Gallium oxide crystal storage environment monitoring system and method based on artificial intelligence

Abstract

The invention discloses a gallium oxide crystal storage environment monitoring system and method based on artificial intelligence, and relates to the technical field of intelligent industrial monitoring, and the method comprises the steps: inputting a multi-source coupling data set into a space-time hypergraph model, carrying out the multi-source data coupling of a hypergraph construction layer, and carrying out the spatial dependence capture and time sequence feature extraction of a space-time convolution layer, the method comprises the steps of forming an environment parameter prediction matrix, performing damage quantification and probability mapping on a gallium oxide crystal damage image, obtaining a reference damage distribution diagram, performing coupling association on the environment prediction matrix and the reference damage distribution diagram, outputting an environment-damage weight matrix, and performing defect positioning on the reference damage distribution diagram according to the environment-damage weight matrix. And obtaining a damage sensitive area. According to the invention, through the space-time hypergraph model, the DBSCAN clustering analysis and the risk grade division strategy, the accuracy and the real-time performance of the monitoring scheme are enhanced.

Description

Technical Field

[0001] This invention relates to the field of intelligent industrial monitoring technology, and in particular to a gallium oxide crystal storage environment monitoring system and method based on artificial intelligence. Background Technology

[0002] Gallium oxide (GaO) crystals, as a novel wide-bandgap semiconductor material, have shown great application potential in high-power electronic devices and ultraviolet photodetectors due to their excellent electrical properties and thermal stability. With the increasing demand for GaO crystals, ensuring their stability and reliability during storage has become a key research focus. Traditional storage environment monitoring technologies mainly rely on sensor networks and simple data processing algorithms to assess crystal safety by monitoring changes in basic environmental parameters such as temperature and humidity.

[0003] However, existing technologies still have some limitations. First, most existing environmental prediction models fail to fully integrate multi-source heterogeneous data, resulting in an insufficient and in-depth understanding of the storage environment and limiting the accuracy of storage environment monitoring solutions. Second, existing technologies struggle to accurately identify potential risk areas, leading to inaccurate location of damage-sensitive areas. This makes it impossible to achieve dynamic early warning and refined control of crystal damage risks under complex and ever-changing environmental conditions, thus limiting the real-time performance of storage environment monitoring solutions. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an artificial intelligence-based method for monitoring gallium oxide crystal storage environments to address the problems of insufficient accuracy and low real-time performance in existing storage environment monitoring schemes.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for monitoring the storage environment of gallium oxide crystals based on artificial intelligence, comprising,

[0008] The multi-source coupled dataset is input into the spatiotemporal hypergraph model. The hypergraph construction layer couples the multi-source data, and the spatiotemporal convolutional layer performs spatial dependency capture and temporal feature extraction to form an environmental parameter prediction matrix.

[0009] Damage quantization and probability mapping are performed on gallium oxide crystal damage images to obtain a baseline damage distribution map. The environmental prediction matrix is ​​coupled and correlated with the baseline damage distribution map to output an environment-damage weight matrix.

[0010] Based on the environment-damage weight matrix, defects are located on the baseline damage distribution map to obtain damage-sensitive areas. DBSCAN cluster analysis is then performed on the damage-sensitive areas to generate a set of high-risk area coordinates.

[0011] The risk level of the high-risk area coordinate set is classified and the area is aggregated to obtain a risk level distribution map. Based on the risk level distribution map, the storage environment is monitored in real time, and an environment monitoring plan is output synchronously.

[0012] As a preferred embodiment of the artificial intelligence-based gallium oxide crystal storage environment monitoring method of the present invention, the multi-source coupled dataset includes gallium oxide crystal damage images, storage environment change data, and crystal characteristic change data.

[0013] As a preferred embodiment of the gallium oxide crystal storage environment monitoring method based on artificial intelligence described in this invention, the step of forming an environmental parameter prediction matrix specifically includes the following steps.

[0014] A hypergraph construction layer and a spatiotemporal convolutional layer are constructed, and skip connections are used to perform high-order information aggregation and residual stacking on the hypergraph construction layer and the spatiotemporal convolutional layer to construct a spatiotemporal hypergraph model.

[0015] The multi-source coupled dataset is input into the spatiotemporal hypergraph model. The hypergraph construction layer couples the multi-source data through an attention mechanism to obtain the hypergraph associated feature tensor.

[0016] Spatiotemporal convolutional layers apply gated convolution and hypergraph convolution to perform spatial dependency capture and temporal feature extraction, generating spatiotemporal fusion features;

[0017] The hypergraph association feature tensor and spatiotemporal fusion features are concatenated to form an environmental parameter prediction matrix.

[0018] As a preferred embodiment of the artificial intelligence-based gallium oxide crystal storage environment monitoring method of the present invention, the step of obtaining the benchmark damage distribution map specifically includes the following steps.

[0019] Multidimensional scanning is performed on gallium oxide crystal damage images to obtain damage depth sequence images; Delaunay triangulation and damage quantization are performed on the damage depth sequence images to generate damage volume and mass matrices.

[0020] A baseline damage distribution map is obtained by performing a probability mapping between the damage volume and mass matrix.

