A multi-parameter automatic detection system and method for a laser communication device

Abstract

The present invention relates to the field of communication technologies, and discloses a multi-parameter automatic detection system and method for a laser communication device. The method includes the following steps: data acquisition and forward scattering compensation, feature extraction and modeling, virtual debugging and enhanced simulation, parameter optimization and noise filtering, and result verification and feedback adjustment. Through the deep integration of multi-modal perception, intelligent modeling, and closed-loop optimization, the present invention has achieved a leapfrog upgrade in the detection of laser communication parameters. With a core breakthrough in improving the signal-to-noise ratio, the system bit error rate has been reduced to a small magnitude, and at the same time, the transmission distance has been extended to 182 km. Through the collaborative innovation of the federated learning framework and digital twin technology, the training efficiency has been improved, the passing rate of extreme working condition verification has been increased, and at the same time, a super high energy efficiency ratio and a low annual failure rate have been achieved.

Description

Technical Field

[0001] The present invention relates to the field of communications, and more specifically, to a multi-parameter automatic detection system and method for a laser communication device. Background Art

[0002] With the rapid development of new communication technologies such as space laser communication and quantum secure communication, the traditional parameter adjustment method based on empirical models has been difficult to meet the real-time and accuracy requirements in complex dynamic environments;

[0003] In space laser communication in complex dynamic environments, it faces multiple interferences such as the dynamic channels of low-earth orbit satellites, urban haze, and underwater turbulence. The traditional parameter adjustment relies on lagging empirical formulas, and there are bottlenecks such as difficulties in high-dimensional coupling analysis, a 56% decline in cross-scenario migration performance, and insufficient accuracy of the verification model. Summary of the Invention

[0004] The present invention provides a multi-parameter automatic detection system and method for a laser communication device to solve the technical problems in the related art.

[0005] The present invention provides a multi-parameter automatic detection method for a laser communication device, including the following steps:

[0006] S100, data acquisition and forward scattering compensation: Configure a multi-type sensor network to synchronously collect raw data and environmental parameters, correct atmospheric interference through the forward scattering compensation algorithm, complete data cleaning and standardization preprocessing, and generate a high-quality input vector;

[0007] S200, feature extraction and modeling: Based on a deep reinforcement learning model, screen key features from the compensated data, perform non-linear transformation and manifold dimensionality reduction, construct a low-dimensional and high-information feature space, and achieve iterative optimization of the prediction model through dynamic tuning;

[0008] S300, virtual commissioning and enhanced simulation: Load the prediction model in a digital twin environment, reconstruct the optical path propagation in combination with the neural radiance field technology, simulate the device behavior under extreme working conditions, and optimize the accuracy of the virtual model through multiple rounds of parameter adjustment;

[0009] S400, parameter optimization and noise filtering: Adopt a federated learning framework to achieve multi-node collaborative training, design an objective function with a noise filtering term, suppress interference through adaptive Kalman filtering, and complete distributed parameter optimization and global aggregation;

[0010] S500, result verification and feedback adjustment: Deploy the optimized parameters in the real environment, collect full-condition test data to quantify the performance indicators, construct a feedback control closed-loop to drive parameter iteration, and finally verify the reliability of the system through a golden sample set.

[0011] Further, S100 specifically includes the following steps:

[0012] Configure a set of sensors, set the start time of data acquisition, obtain the raw data from each sensor, and integrate all the data into a vector.

[0013] Measure the current environmental parameters, perform forward scattering compensation on the raw data, and generate a compensated data vector.

[0014] Preprocess the compensated data vector to obtain a high-quality data vector.

[0015] Further, the set of sensors includes the following six types of sensors: a laser power sensor, a wavelength monitor, a beam quality analyzer, an environmental temperature and humidity meter, an aerosol monitor, and a high-speed photodetector.

[0016] Further, S200 specifically includes the following steps:

[0017] Load a pre-trained DBN-DRL model, set the initial parameters, select relevant features from the compensated data vector, and generate a feature set.

[0018] Transform the selected features to generate a transformed feature set.

[0019] Use dimensionality reduction techniques to reduce the feature dimension to a more manageable scale and generate a dimensionality-reduced feature set.

[0020] Use the transformed feature set as the input data of the model, evaluate the quality of the model prediction results, and adjust the model parameters according to the evaluation results.

