164] For each performance index, calculate the absolute value of the difference between the performance index values ​​corresponding to the two parameter combinations, and divide this absolute value by the allowable range of variation of the corresponding performance index to obtain the normalized performance variation.

[0165] Calculate the absolute value of the difference between two values ​​of the parameter to be analyzed, and divide the absolute value by the sum of the allowable range of variation of the corresponding parameter and the preset stability constant to obtain the normalized parameter change.

[0166] The ratio between the normalized performance change and the normalized parameter change is determined as the parameter sensitivity of the parameter to be analyzed relative to the corresponding performance index.

[0167] When the parameter to be analyzed can only be subjected to unidirectional perturbation, the normalized difference between the perturbed performance index value and the original performance index value is calculated, and the normalized difference is divided by the corresponding change in the normalized parameter to obtain the parameter sensitivity.

[0168] By using the importance weights corresponding to each performance index, the sensitivity of multiple parameters corresponding to the same antenna parameter is weighted to obtain the comprehensive sensitivity of the corresponding parameter.

[0169] The antenna parameters were sorted from highest to lowest overall sensitivity to obtain the parameter sensitivity analysis results.

[0170] The direction of change of the sensitivity of each parameter is retained to indicate the direction of change of the corresponding performance index when the antenna parameters increase, decrease or switch states, thus forming a parameter-performance correlation.

[0171] Based on the design allowable range of each performance indicator, calculate the extent to which the current performance indicator value exceeds the corresponding allowable range, and divide the excess by the corresponding allowable deviation to obtain the degree of performance indicator abnormality. If the performance indicator does not exceed the corresponding allowable range, the corresponding degree of abnormality is set to zero.

[0172] The degree of abnormality of each performance indicator is weighted by the importance weight of the performance indicators to obtain the overall degree of abnormality before recovery.

[0173] Based on the comprehensive anomaly degree and parameter-performance correlation before recovery, array elements, power supply control parameters and active circuit parameters with a high degree of correlation with abnormal performance are selected to form a candidate array element set and a candidate parameter set.

[0174] The reference states for candidate array elements are the rated actual feed amplitude, rated actual feed phase, and normal operating state in the antenna design scheme. The reference states for candidate active circuit parameters are the corresponding RF device design nominal values ​​or channel normal calibration values.

[0175] For each candidate array element or candidate parameter, the corresponding actual feed amplitude, actual feed phase, power amplifier gain, phase shifter phase state, channel transmission delay, operating temperature or T / R channel state are restored to the reference state, and the calibrated and improved DeepONet model is called to recalculate the performance index vector.

[0176] The overall anomaly level is recalculated based on the recovered performance index vector to obtain the recovered overall anomaly level.

[0177] Calculate the difference between the comprehensive anomaly level before recovery and the comprehensive anomaly level after recovery, and divide the difference by the sum of the comprehensive anomaly level before recovery and the preset stability constant to obtain the anomaly confidence level of the corresponding candidate array element or candidate parameter.

[0178] The higher the confidence level of the anomaly, the greater the degree of improvement in the anomaly performance after restoring the corresponding array elements or parameters.

[0179] When the anomaly confidence level reaches the preset anomaly confidence level threshold, the corresponding array element is determined as a candidate anomaly array element, the corresponding parameter is determined as a candidate anomaly parameter, and the candidate anomaly array element and candidate anomaly parameter are used as the anomaly performance localization result.

[0180] When the decrease in the overall anomaly level after multiple candidate array elements or multiple candidate active channels are jointly restored is greater than the preset proportion of the sum of the decreases when each candidate object is restored individually, the multiple candidate objects are identified as a set of associated candidate anomalies.

[0181] The antenna structure information, feed control parameters, active circuit parameters, operating conditions, batch performance simulation results, parameter sensitivity analysis results, parameter-performance correlation and abnormal performance location results are correlated to generate a performance simulation analysis report.

