A real-time simulation method for detecting optoelectronic tracking equipment

Abstract

The present invention relates to the field of simulation technology, and specifically discloses a real-time simulation method for detecting photoelectric tracking equipment, comprising: performing feature decoupling processing on photoelectric signals through a dual-channel adaptive neural network architecture, and constructing a device mathematical model based on the processing results using a dynamic gating fusion mechanism; building a multi-physical field coupling simulation environment based on a heterogeneous computing architecture, introducing a device mathematical model, and performing three-field collaborative evolution through light transmission modeling, electromagnetic field distribution calculation, and target motion prediction to generate a dynamic test scenario containing spatiotemporal correlation; the present invention adopts a combination of hybrid feature analysis and fuzzy reasoning to extract multidimensional performance indicators from simulation data, and performs multi-objective dynamic optimization through a strategy combining a quantum evolution algorithm with swarm intelligence optimization; with the help of a digital twin platform and a hardware-in-the-loop interface, real-time verification and closed-loop optimization are achieved, greatly improving the response speed and tracking accuracy of photoelectric tracking equipment in complex environments.

Description

Technical Field

[0001] The invention belongs to the field of simulation technology, and in particular relates to a real-time simulation method for detecting photoelectric tracking equipment. Background Art

[0002] Optoelectronic tracking equipment is widely used in military, aviation, aerospace, and other fields. Its primary function is to accurately detect and track targets. However, with the increasing complexity of the environment, traditional optoelectronic tracking technology faces many challenges, especially in dynamic and complex environments, where its performance is easily affected by various interference factors.

[0003] Most existing simulation methods for optoelectronic tracking devices are based on single physical field modeling, ignoring the interaction between different physical fields. This results in limited accuracy and practicality of the simulation results. Common simulation methods focus on signal processing, but most use simplified linear models, which are difficult to cope with complex scenarios with multiple interferences and noises. In particular, when the signal is affected by electromagnetic interference, atmospheric turbulence and target motion uncertainty, the robustness of traditional methods is poor. In addition, optoelectronic tracking devices usually rely on static optimization schemes, which means that when faced with environmental changes, it is difficult for the device to adapt and adjust its control parameters in real time, resulting in the inability to maintain high tracking accuracy and response speed in dynamic environments. In complex environments, traditional methods often fail to fully consider the coupling effects of multiple physical fields, resulting in insufficient processing capabilities for optoelectronic signals, further affecting the device's anti-interference and real-time response capabilities.

[0004] Therefore, it is necessary to propose a real-time simulation method for detecting photoelectric tracking equipment to solve the problem of how to achieve real-time efficient simulation and dynamic optimization of photoelectric tracking equipment in the prior art.

[0005] The above information disclosed in this background technology is only for enhancing understanding of the background technology of the present invention and therefore it may contain information that does not constitute the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0006] The object of the present invention is to provide a real-time simulation method for detecting a photoelectric tracking device, so as to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A real-time simulation method for detecting a photoelectric tracking device, comprising:

[0009] The photoelectric signal is decoupled and processed using a dual-channel adaptive neural network architecture, and a dynamic gating fusion mechanism is used to build a mathematical model of the device based on the processing results.

[0010] A multi-physics field coupling simulation environment is built based on a heterogeneous computing architecture. The mathematical model of the device is introduced. Through light transmission modeling, electromagnetic field distribution calculation, and target motion prediction, three fields are co-evolved to generate dynamic test scenarios with spatiotemporal correlations.

[0011] Based on dynamic test scenarios, we obtain simulation data of multi-dimensional physical fields. We use a method combining hybrid feature analysis and fuzzy reasoning to extract multi-dimensional performance indicators from the simulation data and establish parameter optimization mapping relationships.

[0012] Based on the parameter optimization mapping relationship, a hybrid strategy integrating quantum evolutionary algorithm and swarm intelligence optimization is adopted to perform multi-objective dynamic optimization of control parameters through the environment perception weighting mechanism to generate multi-objective dynamic optimization results;

[0013] A virtual-real data interaction channel is built through the digital twin platform, the multi-objective dynamic optimization results are input into the photon-level simulation engine, and the closed-loop iterative optimization of the control algorithm is completed through real-time verification through the hardware-in-the-loop interface.

[0014] Preferably, the dual-channel adaptive neural network architecture includes a spatial noise processing channel and a time-varying interference processing channel;

[0015] Using the spatial noise processing channel to perform non-local mean denoising on the photoelectric signal, retaining high-frequency edge features, and obtaining a denoised photoelectric signal;

[0016] The denoised photoelectric signal is input into a fourth-order residual network to extract shallow texture features, and the cross-attention module of SwinTransformer is used to perform global context enhancement on the shallow texture features to generate a feature matrix;

[0017] Processing the feature matrix through multi-scale spatial convolution and adaptive entropy weighted fusion to generate a noise distribution map;

[0018] Processing the noise distribution map through a pre-trained conditional generative adversarial network to generate a pixel-level noise compensation signal;

[0019] generating a time-frequency spectrum by performing a short-time Fourier transform on the photoelectric signal using the time-varying interference processing;

[0020] Based on the time-spectrogram, extracting temporal features through a causal dilation convolution layer of a probabilistic temporal convolutional network;

[0021] Probabilistically modeling the phase jitter of the time series feature through a variational autoencoder to generate an interference feature code;

[0022] The interference feature code is post-processed using a Kalman smoothing algorithm to generate a time-varying interference suppression compensation signal.

