Digital twin experience interaction system and method based on lightweight container

Abstract

The invention relates to the field of digital twinning interaction, and discloses a digital twinning experience interaction system and method based on a light-weight container, and the method comprises the steps: carrying out the packaging initialization of a digital twinning unit based on a light-weight container mirror image, generating a computing power portrait, recording a container instance ID initialization scene, executing semantic, spatial and logical fragmentation, generating a fragmentation set, and carrying out the recognition of the fragmentation set. A bounding box, a semantic tag, an update frequency, an LOD level and texture hash are calculated for each piece, a change detection identifier is established, a client continuously collects an input event stream and view angle transformation information, a time sequence prediction module is executed at a container end, state difference is calculated, and a cloud-end-to-end cooperation ratio, a difference packet coding level and a transmission sequence are determined according to a computing power portrait. And outputting a prediction difference packet, executing incremental three-dimensional reconstruction and local rendering by the client, collecting frame rate, delay and lagging operation data at the same time, and reporting the data to a scheduler. The method has the advantage of improving the intelligent collaborative interaction capability.

Description

Technical Field

[0001] This invention relates to the field of digital twin interaction, specifically to a digital twin experience interaction system and method based on lightweight containers. Background Technology

[0002] Current high-fidelity digital twin systems face multiple technical bottlenecks in practical applications. These include slow loading of digital twin scenarios, high resource consumption, complex deployment affecting startup and elastic scaling, limited network bandwidth leading to significant operational latency, and the difficulty of guaranteeing high real-time performance and high-fidelity display using traditional video streaming or WebGL solutions. Cloud GPU resources are consumed in large quantities and are costly, terminal computing power is underutilized, real-time interaction in complex scenarios is limited, session management and container scheduling are difficult under high concurrency, stability is low, and there is pressure for expansion and fault recovery. Low-computing-power terminals have insufficient performance, cross-platform experiences are inconsistent, and user interaction latency is significant. Existing technologies struggle to simultaneously achieve high-fidelity display, low-latency interaction, resource optimization, and cross-platform consistency, failing to meet the comprehensive requirements of digital twin applications in terms of timeliness, reliability, accuracy, and efficiency. Therefore, designing a lightweight container-based digital twin experience interaction system and method to improve intelligent collaborative interaction capabilities is essential. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a digital twin experience interaction system and method based on lightweight containers, which has the advantage of improving intelligent collaborative interaction capabilities and solves the problems mentioned in the background technology.

[0004] To achieve the aforementioned goal of improving intelligent collaborative interaction capabilities, this invention provides the following technical solution: a digital twin experience interaction system and method based on lightweight containers, comprising the following steps: The digital twin unit is encapsulated and initialized based on a lightweight container image, and GPU frequency, rendering throughput, shader level, video memory and CPU performance are collected in real time to generate a computing power profile and record the container instance ID. Initialize the scene with container instance ID to perform semantic, spatial and logical sharding, generate a shard set, calculate bounding box, semantic label, update frequency, LOD level and texture hash for each shard, establish change detection identifier, and form shard metadata table; Based on the fragmented metadata table, the client continuously collects input event streams and viewpoint change information, and executes the timing prediction module on the container side to perform multi-step prediction of the view frustum parameters and viewpoint trajectory for future time steps, generating a set of potentially visible fragments. By combining the sharding metadata table and the set of potentially visible shards, the state difference is calculated, spatial compression and hierarchical coding are applied according to semantic weights, the cloud-edge collaboration ratio, differential packet coding level and transmission order are determined based on the computing power profile, and the predicted differential packet is output. The client receives the prediction difference packet, performs incremental 3D reconstruction and local rendering, and collects frame rate, latency and stuttering data and reports them to the scheduler to adjust the coordination ratio, container migration and scaling.

