Light and shadow scheme optimization method and system based on digital twinborn and artificial intelligence

Abstract

The invention discloses a light and shadow scheme optimization method and system based on digital twinning and artificial intelligence, and belongs to the technical field of data processing, and the method comprises the steps: converting a user demand into a semantic vector, mapping a user intention vector to obtain a light and shadow control parameter, carrying out the simulation calculation of the light and shadow control parameter, and outputting a simulation image and a light and shadow physical index. Extracting key physical quantities in a light and shadow state from the simulation image, integrating to form a simulation feedback index, calculating a difference between a simulation semantic vector and a user intention vector, obtaining a semantic error vector, inputting the semantic error vector and a light and shadow control parameter, outputting an update quantity of the light and shadow control parameter, establishing a convergence data function, and obtaining a simulation result; and judging an optimization condition, terminating the operation when a preset scene is obtained, and outputting a final result. According to the light and shadow scheme optimization method and system based on digital twinning and artificial intelligence, a closed-loop adaptive optimization process is realized by constructing a linkage mechanism from semantic understanding and parameter reasoning to feedback correction.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and particularly relates to a method and system for optimizing a light and shadow scheme based on digital twin and artificial intelligence. Background Art

[0002] With the rapid development of digital twin technology and artificial intelligence, optimizing the physical environment based on a virtual simulation environment has become a key means in multiple industries (such as architectural design, urban lighting, stage setting, film and television production, etc.). Among them, the design and optimization of the light and shadow scheme, as an important part of the spatial visual experience, increasingly rely on fine-grained simulation and intelligent assistance systems. In practical applications, professional designers usually need to manually set a series of light and shadow parameters (such as light source type, position, direction, illuminance intensity, color temperature, material reflectivity, etc.), and then repeatedly test and adjust through a simulation system to obtain a light and shadow effect that meets the requirements of a specific scene. However, this process not only highly depends on professional knowledge and experience, but also has a long debugging cycle and low efficiency, and it is difficult to meet the actual needs of rapid response and personalized design.

[0003] To solve the problems of low efficiency of manual parameter adjustment and unfriendly user interaction, in recent years, some studies have begun to try to introduce artificial intelligence to assist in light and shadow optimization. For example, predicting user preferences through a machine learning model, or recommending reference designs with similar light and shadow styles based on image retrieval. Although these attempts have improved the efficiency of certain links, overall, the existing technologies still have obvious defects when dealing with the following three key challenges: (1) Mismatch in the expression of user needs: When users describe their light and shadow needs, they often use subjective and vague language (such as "bright and transparent", "warm and dreamy"), while the simulation system requires clear structured parameters, and it is difficult for existing methods to achieve an accurate conversion from subjective intentions to physical parameters; (2) Lack of a feedback closed-loop in the optimization process: Most systems regard the AI model as a static recommender, ignoring that the digital twin simulation system can provide high-value light feedback data, and fail to form a dynamic optimization mechanism of "intention → parameter → simulation → reverse correction"; (3) Lack of interpretability and adaptability in the parameter adjustment process: Although traditional black-box optimization methods (such as genetic algorithms, particle swarm algorithms) can perform parameter search, the process is not interpretable, and it is difficult to make effective adaptations to complex building space structures or user personality preferences. Most existing methods are "static one-way processes": user expression → parameter recommendation → simulation verification. Once the recommended scheme deviates from the expectation, it needs to be re-entered, lacking the ability to learn and adjust, which severely limits the intelligent application potential of the digital twin system in light and shadow design.

[0004] Therefore, we propose a method and system for optimizing a light and shadow scheme based on digital twin and artificial intelligence to solve the above problems. Summary of the Invention

[0005] The object of the present invention is to solve the problems of difficult parameter conversion and lack of feedback in optimization in the prior art, and to propose an optimization method and system for lighting schemes based on digital twins and artificial intelligence.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] An optimization method for lighting schemes based on digital twins and artificial intelligence, comprising:

[0008] Input the user requirements, convert the user requirements into semantic vectors, and obtain the user intention vector by weighted averaging the semantic vectors generated for different category inputs;