[0021] As a preferred embodiment of the AI-based gallium oxide crystal storage environment monitoring method of the present invention, the output environment-damage weight matrix specifically includes the following steps.

[0022] Spatial registration and kriging interpolation are performed on the environmental prediction matrix to generate a continuous environmental prediction distribution.

[0023] The continuous environment prediction distribution and the baseline damage distribution map are coupled and probabilistically fused by the Hadamard product to output the environment-damage weight matrix.

[0024] As a preferred embodiment of the AI-based gallium oxide crystal storage environment monitoring method of the present invention, the step of generating a set of high-risk area coordinates specifically includes the following steps.

[0025] The environment-damage weight matrix is ​​weighted to obtain the damage sensitivity coefficient; based on the damage sensitivity coefficient, the baseline damage distribution map is spatially sampled to generate a set of damage area points.

[0026] Neighborhood expansion is performed on the point set of the damaged area to form the damaged sensitive area; DBSCAN clustering is applied to the damaged sensitive area for density clustering to obtain the coordinate set of high-risk areas.

[0027] As a preferred embodiment of the gallium oxide crystal storage environment monitoring method based on artificial intelligence described in this invention, the formation of the environment monitoring scheme specifically includes the following steps.

[0028] The coordinate set of high-risk areas is divided into high-risk, medium-risk, and low-risk risk levels using a three-level threshold.

[0029] Simultaneously perform regional aggregation on the coordinate set of high-risk areas to form continuous closed risk areas; perform GIS spatial overlay analysis on the divided risk levels and continuous closed risk areas to obtain a risk level distribution map;

[0030] Based on the risk level distribution map, the storage environment is monitored in different zones and levels. Real-time storage feedback data is collected and integrated to form an environmental monitoring plan.

[0031] Secondly, the present invention provides an artificial intelligence-based gallium oxide crystal storage environment monitoring system, comprising,

[0032] The environment prediction module is used to input multi-source coupled datasets into the spatiotemporal hypergraph model. The hypergraph construction layer couples multi-source data, and the spatiotemporal convolutional layer performs spatial dependency capture and temporal feature extraction to form an environment parameter prediction matrix.

[0033] The coupling and correlation module is used to perform damage quantization and probability mapping on gallium oxide crystal damage images, obtain a baseline damage distribution map, couple and correlate the environmental prediction matrix with the baseline damage distribution map, and output the environment-damage weight matrix.

[0034] The localization and clustering module is used to locate defects on the baseline damage distribution map based on the environment-damage weight matrix, obtain damage-sensitive areas, and perform DBSCAN clustering analysis on the damage-sensitive areas to generate a set of high-risk area coordinates.

[0035] The real-time monitoring module is used to classify and aggregate the coordinates of high-risk areas to obtain a risk level distribution map. Based on the risk level distribution map, the storage environment is monitored in real time, and an environment monitoring plan is output synchronously.

[0036] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the method for monitoring the gallium oxide crystal storage environment based on artificial intelligence as described in the first aspect of the present invention.

[0037] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the gallium oxide crystal storage environment monitoring method based on artificial intelligence as described in the first aspect of the present invention.

[0038] The beneficial effects of this invention are as follows: The application of a spatiotemporal hypergraph model enables deep fusion and high-order feature extraction of multi-source coupled data, thereby significantly improving the accuracy of environmental parameter prediction and enhancing the precision of the monitoring scheme. By combining DBSCAN clustering analysis with a risk level classification strategy, high-risk areas can be efficiently identified and aggregated, thus supporting real-time monitoring and tiered response of the storage environment, effectively improving the real-time performance of the monitoring scheme. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of an artificial intelligence-based method for monitoring the storage environment of gallium oxide crystals.

[0041] Figure 2 This is a schematic diagram of an artificial intelligence-based gallium oxide crystal storage environment monitoring system.

[0042] Figure 3 A flowchart for generating the environment-damage weight matrix.

[0043] Figure 4 A flowchart for generating the coordinate set of high-risk areas. Detailed Implementation

[0044] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0045] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0046] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0047] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for monitoring the gallium oxide crystal storage environment based on artificial intelligence, including the following steps:

[0048] S1. Input the multi-source coupled dataset into the spatiotemporal hypergraph model. The hypergraph construction layer couples the multi-source data, and the spatiotemporal convolutional layer performs spatial dependency capture and temporal feature extraction to form an environmental parameter prediction matrix.

[0049] Specifically, the operations include the following:

[0050] S1.1 Acquire multi-source coupled datasets, which include gallium oxide crystal damage images, storage environment change data, and crystal property change data.

[0051] Damage images of gallium oxide crystals were acquired through non-contact scanning of gallium oxide crystals using a high-resolution optical microscope and a confocal scanning imaging unit.

[0052] The storage environment change data includes temperature, humidity, air pressure and vibration parameters; temperature and humidity are collected by high-precision temperature and humidity sensors, air pressure is collected by barometers, and vibration parameters are monitored in real time using triaxial accelerometers.

[0053] The data on changes in crystal properties include crystal resistivity, defect density, and photoluminescence intensity. Crystal resistivity was acquired using a four-probe tester, defect density was acquired using an electron beam induced current unit (EBIC), and photoluminescence intensity was acquired using a spectrometer.