[0005] Preferably, the process of generating a computing power profile and recording the container instance ID is as follows: During the container initialization phase, multi-dimensional benchmark tests are performed on the performance parameters of GPU, CPU, and video memory, including rendering throughput evaluation, shader execution performance analysis, video memory access latency, and multi-threaded CPU load capacity testing. Real-time monitoring of the dynamic status of various hardware resources, including temperature changes, frequency fluctuations, and current task load, generates continuous time series data; By combining benchmark test results with real-time monitoring data, a multi-dimensional computing power profile is formed according to hardware identifiers and performance characteristics, and a unique container instance ID is generated for each digital twin unit.

[0006] Preferably, the process of generating the fragmented set is as follows: Based on the computing power profile and initialization parameters of the container instance ID, dynamically adjust the sharding granularity and LOD level; Based on the digital twin scene model, the scene is segmented according to spatial location, object type, logical relationship and segmentation dependency; Calculate the bounding box size, texture hash, and semantic tag for each fragment to form a complete fragment metadata set.

[0007] Preferably, the process of forming the sharded metadata table is as follows: Based on the complete set of fragment metadata, calculate the geometric boundary and bounding box size of each fragment; Based on the semantic information of the fragments and the interactive attributes, the object categories are labeled, the update frequency of each fragment is counted, and a texture hash value is generated and associated with the fragment priority information. Establish change detection flags and access flags for each shard, and organize the information into a structured shard metadata table.

[0008] Preferably, the process of executing the timing prediction module on the container side is as follows: The client continuously records user operation events according to time steps, including mouse operations, touch gestures and controller inputs, while also collecting view position, direction and movement speed; The container receives the event stream and, in conjunction with the fragment geometric boundaries, semantic tags, update frequency, LOD level, and texture hash recorded in the structured fragment metadata table, executes a time-series prediction algorithm. Model short-term and multi-step view trajectories and potential visible segments to generate a list of predicted visible segments and segment priority information.

[0009] Preferably, the process of generating a potentially visible set of fragments is as follows: The container-side uses a time-series prediction module to predict the visible fragment list and fragment priority information, and simulates the range of changes in the position, orientation and viewing angle of the view frustum at several future time steps. Mark the fragments that enter the visible area as potentially visible fragments; By combining historical user behavior data, shard access frequency, and shard priority information, a multi-step prediction list is generated, forming a set of potentially visible shards.

[0010] Preferably, the process of applying spatial compression and hierarchical coding according to semantic weights is as follows: The set of potentially visible fragments is compared piece by piece with the state record of the potentially visible fragments in the previous time step to identify differences in geometric deformation, texture update, LOD change and semantic attribute, and the state difference information is recorded in the differential data structure. Based on the semantic tags, interaction importance, and shard access frequency in the shard metadata table, the state differential data is weighted and sorted, and the update priority is marked. By combining the LOD level of the fragments, the texture compression scheme, and the spatial distribution characteristics, layered encoding and region compression processing are performed; The encoded differential data packets are matched with the predicted visible fragment list to generate the final data transmission priority list.

[0011] Preferably, the process of outputting the prediction difference packet is as follows: Utilize container instance computing power profiles to assess the number of shards a terminal can handle, its rendering throughput, and its real-time load status; The set of fragments that can be preloaded and rendered in real time on the client side is determined by combining the fragment priority list; Based on the data transmission priority list, each fragment differential packet is weighted and sorted, and the cloud-to-device fragment data transmission ratio, differential packet compression level, encoding method and transmission order are dynamically adjusted. Incremental differential compression and hierarchical coding are applied to high-priority fragments to generate and output complete prediction differential packets.

[0012] Preferably, the process of adjusting the collaboration ratio, container migration, and scaling is as follows: The client decodes the packets based on the predicted differential packets, according to the order of their fragmentation priorities. Incremental 3D reconstruction is performed based on the decoded fragmented data; Real-time collection of frame rate, rendering latency, stuttering, and chunk loading status data, and reporting performance metrics along with chunk rendering information to the scheduler; The scheduler dynamically adjusts the edge-cloud collaboration ratio, container migration strategy, and elastic scaling scheme based on performance feedback and sharding status.

