Intelligent component warehousing management system and method based on digital twinning

Abstract

The application relates to the technical field of warehouse management, and discloses an intelligent component warehouse management system and method based on digital twinning, which comprises a stress management unit, a conduction management unit, a change management unit, a compensation management unit and a twinning management unit. The stress management tensor generated by the stress management unit can accurately capture the time-domain-frequency-domain cross characteristics of temperature and humidity data and vibration data, avoids missing the nonlinear interaction of multiple physical fields, the lattice distortion management graph generated by the conduction management unit can identify the distortion risk of a quantum dot active region in advance, the phonon energy management spectrum of each quantum dot generated by the change management unit can track the dynamic change of lattice distortion, the spectrum shape compensation signal generated by the compensation management unit in the reverse direction of each quantum dot can offset the central wavelength drift, and the twinning management instruction generated by the twinning management unit can avoid management lag, reduce the damage probability of a quantum dot laser, and improve the reliability of warehouse management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of warehouse management, in particular to an intelligent component warehouse management system and method based on digital twinning. BACKGROUND

[0002] At present, when the quantum dot laser and other intelligent components are managed in the warehouse, the management is based on neural networks, and independent shallow neural networks are usually built based on temperature, humidity and vibration, each neural network only outputs a single environmental parameter, and the parameters are visualized through digital twinning to realize the management of intelligent components.

[0003] However, the above management method still has the following defects when the air conditioner is frequently started and stopped in summer: in summer, the warehouse air conditioner will be frequently started and stopped due to high load, resulting in changes in temperature and humidity, and the AGV in the warehouse is prone to vibration when moving or restocking. Under such an environment, thermal stress causes the active region of the quantum dot to distort, moisture penetration causes condensation on the packaged optical window, and environmental vibration exacerbates the existing interlayer dislocation slip in the active region. The combined effects of the above factors cause the central wavelength of the laser to drift. However, the neural networks of the existing management method are usually independent, and even simple interactions cannot capture the comprehensive effects of nonlinear interactions between different physical fields on intelligent components, resulting in lag in the management of quantum dot lasers, which makes the quantum dot lasers prone to damage. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides an intelligent component warehouse management system and method based on digital twinning, which solves the above problems.

[0005] The above technical purpose of the present application is realized by the following technical scheme:

[0006] An intelligent component warehouse management system based on digital twinning, comprising:

[0007] A stress management unit is configured to acquire temperature and humidity data and vibration data in a target management warehouse in real time, and acquire a central wavelength of a target management object in real time, take the central wavelength as a reference wave value, and use a convolutional neural network in a digital twinning platform to input the temperature and humidity data and the vibration data, extract time-frequency cross features through a convolution kernel, and perform nonlinear mapping based on a fully connected layer to output a stress management tensor acting on a multi-quantum dot active region of the target management object, wherein the target management object is a quantum dot laser.

[0008] A conduction management unit is configured to construct a plurality of quantum dot active regions as a node-edge graph structure through a graph neural network in a digital twin platform, take a stress management tensor as a graph node feature, learn a correlation relationship of a high stress gradient region through a graph convolution layer, output a probability distribution of a lattice strain at each position, and generate a lattice distortion management graph;

[0009] A change management unit is configured to take the lattice distortion management graph as an input, capture a dynamic change feature of the lattice distortion over time through a gating mechanism of a gated recurrent unit neural network, analyze a nonlinear change of a quantum confinement effect, and generate a phonon energy management spectrum of each quantum dot;

[0010] A compensation management unit is configured to map the phonon energy management spectrum of each quantum dot to a drift amount of a center wavelength of a target management object, and generate a spectrum-shaped compensation signal of each quantum dot in a reverse direction;

[0011] A twin management unit is configured to compensate the target management object through the spectrum-shaped compensation signal, obtain a center wavelength of the compensated target management object, take the center wavelength as a feedback wave value, compare the feedback wave value with a reference wave value, and generate a twin management instruction.

[0012] Further, in the digital twin platform, a convolutional neural network is used to take temperature and humidity data and vibration data as inputs, extract time-frequency cross features through a convolution kernel, and perform nonlinear mapping based on a fully connected layer to output a stress management tensor acting on a plurality of quantum dot active regions in the target management object, including:

[0013] The temperature and humidity data and the vibration data after preprocessing are calculated to obtain dynamic cooperative changes in the same time dimension to generate an environment time domain cooperative matrix;

[0014] The environment time domain cooperative matrix is input into a first layer convolution kernel of the convolutional neural network, the cooperative change law in the environment time domain cooperative matrix is analyzed, time domain features reflecting the influence of the environment on the lattice are extracted, and a time ring-lattice correlation management graph is generated;

[0015] The time ring-lattice correlation management graph is converted into a frequency domain, and a second layer convolution kernel of the convolutional neural network is used to filter regions with strong correlation with time domain correlation features in the frequency domain, analyze the effect of environmental changes on the lattice, and generate a frequency ring-lattice correlation management graph;

[0016] A third layer convolution kernel of the convolutional neural network is used to fuse the time ring-lattice correlation management graph and the frequency ring-lattice correlation management graph to generate a domain ring-lattice fusion feature tensor.

[0017] Further, in the digital twin platform, the temperature and humidity data and the vibration data are input through a convolutional neural network, cross-time domain-frequency domain features are extracted through a convolution kernel, nonlinear mapping is performed based on a fully connected layer, and a stress management tensor acting on a multi-quantum dot active region in the target management object is output, and the stress management tensor further includes:

[0018] The domain ring crystal fusion feature tensor is input into the fully connected layer of the third layer convolution kernel, the size data of the multi-quantum dot active region in the quantum dot laser is combined, the fusion features are converted into features related to the crystal lattice stress, and a stress-related feature vector is generated;

[0019] The three-dimensional spatial coordinates of each quantum dot are obtained, and the distances between each quantum dot are calculated based on the three-dimensional spatial coordinates of each quantum dot.

[0020] Based on the stress-related feature vector, the distances between each quantum dot are combined for dimension reconstruction, the features of the stress-related feature vector are assigned to the corresponding spatial positions, and a stress management tensor acting on the multi-quantum dot active region in the target management object is generated.

[0021] Further, in the digital twin platform, the multi-quantum dot active region is constructed as a node-edge graph structure through a graph neural network, the stress management tensor is used as a graph node feature, the correlation relationship of the high stress gradient area is learned through a graph convolution layer, the probability distribution of the lattice strain at each position is output, and a lattice distortion management graph is generated, including:

[0022] Based on the stress values of each quantum dot in the stress management tensor, the three-dimensional size of each quantum dot is combined, and a node stress bearing coefficient is calculated;

[0023] The distances between each quantum dot and the node stress bearing coefficient are analyzed to obtain a stress conduction correlation degree;

[0024] The components related to the time domain features in the domain ring crystal fusion feature tensor and the node stress bearing coefficient are feature fused to generate a node stress-time domain feature vector;

[0025] The node stress-time domain feature vector and the stress conduction correlation degree are input into the graph convolution layer, the mean value of the stress conduction correlation degree is used as a threshold, the features of adjacent nodes whose stress conduction correlation degree exceeds the threshold are aggregated, and a high correlation stress management value is generated.

[0026] Further, in the digital twin platform, the multi-quantum dot active region is constructed as a node-edge graph structure through a graph neural network, the stress management tensor is used as a graph node feature, the correlation relationship of the high stress gradient area is learned through a graph convolution layer, the probability distribution of the lattice strain at each position is output, and a lattice distortion management graph is generated, further including:

[0027] The thickness of the spacer layer of the multi-quantum dot active region is calculated to obtain a stress attenuation management factor, the high correlation stress management value is corrected based on the stress attenuation management factor, and a corrected node strain value is generated;

[0028] Based on the three-dimensional spatial coordinates of each quantum dot, the distribution density of each quantum dot is analyzed and combined with the node strain value to calculate the node strain probability value;

[0029] Based on the three-dimensional spatial coordinates of each quantum dot, the node strain probability value is mapped to a spatial thermal force distribution, and the strength of the stress conduction correlation is superimposed to generate a lattice distortion management map.

[0030] Further, the lattice distortion management map is taken as input, the dynamic change characteristics of the lattice distortion over time are captured through the gating mechanism of the gated recurrent unit neural network, the nonlinear change of the quantum confinement effect is analyzed, and the phonon energy management spectrum of each quantum dot is generated, including:

[0031] The node strain probability value is combined with the stress conduction correlation and the timing data of the digital twin platform to calculate the ratio of the strain probability change amplitude to the stress conduction correlation change amplitude in the same time interval, and a strain-conduction management timing coefficient is generated;

[0032] The strain-conduction management timing coefficient is input into the reset gate and update gate of the gated recurrent unit, and the domain ring crystal fusion feature tensor is combined to calculate the weighted proportion of the strain-conduction management timing coefficient and the frequency domain feature contribution degree, and a spatiotemporal frequency three-dimensional gating weight is generated;

[0033] Based on the spatiotemporal frequency three-dimensional gating weight, the strain-conduction management timing coefficient is hierarchically screened to generate a spatiotemporal frequency correlation timing feature vector;

[0034] Based on the spatiotemporal frequency correlation timing feature vector and the distribution density of each quantum dot, the change correlation of phonon energy in different timing segments is analyzed to generate an energy cooperative deviation management value;

[0035] The energy cooperative deviation management value is distributed according to the three-dimensional spatial coordinates of each quantum dot, and the spatiotemporal frequency three-dimensional gating weight is superimposed to generate a phonon energy management spectrum for each quantum dot.