[0021] Further, the pre-trained DBN-DRL structure includes:

[0022] Visible layer: 43 neurons, consistent with the output dimension in S100;

[0023] Hidden layer architecture: a 4-layer DBN network structure;

[0024] Weight initialization: Xavier normal distribution ;

[0025] where represents the connection weight from the i-th input neuron to the j-th output neuron, represents the number of neurons in the input layer, represents the number of neurons in the output layer, represents the total number of neurons;

[0026] Policy network: PPO algorithm, discount factor ;

[0027] Value function: Estimation .

[0028] Further, S300 specifically includes the following steps:

[0029] Configure and start a virtual debugging environment, set the simulation start time, import the prediction results obtained in the feature extraction and modeling phase into the virtual environment to generate input data;

[0030] Set the initial conditions and device status in the virtual environment, use the NeRF technology to perform 3D reconstruction of the beam propagation path, and combine with the forward scattering compensation simulator to enhance the simulation effect, simulate the dynamic changes in the virtual environment, and update the environmental parameters and device status;

[0031] Evaluate the simulation results, calculate the error, and adjust the model parameters and initial conditions according to the evaluation results, apply the optimized parameters to the virtual environment to generate the final simulation results.

[0032] Further, perform 3D reconstruction of the beam propagation path:

[0033]

[0034] where represents the radiation field, and its number of sampling points , represents the density network, represents the color network, represents the cumulative transmittance, represents the starting boundary distance of the ray casting, represents the ending boundary distance of the ray casting.

[0035] Further, S400 specifically includes the following steps:

[0036] Configure and start a distributed optimization framework based on federated learning, set the initial model parameters, define the loss function, regularization term, and objective function, and introduce a noise filtering term;

[0037] Configure an adaptive noise filter to ensure that the noise filtering coefficient is reasonably set, perform local parameter updates on each node, and send the results to the central server for aggregation;

[0038] Adjust the initial parameters according to the global parameter update results, use the feedback mechanism to evaluate the optimization effect, apply the optimized global parameters to the actual system, verify the effect of the optimized parameters in the actual environment, and calculate the performance metrics.

[0039] Further, S500 specifically includes the following steps:

[0040] Collect test data from the actual environment, apply the optimized parameters to the actual system, run the laser communication system and record the output data, and calculate the performance metrics of the system on the test set;

[0041] Evaluate the system performance using predefined criteria, compare the actual performance with the target performance, generate feedback information based on the evaluation results, and use the feedback information to adjust the model parameters;

[0042] If the performance does not meet the expectations, re-execute the parameter optimization and noise filtering steps, verify again, and conduct a final verification on the final optimization results to confirm that the system performance meets the requirements.

[0043] The present invention also proposes a multi-parameter automatic detection system for a laser communication device, including:

[0044] Multi-modal data acquisition module: Corresponding to step S100, responsible for data acquisition and forward scattering compensation;

[0045] Intelligent feature engineering module: Corresponding to step S200, dealing with feature extraction and dimensionality reduction;

[0046] Digital twin simulation module: Corresponding to step S300, conducting virtual debugging and optical path simulation;

[0047] Federated optimization control module: Corresponding to step S400, realizing distributed parameter optimization;

[0048] Closed-loop verification feedback module: Corresponding to step S500, verifying the optimization results and providing feedback for adjustment;

[0049] Intelligent central management module: Coordinating the processes of steps S100 - S600 to ensure the smooth execution from data acquisition to verification.

[0050] The beneficial effects of the present invention are as follows:

[0051] Through the deep integration of multi-modal perception, intelligent modeling, and closed-loop optimization, the present invention has achieved a leapfrog upgrade in the detection of laser communication parameters. With the core breakthrough of improving the signal-to-noise ratio, the system bit error rate has been reduced to a small magnitude, and at the same time, the transmission distance has been extended to 182 km. Through the collaborative innovation of the federated learning framework and digital twin technology, the training efficiency has been improved, the passing rate of extreme working condition verification has been increased, and at the same time, a super high energy efficiency ratio and a low annual failure rate have been achieved, effectively solving the problems of difficult high-dimensional coupling analysis, decreased cross-scenario migration performance, and insufficient verification model accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flowchart of a multi-parameter automatic detection method for a laser communication device proposed by the present invention;

[0053] Figure 2It is a structural block diagram of a multi-parameter automatic detection system for a laser communication device proposed by the present invention.