[0013] A lightweight container-based digital twin experience interaction system includes: Container encapsulation module: Based on lightweight container images, it encapsulates digital twin units, collects computing power information, and records container instances; The sharding management module performs semantic, spatial, and logical sharding on the scene and generates a sharding set and metadata table. Prediction Calculation Module: Performs multi-step view frustum and view trajectory prediction at the container end to generate a set of potentially visible slices; Differential transmission module: Combines metadata and prediction results to calculate state differences, performs compression encoding and collaborative allocation according to semantic weights, and outputs prediction difference packets; Rendering optimization module: The client performs incremental 3D reconstruction and local rendering, while collecting runtime data and reporting it to the scheduler to dynamically adjust the resource and collaboration ratio.

[0014] Compared with existing technologies, the present invention provides a digital twin experience interaction system and method based on lightweight containers, which has the following beneficial effects: This invention achieves adaptive scheduling for edge-cloud collaboration by lightweight containerizing and encapsulating digital twin units and combining them with real-time acquired computing power profiles. It dynamically adjusts fragmentation transmission and encoding strategies based on the computing power and rendering performance of different terminals, thereby improving intelligent collaborative interaction capabilities. Through semantic, spatial, and logical fragmentation and multi-step view frustum prediction, it significantly improves the efficiency and predictability of incremental 3D data transmission. Utilizing layered differential coding and edge-cloud ratio optimization, it achieves efficient data compression and transmission. Simultaneously, incremental reconstruction and local rendering are performed on the client side, and the scheduling strategy is continuously adjusted through real-time acquisition of frame rate, latency, and stuttering information, achieving continuous optimization of rendering smoothness and user experience. This invention constructs a dynamically adaptable, multi-step predictive, and experience-driven edge-cloud digital twin system that significantly reduces data transmission volume and system resource consumption while ensuring real-time interaction and rendering quality, improving system response speed and stability. This demonstrates the technical advantages and innovation of the overall solution in edge-cloud collaboration, intelligent interaction, and real-time 3D rendering. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the method of the present invention; Figure 2 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0016] 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.

[0017] Example 1: Please refer to Figure 1 As shown in the embodiment of the present invention, the digital twin experience interaction method based on lightweight containers includes the following steps: S1: Based on a lightweight container image, the digital twin unit is encapsulated and initialized, and GPU frequency, rendering throughput, shader level, video memory and CPU performance are collected in real time to generate a computing power profile and record the container instance ID.

[0018] The process of generating a computing power profile and recording the container instance ID in S1 is as follows: During the container initialization phase, multi-dimensional benchmark tests are performed on the performance parameters of GPU, CPU, and video memory, including rendering throughput evaluation, shader execution performance analysis, video memory access latency, and multi-threaded CPU load capacity testing. During the container initialization phase, multi-dimensional benchmark tests are conducted on the performance parameters of the GPU, CPU, and video memory. GPU performance testing involves continuously rendering a series of standard 3D scenes, including models with different polygon counts and lighting complexities, recording the rendering time, rendering throughput, and shader execution time for each frame. Shader performance analysis uses typical rendering pipeline tasks to test vertex shading, pixel shading, and computational shading units, recording execution time and resource usage. Video memory performance testing involves continuously accessing data blocks of different sizes to measure access latency and bandwidth utilization. CPU multi-threaded load capacity testing uses multi-core matrix operations, floating-point operations, and task scheduling stress tests, recording the execution rate and task completion time of each core. During the test, timestamps, thread IDs, and hardware identification information of all operations are recorded to form a preliminary performance dataset.

[0019] Real-time monitoring of the dynamic status of various hardware resources, including temperature changes, frequency fluctuations, and current task load, generates continuous time series data; During container operation, the dynamic status of various hardware resources is continuously monitored, including GPU temperature and frequency fluctuations, changes in memory usage, CPU core load, and total system load. Data acquisition is performed by reading sensor information through hardware interfaces or drivers, with data collected once per second to form continuous time series data.