[0036] Further, the phonon energy management spectrum of each quantum dot is mapped to the drift amount of the center wavelength of the target management object to generate a spectrum shape compensation signal in the opposite direction of each quantum dot, including:

[0037] The phonon energy management spectrum of each quantum dot is combined with the spatiotemporal frequency three-dimensional gating weight and the node strain probability value to calculate the correlation strength ratio of phonon energy timing fluctuation, spatiotemporal frequency characteristics, and strain probability, and a phonon-strain-frequency domain management coefficient is generated;

[0038] Analyze the phonon-strain-frequency domain management coefficient and the reference wave value to generate a preliminary central wavelength drift amount;

[0039] Calculate the reference wave value and the preliminary central wavelength drift amount to generate a double-constraint correction factor;

[0040] Based on the double-constraint correction factor, calculate the size-sensitive calibration coefficient in combination with the three-dimensional size of each quantum dot, and perform double calibration on the preliminary central wavelength drift amount through the double-constraint correction factor and the size-sensitive calibration coefficient to obtain an accurate drift compensation mapping value;

[0041] Analyze the inverse correlation between the accurate drift compensation mapping value and the phonon energy management spectrum to generate a spectrum shape compensation signal in the opposite direction of each quantum dot.

[0042] Further, compensate the target management object through the spectrum shape compensation signal to obtain the central wavelength of the compensated target management object, and compare the reference wave value with the feedback wave value to generate a twin management instruction, including:

[0043] Combine the spectrum shape compensation signal of each quantum dot with the three-dimensional spatial coordinates of each quantum dot to calculate the weight proportion of the compensation signal of the quantum dot at different spatial positions on the wavelength influence of the target management object, and generate a compensation priority coefficient;

[0044] Based on the compensation priority coefficient, apply the spectrum shape compensation signal to the multi-quantum dot active region of the target management object in the order of quantum dot spatial distribution, and after the compensation operation, real-time collect the central wavelength of the target management object as the feedback wave value;

[0045] Based on the reference wave value, the feedback wave value, and the phonon energy management spectrum of each quantum dot, calculate the correlation degree of the difference between the two wave values and the phonon energy fluctuation to generate a wavelength deviation correlation degree;

[0046] Analyze the influence of the fluctuation amplitude of the temperature and humidity data on the wavelength deviation correlation degree, calculate the stability index of the compensation effect under environmental interference, and generate a compensation dynamic coefficient.

[0047] Further, compensate the target management object through the spectrum shape compensation signal to obtain the central wavelength of the compensated target management object, and compare the reference wave value with the feedback wave value to generate a twin management instruction, also including:

[0048] Calculate the upper limit of the wavelength deviation corresponding to the maximum fluctuation of temperature and humidity and the mean value of the node strain probability value to obtain a dynamic deviation threshold;

[0049] Combine the stress conduction correlation degree with the spacer layer thickness to obtain a stability coefficient threshold;

[0050] The wavelength deviation correlation degree is compared with the dynamic deviation threshold, the compensation dynamic coefficient and the stability coefficient threshold respectively, and twin management instructions are generated.

[0051] Furthermore, a digital twin-based intelligent component warehouse management method, applying the aforementioned digital twin-based intelligent component warehouse management system, includes:

[0052] Step S1: Real-time acquisition of temperature, humidity, and vibration data within the target management warehouse, and real-time acquisition of the center wavelength of the target management object. Using this center wavelength as a reference wave eigenvalue, a convolutional neural network is used in the digital twin platform. With temperature, humidity, and vibration data as input, time-domain and frequency-domain cross features are extracted through convolutional kernels, and nonlinear mapping is performed based on fully connected layers. The output is a stress management tensor acting on the multi-quantum dot active region of the target management object; the target management object is a quantum dot laser.

[0053] Step S2: In the digital twin platform, the active region of multiple quantum dots is constructed as a node-edge graph structure through a graph neural network. The stress management tensor is used as the graph node feature. The correlation between high stress gradient regions is learned through graph convolutional layers. The probability distribution of lattice strain at each position is output to generate a lattice distortion management graph.

[0054] Step S3: Using the lattice distortion management map as input, the dynamic change characteristics of lattice distortion over time are captured through the gating mechanism of the gated recurrent unit neural network, the nonlinear change of quantum confinement effect is analyzed, and the phonon energy management spectrum of each quantum dot is generated.

[0055] Step S4: Map the phonon energy management spectrum of each quantum dot to the drift of the center wavelength of the target managed object, and generate the reverse spectral compensation signal for each quantum dot.

[0056] Step S5: Compensate the target management object with the spectral compensation signal to obtain the center wavelength of the compensated target management object. Use the center wavelength as the feedback wave eigenvalue and compare the reference wave eigenvalue and the feedback wave eigenvalue to generate twin management instructions.

[0057] In summary, the present invention has the following main beneficial effects:

[0058] The stress management unit acquires real-time temperature and humidity data, vibration data, and baseline wave characteristics of the target managed object. It extracts time-domain and frequency-domain cross-features using a three-layer convolutional neural network kernel. Combined with the size data of the multi-quantum dot active region, the three-dimensional spatial coordinates of each quantum dot, and the distance between quantum dots, it generates a stress management tensor acting on the multi-quantum dot active region. This accurately captures the comprehensive impact of the coordinated changes in temperature, humidity, and vibration in the time and frequency dimensions on the lattice, avoiding the one-sidedness of single environmental parameter analysis and ensuring no nonlinear interactions are overlooked. Meanwhile, the conduction management unit constructs a node-edge graph structure using a graph neural network, calculates the node stress bearing capacity coefficient and stress conduction correlation, and corrects highly correlated stress management values ​​based on the spacing layer thickness. This generates a lattice distortion management map containing spatial thermal distribution and stress conduction correlation strength, intuitively presenting the correlation relationship of high stress gradient regions and the lattice strain probability at each location, allowing for early identification of active region distortion risks caused by thermal stress.

[0059] The change management unit, leveraging the gating mechanism of the gated loop unit and combining the spatiotemporal frequency three-dimensional gating weights with the quantum dot distribution density, generates a phonon energy management spectrum containing temporal fluctuations, spatial distribution, and frequency domain correlations. This allows for real-time tracking of dynamic changes in lattice distortion and precise analysis of nonlinear changes in quantum confinement effects. The compensation management unit maps the phonon energy management spectrum to a center wavelength drift. Through dual-constraint correction factors and size-sensitive calibration coefficients, it generates an inverse spectral compensation signal that matches the spatiotemporal frequency dimensions of the phonon energy management spectrum. This precisely offsets the center wavelength drift caused by thermal stress, moisture penetration, and environmental vibration, avoiding potential hazards caused by condensation in the encapsulation optical window and interlayer dislocation slippage. The twin management unit calculates the compensation priority coefficient based on the quantum dot's three-dimensional spatial coordinates and the compensation signal strength. After compensation, a wave eigenvalue is obtained. By comparing and analyzing the dynamic deviation threshold and stability coefficient threshold, instructions for maintaining management or re-compensation are generated, dynamically optimizing the management process, avoiding management lag, significantly reducing the probability of quantum dot laser damage, and ensuring the effectiveness of warehouse management. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of an intelligent component warehouse management system based on digital twins according to the present invention;

[0061] Figure 2 This is a flowchart illustrating the steps of an intelligent component warehouse management method based on digital twins according to the present invention. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] refer to Figure 1 and Figure 2 A digital twin-based intelligent component warehouse management system includes:

[0064] The stress management unit is used to acquire temperature, humidity and vibration data in the target management warehouse in real time, and to acquire the center wavelength of the target management object in real time. The center wavelength is used as the reference wave eigenvalue. In the digital twin platform, the temperature, humidity and vibration data are used as inputs. The convolutional neural network extracts the time-domain-frequency domain cross features through the convolutional kernel and performs nonlinear mapping based on the fully connected layer. The output is a stress management tensor acting on the multi-quantum dot active region in the target management object. The target management object is a quantum dot laser. In the quantum dot laser, the active region is the core region for laser generation.

[0065] The conduction management unit is used to construct the active region of multiple quantum dots into a node-edge graph structure in the digital twin platform through a graph neural network. The stress management tensor is used as the graph node feature. The correlation of high stress gradient regions is learned through graph convolutional layers, and the probability distribution of lattice strain at each position is output to generate a lattice distortion management graph.

[0066] The change management unit is used to take the lattice distortion management map as input, capture the dynamic change characteristics of lattice distortion over time through the gating mechanism of the gated recurrent unit neural network, analyze the nonlinear change of quantum confinement effect, and generate the phonon energy management spectrum of each quantum dot.

