Dynamic time-varying scene-oriented digital twin data transmission method and system

Abstract

The invention discloses a dynamic time-varying scene-oriented digital twin data transmission method and system. The method comprises the following steps of: acquiring multi-modal data through a sensing node; processing the multi-modal data through a deep learning model deployed by edge calculation to obtain input data; performing hierarchical compression on the input data; transmitting the input data to cloud equipment through progressive transmission; performing data reconstruction on the cloud equipment by using hierarchical decompression; and the cloud equipment updates the digital twin model through the reconstructed data. According to the method, efficient hierarchical processing of data is realized through a collaborative compression mechanism of deep learning and a traditional algorithm. Channel fluctuation is adapted through progressive transmission and scheduling optimization, and the transmission efficiency is improved. And the reconstruction quality and real-time performance are ensured through cloud hierarchical decompression and feedback control.

Description

Technical Field

[0001] This invention relates to the field of data transmission technology, and in particular to a digital twin data transmission method and system for dynamic, time-varying scenarios. Background Technology

[0002] In the field of data transmission in digital twins and dynamic time-varying scenarios, image compression and transmission in large-scale time-varying environments face multiple technical bottlenecks. While traditional compression methods can achieve efficient compression of static data, they exhibit significant limitations in dynamic time-varying scenarios: First, under complex channel conditions such as low signal-to-noise ratio and bandwidth fluctuations, fixed compression ratio strategies cannot guarantee reliable transmission of basic content; second, existing progressive transmission technologies mostly rely on single-modality or fixed layering, failing to adapt to the heterogeneous characteristics and real-time requirements of multimodal data; third, while deep learning compression models can improve compression efficiency, they lack a collaborative mechanism with traditional algorithms, resulting in insufficient optimization of residual redundancy and impacting transmission efficiency. Furthermore, existing systems lack integrated design in dynamic layered compression, adaptive channel scheduling, and cloud feedback control, making it difficult to meet the stringent requirements of real-time performance, flexibility, and reconstruction quality in digital twin scenarios. Summary of the Invention

[0003] This invention provides a digital twin data transmission method and system for dynamic time-varying scenarios, which addresses the problems in existing technologies where fixed compression ratio strategies cannot guarantee reliable transmission of basic content under complex channel conditions such as low signal-to-noise ratio and bandwidth fluctuations. Existing progressive transmission technologies often rely on a single mode or fixed layering, which cannot adapt to the heterogeneous characteristics and real-time requirements of multimodal data. Furthermore, deep learning compression models lack a collaborative mechanism with traditional algorithms, resulting in insufficient optimization of residual redundancy and affecting transmission efficiency.

[0004] On one hand, embodiments of the present invention provide a digital twin data transmission method for dynamic, time-varying scenarios, including:

[0005] Multimodal data is collected through sensing nodes;

[0006] The input data is obtained by cleaning, feature optimization, and standardization of the multimodal data through a deep learning model deployed by edge computing;

[0007] The input data is compressed in layers;

[0008] The input data, after being compressed in layers, is transmitted to the cloud device via progressive transmission;

[0009] The cloud device uses layered decompression to reconstruct the input data to obtain reconstructed data.

[0010] The cloud device updates the digital twin model using the reconstructed data.

[0011] In one possible implementation, the acquisition of multimodal data through sensing nodes includes:

[0012] The sensing node acquires the multimodal data through a multi-source heterogeneous network;

[0013] The multi-source heterogeneous network includes industrial cameras, infrared devices, text recognition engines, and microphones;

[0014] The multimodal data includes image, text, and audio data.

[0015] In one possible implementation, the hierarchical compression of the input data includes:

[0016] The input data is divided into a base layer, an enhancement layer, and a transport adaptation layer, and compressed separately.

[0017] The layering of the input data is performed using a deep learning model, a hyper-prior encoder, and a vector quantization generative adversarial network encoder.

[0018] In one possible implementation, the base layer generates minimum quality, fully identifiable data;

[0019] The enhancement layer provides multi-level quality enhancement data on the base layer;

[0020] The enhancement layer is used to improve the high-frequency details of the base layer;

[0021] The transport adaptation layer encapsulates the compressed, layered input data into a progressive transport format.