[0023] Preferably, the step of constructing a device mathematical model based on the processing results using a dynamic gating fusion mechanism includes:

[0024] Performing feature extraction on the pixel-level noise compensation signal and the time-varying interference suppression compensation signal respectively to generate a spatial feature vector and a time-varying feature vector;

[0025] Based on the spatial feature vector and the time-varying feature vector, a dual-channel feature similarity matrix is constructed using a multi-layer perceptron and a cosine similarity function;

[0026] Based on the dual-channel feature similarity matrix, a policy gradient reinforcement learning network is used to generate dynamic fusion weights using the weighted tracking error entropy and control efficiency as the reward function;

[0027] Designing the gating function Perform probabilistic channel selection, maximize cumulative rewards, and output device mathematical models;

[0028] ;

[0029] Where, is the action-value function, Select an action for the channel, is the temperature coefficient.

[0030] Preferably, the building of a multi-physics field coupling simulation environment based on a heterogeneous computing architecture includes:

[0031] Based on the configuration information of heterogeneous computing architecture, the adaptive resource scheduling algorithm is used to dynamically identify and initialize the heterogeneous computing architecture;

[0032] Based on the initialized heterogeneous computing architecture, a load balancing algorithm is used to predict and allocate various computing resources according to the requirements of multi-physics field coupling simulation;

[0033] Introducing mathematical models and interaction rule sets for various physical fields, analyzing spatiotemporal correlation parameters through intelligent fusion algorithms, and establishing a multi-physics field coupling topology architecture;

[0034] Based on the simulation target constraints and historical data sets, the intelligent optimization algorithm is triggered to retrieve the strategy library and generate the parameter configuration table for each physical field;

[0035] The configuration parameters of each physical field are input into the established multi-physics field coupling topology architecture, and the data exchange protocol is defined, and a time step alignment mechanism for each physical field is established to form a multi-physics field coupling simulation environment.

[0036] Preferably, the light transmission modeling includes:

[0037] Initialize atmospheric environment parameters to construct a dynamic meteorological field, and use an improved Mie scattering model and GPU-accelerated Monte Carlo ray tracing algorithm to generate a light transmission attenuation field that changes dynamically with the meteorological field;

[0038] Through the dynamic texture synthesis algorithm driven by weather radar data, the turbulence intensity distribution is modulated in real time and the light transmission attenuation field parameters are updated;

[0039] The electromagnetic field distribution calculation includes:

[0040] Construct a coupling model of the mixed potential integral equation and the equivalent dipole method to divide the high- and low-frequency calculation domains;

[0041] Based on the high and low frequency computational domains, OpenMP is used to parallelize the low frequency electromagnetic field, and cuFFT is used to accelerate the solution of high frequency components to generate the full frequency band electromagnetic field distribution.

[0042] Based on the full-band electromagnetic field distribution, a multivariable coupling solver is used to perform real-time mapping of electromagnetic field and flow field data, and a time-step alignment mechanism is used to perform spatiotemporal consistency alignment of multiple physical fields.

[0043] The target motion prediction includes:

[0044] The LSTM-VAM encoder is trained using the target's historical trajectory data to generate a sequence of latent variables for motion intention.

[0045] Through inverse reinforcement learning, the reward function corresponding to the latent variables is analyzed and a Markov decision process based on maximum entropy is constructed.

[0046] Model predictive control is used to optimize the Markov decision process in a rolling manner to generate trajectory prediction results within a preset time period.

[0047] Preferably, the method of obtaining simulation data of a multi-dimensional physical field based on a dynamic test scenario, extracting multi-dimensional performance indicators from the simulation data and establishing a parameter optimization mapping relationship by combining hybrid feature analysis with fuzzy reasoning, includes:

[0048] The wavelet packet transform algorithm is used to decompose the simulation data in the time-frequency domain, and the nonlinear feature subset is extracted through kernel principal component analysis to generate a mixed feature matrix;

[0049] Define input / output language variables, initialize fuzzy set membership functions based on expert rule base, and build fuzzy inference system;

[0050] A Mamdani inference engine and a centroid defuzzifier are configured, and the mixed feature matrix is input into the fuzzy inference system to train the membership function parameters, thereby generating a mapping relationship between feature indicators and control parameters.

[0051] Preferably, the method comprises: based on the parameter optimization mapping relationship, adopting a hybrid strategy integrating quantum evolutionary algorithm and swarm intelligence optimization, performing multi-objective dynamic optimization of control parameters through an environment perception weighting mechanism, and generating a multi-objective dynamic optimization result, including:

[0052] The control parameter population is initialized using the quantum bit encoding strategy, and the initial quantum state sequence is generated through chaotic logistic mapping;

[0053] Based on the initial quantum state training, a dual-objective optimization function is defined: the root mean square of the tracking error and the rate of change of the control variable, and an initial range of the quantum gate rotation angle is set;

[0054] Initialize Pareto frontier archive and set crowding distance threshold , establish a non-dominated sorting layer;

[0055] Based on the initial range of the quantum gate rotation angle, the quantum rotation gate operation is performed on the current population, and the phase update amount Dynamic adjustment based on individual crowding ranking:

[0056] ;

[0057] Collapse the quantum state into a classical parameter solution, calculate the value of the dual-objective optimization function, and perform non-dominated sorting based on the calculation results;

[0058] Extracting parameter encoding patterns of the top 10% individuals based on the non-dominated sorting, and constructing a graph structure containing historical optimization paths;

[0059] The graph attention network is used to learn the associated features of graph structure parameters, and the Transformer encoder is used to generate feature embedding vectors. The feature embedding vectors are input into the multi-layer perceptron to generate dynamic weight distribution.