[0054] In the figure: 101, multi-modal data acquisition module; 102, intelligent feature engineering module; 103, digital twin simulation module; 104, federated optimization control module; 105, closed-loop verification feedback module; 106, intelligent central management module. Specific implementation manners

[0055] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described relative to some examples can also be combined in other examples.

[0056] As Figure 1 shown, a multi-parameter automatic detection method for a laser communication device includes the following steps:

[0057] S100, data acquisition and forward scattering compensation: Configure a multi-type sensor network to synchronously acquire raw data and environmental parameters, correct atmospheric interference through the forward scattering compensation algorithm, complete data cleaning and standardization preprocessing, and generate a high-quality input vector;

[0058] In an embodiment of the present invention, the specific steps are as follows:

[0059] Configure a sensor set, set the start time of data acquisition, obtain raw data from each sensor, and integrate all the data into a vector;

[0060] Measure the current environmental parameters, perform forward scattering compensation on the raw data, and generate a compensated data vector;

[0061] Preprocess the compensated data vector to obtain a high-quality data vector;

[0062] In an embodiment of the present invention, the specific steps are as follows:

[0063] S110, configure a sensor set: The sensor set includes the following 6 types of sensors: laser power sensor, wavelength monitor, beam quality analyzer, environmental temperature and humidity meter, aerosol monitor, and high-speed photodetector;

[0064] It should be supplemented and explained that the measurement range of the laser power sensor is 10 - 500 mW, ;

[0065] It should be noted that the range accuracy of the wavelength monitor is ±0.02nm, ;

[0066] It should be noted that the beam quality analyzer analyzes through Zernike coefficients, ;

[0067] It should be noted that the PM2.5 detection range of the aerosol monitor is 0 - 500 μg / m³, ;

[0068] It should be noted that the range of the environmental temperature and humidity meter is a resolution of 0.1℃ / 0.1%RH, ;

[0069] It should be noted that the sampling rate of the high-speed photodetector is 1GHz, ;

[0070] In the above sensor set, the acquisition time adopts the IEEE 1588 precise clock protocol to ensure that the maximum time difference is less than or equal to , that is: ;

[0071] And set the reference sampling period: ;

[0072] S120, obtain the raw data from each sensor in the sensor set and vectorize the raw data;

[0073] Construct a time-domain data matrix:

[0074] ;

[0075] where represents the original time-domain data matrix, represents the instantaneous laser power, represents the working wavelength, represents the Zernike coefficients of the beam wavefront, represents the temperature and humidity of the environment, , represents the aerosol concentration classification measurement value, represents the photocurrent intensity;

[0076] Time window accumulation:

[0077] ;

[0078] where represents the step period of the time window, represents the accumulated data matrix of the kth time window;

[0079] Perform forward scattering compensation on the original data:

[0080] Apply an improved Beer-Lambert correction model:

[0081] ;

[0082] where represents the component-related physical quantity calculated by the Beer-Lambert law, represents the reference scattering coefficient, , represents the transmission distance (km), represents a constant or calibration parameter, represents the instantaneous power in the time-domain data;

[0083] where represents the compensation error compensation factor, , represents the time-related frequency or scale factor;

[0084] S130, perform data preprocessing on the vectorized original data to generate high-quality data vectors;

[0085] In one embodiment of the present invention, the data preprocessing includes the following:

[0086] Normalization processing:

[0087] ;

[0088] where represents the moving window mean, represents the processed component signal, represents the moving window standard deviation, represents the normalized data;

[0089] Wavelet denoising:

[0090] Adopt the DB8 wavelet basis function and set the threshold:

[0091] ;

[0092] where represents the discrete wavelet transform for extracting the characteristic components in the signal, represents the threshold;

[0093] Generate high-quality data vectors:

[0094] ;

[0095] where is the pre-trained filter matrix, , represents the denoised signal, represents the real-valued weight function for filtering or weighting the signal, represents the weight parameter for generating high-quality data vectors;

[0096] S200, Feature Extraction and Modeling: Based on the Deep Belief Network - Deep Reinforcement Learning Model (DBN - DRL model), key features are screened from the compensation data, non-linear transformation and manifold dimensionality reduction are performed to construct a low-dimensional and high-information feature space, and the iterative optimization of the prediction model is achieved through dynamic tuning;