[0054] S1.2. Preprocessing of the multi-source coupled dataset: Specifically, for gallium oxide crystal damage images, Gaussian filtering is applied for image denoising to eliminate high-frequency noise, and the morphological closing operation function of the OpenCV library is used for image filling to ensure the contour integrity of the gallium oxide crystal damage images; for data on changes in storage environment, Z-score normalization is used for data normalization, and Savitzky-Golay filtering is applied for smoothing to suppress sampling fluctuations, while linear interpolation is used to fill missing values ​​to maintain data continuity; for data on changes in crystal characteristics, mean normalization is used for normalization, and principal component analysis is used for feature dimensionality reduction to eliminate redundant information, outputting the preprocessed multi-source coupled dataset.

[0055] S1.3. Construct and train the spatiotemporal hypergraph model. Specifically, in the PyTorch framework, a multi-head attention architecture is called using the `nn.Module` parameter, with the number of heads set to 8, the embedding dimension to 256, and dropout set to 0.1. After the multi-head attention architecture, LayerNorm is applied for feature normalization, and GeLU activation is used for non-linear transformation to enhance feature representation, thus completing the hypergraph construction layer. The spatiotemporal convolutional architecture is then called using the `nn.Conv2d` function, embedding gated convolutions and hypergraph convolutions. The kernel size, dilation factor, and hypergraph convolution order are set to 3. Finally, a fully connected layer is applied after the spatiotemporal convolutional architecture, with the feature dimension adjusted to adapt to downstream tasks, completing the spatiotemporal convolutional layer construction.

[0056] Skip connections are used to concatenate features between the hypergraph construction layer and the spatiotemporal convolutional layer to obtain enhanced spatiotemporal features. These enhanced spatiotemporal features are then batch-normalized to generate a standardized feature matrix. High-order information aggregation is performed on the standardized feature matrix to obtain global spatiotemporal features. Fully connected layers are applied to linearly project these global spatiotemporal features to generate environmental parameter prediction weights. Based on these environmental parameter prediction weights, residual connections are used to stack the hypergraph construction layer and the spatiotemporal convolutional layer, and the hierarchical sequence is performed using the nn.Sequential function to complete the construction of the spatiotemporal hypergraph model.

[0057] Next, the spatiotemporal hypergraph model is trained. Further, the historical multi-source coupled dataset is divided into a sample set, a training set, and a validation set. On the sample set, a DataLoader is used to load data, and random numbers are generated through random sampling to form training batches. On the training set, the AdamW optimizer is used to update the parameters of the training batches, and a cosine annealing scheduler is applied simultaneously to adjust the learning rate, obtaining optimized spatiotemporal hypergraph model parameters. On the validation set, the mean squared error loss function is used to quantize the loss of the optimized spatiotemporal hypergraph model parameters, obtaining the training loss value. When the training loss value exceeds the convergence threshold for five consecutive rounds, training terminates, and the trained spatiotemporal hypergraph model is output using torch.save.

[0058] It should be noted that the convergence threshold is defined based on the fluctuation range of the training loss value, and the value range is [0.01~0.05].

[0059] S1.4 The hypergraph construction layer couples multi-source data through an attention mechanism to obtain the hypergraph associated feature tensor. Specifically, the multi-source coupled dataset is input into the spatiotemporal hypergraph model through a real-time stream processing pipeline. The hypergraph construction layer applies 3×3 convolution to perform local feature extraction on the preprocessed gallium oxide crystal damage image to obtain a spatial feature map. Global average pooling is used to perform dimensionality reduction and compression on the spatial feature map to generate high-dimensional abstract image features. Temporal encoding and weighted aggregation are performed on the preprocessed storage environment change data through a gating mechanism to generate temporal environment features. A fully connected layer is used to perform linear transformation and nonlinear activation on the preprocessed crystal characteristic change data to generate crystal characteristic features. Dimensional alignment and fusion are performed on the high-dimensional abstract image features, temporal environment features, and crystal characteristic features to obtain the multi-source coupled feature tensor.

[0060] An attention mechanism is applied to the multi-source coupled feature tensor to perform feature interaction, resulting in multi-head attention features. Dot product operations are performed on the multi-head attention features to quantify the correlation strength between features, forming the original attention score. The Softmax function is applied to normalize the original attention score to obtain the correlation weight distribution. Based on the correlation weight distribution, the multi-head attention features are weighted and aggregated to obtain the hyperedge feature representation. Hyperedge connection and hypergraph sparsification are performed on the hyperedge feature representation to obtain the spatiotemporal dependent hypergraph structure. Hypergraph convolution is performed on the spatiotemporal dependent hypergraph structure to obtain the hypergraph correlation feature tensor.