[0067] The compensation management unit is used to map the phonon energy management spectrum of each quantum dot to the drift amount of the center wavelength of the target managed object, and generate the reverse spectral compensation signal for each quantum dot;

[0068] The twin management unit is used to compensate the target management object through the spectral compensation signal, obtain the center wavelength of the target management object after compensation, use the center wavelength as the feedback wave characteristic value, compare the reference wave characteristic value and the feedback wave characteristic value, and generate twin management instructions.

[0069] By employing multi-unit collaboration to accurately capture the nonlinear interactions of different physical fields, the stress management unit uses a convolutional neural network to extract the time-frequency cross-features of temperature, humidity, and vibration data, outputting a stress management tensor acting on the active region of the quantum dot laser. The conduction management unit constructs a graph structure and learns the correlation relationships of high stress gradient regions, generating a lattice distortion management graph. The change management unit, based on a gated recurrent unit neural network, captures dynamic changes in lattice distortion and generates a phonon energy management spectrum. The compensation management unit maps the center wavelength drift to generate a spectral compensation signal. The twin management unit compares the reference wave eigenvalue with the feedback wave eigenvalue to output management commands, thereby avoiding management lag and effectively preventing damage to the quantum dot laser caused by the combined effects of thermal stress, moisture penetration, and environmental vibration, thus improving management effectiveness.

[0070] In one embodiment, within a digital twin platform, a convolutional neural network is used to extract time-domain and frequency-domain cross-features through convolutional kernels, and nonlinear mapping is performed based on fully connected layers. The output is a stress management tensor acting on the multi-quantum dot active region of the target object, including:

[0071] The dynamic co-variation of preprocessed temperature and humidity data and vibration data over the same time dimension is calculated to generate an environmental temporal co-variation matrix. Specifically, this involves: first, aligning the timestamps of the preprocessed temperature and humidity data with the vibration data to ensure that the same time point corresponds to two sets of data; then, using a fixed-duration sliding window, dividing the aligned data into continuous time periods; for the data within a single window in each time period, calculating the average value of the temperature and humidity data and the average value of the vibration data within that window; then, subtracting the average value from the temperature and humidity value at each time point to obtain the temperature and humidity deviation, and subtracting the average value from the vibration value at each time point to obtain the vibration deviation; then, multiplying the temperature and humidity deviation and vibration deviation at the same time point, summing all the products, and dividing by the number of time points to obtain the co-variation value for that window. The co-variation value reflects the degree of correlation between temperature, humidity, and vibration during that time period; finally, arranging the co-variation values ​​of each window sequentially according to the time order of the sliding window to obtain the environmental temporal co-variation matrix.

[0072] The environmental temporal coordination matrix is ​​input into the first layer convolutional kernel of a convolutional neural network. The coordination variation law in the environmental temporal coordination matrix is ​​analyzed, and temporal features that reflect the influence of the environment on the crystal lattice are extracted to generate a time-loop crystal correlation management map. Specifically, the process includes: inputting the environmental temporal coordination matrix into the first layer convolutional kernel, setting the convolution window to 3×3, and sliding it on the environmental temporal coordination matrix with a stride of 1. At each position, the convolutional kernel weights are multiplied by the coordination values ​​of the environmental temporal coordination matrix one by one, and all products are summed to obtain the initial feature value at that position; using the ReLU activation function to filter effective features: judging each initial feature value one by one, if it is positive, it is retained, and if it is negative, it is set to 0, thus filtering out effective temporal feature values. Finally, all effective temporal feature values ​​are arranged according to the sliding order of the convolutional kernel, and the resulting matrix is ​​the time-loop crystal correlation management map.

[0073] The time-loop crystal correlation management map is transformed into a frequency domain, and then the second layer of a convolutional neural network is used to filter regions in the frequency domain that are strongly correlated with the time-domain correlation features. The effect of environmental changes on the crystal lattice is analyzed to generate a frequency-loop crystal correlation management map. Specifically, this involves: converting the effective time-domain feature values ​​at individual positions in the time-loop crystal correlation management map into frequency-varying intensity values ​​using a Fast Fourier Transform (FFT) algorithm. For example, the intensity is calculated to be 4 at 1 Hz and 2 at 2 Hz. Here, frequency represents the repetition period of environmental changes, and intensity represents the magnitude of the environmental influence on the crystal lattice at that frequency. The effective time-domain feature values ​​at all positions in the matrix are then calculated using FFT. The same transformation is performed, rearranging these effective time-domain feature values ​​according to the row and column positions of the original matrix to obtain the frequency-domain feature matrix. The frequency-domain feature matrix is ​​then input into the second convolutional kernel, with a convolution window of 3×3 and a stride of 1. During sliding, the convolutional kernel weights are multiplied by the corresponding effective time-domain feature values ​​and summed to obtain the initial frequency-domain feature values. The ReLU activation function is then used to filter the initial frequency-domain feature values: if the initial frequency-domain feature value is positive, it is retained; if the initial frequency-domain feature value is negative, it is set to 0, thus obtaining the effective frequency-domain feature values. Finally, according to the order of sliding of the convolutional kernel, first horizontally and then vertically, all the effective frequency-domain feature values ​​are arranged into a new matrix, which is the frequency-ring crystal correlation management diagram.

[0074] The temporal and frequency-domain ring-crystal association management maps are fused using the third-layer convolutional kernel of a convolutional neural network to generate a domain-domain ring-crystal fusion feature tensor. Specifically, the temporal and frequency-domain ring-crystal association management maps are stacked according to the channel dimension to form a two-channel dual-domain feature matrix. The dual-domain feature matrix is ​​input into the third-layer convolutional kernel with a 3×3 window and a stride of 1. During sliding, for each position, the kernel weights are multiplied by the corresponding feature values ​​(effective temporal and effective frequency-domain feature values) of the two channels of the dual-domain feature matrix, and the product results of the two channels are added to obtain the initial fusion feature value. The initial fusion feature value is then processed by the ReLU activation function to retain the effective fusion feature values. All effective fusion feature values ​​are arranged in sliding order into a three-dimensional structure with spatial dimension + channel dimension. This structure is the domain-domain ring-crystal fusion feature tensor.

[0075] In one embodiment, the digital twin platform uses a convolutional neural network to extract time-domain and frequency-domain cross features from temperature and humidity data and vibration data as input. Based on a fully connected layer, a nonlinear mapping is performed to output a stress management tensor acting on the multi-quantum dot active region of the target object. The system also includes:

[0076] Obtain the dimensional data of the active region of multiple quantum dots in a quantum dot laser. The dimensional data includes: the basic three-dimensional dimensions of a single quantum dot, the total thickness of the active region, and the thickness of the spacer layer, etc.; the thickness of the spacer layer represents the thickness of the isolation layer between two adjacent quantum dot layers.

[0077] The domain-ring crystal fusion feature tensor is input into the fully connected layer of the third convolutional kernel. Combined with the size data of the active regions of multiple quantum dots in the quantum dot laser, the fusion feature is converted into a feature related to lattice stress, generating a stress-related feature vector. Specifically, the domain-ring crystal fusion feature tensor is expanded into a one-dimensional feature vector. The effective fusion feature values ​​at each position in the tensor are extracted sequentially according to the channel, row, and column order and arranged into a series of values. Then, the length, width, and height of individual quantum dots, the total thickness of the active region, the thickness of the spacer layer, etc., are arranged into a one-dimensional size vector in a fixed order, such as the size of individual quantum dots, the total thickness, and the thickness of the spacer layer. The one-dimensional feature vector is concatenated with the one-dimensional size vector to form a comprehensive vector. The comprehensive vector is input into the fully connected layer. Each neuron in the fully connected layer is multiplied by each value in the comprehensive vector with a preset weight. All products are added together to obtain the initial correlation value. The initial correlation value is activated by the ReLU activation function to retain positive values, which are the effective correlation values. Finally, all effective correlation values ​​are arranged according to the order of neurons to generate the stress-related feature vector.

[0078] Obtain the three-dimensional spatial coordinates of each quantum dot, and calculate the distance between each quantum dot based on the three-dimensional spatial coordinates of each quantum dot. Specifically, for any two quantum dots, calculate the difference between their x, y, and z coordinates, square the three differences respectively, add the squared results together and take the square root. The resulting value is the straight-line distance between the two quantum dots. Calculate the distance between each pair of all quantum dots one by one to obtain the distance between each quantum dot.

[0079] Based on stress-related feature vectors, and combined with the distances between quantum dots, dimensional reconstruction is performed. The features of the stress-related feature vectors are assigned to corresponding spatial positions to generate a stress management tensor acting on the active region of multiple quantum dots in the target management object. Specifically, this includes: determining the spatial dimensions of the stress management tensor according to the three-dimensional spatial coordinates of each quantum dot: the x-axis corresponds to the lateral position of the quantum dot, the y-axis corresponds to the longitudinal direction, and the z-axis corresponds to the thickness direction, thus constructing a three-dimensional framework; for the distances between all quantum dots, the maximum distance is used as the baseline distance value; for any two quantum dots, the baseline distance value is subtracted from the distance between any two quantum dots to obtain the distance difference; the distance differences of all quantum dot pairs are summed to obtain the total difference value; and the distance difference of a single quantum dot pair is divided by the total difference value to obtain the distance weight of that pair of quantum dots.

