Digital twin scene cloud edge-end collaborative rendering method supporting multi-agent collaboration

Abstract

The invention provides a digital twin scene cloud edge-end collaborative rendering method supporting multi-agent collaboration, and the method comprises the steps: carrying out the modeling of a space structure, light source distribution and equipment material reflection parameters at a cloud rendering center, and obtaining a light energy propagation frame; layered illumination area division is carried out on a target industrial scene through a light energy propagation framework to obtain a multi-layer light and shadow range, and a cross-layer light and shadow boundary constraint is generated. And rendering each layer of light and shadow range according to the light and shadow boundary constraint, and adjusting the illumination intensity in real time during rendering to obtain a plurality of light and shadow rendering results of different levels. And fusing the light and shadow rendering results of different levels, and identifying an illumination energy abrupt change region in the fused light and shadow rendering results, so as to correct the illumination energy abrupt change region and obtain a continuous light field rendering picture. And performing consistency evaluation on the continuous light field rendering picture, and feeding back a consistency evaluation result to the cloud to adjust the light energy propagation frame and the light and shadow boundary constraint.

Description

Technical Field

[0001] This disclosure belongs to the field of digital twin technology, and more specifically, relates to a cloud-edge-device collaborative rendering method for digital twin scenarios that supports multi-agent collaboration. Background Technology

[0002] Currently, in scenes with extremely strong local bright sources and large-scale shadow areas, such as hot rolling workshops in steel plants, nighttime operations in port yards, and glass furnace workshops, it is difficult to guarantee the physical continuity of spatial light energy distribution at different rendering levels.

[0003] Especially when the distant view is pre-rendered in the cloud, the mid-range view is rendered in real time from the edge, and the foreground view is rendered in detail by the terminal, there are unavoidable physical breaks in the radiation intensity gradient and reflection attenuation trend of each layer. This makes it easy for the existing technology to render images with light field layering misalignment, which in turn can easily mislead operators in high-precision operation and maintenance visualization to judge the real spatial brightness environment. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to overcome the aforementioned deficiencies and proposes a cloud-edge-device collaborative rendering method for digital twin scenes that supports multi-agent collaboration.

[0005] The present invention adopts the following technical solution.

[0006] The first aspect of this invention discloses a cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent cooperation, the method comprising: The spatial structure, light source distribution, and equipment material reflection parameters of the target industrial scene are obtained, and the spatial structure, light source distribution, and equipment material reflection parameters are modeled in the cloud rendering center to obtain the light energy propagation framework. The target industrial scene is divided into layers of illumination areas using the light energy propagation framework to obtain the light and shadow range of multiple layers and generate cross-layer light and shadow boundary constraints. The light and shadow range of each layer is rendered according to the light and shadow boundary constraints, and the light intensity is adjusted in real time during rendering to obtain multiple light and shadow rendering results of different levels. The lighting and shadow rendering results at different levels are fused, and the regions of abrupt changes in lighting energy in the fused lighting and shadow rendering results are identified in order to correct the regions of abrupt changes in lighting energy and obtain a continuous light field rendering image. A consistency assessment is performed on the continuous light field rendering image to obtain a consistency assessment result, and the consistency assessment result is fed back to the cloud to adjust the light energy propagation framework and light and shadow boundary constraints.

[0007] Furthermore, the process of acquiring the spatial structure, light source distribution, and equipment material reflection parameters of the target industrial scene, and modeling these parameters in a cloud rendering center to obtain a light propagation framework, includes: The original light source intensities of different forms in the light source distribution are obtained, and the original light source intensities are normalized to convert the original light source intensities of different forms into the equivalent irradiance intensity of each irradiation source point in the light source distribution. Based on the equivalent irradiance and the spatial structure of the target industrial scene, a spatial light energy propagation path is constructed, and the difference in irradiance attenuation intensity of the same irradiance source in different spatial light energy propagation paths is extracted, so as to determine the spatial light energy attenuation zone according to the difference in irradiance attenuation intensity.

[0008] Furthermore, the process of acquiring the spatial structure, light source distribution, and equipment material reflection parameters of the target industrial scene, and modeling these parameters in a cloud rendering center to obtain a light propagation framework, also includes: Obtain the device material reflection parameters within the spatial light energy attenuation zone, and calculate the equivalent reflection brightness of the device material surface based on the device material reflection parameters. Map the equivalent reflection brightness to the corresponding spatial unit according to the device material type to obtain the material reflection propagation parameter set. The equivalent irradiance, spatial light energy attenuation partition, and material reflection and propagation parameter set are encapsulated to construct the light energy propagation framework, and the reference brightness level corresponding to each spatial unit in the light energy propagation framework is determined. The reference brightness level is used to divide the light and shadow range of different levels.

[0009] Furthermore, the step of dividing the target industrial scene into layered illumination regions using the light energy propagation framework to obtain multi-layered light and shadow ranges and generating cross-layered light and shadow boundary constraints includes: The light propagation framework is read through the edge rendering collaboration node, and the light and shadow range corresponding to the edge rendering collaboration node is divided into multiple rendering space units to calculate the dominant lighting contribution value of each rendering space unit. The rendering space unit is divided into multiple light and shadow ranges of different levels according to the dominant lighting contribution value, and the width of the light and shadow transition band between adjacent light and shadow ranges is calculated. Calculate the constraint coefficient of each rendering space unit within the width of the light and shadow transition zone, and encapsulate the constraint coefficient with the light and shadow range of the corresponding layer to construct the light and shadow boundary constraint across layers.

[0010] Furthermore, the rendering of each layer of light and shadow range is performed according to the light and shadow boundary constraints, and the light intensity is adjusted in real time during rendering to obtain multiple light and shadow rendering results at different levels, including: Calculate the initial rendering load of the light and shadow range of each layer in each rendering space unit, and transform the light and shadow boundary constraints into layered rendering tasks to be executed based on the initial rendering load; The layered rendering task is read by cloud nodes, edge rendering collaboration nodes and terminal nodes respectively, and initial lighting and shadow rendering is performed in the rendering layers corresponding to the cloud nodes, edge rendering collaboration nodes and terminal nodes respectively, so as to obtain the initial lighting and shadow rendering results of each node after the initial lighting and shadow rendering is performed. Calculate the intensity of the local illumination change trend corresponding to each node in the initial lighting and shadow rendering result, and align the intensity of the local illumination change trend of each node to generate the continuous modulation amount of each node. The continuous modulation amount is used to adjust the intensity of the local illumination change trend in real time, and outputs the corresponding light and shadow rendering results for each level after real-time adjustment.

[0011] Furthermore, the process of fusing the lighting and shadow rendering results at different levels and identifying abrupt changes in lighting energy in the fused lighting and shadow rendering results, in order to correct these abrupt changes in lighting energy and obtain a continuous light field rendering image, includes: The terminal node receives the cloud-based distant view lighting and shadow rendering results and the edge mid-range lighting and shadow rendering results, and extracts the pre-stored local near-view rendering results, so as to map the cloud-based distant view lighting and shadow rendering results, the edge mid-range lighting and shadow rendering results, and the local near-view rendering results into the spatial units within the terminal node's field of view. Calculate the reference brightness before fusion for each spatial unit, and calculate the brightness change gradient between adjacent spatial units based on the reference brightness before fusion, so as to identify the light energy abrupt change region where the brightness change gradient exceeds a set change threshold.

[0012] Furthermore, the process of fusing the lighting and shadow rendering results at different levels and identifying abrupt changes in lighting energy in the fused lighting and shadow rendering results, in order to correct these abrupt changes in lighting energy and obtain a continuous light field rendering image, also includes: Calculate the gradient deviation between the brightness change gradient corresponding to the region of sudden change in illumination energy and the preset target gradient, and generate a continuous correction intensity based on the gradient deviation. The brightness of the reference brightness before fusion is corrected according to the continuity correction intensity and spatial unit order to obtain the corrected output brightness, and the continuous light field rendering image is generated based on the output brightness of all spatial units.

