A one-click publishing and link generation method and system for a virtual space application

Abstract

The application discloses a one-key publishing and link generation method and system for virtual space application, and the method comprises the following steps: receiving virtual scene data to perform Gaussian spherical harmonic decomposition, converting the scene data into multi-frequency band rendering components, and constructing a frequency domain deployment mapping table; generating an interleaved publishing channel through frequency domain inverse transformation and time sequence misplacement analysis; identifying network conflict points by analyzing channel phase difference and transmission delay distribution, determining avoidance opportunity to generate staggered transmission paths; performing load conduction analysis on deployment flow, identifying load aggregation areas, and reversely deducing load release channel position to set a shunt node; constructing a gradual resource pool based on the rendering components, generating a priority sequence and an elastic access matrix through entropy value sorting; topologically fusing the access matrix and the transmission path to generate an access convergence point through weight analysis; and generating a unique link identifier based on the space-time characteristics of the convergence point to complete intelligent deployment of the virtual space application and realize one-key publishing of network links.

Description

Technical Field

[0001] This invention relates to the field of virtual reality technology, and in particular to a method and system for one-click publishing and link generation of virtual space applications. Background Technology

[0002] With the rapid development of virtual reality technology, new scene representation technologies such as 3DGS (3D Gaussian Splatting) have brought unprecedented visual experiences to virtual space applications. However, the release of virtual space applications faces many challenges: the scene data volume is enormous, containing a massive number of 3D primitives and rendering parameters; the performance differences between different terminal devices are huge, making it difficult to provide a consistent user experience; network transmission is prone to congestion and resource contention, affecting access quality.

[0003] Existing virtual space publishing solutions mostly employ static data packaging and fixed transmission strategies, failing to dynamically adjust based on actual network conditions and device capabilities. In scenarios with multiple users accessing the site concurrently, traditional solutions lack effective load balancing and resource scheduling mechanisms, frequently resulting in some nodes being overloaded while others remain idle. Furthermore, existing solutions lack intelligent resource location capabilities in the generated access links, leading to a poor user experience. Therefore, a method is urgently needed to address at least one of these problems. Summary of the Invention

[0004] This invention discloses a one-click publishing and link generation method and system for virtual space applications. It aims to convert virtual scene data into multi-layered rendering components through frequency domain decomposition technology to achieve platform adaptive deployment; optimize data transmission paths through network conflict identification and peak-shifting transmission strategies; achieve load balancing through load analysis and distribution node deployment; and provide users with a convenient one-click publishing experience and a unified link access portal through intelligent topology fusion and identifier generation technology.

[0005] The first aspect of this invention proposes a one-click publishing and link generation method for virtual space applications, comprising the following steps:

[0006] Receive the edited 3DGS virtual scene data, perform Gaussian spherical harmonic decomposition on the 3DGS virtual scene data to generate multi-band rendering components, and construct a frequency domain deployment mapping table through the multi-band rendering components;

[0007] Based on the frequency domain deployment mapping table, perform inverse frequency domain transformation to obtain redundant rendering paths, perform timing misalignment analysis on the redundant rendering paths to obtain idle processing windows, and perform interleaving scheduling on the idle processing windows to generate interleaved publishing channels.

[0008] Phase difference and transmission delay distribution are extracted from the interleaved publishing channel. Network conflict points are identified based on the phase difference. Avoidance timing is determined through the transmission delay distribution. Peak-shifting transmission paths are generated by combining the network conflict points and the avoidance timing.

[0009] Based on the peak-shifting transmission path, a conflict-free deployment flow is constructed. Load transmission analysis is performed on the conflict-free deployment flow to form a deployment load gradient. Load clustering areas are identified based on the deployment load gradient. Based on the load clustering areas, the load release channel position is deduced in reverse to generate a split deployment node.

[0010] A gradient resource pool is constructed based on the traffic splitting deployment node and the multi-band rendering components. The gradient resource pool is sorted by entropy decreasing to generate a priority sequence. A tiered push strategy is generated based on the priority sequence to form an elastic access matrix.

[0011] The elastic access matrix and the off-peak transmission path are topologically fused to generate access convergence points. A unique link identifier is generated based on the access convergence points to complete the one-click publishing and link generation of virtual space applications.

[0012] A second aspect of this invention proposes a one-click publishing and link generation system for virtual space applications, comprising:

[0013] The scene parsing module is used to receive the edited 3DGS virtual scene data, perform Gaussian spherical harmonic decomposition on the 3DGS virtual scene data to generate multi-band rendering components, and construct a frequency domain deployment mapping table through the multi-band rendering components.

[0014] The intelligent scheduling module is used to perform inverse frequency domain transformation based on the frequency domain deployment mapping table to obtain redundant rendering paths, perform time-series misalignment analysis on the redundant rendering paths to obtain idle processing windows, and perform interleaving scheduling on the idle processing windows to generate interleaved publishing channels.

[0015] The conflict prevention module is used to extract the phase difference and transmission delay distribution from the interleaved publishing channel, identify network conflict points based on the phase difference, determine the avoidance timing through the transmission delay distribution, and generate a staggered transmission path by combining the network conflict points and the avoidance timing.

[0016] The load distribution module is used to construct a conflict-free deployment flow based on the off-peak transmission path, perform load transmission analysis on the conflict-free deployment flow to form a deployment load gradient, identify load clustering areas based on the deployment load gradient, and deduce the load release channel position based on the load clustering areas to generate a distribution deployment node.

[0017] The resource optimization module is used to construct a gradient resource pool based on the traffic splitting deployment node and the multi-band rendering component, sort the gradient resource pool by decreasing entropy value to generate a priority sequence, and generate a tiered push strategy to form an elastic access matrix based on the priority sequence.

[0018] The link building module is used to perform topology fusion of the elastic access matrix and the off-peak transmission path to generate access convergence points, and generate unique link identifiers based on the access convergence points to complete one-click publishing and link generation of virtual space applications.

[0019] The beneficial effects of this invention are reflected in the following points: 1. By converting virtual scene data into multi-band rendering components through Gaussian spherical harmonic decomposition and establishing a frequency domain deployment mapping table, layered representation of scene data and platform adaptive transmission are achieved. Low-frequency components prioritize basic rendering quality, while high-frequency components are loaded on demand to provide detail enhancement. Simultaneously, the efficiency of multi-path parallel transmission is effectively improved through frequency domain inverse transformation and interleaved publishing channel design. 2. Network conflict points are accurately identified through phase difference analysis, and avoidance timing is predicted by combining transmission delay distribution. The generated off-peak transmission path actively avoids network congestion. Furthermore, load transmission analysis locates load clusters, and the reverse derivation of the set distribution nodes effectively disperses access hotspots, achieving balanced utilization of network resources. 3. A gradient resource pool and elastic access matrix constructed based on information entropy values ​​enable differentiated resource management and tiered push, prioritizing critical resources. Access convergence points and unique link identifiers generated through topology fusion simplify complex distributed resource access into a single link operation. Users do not need to concern themselves with the underlying resource distribution and scheduling details, truly achieving a one-click publishing experience for virtual space applications.

[0020] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a one-click publishing and link generation method for a virtual space application according to the present invention.

[0022] Figure 2 This is a structural block diagram of a one-click publishing and link generation system for virtual space applications according to the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0024] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicator will also change accordingly.

[0025] It should also be noted that when a component is described as "fixed to" or "set on" another component, it can be directly on the other component or there may be an intervening component present. When a component is described as "connected to" another component, it can be directly connected to the other component or there may be an intervening component present.

[0026] Furthermore, the use of terms such as "first" and "second" in this application is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed in this application.

[0027] The technical solutions of the embodiments of this application will be described below.

[0028] like Figure 1 As shown, this embodiment of the invention provides a one-click publishing and link generation method for virtual space applications, including the following steps S110-S160:

[0029] Step S110: Receive the edited 3DGS virtual scene data, perform Gaussian spherical harmonic decomposition on the 3DGS virtual scene data to generate multi-band rendering components, and construct a frequency domain deployment mapping table through the multi-band rendering components.

[0030] Specifically, the system receives edited 3DGS virtual scene data. It acquires the processed 3DGS virtual scene data through a standardized data receiving interface. The data uses a 3D Gaussian scatter cloud representation, where each Gaussian primitive contains spatial coordinates (x, y, z), scaling factors (sx, sy, sz), rotation quaternions (qw, qx, qy, qz), opacity α, and a vector of spherical harmonic coefficients. The storage structure of the 3DGS virtual scene data is optimized, using a compact binary format to reduce storage overhead. A single scene contains tens of thousands to millions of Gaussian distributed primitives. During reception, the spatial bounding box volume and geometric dimensions of the scene are calculated, and the total number and spatial distribution characteristics of Gaussian primitives are statistically analyzed. The color attributes of the Gaussian primitives are extracted and organized, including RGB color components, transparency information, and spherical harmonic illumination coefficients, constructing a complete visual feature dataset. The spatial relationships between adjacent Gaussian primitives are analyzed, calculating the distance relationships and geometric overlap between primitives, and establishing spatial connection mappings. The data receiving module adopts a streaming processing architecture, supporting efficient loading of large-scale 3DGS scene data. After receiving the data, the system integrates complete scene information, including key data such as geometric structure, visual attributes, and spatial distribution, to form standardized 3DGS virtual scene data.