[0182] In this embodiment, the improved DeepONet model training process includes: The antenna simulation sample set is divided into training sample set, verification sample set and calibration sample set according to antenna scheme and parameter combination, so as to avoid different spatial sampling points corresponding to the same antenna parameter combination from entering the training sample set and verification sample set at the same time.

[0183] The training sample set includes data corresponding to different numbers of array elements, different array forms, different feed control states, and different active channel states, enabling the array topology-aware branch network to learn the topology function representation of phased array antennas of different sizes.

[0184] In the first training phase, the array topology function is input into the array topology sensing branch network, and the frequency-space query conditions are input into the frequency-space joint backbone network to train the basic operator mapping relationship between the array topology function, the continuous electromagnetic field function, and the port response function.

[0185] In the second training phase, the physical residual correction unit is trained based on the field equation residual, boundary continuity residual, and energy deviation to make the physical correction electromagnetic field result close to the reference electromagnetic field data and satisfy the corresponding electromagnetic physical relationship.

[0186] In the third training phase, the mutual-coupled propagation unit is trained based on reference data of local electromagnetic response, port response, actual power supply state and different active channel states of the array elements, so that the model can generate spatial field mutual-coupled compensation and port mutual-coupled compensation based on the array element mutual-coupled matrix.

[0187] In the fourth training phase, a multi-performance index decoding unit is trained based on reference far-field radiation data and reference port performance indicators, enabling the multi-performance index decoding unit to parse the phased array antenna performance index vector from the corrected electromagnetic field results and the corrected port response results.

[0188] In the fifth training phase, the model calibration parameters in the operator output calibration unit and the multi-performance index decoding unit are updated using the sparse reference electromagnetic simulation data corresponding to the specific phased array antenna to be analyzed, so as to obtain the calibration model for the current phased array antenna.

[0189] Based on the electromagnetic field error, comprehensive physical residual, port response error, and comprehensive performance error corresponding to the validation sample set, it is determined whether the model has reached the training termination state, thus obtaining an improved DeepONet model for performance simulation analysis of phased array antennas.

[0190] In this embodiment, the phased array antenna performance simulation and analysis system includes: The data processing module is used to collect phased array antenna parameters and sparse reference electromagnetic simulation data, perform correlation alignment and standardization processing, and obtain an antenna simulation sample set. Phased array antenna parameters include array structure parameters, element attribute parameters, feed control parameters, active circuit parameters, and operating condition parameters.

[0191] The topology construction module is used to extract the spatial relationships, attributes, feeding states, active channel states, and electromagnetic response relationships of array elements, and obtain the array topology function representation and the array element mutual coupling matrix.

[0192] The operator mapping module, including an array topology-aware branch network, a frequency-space joint backbone network, and an adaptive operator fusion unit, is used to generate initial electromagnetic field operator representations and initial port response representations.

[0193] The electromagnetic field correction module includes a physical residual correction unit and a mutual coupling propagation unit, which are used to physically correct the initial electromagnetic field operator characterization, and perform mutual coupling compensation on the physically corrected electromagnetic field result and the initial port response characterization according to the array element mutual coupling matrix, so as to obtain the corrected electromagnetic field result and the corrected port response result.

[0194] The performance analysis module includes a multi-performance index decoding unit, which is used to analyze the correction electromagnetic field results and correction port response results to obtain the phased array antenna performance index vector.

[0195] The model calibration module includes an operator output calibration unit, which is used to update the model calibration parameters based on the error between the performance index vector and the reference performance index vector, so as to obtain the calibrated improved DeepONet model and performance simulation results.

[0196] The simulation analysis module is used to call the calibrated and improved DeepONet model to perform batch operator inference and obtain parameter sensitivity analysis results and anomaly performance localization results.

[0197] The report output module is used to correlate array structure parameters, power supply control parameters, active circuit parameters, performance indicators, parameter sensitivity analysis results, and abnormal performance location results to generate a performance simulation analysis report.