[0060] Design weighting function Optimizing the real-time allocation of the target dynamic weight distribution and outputting the multi-objective dynamic optimization results;

[0061] ;

[0062] Where, For the The optimization goal is at time The weight of For the The feature embedding vector corresponding to the target, is the real-time environment feature vector, is the vector concatenation operation, is the learnable parameter vector.

[0063] Preferably, the process of constructing a virtual-real data interaction channel through a digital twin platform, inputting the multi-objective dynamic optimization results into a photon-level simulation engine, and completing the closed-loop iterative optimization of the control algorithm through real-time verification through a hardware-in-the-loop interface includes:

[0064] Import the optical CAD model into the digital twin platform, configure the spectral sampling parameters, and initialize the photon tracking parameters;

[0065] Based on the initialized photon tracing parameters, photon tracing data is generated through the Jones matrix and Monte Carlo algorithm, and the scattered field distribution is output to the hardware-in-the-loop interface in combination with the bidirectional reflectance distribution function model;

[0066] Configure an FPGA module at the hardware-in-the-loop interface to perform nanosecond-level real-time simulation, and use a timestamp alignment algorithm to compensate for data transmission delays.

[0067] Perform hardware frequency response calibration using chirp signal injection, generate compensation filters, and load them into the FPGA.

[0068] A bidirectional channel for virtual and real data is constructed, and the unscented Kalman filter algorithm is used for multi-target state estimation and deviation prediction. The deviation compensation function is established by combining differential homeomorphism mapping to generate control parameter corrections.

[0069] Based on the control parameter correction amount, the covariance matrix adaptive evolution strategy is used to evolve the control parameters online. The iteration period is set to 20ms to complete the closed-loop optimization process.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] The present invention adopts a dual-channel adaptive neural network architecture to perform feature decoupling processing on the photoelectric signal, and combines it with a dynamic gating fusion mechanism to build a mathematical model of the device, accurately processing different types of signal interference and improving the anti-noise capability of the photoelectric signal; based on the multi-physical field coupling simulation environment built on the heterogeneous computing architecture, it comprehensively considers the interaction of multiple physical factors such as light transmission, electromagnetic field distribution and target motion, and generates dynamic test scenarios with spatiotemporal correlation, thereby enhancing simulation accuracy and adaptability.

[0072] This invention adopts a combination of hybrid feature analysis and fuzzy reasoning to extract multidimensional performance indicators from simulation data, and performs multi-objective dynamic optimization through a strategy that combines quantum evolution algorithm with swarm intelligence optimization, thereby improving the real-time performance and accuracy of control parameter optimization; with the help of digital twin platform and hardware-in-the-loop interface, real-time verification and closed-loop optimization are realized, greatly improving the response speed and tracking accuracy of optoelectronic tracking equipment in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1This is a flow chart of the real-time simulation method for detecting photoelectric tracking equipment of the present invention. DETAILED DESCRIPTION

[0074] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0075] Example 1:

[0076] See also Figure 1 As shown, a real-time simulation method for detecting an optoelectronic tracking device includes:

[0077] The photoelectric signal is decoupled and processed using a dual-channel adaptive neural network architecture, and a dynamic gating fusion mechanism is used to build a mathematical model of the device based on the processing results.

[0078] The dual-channel adaptive neural network architecture includes a spatial noise processing channel and a time-varying interference processing channel;

[0079] The spatial noise processing channel is used to perform non-local mean denoising on the photoelectric signal, retaining the high-frequency edge features to obtain the denoised photoelectric signal;

[0080] The denoised photoelectric signal is input into a fourth-order residual network to extract shallow texture features, and the cross-attention module of SwinTransformer is used to perform global context enhancement on the shallow texture features to generate a feature matrix;

[0081] The feature matrix is processed through multi-scale spatial convolution and adaptive entropy weighted fusion to generate a noise distribution map;

[0082] The noise distribution map is processed through a pre-trained conditional generative adversarial network to generate a pixel-level noise compensation signal;

[0083] The time-frequency spectrum is generated by performing short-time Fourier transform on the photoelectric signal using time-varying interference processing;

[0084] Based on the time-spectrogram, the causal dilation convolution layer of the probabilistic temporal convolutional network is used to extract time series features. The phase jitter of the time series features is probabilistically modeled with a variational autoencoder to generate interference feature codes.

[0085] The Kalman smoothing algorithm is used to post-process the interference feature code to generate a time-varying interference suppression compensation signal.

[0086] Furthermore, a dual-channel adaptive neural network architecture efficiently decouples and optimizes the features of photoelectric signals. Combined with noise removal, time-varying interference suppression, feature enhancement, and noise compensation, this technology enables precise photoelectric signal processing. By independently processing spatial noise and time-varying interference, signal quality is effectively improved, enhancing the signal's edge and timing characteristics. This dynamically optimizes device performance, thereby enhancing the accuracy, adaptability, and robustness of photoelectric tracking equipment.

[0087] Feature extraction is performed on the pixel-level noise compensation signal and the time-varying interference suppression compensation signal respectively to generate a spatial feature vector and a time-varying feature vector;

[0088] Based on the spatial eigenvector and the time-varying eigenvector, a dual-channel feature similarity matrix is constructed using a multi-layer perceptron and a cosine similarity function.