[0022] In one possible implementation, transmitting the layered compressed input data to the cloud device via progressive transfer further includes:

[0023] The base layer is preferentially used for data transmission at low signal-to-noise ratios.

[0024] The enhancement layer is used to supplement transmission when the channel is improved.

[0025] In one possible implementation, the step of using hierarchical decompression on the cloud device to reconstruct the input data to obtain reconstructed data includes:

[0026] After receiving the progressively transmitted multi-layered input data, the cloud device performs layered decompression according to the transmission order;

[0027] The input data from multiple layers is dequantized and upsampled using a deep learning model with a super-prior decoder and a vector quantization generative adversarial network decoder to generate reconstructed data.

[0028] In one possible implementation, the cloud device updating the digital twin model using the reconstructed data further includes:

[0029] The cloud device updates the digital twin model based on the quality of the reconstructed data;

[0030] The quality of the reconstructed data is evaluated using an image similarity evaluation index.

[0031] On the other hand, embodiments of the present invention provide a digital twin data transmission system for dynamic, time-varying scenarios, including:

[0032] The sensing node module is used to collect multimodal data through sensing nodes;

[0033] The data processing pre-module is used to clean, optimize features, and standardize the multimodal data using a deep learning model deployed through edge computing to obtain input data.

[0034] The data compression module is used to perform layered compression on the input data;

[0035] The data transmission module is used to transmit the layered compressed input data to the cloud device via progressive transmission;

[0036] A cloud module is used to reconstruct the input data using layered decompression on the cloud device to obtain reconstructed data; the cloud device updates the digital twin model using the reconstructed data.

[0037] The digital twin data transmission method and system for dynamic, time-varying scenarios disclosed in this invention have the following advantages:

[0038] (1) Through the collaborative compression mechanism of deep learning and traditional algorithms, efficient hierarchical processing of data is achieved.

[0039] (2) Adapt to channel fluctuations and improve transmission efficiency through progressive transmission and scheduling optimization.

[0040] (3) Ensure reconstruction quality and real-time performance through cloud-based layered decompression and feedback control. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. 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.

[0042] Figure 1A flowchart illustrating a digital twin data transmission method for dynamic, time-varying scenarios provided in this application embodiment;

[0043] Figure 2 A schematic diagram of a digital twin data transmission system for dynamic, time-varying scenarios provided in this application embodiment;

[0044] Figure 3 This is a schematic diagram illustrating the operation of a digital twin data transmission method for dynamic, time-varying scenarios provided in this application. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Figure 1 This is a flowchart illustrating a digital twin data transmission method and system for dynamically time-varying scenarios provided by an embodiment of the present invention. The embodiment of the present invention provides a digital twin data transmission method and system for dynamically time-varying scenarios, including:

[0047] Multimodal data is collected through sensing nodes;

[0048] The input data is obtained by cleaning, feature optimization, and standardization of the multimodal data through a deep learning model deployed by edge computing;

[0049] The input data is compressed in layers;

[0050] The input data, after being compressed in layers, is transmitted to the cloud device via progressive transmission;

[0051] The cloud device uses layered decompression to reconstruct the input data to obtain reconstructed data.

[0052] The cloud device updates the digital twin model using the reconstructed data.

[0053] The multimodal data acquired through the sensing nodes includes:

[0054] The sensing node acquires the multimodal data through a multi-source heterogeneous network;

[0055] The multi-source heterogeneous network includes industrial cameras, infrared devices, text recognition engines, and microphones;

[0056] The multimodal data includes image, text, and audio data.

[0057] The layered compression of the input data includes:

[0058] The input data is divided into a base layer, an enhancement layer, and a transport adaptation layer, and compressed separately.

[0059] The layering of the input data is performed using a deep learning model, a hyper-prior encoder, and a vector quantization generative adversarial network encoder.

[0060] The base layer generates the lowest quality, fully identifiable data.

[0061] The enhancement layer provides multi-level quality enhancement data on the base layer;

[0062] The enhancement layer is used to improve the high-frequency details of the base layer;

[0063] The transport adaptation layer encapsulates the compressed, layered input data into a progressive transport format.