[0198] The outputs of the data processing module are connected to the topology construction module and the model calibration module, respectively. The topology construction module sends the array topology function representation to the operator mapping module and the array element mutual coupling matrix to the electromagnetic field correction module.

[0199] The operator mapping module sends the initial electromagnetic field operator characterization and the initial port response characterization to the electromagnetic field correction module. The electromagnetic field correction module then sends the corrected electromagnetic field results and the corrected port response results to the performance analysis module.

[0200] The performance parsing module sends performance metric vectors to both the model calibration module and the simulation analysis module. The model calibration module updates the model calibration parameters based on the reference performance metric vectors. The simulation analysis module performs batch performance simulations based on the calibrated and improved DeepONet model.

[0201] The report output module receives batch performance simulation results, parameter sensitivity analysis results, and abnormal performance location results, and generates a phased array antenna performance simulation analysis report.

[0202] This invention introduces array topology function representation, frequency-space continuity operator mapping, electromagnetic physical residual correction, current-condition element mutual coupling propagation, and sparse reference data calibration into the performance simulation of phased array antennas. This enables antenna schemes with different numbers of elements, different array configurations, and different active channel states to be simulated using a unified model. By converting active circuit parameters into actual feed amplitude, actual feed phase, and active channel state characteristics, and further using them for array topology function characterization and element mutual coupling matrix construction, the impact of active channel gain variation, phase error, gain compression, transmission delay, operating temperature, and T / R channel state on radiation performance and port performance can be analyzed.

[0203] Example 1: To verify the feasibility of this invention in practice, it is applied to a performance simulation analysis task of an active planar phased array antenna operating in the X-band. This example uses an 8×8 active planar phased array antenna composed of 64 elements for illustration. The 64 elements and 8×8 arrangement are only set for the convenience of illustrating the implementation process of this invention and do not constitute a limitation on the number of elements, array size, or array arrangement. For phased array antennas with other numbers of elements or other array arrangements, the element node set and array topology connection structure can be established in the same way, and the local topological features of different numbers of elements can be converted into fixed-dimensional branch operator features through effective state weighted pooling.

[0204] The improved DeepONet model in this embodiment refers to a neural operator model that introduces array topology awareness, frequency-space joint query, adaptive operator fusion, physical residual correction, array element mutual coupling propagation, and operator output calibration mechanisms on the basis of the DeepONet operator learning structure. The model simultaneously generates initial electromagnetic field operator representations and initial port response representations, and performs physical correction on the electromagnetic field branches and current operating condition mutual coupling compensation on the electromagnetic field and port responses, respectively.

[0205] The phased array antenna parameters collected in this embodiment include array structure parameters, element attribute parameters, feed control parameters, active circuit parameters, and operating condition parameters. Array structure parameters include the number of elements, array type, element spatial coordinates, element spacing, element size, and array aperture; element attribute parameters include element type, conductor properties, dielectric properties, substrate thickness, polarization state, and port position; feed control parameters include target feed amplitude, target feed phase, channel switching state, and beam control code; operating condition parameters include operating frequency, scanning azimuth angle, scanning elevation angle, observation direction, ambient temperature, and simulation boundary conditions.

[0206] Each array element corresponds to one radio frequency (RF) channel. The RF channel includes at least one active circuit selected from the following: power amplifier, low-noise amplifier, phase shifter, attenuator, and T / R switch. Active circuit parameters include at least one of the following: power amplifier gain, gain compression state, amplitude nonlinearity parameter, phase nonlinearity parameter, low-noise amplifier gain and noise state, phase shifter insertion loss, phase shift accuracy and phase quantization state, attenuator insertion loss and attenuation accuracy, T / R switch insertion loss and isolation, channel propagation delay, channel amplitude error, channel phase error, power supply state, operating temperature, channel operating state, and channel calibration parameters.

[0207] In this embodiment, the active circuit parameters are generated using the device design range and simulation setting range, and all relevant data are used as simulation verification data. The power amplifier gain is set to vary around the rated gain, the channel amplitude error is set to vary around the rated output amplitude, the channel phase error and phase shift quantization error vary within a preset angle range, the channel transmission delay varies within the allowable range of the corresponding RF channel, the T / R switch states include normal operation and inactive states, and the operating temperature covers the preset operating temperature range of the target antenna.