[0089] Based on the dual-channel feature similarity matrix, a policy gradient reinforcement learning network is used to generate dynamic fusion weights using the weighted tracking error entropy and control efficiency as the reward function;

[0090] Design a gating function for probabilistic channel selection, maximize cumulative rewards, and output the device mathematical model.

[0091] Furthermore, by extracting and analyzing the features of the noise compensation and interference suppression signals and optimizing dynamic fusion weights using a reinforcement learning network, they achieved intelligent fusion and adaptive adjustment of signal features. Probabilistic channel selection through gating functions effectively improved the accuracy and efficiency of signal processing, reduced tracking errors, and optimized control performance, thereby generating a precise device mathematical model and enhancing the overall performance and adaptability of optoelectronic devices.

[0092] A multi-physics field coupling simulation environment is built based on a heterogeneous computing architecture. The mathematical model of the device is introduced. Through light transmission modeling, electromagnetic field distribution calculation, and target motion prediction, three fields are co-evolved to generate dynamic test scenarios with spatiotemporal correlations.

[0093] Based on the configuration information of heterogeneous computing architecture, the adaptive resource scheduling algorithm is used to dynamically identify and initialize the heterogeneous computing architecture;

[0094] Based on the initialized heterogeneous computing architecture, a load balancing algorithm is used to predict and allocate various computing resources according to the requirements of multi-physics field coupling simulation;

[0095] Introducing mathematical models and interaction rule sets for various physical fields, analyzing spatiotemporal correlation parameters through intelligent fusion algorithms, and establishing a multi-physics field coupling topology architecture;

[0096] Based on the simulation target constraints and historical data sets, the intelligent optimization algorithm is triggered to retrieve the strategy library and generate the parameter configuration table for each physical field;

[0097] The configuration parameters of each physical field are input into the established multi-physics field coupling topology architecture, and the data exchange protocol is defined, and a time step alignment mechanism for each physical field is established to form a multi-physics field coupling simulation environment.

[0098] Furthermore, by constructing a multi-physics field coupling simulation environment based on a heterogeneous computing architecture, the three fields of light transmission, electromagnetic field distribution, and target motion are co-evolved, generating dynamic spatiotemporal correlation test scenarios. Computing resource allocation is optimized through adaptive resource scheduling and load balancing algorithms. Combined with intelligent fusion and optimization algorithms, it is possible to efficiently analyze and configure the parameters of each physical field, forming a precise multi-physics field coupling topology architecture. This approach not only improves simulation accuracy and computational efficiency, but also supports the dynamic adjustment and optimization of complex systems through intelligent optimization strategies, significantly enhancing the adaptability and performance of the simulation environment.

[0099] (1) Light transmission modeling

[0100] Initialize atmospheric environment parameters to construct a dynamic meteorological field, and use an improved Mie scattering model and GPU-accelerated Monte Carlo ray tracing algorithm to generate a light transmission attenuation field that changes dynamically with the meteorological field;

[0101] Through the dynamic texture synthesis algorithm driven by weather radar data, the turbulence intensity distribution is modulated in real time and the light transmission attenuation field parameters are updated;

[0102] (2) Calculation of electromagnetic field distribution

[0103] Construct a coupling model of the mixed potential integral equation and the equivalent dipole method to divide the high- and low-frequency calculation domains;

[0104] Based on the high and low frequency computational domains, OpenMP is used to parallelize the low frequency electromagnetic field, and cuFFT is used to accelerate the solution of high frequency components to generate the full frequency band electromagnetic field distribution.

[0105] Based on the full-band electromagnetic field distribution, a multivariable coupling solver is used to perform real-time mapping of electromagnetic field and flow field data, and a time-step alignment mechanism is used to perform spatiotemporal consistency alignment of multiple physical fields.

[0106] (3) Target motion prediction

[0107] The LSTM-VAM encoder is trained using the target's historical trajectory data to generate a sequence of latent variables for motion intention.

[0108] Through inverse reinforcement learning, the reward function corresponding to the latent variables is analyzed and a Markov decision process based on maximum entropy is constructed.

[0109] Model predictive control is used to optimize the Markov decision process in a rolling manner to generate trajectory prediction results within a preset time period.

[0110] Furthermore, through the comprehensive optimization of three aspects: light transmission modeling, electromagnetic field distribution calculation, and target motion prediction, high-precision, multi-physics field dynamic simulation and prediction are achieved. Light transmission modeling utilizes the atmospheric environment and a GPU-accelerated Monte Carlo ray tracing algorithm to dynamically update the light attenuation field. Electromagnetic field calculation combines high- and low-frequency computing with parallel acceleration technology to accurately simulate the full-band electromagnetic field distribution and map electromagnetic field and flow field data in real time. Target motion prediction optimizes trajectory prediction accuracy through a combination of deep learning and reinforcement learning. Overall, the technology achieves efficient physical field coupling and spatiotemporal consistency alignment, significantly improving the accuracy, real-time performance, and adaptability of the simulation system.

[0111] Based on dynamic test scenarios, we obtain simulation data of multi-dimensional physical fields. We use a method combining hybrid feature analysis and fuzzy reasoning to extract multi-dimensional performance indicators from the simulation data and establish parameter optimization mapping relationships.

[0112] The wavelet packet transform algorithm is used to decompose the simulation data in the time-frequency domain, and the nonlinear feature subset is extracted through kernel principal component analysis to generate a mixed feature matrix;

[0113] Define input / output language variables, initialize fuzzy set membership functions based on expert rule base, and build fuzzy inference system;

[0114] The Mamdani inference engine and the center of gravity method defuzzifier are configured, and the mixed feature matrix is input into the fuzzy inference system to train the membership function parameters and generate the mapping relationship from the feature index to the control parameters.