[0031] In some embodiments, the step of generating multi-band rendering components by performing Gaussian spherical harmonic decomposition on the 3DGS virtual scene data includes: evaluating the rendering load based on the 3DGS virtual scene data to determine the unpacking order, wherein the rendering load includes Gaussian distribution density, spatial dimensional coupling, and spectral energy nonuniformity; setting spherical harmonic decomposition parameters according to the unpacking order; and using the spherical harmonic decomposition parameters to perform frequency domain unpacking on the 3DGS virtual scene data to generate multi-band rendering components.

[0032] The rendering load and unfolding order are determined based on 3DGS virtual scene data assessment. The rendering load includes Gaussian distribution density, spatial dimensional coupling, and spectral energy non-uniformity. First, Gaussian distribution density is calculated based on the 3DGS virtual scene data. All Gaussian primitives in the scene are traversed, their spatial distribution is statistically analyzed, the scene bounding box volume V is calculated, and the total number of Gaussian points N is obtained. The density ρ = N / V reflects the richness of scene detail; higher density indicates the need for more frequency domain components for accurate representation. Spatial dimensional coupling is analyzed based on the 3DGS virtual scene data. The spatial relationships between adjacent Gaussian primitives are examined, and the overlap and connection strength between primitives are calculated. Regions with high coupling form visually continuous surfaces and can be represented with fewer spherical harmonic coefficients, while regions with low coupling contain more independent details and require a higher unfolding order. Spectral energy non-uniformity is assessed based on the 3DGS virtual scene data. Spatial frequency domain analysis is performed on scene color information, and the energy distribution of different frequency components is calculated. Non-uniformity η reflects the concentration of energy in the spectrum. Energy concentrated in low frequencies indicates smooth scene changes, while energy dispersed in high frequencies indicates rich detail and texture. The overall rendering load L_render is calculated by combining three indicators: w1·ρ + w2·C + w3·η, where w1, w2, and w3 are normalized weighting coefficients, ρ is the Gaussian distribution density, C is the spatial dimensional coupling measure, and η is the spectral energy non-uniformity. Based on the rendering load value L_render, the spherical harmonic unfolding order l_max is determined: when L_render < 0.3, the scene is simple, and an unfolding of order 3-4 is sufficient; when 0.3 ≤ L_render < 0.7, the scene is moderately complex, and an unfolding of order 5-6 is used; when L_render ≥ 0.7, the scene is complex, and an unfolding of order 7 or higher is required to maintain visual quality.

[0033] The spherical harmonic decomposition parameters are set according to the expansion order. The total number of spherical harmonic basis functions required is calculated based on the determined expansion order l_max. A complete l-th order spherical harmonic expansion requires (l_max+1)² basis functions, each corresponding to a unique angular distribution pattern. An angular sampling scheme is configured according to the expansion order: 2l_max+1 sampling points are set in the polar angle θ direction, and 4l_max+2 sampling points are set in the azimuth angle φ direction, ensuring that the sampling density meets the accuracy requirements of the selected order. Computational and storage resources are allocated according to the expansion order. Each Gaussian element requires 3×(l_max+1)² floating-point numbers to store the spherical harmonic coefficients of the RGB three channels, with sufficient pre-allocated memory space to avoid dynamic allocation overhead during runtime. An appropriate numerical integration method is selected according to the expansion order: Gaussian quadrature with fixed sampling points is used for lower-order expansions, while adaptive integration algorithms are used for higher-order expansions to improve computational accuracy. Numerical stability control parameters are set according to the expansion order, including minimum coefficient threshold, normalization factor, and condition number limits, to prevent numerical instability in higher-order calculations. The calculation process is optimized based on the expansion order. The recursive relationship of spherical harmonic functions is used to reduce the calculation of trigonometric functions. A lookup table is used to store commonly used values, and the processing efficiency is improved by block parallelism.

[0034] Multi-band rendering components are generated by frequency domain expansion of 3DGS virtual scene data using spherical harmonic decomposition parameters. Using the set spherical harmonic decomposition parameters, each Gaussian primitive in the 3DGS virtual scene data is traversed to extract its color attributes and directional distribution information. A spherical harmonic transformation is performed on each Gaussian primitive, decomposing the 3D color distribution according to its complexity hierarchy: the zero-order component stores ambient lighting, the first-order component encodes directional information, and higher-order components contain texture details. The spherical harmonic coefficients c_lm for each frequency level are calculated through numerical integration, where l (0 ≤ l ≤ l_max) represents the order indicating the complexity level, m represents the degree indicating the directional component, and c_lm serves as a weighting coefficient quantifying the contribution intensity of each frequency component. The calculated spherical harmonic coefficients are used to perform frequency domain decomposition, separating the original color information into independent components of different frequencies, each representing a specific visual feature. The coefficients are grouped based on frequency characteristics: coefficients with l=0 form the zero-frequency component, preserving ambient lighting; coefficients with l=1 form the low-frequency component, encoding directional lighting; coefficients with l=2 and 3 form the mid-frequency component, representing surface reflection characteristics; and coefficients with l≥4 form the high-frequency component, containing fine texture details. Independent rendering components are generated using the grouping results. Each frequency band can be stored, compressed, and transmitted separately, supporting on-demand loading and progressive rendering. The original scene is reconstructed using the generated multi-frequency rendering components, achieving a flexible balance between quality and performance through selective combination of different frequency bands.

[0035] A frequency domain deployment mapping table is constructed using multi-band rendering components. A hierarchical data organization structure is designed based on the generated multi-band rendering components, establishing an index system for different frequency rendering components according to visual importance and data dependencies. Metadata tags are assigned to each frequency band, recording key information such as frequency band identifier, data size, compression format, precision level, and dependencies. Adaptation rules are established based on the hardware characteristics of the target platform, mapping platform capabilities to appropriate frequency band combinations to ensure optimal matching between rendering effects and hardware resources. A bandwidth-aware transmission priority is designed, with low-frequency basic components receiving the highest priority to ensure basic scene visibility, while mid- and high-frequency components are dynamically scheduled based on network conditions. A two-dimensional mapping table structure M[platform][bandwidth]={freq_list,order,size} is constructed, where platform is the platform type, bandwidth is the bandwidth level, freq_list is the list of selected frequency bands, order is the transmission order, and size is the total data volume. The mapping table uses a lookup table format for fast access, pre-calculating common configuration combinations to reduce runtime decision-making overhead.

[0036] Step S120: Based on the frequency domain deployment mapping table, perform inverse frequency domain transformation to obtain redundant rendering paths, perform timing misalignment analysis on the redundant rendering paths to obtain idle processing windows, and perform interleaving scheduling on the idle processing windows to generate interleaved publishing channels.

[0037] Specifically, redundant rendering paths are obtained by performing inverse frequency domain transformation based on a frequency domain deployment mapping table. The organization, storage location, and dependencies of multi-band rendering components are read from the frequency domain deployment mapping table, which contains deployment schemes for each frequency band data under different platform configurations. An inverse spherical harmonic transformation is performed based on the frequency domain deployment mapping table to convert the frequency domain-represented rendering data back to the spatial domain representation, restoring the original color and lighting information. The original function is reconstructed through numerical integration, and the previously calculated spherical harmonic coefficients c_lm are used to perform inverse transformation combinations according to each frequency level, where l is the order and m is the degree. Based on the platform adaptation information of the frequency domain deployment mapping table, an independent rendering process is constructed for each target platform, with different platforms selecting different frequency band combinations based on hardware capabilities. The multiple rendering paths generated by the inverse transformation are analyzed to identify overlapping parts and shareable resources between paths; these overlapping and shared parts constitute redundant rendering paths. Based on the hierarchical structure information of the frequency domain deployment mapping table, an execution dependency graph of the rendering paths is established, marking serially executed parts and parallelizable parts. The redundant rendering paths contain complete rendering task sequences, resource requirement descriptions, and timing constraints, forming the basic data structure for optimized scheduling.

[0038] Perform time-series misalignment analysis on redundant rendering paths to obtain idle processing windows. Traverse all rendering tasks in the redundant rendering paths, extracting the temporal attributes of each task, including estimated start time, execution duration, resource usage type, and usage amount. Construct a temporal relationship graph of rendering tasks, where nodes represent tasks, directed edges represent execution order constraints, and edge weights represent time intervals between tasks. Identify the critical path in the temporal relationship graph, i.e., the task sequence that determines the overall completion time, and tasks with time margins on non-critical paths. Apply time-series misalignment techniques to the redundant rendering paths, shifting tasks that originally overlapped in time to achieve staggered execution and reduce resource contention. Establish a resource usage time table R[t][r], where t is the time slice index and r is the resource type index, recording the usage status of various resources in each time slice. Scan the resource usage time table to find consecutive unoccupied time periods, which are the available idle processing windows. Calculate the attribute parameters of each idle processing window, including window start time, duration, available resource types, maximum processing capacity, and suitable task types. The identification results of the idle processing windows form a window list, and each window is labeled with detailed availability information and scheduling constraints.