[0061] S1.5. The spatiotemporal convolutional layer applies gated convolution and hypergraph convolution to perform spatial dependency capture and temporal feature extraction, generating spatiotemporal fusion features. Specifically, the hypergraph-related feature tensor is synchronously fed into the gated convolution and the parallel branch of the hypergraph convolution in the spatiotemporal convolutional layer: In the hypergraph convolution branch, spectral domain convolution is performed on the hypergraph-related feature tensor. Further, feature decomposition is performed on the hypergraph-related feature tensor to obtain the feature vector matrix. The spectral domain feature representation of the feature vector matrix is ​​extracted through Fourier transform: polynomial filtering is performed on the feature vector matrix to obtain the filtered frequency domain features. Feature scaling is performed on the filtered frequency domain features to obtain normalized frequency domain features. Feature recombination is performed on the normalized frequency domain features to output the spectral domain feature representation; hypergraph convolution is applied to perform spatial transformation on the spectral domain feature representation to generate spatial dependency features.

[0062] In the gated convolution branch, the hypergraph association feature tensor is expanded. Furthermore, temporal feature extraction is performed on the hypergraph association feature tensor through 3×3 depthwise convolution to obtain temporal local features. A 1×1 gated convolution is used to perform linear transformation and sigmoid activation on the temporal local features to generate gate weights. The gate weights are multiplied point by point with the hypergraph association feature tensor to output temporal augmentation features.

[0063] Temporal enhancement features and spatial dependency features are concatenated through feature channels to obtain spatiotemporal fusion features. Dilated causal convolution is used to expand the receptive field of the spatiotemporal fusion features, and residual connections are applied to stack the layers to output multi-scale feature representations. Layer normalization and weighted fusion are performed on the multi-scale feature representations to finally output spatiotemporal fusion features.

[0064] S1.6. Concatenate the hypergraph association feature tensor and the spatiotemporal fusion feature to form an environmental parameter prediction matrix. Specifically, perform feature concatenation and nonlinear activation on the hypergraph association feature tensor and the spatiotemporal fusion feature to obtain the activated features. Then, assign feature channel weights to the activated features to obtain weighted features. Perform 1×1 convolution and sigmoid activation on the weighted features to generate a spatial weight matrix. At the same time, perform temporal attention aggregation on the weighted features to obtain a temporal weight matrix.

[0065] Subsequently, a feature concatenation operation is performed: zero-padding is applied to the spatial weight matrix and the temporal weight matrix along the channel dimension to ensure that the spatial dimensions are the same, resulting in an aligned feature tensor; depthwise separable convolution is used to perform cross-channel feature concatenation on the aligned feature tensor to form a fused feature representation; feature selection and scaling are performed on the fused feature representation through a gating mechanism to obtain a weighted fused feature; and feature projection is performed on the weighted fused feature through a fully connected layer to form an environmental parameter prediction matrix.

[0066] S2. Perform damage quantization and probability mapping on the gallium oxide crystal damage image to obtain a baseline damage distribution map. Couple the environmental prediction matrix with the baseline damage distribution map to output the environmental-damage weight matrix.

[0067] Specifically, the operations include the following:

[0068] S2.1. Multidimensional scanning is performed on the gallium oxide crystal damage image to obtain a damage depth sequence image. Delaunay triangulation and damage quantization are performed on the damage depth sequence image to generate a damage volume and mass matrix. Specifically, a multidimensional scanning of the gallium oxide crystal damage image is performed layer by layer using a confocal microscope to obtain a tomographic image sequence. A segmentation threshold is applied to perform binarization on the tomographic image sequence. Further, grayscale value mapping is performed on the tomographic image sequence to form image pixel values. When the image pixel value is lower than the segmentation threshold, it is classified as a background region. When the image pixel value exceeds the segmentation threshold, it is classified as a damage region. The divided damage regions are connected at the boundary to generate a damage region mask. Spatial interpolation is performed on the damage region mask to output a three-dimensional damage sequence. Voxel resampling and smoothing are performed on the three-dimensional damage sequence to obtain the damage depth sequence image.

[0069] It should be noted that the segmentation threshold is defined based on the local gray-level distribution of the gallium oxide crystal damage image, and the value range is [0.35, 0.65].

[0070] Three-dimensional coordinate mapping is performed on the damage depth sequence image to obtain damage surface point cloud data; voxel downsampling is performed on the damage surface point cloud data to obtain a uniformly distributed point set, and principal component decomposition is performed on the damage surface point cloud data to obtain surface orientation information; the uniformly distributed point set is used as the x-axis in three-dimensional space and the surface orientation information is used as the y-axis in three-dimensional space to establish an initial triangular mesh; then Delaunay triangulation is performed on the initial triangular mesh.

[0071] Furthermore, edge flipping and filtering smoothing are performed on the initial triangular mesh to obtain a stable triangular mesh. At the same time, the stable triangular mesh is resampled to obtain an optimized point set. Importance weighting is performed on the optimized point set to obtain the partitioning weight. Based on the partitioning weight, dynamic partitioning is performed on the stable triangular mesh to obtain the final three-dimensional damaged surface structure.

[0072] The final 3D damaged surface structure is meshed using tetrahedral elements. Volume integration is performed on these tetrahedral elements to generate the volume of the damaged region. Density conversion is then applied to the volume of the damaged region to obtain its mass. The specific mathematical formulas are as follows.