[0020] By combining benchmark test results with real-time monitoring data, a multi-dimensional computing power profile is formed according to hardware identifiers and performance characteristics, and a unique container instance ID is generated for each digital twin unit. First, the performance metrics of GPU, CPU, and video memory are normalized to map throughput, latency, load, etc. to a unified scale. Then, a performance vector for each hardware module is generated based on the hardware unit identifier. All hardware vectors are combined to form the computing power vector of the entire container instance. The time series mean, variance, and maximum value of each metric are recorded. Finally, a unique container instance ID is generated for each digital twin unit.

[0021] S2: Initialize the scene with container instance ID, perform semantic, spatial and logical sharding, generate a shard set, calculate bounding box, semantic label, update frequency, LOD level and texture hash for each shard, establish change detection identifier, and form a shard metadata table.

[0022] The process of generating the fragmented set in S2 is as follows: Based on the computing power profile and initialization parameters of the container instance ID, dynamically adjust the sharding granularity and LOD level; Based on the computing power profile and initialization parameters corresponding to the container instance ID, the computing power, rendering throughput, and real-time load information of the GPU, CPU, and video memory are read. The appropriate fragment size and number are calculated according to the complexity and object density of the digital twin scene. For high-computing-power containers, smaller fragment granularity and higher precision levels are allowed to retain more details. For low-computing-power containers, the fragment size is appropriately increased and the level of detail is reduced to ensure renderability. The LOD level is determined by analyzing the geometric complexity and texture resolution of the object, assigning multiple geometric and texture levels to each fragment, and binding the fragment granularity and LOD parameters to the container instance ID.

[0023] Based on the digital twin scene model, the scene is segmented according to spatial location, object type, logical relationship and segmentation dependency; Based on the digital twin scene model, the entire scene is spatially divided, and the segment boundaries are determined according to the three-dimensional coordinates of the objects. At the same time, the component structure within the segment is identified by combining the object type and logical relationship. During the segmentation process, the dependencies between segments are considered, such as the physical or logical connections between adjacent segments, to ensure the independence of segments and data integrity. Each segment records its object list, spatial boundary coordinates, and dependencies between segments, forming a preliminary structured segmented data table.

[0024] Calculate the bounding box size, texture hash, and semantic tag for each fragment to form a complete fragment metadata set; For each generated fragment, its geometric boundary is calculated and the bounding box size is generated. The minimum and maximum coordinate values ​​of the fragment in 3D space are recorded to form a complete boundary description. At the same time, the texture resources within the fragment are hashed to generate a unique texture identifier for differential detection and version management. Semantic tags are annotated using the type and interaction attributes of objects within the fragment, including static objects, dynamic objects, and interactive object categories. Finally, the fragment boundaries, texture hashes, and semantic tags are summarized to form a complete fragment metadata set.

[0025] The process of forming the sharded metadata table in S2 is as follows: Based on the complete set of fragment metadata, calculate the geometric boundary and bounding box size of each fragment; By reading the three-dimensional spatial coordinates and shape information of each object in the slice, the maximum and minimum boundary coordinates of the slice in three-dimensional space are generated, and the bounding box size is determined.

[0026] Based on the semantic information of the fragments and the interactive attributes, the object categories are labeled, the update frequency of each fragment is counted, and a texture hash value is generated and associated with the fragment priority information. Semantic tags are determined by the type and functional attributes of each object in the fragment, and the update frequency is calculated by combining historical or real-time update records. The texture data used by the fragment is hashed to uniquely identify the texture content, and this information is associated with the fragment access and rendering priority.

[0027] Establish change detection flags and access flags for each shard, and organize the information into a structured shard metadata table; By setting an identifier for each fragment to record state changes and access information, the geometric boundaries, semantic tags, update frequency, texture hashes, and identifiers of the fragments are stored in a unified manner to form a structured data table.

[0028] S3: Based on the fragmented metadata table, the client continuously collects input event streams and viewpoint change information, executes the timing prediction module on the container side, performs multi-step prediction of the view frustum parameters and viewpoint trajectory for future time steps, and generates a set of potentially visible fragments.