[0039] In some embodiments, the step of interleaving the idle processing window to generate an interleaved deployment channel includes: performing task hierarchical identification on the idle processing window to generate primary tasks and auxiliary tasks; using the primary tasks to perform time-series balancing processing on the auxiliary tasks to form a balanced task flow; adjusting the balanced task flow through interleaving optimization to generate an interleaved deployment flow; and establishing a channel connection based on the interleaved deployment flow to generate an interleaved deployment channel.

[0040] The idle processing window is hierarchically identified to generate primary and auxiliary tasks. All rendering tasks awaiting scheduling in the idle processing window are examined, and the contribution of each task to the final rendering effect is evaluated. Tasks directly affecting core visual elements of the scene are categorized as primary tasks, including essential processing steps such as basic geometry rendering, main object drawing, and critical lighting calculation. Tasks used to enhance visual effects but not essential are categorized as auxiliary tasks, such as optional optimization steps like high-frequency texture details, environmental effects, and anti-aliasing. Based on the analysis results of the resource usage timetable R[t][r], the availability of different resource types in each idle processing window is evaluated, including the remaining capacity of CPU time slices, GPU computing units, memory cache, and network bandwidth. The resource capacity of each idle processing window is evaluated to determine the number of primary and auxiliary tasks that can be accommodated, establishing a window-task capacity mapping relationship. Execution priorities are assigned to tasks at different levels, with primary tasks receiving the highest priority for priority scheduling, and auxiliary tasks assigned medium to low priority based on their visual contribution. The resource requirement characteristics of each task are recorded, including required CPU time, GPU time, memory usage, and bandwidth consumption, forming a task resource requirement table. The results of task hierarchical identification include task type labels, priority values, resource requirement vectors, and window adaptability scores, which constitute the basic data for scheduling decisions.

[0041] A balanced task flow is formed by balancing auxiliary tasks with primary tasks in a time-series manner. A timeline is established based on the execution sequence of primary tasks, arranged according to logical dependencies and priorities. The usage patterns of various resources during the execution of primary tasks are analyzed, and the utilization curves of resources such as CPU, GPU, memory, and network are plotted. Low periods of resource utilization in primary tasks are identified, and these periods can be used to insert auxiliary tasks to achieve full resource utilization. The temporal distribution characteristics of resource utilization are calculated, including average, peak, trough, and fluctuation amplitude, to assess the balance of resource use. Auxiliary tasks are inserted into the resource idle periods of primary tasks, adjusting the insertion timing to stabilize the overall resource utilization and avoid drastic fluctuations. A time buffer is set between primary and auxiliary tasks to address the uncertainty of task execution time and prevent cascading effects caused by task delays. By repeatedly adjusting the task timing, a balanced task flow with alternating execution of primary and auxiliary tasks is formed, achieving a uniform distribution of computing resources over time.

[0042] An interleaved deployment flow is generated by adjusting the balanced task flow through interleaving optimization. The execution pattern of the balanced task flow is analyzed to identify opportunities for further optimization through task interleaving, including the parallelization and pipelining of independent tasks. Sequentially executed independent task segments are reorganized into an alternating execution pattern using a time-slice rotation mechanism, where each task executes for a fixed duration before switching to the next task. For task sequences with data dependencies, a pipelined organization approach is adopted, decomposing tasks into multiple processing stages, with different stages executing in parallel on different processing units. The execution granularity of tasks is adjusted, breaking down long-running tasks into multiple smaller tasks to increase scheduling flexibility and interleaving opportunities. The completion time of each task is predicted based on historical execution data, and task switching timing is optimized based on the prediction results to reduce waiting time. A detailed execution sequence diagram is constructed, marking the start and end times, resource consumption, and data transfer relationships of each task segment. The result of interleaving optimization transforms the linear task flow into a multi-dimensional interleaved execution structure, where multiple task segments intertwine in time and space, forming a complex but efficient interleaved deployment flow.

[0043] Interleaved deployment channels are generated by establishing channel connections based on the interleaved deployment flow. Multiple independent execution channels are created according to the topology of the interleaved deployment flow, each carrying a specific type of task sequence. A dedicated resource pool is configured for each channel, including computing units, cache space, and transmission bandwidth, ensuring resource isolation between channels. Data transmission connections are established between channels, using shared memory regions or message passing interfaces to achieve data exchange. A channel synchronization mechanism is designed, setting synchronization barriers at key execution points to ensure that interdependent tasks maintain the correct execution order across different channels. A dynamic load balancing mechanism is implemented, monitoring the real-time load of each channel and migrating some tasks to less loaded channels when a channel is overloaded. Channel expansion and contraction strategies are configured to dynamically adjust the number of active channels based on the overall load level, optimizing resource utilization efficiency. All channels are integrated to form a unified interleaved deployment channel system, providing standardized task submission interfaces and result aggregation mechanisms, presenting a single high-performance deployment service externally. The interleaved deployment channel encapsulates complex parallel scheduling details internally, providing transparent and efficient multi-channel parallel deployment capabilities for 3DGS scenes.

[0044] Step S130: Extract the phase difference and transmission delay distribution from the interleaved publishing channel, identify network conflict points based on the phase difference, determine the avoidance timing through the transmission delay distribution, and generate a staggered transmission path by combining the network conflict points and the avoidance timing.

[0045] Specifically, phase difference and transmission delay distribution are extracted from the interleaved distribution channels. Multiple parallel execution paths of the interleaved distribution channels are scanned, each carrying a fragmented transmission task of 3DGS scene data. Network packet capture tools are used to capture the timing information of data packets for each channel. Timestamps accurate to the microsecond level are extracted from the transmission logs of the interleaved distribution channels, recording the transmission time, reception time, and intermediate node forwarding time of each data packet. The phase relationship between different channels is calculated: phase difference φ = t1 - t2, where t1 is the data transmission time of channel 1, and t2 is the data transmission time of channel 2. The phase difference reflects the timing offset relationship between channels. Transmission delay is statistically analyzed from the end-to-end measurement data of the interleaved distribution channels, including propagation delay, queuing delay, processing delay, and transmission delay. Probabilistic statistical analysis is performed on the collected transmission delay samples to construct the transmission delay distribution, including the probability density function and cumulative distribution function of the delay. The numerical characteristics of the delay are calculated, including mean, variance, skewness, and kurtosis. The phase difference data is organized into an upper triangular matrix according to channel pairs. The symmetry of the matrix reduces storage space, and the matrix elements accurately record the timing offset between each channel pair. Based on the temporal characteristics of transmission delay distribution, transmission performance data is continuously collected from the real-time monitoring system of the interleaved distribution channel to construct a delay time series. The sampling interval is set to 100 milliseconds to ensure that the dynamic change characteristics of delay are captured.

[0046] In some embodiments, identifying network conflict points based on the phase difference includes: performing time-series feature analysis on the phase difference to form uplink traffic and downlink traffic; using the uplink traffic to perform traffic balancing processing on the downlink traffic to form a balanced data stream; dividing the balanced data stream into regional groups to generate grouped conflict zones; and locking nodes in the grouped conflict zones to determine network conflict points.

[0047] For example, performing time-series feature analysis on the phase difference to form uplink and downlink traffic includes: calibrating the phase difference using a transmission time reference to form a transmission time series; separating delay abrupt change points and delay stable points using the transmission time series; segmenting the flow direction using the delay abrupt change points and the delay stable points as boundaries to generate a flow direction distribution; and performing flow direction classification analysis on the flow direction distribution to obtain uplink and downlink traffic.

[0048] The phase difference is calibrated using a transmission time reference to form a transmission time series. A comprehensive review of the clock system of the phase difference data source is conducted, detecting the clock source type of each node and evaluating clock accuracy and stability indicators. Precise measurement of clock deviations in the distributed system is performed, using round-trip time measurement and statistical analysis methods to calculate the clock deviation values ​​between nodes. A multi-level calibration algorithm is applied to the phase difference data: first, coarse calibration eliminates significant clock jumps, then fine calibration handles minor clock drifts, achieving sub-millisecond accuracy. The calibrated phase differences are sorted by timestamp to construct a normalized transmission time series, including timestamps, phase difference values, and data quality markers. The time series is validated for continuity by calculating the interval distribution of adjacent timestamps, identifying anomalous time jumps and missing data, and reconstructing missing data using interpolation methods. Optimal filtering techniques are applied to the time series to eliminate measurement noise and improve the accuracy of phase difference estimation.

[0049] Delay mutation points and delay stability points are separated using transmission time series analysis. Point-by-point transmission delay is calculated using transmission time series analysis to construct a delay sequence, and the statistical characteristics and dynamic behavior of the delay are analyzed. Delay mutations are identified using an adaptive threshold detection method, constructing a dynamic threshold using local statistics; when the delay change exceeds the threshold, it is marked as a mutation point. A change point detection algorithm systematically identifies all mutation points in the delay sequence, and the algorithm can efficiently handle large-scale data. Delay stability is evaluated using sliding window variance analysis; the window length is adaptively adjusted, and when the variance within the window remains consistently small, it is marked as a stable interval. The dynamic characteristics of the delay are characterized using a state transition model, defining three states: stable state, transition state, and mutation state, and estimating the transition probabilities between states. Multi-scale analysis techniques are used to identify the characteristics of the delay signal; mutation points correspond to local extrema of high-frequency components, and delay stability points correspond to flat regions of low-frequency components. The separated feature point sequence accurately marks all transition moments of the network state.