[0064] The process of transmitting the layered compressed input data to the cloud device via progressive transmission also includes:

[0065] The base layer is preferentially used for data transmission at low signal-to-noise ratios.

[0066] The enhancement layer is used to supplement transmission when the channel is improved.

[0067] The reconstructed data obtained by using hierarchical decompression on the cloud device to reconstruct the input data includes:

[0068] After receiving the progressively transmitted multi-layered input data, the cloud device performs layered decompression according to the transmission order;

[0069] The input data from multiple layers is dequantized and upsampled using a deep learning model with a super-prior decoder and a vector quantization generative adversarial network decoder to generate reconstructed data.

[0070] The cloud-based device's updating of the digital twin model using the reconstructed data also includes:

[0071] The cloud device updates the digital twin model based on the quality of the reconstructed data;

[0072] The quality of the reconstructed data is evaluated using an image similarity evaluation index.

[0073] For example, this application extracts image layering features through a deep learning super-prior model and a vector quantization generative adversarial network model, optimizes residual redundancy by combining traditional algorithms, and uses a dynamic layering strategy to adapt to channel conditions, thereby achieving efficient image compression and progressive transmission.

[0074] By collecting multimodal data such as images and text through sensing nodes, and introducing an edge computing architecture, a lightweight preprocessing model is deployed on the sensing nodes to perform feature filtering and preprocessing on the data, reducing the amount of data to be processed in subsequent steps and standardizing the data, thus achieving edge-side collaboration between data acquisition and preprocessing.

[0075] Specifically, it includes:

[0076] Data Acquisition. Sensing nodes acquire multimodal data in real time through a multi-source heterogeneous data acquisition network: image acquisition nodes deploy industrial cameras and infrared equipment to capture images; text nodes use text recognition engines to extract structured text. The acquired image, environmental, and text data are then preliminarily labeled.

[0077] Deep learning preprocessing. Deep learning is used for multi-stage processing: multimodal data cleaning, feature optimization, and standardization to improve data quality and efficiency. Finally, edge computing architecture is used for end-user preprocessing, reducing data volume while maintaining high information entropy, providing efficient and standardized input for model training. Then, layered compression and transmission adaptation are performed. Based on data characteristics and transmission requirements, the compression process is divided into multiple layers to ensure that the receiving end can decode sequentially with increasing quality.

[0078] Specifically, it includes:

[0079] Base layer generation. Generates minimum-quality but complete and identifiable data, ensuring reconstruction of the basic content even under worst-case channel conditions. Structural features are extracted preferentially using a super-prior model and a vector quantization generative adversarial network model, and grey relational coefficient β is used for further analysis. i The formula for analyzing and filtering multimodal data features that are strongly correlated with the target task is as follows:

[0080]

[0081] Where, ξ i (k) is the grey relational coefficient, Δ i (k) is the absolute difference, Δ max Δ min It is the global extremum, and ρ is the resolution coefficient. When β i Data with a correlation coefficient <0.7 is discarded to reduce the amount of data processed later. Run-length encoding (RLE) and fixed dictionary string compression (LZSS) algorithms are applied to the regular regions. The run-length encoding and compression formula for the regular regions is as follows:

[0082] C = {(c1,v1),(c2,v2),…,(c k ,v k )}

[0083] Among them, ci v represents the number of consecutive repetitions. i This corresponds to the pixel value.

[0084] Enhancement layers are layered. Multi-level quality enhancement data is provided above the base layer, allowing it to be transmitted progressively according to channel conditions, improving high-frequency details. Channel filtering and quantization are performed on the latent representation Y of the encoder output, where Mask(Y,k%) retains the top k% of the most important channels.

[0085]

[0086] The residuals of the base layer are hierarchically encoded, and the codebook index c is selected in the m-th stage. m :

[0087]

[0088] Where T i [c i ] represents the codeword of the i-th stage, and x is the input vector.

[0089] Transmission adaptation layer encapsulation. The compressed, layered data is encapsulated into a format suitable for progressive transmission, the priority of each layer is clearly defined, and channel fluctuations are dynamically adapted.

