Marine digital twinning optimization method and system based on multi-scale feature fusion

Abstract

The invention relates to the technical field of ocean data monitoring, in particular to an ocean digital twinning optimization method and system based on multi-scale feature fusion. The method comprises the following steps: acquiring marine environment multi-source data; performing data preprocessing on the acquired marine environment multi-source data; performing feature extraction and feature fusion on the multi-source data based on a digital twin model; optimizing the digital twin model based on the fused features; and performing marine environment multi-source data prediction by using the optimized digital twinborn model. According to the method, fusion modeling is carried out on multi-modal and multi-scale ocean data, a digital twinning closed-loop system covering the whole life cycle is constructed, the defects existing in the prior art are overcome, the modeling precision, real-time performance and adaptability of ocean digital twinning are improved, and innovative technical support is provided for ocean equipment and environment management.

Description

Technical Field

[0001] The present invention relates to the technical field of ocean data monitoring, and in particular to an ocean digital twin optimization method and system based on multi-scale feature fusion. Background Art

[0002] The marine environment is complex and changeable. Traditional testing and development methods have problems of low efficiency, high cost and high risk. Especially in the research and development and testing of marine equipment, it has long relied on traditional physical model tests, which are difficult to dynamically adapt to real sea conditions and environmental changes, resulting in long development cycles and low iteration efficiency. In recent years, the rise of digital twin technology has provided a new idea for solving this problem. Digital twins support the optimization management of the entire life cycle of equipment from design, manufacturing to use and maintenance by mapping the behavior and state of physical objects with virtual environments in real time. Especially in marine scenarios, digital twins can achieve efficient modeling and accurate simulation of complex marine environments and equipment operating states by integrating multi-source heterogeneous data and multi-scale information. The existing methods can solve the above problems to a certain extent, but still show the following deficiencies: 1. Technical bottlenecks in multi-scale feature fusion. The data dimensions in the marine environment are complex, and the features span multiple scales such as time domain and space domain. Traditional methods are difficult to effectively fuse these multi-scale features. 2. Insufficient real-time and robustness. The existing models have low real-time performance when processing large-scale marine data, poor robustness to noise and abnormal environments, and cannot adapt to dynamic changes in practical applications. 3. Insufficient integration throughout the entire life cycle. There have been many studies on the application of digital twin technology in the design, manufacturing, and operation stages, but there is a lack of full-process integrated solutions from data collection to feedback optimization.

[0003] At present, there are still the following deficiencies in the optimization of marine digital twin data: 1. Insufficient multi-scale feature modeling: Existing methods are insufficient in expressing and integrating multi-scale features in the time domain, spatial domain and semantic domain of the marine environment, and cannot fully and accurately characterize the complex marine dynamic environment. 2. Poor real-time and robustness: When faced with complex changes in the marine environment, traditional models often find it difficult to ensure efficient processing while having sufficient robustness, and are easily affected by environmental noise and data loss. 3. Limited full life cycle integration: Current technologies are mostly concentrated in a specific link, lacking a full life cycle integration framework covering data collection to feedback optimization, and are difficult to provide closed-loop support for the design, manufacturing, operation and maintenance of marine equipment. 4. Weak system adaptability: The optimization methods for different task requirements lack the ability to dynamically adjust, resulting in poor versatility and adaptability of the system, limiting the breadth and flexibility of practical applications. Summary of the invention

[0004] In order to solve the above-mentioned problems, the present invention provides an ocean digital twin optimization method and system based on multi-scale feature fusion.

[0005] In the first aspect, the present invention provides an ocean digital twin optimization method based on multi-scale feature fusion, which adopts the following technical solution: A marine digital twin optimization method based on multi-scale feature fusion, comprising: Obtain multi-source data on the marine environment; Carry out data preprocessing on the acquired multi-source data of marine environment; Extract and fuse features from multi-source data based on the digital twin model; The fused features are transformed into a unified feature representation through dimension alignment; Optimize the digital twin model based on the fused features; Use the optimized digital twin model to predict multi-source data of the marine environment.

[0006] Furthermore, the acquisition of multi-source data of the marine environment includes multi-modal and multi-scale data collection by using a multi-sensor network and a collaborative operation of fixed sensors, unmanned aerial vehicles, and satellite multi-platform systems. The time series data generated by the sensors is expressed as: in, For the The output of a sensor at time t is m is the feature dimension; Using drones and satellites to collect multi-channel images from multiple angles, the image data is represented as: Among them, H is the image height, W is the image width, and C is the number of channels.