[0013] Furthermore, the process of performing a consistency evaluation on the continuous light field rendered image, obtaining a consistency evaluation result, and feeding the consistency evaluation result back to the cloud to adjust the light energy propagation framework and light and shadow boundary constraints includes: The current field of view is obtained based on the continuous light field rendering image, and the current field of view is divided into multiple partitions to calculate the light and shadow consistency score of each partition. The intensity of light and shadow deviation in each partition is determined by the light and shadow consistency score, and the light propagation framework and light and shadow boundary constraints are corrected according to the intensity of light and shadow deviation to obtain the updated light propagation framework and light and shadow boundary constraints.

[0014] The second aspect of this invention discloses a cloud-edge-device collaborative rendering apparatus for digital twin scenes supporting multi-agent collaboration, used to implement the cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration as described in any of the first aspects, the apparatus comprising: The propagation framework construction module is used to obtain the spatial structure, light source distribution, and equipment material reflection parameters of the target industrial scene, and to model the spatial structure, light source distribution, and equipment material reflection parameters in the cloud rendering center to obtain the light energy propagation framework; The boundary constraint generation module is used to divide the target industrial scene into layers of illumination areas through the light energy propagation framework to obtain the light and shadow range of multiple layers and generate cross-layer light and shadow boundary constraints. The layered lighting and shadow rendering module is used to render the lighting and shadow range of each layer according to the lighting and shadow boundary constraints, and to adjust the light intensity in real time during rendering to obtain multiple lighting and shadow rendering results of different levels. The mutation region correction module is used to fuse the light and shadow rendering results of different levels and identify the light energy mutation regions in the fused light and shadow rendering results, so as to correct the light energy mutation regions and obtain a continuous light field rendering image. The rendering evaluation module is used to perform a consistency evaluation on the continuous light field rendering, obtain a consistency evaluation result, and feed the consistency evaluation result back to the cloud to adjust the light energy propagation framework and light and shadow boundary constraints.

[0015] A third aspect of the present invention discloses a terminal, including a processor and a storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method described in the first aspect.

[0016] A fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0017] The beneficial effects of the present invention are as follows: Compared with the prior art, the present invention has the following advantages: (1) This invention performs unified modeling of the light source distribution, spatial structure, and material reflection characteristics of the target industrial scene in a cloud rendering service center, generates a light energy propagation reference framework covering the entire scene, and synchronously distributes this reference framework to edge nodes and terminal agents, so that rendering on all three ends is performed based on the same light propagation constraint. Then, the edge rendering collaborative node divides the scene into spatial lighting contribution domains according to the light energy propagation reference framework, clarifies the influence range of distant light and shadow, the influence range of mid-range light and shadow, and the influence range of near-range light and shadow, and generates cross-layer light and shadow boundary constraint rules to limit the light effect range and energy distribution boundary of each level of rendering. This enables cloud, edge, and terminal nodes to execute rendering tasks of corresponding levels under light and shadow boundary constraints, and to modulate the local light intensity change trend in real time during the rendering process, so that the light and shadow distribution output by each node remains continuous in the spatial propagation direction. Finally, it achieves physical consistency in radiation gradient, reflection attenuation, and spatial lighting continuity of different levels of rendering output, thereby avoiding light field misalignment after multi-layer rendering fusion and improving the realistic expression capability of the digital twin system in highly dynamic lighting industrial scenes.

[0018] (2) After receiving the distant and mid-range light and shadow rendering results from the cloud through the terminal intelligent rendering agent, the present invention performs step-by-step fusion of light and shadow results at different levels under the constraint of a unified light energy propagation reference frame, and performs continuous correction on the regions of abrupt changes in light energy gradient, thereby generating a final rendered image with a continuous light field distribution. Finally, the final rendered image is evaluated for light and shadow consistency through an adaptive optimization agent, and the evaluation results are fed back to the global scheduling center in the cloud to dynamically adjust the light energy propagation reference frame and light and shadow boundary constraint rules, so that the system can maintain the physical consistency of cross-layer light and shadow distribution in subsequent rendering tasks, further improving the real expression capability of the digital twin system in highly dynamic lighting industrial scenes. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the cloud-edge-device collaborative rendering method for digital twin scenes that supports multi-agent collaboration, provided by the present invention.

[0020] Figure 2 This is a schematic diagram of the structure of the digital twin scene cloud-edge-device collaborative rendering device that supports multi-agent collaboration provided by the present invention. Detailed Implementation

[0021] The present application will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be construed as limiting the scope of protection of the present application.

[0022] like Figure 1 As shown, in one embodiment, a cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration includes the following steps: Step S110: Obtain the spatial structure, light source distribution, and equipment material reflection parameters of the target industrial scene, and model the spatial structure, light source distribution, and equipment material reflection parameters in the cloud rendering center to obtain the light energy propagation framework.

[0023] In some embodiments, the cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration provided by the present invention includes the following steps in step S110: Step S111: Obtain the original light source intensities of different forms in the light source distribution, and normalize the original light source intensities to convert the original light source intensities of different forms into the equivalent irradiance intensities of each irradiation source point in the light source distribution.

[0024] Step S112: Based on the equivalent irradiance intensity and the spatial structure of the target industrial scene, construct the spatial light energy propagation path, and extract the difference in irradiance attenuation intensity of the same irradiance source in different spatial light energy propagation paths, so as to determine the spatial light energy attenuation zone according to the difference in irradiance attenuation intensity.

[0025] In some embodiments, the cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration provided by the present invention further includes the following steps in step S110: Step S113: Obtain the device material reflection parameters within the spatial light energy attenuation partition, and calculate the equivalent reflection brightness of the device material surface based on the device material reflection parameters, so as to map the equivalent reflection brightness to the corresponding spatial unit according to the device material type, and obtain the material reflection propagation parameter set.

[0026] Step S114: The equivalent irradiance, spatial light energy attenuation partition, and material reflection propagation parameter set are encapsulated to construct a light energy propagation framework, and the reference brightness level corresponding to each spatial unit in the light energy propagation framework is determined. The reference brightness level is used to divide the light and shadow range of different levels.

[0027] In a specific embodiment, the cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration provided by the present invention includes steps 1 to 5: Step 1: Construct a unified reference framework for light propagation in industrial scenarios.

[0028] In the cloud rendering service center, the light source distribution, spatial structure, and material reflection characteristics of the target industrial scene are uniformly modeled to generate a light propagation reference framework covering the entire scene. This reference framework is then synchronously distributed to edge nodes and terminal agents, ensuring that rendering on all three ends is performed based on the same light propagation constraint. This includes the following sub-steps: Sub-step 1.1: Construct a set of equivalent irradiance distributions for a unified light source.

[0029] Specifically, in the cloud rendering service center, a unified light source model is first performed on all light-emitting objects in the industrial scene. Because there are point-like high-brightness sources, strip-shaped high-temperature radiation sources, and area lighting sources coexisting in the hot rolling workshop of the steel plant, the night operation area of ​​the port yard, and the glass melting furnace plant, a single lamp model cannot be directly used. Instead, it is necessary to convert the different types of light sources into equivalent irradiance that can be calculated uniformly. That is, the power diffusion attenuation, directional projection attenuation, and medium propagation attenuation are unified into the same irradiance framework, so that different types of light sources can enter the same subsequent calculation chain.

[0030] In this embodiment, the cloud first performs three-dimensional sampling of the scene according to the spatial structure model, with a sampling interval of 0.2-2 meters. Then, it iterates through all visible light sources for each sampling point and sums them to form a unified set of equivalent irradiance distributions of light sources.

[0031] Sub-step 1.2 generates spatial light energy attenuation partitioning results.