[0115] Furthermore, by combining multidimensional physical field simulation data obtained from dynamic test scenarios with hybrid feature analysis and fuzzy reasoning, they achieved precise extraction and optimization of system performance. Wavelet packet transform and kernel principal component analysis were used to extract nonlinear features, construct a hybrid feature matrix, and utilize a fuzzy reasoning system to model and optimize the mapping relationship between features and control parameters. The resulting optimized mapping relationship accurately reflects the complex relationship between multidimensional performance indicators and control parameters.

[0116] Based on the parameter optimization mapping relationship, a hybrid strategy integrating quantum evolutionary algorithm and swarm intelligence optimization is adopted to perform multi-objective dynamic optimization of control parameters through the environment perception weighting mechanism to generate multi-objective dynamic optimization results;

[0117] The control parameter population is initialized using the quantum bit encoding strategy, and the initial quantum state sequence is generated through chaotic logistic mapping;

[0118] Based on the initial quantum state training, a dual-objective optimization function is defined: the root mean square of the tracking error and the rate of change of the control variable, and the initial range of the quantum gate rotation angle is set;

[0119] Initialize the Pareto frontier archive, set the crowding distance threshold, and establish the non-dominated sorting layer;

[0120] Based on the initial range of quantum gate rotation angle, the quantum rotating gate operation is performed on the current population, and the phase update amount is dynamically adjusted by the individual crowding ranking;

[0121] Collapse the quantum state into a classical parameter solution, calculate the value of the dual-objective optimization function, and perform non-dominated sorting based on the calculation results;

[0122] Extract the parameter encoding patterns of the top 10% individuals based on non-dominated sorting and construct a graph structure containing historical optimization paths;

[0123] The graph attention network is used to learn the associated features of graph structure parameters, and the Transformer encoder is used to generate feature embedding vectors. The feature embedding vectors are then input into the multi-layer perceptron to generate dynamic weight distribution.

[0124] The weighting function is designed to optimize the real-time allocation of dynamic weight distribution of the target and output the multi-objective dynamic optimization results.

[0125] Furthermore, by integrating a quantum evolutionary algorithm with a swarm intelligence optimization strategy and incorporating an environmentally aware weighting mechanism, they achieved multi-objective dynamic optimization of control parameters. Through quantum bit encoding, chaotic mapping, and quantum rotation operations, they optimized the dual objective function (tracking error and rate of change of the control variable), and extracted the optimal solution through non-dominated sorting and Pareto frontier optimization. By combining a graph attention network with a Transformer encoder, they dynamically learned parameter association features and implemented dynamic weight assignment of optimization objectives in real time through a weighting function, generating efficient multi-objective optimization results.

[0126] By building a virtual-real data interaction channel through the digital twin platform, the multi-objective dynamic optimization results are input into the photon-level simulation engine, and the closed-loop iterative optimization of the control algorithm is completed through real-time verification through the hardware-in-the-loop interface;

[0127] Import the optical CAD model into the digital twin platform, configure the spectral sampling parameters, and initialize the photon tracking parameters;

[0128] Based on the initialized photon tracing parameters, the photon tracing data is generated through the Jones matrix and Monte Carlo algorithm, and the scattered field distribution is output to the hardware-in-the-loop interface in combination with the bidirectional reflectance distribution function model;

[0129] Configure an FPGA module at the hardware-in-the-loop interface to perform nanosecond-level real-time simulation, and use a timestamp alignment algorithm to compensate for data transmission delays.

[0130] Perform hardware frequency response calibration using chirp signal injection, generate compensation filters, and load them into the FPGA.

[0131] A bidirectional channel for virtual and real data is constructed, and the unscented Kalman filter algorithm is used for multi-target state estimation and deviation prediction. The deviation compensation function is established by combining differential homeomorphism mapping to generate control parameter corrections.

[0132] Based on the control parameter correction amount, the covariance matrix adaptive evolution strategy is used to evolve the control parameters online. The iteration period is set to 20ms to complete the closed-loop optimization process.

[0133] Furthermore, a digital twin platform and a photonic-level simulation engine enable real-time interaction and closed-loop optimization of virtual and real data. The accuracy and efficiency of the control algorithm are verified and optimized using optical CAD models, photon tracking data, and hardware-in-the-loop interfaces. Nanosecond-level real-time simulation and a timestamp alignment algorithm, combined with an FPGA module, ensure compensation for data transmission delays. Unscented Kalman filtering and diffeomorphism mapping are used for multi-objective state estimation and deviation prediction. Control parameters are then corrected and online optimization is achieved using a covariance matrix adaptive evolutionary strategy.

[0134] Example 2:

[0135] Application example: Real-time simulation of the working status of photoelectric tracking equipment

[0136] Optoelectronic tracking devices play a vital role in military, aviation, and aerospace applications, requiring precise detection and tracking of targets. However, complex and changing environmental conditions, such as electromagnetic interference, atmospheric turbulence, and uncertainty in target motion, pose significant challenges to their performance. To evaluate and optimize the performance of optoelectronic tracking devices in these complex environments, we employed a real-time simulation method based on a dual-channel adaptive neural network architecture and a multi-physics coupled simulation environment.

[0137] 1. Application Goals and Requirements

[0138] (1) Accurately simulate complex environments: simulate multi-physics environments including electromagnetic interference, atmospheric turbulence, and dynamic target motion.