[0090] Next, progressive transmission and scheduling optimization are performed. The progressive transmission mechanism dynamically adjusts the transmitted content of compressed data according to channel conditions to optimize transmission efficiency. Specifically, this includes:

[0091] Transmit the base layer. At low signal-to-noise ratios, the base layer is transmitted first to quickly reconstruct low-quality data. The receiver only obtains the base layer codebook index and generates initial data through dequantization. The initial data is...

[0092]

[0093] The resolution is low, but it can meet basic visual needs.

[0094] Transmission enhancement layer. When the channel improves, a transmission enhancement layer is added, sequentially transmitting residual codebook indices. The receiver then uses residual superposition. Gradually refine multimodal data to progressively improve quality. Mathematically, this can be represented as:

[0095]

[0096] Where M stages For the number of residual levels transmitted, each level of codebook CB m This corresponds to more refined residual information.

[0097] Transmission scheduling optimization. Relay nodes employ a balanced shortest path tree, allocating transmission time slots based on node energy and distance to avoid channel contention. Simultaneously, the number of available bits N is calculated in real-time based on channel conditions. bits Select the maximum transmittable codebook level M stages The bit budget N for matching asymptotic transmission bits :

[0098]

[0099] in Ensure that the transmitted data does not exceed the budget.

[0100] Finally, cloud-based layered decompression and feedback control. The cloud receives multi-layered data transmitted progressively, decompresses and reconstructs the data sequentially, and optimizes subsequent transmission strategies. Specifically, this includes:

[0101] Layered decompression and reconstruction are performed. The corresponding deep learning model's super-prior decoder and vector quantization generative adversarial network decoder are used for dequantization and upsampling to generate reconstructed data. The base layer decodes to generate initial data. The formula is:

[0102]

[0103] Residuals are gradually stacked in the reinforcement layer:

[0104]

[0105] Finally, through decoder g s Generate high-resolution data:

[0106]

[0107] Intelligent feedback control. The cloud platform updates the digital twin model based on decompression quality and analyzes transmission performance. Reconstruction quality is evaluated using image similarity metrics PSNR and SSIM.

[0108]

[0109] If PSNR < threshold, indicating insufficient reconstruction quality, adjust the sampling frequency and code number level of the sensing nodes, and update the maximum codebook level according to the channel state. Where B is the bandwidth, T slot For transmission time slots:

[0110]

[0111] If the channel continues to deteriorate, add a channel coding-assisted advanced prior model.

[0112] like Figure 3As shown, in the data compression stage, this application extracts structured prior features of the image through a super-prior model, generates multi-scale visual semantic features by combining a vector quantization generative adversarial network model, and uses grey relational analysis to screen key features strongly related to the target task, thereby reducing redundant data. Simultaneously, traditional algorithms are applied to regular regions to further optimize residual redundancy. The design philosophy is to integrate the semantic feature extraction capabilities of deep learning with the efficient redundancy elimination advantages of traditional algorithms, forming a complementary compression mechanism. Through staged feature extraction and optimization, compression efficiency and reconstruction quality are significantly improved.

[0113] This application introduces a transmission adaptation layer encapsulation technology. The base layer employs a super-prior model and serial list compression algorithm for algorithm compression, ensuring decodeability under minimal channel conditions. The enhancement layer optimizes detail reconstruction through vector quantization-based generative adversarial network layer number and residual quantization, dynamically transmitting according to channel state. The transmission adaptation layer adapts to channel fluctuations in real time through dynamic priority adjustment and adaptive modulation and coding techniques. Its principle is based on dynamically adjusting the transmission content and coding strategy according to channel state feedback, achieving progressive transmission of "preserving the basics despite channel degradation and enhancing details through channel optimization," significantly improving transmission reliability and bandwidth utilization in complex time-varying scenarios.

[0114] The image is reconstructed stepwise by dequantization using a super-prior decoder and a vector quantization generative adversarial network decoder, with real-time evaluation of the reconstruction quality. If the quality is insufficient, the sampling frequency of the sensing nodes is dynamically adjusted or the focus is shifted to the region of interest; if the channel continues to deteriorate, channel coding redundancy is increased to assist the super-prior model decoding. The idea is to use end-to-end closed-loop optimization to feed back the transmission effect to the front-end data acquisition and compression stages, forming a collaborative link of "compression-transmission-decompression-feedback" to ensure the real-time performance and data fidelity of the digital twin model.