[0208] The actual complex gain of the RF channel is determined based on the feed control parameters and active circuit parameters corresponding to each array element. The actual complex gain includes the actual amplitude transmission state and the actual phase transmission state. The actual amplitude transmission state is jointly determined by the target feed amplitude, power amplifier gain, gain compression state, phase shifter insertion loss, attenuator insertion loss, and channel amplitude error. The actual phase transmission state is jointly determined by the target feed phase, phase shifter phase quantization state, channel phase error, phase nonlinearity parameters, and channel transmission delay.

[0209] The actual feed control parameters of each array element are corrected using the actual complex gain to obtain the actual feed amplitude and actual feed phase for each element. The actual feed amplitude, actual feed phase, channel operating state, power supply state, and operating temperature are encoded as active channel state features, and these active channel state features are fused with the array element node attributes to form array element state features. When the power amplifier enters gain compression mode, the actual feed amplitude is reduced based on the amplitude nonlinearity parameter. When the channel has phase nonlinearity or transmission delay, the actual feed phase is corrected based on the phase nonlinearity parameter and the channel transmission delay. When the channel is off or the power supply is abnormal, the active excitation state of the corresponding array element is set to an inactive state.

[0210] The parameters are aligned and associated according to the antenna scheme number, array element number, RF channel number, frequency point, and scanning direction. The array element spatial coordinates, array element size, array element spacing, operating frequency, active channel gain, actual feed amplitude, and electromagnetic field amplitude are standardized. The target feed phase, actual feed phase, scanning azimuth angle, and scanning elevation angle are periodically and continuously encoded to form a data record that corresponds one-to-one with each array element and RF channel.

[0211] A candidate parameter space is established based on the frequency range, scanning range, element size range, element spacing range, power supply control range, and active circuit parameter range. From this candidate parameter space, parameter combinations corresponding to frequency boundaries, large scanning angles, high mutual coupling states, power amplifier gain compression, channel phase shifts, and T / R channel shutdown are selected. A full-wave electromagnetic simulation program is then used to obtain spatial electric field components, spatial magnetic field components, far-field radiation data, port incident response, port reflection response, inter-port transmission response, and corresponding performance indicators, forming the basic reference electromagnetic simulation data.

[0212] For the remaining candidate parameter combinations, multi-model joint inference is performed to calculate the performance index prediction dispersion, physical residual evaluation value, and local performance change evaluation value. These are then weighted according to weights of 0.40, 0.35, and 0.25 to obtain the prediction uncertainty. Parameter combinations with higher prediction uncertainty and larger distances from the normalized parameters of existing reference parameter combinations are selected for supplementary full-wave simulation to form sparse reference electromagnetic simulation data.

[0213] The array element topology connection structure is established based on the spatial coordinates, arrangement, and feeding connection of the array elements. The array element attributes and active channel state characteristics are written into the corresponding array element nodes, and the array element state characteristics are propagated according to the array topology connection structure to obtain the local topology features of the array elements. Effective state weighted pooling and gated fusion are performed on the local topology features of the array elements to obtain fixed-dimensional branch operator features.

[0214] A basic mutual coupling relationship is established based on the impact of source element excitation changes on the target element port response and local electromagnetic field. This basic mutual coupling relationship is then dynamically corrected based on the current operating frequency, scanning direction, actual feed amplitude and phase of each element, active channel gain, channel propagation delay, operating temperature, and channel operating status, forming the element mutual coupling matrix corresponding to the current operating condition. When the active channel gain decreases or enters a gain compression state, the mutual coupling propagation strength of the corresponding source element is reduced. When the channel phase error or propagation delay changes, the mutual coupling propagation direction is corrected based on the actual phase difference. When the T / R channel is inactive, the active mutual coupling propagation connection of the corresponding element is set to inactive.