[0139] (2) Real-time processing of photoelectric signals: Real-time processing of photoelectric signals, including denoising, feature extraction and interference suppression.

[0140] (3) Dynamic optimization of control parameters: Based on the simulation results, the control parameters of the optoelectronic tracking device are dynamically optimized to improve the tracking accuracy and response speed.

[0141] (4) Closed-loop iterative optimization: The control algorithm is verified in real time through the hardware-in-the-loop interface to achieve closed-loop iterative optimization.

[0142] 2. Application steps

[0143] 1. Build a dual-channel adaptive neural network architecture

[0144] ①Spatial noise processing channel:

[0145] The photoelectric signal is subjected to non-local mean denoising to retain high-frequency edge features.

[0146] The denoised signal is input into a fourth-order residual network to extract shallow texture features, and the cross-attention module of the Swin Transformer is used for global context enhancement.

[0147] The noise distribution map is generated through multi-scale spatial convolution and adaptive entropy weighted fusion, and then processed by the conditional generative adversarial network to generate a pixel-level noise compensation signal.

[0148] ②Time-varying interference processing channel:

[0149] Perform short-time Fourier transform on the photoelectric signal to generate a time-frequency spectrum.

[0150] The causal dilation convolution layer of the probabilistic temporal convolutional network is used to extract temporal features, and the phase jitter is probabilistically modeled in combination with a variational autoencoder.

[0151] The Kalman smoothing algorithm is used to post-process the interference feature code to generate a time-varying interference suppression compensation signal.

[0152] 2. Build a multi-physics coupling simulation environment

[0153] (1) Initializing heterogeneous computing architecture: Based on configuration information, the heterogeneous computing architecture is dynamically identified and initialized through an adaptive resource scheduling algorithm.

[0154] (2) Allocate computing resources: Based on the requirements of multi-physics field coupling simulation, a load balancing algorithm is used to predict and allocate various computing resources.

[0155] (3) Introducing physical field models: Introducing mathematical models and interaction rule sets for light transmission, electromagnetic field distribution, and target motion prediction.

[0156] (4) Establish a coupled topological architecture: Analyze the spatiotemporal correlation parameters through intelligent fusion algorithms and establish a multi-physics field coupled topological architecture.

[0157] (5) Generate dynamic test scenarios: Generate dynamic test scenarios with spatiotemporal correlations based on simulation target constraints and historical data sets.

[0158] 3. Light transmission modeling, electromagnetic field distribution calculation, and target motion prediction

[0159] ① Light transmission modeling:

[0160] Initialize atmospheric environment parameters and construct a dynamic meteorological field.

[0161] An improved Mie scattering model and GPU-accelerated Monte Carlo ray tracing algorithm are used to generate a light transmission attenuation field that changes dynamically with the meteorological field.

[0162] The turbulence intensity distribution is modulated in real time through a dynamic texture synthesis algorithm driven by weather radar data.

[0163] ② Calculation of electromagnetic field distribution:

[0164] A coupling model of the mixed potential integral equation and the equivalent dipole method is constructed to divide the high-frequency and low-frequency calculation domains.

[0165] OpenMP is used to parallelize the computation of low-frequency electromagnetic fields, and cuFFT is used to accelerate the computation of high-frequency components.

[0166] Perform real-time mapping of electromagnetic field and flow field data to achieve spatial and temporal consistency of multi-physics fields.

[0167] ③Target motion prediction:

[0168] The LSTM-VAM encoder is used to generate the motion intention latent variable sequence.

[0169] Constructing a maximum entropy-based Markov decision process through inverse reinforcement learning.

[0170] Rolling optimization trajectory prediction results using model predictive control.

[0171] 4. Extract multidimensional performance indicators and establish parameter optimization mapping relationships

[0172] (1) Hybrid feature analysis: The wavelet packet transform algorithm is used to decompose the simulation data in the time-frequency domain, and the nonlinear feature subset is extracted through kernel principal component analysis.

[0173] (2) Fuzzy reasoning: Construct a fuzzy reasoning system, train the membership function parameters, and generate a mapping relationship from characteristic indicators to control parameters.

[0174] 5. Multi-objective dynamic optimization control parameters

[0175] (1) Initialize the quantum state: Use the quantum bit encoding strategy to initialize the control parameter population, and generate the initial quantum state sequence through chaotic logistic mapping.

[0176] (2) Define the optimization function: Define the dual-objective optimization function (the root mean square of the tracking error and the rate of change of the control variable) and set the initial range of the quantum gate rotation angle.

[0177] (3) Perform quantum rotating gate operation: Perform quantum rotating gate operation on the current population based on the initial range of quantum gate rotation angle, and the phase update amount is dynamically adjusted by the individual crowding ranking.

[0178] (4) Generate optimization results: collapse the quantum state into a classical parameter solution, calculate the value of the dual-objective optimization function, perform non-dominated sorting, and output the multi-objective dynamic optimization results.

[0179] 6. Closed-loop iterative optimization

[0180] (1) Constructing a virtual-real data interaction channel: Importing the optical CAD model into the digital twin platform and initializing the photon tracking parameters.

[0181] (2) Hardware-in-the-loop interface: Configure the FPGA module to perform nanosecond-level real-time simulation and use a timestamp alignment algorithm to compensate for data transmission delays.

[0182] (3) Frequency response calibration: Perform hardware frequency response calibration through chirp signal injection method, generate compensation filter and load it into FPGA.