[0080] For each element in the stress-related feature vector, its value is multiplied by the distance weights of all quantum dot pairs corresponding to that element to obtain the contribution value of each quantum dot pair. All contribution values ​​are summed to obtain the component stress value generated by a single quantum dot. This operation is repeated for all elements in the stress-related feature vector to sum all component stress values ​​of a single quantum dot to obtain the total stress feature value of that quantum dot. The total stress feature value is then filled into the corresponding position of the three-dimensional frame according to the x, y, z coordinate order of each quantum dot to form a stress management tensor containing stress feature values.

[0081] By aligning timestamps and calculating collaborative values ​​using a sliding window, an environmental time-domain collaborative matrix is ​​constructed to accurately capture the dynamic collaborative changes of temperature, humidity, and vibration data. Then, through three layers of convolutional kernels, time-domain features are progressively extracted, generating a time-ring crystal correlation management map. Simultaneously, frequency-domain features are transformed and filtered to generate a frequency-ring crystal correlation management map. Finally, dual-domain features are fused to generate a domain-ring crystal fusion feature tensor. Combining the size data of the multi-quantum dot active region, three-dimensional spatial coordinates, and distances between quantum dots, the fused features are assigned to corresponding spatial locations through nonlinear mapping and dimensional reconstruction using a fully connected layer. This generates a stress management tensor that precisely acts on the multi-quantum dot active region, reflecting the comprehensive impact of temperature, humidity, and vibration on the lattice. This avoids management lag and prevents damage to the quantum dot laser caused by the synergy of thermal stress, moisture penetration, and environmental vibration, ensuring the reliability of warehouse management.

[0082] In one embodiment, the active regions of multiple quantum dots are constructed as a node-edge graph structure using a graph neural network in the digital twin platform. The stress management tensor is used as the graph node feature. The correlation between high stress gradient regions is learned through graph convolutional layers, outputting the probability distribution of lattice strain at each location, thus generating a lattice distortion management graph, including:

[0083] Based on the stress values ​​of each quantum dot in the stress management tensor, and combined with the three-dimensional dimensions of each quantum dot, the node stress bearing capacity coefficient is calculated. Specifically, this involves: multiplying the three-dimensional dimensions of the quantum dot (length multiplied by width and then by height) to obtain the spatial volume of a single quantum dot; multiplying the stress characteristic value of the quantum dot by the spatial volume to obtain the total stress load that a single quantum dot can withstand; finding the maximum total stress bearing capacity of all quantum dots as a benchmark value; and dividing the total stress load of a single quantum dot by the benchmark value to obtain the node stress bearing capacity coefficient. The node stress bearing capacity coefficient ranges from 0 to 1, with a larger value indicating a stronger stress bearing capacity of the node.

[0084] The stress transmission correlation degree is obtained by analyzing the distance between quantum dots and the nodal stress bearing coefficient. Specifically, for any pair of quantum dots, the maximum distance between all quantum dots is found, the distance between the pair of quantum dots is subtracted from the maximum distance and then divided by the maximum distance to obtain a distance influence factor between 0 and 1; the nodal stress bearing coefficients of the pair of quantum dots are added together and divided by 2 to obtain the average bearing coefficient; the distance influence factor is multiplied by the average bearing coefficient to obtain the stress transmission correlation degree of the pair of quantum dots. The stress transmission correlation degree ranges from 0 to 1. The larger the stress transmission correlation degree value, the stronger the stress transmission correlation between the pair of quantum dots.

[0085] The components related to time-domain features in the domain ring crystal fusion feature tensor are fused with the nodal stress bearing coefficients to generate a nodal stress-time-domain feature vector. Specifically, this involves: expanding all effective time-domain feature values ​​of the time-domain channels in the domain ring crystal fusion feature tensor into a one-dimensional time-domain feature vector in the order of channel, row, and column. Each quantum dot corresponds to one nodal stress bearing coefficient, which is repeatedly filled to expand it into a coefficient vector with the same length as the one-dimensional time-domain feature vector; multiplying the one-dimensional time-domain feature vector with the corresponding elements of the coefficient vector one by one to obtain the fused feature management value; arranging all the fused feature management values ​​in the original order to form a one-dimensional vector is the nodal stress-time-domain feature vector.

[0086] The node stress-temporal feature vector and stress transmission correlation are input into the graph convolutional layer. Using the mean of the stress transmission correlation as a threshold, neighboring nodes whose stress transmission correlation exceeds the threshold are aggregated to generate a high-correlation stress management value. Specifically, this involves: summing the stress transmission correlation of all quantum dot pairs and dividing by the total number of quantum dot pairs to obtain the mean correlation, which is then used as the screening threshold; for each quantum dot (representing a node), iterating through all its paired nodes (representing other quantum dots), retaining nodes whose stress transmission correlation exceeds the mean threshold as high-correlation neighbors; adding the node stress-temporal feature vector of the current node to the corresponding elements of the node stress-temporal feature vectors of all high-correlation neighbors to obtain the aggregated feature sum; and dividing the aggregated feature sum by the total number of nodes participating in the aggregation to obtain the high-correlation stress management value for that node.

[0087] In one embodiment, the active regions of multiple quantum dots are constructed as a node-edge graph structure using a graph neural network in the digital twin platform. The stress management tensor is used as the graph node feature. The correlation between high stress gradient regions is learned through graph convolutional layers, outputting the probability distribution of lattice strain at each location to generate a lattice distortion management graph. The method also includes:

[0088] The spacer layer thickness of the multi-quantum dot active region is calculated to obtain the stress attenuation management factor. Based on the stress attenuation management factor, the highly correlated stress management value is corrected to generate the corrected nodal strain value. Specifically, this involves: for each pair of adjacent quantum dots in the multi-quantum dot active region, finding the maximum value among all adjacent quantum dot pairs of spacer layer thicknesses as the reference thickness; dividing the reference thickness by the spacer layer thickness of a certain pair of adjacent nodes to obtain the stress attenuation management factor of that pair of nodes; and multiplying the highly correlated stress management value of that node by the mean of the stress attenuation management factors of that node and all adjacent nodes. The result is the corrected nodal strain value.

[0089] Based on the three-dimensional spatial coordinates of each quantum dot, the distribution density of each quantum dot is analyzed and combined with the node strain value to calculate the node strain probability value. Specifically, this includes: setting a cube region with a fixed side length centered on a single quantum dot, where the fixed side length is the average distance of all adjacent quantum dots to ensure coverage of surrounding nodes; for the number of all quantum dots in the cube region, the number is divided by the cube volume (volume = side length × side length × side length) to obtain the distribution density of that quantum dot; the maximum distribution density and the maximum high-correlation stress management value of all quantum dots are identified as benchmarks; the distribution density of a single quantum dot is divided by the maximum density to obtain the density ratio, and the high-correlation stress management value of that node is divided by the maximum high-correlation stress management value to obtain the strain ratio; the density ratio and the strain ratio are added together and divided by 2 to obtain the node strain probability value. The node strain probability value ranges from 0 to 1, and the larger the node strain probability value, the higher the probability of strain occurring at that node.

[0090] Based on the three-dimensional spatial coordinates of each quantum dot, the strain probability values ​​of the nodes are mapped to a spatial thermal distribution. Simultaneously, the strength of the stress conduction correlation is superimposed to generate a lattice distortion management map. Specifically, this involves: marking the position of each quantum dot in the three-dimensional space of the digital twin platform according to its x, y, and z three-dimensional coordinates; setting a color gradient according to the node strain probability values ​​(e.g., blue for a probability of 0, red for a probability of 1, with higher probability values ​​resulting in darker colors); matching the node strain probability value of each quantum dot to its corresponding color and displaying that color at its coordinate position to form the basic thermal distribution; setting the thickness of the connecting lines for each pair of quantum dots according to the stress conduction correlation (higher correlation results in thicker lines); drawing a connecting line between the two points to reflect the strength of the conduction relationship; and finally integrating the colored quantum dots and the thickened connecting lines into the same three-dimensional spatial map, which is the lattice distortion management map.

[0091] By calculating the spatial volume and total stress load, the nodal stress bearing capacity coefficient is obtained. Then, the stress transmission correlation degree is calculated based on the distance between quantum dots and the bearing capacity coefficient. At the same time, the temporal component of the domain ring crystal fusion feature tensor is fused to generate the nodal stress-temporal feature vector. The graph convolutional layer aggregates the features of highly correlated nodes with the mean of the correlation degree as the threshold. The nodal strain value is obtained by combining the thickness of the spacer layer. The nodal strain probability value is calculated by combining the quantum dot distribution density. Finally, it is mapped to the spatial thermal distribution and superimposed with the stress transmission correlation degree to generate a lattice distortion management map. This can present the correlation relationship of high stress gradient regions and the lattice strain probability, avoiding management lag.