[0215] The array topology function representation and frequency-space query conditions are input into the improved DeepONet model to obtain the initial electromagnetic field operator representation and the initial port response representation. The initial electromagnetic field operator representation is physically corrected based on the field equation residuals, boundary continuity residuals, and energy deviations. Spatial field mutual coupling compensation and port mutual coupling compensation are generated according to the array element mutual coupling matrix corresponding to the current operating condition, yielding the corrected electromagnetic field result and the corrected port response result.

[0216] Near-field to far-field conversion is performed on the corrected electromagnetic field results to analyze the gain, beam direction, beamwidth, sidelobe level, scanning loss, and front-to-back ratio. Based on the port incident response, port reflection response, and inter-port transmission response in the corrected port response results, the active reflection coefficient, return loss, and active VSWR are analyzed to form a vector of phased array antenna performance indicators.

[0217] During the model training phase, 420 sets of parameter combinations from various historical phased array structures and simulation conditions were selected as offline training samples, and 80 sets of parameter combinations were selected as verification samples. For the 64-element target phased array in this embodiment, 60 sets of sparse reference parameter combinations were selected as calibration samples, and another 60 sets of parameter combinations that do not overlap with the training samples and calibration samples were selected as independent test samples.

[0218] During the model calibration phase, normalized index errors for gain, beam direction, beamwidth, sidelobe level, scan loss, return loss, and active VSWR are calculated separately, and weighted according to weights of 0.20, 0.15, 0.10, 0.20, 0.10, 0.10, and 0.15 to obtain the comprehensive performance error. Based on the radiation performance error and port performance error, the operator output calibration unit and the multi-performance index decoding unit are updated to obtain the calibrated and improved DeepONet model.

[0219] After calibration, batch inference is performed on multiple antenna parameter combinations, and the array element spacing, array element size, target feed amplitude, target feed phase, power amplifier gain, phase shifter phase error, channel transmission delay, operating temperature and T / R channel status are disturbed respectively. The ratio of normalized performance change to normalized parameter change is calculated to obtain parameter sensitivity and parameter-performance correlation.

[0220] For operating conditions where performance exceeds the design tolerances, candidate array elements and candidate active circuit parameters are selected based on the parameter-performance correlation. After restoring them to their nominal design state, model inference is re-executed, and the anomaly confidence level is calculated based on the relative decrease in the overall anomaly severity before and after restoration. When the anomaly confidence level reaches 0.60, the corresponding array element is identified as a candidate anomaly array element, and the corresponding active circuit parameter is identified as a candidate anomaly parameter, thus forming the anomaly performance localization result.

[0221] Table 1. Performance simulation results under different array and active channel conditions.

[0222] The absolute gain error in Table 1 is the absolute value of the difference between the predicted gain of this invention and the reference gain of the full-wave electromagnetic simulation. The absolute sidelobe level error is the absolute value of the difference between the predicted sidelobe level of this invention and the reference sidelobe level of the full-wave electromagnetic simulation. The overall performance error is formed by weighting the normalized errors of seven indicators: gain, beam direction, beamwidth, sidelobe level, scanning loss, return loss, and active VSWR. The weights of the seven indicators are 0.20, 0.15, 0.10, 0.20, 0.10, 0.10, and 0.15, respectively. Traditional DeepONet and this invention use the same test samples, reference performance indicators, allowable variation range, and evaluation weights.

[0223] Traditional DeepONet employs conventional branch and backbone networks, lacking array topology-aware encoding, adaptive operator fusion, physical residual correction, current-condition co-location propagation, and operator output calibration mechanisms. This invention uses the same offline training samples, independent test samples, and training epochs as traditional DeepONet to ensure that the comparison results primarily reflect the performance changes brought about by the improved model structure.