[0009] Map the user intention vector in a graph neural network to obtain lighting control parameters, wherein the mapping uses the semantic features in the user intention vector as nodes in the graph neural network, and the lighting control parameters as the mapping targets of the nodes, and adjusts the edge weights of the nodes in the graph to ensure the accuracy of the mapping;

[0010] Perform simulation calculations on the lighting control parameters, output simulation images and lighting physical indicators, extract key physical quantities in the lighting state from the simulation images, and integrate them to form simulation feedback indicators, wherein the simulation calculations are performed based on a preset digital twin three-dimensional space for performing lighting simulations, and the simulation images are the rendering results of the scene under given parameters;

[0011] Align the user intention vector with the simulation feedback indicators, map the simulation feedback indicators to obtain a simulation semantic vector by constructing a cross-modal mapping relationship, calculate the difference between the simulation semantic vector and the user intention vector to obtain a semantic error vector, and perform perceptual normalization processing on the semantic error vector to obtain a global weighted error; wherein, in the cross-modal mapping relationship, the nodes represent semantic labels, and the edges represent the dependence relationships between the semantic labels and the physical indicators;

[0012] Establish an optimization strategy model, input the semantic error vector and the lighting control parameters, and output the update amount of the lighting control parameters, wherein the optimization strategy model adopts a graph attention network structure and is set to map the semantic error vector to the gradient adjustment amount of the lighting control parameters;

[0013] Establish a convergence data function, input the semantic error vector, the global weighted error and the adjustment amount of the lighting control parameters for judgment, and when a preset scenario is obtained, terminate the operation and output the final result.

[0014] Preferably, the conversion of user requirements includes natural language processing and image input processing. The natural language processing converts user requirements into vector representations through a pre-trained language model; the image input processing extracts visual features of images through a convolutional neural network to obtain vector representations.

[0015] Preferably, the features extracted in natural language processing include color temperature, brightness, light source, and time.

[0016] Preferably, the features extracted in image input processing include light source position, color temperature, and brightness distribution.

[0017] Preferably, the key physical quantities extracted from the simulation images include global average illuminance, lighting uniformity, maximum brightness and regional contrast, main spot area and distribution range, and shadow boundary clarity.

[0018] Preferably, an orthogonality regularization term is introduced into the optimization strategy model, and the difference between the combined gradient and the independent gradient is used as the regularization term to penalize non-orthogonal gradient directions and suppress the direction cancellation of error signals from different semantic directions on the adjustment of the same physical parameter.

[0019] Preferably, the preset scenarios include:

[0020] When the current weighted semantic error is less than the global threshold, it is considered that the user's goal has been achieved and the process automatically stops;

[0021] When the error changes in the past w rounds tend to be stable, that is, the error change trend is less than the trend stability threshold, it indicates that the optimization has entered a stable fluctuation region and converges actively;

[0022] When the parameter update amplitude continuously shrinks, it indicates that the strategy approaches ineffective adjustment and converges actively.

[0023] A lighting scheme optimization system based on digital twin and artificial intelligence, comprising:

[0024] An input processing module, configured to convert the input lighting requirements into structured semantic vectors, where the input lighting requirements include natural language input and / or image input; and obtain a user intention vector by weighted averaging the semantic vectors generated by different types of inputs;

[0025] A parameter mapping module, configured to convert the user semantic vector into specific lighting control parameters;

[0026] A simulation calculation module, configured to perform simulation calculations on the lighting control parameters, output simulation images and lighting physical indicators, extract key physical quantities in the lighting state from the simulation images, and integrate them into simulation feedback indicators;

[0027] An error evaluation module, which is configured to align the user intention vector with the simulation feedback metrics. By constructing a cross-modal mapping relationship, the simulation feedback metrics are mapped to obtain a simulation semantic vector, and the difference between the simulation semantic vector and the user intention vector is calculated to obtain a semantic error vector. The semantic error vector is subjected to perceptual normalization processing to obtain a global weighted error;

[0028] A strategy optimization module, which is configured to input the semantic error vector and the lighting control parameters and output an update amount of the lighting control parameters;

[0029] An optimized control and output module, which is configured to make a judgment based on a convergence judgment function, input the semantic error vector, the global weighted error, and the adjustment amount of the lighting control parameters. When a preset scenario is obtained, the operation is terminated and the final result is output.