[0029] The process of executing the timing prediction module on the container side in S3 is as follows: The client continuously records user operation events according to time steps, including mouse operations, touch gestures and controller inputs, while also collecting view position, direction and movement speed; The client collects user input information at fixed time intervals, maps input events into structured data, including event type, timestamp and corresponding three-dimensional coordinates or operation parameters, and simultaneously records the position coordinates, orientation vector and movement speed vector of the viewpoint in three-dimensional space, forming a continuous time-series input stream.

[0030] The container receives the event stream and, in conjunction with the fragment geometric boundaries, semantic tags, update frequency, LOD level, and texture hash recorded in the structured fragment metadata table, executes a time-series prediction algorithm. The container associates the received time-series operation data with the shard metadata, indexes the geometric boundaries, semantic categories, update history and hierarchical details of each shard to form an input matrix, and the algorithm constructs a state transition model based on the operation event sequence and shard attributes, outputting the probability and position state of each shard that may be covered by the user's perspective in future time steps.

[0031] Model short-term and multi-step view trajectories and potential visible segments to generate a list of predicted visible segments and segment priority information; The time series prediction results are extended in multiple steps. By combining the fragment access history and geometric visibility conditions, a set of potentially visible fragments for several future time steps is constructed. The access priority of each fragment is calculated, and the sorted list of predicted visible fragments is output.

[0032] The process of generating a potential visible fragment set in S3 is as follows: The container-side uses a time-series prediction module to predict the visible fragment list and fragment priority information, and simulates the range of changes in the position, orientation and viewing angle of the view frustum at several future time steps. The container uses the generated list of predicted visible segments and the priority information of each segment to combine the spatial position of the segments with the view frustum geometry model. It calculates the possible spatial position, orientation changes and rotation range of the viewpoint in the short and multiple steps in the future at fixed time steps, forming a continuous view frustum trajectory sequence.

[0033] Mark the fragments that enter the visible area as potentially visible fragments; The container calculates the intersection between the simulated future view frustum coverage and the spatial boundary of the predicted visible patches. It then identifies patches that spatially intersect with the view frustum at any time step, recording their attributes, location, and potential visible time step in the potential visible patch identifier structure. By combining historical user behavior data, shard access frequency, and shard priority information, a multi-step prediction list is generated, forming a set of potentially visible shards; The container integrates the historical access records, user interaction behavior patterns, and the priority of each potentially visible shard in multi-step prediction, generates a multi-step prediction list for each future time step according to the time series, sorts the shards in the list and marks their priority and visible time window, and finally forms a structured set of potentially visible shards.

[0034] S4: Combine the sharding metadata table and the set of potentially visible shards to calculate the state difference, apply spatial compression and hierarchical coding according to semantic weights, determine the cloud-edge collaboration ratio, differential packet coding level and transmission order based on the computing power profile, and output the predicted differential packet.

[0035] The process of applying spatial compression and hierarchical coding according to semantic weights in S4 is as follows: The set of potentially visible fragments is compared piece by piece with the state record of the potentially visible fragments in the previous time step to identify differences in geometric deformation, texture update, LOD change and semantic attribute, and the state difference information is recorded in the differential data structure. For each potentially visible fragment, the container compares its current frame's spatial boundary, mesh vertex coordinates, material texture hash value, LOD level, and semantic tag with the corresponding fragment state from the previous time step. By calculating vertex offset, texture hash changes, and LOD changes, it generates state difference entries.

[0036] Based on the semantic tags, interaction importance, and shard access frequency in the shard metadata table, the state differential data is weighted and sorted, and the update priority is marked. Read the semantic tags, user interaction importance level and historical access frequency information defined in the metadata table for each shard, calculate the weighted score for each state difference entry, and sort the entries by comprehensively considering geometric changes, texture updates and shard priorities to generate a differential data list with update priority identifiers.