[0224] As shown in Table 1, the present invention maintains low errors under uniform array, non-uniform array, and active channel state variation conditions. The absolute gain error for ten typical operating conditions ranges from 0.06 dB to 0.18 dB, with an average of 0.11 dB; the absolute sidelobe level error ranges from 0.12 dB to 0.31 dB, with an average of 0.21 dB. Compared with the traditional DeepONet, the average overall performance error of the present invention is reduced from 5.95% to 1.66%.

[0225] Under the condition that the amplification gain of channels 18 and 37 decreases by 3dB, this invention corrects the actual amplitude transmission state of the corresponding channels based on the power amplifier gain change, and updates the array element state characteristics and the array element mutual coupling matrix under the current operating condition using the actual feed amplitude. The overall performance error is 2.06%, which is lower than the 7.08% of the traditional DeepONet. This result shows that by incorporating the active channel gain and gain compression state into the array element state characteristics, the prediction deviation caused by the inconsistency between the target feed amplitude at the control end and the actual output amplitude of the array element can be reduced.

[0226] Under the condition of a 35° phase shift error in channel 29, this invention corrects the actual feed phase using the phase shifter phase error and channel transmission state, and corrects the mutual coupling propagation direction based on the actual phase difference. The overall performance error is 1.98%, lower than the 6.87% of the traditional DeepONet. This result shows that the phase parameters of the active circuit not only affect the radiation phase of a single array element, but also change the mutual coupling propagation state with adjacent array elements. Therefore, it is necessary to simultaneously consider the array element state characteristics and the array element mutual coupling matrix.

[0227] When T / R channels 12 and 13 are not operating, this invention sets the active excitation state and active mutual coupling propagation connection of the corresponding array elements to an inactive state, with a comprehensive performance error of 2.55%, lower than the 8.24% of the traditional DeepONet. The error in this condition is higher than in other conditions because the simultaneous cessation of two adjacent active channels will simultaneously change the local array excitation, port response, and mutual coupling propagation path, but this invention can still maintain a low overall prediction bias.

[0228] The overall performance error includes both return loss and the normalized error corresponding to the active VSWR. Therefore, the results in the table not only reflect radiation performance such as gain, beam direction, beamwidth, sidelobe level, and scanning loss, but also the impact of power amplifier gain changes, phase shift error, and T / R channel status on port performance. By characterizing the initial port response, calculating the port mutual coupling compensation, and correcting the port response, a complete analytical chain from active channel status to port performance indicators can be formed.

[0229] In the parameter sensitivity analysis stage, this invention can separately perturb the power amplifier gain, channel amplitude error, phase shifter phase error, channel propagation delay, operating temperature, and T / R channel state to calculate the parameter sensitivity of each active circuit parameter relative to radiation performance and port performance. In the anomaly location stage, the candidate channel or candidate active circuit parameters are restored to their design nominal state, and the anomaly confidence level is calculated based on the relative decrease in the overall anomaly degree before and after restoration, thereby determining the candidate anomaly array elements and candidate anomaly active circuit parameters.

[0230] In terms of computational efficiency, the average time per set for full-wave electromagnetic simulation is 503.8 s, while the average time per set for this invention is 0.45 s, representing a simulation speedup of approximately 1119.56 times. This result demonstrates that this invention can reduce the computational load of repeatedly performing mesh generation and full-wave electromagnetic solution under multiple operating conditions, while considering active circuit parameters, array element mutual coupling, and port responses.

[0231] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for simulating and analyzing the performance of a phased array antenna, characterized in that, Includes the following steps: S1. Collect phased array antenna parameters and sparse reference electromagnetic simulation data, and obtain the antenna simulation sample set after correlation alignment and standardization processing; S2. Extract the spatial relationships, attributes, feeding states and electromagnetic response relationships of array elements from the antenna simulation sample set, and construct the array topology function representation and array element mutual coupling matrix; S3. Input the array topology function representation and frequency-space query conditions into the improved DeepONet model, perform topology-aware operator mapping, and obtain the initial electromagnetic field operator representation and the initial port response representation. S4. Based on the electromagnetic physical residual and the array element mutual coupling matrix, physical correction and mutual coupling compensation are performed on the two types of initial characterizations to obtain the corrected electromagnetic field results and the corrected port response results. S5. Perform multi-performance index analysis on the corrected electromagnetic field results and the corrected port response results to obtain the phased array antenna performance index vector. S6. Perform error verification between the performance index vector and the reference performance index vector, and update the model calibration parameters to obtain the calibrated improved DeepONet model and performance simulation results. S7. Call the calibrated and improved DeepONet model to perform batch operator inference, and obtain parameter sensitivity analysis results, abnormal performance location results and performance simulation analysis report.