[0032] Specifically, after obtaining the equivalent irradiance intensity at each sampling point in the scene, the attenuation partitions of the spatial light propagation path are further constructed by combining the large equipment obstruction, column arrays, crane beams, pipe corridors, and enclosure structures in the industrial space. Instead of simply dividing the space into light and dark areas, we extract the difference in attenuation intensity formed after the same irradiance source propagates through different structural channels; that is, we calculate the comprehensive propagation fidelity coefficient for each spatial unit, expressed as: ; In the formula, The overall propagation fidelity coefficient of a spatial unit is dimensionless. This represents the length of the main propagation path, in meters, and is determined by the traversable propagation distance from the sampling point to the dominant light source. The dielectric attenuation coefficient; This represents the number of times the main occlusion contour is traversed along the propagation path. It is a non-negative integer and is obtained through statistics of occlusion boundaries in the spatial structure model. This represents the occlusion penalty coefficient, with a value ranging from 0.1 to 1. It represents the structural enclosure degree of the space unit, is dimensionless, and is calculated by the ratio of the area of ​​the enclosing surface to the area of ​​the open surface; This represents the enclosure correction coefficient, with a value ranging from 0.05 to 0.5. This formula couples distance attenuation, cumulative occlusion attenuation, and spatial closure attenuation into a single spatial propagation fidelity, thereby dividing industrial scenes into high-fidelity propagation zones, transitional propagation zones, and heavily attenuated propagation zones.

[0033] In this embodiment, the cloud rendering service center adopts a voxel-based spatial partitioning method, with voxel side lengths ranging from 0.5 to 3 meters. For each voxel unit, the path length and number of occlusions are first counted along the main light propagation direction, then the overall propagation fidelity coefficient is obtained in combination with the surrounding enclosure structure, and finally, the light energy attenuation partition is determined based on the numerical range of the overall propagation fidelity coefficient.

[0034] Sub-step 1.3: Construct the set of material reflection propagation parameters.

[0035] Specifically, in industrial scenarios, even with identical spatial propagation conditions, different surfaces will exhibit drastically different brightness responses due to their varying materials. For example, anodized steel plates exhibit more diffuse reflection, polished metal shields more specular reflection, refractory brick surfaces show rough absorption, and glass furnace windows also exhibit transmission overlay. Therefore, it is necessary to establish unified reflection propagation parameters for typical materials within each zone, unifying diffuse base brightness, specular peak brightness, and surface adhesion attenuation into a single apparent brightness expression. This ensures that subsequent rendering across cloud, edge, and endpoint platforms follows the same material optical rules. During execution, a parameter table is first established according to material category, and then the parameter table is mapped to spatial partition units, forming a set of material reflection propagation parameters.

[0036] Sub-step 1.4: Construct a unified reference framework for light propagation in industrial scenarios.

[0037] Specifically, the aforementioned unified light source equivalent irradiance distribution, spatial light energy attenuation partitioning results, and material reflection and propagation parameter set are encapsulated to generate a unified light energy propagation reference framework for use across the cloud, edge nodes, and terminal agents. To enable this framework to directly support subsequent layered rendering task partitioning, the reference brightness level for each spatial unit needs to be further calculated, expressed as: ; In the formula, The reference brightness level base value for a spatial unit, expressed in watts per square meter per steradian. Indicates the material's equivalent reflectance; Indicates the overall transmission fidelity coefficient; This represents the line-of-sight perturbation correction factor, with a value range of 0.05-0.5; This represents the degree of occlusion disturbance between the spatial unit and the main observation channel. It is dimensionless and calculated jointly by the density of the occlusion boundary and the frequency of changes in the observation direction. This formula combines the amount of surface energy that can be brightened, the amount of propagation energy that can be retained, and the amount that can be seen by the observer into the final reference brightness level, so that the three ends have a unified brightness constraint base before performing layered rendering.

[0038] After encapsulation, the cloud distributes the reference framework to edge nodes and terminal proxies according to a hierarchical index of scene number, spatial partition number, and material category number. Edge nodes store a complete region-level copy, while terminal proxies store a clipped copy of the current viewport, thus ensuring that the three endpoints use the same lighting propagation benchmark during subsequent layered rendering, rather than inferring it independently.

[0039] Step S120: Divide the target industrial scene into layered lighting areas using a light energy propagation framework to obtain the light and shadow range of multiple layers and generate cross-layer light and shadow boundary constraints.

[0040] In some embodiments, the cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration provided by the present invention includes the following steps in step S120: Step S121: Read the light propagation framework through the edge rendering co-working node, and divide the light and shadow range corresponding to the edge rendering co-working node into multiple rendering space units to calculate the dominant lighting contribution value of each rendering space unit.

[0041] Step S122: Divide the rendering space unit into multiple light and shadow ranges of different levels according to the dominant lighting contribution value, and calculate the width of the light and shadow transition zone between adjacent light and shadow ranges.

[0042] Step S123: Calculate the constraint coefficient of each rendering space unit within the width of the light and shadow transition zone, and encapsulate the constraint coefficient with the light and shadow range of the corresponding layer to construct cross-layer light and shadow boundary constraints.

[0043] Generally, within the cross-layer light and shadow transition zone, the basic action range of each level can be determined first according to the light and shadow boundary constraint rules. Then, the allowable participation ratio of each level can be calculated according to the relative position of the spatial unit in the transition zone, and this ratio can be combined with the energy upper limit of the corresponding level to form a constraint action coefficient.

[0044] In a specific embodiment, the cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration provided by the present invention includes step 2, which involves performing layered illumination contribution domain division and light and shadow boundary constraint generation. Based on the light propagation reference frame generated in step 1, the edge rendering collaborative node divides the scene into spatial illumination contribution domains, clarifies the influence range of distant, mid-range, and near-range light and shadow, and generates cross-layer light and shadow boundary constraint rules to limit the illumination range and energy distribution boundaries of each rendering layer. This includes the following sub-steps: Sub-step 2.1: Calculate the dominant illumination contribution value of the spatial cell.

[0045] Specifically, the edge rendering collaboration node first reads the unified light propagation reference frame for the industrial scene issued in step 1, and divides the scene area it is responsible for into multiple spatial units according to the regional rendering management method. The side length of each spatial unit is 0.5-2 meters. When there are high-temperature furnace openings, crane lighting strips, glass windows, or large metal reflective surfaces in the scene, a finer-grained division is performed on the local area. Subsequently, the dominant illumination contribution value of each spatial unit to the current observation area is calculated to determine whether the spatial unit should be mainly attributed to the far-field illumination layer, the mid-field illumination layer, or the near-field illumination layer.

[0046] In this embodiment, the expression for the dominant illumination contribution value of the spatial unit is: ; In the formula, This represents the dominant illumination contribution of a spatial unit to the current observation area, expressed in watts per square meter per steradian. The reference luminance level is expressed in watts per square meter per steradian. This represents the path attenuation coefficient, expressed in meters, with a value range of 0.01-0.3. This represents the distance from the center of the spatial unit to the current observation center, in meters. Indicates the angle between the normal of the main reflecting surface of the spatial unit and the viewing direction; This represents the structural occlusion correction factor, with a value range of 0.1-1; It represents the occlusion level on the observation path and is a dimensionless quantity, calculated based on the number of occluded surfaces, occlusion thickness, and continuous occlusion length.

[0047] During execution, edge nodes calculate the dominant illumination contribution value for each spatial unit, and form a set of dominant illumination contribution values ​​for each spatial unit. For moving viewpoints, the calculation is repeated every 0.1-1 seconds to keep the contribution domain division synchronized with the actual viewpoint.

[0048] Sub-step 2.2: Divide the initial layered illumination contribution domain.