[0097] In one embodiment of the present invention, it specifically includes the following steps:

[0098] Load the pre-trained DBN - DRL model, set the initial parameters, select relevant features from the compensated data vector, and generate a feature set;

[0099] Transform the selected features to generate a transformed feature set;

[0100] Use dimensionality reduction technology to reduce the feature dimension to a more manageable scale and generate a dimensionality-reduced feature set;

[0101] Use the transformed feature set as the input data of the model, evaluate the quality of the model prediction results, and adjust the model parameters according to the evaluation results;

[0102] S210, Initialization of the Deep Belief Network:

[0103] Load the pre-trained DBN - DRL structure:

[0104] Visible layer: 43 neurons (matching the output dimension of S100);

[0105] Hidden layer architecture: 4 - layer DBN network structure (43→256→128→64→32);

[0106] Weight initialization: Xavier normal distribution ;

[0107] where represents the connection weight from the i-th input neuron to the j-th output neuron, represents the number of neurons in the input layer (here it is 43), represents the number of neurons in the output layer (corresponding to the hidden layer architecture);

[0108] Reinforcement learning module configuration:

[0109] Policy network: PPO algorithm, discount factor ;

[0110] Value function: Estimation ;

[0111] S220, Multimodal feature selection:

[0112] Perform mutual information screening:

[0113] ;

[0114] Where represents mutual information, represents the i-th feature variable, Y represents the target output variable, and respectively represent and marginal probability density functions of represents and joint probability density function of Retain 18 core features that satisfy

[0115] L1-regularized sparse filtering:

[0116] ;

[0117] Where represents the j-th parameter weight of the model, set , Screen features corresponding to non-zero weights;

[0118] S230, Nonlinear feature engineering:

[0119] Polynomial expansion:

[0120] ;

[0121] Where and respectively represent the i-th and j-th features in the original feature vector, is the feature expansion term;

[0122] Expanded feature dimension: 18 → 72;

[0123] Wavelet time-frequency transform:

[0124]

[0125] Where represents the scale parameter, which controls the frequency resolution of wavelet analysis, represents the displacement parameter, which determines the position of the analysis time window, represents the complex conjugate of the Morlet wavelet function, and the scale parameter of the Morlet wavelet is adopted ;

[0126] S240, Manifold learning dimensionality reduction:

[0127] Improve t-SNE projection:

[0128] ;

[0129] where represents the adaptive bandwidth parameter of the i-th data point, set the perplexity , learning rate ;

[0130] Supervised PCA compression: ;

[0131] where represents the projection direction vector, represents the input feature matrix, represents the target variable matrix, where 95% of the variance is retained and finally reduced to 12 dimensions;

[0132] S250, Dynamic model tuning:

[0133] Construct a hybrid loss function:

[0134] ;

[0135] where the coefficients of the loss function are set , , , represents the set of all parameter matrices of the model;

[0136] Adaptive learning rate mechanism:

[0137] ;

[0138] where represents the adaptive learning rate, represents the initial learning rate, , warm-up period ;

[0139] S300, Virtual debugging and enhanced simulation: Load the prediction model in the digital twin environment, reconstruct the optical path propagation in combination with the Neural Radiance Field (NeRF) technology, simulate the equipment behavior under extreme conditions, and optimize the virtual model accuracy through multiple rounds of parameter adjustment;

[0140] In one embodiment of the present invention, it specifically includes the following steps:

[0141] Configure and start the virtual debugging environment, set the simulation start time, import the prediction results obtained in the feature extraction and modeling stage into the virtual environment, and generate input data;

[0142] Set the initial conditions and device status in the virtual environment, use NeRF technology to perform 3D reconstruction of the light beam propagation path, and combine with the forward scattering compensation simulator to enhance the simulation effect. Simulate dynamic changes in the virtual environment, such as atmospheric turbulence and device aging, and update the environmental parameters and device status;

[0143] Evaluate the simulation results, calculate the error, and adjust the model parameters and initial conditions according to the evaluation results. Apply the optimized parameters to the virtual environment to generate the final simulation results;

[0144] In one embodiment of the present invention,

[0145] S310, full-body virtual environment construction:

[0146] Build a simulation platform using the ROS Gazebo framework:

[0147] Time synchronization parameters:

[0148] Virtual clock accuracy: , Indicates the virtual time difference;

[0149] Physical engine step size: ;

[0150] Spatial coordinate system:

[0151] World coordinate system: ;

[0152] Device coordinate system: ;

[0153] S320, multi-source data fusion injection:

[0154] Feature data conversion:

[0155] ;

[0156] Wherein Indicates the weight of the converted feature data, Indicates the parameter of the converted feature data, Indicates the virtual feature data, Indicates the size of The input feature matrix of dimension;

[0157] Conversion matrix dimension: ;

[0158] Spatio-temporal alignment processing: ;

[0159] where represents virtual time, represents real time, represents the offset time difference;

[0160] S330, Beam Propagation NeRF Modeling:

[0161] Perform 3D reconstruction on the beam propagation path, and construct the following neural radiance field:

[0162] ;

[0163] where represents the radiance field, and its number of sampling points , represents the density network, , represents the color network, , represents the cumulative transmittance, represents the starting boundary distance of ray casting, represents the ending boundary distance of ray casting;

[0164] S340, Dynamic Interference Simulator:

[0165] Atmospheric Turbulence Modeling:

[0166] ;

[0167] where represents the Reynolds number, reflecting the ratio of fluid inertial force to viscous force, represents the energy spectral density or power of the nth mode, represents the wave number, n represents the mode number, x represents the spatial coordinate, t represents the time variable, and ωₙ represents the angular frequency of the nth mode;

[0168] Phase Screen Parameters:

[0169] Refractive Index Structure Constant: ;

[0170] Wind Speed Profile: ;

[0171] Equipment Aging Model:

[0172] ;

[0173] where the aging coefficient , and the noise term ;

[0174] S350, Forward Scattering Enhancement Simulation:

[0175] Modified Scattering Compensation:

[0176] ;

[0177] where represents the environmental coupling coefficient ;

[0178] Power Attenuation Reconstruction:

[0179] ;

[0180] where the perturbation frequency ;

[0181] S360, Online Parameter Optimization:

[0182] Error Evaluation Function:

[0183] ;

[0184] where the weight setting , , ;

[0185] Gradient Adaptive Adjustment:

[0186] ;

[0187] where the learning rate , the smoothing factor , represents the electric field strength vector;

[0188] S400, Parameter Optimization and Noise Filtering: Implement multi-node collaborative training using a federated learning framework, design an objective function with a noise filtering term, suppress interference through adaptive Kalman filtering, and complete distributed parameter optimization and global aggregation;

[0189] In one embodiment of the present invention, it specifically includes the following steps:

[0190] Configure and start a distributed optimization framework based on federated learning, set initial model parameters, define a loss function, a regularization term, and an objective function, and introduce a noise filtering term;

[0191] Configure an adaptive noise filter to ensure reasonable setting of the noise filtering coefficient, perform local parameter updates on each node, and send the results to the central server for aggregation;

[0192] Adjust the initial parameters according to the global parameter update result, evaluate the optimization effect using the feedback mechanism, apply the optimized global parameters to the actual system, verify the effect of the optimized parameters in the actual environment, and calculate the performance metrics;

[0193] S410, Federated learning architecture deployment:

[0194] Distributed node configuration:

[0195] Number of participating nodes ;

[0196] Synchronization period ;

[0197] Initial parameter inheritance: ;

[0198] Secure communication protocol:

[0199] Encryption method: AES-256;

[0200] Bandwidth constraint: ;

[0201] S420, Hybrid objective function construction:

[0202] Regularized loss combination:

[0203] ;

[0204] Where represents the regularization coefficient, represents the noise suppression coefficient, coefficient setting , , represents the noise residual amount of the i-th sample;

[0205] Noise filtering term design:

[0206] ;

[0207] Where the adaptive variance ;

[0208] S430, Adaptive Kalman filtering:

[0209] State observation equation:

[0210] ;

[0211] Where represents the state transition matrix, , represents the observation matrix, ;

[0212] Gain matrix update:

[0213] ;

[0214] where represents the Kalman gain matrix at time t, , represents the prior estimation error covariance, , represents the observation noise, ;

[0215] S440, Distributed parameter training:

[0216] Local SGD optimization:

[0217] ;