[0030] In summary, the technical effects and advantages of the present invention: The lighting scheme optimization method and system based on digital twin and artificial intelligence start from the user's subjective expression, automatically generate structured lighting parameters, verify through the digital twin environment, and reverse correct the internal model of the system through the simulation results, gradually optimizing the output scheme, realizing a closed-loop adaptive optimization process. Description of the Drawings

[0031] Figure 1 It is a flowchart of the method in the present invention;

[0032] Figure 2 It is a schematic structural diagram of the system in the present invention. Detailed Embodiments

[0033] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments.

[0034] As Figure 1 shown, a lighting scheme optimization method based on digital twin and artificial intelligence includes:

[0035] Input the user requirements, convert the user requirements into a semantic vector, and obtain a user intention vector by weighted averaging the semantic vectors generated by different categories of inputs;

[0036] Map the user intention vector in the graph neural network to obtain the lighting control parameters. Among them, the mapping takes the semantic features in the user intention vector as nodes and the lighting control parameters as the mapping targets of the nodes in the graph neural network, and adjusts the edge weights of the nodes in the graph to ensure the accuracy of the mapping;

[0037] Perform simulation calculations on the light and shadow control parameters, output the simulation images and light and shadow physical indicators, extract the key physical quantities in the light and shadow state from the simulation images, and integrate them to form the simulation feedback indicators. Among them, the simulation calculation is performed based on the preset digital twin three-dimensional space for light simulation, and the simulation image is the rendering result of the scene under the given parameters;

[0038] Align the user intention vector with the simulation feedback indicators. By constructing a cross-modal mapping relationship, map the simulation feedback indicators to obtain the simulation semantic vector, calculate the difference between the simulation semantic vector and the user intention vector to obtain the semantic error vector, and perform perceptual normalization processing on the semantic error vector to obtain the global weighted error; among them, in the cross-modal mapping relationship, the nodes represent semantic labels, and the edges represent the dependency relationships between the semantic labels and the physical indicators;

[0039] Establish an optimization strategy model, input the semantic error vector and the light and shadow control parameters, and output the update amount of the light and shadow control parameters. Among them, the optimization strategy model adopts a graph attention network structure and is set to map the semantic error vector to the gradient adjustment amount of the light and shadow control parameters;

[0040] Establish a convergence data function, input the semantic error vector, the global weighted error and the adjustment amount of the light and shadow control parameters for judgment. When the preset scenario is obtained, terminate the operation and output the final result.

[0041] The specific steps of this embodiment are as follows,

[0042] Step 1: Semantic structured expression of user intention

[0043] The goal of this step is to transform the user's light and shadow requirements (whether described by natural language or input by image) into a structured semantic vector . This vector contains the specific characteristics of the light and shadow effects expected by the user and becomes the basic input for the subsequent steps, driving the entire optimization process. The user's light and shadow requirements are usually vague and subjective. How to effectively transform these requirements into computable light and shadow parameters is the core problem to be solved in this step.

[0044] First, the user can express their requirements through natural language description (such as "There is warm sunset light in the room") or upload an image (such as a reference photo with light and shadow effects). For different input forms, the system needs to adopt different processing methods and finally generate a general semantic vector , so that the subsequent steps can use it.

[0045] Natural language processing

[0046] For natural language input, the system converts the input text into a vector representation through a pre-trained language model (such as GPT-4 or BERT). The language model can not only understand the meaning of words but also capture complex semantic associations in the context. For example, when processing "warm sunset light", the system can extract the following key information from the text:

[0047] Color temperature: Warm tone, indicating that the color temperature of the light source is low, close to red or orange.

[0048] Brightness: Soft brightness, referring to relatively uniform light without strong light spots or high contrast.

[0049] Light source: Natural light at sunset, usually accompanied by low-intensity illumination and gradually changing brightness.

[0050] Time: Sunset, usually referring to the twilight period, meaning that the direction and intensity of the light will change with time.

[0051] By converting these features into a high-dimensional vector, the semantic vector generated by the system contains the above information.

[0052] Image input processing

[0053] If the user provides image input, the system will use a convolutional neural network (CNN) to extract the visual features of the image, especially information such as the light and shadow distribution, brightness change, and color tone in the image. The purpose of image processing is to extract information such as the light source, brightness, and light spot distribution to form a structured vector .