[0092] In one embodiment, the lattice distortion management map is used as input, and the dynamic changes in lattice distortion over time are captured through the gating mechanism of a gated recurrent unit neural network. The nonlinear changes in the quantum confinement effect are analyzed to generate the phonon energy management spectrum for each quantum dot, including:

[0093] By combining the nodal strain probability value and stress conduction correlation degree with the time series data of the digital twin platform, the ratio of the change amplitude of strain probability to the change amplitude of stress conduction correlation degree within the same time interval is calculated to generate the strain-conduction management time series coefficient. Specifically, this involves: retrieving continuous time series data from the digital twin platform, setting a fixed time interval (10 seconds), finding the nodal strain probability value and stress conduction correlation degree between two consecutive moments within the interval; subtracting the nodal strain probability value of the previous moment from the nodal strain probability value of the next moment, and taking the absolute value of the calculation result as the strain probability change amplitude; subtracting the stress conduction correlation degree of the previous moment from the stress conduction correlation degree of the next moment, and taking the absolute value of the calculation result as the stress conduction correlation degree change amplitude; dividing the strain probability change amplitude by the stress conduction correlation degree change amplitude, and normalizing the calculation result to 0-1 to obtain the strain-conduction management time series coefficient within that time interval.

[0094] The strain-conduction management timing coefficients are input into the reset and update gates of the gated loop unit, and combined with the domain-ring crystal fusion feature tensor, the weighted proportion of the contribution of the strain-conduction management timing coefficients and the frequency domain features is calculated to generate a spatiotemporal-frequency three-dimensional gating weight. Specifically, this includes: expanding the strain-conduction management timing coefficients (time dimension features) into a vector with the same spatial dimension as the domain-ring crystal fusion feature tensor (spatial and frequency domain dimension features), ensuring that each quantum dot corresponds to one strain-conduction management timing coefficient; calculating the sum of all strain-conduction management timing coefficients, dividing the individual strain-conduction management timing coefficient by the sum to obtain the timing weight of the timing coefficient; adding all effective frequency domain feature values ​​in the domain-ring crystal fusion feature tensor to obtain the sum, dividing the individual effective frequency domain feature value by the sum to obtain the frequency domain weight of the frequency domain feature.

[0095] The adapted strain-conduction management timing coefficients and effective frequency domain feature values ​​are input into the reset and update gates, respectively. Within each gate, the initial weighting value for each quantum dot is obtained by multiplying the strain-conduction management timing coefficient by the timing weight and adding the effective frequency domain feature value by the frequency domain weight. This initial weighting value reflects the synergistic contribution of time and frequency domain features. The sigmoid algorithm is used to convert the initial weighting values ​​to values ​​between 0 and 1, obtaining the weight of each quantum dot at each time point, forming a two-dimensional space-time weight table. All relevant frequencies are extracted from the domain ring crystal fusion feature tensor, and the values ​​of each frequency are statistically analyzed. Sum all corresponding intensity values ​​(intensities of the same frequency at different spatial locations) to obtain the total intensity of that frequency; calculate the sum of the total intensities of all frequencies as the baseline value; divide the total intensity of a single frequency by the baseline value to obtain the frequency domain weight of that frequency, with the frequency domain weight ranging from 0 to 1; multiply each value in the two-dimensional weight table by the frequency domain weight of that frequency to obtain the spatial-temporal weight at that frequency; arrange the tables of spatial-temporal weights of all frequencies in ascending order of frequency to form a three-dimensional table of spatial location-time point-frequency, which is the spatiotemporal frequency three-dimensional gating weight.

[0096] Based on a three-dimensional gating weight system of spatiotemporal frequency, strain-transmission management time series coefficients are hierarchically screened to generate a spatiotemporal frequency-related time series feature vector. Specifically, this involves: using the mean of all spatial-temporal weights at a given frequency as a threshold, retaining strain-transmission management time series coefficients corresponding to spatial locations and time points where the spatial-temporal weights exceed the threshold; arranging the retained strain-transmission management time series coefficients at each frequency in ascending order of frequency according to their original spatial location and time point, forming a sub-vector for that frequency; and concatenating the sub-vectors of all frequencies end-to-end to form a one-dimensional vector, which is the spatiotemporal frequency-related time series feature vector. The length of the spatiotemporal frequency-related time series feature vector is the sum of the number of strain-transmission management time series coefficients after screening at each frequency.

[0097] Based on the spatiotemporal frequency correlation time-series feature vector and the distribution density of each quantum dot, the correlation degree of phonon energy change in different time segments is analyzed to generate an energy coordination deviation management value. Specifically, the spatiotemporal frequency correlation time-series feature vector is divided into multiple time segments at fixed time intervals (10 seconds), with each segment corresponding to a set of feature elements; the distribution density of a single quantum dot is divided by the sum of the distribution densities of all quantum dots to obtain the density weight; each feature element in the segment is multiplied by its corresponding density weight and then summed to obtain the phonon energy correlation value of that segment; for any two time segments, the energy correlation value of the latter segment is subtracted from the energy correlation value of the former segment, and the absolute value of the calculation result is taken as the phonon energy change correlation degree between the two segments; the maximum value of all phonon energy change correlation degrees is found, and 1 is subtracted from (the correlation degree of a single phonon energy change divided by the maximum value), and the calculation result is normalized to 0-1 to obtain the energy coordination deviation management value.

[0098] The energy coordination deviation management value is allocated according to the three-dimensional spatial coordinates of each quantum dot, and a three-dimensional gating weight of time, space and frequency is superimposed to generate the phonon energy management spectrum of each quantum dot. Specifically, this includes: matching each quantum dot with one energy coordination deviation management value according to its own coordinates; for each quantum dot, at each time point and at each frequency, multiplying its energy coordination deviation management value by the three-dimensional gating weight corresponding to that time and space frequency to obtain the phonon energy value of that quantum dot at that time and space frequency; arranging the phonon energy values ​​of the same quantum dot at different time points in chronological order to reflect temporal fluctuations; arranging the phonon energy values ​​of different quantum dots at the same time according to three-dimensional coordinates to reflect spatial distribution; and displaying the phonon energy values ​​of the same quantum dot at different frequencies in frequency layers to reflect frequency domain correlation. After integration, a phonon energy management spectrum containing temporal fluctuations, spatial distribution, and frequency domain correlation is formed.

[0099] By combining time-series data from a digital twin platform, strain-conduction management timing coefficients are calculated to capture the dynamic changes of lattice distortion over time. Then, through the reset and update gates of the gated loop unit, the spatiotemporal frequency three-dimensional gating weights are generated from the fusion domain ring crystal fusion feature tensor, achieving deep synergy of spatial, temporal, and frequency domain features. This avoids the one-sidedness of single-dimensional analysis. The spatiotemporal frequency correlation timing feature vector generated by the two-layer screening can extract key correlation features. Combined with the quantum dot distribution density, the energy coordination deviation management value is calculated to analyze the correlation between the nonlinear changes of quantum confinement effect and phonon energy changes. Finally, the energy coordination deviation management value is allocated according to three-dimensional coordinates and superimposed with three-dimensional gating weights to generate a phonon energy management spectrum containing temporal fluctuations, spatial distribution, and frequency domain correlation. This allows for early detection of phonon energy anomalies, real-time reflection of the impact of multi-physics field interactions on the active region, solving the problem of management lag, and significantly reducing the probability of quantum dot laser damage.

[0100] In one embodiment, the phonon energy management spectrum of each quantum dot is mapped to the drift of the center wavelength of the target object, generating a reversed spectral shape compensation signal for each quantum dot, including:

[0101] By combining the phonon energy management spectrum of each quantum dot with the three-dimensional gating weights of spatiotemporal frequency and the node strain probability value, the correlation strength ratio of phonon energy temporal fluctuation, spatiotemporal frequency characteristics, and strain probability is calculated to generate the phonon-strain-frequency domain management coefficient. Specifically, this involves: extracting the phonon energy values ​​of the same quantum dot at different time points from the phonon energy management spectrum; subtracting the energy value of the previous time point from the energy value of the later time point and taking the absolute value to obtain the phonon energy temporal fluctuation amplitude; extracting the three-dimensional gating weight values ​​corresponding to the two time points for the same quantum dot and the same frequency; subtracting the weight value of the previous time point from the weight value of the later time point, and the absolute value of this calculation result is the quantum dot at that frequency. The temporal variation amplitude of the three-dimensional gating weights of spatiotemporal frequency is calculated. Then, the temporal variation amplitude of phonon energy is multiplied by the temporal variation amplitude to obtain the first correlation strength of phonon energy temporal variation-spatiotemporal frequency characteristics. The temporal variation amplitude of phonon energy is multiplied by the nodal strain probability value of the quantum dot to obtain the second correlation strength of phonon energy temporal variation-nodal strain probability. The first and second correlation strengths are added together to obtain the total correlation strength. The first and second correlation strengths are divided by the total correlation strength to obtain their respective correlation proportions. The two correlation proportions are then combined in the order of phonon-spatiotemporal frequency proportion and phonon-strain probability proportion to form a two-dimensional value, which is the phonon-strain-frequency domain management coefficient.