[0050] The flow direction is segmented using delay abrupt change points and delay stabilization points as boundaries to generate a flow direction distribution. The identified set of delay abrupt change points serves as the segmentation boundary, dividing the transmission time series into multiple independent intervals. Using the stable time periods determined by the delay stabilization points as units, the phase difference distribution characteristics within each time period are analyzed, and the statistical parameters of the phase difference are calculated. The flow direction switching mode is inferred based on the type of abrupt change point; an increase in delay usually corresponds to a transition from light load to heavy load, which may be accompanied by a change in flow direction. The dominant flow direction is determined by the phase difference sign distribution within each segment: a dominant positive phase difference is marked as an uplink segment, a dominant negative phase difference as a downlink segment, and a balanced distribution as a mixed segment. A weighted flow direction distribution is calculated using segment length and flow intensity as weights, reflecting the time proportion of different flow directions. A visualization chart of the flow direction distribution is constructed with time as the horizontal axis, using different labels to distinguish between uplink, downlink, and mixed states, intuitively displaying the time evolution pattern of the flow direction. The generated flow direction distribution contains precise segmentation information, flow direction labels, and statistical characteristics, fully characterizing the time-varying characteristics of bidirectional flow.

[0051] The process involves classifying and analyzing the traffic direction distribution to obtain uplink and downlink traffic. Data is extracted from all time periods marked as uplink, and all data packet information within these periods, including packet size, timestamp, and protocol type, is aggregated to form the raw uplink traffic dataset. The same data extraction process is performed on the time periods marked as downlink to construct the raw downlink traffic dataset, maintaining data integrity and temporal sequence. Deep analysis is conducted on mixed segments, parsing the protocol information of each data packet to accurately determine the data direction based on protocol semantics. Multi-dimensional feature extraction is performed on the separated uplink traffic, calculating key parameters such as average rate, rate variance, burst factor, and packet size distribution. The same dimensional feature parameters are extracted from the downlink traffic to construct feature vectors and establish a feature comparison matrix for uplink and downlink traffic. Temporal correlation analysis is performed on the uplink and downlink traffic, calculating cross-correlation functions to identify request-response time delay patterns. Cross-validation of the classification results is performed, using auxiliary information such as port number and transport layer flags to verify the accuracy of the flow direction determination and ensure classification reliability. The final uplink and downlink traffic datasets not only contain the raw data but also include rich statistical features and metadata annotations.

[0052] A balanced data stream is formed by balancing uplink and downlink traffic. Real-time monitoring data of uplink traffic is used to construct an uplink bandwidth occupancy function, which describes the time-varying demand of uplink traffic on network resources. An uplink traffic prediction model is used to predict future uplink traffic trends through historical data analysis, providing a reference for downlink traffic scheduling. Using a remaining bandwidth calculation method, uplink occupancy and safety reserves are subtracted from the total bandwidth to obtain the bandwidth resources available for downlink traffic. A priority inversion mechanism is used to temporarily adjust the priority of downlink batch transmissions when real-time services are detected in the uplink traffic, ensuring the quality of service for latency-sensitive services. A dual-buffer structure is used to manage uplink and downlink data separately; the uplink buffer uses a priority queue, and the downlink buffer uses a fair queue. The timing of transmission for both types of traffic is coordinated through buffer level control. A closed-loop control method is used to achieve traffic balancing, dynamically adjusting control parameters based on real-time balance deviations. The resulting balanced data stream achieves harmonious coexistence of uplink and downlink traffic in the time dimension, eliminating directional congestion.

[0053] The balanced data flow is divided into regional groups to generate packet conflict zones. Based on the hierarchical structure of the Internet, the balanced data flow is mapped to network regions of different granularities, including access networks, metropolitan area networks, and backbone networks, each with different network characteristics and capacity limitations. Data flows are allocated to corresponding regional groups based on geographical location information. A geographic database of IP addresses is used to determine the source and destination regions of the traffic, and the ratio of intra-regional to inter-regional traffic is calculated. Logical groups are formed based on Autonomous System (AS) numbers. Traffic within the same AS follows a unified routing policy and quality of service (QoS) guarantee, while inter-AS traffic requires passing through peer interconnection points. The traffic density ρ = F / C is calculated for each regional group, where F is the total regional traffic and C is the regional network capacity; the density value reflects the resource utilization level. Regions with traffic density exceeding a safety threshold are identified, indicating intense resource competition within these regions, and are defined as packet conflict zones. The spatial correlation of packet conflict zones is analyzed; traffic exchange between adjacent conflict zones exacerbates congestion. Graph theory methods are used to establish the adjacency relationships of conflict zones. The conflict zone map obtained through regional grouping visually displays the congestion distribution and hotspot locations in the network.

[0054] Node locking is performed to identify network conflict points within packet conflict zones. Within each identified packet conflict zone, a distributed node performance monitoring system is deployed to periodically acquire key metrics such as node processor, memory, and interface queues. The traffic matrix within the conflict zone is decomposed, and traffic exchange matrices between nodes are constructed using traffic data to identify hotspot nodes where traffic converges. Node load is evaluated using comprehensive metrics, considering processor utilization, memory usage, and queue length, to calculate the node load index. In-depth analysis is performed on high-load nodes to examine traffic composition and application types, identifying the main traffic sources causing congestion. Active probing techniques are used to verify the actual processing capacity of nodes by sending test packets of different sizes and measuring the changes in forwarding latency and packet loss rate as a function of load. Node configuration parameters and firmware versions are analyzed to identify configuration defects that may lead to performance bottlenecks, such as improper buffer size settings or routing table exceeding limits. The locked network conflict points form a hierarchical set of conflict points, including backbone routers at the core layer, switches at the aggregation layer, and gateway devices at the access layer. Detailed records of location information, device parameters, performance metrics, and conflict characteristics are maintained for each conflict point.

[0055] Analyzing the transmission delay distribution determines the timing for yielding. The probability density function of the transmission delay distribution identifies the time-varying characteristics of network load, calculating the expected delay values ​​for different time periods. Periods with lower expected delays indicate relatively abundant network resources. Calculating the quantiles of the transmission delay distribution determines a reliable transmission window; the 90th quantile is selected as the criterion, and a period is considered suitable for transmission when this quantile is below a preset threshold. Evaluating the autocorrelation function of the transmission delay analyzes the temporal correlation of the delay; strong correlation indicates predictable delay patterns, which is beneficial for transmission planning. Spectral analysis identifies the periodic components of the delay; the frequencies corresponding to spectral peaks reveal the periodic patterns of network load. A time window segmentation for the transmission delay distribution is established, dividing the day into multiple time slices. The delay characteristics of each time slice are statistically analyzed, forming a refined time-segment-delay mapping table. Time series forecasting methods are used to estimate the delay values ​​for future periods, and the prediction accuracy is evaluated using the root mean square error. The determined set of yielding opportunities includes detailed time interval attributes: start and end times, expected delay range, estimated available bandwidth, and recommended transmission rate, forming complete characteristic data of the yielding opportunity distribution.

[0056] This paper proposes a method to generate off-peak transmission paths by combining network conflict points and avoidance opportunities. A four-dimensional spatiotemporal path planning problem is constructed using the spatial coordinates of network conflict points and the time intervals of avoidance opportunities, aiming to minimize transmission costs while avoiding conflicts. Routing strategies are redesigned based on the topology information of conflict points, using an improved shortest path algorithm that incorporates conflict probability as a key factor in path selection. A transmission scheduling plan is developed based on the distribution characteristics of avoidance opportunities, decomposing large file transfer tasks into multiple sub-tasks, each matched with an avoidance opportunity window. Path selection is dynamically adjusted based on real-time network status information, establishing a path quality scoring mechanism that comprehensively considers multiple dimensions such as latency, packet loss rate, jitter, and available bandwidth. Path decisions are optimized using historical transmission data, extracting path selection patterns from historical records to predict the transmission success rate of different paths. Load balancing is achieved through multiple alternative paths; when the primary path approaches saturation, some traffic is automatically switched to backup paths, realizing dynamic traffic allocation. The generated set of off-peak transmission paths includes complete spatial routing, temporal scheduling, and priority setting information for each path.

[0057] Step S140: Construct a conflict-free deployment flow based on the off-peak transmission path, perform load transmission analysis on the conflict-free deployment flow to form a deployment load gradient, identify load clustering areas based on the deployment load gradient, and deduce the load release channel location based on the load clustering areas to generate a split deployment node.

[0058] Specifically, a conflict-free deployment flow is constructed based on off-peak transmission paths. Traffic allocation is performed using the bandwidth characteristics of these paths, employing a traffic allocation algorithm to rationally distribute total data traffic across all paths, ensuring balanced utilization of each path and preventing it from exceeding capacity limits. Timing constraints require a strict synchronization control mechanism, using clock synchronization to ensure precise coordination of transmission actions across all paths. A rate control mechanism implements fine-grained traffic control, controlling the transmission rate of each path, with the token generation rate dynamically adjusted based on real-time network conditions. Detailed data transmission logs are continuously recorded during the deployment flow's operation, including key performance indicators such as data traffic statistics, processing latency measurements, and queue length monitoring for each network node. A distributed monitoring system sets up detectors at key nodes along the off-peak transmission paths to collect real-time network performance data such as throughput, round-trip time, and packet loss rate. The conflict-free deployment flow constructed through all off-peak transmission paths achieves complete conflict isolation, ensuring reliable transmission of 3DGS scene data in complex network environments through precise spatiotemporal scheduling, while generating rich transmission records and node load information.