[0218] where represents the local learning rate, , momentum term ;

[0219] Gradient clipping:

[0220] ;

[0221] Threshold setting ;

[0222] S450, Federated model aggregation:

[0223] Weighted average strategy:

[0224] ;

[0225] Data volume weight , is the federated global parameter;

[0226] Momentum accelerated update:

[0227] ;

[0228] S460, Online feedback regulation:

[0229] Performance improvement evaluation:

[0230] ;

[0231] where represents the bit error rate benchmark value before optimization, represents the measured bit error rate after optimization, represents the performance evaluation value, and its trigger threshold ;

[0232] Dynamic regularization adjustment:

[0233] ;

[0234] where represents the regularized parameter after adjustment, represents the regularized parameter before adjustment;

[0235] S500, Result verification and feedback adjustment: Deploy the optimized parameters in the real environment, collect full operating condition test data to quantify performance metrics (bit error rate / power consumption / latency), construct a feedback control closed loop to drive parameter iteration, and finally verify the system reliability through the golden sample set;

[0236] In an embodiment of the present invention, it specifically includes the following steps:

[0237] Collect test data from the actual environment, apply the optimized parameters to the actual system, run the laser communication system and record the output data, and calculate the performance metrics of the system on the test set, such as bit error rate and power consumption;

[0238] Evaluate the system performance using predefined criteria, compare the actual performance with the target performance, generate feedback information based on the evaluation results, and use the feedback information to adjust the model parameters;

[0239] If the performance does not meet the expectation, re - execute the parameter optimization and noise filtering steps, verify again, and conduct a final verification of the final optimization result to confirm that the system performance meets the requirements.

[0240] S510, Multi - modal test data acquisition:

[0241] Optoelectronic signal sampling parameters:

[0242] Detector response time: ;

[0243] Sampling frequency: ;

[0244] Storage bandwidth: ;

[0245] Environmental parameter synchronous recording:

[0246] ;

[0247] where represents the environmental temperature monitoring range;

[0248] S520, Parameter loading and system re - configuration:

[0249] Global parameter injection:

[0250] ;

[0251] where represents the calibration matrix, with the condition number κ(M) ≤ 10², represents the device's inherent calibration parameters, and represents the system calibration parameters;

[0252] Secure loading protocol:

[0253] Adopts double redundancy check, CRC - 32 check threshold: ≤ 3bit error

[0254] S530, full - condition performance test:

[0255] Test case set:

[0256] ;

[0257] where represents the feature vector of a single test case, and the single - test duration , represents the test case set, and represents the number of tests;

[0258] Output data record:

[0259] Timestamp alignment accuracy ;

[0260] Data encapsulation format: IEEE 1588v2 standard

[0261] S540, Key - index quantitative analysis:

[0262] Bit - error rate calculation:

[0263] ;

[0264] where represents the total number of bit errors, represents the total number of transmitted bits,

[0265] Confidence requirement: 99% confidence interval width ≤ 10⁻ 7

[0266] Energy - efficiency evaluation model:

[0267] ;

[0268] where the benchmark requirement , represents the average power at the receiving end, and represents the power consumption of the signal - processing unit Represents the average power of the transmitting end;

[0269] S550, Adaptive feedback regulation:

[0270] Calculation of performance difference degree:

[0271] ;

[0272] Wherein Represents the measured value of the i-th index, Represents the design target value of the i-th index, Represents the weight vector of performance indicators, ;

[0273] Parameter correction model:

[0274] ;

[0275] Wherein Difference degree threshold, feedback learning rate ;

[0276] S560, Convergence test:

[0277] Golden sample test set:

[0278] ;

[0279] Wherein Represents the golden test set, that is, 500 sets of extreme working condition combinations, and the samples cover all severe working condition combinations;

[0280] Stability verification:

[0281] ;

[0282] Wherein Represents the constraint of the coefficient of variation of the bit error rate, Represents the power stability constraint, continuous observation duration .