[0049] Specifically, after obtaining the dominant lighting contribution value of each spatial unit, the edge rendering collaboration node needs to further divide these spatial units into far-field lighting contribution domains, mid-field lighting contribution domains, and near-field lighting contribution domains. This cannot be simply layered by distance, because in a steel mill's hot rolling mill or a glass melting furnace, a distant, bright heat source may still have a strong illuminating effect on a nearby shadow area; similarly, nearby metal components may only have a weak effect on a localized area due to severe occlusion.

[0050] In this embodiment, a hierarchical attribution determination value is introduced, with the expression as follows: ; In the formula, The value representing the stratification of a spatial unit is expressed in watts per square meter per steradian. Indicates the dominant light contribution value; This represents the structural closure correction factor, with a value range of 0.05-0.8; It represents the enclosure level of the area where the spatial unit is located. It is a dimensionless quantity and is determined by a combination of enclosure ratio, equipment density ratio and the degree of enclosure of tall components. This represents the local highlighting traction coefficient, with units of watts per square meter per steradian multiplied by meters, and a value range of 0.1-20. For observation distance; This represents the distance stability constant, in meters, and ranges from 0.2 to 3.

[0051] In practice, edge nodes are first sorted according to their hierarchical classification values, and then spatial units are allocated into near-field, mid-field, and far-field lighting contribution domains based on preset proportional thresholds. Typically, the top 10%-20% with the highest hierarchical classification values ​​are assigned to the near-field lighting contribution domain, the middle 30%-50% to the mid-field, and the remainder to the far-field. For areas with high continuity requirements, such as below crane tracks, near glass observation windows, or in front of large heat sources, regional connectivity correction is used to prevent the same physical area from being mechanically cut off to different levels.

[0052] Sub-step 2.3: Determine the cross-layer light and shadow transition zone.

[0053] Specifically, since the initial set of layered lighting contribution domains is formed through numerical determination, abrupt changes in layer edges often occur between the foreground, midground, and background. If these boundaries are directly used for subsequent cloud, edge, and end-level layered rendering, lighting jumps will occur near the layer boundaries, still failing to resolve the cross-layer lighting mismatch problem. Therefore, the edge rendering collaboration node needs to further calculate the width of the lighting transition band between adjacent layers, expressed as: ; In the formula, This indicates the width of the light and shadow transition zone between two adjacent layers, in meters. This indicates the basic transition width, in meters, with a value range of 0.3-5. This represents the contribution difference amplification factor, in meters, with a value range of 0.1-2. It represents the absolute value of the difference in dominant illumination contribution values ​​between the spatial units on both sides of the boundary, in watts per square meter per steradian. This represents the dynamic light source correction factor, with a value range of 0.1-1; It represents the level of dynamic light source disturbance near the boundary. It is a dimensionless quantity and is obtained by statistical analysis based on the frequency of changes in crane moving lights, maintenance lights, or high-temperature opening areas. This represents the structural stability inhibition coefficient, with a value range of 0.05-0.5; It represents the structural uniformity near the boundary and is a dimensionless quantity. The higher the uniformity, the more stable the boundary.

[0054] The transition band automatically expands when the difference in lighting contribution is greater and the dynamic disturbance is stronger, while compressing the width of the transition band in areas with continuous and stable structures. This ensures that the transition band can suppress brightness jumps without excessively encroaching on the actual rendering layer range. In implementation, edge nodes calculate the width of the light and shadow transition band segment by segment along the far-mid-ground boundary and the mid-ground-foreground boundary, and generate cross-layer light and shadow transition band sets of corresponding widths on both sides of the boundary. The spatial units within the transition band are no longer simply assigned to a single layer, but are instead treated as key processing areas when generating subsequent boundary constraints.

[0055] Sub-step 2.4: Construct cross-layer light and shadow boundary constraint rules.

[0056] Specifically, the edge rendering collaboration nodes, based on the cross-layer lighting transition zone set, further formulate boundary constraint rules that can be directly executed by subsequent distributed collaborative rendering. The core of these constraint rules is not to limit a certain layer to rendering only a fixed area, but rather to limit the proportion of lighting effects that each layer is allowed to undertake near the boundary and the range of energy that can be released. To this end, the layer's permissible effect coefficient is calculated for any spatial unit within the transition zone, expressed as: ; In the formula, It represents the permissible action coefficient of a certain level on the spatial unit within the transition zone. It is dimensionless and ranges from 0 to 1. This represents the boundary steepness adjustment coefficient, with a value range of 4-20; This indicates the distance from the spatial unit to the starting boundary of the transition zone, in meters; This represents the width of the corresponding transition zone, in meters. This formula utilizes a continuously monotonically changing function to ensure that the layered effect capability changes smoothly within the transition zone, rather than abruptly changing from fully permitted to fully prohibited at the boundary line. In this way, the foreground, midground, and background layers each have their own controlled participation proportions within the transition zone, providing clear boundary constraints for subsequent layered rendering.

[0057] In execution, the edge nodes generate permissible action coefficient curves for the far-field, mid-field, and near-field layers for each transition zone. These curves are then encapsulated by combining the basic contribution domain number, spatial unit number, and corresponding energy upper limit value for each layer, ultimately yielding cross-layer lighting boundary constraint rules. These rules include at least three components: the basic action range of each layer, the permissible action coefficient of the layer within the transition zone, and the maximum energy overlap limit between layers. After encapsulation, these rules are sent to the cloud, edge, and terminal nodes corresponding to step 3, serving as the direct constraint basis for distributed collaborative lighting modulation rendering execution.

[0058] Step S130: Render the light and shadow range of each layer according to the light and shadow boundary constraints, and adjust the light intensity in real time during rendering to obtain multiple light and shadow rendering results of different levels.

[0059] In some embodiments, the cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration provided by the present invention includes the following steps in step S130: Step S131: Calculate the initial rendering load of the light and shadow range of each layer in each rendering space unit, and transform the light and shadow boundary constraints into layered rendering tasks to be executed based on the initial rendering load.

[0060] Step S132: Read the layered rendering task through the cloud node, edge rendering collaboration node and terminal node respectively, and perform initial lighting and shadow rendering in the rendering layer corresponding to each of the cloud node, edge rendering collaboration node and terminal node to obtain the initial lighting and shadow rendering result of each node after the initial lighting and shadow rendering is executed.

[0061] Step S133: Calculate the intensity of local illumination change trend corresponding to each node in the initial lighting and shadow rendering result, and align the intensity of local illumination change trend of each node to generate the continuous modulation amount of each node.

[0062] Among them, the continuous modulation amount is used to adjust the intensity of local illumination change trends in real time, and outputs the corresponding light and shadow rendering results for each level after real-time adjustment.

[0063] In a specific embodiment, the cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration provided by the present invention includes step 3, which involves multi-node collaborative light and shadow modulation rendering. Cloud, edge, and terminal nodes execute rendering tasks at their respective levels under the light and shadow boundary constraints formed in step 2, and modulate the local illumination intensity change trend in real time during the rendering process to ensure that the light and shadow distribution output by each node remains continuous in the spatial propagation direction. This includes the following sub-steps:

[0064] Sub-step 3.1: Determine the layered rendering execution task.