2. The method for performance simulation and analysis of a phased array antenna according to claim 1, characterized in that, S2 includes the following steps: S21. Establish a set of array element nodes according to the array element number, array element spatial coordinates and array element type, and write the array element size, material properties and polarization state into the corresponding array element node. S22. Establish the array element topology edges based on the array element spacing, arrangement relationship and power supply connection relationship to obtain the array topology connection structure; S23. Encode the feeding amplitude, feeding phase and array element working state into array element excitation features, and fuse them with array element node attributes to obtain array element state features. S24. Jointly encode the array element state characteristics and array topology connection structure to form an array topology function representation, and form an array element mutual coupling matrix based on the array element electromagnetic response relationship under the current operating conditions.

3. The method for performance simulation and analysis of a phased array antenna according to claim 2, characterized in that, S3 includes the following steps: S31. The array topology function is used to represent the input array topology-aware branch network. The array element state features are propagated according to the array topology connection structure to obtain the local topology features of the array elements. S32. Perform effective state-weighted pooling and gated fusion on the local topological features of the pair elements to obtain fixed-dimensional branch operator features; S33. Extract the operating frequency, spatial coordinates and scanning direction from the antenna simulation sample set, construct the frequency-space query conditions, and input the frequency-space query conditions into the frequency-space joint backbone network to obtain the backbone query features; S34. Using the adaptive operator fusion unit, the branch operator features and the main query features are fused in a state-related manner to obtain the initial electromagnetic field operator characterization and the initial port response characterization.

4. The method for performance simulation and analysis of a phased array antenna according to claim 3, characterized in that, S34 includes the following steps: S341. Determine the operator query status corresponding to the current working frequency, spatial location, and scanning direction based on the trunk query characteristics; S342. Generate candidate operator features and corresponding state-related weights based on the operator query state, branch operator features, and local topological features of the array elements. S343. Use state-related weights to perform weighted aggregation of candidate operator features, and combine the aggregation results with the main query features dimension by dimension. S344. Perform electromagnetic field component mapping and port response mapping on the dimension-by-dimensional combination results to obtain the initial electromagnetic field operator characterization and the initial port response characterization.

5. The method for performance simulation and analysis of a phased array antenna according to claim 1, characterized in that, S4 includes the following steps: S41. Analyze the electric field components and magnetic field components from the initial electromagnetic field operator characterization, and extract the port response characteristics from the initial port response characterization. S42. Determine the field equation residuals, boundary continuity residuals, and energy deviations based on the electric field components, magnetic field components, medium properties, and boundary conditions. S43. Jointly encode the field equation residuals, boundary continuity residuals, and energy deviations to generate electromagnetic field component correction quantities and obtain the physical correction electromagnetic field results. S44. Based on the propagation of local electromagnetic response and port response of the array element mutual coupling matrix, the corrected electromagnetic field result and the corrected port response result are obtained.

6. The method for performance simulation and analysis of a phased array antenna according to claim 5, characterized in that, S44 includes the following steps: S441. Establish a response mapping index based on the spatial coordinates of the array elements and the port positions, and extract the local electromagnetic response characteristics and port response characteristics of each array element. S442. Generate the mutual coupling propagation coefficient based on the mutual coupling strength, mutual coupling direction, array element distance, and feed phase difference in the array element mutual coupling matrix; S443. By aggregating the two types of response characteristics of adjacent array elements using the mutual coupling propagation coefficient, the spatial field mutual coupling compensation amount and the port mutual coupling compensation amount are obtained. S444. The two types of mutual coupling compensation quantities are respectively fused with the physical correction electromagnetic field results and the initial port response characterization to obtain the correction electromagnetic field results and the correction port response results.