[0007] Furthermore, the acquired marine environment multi-source data is preprocessed, including adopting difference optimization strategies for time series data and image data respectively to eliminate the differences between the data, wherein, for time series data, linear interpolation is applied to fill in missing values ​​to ensure the continuity and integrity of the data, and then the eigenvalues ​​are adjusted to a uniform range through normalization processing; for the preprocessing of image data, Gaussian filtering is first used to remove noise to improve the clarity and stability of the image.

[0008] Furthermore, the digital twin model is used to extract and fuse features from multiple source data, including extracting dynamic features of time series data through an improved spatiotemporal convolutional network ST-CNN to capture its changing trend in the time dimension, and extracting multi-scale features of image data using a median enhanced spatial channel attention network MECS to enhance its ability to express spatial distribution, wherein the enhanced features are decomposed to extract features of different resolutions, and grouped according to channels to obtain features of different channels, and finally, the time series features and image features of different channels are weightedly fused through a dual-time fusion network BFM to integrate multimodal data into a unified feature representation for modeling.

[0009] Furthermore, the fused features are converted into a unified feature representation through dimensional alignment, including mapping the two modalities to a unified feature space through linear projection and broadcasting mechanisms, wherein the temporal features are first projected and expanded, and then the image features are projected, and attention distribution is generated after feature splicing, the features of the two modalities are weightedly fused according to the attention weights, and the fusion results are subjected to multi-scale convolution optimization to extract local and global feature information.

[0010] Furthermore, the digital twin model is optimized based on the fused features, including training the model according to the fused features using a loss function consisting of a prediction error loss and a regularization term, wherein the prediction error part uses the mean square error (MSE) to evaluate the accuracy of the prediction result, and in order to prevent the model from losing generalization performance due to overfitting during the training process, a regularization term is introduced into the loss function to constrain the complexity of the model parameters; at the same time, the Adam optimization algorithm is used to dynamically adjust the learning rate and combine the accumulated information of the first-order and second-order momentum to optimize the model parameters in the non-convex optimization problem, and gradient clipping is introduced on the basis of Adam to prevent the gradient explosion from interfering with the training process.

[0011] Furthermore, the use of the optimized digital twin model to predict multi-source data of the marine environment includes deploying the optimized digital twin model to an actual scene, performing real-time monitoring and simulation, and calculating the error by comparing the target variable predicted by the model with the observed value, wherein for time series data, based on the mean square error MSE between the predicted value and the observed value, the threshold is set to 0.05, that is, the square average of the error between the predicted value and the true value is less than 5%; for image feature prediction, based on the mean absolute difference MAE of pixel point errors, the threshold is set to 3%, that is, the proportion of pixel value error to the image gray value range is less than 3%, and the calculated error is expressed as: Among them, the target variable , observed value .

[0012] The second aspect is an ocean digital twin optimization system based on multi-scale feature fusion, including: The data acquisition module is configured to acquire multi-source data of the marine environment; The preprocessing module is configured to perform data preprocessing on the acquired marine environment multi-source data; The feature module is configured to perform feature extraction and feature fusion on multi-source data based on the digital twin model; The conversion module is configured to convert the fused features into a unified feature representation through dimension alignment; The optimization module is configured to optimize the digital twin model based on the fused features; The prediction module is configured to use the optimized digital twin model to predict multi-source data of the marine environment.

[0013] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for example, a method for optimizing an ocean digital twin based on multi-scale feature fusion.

[0014] In a fourth aspect, the present invention provides a terminal device comprising a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store multiple instructions, the instructions being suitable for being loaded by the processor and executing the ocean digital twin optimization method based on multi-scale feature fusion.

[0015] In summary, the present invention has the following beneficial technical effects: This invention builds a digital twin closed-loop system covering the entire life cycle by fusing and modeling multi-modal and multi-scale ocean data, thereby addressing the shortcomings of current technologies, improving the modeling accuracy, real-time performance, and adaptability of ocean digital twins, and providing innovative technical support for marine equipment and environmental management. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Schematic diagram of the method of Example 1 of the present invention. DETAILED DESCRIPTION

[0017] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0018] Example 1 Reference Figure 1 , an ocean digital twin optimization method based on multi-scale feature fusion in this embodiment includes: Obtain multi-source data on the marine environment; Carry out data preprocessing on the acquired multi-source data of marine environment; Extract and fuse features from multi-source data based on the digital twin model; The fused features are transformed into a unified feature representation through dimension alignment; Optimize the digital twin model based on the fused features; Use the optimized digital twin model to predict multi-source data of the marine environment.