[0059] Load propagation analysis is performed on the conflict-free deployment flow to form a deployment load gradient. Node load information in the conflict-free deployment flow is converted into network load metrics; nodes with higher loads correspond to greater network loads. Load information for each node, including key metrics such as data traffic, processing latency, and queue length, is extracted from the data transmission records of the conflict-free deployment flow to construct a node load evaluation system. Similar to data transmission in 3DGS scenes, a busy source server propagates its processing load to downstream nodes, creating a cascading propagation effect of network load. A load propagation model is established for each network node traversed by the conflict-free deployment flow, using the output load of upstream nodes as the input load of downstream nodes, considering the limitations of node processing capacity and buffer capacity constraints. The node load propagation coefficient k = P_out / P_in is calculated, where P_in is the input load and P_out is the output load. The propagation coefficient reflects the amplification or attenuation effect of a node on the load. A hierarchical analysis of the topology of the conflict-free deployment flow is performed, calculating load propagation layer by layer starting from the source node. The load value is updated according to the propagation coefficient at each node. Construct the load field function P(x,y,z), where (x,y,z) are the spatial coordinates of the node, and the function value represents the deployment load intensity at that location. Calculate the load gradient. ,in , , These are the partial derivatives of the load field function P with respect to the x, y, and z coordinates, respectively. The gradient vector points in the direction of the fastest load growth, and the magnitude of the gradient reflects the drastic change in load.

[0060] Identify load clusters based on deployment load gradients. Based on the numerical distribution of the deployment load gradient field, draw a load distribution map using the isoload line method; areas with dense isoload lines indicate drastic load changes. Based on gradient magnitude... A threshold is set; when the gradient magnitude exceeds the threshold, it indicates a significant load accumulation phenomenon in the region. Cluster analysis is performed based on the local maxima of the load field, grouping spatially similar nodes with similar load values ​​into the same load cluster. The total load of each cluster is calculated as Q = ∫∫∫_VP(x,y,z)dV, where V is the volume of the cluster and Q represents the total accumulated load within the region. The causes of load cluster formation are analyzed, including network bottlenecks, node failures, and route convergence, establishing a correspondence between load causes and clustering patterns. The stability of load clusters is assessed; time series analysis is used to determine whether the load is continuously accumulating or fluctuating periodically, with stable high-load areas requiring close monitoring. Based on the load clusters identified by the deployed load gradient, a heatmap of network load distribution is generated, clearly marking key areas requiring load mitigation.

[0061] In some embodiments, the step of generating a load distribution node by reverse deducing the load release channel location based on the load aggregation area includes: performing load gradient tracing from the load aggregation area outwards to form a diffusion path; performing impedance evaluation on the diffusion path to identify low-impedance channels; establishing a load release channel based on the low-impedance channels; and setting up a load distribution node at a key location on the load release channel.

[0062] Load gradient tracing is performed outwards from the load cluster area to form a diffusion path. Starting from the center point of the load cluster area, tracing proceeds along the negative direction of the load gradient, which points in the direction of the fastest load reduction. Similar to 3DGS data flow finding the fastest evacuation channel in an overloaded server, it automatically seeks the direction of the steepest network load decrease. In each tracing step, the gradient vector at the current position is calculated, and the direction of the steepest gradient decrease is selected as the next movement direction, with the step size adaptively adjusted according to the gradient magnitude. All nodes traversed during the tracing process are recorded, forming a complete path from the high-load area to the low-load area, with each path representing a potential load release channel. When tracing reaches an area where the load value is below a set threshold, tracing stops and the end point of the path is marked; the end point area has the capacity to absorb additional traffic. Parallel tracing is performed on multiple starting points to generate multiple diffusion paths, which may converge or diverge at some nodes, forming a tree-like diffusion network. The effectiveness of each diffusion path is evaluated by calculating the path length, number of nodes traversed, total load reduction, and network link quality, and the path with the best load release effect is selected.

[0063] Impedance assessment is performed on diffusion paths to identify low-impedance channels. Impedance is measured on network links along each diffusion path, taking into account the combined effects of transmission delay, bandwidth limitations, packet loss rate, and other factors. The link impedance is calculated as Z = α·D + β / B + γ·L, where D is the transmission delay, B is the available bandwidth, L is the packet loss rate, and α, β, and γ are weighting coefficients. The total impedance of the diffusion path is calculated; the total impedance equals the sum of the impedances of each link segment. A lower total impedance indicates a more suitable path for load shedding. The time-varying characteristics of impedance are analyzed; the impedance of some links may fluctuate over time. Paths with stable and consistently low impedance are selected as reliable load shedding channels. The impedance characteristics of different diffusion paths are compared to identify several paths with the lowest impedance. These low-impedance channels can efficiently dissipate load. Capacity assessment is performed on low-impedance channels to ensure that their carrying capacity meets the load shedding requirements and avoids shifting load to new bottlenecks. The low-impedance channels identified through impedance assessment provide optimal path selection for establishing an efficient load shedding mechanism.

[0064] A load release channel is established based on low-impedance channels. A dedicated load release channel is designed based on the identified low-impedance channels to provide a fast evacuation path for traffic in high-load areas. A resource reservation mechanism is configured along the low-impedance channel to ensure that the channel always maintains sufficient available bandwidth and is not occupied by regular traffic. An access control policy is established for the channel, allowing only traffic from load-concentrated areas to use the channel, ensuring the priority of load release. Traffic shaping devices are deployed at key nodes of the channel to perform rate control and burst management on incoming traffic, preventing new congestion caused by excessive instantaneous traffic. A dynamic adjustment mechanism for the channel is set up to expand or shrink the channel width according to real-time load conditions, increasing resource allocation during peak load periods. Monitoring points are established along the channel to track traffic flow and load release effects in real time, ensuring the normal functioning of the channel. The load release channel established based on low-impedance channels forms a dedicated fast load diversion channel, effectively improving the network's load resistance.

[0065] Deploy traffic splitting nodes at key locations in the load release channel. To address the large-scale concurrent access demands of 3DGS virtual scenes, intelligent traffic splitting devices need to be deployed at key network nodes to achieve automatic traffic allocation and load balancing. A first-level splitting node is set up at the entry point of the load release channel to identify traffic from load-concentrated areas and guide it into the channel. Splitting decision nodes are set up at the channel's forks to intelligently allocate traffic to different load release paths based on the real-time load status of each branch. Buffer nodes equipped with large-capacity caches are deployed at the channel's bottleneck locations to temporarily store bursts of traffic and smooth the load release process. Aggregation nodes are set up at the channel's exit area to reorganize the dispersed load release traffic, ensuring data integrity and order. Intelligent routing functionality is configured for each splitting deployment node, enabling dynamic adjustment of forwarding strategies based on network conditions to optimize load distribution. A collaborative mechanism is established between nodes to share load information and resource status through signaling interaction, achieving distributed load management.

[0066] Step S150: Construct a gradient resource pool based on the split deployment nodes and multi-band rendering components, sort the gradient resource pool by decreasing entropy value to generate a priority sequence, and generate a tiered push strategy based on the priority sequence to form an elastic access matrix.

[0067] Specifically, a gradient resource pool is constructed based on distributed deployment nodes and multi-band rendering components. Detailed resource information for each node is extracted from the distributed deployment node network, including CPU processing power, GPU rendering performance, memory capacity, storage space, network interface bandwidth, and geographical coordinates, to build a complete node resource profile. Multi-band rendering components are precisely classified according to the order of their spherical harmonic expansion: the zero-order component contains ambient lighting information, the first-order component encodes directional light sources, and higher-order components preserve surface details and texture variations. Based on the network topology and physical location distribution of the distributed deployment nodes, a multi-layered resource pool architecture is designed. The core layer deploys high-performance nodes to store critical data, while the edge layer utilizes ordinary nodes to carry auxiliary information. An intelligent mapping mechanism from rendering components to deployment nodes is established, considering factors such as data access frequency, transmission distance, and node load, and using a bipartite graph matching algorithm to optimize the allocation scheme. A hierarchical index structure for the resource pool is constructed, using a B+ tree to organize metadata, supporting fast location and range queries. Index nodes contain attributes such as resource type, storage location, data size, and access permissions. A load balancing strategy for resource access is designed, using a consistent hashing algorithm to distribute requests and avoid the formation of hotspot nodes.

[0068] In some embodiments, the step of generating a priority sequence by sorting the gradient resource pool by decreasing entropy values ​​includes: scanning the information density of the gradient resource pool to obtain an entropy distribution spectrum; identifying information aggregation peaks and information sparsity valleys based on the entropy distribution spectrum; establishing a gradient sorting benchmark using the information aggregation peaks and the information sparsity valleys; and rearranging resources according to the gradient sorting benchmark to generate a priority sequence.