[0037] By combining the LOD level of the fragments, the texture compression scheme, and the spatial distribution characteristics, layered encoding and region compression processing are performed; Based on the LOD level of each piece, the corresponding mesh refinement layer is selected, and the corresponding texture compression algorithm is applied to encode the material data. At the same time, the distribution information of the pieces in space is combined to perform region segmentation compression, and the geometric, texture and semantic difference data of the pieces are organized in layers to form compressed encoding units.

[0038] The encoded differential data packets are matched with the predicted visible fragment list to generate the final data transmission priority list; Each encoded differential data packet is associated with the position and priority of the corresponding potential visible fragment in the predicted visible fragment list, and a complete data transmission priority list is generated by sorting according to weighted priority and time series.

[0039] The process of outputting the prediction difference packet in S4 is as follows: Utilize container instance computing power profiles to assess the number of shards a terminal can handle, its rendering throughput, and its real-time load status; The system reads GPU, CPU, and video memory performance parameters from the container instance's computing power profile, and calculates the number of fragments that the terminal can process per unit time, the throughput of rendering instructions it can handle, and the concurrent rendering capability, based on the current hardware load, temperature, frequency fluctuations, and task usage. It then generates a time-series load prediction.

[0040] The set of fragments that can be preloaded and rendered in real time on the client side is determined by combining the fragment priority list; Read the list of predictable shards and the priority information defined for each shard in the shard metadata table. Based on the range of the number of shards that can be carried on the edge, allocate high-priority shards as real-time rendering sets and low-priority shards as pre-loading sets.

[0041] Based on the data transmission priority list, each fragment differential packet is weighted and sorted, and the cloud-to-device fragment data transmission ratio, differential packet compression level, encoding method and transmission order are dynamically adjusted. Based on the order and weight information of the differential data packets in the data transmission priority list, a weighted score is calculated for each differential packet, placing high-priority and high-frequency access fragment data packets at the front. At the same time, the cloud-to-end transmission ratio is dynamically adjusted according to the processing capacity, network bandwidth, and latency of the end side. An appropriate compression level and encoding method are selected for each differential packet, and a complete fragment data transmission order table is generated.

[0042] Incremental differential compression and hierarchical coding are applied to high-priority fragments to generate and output complete prediction differential packets. For the fragmented data packets marked as high priority, the geometric changes, texture updates, and semantic differences are compressed according to the hierarchical coding strategy to generate incremental differential data units. The coding results of each fragment are then summarized in the transmission order to form the final complete predictive differential packet data structure.

[0043] S5: The client receives the prediction difference packet, performs incremental 3D reconstruction and local rendering, and collects frame rate, latency and stuttering data and reports them to the scheduler to adjust the coordination ratio, container migration and scaling.

[0044] The process of adjusting collaboration ratios, container migration, and scaling in S5 is as follows: The client decodes the packets based on the predicted differential packets, according to the order of their fragmentation priorities. Read the data of each fragment in the complete prediction differential packet, decode the data of each fragment one by one according to the priority order defined in the fragment metadata table, including geometric vertex information, texture data and LOD level information, and cache the decoded data in the edge memory.

[0045] Incremental 3D reconstruction is performed based on the decoded fragmented data; The decoded fragmented geometry and texture data are assembled into the overall 3D scene according to the spatial boundaries and dependencies of the fragments. For fragments that have been loaded in the previous time step, only the changed vertex, texture or LOD information is updated. The vertex buffer and texture map of the scene are updated by incremental calculation to realize continuous incremental reconstruction of the 3D scene. At the same time, the spatial index and rendering level of each fragment are maintained to support efficient rendering.

[0046] Real-time collection of frame rate, rendering latency, stuttering, and chunk loading status data, and reporting performance metrics along with chunk rendering information to the scheduler; During each rendering frame cycle, GPU rendering frame rate, CPU rendering latency, texture upload and geometry processing time are collected, while the fragment loading progress and rendering status are recorded. This data is packaged in time sequence and associated with the priority and status information of each fragment. It is then uploaded to the scheduler module in real time through the network interface to provide real-time feedback on edge performance and resource usage.