[0102] The phonon-strain-frequency domain management coefficient and the reference wave eigenvalue are analyzed to generate a preliminary center wavelength drift. Specifically, this includes: multiplying the phonon energy temporal fluctuation amplitude by the reference wave eigenvalue to obtain the phonon-spatial-frequency correlation drift component; multiplying the nodal strain probability value by the reference wave eigenvalue to obtain the phonon-strain correlation drift component; and multiplying the phonon-spatial-frequency proportion by the phonon-spatial-frequency correlation drift component plus the phonon-strain probability proportion by the phonon-strain correlation drift component to obtain the preliminary center wavelength drift relative to the reference wave eigenvalue.

[0103] The reference wave eigenvalue and the initial center wavelength drift are calculated to generate a dual-constraint correction factor. Specifically, this includes: using 1% of the reference wave eigenvalue as the maximum allowable drift threshold to constrain the reasonable range of wavelength drift; dividing the absolute value of the initial center wavelength drift by the maximum allowable drift threshold to obtain the drift percentage in the 0-1 range; dividing the reference wave eigenvalue by (reference wave eigenvalue + absolute value of the initial drift) to obtain the reference stability coefficient in the 0-1 range; and multiplying the drift percentage by the reference stability coefficient and normalizing it to 0-1 to obtain the dual-constraint correction factor.

[0104] Based on the dual-constraint correction factor, the size-sensitive calibration coefficient is calculated by combining the three-dimensional dimensions of each quantum dot. The initial center wavelength drift is then double-calibrated using both the dual-constraint correction factor and the size-sensitive calibration coefficient to obtain the accurate drift compensation mapping value. Specifically, this involves: calculating the volume of a single quantum dot by multiplying its length by its width by its height, and then calculating the average volume of all quantum dots as the size reference; dividing the volume of a single quantum dot by the size reference and normalizing the result to 0-1 to obtain the size-sensitive calibration coefficient; multiplying the initial center wavelength drift by the dual-constraint correction factor to obtain the drift after the first calibration; and finally multiplying the drift after the first calibration by the size-sensitive calibration coefficient to obtain the accurate drift compensation mapping value.

[0105] The inverse correlation between the precise drift compensation mapping value and the phonon energy management spectrum is analyzed to generate the inverse spectral compensation signal for each quantum dot. Specifically, this involves: extracting the phonon energy value of a single quantum dot at each time point and frequency from the phonon energy management spectrum; dividing the phonon energy value by the sum of the phonon energy values ​​at all spatiotemporal frequencies of the quantum dot to obtain the energy proportion at that spatiotemporal frequency; multiplying the precise drift compensation mapping value by the energy proportion to obtain the drift contribution component at that spatiotemporal frequency; since inverse compensation is required, the drift contribution component is negative to obtain the inverse compensation component corresponding to each spatiotemporal frequency; arranging them in chronological order, and within each time point arranged in ascending frequency order, the inverse compensation components of all spatiotemporal frequencies are arranged sequentially to form a signal that perfectly matches the spatiotemporal frequency dimension of the phonon energy management spectrum and has the inverse value, which is the inverse spectral compensation signal of the quantum dot.

[0106] By combining the phonon energy management spectrum, the spatiotemporal frequency three-dimensional gating weights, and the nodal strain probability values, the phonon-strain-frequency domain management coefficients are calculated, and then the reference wave eigenvalues ​​are correlated to generate the initial center wavelength drift. Subsequently, through dual calibration using a dual-constraint correction factor and a size-sensitive calibration coefficient, the accurate drift compensation mapping value is obtained. Finally, based on the inverse correlation, an inverse spectral shape compensation signal matching the spatiotemporal frequency dimension of the phonon energy management spectrum is generated. This process can offset the wavelength drift caused by the synergistic effects of thermal stress and vibration in real time, avoid management lag, effectively prevent active region distortion and interlayer dislocation slip, significantly reduce the risk of quantum dot laser damage, and ensure the effectiveness of warehouse storage and management.

[0107] In one embodiment, the target managed object is compensated using a spectral compensation signal to obtain the center wavelength of the compensated target managed object. This center wavelength is used as a feedback wave eigenvalue. The reference wave eigenvalue and the feedback wave eigenvalue are compared to generate a twin management command, including:

[0108] By combining the spectral compensation signal of each quantum dot with the three-dimensional spatial coordinates of each quantum dot, the weight ratio of the compensation signal of the quantum dot at different spatial locations to the wavelength of the target object is calculated, and a compensation priority coefficient is generated. Specifically, this includes: calculating the square root of the sum of the squares of the differences between the quantum dot coordinates and the core region (the core region is the active region of multiple quantum dots) x, y, z, to obtain the straight-line distance from each quantum dot to the core region; taking the maximum straight-line distance of all quantum dots as the benchmark, dividing the benchmark by the straight-line distance of a single quantum dot, and normalizing the result to 0-1, which is the spatial influence weight of that quantum dot;

[0109] For all the inverse compensation components at all time and space frequencies in the inverse spectral compensation signal, calculate the absolute value of each component and sum them to obtain the total intensity of the compensation signal of the quantum dot. The sum of the total intensity of the compensation signals of all quantum dots is used as the reference intensity. Divide the total intensity of the compensation signal of a single quantum dot by the reference intensity and normalize the calculation result to 0-1, which is the signal intensity weight of the quantum dot.

[0110] The initial priority value is obtained by multiplying the spatial influence weight of a single quantum dot by the signal strength weight; the sum of the initial priority values ​​of all quantum dots is calculated to obtain the total initial value; the individual initial priority value is divided by the total initial value and the result is normalized to 0-1, which is the compensation priority coefficient of that quantum dot.

[0111] Based on the compensation priority coefficient, spectral compensation signals are applied to the active region of the target quantum dot in the order of quantum dot spatial distribution. After the compensation operation is completed, the center wavelength of the target quantum dot is acquired in real time and used as the feedback wave eigenvalue. Specifically, this involves: sorting all quantum dots in descending order of compensation priority coefficient, then determining the final spatial application order based on the three-dimensional coordinates of the sorted quantum dots, where the x-axis is from left to right, the y-axis is from front to back, and the z-axis is from bottom to top. Following this order, the spectral compensation signal of each quantum dot is precisely applied to its corresponding spatial position in the active region of the multi-quantum dot, completing the compensation operation for all quantum dots. After compensation, the center wavelength of the target quantum dot is acquired in real time, and this center wavelength is the feedback wave eigenvalue.

[0112] Based on the reference wave eigenvalue, feedback wave eigenvalue, and the phonon energy management spectrum of each quantum dot, the correlation between the difference between the two wave eigenvalues ​​and phonon energy fluctuations is calculated to generate a wavelength deviation correlation degree. Specifically, this includes: subtracting the reference wave eigenvalue from the feedback wave eigenvalue and using the absolute value of the calculation result as the wavelength difference between the reference and feedback waves; identifying all phonon energy values ​​for each quantum dot from the phonon energy management spectrum of each quantum dot, finding the maximum and minimum energy values ​​for a single quantum dot, and subtracting the maximum and minimum values ​​to obtain the energy fluctuation range of that quantum dot; calculating the average energy fluctuation range of all quantum dots and using the average value as the overall phonon energy fluctuation amplitude; dividing the wavelength difference by the overall phonon energy fluctuation amplitude to obtain the correlation ratio; and normalizing the correlation ratio to (correlation ratio + 1) to obtain the wavelength deviation correlation degree. The closer the wavelength deviation correlation degree is to 0, the weaker the correlation between the wavelength deviation and phonon energy fluctuations, and the more stable the compensation effect; the closer the wavelength deviation correlation degree is to 1, the stronger the correlation, and the less stable the compensation effect.

[0113] The influence of temperature and humidity data fluctuations on wavelength deviation correlation is analyzed, and the stability index of compensation effect under environmental interference is calculated to generate compensation dynamic coefficients. Specifically, this includes: finding the absolute values ​​of temperature and humidity deviations at all time points, calculating the mean of these absolute values ​​as the temperature and humidity fluctuation amplitude; dividing the temperature and humidity fluctuation amplitude by the wavelength deviation correlation to obtain the environmental-deviation correlation coefficient; subtracting the environmental-deviation correlation coefficient from 1 and normalizing it to 0-1 to obtain the stability index; multiplying the stability index by the wavelength deviation correlation and normalizing the calculation result to 0-1 to obtain the compensation dynamic coefficients.

[0114] In one embodiment, the target managed object is compensated using a spectral compensation signal to obtain the center wavelength of the compensated target managed object. This center wavelength is used as a feedback wave eigenvalue. The reference wave eigenvalue and the feedback wave eigenvalue are compared to generate a twin management command. The method also includes:

[0115] The dynamic deviation threshold is obtained by calculating the upper limit of wavelength deviation corresponding to the maximum temperature and humidity fluctuation and the mean of the nodal strain probability values. Specifically, this involves: finding the absolute values ​​of temperature and humidity deviations at all time points, identifying the largest absolute value as the maximum temperature and humidity fluctuation; multiplying the maximum temperature and humidity fluctuation by the reference wave eigenvalue to obtain the upper limit of wavelength deviation corresponding to the maximum temperature and humidity fluctuation; summing the nodal strain probability values ​​of all quantum dots to obtain the total strain probability value, dividing the total strain probability value by the total number of quantum dots to obtain the mean of the nodal strain probability values; and summing the upper limit of wavelength deviation and the mean of the nodal strain probability values ​​and dividing by 2 to obtain the dynamic deviation threshold.