[0069] An information density scan is performed on the gradient resource pool to obtain the entropy distribution spectrum. A comprehensive resource pool scan program is initiated, traversing each storage block of the gradient resource pool, reading the metadata information in the block header and the actual stored rendering data. A block-parallel scanning technique is employed to improve efficiency, dividing the resource pool into multiple scan partitions, each processed by an independent scan thread, with threads exchanging progress information via shared memory. Local information density is calculated at each scan location, statistically analyzing the distribution of different information types within a unit of storage space, including geometric information, texture information, and lighting information. For example, complex architectural model areas have high information density, while open sky areas have low information density; this difference can be quantified through entropy calculation. After calculating the probability distribution of each type of information, the local entropy value is calculated using the information entropy formula H=-Σpilog2(pi), where pi is the probability of the i-th type of information. A larger entropy value indicates a more uniform information distribution and higher uncertainty. A sliding window algorithm is used to calculate continuous local information entropy, with the window size adaptively adjusted based on the spatial correlation of the data; large windows are used for areas with strong correlation, and small windows are used for areas with drastic changes. The calculated entropy values ​​are organized according to the three-dimensional coordinates of the resource pool to construct a spatial data structure for the entropy value distribution spectrum, supporting fast spatial queries and neighborhood analysis. Multi-resolution analysis is performed on the entropy value distribution spectrum, using wavelet transform to decompose entropy changes at different scales. The coarse scale reflects the global distribution trend, while the fine scale captures local anomalies.

[0070] Information clustering peaks and information sparsity valleys are identified based on entropy distribution spectra. An adaptive peak detection algorithm is implemented based on the entropy distribution spectrum, using a local maximum search method. The entropy values ​​of each point are compared within its neighborhood, and a peak is marked as a candidate peak when the entropy value of the center point is strictly greater than that of all its neighbors. Hierarchical clustering analysis is performed on the candidate peaks to merge spatially adjacent peaks with similar entropy values, forming information clustering regions with a certain spatial range. The significance index of each peak is calculated, including peak height (difference from the surrounding average entropy value), peak width (spatial range of the high entropy region), and peak isolation (nearest distance to other peaks). Valley locations are searched based on the entropy distribution spectrum, and a watershed algorithm is used to identify local minimum entropy regions, which typically correspond to transitional zones with low information content. The characteristic parameters of the valleys are evaluated, including valley depth (difference from the surrounding entropy value), valley area (coverage of the low entropy region), and valley connectivity (connection relationships with other valleys).

[0071] A gradient ranking benchmark is established using information clustering peaks and information sparsity valleys. Identified information clustering peaks are used as anchor points for ranking, and weight coefficients are determined based on peak significance indicators (height, width, isolation). Peaks with higher significance receive greater weight. The entropy gradient field is calculated starting from each peak position, and the gradient vector is computed using finite difference squares. ,in , , These are the partial derivatives of the entropy function H with respect to the x, y, and z coordinates, respectively. The gradient points in the direction of the fastest entropy increase. The path from peak to trough is traced along the gradient descent direction, and each point on the path receives a decreasing ranking weight based on its relative position within the path. When multiple peaks exist, the impact on resource points is calculated, and the influence range of the peaks is modeled using a Gaussian radial basis function, with the influence intensity decaying exponentially with distance. A gradient ranking scoring function is designed.

[0072] S(p)=Σ i The formula is wi·Hi·exp(-||p-pi||² / 2σi²), where p is the resource location, wi is the importance weight of the i-th peak, Hi is the peak entropy, pi is the peak location, and σi controls the range of influence. When considering the negative impact of valleys, differentiated penalty strengths are designed based on valley characteristic parameters (depth, area, connectivity). A penalty term is introduced into the ranking benchmark, with resources closer to valleys receiving larger negative corrections to ensure low-information regions receive lower priority.

[0073] Resources are rearranged to generate a priority sequence based on a gradient sorting benchmark. A comprehensive sorting score is calculated for each resource unit in the gradient resource pool, taking into account spatial location, information entropy, and gradient characteristics. A quicksort algorithm is implemented to sort all resource units in descending order of their scores. The algorithm employs a three-way partitioning optimization to handle a large number of identical scores. Local optimizations are performed based on the initial sorting, checking logical dependencies between resources to ensure that the priority of a referenced resource is not lower than that of the resources that reference it. Load balancing constraints are introduced to correct the sorting results; when a node carries too many high-priority resources, the priority of some resources is appropriately reduced and distributed to other nodes. Continuous sorting scores are mapped to discrete priority levels, and a clustering algorithm is used to automatically determine the number of levels and boundary points, ensuring that resources within each level have similar importance. An efficient index structure is established for the generated priority sequence, using skip lists to achieve O(logn) lookup complexity, supporting fast priority queries and dynamic updates. The mapping relationship between the original location and new priority of resources is recorded, generating a resource migration plan to guide the actual data reorganization process.

[0074] A tiered push strategy is generated based on a priority sequence to form an elastic access matrix. Differentiated push timing strategies are designed based on the priority sequence, with the highest priority resources using a pre-push mechanism, proactively transmitted to the edge cache before user requests. For the 3DGS virtual exhibition hall application, high-precision models of core exhibits are pre-pushed to user devices to ensure a smooth interactive experience. Dedicated network resources are configured for different priority levels: critical resources receive a 50% bandwidth guarantee, important resources 30%, ordinary resources 15%, and auxiliary resources 5%. An intelligent push triggering mechanism is established, dynamically determining the push timing by comprehensively considering multiple factors such as user behavior prediction, network idle time, and cache hit rate. A tiered service degradation scheme is designed; when the network is congested, the transmission quality of high-level resources is prioritized, while the bitrate or resolution of low-level resources is appropriately reduced. An elastic access matrix A[u][r][t] is constructed, where the three dimensions represent user type u, resource level r, and time period t, respectively. Matrix elements define specific access strategy parameters. Fine-grained access control rules are configured, including concurrent connection limits, transmission rate limits, and retry constraints, to prevent excessive resource consumption. The resulting elastic access matrix maximizes resource utilization efficiency through precise resource scheduling and flexible access control.

[0075] Step S160: The elastic access matrix and the off-peak transmission path are topologically fused to generate access convergence points. A unique link identifier is generated based on the access convergence points to complete the one-click publishing and link generation of the virtual space application.

[0076] In some embodiments, the step of topologically fusing the elastic access matrix with the off-peak transmission path to generate an access convergence point includes: extracting the access weight distribution of the elastic access matrix and the flow weight distribution of the off-peak transmission path; performing weight coupling on the access weight distribution and the flow weight distribution to generate a composite weight field; identifying a weight convergence center in the composite weight field; and labeling the weight convergence center as the access convergence point.

[0077] The access weight distribution of the elastic access matrix and the flow weight distribution of the off-peak transmission path are extracted. The internal structure of the elastic access matrix is ​​analyzed, comprehensively considering multi-dimensional attributes such as access frequency, priority weight, and historical access volume for each matrix element. An access weight function V_access=Σi(fi·wi·exp(-λ·ti)), where fi is the frequency of the i-th access, wi is the corresponding priority weight, ti is the time decay factor, and λ is the decay coefficient. The discrete matrix element weight values ​​are extended to a continuous space using a three-dimensional interpolation algorithm. A radial basis function (RBF) interpolation method is adopted, and a Gaussian kernel function is selected as the basis function to ensure the smoothness and continuity of the weight field. Dynamic traffic data, including real-time traffic rate, cumulative transmission volume, instantaneous peak value, and average load, are extracted from the off-peak transmission path. A sliding time window is used to calculate various statistical values. A flow weight function, V_flow = ∫(ρ(s)·v²(s) / 2)ds, is defined and integraled along path s, where ρ(s) is the flow density at point s on the path, and v(s) is the transmission rate at that point. The integral result represents the cumulative kinetic energy of the entire path. The flow weights of multiple off-peak transmission paths are spatially superimposed. At path intersections, the vector composition principle is used, considering the directionality of the flow, and the cosine theorem is used to calculate the composite weight. Filtering is applied to preprocess the two weight distributions, adjusting the processing parameters based on the local weight gradient. Regions with large gradients retain details, while regions with small gradients are smoothed. Access weight reflects user demand intensity, while flow weight reflects network transmission capacity. The combination of these two factors determines the optimal access node location for the virtual space application.

[0078] A composite weight field is generated by coupling the access weight distribution and the flow weight distribution. A nonlinear weight coupling model is designed, considering the interaction effect of the two weights. The coupling function is V_total = α·V_access + β·V_flow + γ·V_access·V_flow / (V_access + V_flow + ε), where the third term represents the interaction, and ε is a small positive number to prevent division by zero. The coupling coefficients α, β, and γ are determined using machine learning methods. Historical operating data is collected to train a neural network model. The input is the system state parameters, and the output is the optimal coupling coefficients. The backpropagation algorithm is used to optimize the network weights. Field theory analysis is introduced into the weight coupling process. The access weight is analogous to an electric potential field, the flow weight to a magnetic field, and the composite weight field is similar to the superposition of electromagnetic fields, following similar superposition principles and boundary conditions. Singularities in the weight coupling are handled. When both weights are simultaneously zero or infinite at a certain point, regularization techniques are used to eliminate singularities by adding small perturbations. Key features of the composite weight field are calculated, including the gradient. divergence curl These quantities describe the directionality, source-sink distribution, and circulation characteristics of the weight field, respectively. The spectral characteristics of the composite weight field are analyzed using Fourier transform; low-frequency components reflect the global weight distribution trend, while high-frequency components reflect local weight fluctuations. Frequency filtering is then performed as needed.