[0019] Specifically, the following steps are included: S1: Collect and perceive multi-source data, In complex marine environments, sensors, drones and satellite remote sensing equipment are used to obtain multi-modal and multi-scale data. These data include temperature, velocity, salinity physical time series data and high-definition local images and global remote sensing image data, showing the dynamic marine environment and providing rich input data for subsequent steps.

[0020] S2: Data preprocessing and standardization, The collected data is preprocessed to ensure the consistency and quality of multimodal data. Time series data is processed by missing value completion and normalization to eliminate noise and outliers; image data is improved in quality and diversity through denoising, standardization and data enhancement techniques. This step lays a solid foundation for subsequent feature modeling.

[0021] S3: Construct a digital twin model. The digital twin model is constructed using the spatiotemporal convolutional network as the backbone network to simulate the dynamic changes of the marine environment and the operating status of the equipment. The spatiotemporal convolutional network is used to extract dynamic features in the time series data, and the multi-scale spatial features of the image data are combined. The fusion technology is used to transform the multimodal data into a unified feature representation. The model provides core data support for training and optimization.

[0022] S4: Training and optimizing digital twin models, Based on the fused feature data, the digital twin model is trained and optimized. The error between the predicted value and the true value is measured by defining a loss function, and the model parameters are continuously adjusted in combination with the optimization algorithm to improve the prediction accuracy and robustness of the model. The trained model can efficiently adapt to complex dynamic environments.

[0023] S5: Twin system application and feedback optimization, The optimized digital twin model is deployed to the actual system for real-time monitoring and dynamic simulation. By monitoring the error between the predicted results output by the system and the actual observed values, the feedback mechanism is activated, new data is collected and returned to S1, the model is updated, and closed-loop optimization of data and models is achieved. This step ensures the high adaptability and real-time performance of the system to dynamic environments.

[0024] Specifically: S1's time series data collection and image data collection use multi-modal and multi-scale data collection in a collaborative way using a multi-sensor network and a multi-platform system of fixed sensors, drones, and satellites. Fixed sensors are used to provide continuous and stable time series data to ensure that the time series changes of dynamic environments can be accurately captured; at the same time, drones are used to obtain high-resolution images from a local perspective to supplement the detailed description of key areas; satellites provide global background data through large-scale remote sensing to make up for the limitations of a single sensor. The multi-platform design ensures data diversity and spatial coverage. The time series data generated by the sensor can be expressed as: in, For the The output of a sensor at time t is m is the feature dimension.

[0025] Using drones and satellites to collect multi-channel images from multiple angles, the image data is represented as: Among them, H is the image height, W is the image width, and C is the number of channels.

[0026] The S2 preprocesses the collected data and adopts different optimization strategies for time series data and image data to eliminate the differences between the data and improve the data quality.

[0027] For time series data, linear interpolation is first used to fill in missing values ​​to ensure the continuity and integrity of the data: in, It is Features in time The interpolation result on . and It is a known time point and On feature values. and are two known time points, and t1 <t<t2。

[0028] Then, the eigenvalues ​​are adjusted to a uniform range through normalization to facilitate subsequent model processing: in, and are the mean and standard deviation of the features respectively.

[0029] For the preprocessing of image data, Gaussian filtering is first used to remove noise and improve the clarity and stability of the image. The specific formula of Gaussian filtering is: in, is the weight of the Gaussian filter, , is the horizontal and vertical coordinate offset of the pixel, is the standard deviation of the Gaussian distribution and determines the smoothness of the filter.

[0030] Next, to further enhance the image features, the spatial module of the median enhanced channel attention module (MECS) is used. First, the global features of each channel are enhanced through channel attention. The calculation formula is as follows: in, is a 1×1 convolution operation, and Represent the global average pooling and maximum pooling operations respectively, is the Sigmoid activation function. This formula implements dynamic weight allocation for channel importance.