[0047] The scheduler dynamically adjusts the end-to-cloud collaboration ratio, container migration strategy, and elastic scaling scheme based on performance feedback and sharding status. It receives uploaded performance data and sharding status information, calculates and analyzes the number of shards that can be processed on the edge, rendering throughput and latency indicators, reallocates the collaborative processing ratio between the cloud and the edge based on this data, decides whether to migrate container instances to optimize load distribution, and triggers container scaling up or down operations based on current resource usage and load conditions, generates dynamic scheduling instructions and sends them to the edge and cloud for execution, realizing real-time adaptive adjustment of sharding processing and resources.

[0048] Example 2: Figure 2 As shown, a digital twin experience interaction system based on lightweight containers includes: Container encapsulation module: Based on lightweight container images, it encapsulates digital twin units, collects computing power information, and records container instances; The sharding management module performs semantic, spatial, and logical sharding on the scene and generates a sharding set and metadata table. Prediction Calculation Module: Performs multi-step view frustum and view trajectory prediction at the container end to generate a set of potentially visible slices; Differential transmission module: Combines metadata and prediction results to calculate state differences, performs compression encoding and collaborative allocation according to semantic weights, and outputs prediction difference packets; Rendering optimization module: The client performs incremental 3D reconstruction and local rendering, while collecting runtime data and reporting it to the scheduler to dynamically adjust the resource and collaboration ratio.

[0049] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0050] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital twin experience interaction method based on lightweight containers, characterized in that, Includes the following steps: The digital twin unit is encapsulated and initialized based on a lightweight container image, and GPU frequency, rendering throughput, shader level, video memory and CPU performance are collected in real time to generate a computing power profile and record the container instance ID. Initialize the scene with container instance ID to perform semantic, spatial and logical sharding, generate a shard set, calculate bounding box, semantic label, update frequency, LOD level and texture hash for each shard, establish change detection identifier, and form shard metadata table; Based on the fragmented metadata table, the client continuously collects input event streams and viewpoint change information, and executes the timing prediction module on the container side to perform multi-step prediction of the view frustum parameters and viewpoint trajectory for future time steps, generating a set of potentially visible fragments. By combining the sharding metadata table and the set of potentially visible shards, the state difference is calculated, spatial compression and hierarchical coding are applied according to semantic weights, the cloud-edge collaboration ratio, differential packet coding level and transmission order are determined based on the computing power profile, and the predicted differential packet is output. The client receives the prediction difference packet, performs incremental 3D reconstruction and local rendering, and collects frame rate, latency and stuttering data and reports them to the scheduler to adjust the coordination ratio, container migration and scaling.

2. The digital twin experience interaction method based on lightweight containers according to claim 1, characterized in that, The process of generating a computing power profile and recording the container instance ID is as follows: During the container initialization phase, multi-dimensional benchmark tests are performed on the performance parameters of GPU, CPU, and video memory, including rendering throughput evaluation, shader execution performance analysis, video memory access latency, and multi-threaded CPU load capacity testing. Real-time monitoring of the dynamic status of various hardware resources, including temperature changes, frequency fluctuations, and current task load, generates continuous time series data; By combining benchmark test results with real-time monitoring data, a multi-dimensional computing power profile is formed according to hardware identifiers and performance characteristics, and a unique container instance ID is generated for each digital twin unit.

3. The digital twin experience interaction method based on lightweight containers according to claim 2, characterized in that, The process of generating a sharded set is as follows: Based on the computing power profile and initialization parameters of the container instance ID, dynamically adjust the sharding granularity and LOD level; Based on the digital twin scene model, the scene is segmented according to spatial location, object type, logical relationship and segmentation dependency; For each fragment, the bounding box size, texture hash, and semantic tag are calculated to form a complete fragment metadata set.

4. The digital twin experience interaction method based on lightweight containers according to claim 3, characterized in that, The process of creating the sharded metadata table is as follows: Based on the complete set of fragment metadata, calculate the geometric boundary and bounding box size of each fragment; Based on the semantic information of the fragments and the interactive attributes, the object categories are labeled, the update frequency of each fragment is counted, and a texture hash value is generated and associated with the fragment priority information. Establish change detection flags and access flags for each shard, and organize the information into a structured shard metadata table.