7. The method for performance simulation and analysis of a phased array antenna according to claim 1, characterized in that, S5 includes the following steps: S51. Perform near-field to far-field conversion on the corrected electromagnetic field results to obtain far-field radiation data corresponding to different frequencies and scanning directions. S52. Extract the main beam, side lobes, nulls and polarization features from the far-field radiation data to obtain a set of radiation pattern features; S53. Based on the set of radiation pattern features, analyze the gain, beamwidth, sidelobe level, scanning loss, and front-to-back ratio to obtain the radiation performance indicators; S54. Analyze the active reflection coefficient, return loss, and active VSWR based on the correction port response results, and combine them with radiation performance indicators to form a performance index vector.

8. The method for performance simulation and analysis of a phased array antenna according to claim 1, characterized in that, S6 includes the following steps: S61. Extract reference radiation performance indicators and reference port performance indicators from sparse reference electromagnetic simulation data to form a reference performance indicator vector. S62. Associate and match the performance index vector with the reference performance index vector to obtain the multi-index error set and the corresponding index calibration weights. S63. Based on the indicator calibration weight, fuse the multi-indicator error set to generate operator output correction parameters and performance indicator deviation compensation parameters; S64. Update the corresponding calibration unit based on the two types of calibration parameters and re-infer to obtain the calibrated improved DeepONet model and performance simulation results.

9. The method for performance simulation and analysis of a phased array antenna according to claim 1, characterized in that, S7 includes the following steps: S71. Generate multiple antenna parameter combinations according to the element spacing, element size, feed amplitude and phase of each element, element operating status, operating frequency and scanning direction. S72. Call the calibrated and improved DeepONet model to calculate the performance index vector corresponding to each antenna parameter combination and obtain the batch performance simulation results. S73. Based on the correlation between the changes in antenna parameters and the changes in performance indicators, generate parameter sensitivity analysis results and parameter-performance correlation relationships; S74. Based on the degree of performance index anomaly and the parameter-performance correlation, determine the candidate anomaly array elements and candidate anomaly parameters, use them as the anomaly performance localization results, and generate a performance simulation analysis report.

10. A phased array antenna performance simulation and analysis system, used to implement the phased array antenna performance simulation and analysis method according to any one of claims 1 to 9, characterized in that, include: The data processing module is used to collect phased array antenna parameters and sparse reference electromagnetic simulation data, perform correlation alignment and standardization processing, and obtain an antenna simulation sample set. The topology construction module is used to extract the spatial relationships, attributes, feeding states and electromagnetic response relationships of array elements, and obtain the array topology function representation and array element mutual coupling matrix. The operator mapping module includes an array topology sensing branch network, a frequency-space joint backbone network, and an adaptive operator fusion unit, which are used to generate initial electromagnetic field operator representations and initial port response representations. The electromagnetic field correction module includes a physical residual correction unit and a mutual coupling propagation unit, which are used to perform physical consistency correction and mutual coupling compensation to obtain the corrected electromagnetic field results and the corrected port response results. The performance analysis module includes a multi-performance index decoding unit, which is used to analyze the correction electromagnetic field results and correction port response results to obtain the phased array antenna performance index vector. The model calibration module includes an operator output calibration unit, which is used to update the model calibration parameters based on the error between the performance index vector and the reference performance index vector, so as to obtain the calibrated improved DeepONet model and performance simulation results. The simulation analysis module is used to call the calibrated and improved DeepONet model to perform batch operator inference and obtain parameter sensitivity analysis results and anomaly performance localization results. The report output module is used to associate antenna parameters, performance indicators, parameter sensitivity analysis results, and abnormal performance location results to generate a performance simulation analysis report.

Images