[0073]

[0074] Where M represents the quality of the damaged region, N represents the total number of tetrahedral elements, i represents the index of the tetrahedral element, and V i Let n represent the volume of the i-th tetrahedral element. i Let ρ(n) represent the normal vector of the i-th tetrahedral element. i ) represents the density value of gallium oxide crystal at the normal vector of the i-th tetrahedral unit;

[0075] It should be noted that the normal vector of the tetrahedral element is obtained by performing a cross product and normalization operation on the tetrahedral element.

[0076] The quality of the damaged area is normalized and clustered to form a damage quality distribution; uniform sampling is performed on the damage quality distribution to obtain crystal quality parameters; based on the crystal quality parameters, the final three-dimensional damaged surface structure is quantified to obtain damage degree quantification values; the damage degree quantification values ​​are parameterized to generate damage quantification parameters; the damage quantification parameters are matrix reorganized and formatted to generate a damage volume and mass matrix.

[0077] S2.2. Probabilistic mapping is performed on the damage volume and mass matrices to obtain a baseline damage distribution map. Specifically, the damage volume and mass matrices are normalized by Min-Max scaling to eliminate dimensional differences and output a standardized matrix. Then, kernel density estimation (KDE) is used for probability mapping. Further, Gaussian convolution is performed on the standardized matrix to obtain discrete probability values. The discrete probability values ​​are smoothed by three-dimensional Gaussian filtering to eliminate sampling noise and obtain a smooth probability distribution. Region clustering is performed on the smooth probability distribution to output the damage probability region. Probabilistic isosurface mapping is performed on the damage probability region to generate an equiprobability surface. Nonlinear fitting is performed on the equiprobability surface using the radial basis function method. Further, feature point sampling is performed on the equiprobability surface to obtain a discrete isopoint parameter set. Surface fitting is performed on the discrete isopoint parameter set to generate a parameterized surface. The parameterized surface is then meshed and dynamically rendered to output the baseline damage distribution map.

[0078] S2.3. Perform spatial registration and kriging interpolation on the environmental prediction matrix to generate a continuous environmental prediction distribution. Specifically, in the spatial registration stage, feature points are extracted from the environmental prediction matrix using fast Fourier transform. Further, frequency domain filtering is applied to the environmental prediction matrix to obtain an enhanced spectrogram. Phase quantization is performed on the enhanced spectrogram to obtain displacement vectors. Spatial mapping is then performed on the displacement vectors to generate a feature point set. Nearest neighbor matching is performed on the feature point set to obtain optimized feature point pairs. Affine transformation is then applied to the optimized feature point pairs for spatial registration to obtain registered feature point pairs. The registered feature point pairs are then resampled to obtain a regularized point set. Coordinate transformation is then performed on the regularized point set to obtain geographic reference data.

[0079] In the interpolation stage, matrix decomposition is performed on the geographic reference data to form a covariance matrix. Feature extraction is performed on the covariance matrix to obtain semivariance features. The semivariance features are then parametrically transformed to obtain interpolation parameters. Based on the interpolation parameters, kriging interpolation is performed on the geographic reference data to obtain preliminary interpolation results. The preliminary interpolation results are then averaged in a neighborhood using linear weighting. Furthermore, the preliminary interpolation results are matrix filled to generate a regularized interpolation matrix. The regularized interpolation matrix is ​​then gridded to obtain a continuous prediction surface. The continuous prediction surface is then averaged in a neighborhood to output the continuous environmental prediction distribution.

[0080] S2.4. The continuous environmental prediction distribution and the baseline damage distribution map are coupled and probabilistically fused using the Hadamard product to output an environment-damage weight matrix. Specifically, in the coupling and association stage, the continuous environmental prediction distribution and the baseline damage distribution map are spatially registered and their resolution normalized to ensure complete matching in spatial dimension and resolution. Matrix transformation is performed simultaneously to output aligned environmental prediction and baseline damage distribution matrices. The environmental prediction matrix and the baseline damage distribution matrix are then multiplied point-by-point using the Hadamard product to generate an initial weight matrix, which is then normalized to obtain the normalization coefficients. The specific mathematical formula is as follows.

[0081]

[0082] Where α represents the normalization coefficient, E represents the environmental prediction matrix, D represents the baseline damage distribution matrix, and ⊙ represents the Hadamard product operator;

[0083] Based on the normalization coefficients, the continuous environment prediction distribution and the baseline damage distribution map are coupled and correlated using the Bayesian probabilistic fusion method. Furthermore, the continuous environment prediction distribution and the baseline damage distribution map are spatially convolved to generate a joint weight matrix. The joint weight matrix is ​​regularized to obtain an optimized weight matrix. The optimized weight matrix is ​​nonlinearly compressed using the Sigmoid function to output the coupled weight matrix.

[0084] In the probabilistic fusion stage, kernel density estimation is used to perform probability density transformation on the coupled weight matrix. Further, eigenvalue decomposition is performed on the coupled weight matrix to obtain probability distribution parameters, which are then normalized to obtain a standardized probability distribution. Feature extraction is performed on the standardized probability distribution to obtain probability distribution features. Spatial smoothing and feature enhancement are applied to the probability distribution features to obtain the fused probability distribution. The expectation-maximization algorithm is used to optimize the parameters of the obtained fused probability distribution. Further, iterative convergence is performed on the fused probability distribution to obtain an initial probability matrix, which is then spatially interpolated to obtain the environment-damage joint probability. The environment-damage joint probability is then matrix formatted and dynamically weighted to form the environment-damage weight matrix.