[0065] Specifically, before actual rendering, the cloud, edge, and terminal nodes first transform the cross-layer lighting and shadow boundary constraint rules generated in step 2 into an executable set of layered rendering tasks. The key here is not simply assigning distant views to the cloud, mid-range views to the edge, and foreground views to the terminal, but rather clearly defining the amount of lighting computation each layer is allowed to handle within its basic scope, the maximum allowed brightness, and its participation ratio in the transition zone. To this end, the initial rendering workload of each layer in each spatial unit is first calculated, expressed as: ; In the formula, This indicates the initial rendering load of a certain level in a certain spatial unit, expressed in watts per square meter per steradian. This represents the permissible action coefficient of this level on this spatial unit. It is dimensionless and ranges from 0 to 1. This represents the dominant illumination contribution value corresponding to the spatial unit, expressed in watts per square meter per steradian. This represents the resource consumption suppression coefficient, with a value range of 0.1-2. This indicates the current rendering resource usage level of the node. It is dimensionless and is calculated from the current graphics processor usage rate, video memory usage rate, and task queue length of the node. This represents the penalty coefficient for frequent level switching, expressed in units of each time, with a value range of 0.05-0.5. This represents the cumulative number of times the layer affiliation of the spatial unit changes within adjacent refresh cycles. Based on the boundary allowable effect, this formula incorporates the node's own carrying capacity and layer switching stability to obtain a numerical expression of how much actual lighting rendering this layer should currently handle. This avoids situations where a layer, theoretically allowed to participate at the boundary, is forcibly pulled in when the current node load is too high or the layer fluctuates frequently, leading to unstable subsequent output.

[0066] In this embodiment, cloud nodes primarily create task entries for low-proportion participating units in the distant view area and its transition zone, edge nodes primarily create task entries for the mid-view area and its adjacent transition zone, and terminal nodes primarily create task entries for high-proportion participating units in the near view area, interactive hotspots, and transition zones. Each task entry includes at least a spatial unit number, a layer number, an initial rendering load, an allowed brightness limit, and a refresh cycle number. Finally, all task entries are packaged to form a layered rendering execution task set.

[0067] Sub-step 3.2 generates the initial lighting and shadow rendering results for each node.

[0068] Specifically, the cloud, edge, and terminal nodes each read the layered rendering execution task set and perform initial lighting and shadow rendering within their respective rendering layers. This initial lighting and shadow rendering refers to the process where each node independently generates a version of lighting and shadow results controlled by boundary constraints based on a unified light energy propagation reference frame and its own layer's task set before cross-node continuous modulation. To ensure that different nodes use the same intensity benchmark in their rendering representations, it is necessary to calculate the initial output brightness for each spatial unit, expressed as: ; In the formula, This represents the initial brightness output of a node to a spatial unit, expressed in watts per square meter per steradian. Indicates the initial rendering load, in watts per square meter per steradian. This represents the overall propagation fidelity coefficient, which is dimensionless. This represents the boundary compressibility coefficient, with a value range of 0.05-1; It represents the normalized proximity of the spatial unit to the boundary of the basic functional area of ​​this layer. It is dimensionless, and the closer it is to the boundary, the larger the value. This represents the local brightness compensation coefficient, with a value range of 0-0.5; This represents the visual attention level of the local area where the spatial unit is located. It is dimensionless and determined comprehensively based on the current viewpoint center distance, interaction hotspot density, and device importance indicators. This formula, based on how much attention a node should handle, combines propagation fidelity, boundary compression, and local attention enhancement to obtain the actual brightness value output by the node to the image. Boundary compression is introduced to prevent layers from over-brightening near the boundaries, leading to excessive superposition during subsequent fusion. Local brightness compensation is introduced to prevent undue brightness collapse in key interaction areas due to excessively strong hierarchical constraints.

[0069] During execution, the cloud calculates the initial lighting and shadows of the distant large area with a relatively long refresh cycle of 0.1-1 seconds; the edge nodes calculate the regional dynamic lighting and shadows with a medium refresh cycle of 0.05-0.5 seconds; and the terminal nodes calculate the lighting and shadows of the near details and interactive areas with a shorter refresh cycle of 0.01-0.1 seconds. After completion, each node forms its own initial lighting and shadow rendering result set.

[0070] Sub-step 3.3: Extract local illumination change trend parameters.

[0071] Specifically, after the initial rendering results of multiple nodes are generated, the final output cannot be directly entered. This is because, although the cloud, edge, and terminal all conform to boundary constraints, their brightness gradient change directions may still differ. For example, the distant layer may be in a region of gradual brightness decrease, while the mid-range layer may experience a sudden surge in local peak brightness due to the presence of local bright reflective surfaces. The terminal's near-field layer may further increase in brightness due to enhanced interactive hotspots. This inconsistency in change trends is the direct root cause of cross-layer lighting mismatch. Therefore, it is necessary to extract the local lighting change trends in the output of each node.

[0072] First, calculate the intensity of the local illumination variation trend of the spatial cell, expressed as: ; In the formula, This indicates the intensity of local illumination variation along the direction of spatial propagation within the spatial unit, expressed in watts per square meter per steradian per meter. , , These represent the initial output brightness of three consecutive adjacent spatial units along the main propagation direction, in watts per square meter per steradian. This represents the center-to-center distance between the first adjacent units, in meters. This indicates the center-to-center distance between the second adjacent units, in meters. This represents the roughness suppression coefficient, with a value ranging from 0.05 to 0.8. This represents the surface roughness level of the material, is dimensionless, and is calculated based on the material reflection and propagation parameters in step 1. This formula uses the average of the brightness change rates of adjacent spatial units to characterize whether the brightness changes gradually or abruptly, and utilizes the surface roughness level to suppress overly sensitive high-frequency fluctuations, making the extracted trend closer to the real lighting transition perceived by the naked eye.

[0073] During execution, each node calculates the intensity of the change trend along both the main light propagation direction and the main observation direction in its own output set. The results from both directions are then combined to form the local illumination change trend parameter corresponding to that node. Subsequently, the cloud, edge, and terminal nodes organize the change trend parameters they extracted into a unified format to form a local illumination change trend parameter set.

[0074] Sub-step 3.4 outputs the collaborative lighting and shadow rendering result after continuous modulation.

[0075] Specifically, after obtaining the local illumination change trend parameters of each node, the edge rendering collaboration node, as the regional coordination core, aligns the local trends of the three types of nodes—cloud, edge, and terminal—and generates continuous modulation amounts for each node. Instead of directly modifying the overall brightness, it focuses on modifying the brightness change amplitude of the transition zone and its adjacent areas, ensuring that the brightness gradient in the spatial propagation direction is seamlessly connected.

[0076] In this embodiment, the expression for the output brightness after continuous modulation is: ; In the formula, This indicates the output brightness after continuous modulation, expressed in watts per square meter per steradian. Indicates the intensity of local light variation trends, with units of watts per square meter per steradian. This represents the trend deviation suppression coefficient, with a value range of 0.2-3; This indicates the intensity of the local illumination variation trend of the node in the spatial unit, expressed in watts per square meter per steradian per meter; It represents the intensity of the target change trend obtained after cross-node comprehensive comparison of the same spatial unit. The units are the same, and it is obtained by the edge node by taking the weighted average of the results of the adjacent layers. This represents the energy compensation coefficient, with a value ranging from 0 to 0.5. This represents the equivalent irradiance corresponding to the space unit. This represents the compensation attenuation coefficient, expressed in meters, and ranging from 0.01 to 0.3. This indicates the distance from the spatial unit to the main action center of this layer, in meters.

[0077] The above formula uses a trend deviation suppression term to compress the brightness output that deviates too much from the target trend, and then uses an energy compensation term controlled by the propagation distance to compensate for the brightness collapse that may occur after compression. This ensures trend alignment and prevents the image from becoming gray, dull, or locally distorted due to excessive compression.

[0078] In this embodiment, the edge node first compares the trend parameters of the three types of nodes to obtain the target change trend intensity of each key spatial unit; then, it sends the target change trend intensity and the corresponding modulation parameters to the cloud, edge, and terminal. Each node recalculates its own modulated output brightness accordingly, forming a continuously modulated collaborative lighting and shadow rendering result set. For dynamic high-brightness source scenes, such as the hot-rolled red steel passage area, the swinging area of ​​the mobile maintenance lamp, or the opening and closing area of ​​the furnace observation window, a higher trend deviation suppression coefficient is used to enhance trend constraint, with a value range of 1-3; the energy compensation coefficient is used to avoid excessive local compensation, with a value range of 0-0.2. For stable lighting scenes, the trend deviation suppression coefficient is appropriately reduced to 0.2-1, and the energy compensation coefficient is increased to 0.1-0.5 to maintain overall brightness fullness.