[0283] As Figure 2 shown, the present invention also proposes a multi-parameter automatic detection system for a laser communication device, which executes the steps in the foregoing multi-parameter automatic detection method for a laser communication device, and includes the following modules:

[0284] Multi-modal data acquisition module 101: Corresponding to step S100, mainly responsible for data acquisition and forward scattering compensation;

[0285] The module includes a sensor array for real-time acquisition of optoelectronic signals and environmental parameters (temperature, humidity, PM2.5). The forward scattering compensation unit uses algorithms to correct atmospheric interference, and the data cleaning engine applies the 5σ criterion to filter abnormal data to ensure input quality;

[0286] Intelligent Feature Engineering Module 102: Corresponding to step S200, it processes feature extraction and dimensionality reduction;

[0287] The module uses a Deep Belief Network (DBN) for feature initialization and extracts high-level features through a multi-layer neural network (such as 256-128-64-32).

[0288] Manifold learning components such as the improved t-SNE algorithm reduce high-dimensional features to low dimensions while retaining key information;

[0289] The dynamic tuning unit optimizes model parameters through a hybrid loss function (MAE, KL divergence, regularization) to ensure the effectiveness of features;

[0290] Digital Twin Simulation Module 103: Corresponding to step S300, it performs virtual debugging and optical path simulation;

[0291] The module reconstructs the laser propagation path through the NeRF engine and uses high-resolution pixels (1024³) to simulate light scattering and attenuation in a real environment;

[0292] The extreme working condition simulator generates different environmental conditions (such as turbulence, vibration) to test the performance of the device under adverse conditions;

[0293] The virtual debugging interface allows injecting parameters to adjust the model to match the actual device behavior;

[0294] Federated Optimization Control Module 104: Corresponding to step S400, it realizes distributed parameter optimization;

[0295] The module coordinates multiple nodes for local training through the federated learning framework and uses encrypted communication (AES-256) to ensure data security;

[0296] The hybrid objective function combines the MSE loss and the regularization term, and at the same time, the Kalman filter reduces the influence of noise;

[0297] The parameter aggregation center weights and averages the node models to update the global parameters;

[0298] Closed-loop Verification Feedback Module 105: Corresponding to step S500, it verifies the optimization results and provides feedback for adjustment;

[0299] The module runs preset extreme scenarios through the golden test set executor and collects performance data;

[0300] The performance analyzer calculates key metrics (bit error rate, energy efficiency) to evaluate whether the system meets the standards;

[0301] The feedback controller dynamically adjusts parameters according to the degree of difference, for example, adjusts the learning rate through the tanh function;

[0302] Intelligent central management module 106: Coordinates the entire system process to ensure the smooth execution from data collection to verification;

[0303] The module manages the sequence and resource allocation of each step through a task scheduler, and the data lake stores a large amount of real-time data;

[0304] The security protocol protects data transmission to prevent man-in-the-middle attacks or data leakage.

[0305] At least one embodiment disclosed by the present invention provides a storage medium storing non-transitory computer-readable instructions for executing one or more steps in the process optimization method of the multi-part adaptable self-adjusting stamping die described above.

[0306] The computer program can be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems. Any reference signs in the claims shall not be construed as limiting the scope.

[0307] The above describes the embodiments of this embodiment, but this embodiment is not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of this embodiment.

Claims

1. A method for automatically detecting multiple parameters of a laser communication device, characterized in that It includes the following steps: S100, Data acquisition and forward scattering compensation: Configure a multi-type sensor network to synchronously acquire raw data and environmental parameters, correct atmospheric interference through the forward scattering compensation algorithm, complete data cleaning and standardization preprocessing, and generate high-quality input vectors; S200, Feature extraction and modeling: Based on the deep reinforcement learning model, screen key features from the compensated data, perform non-linear transformation and manifold dimensionality reduction, construct a low-dimensional and high-information feature space, and achieve iterative optimization of the prediction model through dynamic tuning; Specifically, S200 includes the following steps: Load the pre-trained DBN-DRL model and set the initial parameters, select relevant features from the compensated data vectors to generate a feature set; Transform the selected features to generate a transformed feature set; Use dimensionality reduction techniques to reduce the feature dimension to a more manageable scale to generate a dimension-reduced feature set; Use the transformed feature set as the input data of the model, evaluate the quality of the model prediction results, and adjust the model parameters according to the evaluation results; S300, Virtual commissioning and enhanced simulation: Load the prediction model in the digital twin environment, reconstruct the optical path propagation in combination with the neural radiance field technology, simulate the device behavior under extreme working conditions, and optimize the virtual model accuracy through multiple rounds of parameter adjustment; Specifically, S300 includes the following steps: Configure and start the virtual commissioning environment, set the simulation start time, import the prediction results obtained in the feature extraction and modeling stage into the virtual environment to generate input data; Set the initial conditions and device status in the virtual environment, use NeRF technology to perform 3D reconstruction of the beam propagation path, and enhance the simulation effect in combination with the forward scattering compensation simulator, simulate the dynamic changes in the virtual environment, and update the environmental parameters and device status; Evaluate the simulation results, calculate the error, adjust the model parameters and initial conditions according to the evaluation results, apply the optimized parameters to the virtual environment to generate the final simulation results; S400, Parameter optimization and noise filtering: Adopt the federated learning framework to achieve multi-node collaborative training, design an objective function with a noise filtering term, suppress interference through adaptive Kalman filtering, and complete distributed parameter optimization and global aggregation; S500, Result verification and feedback adjustment: Deploy the optimized parameters in the real environment, collect full-condition test data to quantify the performance indicators, construct a feedback control closed-loop to drive parameter iteration, and finally verify the system reliability through the golden sample set.