[0085] S3. Based on the environment-damage weight matrix, locate defects on the baseline damage distribution map, obtain damage-sensitive areas, and perform DBSCAN cluster analysis on the damage-sensitive areas to generate a set of high-risk area coordinates.

[0086] Specifically, the operations include the following:

[0087] S3.1. Perform weight allocation on the environment-damage weight matrix to obtain the damage sensitivity coefficient. Specifically, this includes the following operations: perform weight allocation on the environment-damage weight matrix through singular value decomposition (SVD); further, perform eigenvalue decomposition on the environment-damage weight matrix to extract eigenvalues; sort the eigenvalues ​​in descending order to capture the top k significant singular values; then perform normalized weight allocation on the top k significant singular values ​​to generate the dominant mode weight vector.

[0088] Next, the Local Sensitive Hash (LSH) algorithm is used to perform sensitivity quantification on the dominant mode weight vector. Further, a fast nearest neighbor search is performed on the dominant mode weight vector to obtain similar feature regions. Spectral clustering is then performed on these similar feature regions to generate high-sensitivity region clusters. Next, dynamic weighted fusion and numerical normalization are performed on these high-sensitivity region clusters to obtain initial sensitivity coefficients. The initial sensitivity coefficients are then discretized using tensor product mapping and nonlinearly compressed using the hyperbolic tangent function, reducing the numerical range of the initial sensitivity coefficients to the (-1, 1) interval, thus generating damage sensitivity coefficients. Damage sensitivity coefficients represent the degree of sensitivity to external environmental influences, where positive values ​​indicate that the environment exacerbates damage, and negative values ​​indicate that the environment inhibits damage.

[0089] S3.2. Based on the damage sensitivity coefficient, spatial probability sampling is performed on the baseline damage distribution map to generate a damage area point set. Specifically, this includes the following operations: segmenting the baseline damage distribution map into sub-blocks to obtain candidate sampling areas; performing spatial probability sampling on the candidate sampling areas using the Markov chain Monte Carlo method; further, performing grid division and random sampling on the weighted sampling areas to generate initial sampling points; performing spatial clustering and redundant point elimination on the initial sampling points to obtain an optimized sampling point set; using the damage sensitivity coefficient as a guiding factor, performing spatial density adjustment on the optimized sampling point set to obtain a candidate damage point set; performing Gaussian filtering and coordinate transformation on the candidate damage point set to finally output the damage area point set.

[0090] S3.3. Perform neighborhood expansion on the damaged area point set to form a damaged sensitive area; apply DBSCAN clustering to perform density clustering on the damaged sensitive area to obtain a set of high-risk area coordinates. Specifically, this includes the following operations: using the K-nearest neighbor algorithm to construct neighborhood relationships for the damaged area point set; further, defining a neighborhood radius based on the local density distribution of the damaged area point set; expanding the damaged area point set within the neighborhood radius to obtain an expanded point set; performing topological connections on the expanded point set to generate connected regions; and performing region fusion on the connected regions to form the damaged sensitive area.

[0091] The DBSCAN clustering algorithm is used to perform cluster analysis on the damage-sensitive areas. Further, density clustering and region division are performed on the damage-sensitive areas to obtain the initial density clustering results. Boundary fitting is performed on the initial density clustering results to obtain the cluster region boundaries. The cluster region boundaries are then weighted and aggregated to generate high-risk region clusters. Coordinate mapping and parameter integration are performed on the high-risk region clusters to output the high-risk region coordinate set.

[0092] S4. Divide the coordinate set of high-risk areas into risk levels and aggregate the regions to obtain a risk level distribution map. Based on the risk level distribution map, monitor the storage environment in real time and output the environment monitoring plan simultaneously.

[0093] Specifically, the operations include the following:

[0094] S4.1. A three-tiered threshold system is used to classify the coordinate set of high-risk areas into high-risk, medium-risk, and low-risk levels. Specifically, the three-tiered thresholds are defined based on the damage probability density of the high-risk area coordinate set. For example, the first-tier threshold is set to [0.75-1.0], the second-tier threshold to [0.45-0.75], and the third-tier threshold to [0.0-0.45]. Simultaneously, K-means clustering is applied to quantify the risk level of the high-risk area coordinate set. Further, spatial interpolation and matrix filling are performed on the high-risk area coordinate set to generate a risk density matrix. The risk density matrix is ​​then discretized using equal-frequency methods to obtain risk level intervals. A weighted average of these risk level intervals is then calculated to form the regional risk value. When the regional risk value falls within the first-tier threshold range, it is defined as high-risk, requiring priority handling and enhanced monitoring. When the regional risk value falls within the second-tier threshold range, it is defined as medium-risk, requiring regular inspections and dynamic tracking. When the regional risk value falls within the third-tier threshold range, it is defined as low-risk, requiring routine monitoring and preventative maintenance.