[0031] Then, the spatial information expression ability of the image is enhanced through the spatial attention module, and its specific formula is: in, It represents the channel dimension concatenation of the results of the global average and maximum pooling operations of the input feature map F. It is a 7×7 convolution operation used to capture local spatial context information.

[0032] The enhanced image features are expressed as: in, represents a pixel-by-pixel weighted operation, is the enhanced image feature.

[0033] Preprocessed time series data and augmented image data as input for the next step.

[0034] The multi-scale feature modeling and fusion of S3 first extracts the dynamic features of time series data through an improved spatiotemporal convolutional network (ST-CNN) to capture its changing trend in the time dimension: in, The convolution kernels are of different scales. The present invention uses convolution kernels of 3×C, 5×C, and 7×C. C is the dimension of the feature. Each convolution kernel can capture the dynamic features of a specific time scale. It is a dynamic feature extracted by time gating mechanism or recurrent convolutional layer, which can guide the network to focus on the long-term dependency features of time series.

[0035] After the enhancement, the image data Perform multi-scale decomposition to extract features of different spatial resolutions. Group by channel: Each group feature is then downsampled: Then perform upsampling and feature recovery to restore features of different resolutions to the original resolution: in, represents deep convolution, which is used to extract local features. Represents upsampling, which is used to restore the feature size.

[0036] Then the features of different channels are spliced ​​and fused, and then Convolution reduces the number of channels and integrates features of different scales: is the fused image feature.

[0037] Then The MECS module is used to further optimize the multi-scale fused features: first, channel attention optimization is performed, and the global information of the feature channel is extracted through global average pooling, maximum pooling, and median pooling: in, Represents the channel attention map. The features after channel optimization generated by global pooling and MLP are: in, Represents element-wise multiplication for attention weighting.

[0038] Then perform spatial attention optimization and use MECS's multi-scale deep convolution module to capture spatial dependencies: Then after Convolution generates spatial attention map: Output Features As the final result of multi-scale feature extraction, it has stronger robustness and expression ability: in, Represents element-by-element multiplication for attention weighting. Finally, the optimized image features are fused.

[0039] S3 uses fusion technology to transform multimodal data into a unified feature representation. Since time series features and image features are represented in different forms (time series features are in vector form, and image features are in tensor form), dimension alignment is required. The two modalities are mapped to a unified feature space through linear projection and broadcasting mechanisms.

[0040] First, the time series features are projected and expanded: in, Represents the time series characteristics, represents the projection matrix, represents the bias term, Time series features mapped to a unified dimension.

[0041] Then Expanded to the same spatial dimension as the image features: in, Indicates a broadcast operation, which converts the timing characteristics Broadcast to the same spatial dimensions as the image features Consistent, expanded to a three-dimensional tensor.

[0042] Then perform image feature projection: in, represents the projection matrix, Indicates bias, Represents image features mapped to a uniform dimension.

[0043] In order to dynamically balance the contribution of the two modal features, an attention weight mechanism is introduced to calculate the weight distribution of the two modal features.

[0044] Feature fusion weight calculation: Generate attention distribution by concatenating features. in, represents the fusion weight matrix, represents the bias term, Represents the attention weight of temporal features.

[0045] The weights of image features are: , The features of the two modalities are weighted and fused according to the attention weights: in, represents the fused features, Represents an element-wise multiplication operation.

[0046] In order to further enhance the expressive power of fusion features, the fusion results Perform multi-scale convolution optimization to extract local and global feature information.

[0047] Multi-scale convolution operation: Among them, k=11 is used to extract global features, and k=3 is used to extract local features.

[0048] Fusion of global and local features: in, Represents a splicing operation, represents the linear transformation matrix, represents the bias term, Represents the final fusion feature.

[0049] Through the above operations, we can obtain the core data information of training and optimization. Then use the training loss function and optimization algorithm for optimization.

[0050] In the model training and optimization of S4, in order to effectively improve the prediction accuracy and generalization ability of the digital twin model, a multi-level optimization strategy based on the combination of loss function design and optimization algorithm was adopted.

[0051] First, in order to measure the deviation between the model prediction value and the true value, the present invention designs a comprehensive loss function. The loss function consists of prediction error loss and regularization term. The prediction error part uses mean square error (MSE) to evaluate the accuracy of the prediction result, and its formula is: in, is the predicted value, is the true value, is the sample size.