[0183] (4) State estimation and deviation prediction: The unscented Kalman filter algorithm is used for multi-target state estimation and deviation prediction, and the deviation compensation function is established by combining differential homeomorphism mapping.

[0184] (5) Online evolution of control parameters: Based on the control parameter correction amount, the covariance matrix adaptive evolution strategy is used to evolve the control parameters online, and the iteration period is set to 20ms.

[0185] 3. Application Effect

[0186] This real-time simulation method enables us to accurately simulate the operating conditions of optoelectronic tracking devices in complex environments, process optoelectronic signals in real time, and dynamically optimize control parameters. This not only improves the tracking accuracy and response speed of the optoelectronic tracking device, but also enhances its adaptability and robustness in complex environments. Furthermore, through closed-loop iterative optimization, we achieve continuous optimization of the control algorithm, further enhancing the overall performance of the device.

[0187] Example 3:

[0188] An embodiment of the present invention further provides a computer-readable storage medium storing a program for a real-time simulation method for detecting an optoelectronic tracking device, such as any of the above-described methods. When executed by a processor, the program implements each process of the above-described real-time simulation method embodiments and achieves the same technical effects. To avoid repetition, the program is not further described here. The computer-readable storage medium may be, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0189] In the description of this specification, the reference terms "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0190] The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to common designs. In the absence of conflicts, the same embodiment and different embodiments of the present invention may be combined with each other.

[0191] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0192] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A real-time simulation method for detecting photoelectric tracking equipment, characterized in that: include: The photoelectric signal is decoupled and processed using a dual-channel adaptive neural network architecture, and a dynamic gating fusion mechanism is used to build a mathematical model of the device based on the processing results. A multi-physics field coupling simulation environment is built based on a heterogeneous computing architecture. The mathematical model of the device is introduced. Through light transmission modeling, electromagnetic field distribution calculation, and target motion prediction, three fields are co-evolved to generate dynamic test scenarios with spatiotemporal correlations. Based on dynamic test scenarios, we obtain simulation data of multi-dimensional physical fields. We use a method combining hybrid feature analysis and fuzzy reasoning to extract multi-dimensional performance indicators from the simulation data and establish parameter optimization mapping relationships. Based on the parameter optimization mapping relationship, a hybrid strategy integrating quantum evolutionary algorithm and swarm intelligence optimization is adopted to perform multi-objective dynamic optimization of control parameters through the environment perception weighting mechanism to generate multi-objective dynamic optimization results; By building a virtual-real data interaction channel through the digital twin platform, the multi-objective dynamic optimization results are input into the photon-level simulation engine, and the closed-loop iterative optimization of the control algorithm is completed through real-time verification through the hardware-in-the-loop interface; The dual-channel adaptive neural network architecture includes a spatial noise processing channel and a time-varying interference processing channel; Using the spatial noise processing channel to perform non-local mean denoising on the photoelectric signal, retaining high-frequency edge features, and obtaining a denoised photoelectric signal; The denoised photoelectric signal is input into a fourth-order residual network to extract shallow texture features, and the cross-attention module of the Swin Transformer is used to perform global context enhancement on the shallow texture features to generate a feature matrix; Processing the feature matrix through multi-scale spatial convolution and adaptive entropy weighted fusion to generate a noise distribution map; Processing the noise distribution map through a pre-trained conditional generative adversarial network to generate a pixel-level noise compensation signal; generating a time-frequency spectrum by performing a short-time Fourier transform on the photoelectric signal using the time-varying interference processing; Based on the time-spectrogram, extracting temporal features through a causal dilation convolution layer of a probabilistic temporal convolutional network; Probabilistically modeling the phase jitter of the time series feature through a variational autoencoder to generate an interference feature code; The interference feature code is post-processed using a Kalman smoothing algorithm to generate a time-varying interference suppression compensation signal.

2. A real-time simulation method for detecting a photoelectric tracking device according to claim 1, characterized in that: The step of constructing a device mathematical model based on the processing results using a dynamic gating fusion mechanism includes: Performing feature extraction on the pixel-level noise compensation signal and the time-varying interference suppression compensation signal respectively to generate a spatial feature vector and a time-varying feature vector; Based on the spatial feature vector and the time-varying feature vector, a dual-channel feature similarity matrix is constructed using a multi-layer perceptron and a cosine similarity function; Based on the dual-channel feature similarity matrix, a policy gradient reinforcement learning network is used to generate dynamic fusion weights using the weighted tracking error entropy and control efficiency as the reward function; Designing the gating function Perform probabilistic channel selection, maximize cumulative rewards, and output device mathematical models; ； Where, is the action-value function, Select an action for the channel, is the temperature coefficient.

3. A real-time simulation method for detecting a photoelectric tracking device according to claim 2, characterized in that: The multi-physics field coupling simulation environment is built based on a heterogeneous computing architecture, including: Based on the configuration information of heterogeneous computing architecture, the adaptive resource scheduling algorithm is used to dynamically identify and initialize the heterogeneous computing architecture; Based on the initialized heterogeneous computing architecture, a load balancing algorithm is used to predict and allocate various computing resources according to the requirements of multi-physics field coupling simulation; Introducing mathematical models and interaction rule sets for various physical fields, analyzing spatiotemporal correlation parameters through intelligent fusion algorithms, and establishing a multi-physics field coupling topology architecture; Based on the simulation target constraints and historical data sets, the intelligent optimization algorithm is triggered to retrieve the strategy library and generate the parameter configuration table for each physical field; The configuration parameters of each physical field are input into the established multi-physics field coupling topology architecture, and the data exchange protocol is defined, and a time step alignment mechanism for each physical field is established to form a multi-physics field coupling simulation environment.