[0079] Step S140: The light and shadow rendering results of different levels are fused, and the light energy change areas in the fused light and shadow rendering results are identified in order to correct the light energy change areas and obtain a continuous light field rendering image.

[0080] In other words, brightness compression and propagation constraint correction are first performed on the gradient abrupt region, and then energy is replenished according to the material reflection characteristics and spatial propagation conditions. At the same time, the propagation path is gradually reduced and corrected, and finally a light field distribution with continuous brightness and conforming to the law of light energy propagation is formed.

[0081] In some embodiments, the cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration provided by the present invention includes the following steps in step S140: Step S141: Receive the cloud-based distant view lighting and shadow rendering results and the edge mid-range lighting and shadow rendering results through the terminal node, and extract the pre-stored local near-view rendering results, so as to map the cloud-based distant view lighting and shadow rendering results, the edge mid-range lighting and shadow rendering results, and the local near-view rendering results into the spatial units within the terminal node's field of view.

[0082] Understandably, local near-field rendering results focus on the details of lighting and shadows and the response of interactive areas within the terminal's field of view; far-field lighting and shadow rendering results are generated in the cloud and used to express the propagation of low-frequency light over a large area; edge and mid-field lighting and shadow rendering results are used for dynamic lighting and shadow transitions and structural occlusion responses at the regional scale.

[0083] Step S142: Calculate the reference brightness before fusion for each spatial unit, and calculate the brightness change gradient between adjacent spatial units based on the reference brightness before fusion, so as to identify the light energy abrupt change region where the brightness change gradient exceeds the set change threshold.

[0084] Typically, the three layers of lighting results are first aligned to generate a baseline brightness before fusion. Then, the brightness change rate of adjacent cells is calculated unit by unit along the propagation direction. The local brightness gradient distribution is obtained through continuous scanning, and spatial cells with significantly higher change amplitudes than the surrounding average level are selected as regions of abrupt change in illumination energy. The preferred change threshold is 1.5 to 3 times the average brightness gradient of the region, and it can be increased to 2 to 4 times for areas near high-brightness heat sources.

[0085] In some embodiments, the cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration provided by the present invention further includes the following steps in step S140: Step S143: Calculate the gradient deviation between the brightness change gradient corresponding to the region of sudden change in light energy and the preset target gradient, and generate a continuity correction intensity based on the gradient deviation.

[0086] Step S144: The reference brightness before fusion is corrected according to the continuity correction intensity and spatial unit order to obtain the corrected output brightness, and a continuous light field rendering screen is generated based on the output brightness of all spatial units.

[0087] In a specific embodiment, the cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration provided by the present invention includes step 4, which involves cross-layer light and shadow energy fusion and continuity correction. After receiving the distant light and shadow rendering results from the cloud and the mid-range light and shadow rendering results from the edge, the terminal intelligent rendering agent fuses the light and shadow results at different levels step by step under the constraint of a unified light energy propagation reference frame, and performs continuity correction on the regions where the light energy gradient changes abruptly, thereby generating a final rendered image with a continuous light field distribution. This includes the following sub-steps: Sub-step 4.1: Construct cross-layer co-located light and shadow alignment results.

[0088] Specifically, the terminal's intelligent rendering agent first receives the distant and mid-range lighting rendering results from the cloud, and then retrieves the local near-field rendering results. It then maps these three types of results into the same spatial unit system within the terminal's current field of view. The key here is not simply overlaying images, but rather achieving co-location alignment, ensuring that brightness results from different levels but pointing to the same physical location have consistent spatial placement and a consistent order of application. To this end, the terminal calculates the baseline brightness before fusion for each spatial unit, expressed as: ; In the formula, The reference brightness before fusion of the same spatial unit is expressed in watts per square meter per steradian. This indicates the modulated brightness of the distant view in the cloud, measured in watts per square meter per steradian. This indicates the edge-to-mid-field modulation brightness, measured in watts per square meter per steradian. This indicates the near-field modulation brightness of the terminal, expressed in watts per square meter per steradian. Indicates cloud-based brightness weight; Indicates edge brightness weight; The above three weights represent the terminal brightness weights; all three weights are dimensionless and range from 0 to 1.5.

[0089] It should be noted that the above weights are not fixed constants, but are determined comprehensively based on the hierarchical allowable effect coefficient, the current observation distance, and the proximity of the hierarchical boundaries. Generally, the cloud brightness weight in the far-field dominant area is 0.6-1.5, the edge brightness weight is 0-0.6, and the terminal brightness weight is 0-0.3; the edge brightness weight in the mid-field dominant area is 0.5-1.2; and the terminal brightness weight in the near-field dominant area is 0.7-1.5. This formula compresses the modulated brightness of different levels onto a unified brightness benchmark at the same physical location, providing a uniform input for subsequent gradient abrupt change detection.

[0090] During the execution process, the terminal first completes the result registration according to the spatial unit number, and then writes the corresponding weights according to the hierarchical dominance relationship in the current field of view, and finally forms a cross-layer co-located light and shadow alignment result set.

[0091] Sub-step 4.2: Identify regions of abrupt changes in the light energy gradient.

[0092] Specifically, after obtaining the baseline brightness before fusion, the terminal intelligent rendering agent continuously scans the brightness gradient between adjacent spatial units along the main propagation direction and the main observation direction to identify locations where the brightness change rate significantly exceeds the surrounding average level. These locations are often the areas most prone to light field misalignment after cross-layer fusion, such as the boundary between the high-brightness area of ​​hot-rolled steel billets and the shadow area of ​​the surrounding steel structure, the edge of the observation window of the glass furnace, and the edge of the high-mast lighting in the port yard.

[0093] In this embodiment, the expression for the intensity of the abrupt change in the light energy gradient is: ; In the formula, This represents the gradient abrupt change intensity at the target spatial unit, expressed in watts per square meter per steradian per meter. , , These represent the reference brightness of three consecutive spatial units along the selected propagation direction before fusion, both in watts per square meter per steradian. The formula obtains the brightness gradient by dividing the difference between adjacent brightness segments by the spatial distance, and then uses the average of the two gradients to characterize the intensity of local brightness changes, thus identifying abrupt changes in light energy in an engineering-friendly manner.

[0094] In practice, the terminal can perform a combination of row and column scanning. When the gradient abrupt change intensity exceeds a set threshold, the spatial unit and its adjacent units are marked as gradient abrupt change candidate regions. The threshold is preferably 1.5-3 times the average gradient of the surrounding area; for areas near high-brightness heat sources, the threshold is appropriately increased to 2-4 times to avoid misjudging normal strong light transitions as abnormal abrupt changes. After screening, the terminal forms a set of light energy gradient abrupt change regions.

[0095] Sub-step 4.3: Determine the continuity correction parameters.

[0096] Specifically, for each gradient abrupt change region, the terminal intelligent rendering agent does not directly modify the brightness, but first calculates whether the region needs to be compressed, smoothed, or compensated. To do this, it is necessary to establish the deviation between the target gradient and the actual gradient, and generate a continuous correction intensity, expressed as: ; In the formula, The continuity correction intensity of the target spatial unit is expressed in watts per square meter per steradian. This represents the gradient bias suppression coefficient, with a value range of 0.2-3; Indicates the actual gradient abruptness intensity, expressed in watts per square meter per steradian. This represents the gradient intensity of the reference target, with the same units, and is determined by a unified light energy propagation reference frame combined with the normal propagation gradient of the current region. This represents the energy recovery coefficient, with a value ranging from 0 to 0.6. This represents the equivalent irradiance of the space unit; This represents the overall reflectance coefficient of the material corresponding to the spatial unit, with a value ranging from 0.05 to 0.95. This indicates the distance from the spatial unit to the main action center of this layer, in meters; Let π be the mathematical constant pi. This formula uses the first term to suppress excessive gradient deviation and the second term to compensate for energy loss caused by compression when necessary.