[0095] S4.2. Simultaneously perform regional aggregation on the coordinate set of high-risk areas to form continuous closed risk areas; perform GIS spatial overlay analysis on the divided risk levels and continuous closed risk areas to obtain a risk level distribution map. In specific operations, hierarchical clustering is used to perform regional aggregation on the coordinate set of high-risk areas. Further, spatial clustering is performed on the coordinate set of high-risk areas to obtain initial risk areas; grid filling and topological connection are performed on the initial risk areas to generate closed risk areas; the closed risk areas are smoothed to output continuous closed risk areas.

[0096] Subsequently, the risk levels and continuous closed risk areas are analyzed using GIS spatial overlay. Furthermore, the spatial attributes of the risk levels and continuous closed risk areas are associated using the spatial join function of GIS geographic information units to generate a risk level attribute table. Spatial coordinate binding and attribute mapping are performed on the risk level attribute table to obtain a spatial overlay dynamic view. Layer overlay and hierarchical rendering are performed on the spatial overlay dynamic view to generate a risk level distribution map.

[0097] S4.3. Based on the risk level distribution map, the storage environment is monitored in zones and levels. Real-time storage feedback data is collected and integrated to form an environmental monitoring plan. In specific operations, a mapping relationship is established between the risk level distribution map and the physical space of the warehouse. Furthermore, the risk level distribution map is spatially matched using the GIS spatial query function to obtain the monitoring areas, and the monitoring areas are classified into risk levels. For example, monitoring areas with a damage probability density ≥ 0.80 are classified as high-risk areas, monitoring areas with a damage probability density < 0.80 are classified as medium-risk areas, and monitoring areas with a damage probability density < 0.80 are classified as low-risk areas. The monitoring area of ​​0.50 is divided into low-risk zones, and equipment deployment and topology association are performed in the divided monitoring areas to generate space-device mapping relationships. Based on the space-device mapping relationships, the warehouse storage environment is monitored in different zones and levels. Furthermore, corresponding sensor node groups are bound to the monitoring areas. For example, a high-density sensor network (temperature, humidity, vibration, and gas multi-parameter sensors, sampling frequency ≥1Hz) is deployed in high-risk areas, conventional monitoring nodes (temperature, humidity, and basic gas sensors, sampling frequency 0.5Hz) are configured in medium-risk areas, and mobile inspection equipment (handheld terminals collect data periodically) is used in low-risk areas.

[0098] It should be noted that the damage probability density is obtained by performing spatial probability sampling on the monitored area.

[0099] During the monitoring process, real-time storage feedback data of the warehouse environment is collected, including parameters such as temperature, humidity, vibration frequency, and gas concentration. The real-time storage feedback data is then integrated, and further, data cleaning and standardization are performed on the real-time storage feedback data to output structured monitoring data. The structured monitoring data is then weighted and fused to output an environmental monitoring solution.

[0100] This embodiment also provides an artificial intelligence-based gallium oxide crystal storage environment monitoring system, including: an environment prediction module, used to input multi-source coupled datasets into a spatiotemporal hypergraph model, a hypergraph construction layer to couple multi-source data, and a spatiotemporal convolutional layer to perform spatial dependency capture and temporal feature extraction to form an environment parameter prediction matrix;

[0101] The coupling and correlation module is used to perform damage quantization and probability mapping on gallium oxide crystal damage images, obtain a baseline damage distribution map, couple and correlate the environmental prediction matrix with the baseline damage distribution map, and output the environment-damage weight matrix.

[0102] The localization and clustering module is used to locate defects on the baseline damage distribution map based on the environment-damage weight matrix, obtain damage-sensitive areas, and perform DBSCAN clustering analysis on the damage-sensitive areas to generate a set of high-risk area coordinates.

[0103] The real-time monitoring module is used to classify and aggregate the coordinates of high-risk areas to obtain a risk level distribution map. Based on the risk level distribution map, the storage environment is monitored in real time, and an environment monitoring plan is output synchronously.

[0104] This embodiment also provides a computer device applicable to the case of a gallium oxide crystal storage environment monitoring method based on artificial intelligence, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the gallium oxide crystal storage environment monitoring method based on artificial intelligence as proposed in the above embodiment.

[0105] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0106] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the method for monitoring the gallium oxide crystal storage environment based on artificial intelligence as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0107] In summary, this invention achieves deep fusion and high-order feature extraction of multi-source coupled data through a spatiotemporal hypergraph model, thereby significantly improving the accuracy of environmental parameter prediction and enhancing the precision of the monitoring scheme. By combining DBSCAN clustering analysis with a risk level classification strategy, high-risk areas can be efficiently identified and aggregated, thus supporting real-time monitoring and tiered response of the storage environment and effectively improving the real-time performance of the monitoring scheme.