[0115] Figure 2 This is a schematic diagram of a digital twin data transmission system for dynamically time-varying scenarios provided by an embodiment of the present invention; the present invention also provides a digital twin data transmission system for dynamically time-varying scenarios, comprising:

[0116] The sensing node module is used to collect multimodal data through sensing nodes;

[0117] The data processing pre-module is used to clean, optimize features, and standardize the multimodal data using a deep learning model deployed through edge computing to obtain input data.

[0118] The data compression module is used to perform layered compression on the input data;

[0119] The data transmission module is used to transmit the layered compressed input data to the cloud device via progressive transmission;

[0120] A cloud module is used to reconstruct the input data using layered decompression on the cloud device to obtain reconstructed data; the cloud device updates the digital twin model using the reconstructed data.

[0121] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0122] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for digital twin data transmission for dynamic time-varying scenarios, characterized in that, include: Multimodal data is collected through sensing nodes; The input data is obtained by cleaning, feature optimization, and standardization of the multimodal data through a deep learning model deployed by edge computing; The input data is compressed in layers; The input data, after being compressed in layers, is transmitted to the cloud device via progressive transmission; The cloud device uses layered decompression to reconstruct the input data to obtain reconstructed data. The cloud device updates the digital twin model using the reconstructed data.

2. The digital twin data transmission method for dynamic time-varying scene according to claim 1, characterized in that, The multimodal data acquired through the sensing nodes includes: The sensing node acquires the multimodal data through a multi-source heterogeneous network; The multi-source heterogeneous network includes industrial cameras, infrared devices, text recognition engines, and microphones; The multimodal data includes image, text, and audio data.

3. The digital twin data transmission method for dynamic time-varying scene according to claim 1, characterized in that, The layered compression of the input data includes: The input data is divided into a base layer, an enhancement layer, and a transport adaptation layer, and compressed separately. The layering of the input data is performed using a deep learning model, a hyper-prior encoder, and a vector quantization generative adversarial network encoder.

4. The digital twin data transmission method for dynamically time-varying scenarios according to claim 3, characterized in that, The base layer generates minimum quality, fully identifiable data; The enhancement layer provides multi-level quality enhancement data on the base layer; The enhancement layer is used to improve the high-frequency details of the base layer; The transport adaptation layer encapsulates the compressed, layered input data into a progressive transport format.

5. A digital twin data transmission method for dynamically time-varying scenarios according to claim 3, characterized in that, The process of transmitting the layered compressed input data to the cloud device via progressive transmission also includes: The base layer is preferentially used for data transmission at low signal-to-noise ratios. The enhancement layer is used to supplement transmission when the channel is improved.

6. The digital twin data transmission method for dynamically time-varying scenarios according to claim 1, characterized in that, The reconstructed data obtained by using hierarchical decompression on the cloud device to reconstruct the input data includes: After receiving the progressively transmitted multi-layered input data, the cloud device performs layered decompression according to the transmission order; The input data from multiple layers is dequantized and upsampled using a deep learning model with a super-prior decoder and a vector quantization generative adversarial network decoder to generate reconstructed data.

7. A digital twin data transmission method for dynamically time-varying scenarios according to claim 1, characterized in that, The cloud-based device's updating of the digital twin model using the reconstructed data also includes: The cloud device updates the digital twin model based on the quality of the reconstructed data; The quality of the reconstructed data is evaluated using an image similarity evaluation index.

8. A digital twin data transmission system for dynamic, time-varying scenarios, characterized in that, include: The sensing node module is used to collect multimodal data through sensing nodes; The data processing pre-module is used to clean, optimize features, and standardize the multimodal data using a deep learning model deployed through edge computing to obtain input data. The data compression module is used to perform layered compression on the input data; The data transmission module is used to transmit the layered compressed input data to the cloud device via progressive transmission; A cloud module is used to reconstruct the input data using layered decompression on the cloud device to obtain reconstructed data; the cloud device updates the digital twin model using the reconstructed data.

Images