[0097] When setting parameters, a larger gradient deviation suppression coefficient indicates that the system tends to strongly suppress gradient abrupt changes; this is suitable for areas with very sharp boundaries, such as the edges of high-brightness windows in glass furnaces. A larger energy recovery coefficient indicates that the system emphasizes brightness recovery; this is suitable for scenarios where the overall brightness cannot collapse, such as large-area lighting areas in port yards. The terminal calculates the continuity correction intensity based on the material type, distance from the main light source, and boundary type of each abrupt change area, forming a continuity correction parameter set.

[0098] Sub-step 4.4 outputs the final continuous light field rendering image.

[0099] Specifically, the terminal intelligent rendering agent performs unit-by-unit brightness correction on the baseline brightness before fusion based on the continuous correction parameter set, focusing on the spatial units in gradient abrupt regions and their adjacent propagation paths to obtain the final output brightness. In the process of generating the final output brightness, the brightness with excessive gradient deviation is first compressed, and then the correction intensity after structural constraint screening is injected into the target unit, so that the final brightness conforms to the unified light energy propagation law and does not produce unrealistic overcompensation in high occlusion and high encirclement regions.

[0100] In this embodiment, the terminal first generates the final output brightness for the units within the mutation zone, and then extends outwards to three adjacent units to perform decreasing corrections to prevent the formation of new secondary mutations caused by only correcting the center and not the periphery. After all spatial units have been corrected, the terminal refreshes and outputs the final continuous light field rendering image according to the current display frame. The light and shadow distribution of the distant, middle, and near scenes in this image remains continuous in the spatial propagation direction, thereby completing the cross-layer light and shadow energy fusion and continuity correction.

[0101] Step S150: Perform a consistency assessment on the continuous light field rendering image, obtain the consistency assessment result, and feed the consistency assessment result back to the cloud to adjust the light energy propagation framework and light and shadow boundary constraints.

[0102] In some embodiments, the cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration provided by the present invention includes the following steps in step S150: Step S151: Obtain the current field of view based on the continuous light field rendering image, and divide the current field of view into multiple partitions to calculate the light and shadow consistency score of each partition.

[0103] Step S152: Determine the intensity of light and shadow deviation in each partition through light and shadow consistency score, and correct the light propagation framework and light and shadow boundary constraints according to the intensity of light and shadow deviation to obtain the updated light propagation framework and light and shadow boundary constraints.

[0104] For example, the light and shadow consistency score can be calculated by dividing the field of view into regions, the region with a persistently low score can be identified and the intensity of the deviation can be quantified, and then the cloud scheduling center can gradually adjust the light propagation reference parameters and light and shadow boundary constraints according to the degree of deviation to achieve iterative optimization.

[0105] In a specific embodiment, the cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration provided by the present invention includes step 5, which involves constructing a lighting consistency feedback and rendering strategy iteration mechanism. The adaptive optimization agent evaluates the lighting consistency of the final rendered image formed in step 4 and feeds the evaluation results back to the cloud-based global scheduling center. This is used to dynamically adjust the light energy propagation reference frame and lighting boundary constraint rules, ensuring that the system continuously maintains the physical consistency of cross-layer lighting distribution in subsequent rendering tasks. This includes the following sub-steps: Sub-step 5.1: Calculate the light and shadow consistency evaluation index.

[0106] Specifically, the adaptive optimization agent first performs a partitioned evaluation of the final continuous light field rendering, dividing the current viewport into four regions: distant region, middle region, near region, and transitional region. It then extracts the brightness uniformity, gradient smoothness, and boundary continuity for each region. To facilitate subsequent unified feedback, the lighting consistency evaluation index for a single spatial unit is first calculated, expressed as: ; In the formula, This represents the light and shadow consistency score of a single spatial unit; This represents the brightness deviation compensation coefficient, with a value range of 0.2-1; This represents the luminance compression factor, with a value ranging from 0.1 to 2. This indicates the final output brightness of the spatial unit, expressed in watts per square meter per steradian. This represents the reference brightness of the spatial unit within a unified light propagation reference frame.

[0107] Sub-step 5.2, regional-level light and shadow deviation feedback.

[0108] Specifically, after obtaining the consistency scores for each region, the adaptive optimization agent further identifies which regions have persistent biases, highlighting continuous distortion areas near cross-layer boundaries while avoiding excessive triggering of overall scheduling due to localized minor unit anomalies. During execution, the bias intensity is determined for the distant, middle, near, and cross-layer transition zones, and sorted from high to low bias intensity to form a regional-level light and shadow bias feedback set.

[0109] Sub-step 5.3 generates rendering strategy correction values.

[0110] Specifically, after receiving the regional-level lighting and shadow deviation feedback set, the cloud-based global scheduling center generates the correction amount required for the next round of rendering for regions with high deviations. Instead of directly changing all parameters, it corrects the light propagation reference frame parameters in step 1 and the lighting and shadow boundary constraint parameters in step 2. When a region continues to show deviations, the correction intensity of its propagation reference parameters and boundary constraint parameters is gradually increased; if the deviation is only occasional, a time smoothing term is used to prevent frequent system fluctuations.

[0111] Sub-step 5.4: Update the light energy propagation reference frame and the light and shadow boundary constraint rules.

[0112] Specifically, the cloud-based global scheduling center, based on the rendering strategy correction set, quantitatively updates the reference brightness baseline value and propagation attenuation-related parameters in step 1, as well as the transition band width and hierarchical allowable action coefficient in step 2, and synchronously distributes the update results to edge nodes and terminal agents. The update employs a small-step correction and round-by-round approximation approach, meaning that each round only adjusts the parameters corresponding to the deviation area by 2%-15%, avoiding excessively large corrections that could lead to new brightness faults.

[0113] After the update is completed, a new light energy propagation reference framework and new light and shadow boundary constraint rules are formed for the next round of calls. This forms a closed-loop mechanism of evaluation, feedback, correction and re-rendering, so that the system can maintain the physical consistency of cross-layer light and shadow distribution in subsequent rendering tasks.

[0114] The following describes the cloud-edge-device collaborative rendering device for digital twin scenes that supports multi-agent collaboration, provided by the present invention. The cloud-edge-device collaborative rendering device for digital twin scenes that supports multi-agent collaboration described below can be referred to in correspondence with the cloud-edge-device collaborative rendering method for digital twin scenes that supports multi-agent collaboration described above.

[0115] like Figure 2 As shown, in one embodiment, a cloud-edge-device collaborative rendering device for digital twin scenes that supports multi-agent collaboration includes a propagation framework construction module, a boundary constraint generation module, a layered lighting and shadow rendering module, a mutation region correction module, and a rendering image evaluation module.

[0116] The propagation framework construction module is used to obtain the spatial structure, light source distribution, and equipment material reflection parameters of the target industrial scene, and to model the spatial structure, light source distribution, and equipment material reflection parameters in the cloud rendering center to obtain the light energy propagation framework.

[0117] The boundary constraint generation module is used to divide the target industrial scene into layers of lighting areas using a light energy propagation framework, so as to obtain the light and shadow range of multiple layers and generate cross-layer light and shadow boundary constraints.

[0118] The layered lighting and shadow rendering module is used to render the lighting and shadow range of each layer according to the lighting and shadow boundary constraints, and to adjust the light intensity in real time during rendering to obtain multiple lighting and shadow rendering results of different levels.

[0119] The mutation region correction module is used to fuse the lighting and shadow rendering results of different levels and identify the light energy mutation regions in the fused lighting and shadow rendering results in order to correct the light energy mutation regions and obtain a continuous light field rendering image.