[0108] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for monitoring the storage environment of gallium oxide crystals based on artificial intelligence, characterized in that: include, The multi-source coupled dataset is input into the spatiotemporal hypergraph model. The hypergraph construction layer couples the multi-source data, and the spatiotemporal convolutional layer performs spatial dependency capture and temporal feature extraction to form an environmental parameter prediction matrix. Damage quantization and probability mapping are performed on gallium oxide crystal damage images to obtain a baseline damage distribution map. The environmental prediction matrix is ​​coupled and correlated with the baseline damage distribution map to output an environment-damage weight matrix. Based on the environment-damage weight matrix, defects are located on the baseline damage distribution map to obtain damage-sensitive areas. DBSCAN cluster analysis is then performed on the damage-sensitive areas to generate a set of high-risk area coordinates. The risk level of the high-risk area coordinate set is classified and the area is aggregated to obtain a risk level distribution map. Based on the risk level distribution map, the storage environment is monitored in real time, and an environment monitoring plan is output synchronously.

2. The method for monitoring the gallium oxide crystal storage environment based on artificial intelligence as described in claim 1, characterized in that: The multi-source coupled dataset includes gallium oxide crystal damage images, storage environment change data, and crystal property change data.

3. The method for monitoring the gallium oxide crystal storage environment based on artificial intelligence as described in claim 2, characterized in that: The formation of the environmental parameter prediction matrix specifically includes the following steps. A hypergraph construction layer and a spatiotemporal convolutional layer are constructed, and skip connections are used to perform high-order information aggregation and residual stacking on the hypergraph construction layer and the spatiotemporal convolutional layer to construct a spatiotemporal hypergraph model. The multi-source coupled dataset is input into the spatiotemporal hypergraph model. The hypergraph construction layer couples the multi-source data through an attention mechanism to obtain the hypergraph associated feature tensor. Spatiotemporal convolutional layers apply gated convolution and hypergraph convolution to perform spatial dependency capture and temporal feature extraction, generating spatiotemporal fusion features; The hypergraph association feature tensor and spatiotemporal fusion features are concatenated to form an environmental parameter prediction matrix.

4. The method for monitoring the gallium oxide crystal storage environment based on artificial intelligence as described in claim 1, characterized in that: The acquisition of the baseline damage distribution map specifically includes the following steps. Multidimensional scanning was performed on gallium oxide crystal damage images to obtain damage depth sequence images; Delaunay triangulation and damage quantization are performed on the damage depth sequence image to generate damage volume and mass matrices. A baseline damage distribution map is obtained by performing a probability mapping between the damage volume and mass matrix.

5. The method for monitoring the gallium oxide crystal storage environment based on artificial intelligence as described in claim 4, characterized in that: The output environment-damage weight matrix specifically includes the following steps. Spatial registration and kriging interpolation are performed on the environmental prediction matrix to generate a continuous environmental prediction distribution. The continuous environment prediction distribution and the baseline damage distribution map are coupled and probabilistically fused by the Hadamard product to output the environment-damage weight matrix.

6. The method for monitoring the gallium oxide crystal storage environment based on artificial intelligence as described in claim 1, characterized in that: The process of generating the set of coordinates for high-risk areas specifically includes the following steps. Weights are assigned to the environment-damage weight matrix to obtain the damage sensitivity coefficient; Based on the damage sensitivity coefficient, spatial probability sampling is performed on the baseline damage distribution map to generate a set of damage area points; Perform neighborhood expansion on the point set of the damaged region to form the damaged sensitive region; DBSCAN clustering is used to perform density clustering on damage-sensitive areas to obtain a set of coordinates for high-risk areas.

7. The method for monitoring the gallium oxide crystal storage environment based on artificial intelligence as described in claim 6, characterized in that: The aforementioned environmental monitoring scheme specifically includes the following steps. Using a three-level threshold, the coordinate set of high-risk areas is divided into high-risk, medium-risk, and low-risk risk levels. Simultaneously, regional aggregation is performed on the coordinate set of high-risk areas to form continuous closed risk areas. Perform GIS spatial overlay analysis on the risk levels and continuous closed risk areas to obtain a risk level distribution map; Based on the risk level distribution map, the storage environment is monitored in different zones and levels. Real-time storage feedback data is collected and integrated to form an environmental monitoring plan.

8. An artificial intelligence-based gallium oxide crystal storage environment monitoring system, based on the artificial intelligence-based gallium oxide crystal storage environment monitoring method according to any one of claims 1 to 7, characterized in that: include, The environment prediction module is used to input multi-source coupled datasets into the spatiotemporal hypergraph model. The hypergraph construction layer couples multi-source data, and the spatiotemporal convolutional layer performs spatial dependency capture and temporal feature extraction to form an environment parameter prediction matrix. The coupling and correlation module is used to perform damage quantization and probability mapping on gallium oxide crystal damage images, obtain a baseline damage distribution map, couple and correlate the environmental prediction matrix with the baseline damage distribution map, and output the environment-damage weight matrix. The localization and clustering module is used to locate defects on the baseline damage distribution map based on the environment-damage weight matrix, obtain damage-sensitive areas, and perform DBSCAN clustering analysis on the damage-sensitive areas to generate a set of high-risk area coordinates. The real-time monitoring module is used to classify and aggregate the coordinates of high-risk areas to obtain a risk level distribution map. Based on the risk level distribution map, the storage environment is monitored in real time, and an environment monitoring plan is output synchronously.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the artificial intelligence-based gallium oxide crystal storage environment monitoring method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the artificial intelligence-based gallium oxide crystal storage environment monitoring method according to any one of claims 1 to 7.

Images