[0079] Identify weight convergence centers in a composite weight field. An optimization algorithm searches for the convergence location with the lowest weight in the composite weight field. The algorithm starts from multiple points and moves gradually along the direction of decreasing weight to find the optimal point. In the deployment network of virtual space applications, nodes with high user access density, smooth network transmission, and fast service response naturally form weight convergence centers. These locations have the advantage of handling a large number of user accesses. The stability of each found convergence location is verified by analyzing the changing trends of the weights around the location. Only locations where the weights of the surrounding points are all higher than the center point are confirmed as stable convergence centers. The influence range of each convergence center is divided, and the boundary area of ​​its weight influence is determined. The center's attraction capacity indicators are calculated, including key parameters such as the degree of weight advantage, the size of the influence range, and the traffic absorption capacity. When the influence ranges of different convergence centers overlap, their respective power boundaries are analyzed to ensure that each virtual space access request is directed to the most suitable service node. The identified aggregation centers are categorized and classified into core aggregation centers, regional aggregation centers, and edge aggregation centers based on their weight and influence. Core centers handle the main 3DGS scene publishing tasks, regional centers are responsible for user access in specific regions, and edge centers provide auxiliary load balancing. Through systematic aggregation center identification and classification, a hierarchical access node layout is provided for virtual space applications, ensuring that users' one-click publishing requests can be efficiently aggregated to the optimal service node.

[0080] The weight convergence centers are identified as convergence points. A comprehensive evaluation is performed on each identified weight convergence center, extracting its complete set of feature parameters in the composite weight field, including its three-dimensional spatial coordinates (x, y, z), weight value V, and gradient vector. The Hessian matrix eigenvalues ​​(λ1, λ2, λ3) are used to determine the stability type of convergence points, the weighted basin volume, and the attraction flux, among more than ten other indicators. Based on the weight characteristics and spatial distribution of the convergence centers, a fuzzy C-means clustering algorithm is used for functional classification, dividing the centers into core convergence points, regional convergence points, edge convergence points, and backup convergence points, with each category assigned a different resource allocation strategy. Each access convergence point is precisely labeled in the network topology view, using different graphical symbols to represent different types of convergence points. The symbol size reflects the convergence strength, the color indicates the current load status, and the lines represent the relationships between convergence points. A redundancy configuration scheme is designed for access convergence points, with 2-3 backup convergence points configured for each primary convergence point. The backup points complement the primary node in terms of geographical location, network path, and resource configuration, ensuring rapid switching in case of single-point failure.

[0081] In some embodiments, generating a unique link identifier based on the access convergence point includes: encoding and compressing the spatial coordinates of the access convergence point to generate a location fingerprint; encoding the location fingerprint using user publishing behavior to generate a spatiotemporal identifier seed; performing a hash transformation using the spatiotemporal identifier seed to form an anti-collision identifier; and triggering automatic resource configuration based on the anti-collision identifier to generate a unique link identifier.

[0082] Location fingerprints are generated by encoding and compression based on the spatial coordinates of convergence points. High-precision spatial locations are extracted from the attribute data of these convergence points, represented using the WGS84 coordinate system, with longitude ranging from -180° to 180°, latitude from -90° to 90°, and altitude based on sea level, achieving centimeter-level accuracy. The spherical coordinates are converted to three-dimensional Cartesian coordinates using an Earth ellipsoid model for precise transformation, considering the influence of Earth's oblateness. The conversion formula involves parameters such as the ellipsoid's semi-major axis, semi-minor axis, and first eccentricity. Space-filling curve technology is applied to map the three-dimensional coordinates into a one-dimensional sequence using Z-order curves (Morton coding). Dimensionality reduction is achieved through bit interleaving, maintaining the proximity of spatially adjacent points in the one-dimensional sequence. The Morton coding results are further compressed using adaptive arithmetic coding technology, constructing a probabilistic model based on the statistical distribution characteristics of the coordinates to achieve a compression rate close to the information entropy limit. Error-correcting codes are embedded in the compressed data, using Reed-Solomon coding to provide error detection and correction capabilities, allowing recovery of original coordinate information even if some data is corrupted. The compressed coordinate data is combined with other attributes of the convergence point (such as network address and port number), and a fixed-length location fingerprint is generated using the cryptographic hash function SHA3-256.

[0083] Location fingerprints are encoded using user publishing behavior to generate spatiotemporal identifier seeds. User operational behavior characteristics during the application publishing process in virtual space are recorded, including key behavioral indicators such as duration of stay on the publishing interface, number of preview operations, frequency of parameter adjustments, and hesitation time before confirmation. For example, if a user stays on the publishing interface for 42.3 seconds, clicks the preview button 5 times, adjusts permission settings 3 times, and hesitates for 8.7 seconds before confirming the publication, a behavioral feature vector is formed [42.3s, 5, 3, 8.7s]. This user behavioral feature vector is encoded as a digital fingerprint, and the impact of key operations is highlighted through temporal weight allocation, with frequent operations and long stays receiving higher weight coefficients. A fusion algorithm for behavioral and location fingerprints is designed, interleaving the temporal pattern of user operations with spatial location information to ensure that each user's publishing habits form unique identifiers. Operational stability analysis is introduced during the fusion process; users with similar behavioral patterns in consecutive publications receive higher stability scores, while occasional operational anomalies are automatically filtered and corrected. Behavioral pattern normalization is applied to standardize the fused data, eliminating the impact of individual operation speed differences and ensuring the comparability of behavioral fingerprints from different users. The generated spatiotemporal identifier seed is 256 bits long, of which 128 bits are location information and 128 bits are user behavior fingerprint. It contains the precise spatial coordinates of the access convergence point and the behavioral characteristics of the user's publishing process, which improves user adaptability and service accuracy.

[0084] A collision-resistant identifier is formed by hashing a spatiotemporal identifier seed. A combination of highly secure cryptographic hash functions is selected. First, SHA3-512 is used to perform a preliminary hash on the spatiotemporal identifier seed, generating a 512-bit intermediate result. A key derivation function (KDF) is applied to the intermediate result, using the PBKDF2 algorithm for multiple iterations, with each iteration enhancing resistance to brute-force attacks. A different salt value is added to each stage of the hash chain. The salt value is generated by a hardware security module (HSM) and is 128 bits long, ensuring that even the same input produces different outputs. A multi-algorithm parallel hashing strategy is implemented, using SHA3-512, BLAKE2b-512, and Whirlpool algorithms simultaneously. The three results are mixed at the bit level, ensuring that weaknesses in any one algorithm do not affect the final result. The mixed hash value is formatted, with the first 256 bits used as the primary collision-resistant identifier, the middle 128 bits as version control information, and the last 128 bits as an integrity checksum. The Cuckoo Filter is used to check if the generated identifier already exists. Compared to the Bloom Filter, the Cuckoo Filter supports deletion operations and has a lower false positive rate. When a collision is detected, an incrementing random number is automatically appended to the end of the spatiotemporal identifier seed, and the hashing process is re-executed until a unique identifier is generated.

[0085] The system triggers automatic resource configuration based on anti-collision identifiers to generate unique link identifiers. These identifiers are sent to the resource management system, triggering an automated resource configuration process. The system parses the metadata in the identifiers to determine the required resource types and quantities. The resource configuration engine optimizes resource allocation using a binning algorithm based on the current resource pool status, minimizing resource fragmentation and improving overall utilization while meeting demand. Virtualized containers are created for the allocated resources, and container orchestration technology automatically deploys the required runtime environment, including the operating system, dependency libraries, and configuration files—all completed within seconds. A mapping relationship between identifiers and resource instances is established, using a distributed key-value database to store the mapping information. The database employs a multi-replica mechanism to ensure high availability, requiring confirmation from a majority of replicas for successful writes. Access control policies are configured for the unique link identifiers, using an attribute-based access control (ABAC) model to dynamically determine permissions based on the visitor's identity, role, time, location, and other attributes. A digital signature is embedded in the link identifier, using the Elliptic Curve Digital Signature Algorithm (ECDSA) to sign the identifier and associated metadata. The public key is distributed through a PKI system. Through this complete automatic resource configuration and identifier generation process, a user-accessible network link is generated, enabling one-click publishing.

[0086] To implement the one-click publishing and link generation method for a virtual space application corresponding to the above method embodiments, in order to achieve the corresponding functions and technical effects. See also Figure 2 , Figure 2 This diagram illustrates a structural block diagram of a one-click publishing and link generation system 200 for a virtual space application according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The one-click publishing and link generation system 200 for a virtual space application according to an embodiment of this application includes:

[0087] The scene parsing module 201 is used to receive the edited 3DGS virtual scene data, perform Gaussian spherical harmonic decomposition on the 3DGS virtual scene data to generate multi-band rendering components, and construct a frequency domain deployment mapping table through the multi-band rendering components.