2. The multi-parameter automatic detection method for a laser communication device according to claim 1, characterized in that, Specifically, S100 includes the following steps: Configure the sensor set, set the data acquisition start time, obtain raw data from each sensor, and integrate all data into a vector; Measure the current environmental parameters, perform forward scattering compensation on the raw data to generate a compensated data vector; Preprocess the compensated data vector to obtain a high-quality data vector.

3. The multi-parameter automatic detection method for a laser communication device according to claim 2, characterized in that, The sensor set includes the following six types of sensors: laser power sensor, wavelength monitor, beam quality analyzer, environmental temperature and humidity meter, aerosol monitor, and high-speed photodetector.

4. A multi-parameter automatic detection method for a laser communication device according to claim 1, characterized in that, Among them, the pre-trained DBN-DRL structure includes: Visible layer: 43 neurons, consistent with the output dimension in S100; Hidden layer architecture: 4-layer DBN network structure; Weight initialization: Xavier normal distribution ; wherein represents the connection weight from the i-th input neuron to the j-th output neuron, represents the number of neurons in the input layer, represents the number of neurons in the output layer, represents the total number of neurons; Policy Network: PPO algorithm, discount factor ; Value function: Estimation .

5. A multi-parameter automatic detection method for a laser communication device according to claim 4, characterized in that, Specifically, it includes the following steps in S400: Configure and start a distributed optimization framework based on federated learning, set initial model parameters, define a loss function, a regularization term, and an objective function, and introduce a noise filtering term; Configure an adaptive noise filter to ensure that the noise filtering coefficient is reasonably set, perform local parameter updates on each node, and send the results to the central server for aggregation; Adjust the initial parameters according to the global parameter update results, use a feedback mechanism to evaluate the optimization effect, apply the optimized global parameters to the actual system, verify the effect of the optimized parameters in the actual environment, and calculate performance metrics.

6. The multi-parameter automatic detection method for a laser communication device according to claim 5, characterized in that Specifically, it includes the following steps in S500: Collect test data from the actual environment, apply the optimized parameters to the actual system, run the laser communication system and record the output data, and calculate the performance metrics of the system on the test set; Use predefined criteria to evaluate the system performance, compare the actual performance with the target performance, generate feedback information according to the evaluation results, and use the feedback information to adjust the model parameters; If the performance does not meet the expectations, re-execute the parameter optimization and noise filtering steps, verify again, and perform a final verification on the final optimization results to confirm that the system performance meets the requirements.

7. A multi-parameter automatic detection system for a laser communication device, characterized in that, For executing the steps in a multi-parameter automatic detection method of a laser communication device as described in any one of claims 1-6, including: Multi-modal data acquisition module: Corresponding to step S100, responsible for data acquisition and forward scattering compensation; Intelligent feature engineering module: Corresponding to step S200, dealing with feature extraction and dimensionality reduction; Digital twin simulation module: Corresponding to step S300, performing virtual debugging and optical path simulation; Federated optimization control module: Corresponding to step S400, implementing distributed parameter optimization; Closed-loop verification feedback module: Corresponding to step S500, verifying the optimization results and providing feedback for adjustment; Intelligent central management module: Coordinating the processes of steps S100-S600 to ensure the smooth execution from data acquisition to verification.

Images