[0052] By calculating the minimum value of MSE, the prediction ability of the model can be continuously optimized. At the same time, in order to prevent the model from losing generalization performance due to overfitting during training, a regularization term is introduced into the loss function to constrain the complexity of the model parameters. Specifically, L2 regularization is used, and its formula is: in, are the parameters of the model, is the regularization weight, which is used to control the influence of the regularization term on the loss.

[0053] The final loss function combines these two parts to ensure that the model achieves the best result between accuracy and robustness by balancing prediction error and model complexity.

[0054] When optimizing model parameters, the Adam optimization algorithm is used. Adam can show excellent performance in non-convex optimization problems by dynamically adjusting the learning rate and combining the accumulated information of first-order and second-order momentum. Its parameter update formula is: in, For the t The parameters of the step, is the learning rate, and are the first-order momentum and second-order momentum of the gradient, A small constant to prevent the denominator from being zero.

[0055] In order to further enhance the stability of training, the gradient clipping technology is introduced on the basis of Adam to prevent the gradient explosion from interfering with the training process. Gradient clipping is achieved by limiting the maximum norm of the gradient, and its formula is: in, is the preset gradient clipping threshold, is the gradient norm. Through clipping, the model can maintain stable parameter updates in extreme cases.

[0056] During the training process, in order to ensure that the model can efficiently converge to the optimal solution, the present invention adopts a dynamic learning rate adjustment strategy. The learning rate decay method is used to gradually reduce the learning rate as the training progresses. The formula is: in, is the initial learning rate, is the attenuation factor, tis the current number of training steps. As training progresses, reducing the learning rate can effectively prevent the model from oscillating when approaching the optimal solution, while improving the accuracy of convergence.

[0057] Through the above design, the optimized model not only has high prediction accuracy, but also can maintain robustness in complex dynamic environments. After training is completed, the model will enter the actual deployment stage to provide core support for real-time monitoring and feedback optimization.

[0058] The real-time monitoring and dynamic feedback optimization of S5 deploys the optimized digital twin model to the actual scene for real-time monitoring and simulation. With observed value Comparison, calculation error: Feedback mechanism and closed-loop optimization: For time series data, based on the mean square error (MSE) between the predicted value and the observed value, the threshold is set to 0.05 (that is, the average square of the error between the predicted value and the true value is less than 5%). For image feature prediction, based on the mean absolute difference (MAE) of pixel point errors, the threshold is set to 3% (that is, the proportion of pixel value error to the image grayscale value range is less than 3%). During operation, the system will dynamically adjust the threshold according to the quality of real-time collected data and environmental changes. For example, when the data noise level is high, the threshold is appropriately relaxed; in critical mission scenarios (such as extreme weather warnings), the threshold is tightened to improve the accuracy of the prediction. When the error When the threshold is exceeded, the system activates the feedback mechanism and returns the newly collected data to S1 for retraining, thereby continuously optimizing the model performance and ensuring its real-time responsiveness and adaptability to environmental changes.

[0059] Example 2 This embodiment provides an ocean digital twin optimization system based on multi-scale feature fusion, including: The data acquisition module is configured to acquire multi-source data of the marine environment; The preprocessing module is configured to perform data preprocessing on the acquired marine environment multi-source data; The feature module is configured to perform feature extraction and feature fusion on multi-source data based on the digital twin model; The conversion module is configured to convert the fused features into a unified feature representation through dimension alignment; The optimization module is configured to optimize the digital twin model based on the fused features; The prediction module is configured to use the optimized digital twin model to predict multi-source data of the marine environment.

[0060] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for example, an ocean digital twin optimization method based on multi-scale feature fusion.

[0061] A terminal device includes a processor and a computer-readable storage medium, the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded and executed by the processor. A method for optimizing ocean digital twins based on multi-scale feature fusion.

[0062] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A marine digital twin optimization method based on multi-scale feature fusion, characterized in that: include: Obtain multi-source data on the marine environment; Carry out data preprocessing on the acquired multi-source data of marine environment; Extract and fuse features from multi-source data based on the digital twin model; The fused features are transformed into a unified feature representation through dimension alignment; Optimize the digital twin model based on the fused features; Use the optimized digital twin model to predict multi-source data of the marine environment.