[0120] The rendering evaluation module is used to evaluate the consistency of the continuous light field rendering, obtain the consistency evaluation results, and feed the consistency evaluation results back to the cloud to adjust the light energy propagation framework and light and shadow boundary constraints.

[0121] The applicant of this invention has provided a detailed description of the embodiments of the invention in conjunction with the accompanying drawings. However, those skilled in the art should understand that the above embodiments are merely preferred embodiments of the invention. The detailed description is only intended to help readers better understand the spirit of the invention and is not intended to limit the scope of protection of the invention. On the contrary, any improvements or modifications made based on the inventive spirit of the invention should fall within the scope of protection of the invention.

Claims

1. A cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration, characterized in that, The method includes: The spatial structure, light source distribution, and equipment material reflection parameters of the target industrial scene are obtained, and the spatial structure, light source distribution, and equipment material reflection parameters are modeled in the cloud rendering center to obtain the light energy propagation framework. The target industrial scene is divided into layers of illumination areas using the light energy propagation framework to obtain the light and shadow range of multiple layers and generate cross-layer light and shadow boundary constraints. The light and shadow range of each layer is rendered according to the light and shadow boundary constraints, and the light intensity is adjusted in real time during rendering to obtain multiple light and shadow rendering results of different levels. The lighting and shadow rendering results at different levels are fused, and the regions of abrupt changes in lighting energy in the fused lighting and shadow rendering results are identified in order to correct the regions of abrupt changes in lighting energy and obtain a continuous light field rendering image. A consistency assessment is performed on the continuous light field rendering image to obtain a consistency assessment result, and the consistency assessment result is fed back to the cloud to adjust the light energy propagation framework and light and shadow boundary constraints.

2. The digital twin scene cloud-edge-device collaborative rendering method supporting multi-agent collaboration as described in claim 1, characterized in that, The process involves acquiring the spatial structure, light source distribution, and equipment material reflection parameters of the target industrial scene, and modeling these parameters in a cloud rendering center to obtain a light propagation framework, including: The original light source intensities of different forms in the light source distribution are obtained, and the original light source intensities are normalized to convert the original light source intensities of different forms into the equivalent irradiance intensity of each irradiation source point in the light source distribution. Based on the equivalent irradiance and the spatial structure of the target industrial scene, a spatial light energy propagation path is constructed, and the difference in irradiance attenuation intensity of the same irradiance source in different spatial light energy propagation paths is extracted, so as to determine the spatial light energy attenuation zone according to the difference in irradiance attenuation intensity.

3. The digital twin scene cloud-edge-device collaborative rendering method supporting multi-agent collaboration as described in claim 2, characterized in that, The process of acquiring the spatial structure, light source distribution, and equipment material reflection parameters of the target industrial scene, and modeling the spatial structure, light source distribution, and equipment material reflection parameters in a cloud rendering center to obtain a light propagation framework, also includes: Obtain the device material reflection parameters within the spatial light energy attenuation zone, and calculate the equivalent reflection brightness of the device material surface based on the device material reflection parameters. Map the equivalent reflection brightness to the corresponding spatial unit according to the device material type to obtain the material reflection propagation parameter set. The equivalent irradiance, spatial light energy attenuation partition, and material reflection and propagation parameter set are encapsulated to construct the light energy propagation framework, and the reference brightness level corresponding to each spatial unit in the light energy propagation framework is determined. The reference brightness level is used to divide the light and shadow range of different levels.

4. The cloud-edge-device collaborative rendering method for digital twin scenes supporting multi-agent collaboration as described in claim 1, characterized in that, The step of dividing the target industrial scene into layered illumination regions using the light energy propagation framework to obtain multi-layered light and shadow ranges and generating cross-layered light and shadow boundary constraints includes: The light propagation framework is read through the edge rendering collaboration node, and the light and shadow range corresponding to the edge rendering collaboration node is divided into multiple rendering space units to calculate the dominant lighting contribution value of each rendering space unit. The rendering space unit is divided into multiple light and shadow ranges of different levels according to the dominant lighting contribution value, and the width of the light and shadow transition band between adjacent light and shadow ranges is calculated. Calculate the constraint coefficient of each rendering space unit within the width of the light and shadow transition zone, and encapsulate the constraint coefficient with the light and shadow range of the corresponding layer to construct the light and shadow boundary constraint across layers.

5. The digital twin scene cloud-edge-device collaborative rendering method supporting multi-agent collaboration as described in claim 4, characterized in that, The rendering process is performed on the light and shadow range of each layer according to the light and shadow boundary constraints, and the light intensity is adjusted in real time during rendering to obtain multiple light and shadow rendering results at different levels, including: Calculate the initial rendering load of the light and shadow range of each layer in each rendering space unit, and transform the light and shadow boundary constraints into layered rendering tasks to be executed based on the initial rendering load; The layered rendering task is read by cloud nodes, edge rendering collaboration nodes and terminal nodes respectively, and initial lighting and shadow rendering is performed in the rendering layers corresponding to the cloud nodes, edge rendering collaboration nodes and terminal nodes respectively, so as to obtain the initial lighting and shadow rendering results of each node after the initial lighting and shadow rendering is performed. Calculate the intensity of the local illumination change trend corresponding to each node in the initial lighting and shadow rendering result, and align the intensity of the local illumination change trend of each node to generate the continuous modulation amount of each node. The continuous modulation amount is used to adjust the intensity of the local illumination change trend in real time, and outputs the corresponding light and shadow rendering results for each level after real-time adjustment.

6. The digital twin scene cloud-edge-device collaborative rendering method supporting multi-agent collaboration as described in claim 1, characterized in that, The process of fusing lighting and shadow rendering results at different levels and identifying abrupt changes in lighting energy in the fused lighting and shadow rendering results, in order to correct these abrupt changes in lighting energy and obtain a continuous light field rendering image, includes: The terminal node receives the cloud-based distant view lighting and shadow rendering results and the edge mid-range lighting and shadow rendering results, and extracts the pre-stored local near-view rendering results, so as to map the cloud-based distant view lighting and shadow rendering results, the edge mid-range lighting and shadow rendering results, and the local near-view rendering results into the spatial units within the terminal node's field of view. Calculate the reference brightness before fusion for each spatial unit, and calculate the brightness change gradient between adjacent spatial units based on the reference brightness before fusion, so as to identify the light energy abrupt change region where the brightness change gradient exceeds a set change threshold.

7. The digital twin scene cloud-edge-device collaborative rendering method supporting multi-agent collaboration as described in claim 6, characterized in that, The process of fusing lighting and shadow rendering results at different levels and identifying abrupt changes in lighting energy in the fused lighting and shadow rendering results, in order to correct these abrupt changes in lighting energy and obtain a continuous light field rendering image, further includes: Calculate the gradient deviation between the brightness change gradient corresponding to the region of sudden change in illumination energy and the preset target gradient, and generate a continuous correction intensity based on the gradient deviation. The brightness of the reference brightness before fusion is corrected according to the continuity correction intensity and spatial unit order to obtain the corrected output brightness, and the continuous light field rendering image is generated based on the output brightness of all spatial units.

8. The digital twin scene cloud-edge-device collaborative rendering method supporting multi-agent collaboration as described in claim 1, characterized in that, The process of performing a consistency evaluation on the continuous light field rendered image, obtaining a consistency evaluation result, and feeding the consistency evaluation result back to the cloud to adjust the light energy propagation framework and light and shadow boundary constraints includes: The current field of view is obtained based on the continuous light field rendering image, and the current field of view is divided into multiple partitions to calculate the light and shadow consistency score of each partition. The intensity of light and shadow deviation in each partition is determined by the light and shadow consistency score, and the light propagation framework and light and shadow boundary constraints are corrected according to the intensity of light and shadow deviation to obtain the updated light propagation framework and light and shadow boundary constraints.

9. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-8.

Images