4. A real-time simulation method for detecting a photoelectric tracking device according to claim 3, characterized in that: The light transport modeling includes: Initialize atmospheric environment parameters to construct a dynamic meteorological field, and use an improved Mie scattering model and GPU-accelerated Monte Carlo ray tracing algorithm to generate a light transmission attenuation field that changes dynamically with the meteorological field; Through the dynamic texture synthesis algorithm driven by weather radar data, the turbulence intensity distribution is modulated in real time and the light transmission attenuation field parameters are updated; The electromagnetic field distribution calculation includes: Construct a coupling model of the mixed potential integral equation and the equivalent dipole method to divide the high- and low-frequency calculation domains; Based on the high and low frequency computational domains, OpenMP is used to parallelize the low frequency electromagnetic field, and cuFFT is used to accelerate the solution of high frequency components to generate the full frequency band electromagnetic field distribution. Based on the full-band electromagnetic field distribution, a multivariable coupling solver is used to perform real-time mapping of electromagnetic field and flow field data, and a time-step alignment mechanism is used to perform spatiotemporal consistency alignment of multiple physical fields. The target motion prediction includes: The LSTM-VAM encoder is trained using the target's historical trajectory data to generate a sequence of latent variables for motion intention. Through inverse reinforcement learning, the reward function corresponding to the latent variables is analyzed and a Markov decision process based on maximum entropy is constructed. Model predictive control is used to optimize the Markov decision process in a rolling manner to generate trajectory prediction results within a preset time period.

5. A real-time simulation method for detecting a photoelectric tracking device according to claim 4, characterized in that: The method of obtaining simulation data of a multi-dimensional physical field based on a dynamic test scenario, extracting multi-dimensional performance indicators from the simulation data and establishing a parameter optimization mapping relationship by combining hybrid feature analysis with fuzzy reasoning, includes: The wavelet packet transform algorithm is used to decompose the simulation data in the time-frequency domain, and the nonlinear feature subset is extracted through kernel principal component analysis to generate a mixed feature matrix; Define input / output language variables, initialize fuzzy set membership functions based on expert rule base, and build fuzzy inference system; A Mamdani inference engine and a centroid defuzzifier are configured, and the mixed feature matrix is input into the fuzzy inference system to train the membership function parameters, thereby generating a mapping relationship between feature indicators and control parameters.

6. A real-time simulation method for detecting a photoelectric tracking device according to claim 5, characterized in that: The method uses a hybrid strategy that integrates quantum evolutionary algorithm and swarm intelligence optimization based on parameter optimization mapping relationship, performs multi-objective dynamic optimization of control parameters through an environment perception weighting mechanism, and generates multi-objective dynamic optimization results, including: The control parameter population is initialized using the quantum bit encoding strategy, and the initial quantum state sequence is generated through chaotic logistic mapping; Based on the initial quantum state training, a dual-objective optimization function is defined: the root mean square of the tracking error and the rate of change of the control variable, and the initial range of the quantum gate rotation angle is set; Initialize Pareto frontier archive and set crowding distance threshold , establish a non-dominated sorting layer; Based on the initial range of the quantum gate rotation angle, the quantum rotation gate operation is performed on the current population, and the phase update amount Dynamic adjustment based on individual crowding ranking: ； Collapse the quantum state into a classical parameter solution, calculate the value of the dual-objective optimization function, and perform non-dominated sorting based on the calculation results; Extracting parameter encoding patterns of the top 10% individuals based on the non-dominated sorting, and constructing a graph structure containing historical optimization paths; The graph attention network is used to learn the associated features of graph structure parameters, and the Transformer encoder is used to generate feature embedding vectors. The feature embedding vectors are input into the multi-layer perceptron to generate dynamic weight distribution. Design weighting function Optimizing the real-time allocation of the target dynamic weight distribution and outputting the multi-objective dynamic optimization results; ； Where, For the The optimization goal is at time The weight of For the The feature embedding vector corresponding to the target, is the real-time environment feature vector, is the vector concatenation operation, is the learnable parameter vector.

7. A real-time simulation method for detecting a photoelectric tracking device according to claim 6, characterized in that: The digital twin platform is used to build a virtual-real data interaction channel, input the multi-objective dynamic optimization results into the photon-level simulation engine, and complete the closed-loop iterative optimization of the control algorithm through real-time verification through the hardware-in-the-loop interface, including: Import the optical CAD model into the digital twin platform, configure the spectral sampling parameters, and initialize the photon tracking parameters; Based on the initialized photon tracing parameters, photon tracing data is generated through the Jones matrix and Monte Carlo algorithm, and the scattered field distribution is output to the hardware-in-the-loop interface in combination with the bidirectional reflectance distribution function model; Configure an FPGA module at the hardware-in-the-loop interface to perform nanosecond-level real-time simulation, and use a timestamp alignment algorithm to compensate for data transmission delays. Perform hardware frequency response calibration using chirp signal injection, generate compensation filters, and load them into the FPGA. A bidirectional channel for virtual and real data is constructed, and the unscented Kalman filter algorithm is used for multi-target state estimation and deviation prediction. The deviation compensation function is established by combining differential homeomorphism mapping to generate control parameter corrections. Based on the control parameter correction amount, the covariance matrix adaptive evolution strategy is used to evolve the control parameters online. The iteration cycle is set to 20ms to complete the closed-loop optimization process.

Images