[0116] The stability coefficient threshold is obtained by combining the stress transmission correlation with the spacer layer thickness. Specifically, for each pair of adjacent quantum dots, the spacer layer thickness is divided by the reference thickness to obtain the thickness ratio in the range of 0-1; the stress transmission correlation of the adjacent quantum dot pair is multiplied by the thickness ratio to obtain the preliminary stability coefficient of the adjacent quantum dot pair; the mean of the preliminary stability coefficients of all adjacent quantum dot pairs is calculated and normalized to 0-1 to obtain the stability coefficient threshold.

[0117] The wavelength deviation correlation degree is compared with the dynamic deviation threshold, and the compensation dynamic coefficient is compared with the stability coefficient threshold to generate twin management instructions. Specifically, twin management instructions are output based on the following conditions:

[0118] Condition 1: If the wavelength deviation correlation degree is less than the dynamic deviation threshold and the compensation dynamic coefficient is greater than the stability coefficient threshold, generate the first twin management instruction. The first twin management instruction indicates that the current management should be maintained.

[0119] If condition 1 is not met, a second twin management instruction is generated. The second twin management instruction indicates that the target management object is recompensated from the stress management unit, transmission management unit, change management unit, compensation management unit and twin management unit until condition 1 is met. The recompensation can be repeated up to 3 times. If it is repeated more than 3 times, an alarm will be triggered.

[0120] By using the three-dimensional spatial coordinates of quantum dots and the spectral compensation signal intensity, the spatial influence weight and signal intensity weight are calculated to obtain the compensation priority coefficient. The compensation signal is then applied precisely according to the priority and spatial order. After compensation, the feedback wave eigenvalue is compared with the reference wave eigenvalue. The wavelength deviation correlation is calculated by combining the phonon energy management spectrum, and the dynamic compensation coefficient is calculated by incorporating the temperature and humidity fluctuation amplitude. Finally, a two-way judgment is made through the dynamic deviation threshold and the stability coefficient threshold to generate twin management instructions for maintenance management or re-compensation. This effectively prevents center wavelength drift and active region distortion, avoids damage to the quantum dot laser, and ensures the stability of warehouse management.

[0121] In one embodiment, a digital twin-based intelligent component warehouse management method is applied to the aforementioned digital twin-based intelligent component warehouse management system, comprising:

[0122] Step S1: Real-time acquisition of temperature and humidity data and vibration data within the target management warehouse, and real-time acquisition of the center wavelength of the target management object. Using the center wavelength as the reference wave eigenvalue, in the digital twin platform, a convolutional neural network is used to extract time-domain and frequency-domain cross features through the convolutional kernel, and nonlinear mapping is performed based on the fully connected layer to output the stress management tensor acting on the multi-quantum dot active region of the target management object, where the target management object is a quantum dot laser.

[0123] Step S2: In the digital twin platform, the active region of multiple quantum dots is constructed as a node-edge graph structure through a graph neural network. The stress management tensor is used as the graph node feature. The correlation between high stress gradient regions is learned through graph convolutional layers. The probability distribution of lattice strain at each position is output to generate a lattice distortion management graph.

[0124] Step S3: Using the lattice distortion management map as input, the dynamic change characteristics of lattice distortion over time are captured through the gating mechanism of the gated recurrent unit neural network, the nonlinear change of quantum confinement effect is analyzed, and the phonon energy management spectrum of each quantum dot is generated.

[0125] Step S4: Map the phonon energy management spectrum of each quantum dot to the drift of the center wavelength of the target managed object, and generate the reverse spectral compensation signal for each quantum dot.

[0126] Step S5: Compensate the target management object with the spectral compensation signal to obtain the center wavelength of the compensated target management object. Use the center wavelength as the feedback wave eigenvalue and compare the reference wave eigenvalue and the feedback wave eigenvalue to generate twin management instructions.

[0127] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital twin-based intelligent component warehouse management system, characterized in that, include: The stress management unit is used to acquire temperature, humidity and vibration data in the target management warehouse in real time, and to acquire the center wavelength of the target management object in real time. The center wavelength is used as the reference wave eigenvalue. In the digital twin platform, the temperature, humidity and vibration data are used as inputs. The convolutional neural network extracts the time-domain-frequency domain cross features through the convolutional kernel, and performs nonlinear mapping based on the fully connected layer. The output is the stress management tensor acting on the multi-quantum dot active region of the target management object, which is a quantum dot laser. This includes combining the size data of the active regions of multiple quantum dots, converting the time-domain-frequency-domain cross features into stress-related feature vectors, and reconstructing the dimensions by combining the distance between each quantum dot. The features of the stress-related feature vectors are then assigned to the corresponding spatial positions of each quantum dot to obtain the stress feature values ​​of each quantum dot. These stress feature values ​​are then filled into the corresponding spatial positions to form a stress management tensor containing stress feature values. The stress conduction management unit is used in a digital twin platform to construct a node-edge graph structure of the active regions of multiple quantum dots using a graph neural network. Using the stress management tensor as the graph node feature, a graph convolutional layer aggregates features of adjacent nodes exceeding the mean of the stress conduction correlation degree as a threshold, learns the correlation relationships of high stress gradient regions, outputs the probability distribution of lattice strain at each location, and generates a lattice distortion management map. Specifically, based on the stress feature values ​​and the three-dimensional dimensions of each quantum dot, the stress bearing capacity coefficient of the nodes is calculated; and the stress conduction correlation degree is obtained based on the distance between each quantum dot and the bearing capacity coefficient. The change management unit takes the lattice distortion management map as input and captures the dynamic changes in lattice distortion over time through the gating mechanism of a gated recurrent unit neural network. It analyzes the nonlinear changes in the quantum confinement effect and generates the phonon energy management spectrum for each quantum dot. Specifically, based on the correlation between the node strain probability value and stress conduction in the probability distribution and the time-series data from the digital twin platform, strain-conduction management time-series coefficients are generated. These time-series coefficients are input to the reset and update gates, and combined with the time-domain-frequency domain cross-features, the weighted proportion of the contribution of the time-series coefficients and frequency domain features is calculated to generate a three-dimensional spatiotemporal-frequency gating weight. This time-series coefficient is then subjected to hierarchical filtering to generate a spatiotemporal-frequency correlated time-series feature vector. Based on this feature vector and the distribution density of each quantum dot, the correlation of phonon energy changes in different time segments is analyzed to generate an energy coordination deviation management value. This management value is allocated according to the three-dimensional spatial coordinates of each quantum dot, and simultaneously superimposed with the spatiotemporal-frequency three-dimensional gating weight to generate the phonon energy management spectrum for each quantum dot. The compensation management unit is used to map the phonon energy management spectrum of each quantum dot to the drift amount of the center wavelength of the target managed object, and generate the reverse spectral compensation signal for each quantum dot; The twin management unit is used to compensate the target management object through the spectral compensation signal, obtain the center wavelength of the target management object after compensation, use the center wavelength as the feedback wave characteristic value, compare the reference wave characteristic value and the feedback wave characteristic value, and generate twin management instructions.

2. The intelligent component warehouse management system based on digital twins according to claim 1, characterized in that, In the digital twin platform, a convolutional neural network is used to extract time-domain and frequency-domain cross features through convolutional kernels, and nonlinear mapping is performed based on fully connected layers. The output is a stress management tensor acting on the multi-quantum dot active region of the target object, including: Calculate the dynamic coordinated changes of preprocessed temperature and humidity data and vibration data in the same time dimension to generate an environmental time-domain coordinated matrix. The environmental temporal coordination matrix is ​​input into the first convolutional kernel of the convolutional neural network. The coordination change law in the environmental temporal coordination matrix is ​​analyzed, and the temporal features that reflect the influence of the environment on the crystal lattice are extracted to generate a time-loop crystal correlation management diagram. The time-ring crystal correlation management map is transformed into the frequency domain, and then the second layer convolution kernel of the convolutional neural network is used to filter the regions in the frequency domain that are strongly correlated with the time-domain correlation features. The effect of environmental changes on the crystal lattice is analyzed to generate the frequency-ring crystal correlation management map. The temporal and frequency ring correlation management graphs are fused using the third convolutional kernel of a convolutional neural network to generate a domain ring fusion feature tensor.