2. The ocean digital twin optimization method based on multi-scale feature fusion according to claim 1 is characterized in that: The acquisition of multi-source data of the marine environment includes multi-modal and multi-scale data collection by using a multi-sensor network and a collaborative operation of fixed sensors, unmanned aerial vehicles, and satellite multi-platform systems. The time series data generated by the sensors is expressed as: in, For the The output of a sensor at time t, m is the feature dimension; Using drones and satellites to collect multi-channel images from multiple angles, the image data is represented as: Among them, H is the image height, W is the image width, and C is the number of channels.

3. The ocean digital twin optimization method based on multi-scale feature fusion according to claim 2 is characterized in that: The data preprocessing of the acquired marine environment multi-source data includes adopting difference optimization strategies for time series data and image data respectively to eliminate the differences between the data. For time series data, linear interpolation is used to fill in missing values ​​to ensure the continuity and integrity of the data, and then the eigenvalues ​​are adjusted to a uniform range through normalization. For the preprocessing of image data, Gaussian filtering is first used to remove noise to improve the clarity and stability of the image.

4. The ocean digital twin optimization method based on multi-scale feature fusion according to claim 3 is characterized in that: The digital twin model-based feature extraction and feature fusion of multi-source data includes extracting dynamic features of time series data through an improved spatiotemporal convolutional network ST-CNN to capture its changing trend in the time dimension, and extracting multi-scale features of image data using a median enhanced spatial channel attention network MECS to enhance its ability to express spatial distribution, wherein the enhanced features are decomposed to extract features of different resolutions, and grouped according to channels to obtain features of different channels, and finally, the time series features and image features of different channels are weightedly fused through a dual-time fusion network BFM to integrate multimodal data into a unified feature representation for modeling.

5. The ocean digital twin optimization method based on multi-scale feature fusion according to claim 4 is characterized in that: The fused features are converted into a unified feature representation through dimensional alignment, including mapping the two modalities to a unified feature space through linear projection and broadcasting mechanisms, wherein the temporal features are first projected and expanded, and then the image features are projected, and attention distribution is generated after feature splicing. The features of the two modalities are weightedly fused according to the attention weights, and the fusion results are subjected to multi-scale convolution optimization to extract local and global feature information.

6. The ocean digital twin optimization method based on multi-scale feature fusion according to claim 5 is characterized in that: The digital twin model is optimized based on the fused features, including training the model according to the fused features using a loss function consisting of a prediction error loss and a regularization term, wherein the prediction error part uses the mean square error MSE to evaluate the accuracy of the prediction result, and in order to prevent the model from losing generalization performance due to overfitting during the training process, a regularization term is introduced into the loss function to constrain the complexity of the model parameters; at the same time, the Adam optimization algorithm is used to dynamically adjust the learning rate and combine the accumulated information of the first-order and second-order momentum to optimize the model parameters in the non-convex optimization problem, and gradient clipping is introduced on the basis of Adam to prevent the gradient explosion from interfering with the training process.

7. The ocean digital twin optimization method based on multi-scale feature fusion according to claim 6 is characterized in that: The method of using the optimized digital twin model to predict multi-source data of the marine environment includes deploying the optimized digital twin model to an actual scene, performing real-time monitoring and simulation, and calculating the error by comparing the target variable predicted by the model with the observed value, wherein for time series data, based on the mean square error MSE between the predicted value and the observed value, the threshold is set to 0.05, that is, the square average of the error between the predicted value and the true value is less than 5%; for image feature prediction, based on the mean absolute difference MAE of pixel point errors, the threshold is set to 3%, that is, the proportion of pixel value error to the image gray value range is less than 3%, and the calculated error is expressed as: Among them, the target variable , observed value .

8. An ocean digital twin optimization system based on multi-scale feature fusion, characterized in that: include: The data acquisition module is configured to acquire multi-source data of the marine environment; The preprocessing module is configured to perform data preprocessing on the acquired marine environment multi-source data; The feature module is configured to perform feature extraction and feature fusion on multi-source data based on the digital twin model; The conversion module is configured to convert the fused features into a unified feature representation through dimension alignment; The optimization module is configured to optimize the digital twin model based on the fused features; The prediction module is configured to use the optimized digital twin model to predict multi-source data of the marine environment.

9. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the method according to claim 1 .

10. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is used to implement each instruction; and the computer-readable storage medium is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the method as claimed in claim 1 .

Images