[0088] The intelligent scheduling module 202 is used to perform inverse frequency domain transformation based on the frequency domain deployment mapping table to obtain redundant rendering paths, perform time-series misalignment analysis on the redundant rendering paths to obtain idle processing windows, and perform interleaving scheduling on the idle processing windows to generate interleaved publishing channels.

[0089] The conflict prevention module 203 is used to extract the phase difference and transmission delay distribution from the interleaved publishing channel, identify network conflict points based on the phase difference, determine the avoidance timing through the transmission delay distribution, and generate a staggered transmission path by combining the network conflict points and the avoidance timing.

[0090] The load balancing module 204 is used to construct a conflict-free deployment flow based on the off-peak transmission path, perform load transmission analysis on the conflict-free deployment flow to form a deployment load gradient, identify load clustering areas based on the deployment load gradient, and deduce the load release channel position based on the load clustering areas to generate a split deployment node.

[0091] The resource optimization module 205 is used to construct a gradient resource pool based on the traffic splitting deployment node and the multi-band rendering component, sort the gradient resource pool by decreasing entropy value to generate a priority sequence, and generate a tiered push strategy to form an elastic access matrix based on the priority sequence.

[0092] The link building module 206 is used to perform topological fusion of the elastic access matrix and the off-peak transmission path to generate access convergence points, and generate unique link identifiers based on the access convergence points to complete the one-click publishing and link generation of virtual space applications.

[0093] The aforementioned one-click publishing and link generation system 200 for virtual space applications can implement the one-click publishing and link generation method for virtual space applications described in the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0094] The above description is only a part or preferred embodiment of this application. Neither the text nor the drawings should limit the scope of protection of this application. All equivalent structural transformations made using the content of this application's specification and drawings under the overall concept of this application, or direct / indirect applications in other related technical fields, are included within the scope of protection of this application.

Claims

1. A one-click publish and link generation method for virtual space applications, characterized by, The method comprises the following steps: receiving the edited 3DGS virtual scene data, performing Gaussian spherical harmonic decomposition on the 3DGS virtual scene data to generate multi-frequency band rendering components, and constructing a frequency domain deployment mapping table based on the multi-frequency band rendering components; based on the frequency domain deployment mapping table, performing frequency domain inverse transformation to obtain a redundant rendering path, performing time sequence misplacement analysis on the redundant rendering path to obtain an idle processing window, and performing interleaving scheduling on the idle processing window to generate an interleaving publishing channel, wherein the interleaving publishing channel is integrated by multiple independent execution channels created according to an interleaving deployment flow topology structure, each channel carries a specific type of task sequence and is configured with a dedicated resource pool; extracting phase difference and transmission delay distribution from the interleaving publishing channel, identifying network conflict points based on the phase difference, determining avoidance opportunities through the transmission delay distribution, and generating staggered transmission paths in combination with the network conflict points and the avoidance opportunities; based on the staggered transmission paths, constructing a conflict-free deployment flow, performing load conduction analysis on the conflict-free deployment flow to form a deployment load gradient, identifying a load aggregation area according to the deployment load gradient, and generating a shunt deployment node based on the load aggregation area to deduce the position of a load release channel; based on the shunt deployment node and the multi-frequency band rendering components, constructing a gradual resource pool, performing entropy value decreasing sorting on the gradual resource pool to generate a priority sequence, generating a step-by-step pushing strategy according to the priority sequence, and forming an elastic access matrix; topologically fusing the elastic access matrix and the staggered transmission paths to generate an access convergence point, wherein the access convergence point is a node formed by calibrating and functionally classifying a weight convergence center in a composite weight field, each category is given a different resource configuration strategy, a unique link identifier is generated based on the access convergence point, and one-key publishing and link generation of a virtual space application are completed.

2. The method of claim 1, wherein, The method comprises the following steps: based on the 3DGS virtual scene data, evaluating rendering load to determine the expansion order, wherein the rendering load includes Gaussian distribution density, spatial dimension coupling, and non-uniformity of frequency spectrum energy; setting spherical harmonic decomposition parameters according to the expansion order; using the spherical harmonic decomposition parameters to perform frequency domain expansion on the 3DGS virtual scene data to generate multi-frequency band rendering components.

3. The method of claim 1, wherein, The method comprises the following steps: performing task layering identification on the idle processing window to generate main tasks and auxiliary tasks; using the main tasks to perform time sequence balancing processing on the auxiliary tasks to form a balanced task flow; performing interleaving optimization adjustment on the balanced task flow to generate an interleaving deployment flow; based on the interleaving deployment flow, establishing channel connection to generate an interleaving publishing channel.

4. The method of claim 1, wherein, The method comprises the following steps: performing time sequence feature analysis on the phase difference to form uplink traffic and downlink traffic; using the uplink traffic to perform traffic balancing processing on the downlink traffic to form a balanced data flow; dividing the balanced data flow into regional groups to divide a grouping conflict area; locking nodes in the grouping conflict area to determine network conflict points.

5. The method of claim 1, wherein, The generating a shunt deployment node based on the reverse derivation of the load release channel position of the load aggregation area comprises: forming a diffusion path by performing load gradient tracking from the load aggregation area to the periphery; identifying a low impedance channel by performing impedance evaluation on the diffusion path, wherein the low impedance channel is a path with the minimum total impedance identified after impedance measurement on each network link on the diffusion path, and the impedance is a comprehensive measurement value of transmission delay, bandwidth limitation and packet loss rate; establishing a load release channel based on the low impedance channel; setting a shunt deployment node at a key position of the load release channel.

6. The method of claim 1, wherein, The generating a priority sequence by performing entropy value decreasing sorting on the gradual resource pool comprises: performing information density scanning on the gradual resource pool to obtain an entropy value distribution spectrum; identifying information aggregation peaks and information sparse valleys based on the entropy value distribution spectrum; establishing a gradient sorting benchmark by using the information aggregation peaks and the information sparse valleys; generating a priority sequence by performing resource rearrangement according to the gradient sorting benchmark.

7. The method of claim 1, wherein, The generating an access convergence point by topologically fusing the elastic access matrix and the staggered transmission path comprises: extracting an access weight distribution of the elastic access matrix and a flow weight distribution of the staggered transmission path; performing weight coupling on the access weight distribution and the flow weight distribution to generate a composite weight field; identifying a weight convergence center in the composite weight field; labeling the weight convergence center as an access convergence point.

8. The method of claim 1, wherein, The generating a unique link identifier based on the access convergence point comprises: performing encoding compression based on the spatial coordinates of the access convergence point to generate a location fingerprint; generating a space-time identification seed by performing user publishing behavior coding on the location fingerprint; forming a collision-resistant identifier by performing hash hash transformation using the space-time identification seed; triggering resource automatic configuration based on the collision-resistant identifier to generate a unique link identifier.

9. The method of claim 4, wherein, The forming uplink traffic and downlink traffic by performing timing feature analysis on the phase difference comprises: performing transmission time benchmark calibration on the phase difference to form a transmission time sequence; separating a delay mutation point and a delay stable point through the transmission time sequence; segmenting flow directions by taking the delay mutation point and the delay stable point as boundaries to generate a traffic direction distribution; obtaining uplink traffic and downlink traffic by performing flow direction classification analysis on the traffic direction distribution.

10. A one-click publish and link generation system for a virtual space application, characterized by, The scene analysis module is configured to receive 3DGS virtual scene data edited and complete, perform Gaussian spherical harmonic decomposition on the 3DGS virtual scene data to generate multi-band rendering components, and construct a frequency domain deployment mapping table through the multi-band rendering components. The intelligent scheduling module is configured to perform frequency domain inverse transformation based on the frequency domain deployment mapping table to obtain a redundant rendering path, perform timing dislocation analysis on the redundant rendering path to obtain an idle processing window, perform interleaving scheduling on the idle processing window to generate an interleaving publishing channel, and integrate a plurality of independent execution channels created according to an interleaving deployment flow topology to form the interleaving publishing channel, wherein each channel carries a specific type of task sequence and is configured with a dedicated resource pool. ​ The conflict prevention module is configured to extract a phase difference and a transmission delay distribution from the interleaved publishing channel, identify a network conflict point based on the phase difference, determine an avoidance opportunity through the transmission delay distribution, and generate a staggered transmission path in combination of the network conflict point and the avoidance opportunity. The load diversion module is configured to construct a conflict-free deployment flow based on the staggered transmission path, perform load conduction analysis on the conflict-free deployment flow to form a deployment load gradient, identify a load aggregation area according to the deployment load gradient, and generate a shunt deployment node based on the load aggregation area and a load release channel position. The resource optimization module is configured to construct a gradual resource pool based on the shunt deployment node and the multi-frequency band rendering component, perform entropy value decreasing sorting on the gradual resource pool to generate a priority sequence, generate a stepwise pushing strategy according to the priority sequence, and form an elastic access matrix. The link construction module is configured to perform topological fusion on the elastic access matrix and the staggered transmission path to generate an access convergence point, calibrate and functionally classify a weight convergence center in a composite weight field to form the access convergence point, assign different resource configuration strategies to each category, generate a unique link identifier based on the access convergence point, and complete one-key publishing and link generation of the virtual space application.

Images