3. The intelligent component warehouse management system based on digital twins according to claim 2, characterized in that, In the digital twin platform, a convolutional neural network is used to extract time-domain and frequency-domain cross features through convolutional kernels, and nonlinear mapping is performed based on fully connected layers. The output is a stress management tensor acting on the multi-quantum dot active region of the target object. This also includes: The domain ring crystal fusion feature tensor is input into the fully connected layer of the third convolution kernel. Combined with the size data of the active region of multiple quantum dots in the quantum dot laser, the fusion feature is converted into a feature related to lattice stress, generating a stress-related feature vector. Obtain the three-dimensional spatial coordinates of each quantum dot, and calculate the distance between each quantum dot based on the three-dimensional spatial coordinates of each quantum dot; Based on the stress-related feature vector, the dimensions are reconstructed by combining the distance between each quantum dot, and the features of the stress-related feature vector are assigned to the corresponding spatial positions to generate a stress management tensor that acts on the active region of multiple quantum dots in the target management object.

4. The intelligent component warehouse management system based on digital twins according to claim 3, characterized in that, In the digital twin platform, a graph neural network is used to construct the active region of multiple quantum dots as a node-edge graph structure. Using the stress management tensor as the graph node feature, graph convolutional layers learn the correlation between high stress gradient regions, outputting the probability distribution of lattice strain at each location, and generating a lattice distortion management map, including: Based on the stress eigenvalues ​​of each quantum dot in the stress management tensor, and combined with the three-dimensional dimensions of each quantum dot, the stress bearing capacity coefficient of the node is calculated. The stress transmission correlation was obtained by analyzing the distance between each quantum dot and the stress bearing coefficient of the node. The components related to time-domain features in the domain ring crystal fusion feature tensor are fused with the nodal stress bearing coefficient to generate a nodal stress-time-domain feature vector. The stress-temporal feature vectors of nodes and the stress transmission correlation are input into the graph convolutional layer. The mean value of the stress transmission correlation is used as the threshold. The features of adjacent nodes whose stress transmission correlation exceeds the threshold are aggregated to generate highly correlated stress management values.

5. The intelligent component warehouse management system based on digital twin according to claim 4, characterized in that, In the digital twin platform, a graph neural network is used to construct the active region of multiple quantum dots as a node-edge graph structure. Using the stress management tensor as the graph node feature, graph convolutional layers learn the correlation between high stress gradient regions, outputting the probability distribution of lattice strain at each location, and generating a lattice distortion management map. This also includes: The thickness of the spacer layer in the active region of the multi-quantum dot is calculated to obtain the stress attenuation management factor. Based on the stress attenuation management factor, the highly correlated stress management value is corrected to generate the corrected nodal strain value. Based on the three-dimensional spatial coordinates of each quantum dot, the distribution density of each quantum dot is analyzed and combined with the nodal strain value to calculate the nodal strain probability value. Based on the three-dimensional spatial coordinates of each quantum dot, the strain probability value of the node is mapped to the spatial thermal distribution, and the strength of the stress conduction correlation is superimposed to generate a lattice distortion management map.

6. The intelligent component warehouse management system based on digital twin according to claim 5, characterized in that, The phonon energy management spectrum of each quantum dot is mapped to the drift of the center wavelength of the target object, generating inverse spectral compensation signals for each quantum dot, including: By combining the phonon energy management spectrum of each quantum dot with the three-dimensional gating weights of spatiotemporal frequency and node strain probability, the correlation strength ratio of phonon energy temporal fluctuation, spatiotemporal frequency characteristics, and strain probability is calculated to generate phonon-strain-frequency domain management coefficients. Analyze the phonon-strain-frequency domain management coefficients and reference wave eigenvalues ​​to generate a preliminary center wavelength shift; The reference wave eigenvalues ​​and the initial center wavelength drift are calculated to generate a dual-constraint correction factor; Based on the dual-constraint correction factor, the size-sensitive calibration coefficient is calculated by combining the three-dimensional size of each quantum dot. The initial center wavelength drift is then calibrated by the dual-constraint correction factor and the size-sensitive calibration coefficient to obtain the accurate drift compensation mapping value. The inverse correlation between the precise drift compensation mapping value and the phonon energy management spectrum is analyzed to generate the inverse spectral compensation signal for each quantum dot.

7. The intelligent component warehouse management system based on digital twin according to claim 6, characterized in that, The target managed object is compensated using a spectral compensation signal to obtain the center wavelength of the compensated target managed object. This center wavelength is used as the feedback wave characteristic value. The reference wave characteristic value and the feedback wave characteristic value are compared to generate twin management instructions, including: By combining the spectral compensation signal of each quantum dot with the three-dimensional spatial coordinates of each quantum dot, the weight ratio of the influence of the compensation signal of the quantum dot at different spatial locations on the wavelength of the target management object is calculated, and a compensation priority coefficient is generated. Based on the compensation priority coefficient, the spectral compensation signal is applied to the multi-quantum dot active region of the target management object according to the spatial distribution order of quantum dots. After the compensation operation is completed, the center wavelength of the target management object is collected in real time and used as the feedback wave eigenvalue. Based on the reference wave eigenvalue, feedback wave eigenvalue, and phonon energy management spectrum of each quantum dot, the correlation between the difference between the two wave eigenvalues ​​and phonon energy fluctuations is calculated to generate wavelength deviation correlation degree. The influence of temperature and humidity data fluctuations on wavelength deviation correlation is analyzed, the stability index of compensation effect under environmental interference is calculated, and compensation dynamic coefficients are generated.

8. The intelligent component warehouse management system based on digital twin according to claim 7, characterized in that, The target managed object is compensated using a spectral compensation signal to obtain the center wavelength of the compensated target managed object. This center wavelength is used as the feedback wave characteristic value. The reference wave characteristic value and the feedback wave characteristic value are compared to generate twin management instructions. This also includes: The dynamic deviation threshold is obtained by calculating the average of the wavelength deviation upper limit and the nodal strain probability value corresponding to the maximum temperature and humidity fluctuation. The stability coefficient threshold is obtained by combining the stress transmission correlation degree with the thickness of the spacer layer; The wavelength deviation correlation degree is compared with the dynamic deviation threshold, the compensation dynamic coefficient and the stability coefficient threshold respectively, and twin management instructions are generated.

9. A method for intelligent component warehouse management based on digital twins, applied to the intelligent component warehouse management system based on digital twins as described in any one of claims 1-8, characterized in that, include: Step S1: Real-time acquisition of temperature and humidity data and vibration data within the target management warehouse, and real-time acquisition of the center wavelength of the target management object. Using the center wavelength as the reference wave eigenvalue, in the digital twin platform, a convolutional neural network is used to extract time-frequency cross features through the convolution kernel, with temperature and humidity data and vibration data as input. Nonlinear mapping is then performed based on a fully connected layer to output the stress management tensor acting on the multi-quantum dot active region of the target management object, where the target management object is a quantum dot laser. This includes combining the size data of the active regions of multiple quantum dots, converting the time-domain-frequency-domain cross features into stress-related feature vectors, and reconstructing the dimensions by combining the distance between each quantum dot. The features of the stress-related feature vectors are then assigned to the corresponding spatial positions of each quantum dot to obtain the stress feature values ​​of each quantum dot. These stress feature values ​​are then filled into the corresponding spatial positions to form a stress management tensor containing stress feature values. Step S2: In the digital twin platform, the active regions of multiple quantum dots are constructed as a node-edge graph structure using a graph neural network. The stress management tensor is used as the graph node feature. A graph convolutional layer is used with the mean of the stress conduction correlation degree as a threshold. Neighboring nodes exceeding this threshold are feature-aggregated to learn the correlation relationships of high stress gradient regions, outputting the probability distribution of lattice strain at each location, and generating a lattice distortion management map. Specifically, based on the stress feature values ​​and the three-dimensional dimensions of each quantum dot, the node stress bearing coefficient is calculated. The stress conduction correlation degree is obtained based on the distance between each quantum dot and the bearing coefficient. Step S3: Using the lattice distortion management map as input, the dynamic changes in lattice distortion over time are captured through the gating mechanism of a gated recurrent unit neural network. The nonlinear changes in the quantum confinement effect are analyzed to generate the phonon energy management spectrum for each quantum dot. Specifically, based on the correlation between the node strain probability value and stress conduction in the probability distribution, combined with the time-series data from the digital twin platform, strain-conduction management time-series coefficients are generated. These time-series coefficients are input into the reset and update gates, and combined with the time-frequency cross-features, the weighted proportion of the contribution of the time-series coefficients and the frequency domain features is calculated to generate a three-dimensional spatiotemporal-frequency gating weight. This time-series coefficient is then subjected to hierarchical screening to generate a spatiotemporal-frequency correlated time-series feature vector. Based on this feature vector and the distribution density of each quantum dot, the correlation of phonon energy changes in different time segments is analyzed to generate an energy cooperative deviation management value. This management value is allocated according to the three-dimensional spatial coordinates of each quantum dot, and the spatiotemporal-frequency three-dimensional gating weight is superimposed to generate the phonon energy management spectrum for each quantum dot. Step S4: Map the phonon energy management spectrum of each quantum dot to the drift of the center wavelength of the target managed object, and generate the reverse spectral compensation signal for each quantum dot. Step S5: Compensate the target management object with the spectral compensation signal to obtain the center wavelength of the compensated target management object. Use the center wavelength as the feedback wave eigenvalue and compare the reference wave eigenvalue and the feedback wave eigenvalue to generate twin management instructions.

Images