5. The digital twin experience interaction method based on lightweight containers according to claim 4, characterized in that, The process of executing the timing prediction module on the container side is as follows: The client continuously records user operation events according to time steps, including mouse operations, touch gestures and controller inputs, while also collecting view position, direction and movement speed; The container receives the event stream and, in conjunction with the sharding geometric boundaries, semantic tags, update frequency, LOD level, and texture hash recorded in the structured sharding metadata table, executes a timing prediction algorithm. Model short-term and multi-step view trajectories and potential visible segments to generate a list of predicted visible segments and segment priority information.

6. The digital twin experience interaction method based on lightweight containers according to claim 5, characterized in that, The process of generating a potentially visible set of fragments is as follows: The container-side uses a time-series prediction module to predict the visible fragment list and fragment priority information, and simulates the range of changes in the position, orientation and viewing angle of the view frustum at several future time steps. Mark the fragments that enter the visible area as potentially visible fragments; By combining historical user behavior data, shard access frequency, and shard priority information, a multi-step prediction list is generated, forming a set of potentially visible shards.

7. The digital twin experience interaction method based on lightweight containers according to claim 6, characterized in that, The process of applying spatial compression and hierarchical coding according to semantic weights is as follows: The set of potentially visible fragments is compared with the state record of the potentially visible fragments in the previous time step to identify geometric deformation, texture update, LOD change and semantic attribute differences, and the state difference information is recorded in the differential data structure. Based on the semantic tags, interaction importance, and shard access frequency in the shard metadata table, the state differential data is weighted and sorted, and the update priority is marked. By combining the LOD level of the fragments, the texture compression scheme, and the spatial distribution characteristics, layered encoding and region compression processing are performed; The encoded differential data packets are matched with the predicted visible fragment list to generate the final data transmission priority list.

8. The digital twin experience interaction method based on lightweight containers according to claim 7, characterized in that, The process of outputting the prediction difference packet is as follows: Utilize container instance computing power profiles to assess the number of shards a terminal can handle, its rendering throughput, and its real-time load status; The set of fragments that can be preloaded and rendered in real time on the client side is determined by combining the fragment priority list; Based on the data transmission priority list, each fragment differential packet is weighted and sorted, and the cloud-to-device fragment data transmission ratio, differential packet compression level, encoding method and transmission order are dynamically adjusted. Incremental differential compression and hierarchical coding are applied to high-priority fragments to generate and output complete prediction differential packets.

9. The digital twin experience interaction method based on lightweight containers according to claim 8, characterized in that, The process of adjusting collaboration ratios, container migration, and scaling is as follows: The client decodes the packets based on the predicted differential packets, according to the order of their fragmentation priorities. Incremental 3D reconstruction is performed based on the decoded fragmented data; Real-time collection of frame rate, rendering latency, stuttering, and chunk loading status data, and reporting performance metrics along with chunk rendering information to the scheduler; The scheduler dynamically adjusts the edge-cloud collaboration ratio, container migration strategy, and elastic scaling scheme based on performance feedback and sharding status.

10. A lightweight container-based digital twin experience interaction system, applied to the method described in any one of claims 1-9, characterized in that, include: Container encapsulation module: Based on lightweight container images, it encapsulates digital twin units, collects computing power information, and records container instances; The sharding management module performs semantic, spatial, and logical sharding on the scene and generates a sharding set and metadata table. Prediction Calculation Module: Performs multi-step view frustum and view trajectory prediction at the container end to generate a set of potentially visible slices; Differential transmission module: Combines metadata and prediction results to calculate state differences, performs compression encoding and collaborative allocation according to semantic weights, and outputs prediction difference packets; Rendering optimization module: The client performs incremental 3D reconstruction and local rendering, while collecting runtime data and reporting it to the scheduler to dynamically adjust the resource and collaboration ratio.

Images