[0054] The convolutional neural network will first perform feature extraction on the image to identify the position of the light source, the intensity and direction of the light. For example, in a sunset image, the CNN can extract the following features:

[0055] Light source position: The position of the sun and the direction of illumination;

[0056] Color temperature: The color temperature of the image is warm (low color temperature);

[0057] Brightness distribution: The distribution of bright parts in the image.

[0058] In this way, the CNN converts the image into a feature vector containing the above information . This vector, like the semantic vector of the text input represents the user's specific requirements for light and shadow.

[0059] Multimodal fusion

[0060] To handle cases with different input forms (e.g., both natural language and images), the system fuses the language description and image features to generate a unified semantic vector . By using the method of weighted average, the system can fuse the two into a comprehensive vector according to the contribution degree of each input:

[0061] ;

[0062] Here, is a tuning parameter used to control the relative importance of the language input and the image input in the final semantic vector. Through this weighted method, the system can flexibly adjust the output of the model according to the type of input, so that the contribution of each input method to the result can be appropriately reflected. For example, when the user input is "natural light in the early morning", the text information will have a greater impact on the semantic vector, so may be relatively large; while if the user uploads an image, the weight of the image features will be relatively large, will be relatively small. This way ensures the effective fusion and unified representation of multimodal inputs.

[0063] The generated semantic vector

[0064] Finally, the output of Step 1 is a comprehensive semantic vector , which contains all the key information of the user's light and shadow requirements (such as color temperature, brightness, time, light source type, etc.).

[0065] For example, if the user describes "soft light in the early morning", the system will extract features such as color temperature, brightness, and light source direction from the text, and combine with the possible uploaded image of the early morning scene. The generated semantic vector will contain features such as low color temperature, soft brightness, and low light source intensity in the early morning. These features will directly affect the generation of light and shadow parameters in the subsequent steps.

[0066] Step 2: Mapping Inference from Intent Vector to Physical Parameters

[0067] The goal of this step is to convert the semantic vector generated in Step 1 into specific light and shadow control parameters , which will be used to drive subsequent light and shadow simulations and optimizations. This process not only involves the conversion from the user's light and shadow requirements to physical parameters, but also requires learning how to accurately map abstract light and shadow features to physical control signals through a graph neural network (GNN). Through this step, we will ensure that the user's subjective light and shadow requirements can be accurately reflected in the physical space and provide a basis for subsequent light and shadow simulations.

[0068] The mapping process uses a Graph Neural Network (GNN), where each semantic feature acts as a node in the network, and the physical parameters are the mapping targets of these nodes. The graph neural network learns the relationship between semantic features and physical parameters and ensures that each feature can be accurately mapped to the physical parameters by adjusting the edge weights between nodes in the graph. For example, the feature of "soft sunset light" in the semantic vector can be mapped to a lower light intensity (soft brightness) and a higher color temperature (warm light).

[0069] The calculation formula in the mapping process is as follows:

[0070] ;

[0071] where, is the i-th feature in the semantic vector , representing a light and shadow feature (such as color temperature, brightness, etc.), and is the weight associated with this feature, indicating the degree of influence of this feature on the physical parameter. The weight is obtained through the training of the graph neural network, which determines the contribution of each feature to the final physical parameter . Through this formula, the system can convert the user's light and shadow requirements into specific control parameters.

[0072] In the specific implementation, each dimension of the semantic vector will be mapped to a physical parameter through the graph neural network. Assuming the user inputs "warm sunset light", the system will extract the following features from the semantic vector:

[0073] Color temperature: lower color temperature (warm light source).

[0074] Brightness: soft brightness (low-intensity light).

[0075] Light source type: sunset light source (natural light).

[0076] Light source direction: the light source direction close to the sunset angle.

[0077] These features will be mapped to specific physical parameters. For example, the system will generate control parameters such as light source position, color temperature, and light intensity for subsequent light and shadow simulation.

[0078] To improve the accuracy of the mapping process and avoid overfitting, we introduce a regularization term . The regularization term ensures that the generated physical parameters are within a reasonable range and meet the requirements of the actual physical environment by penalizing overly large weights. The calculation formula of the regularization term is:

[0079] ;​

[0080] Among them, is the th physical parameter generated by the graph neural network, is the target value of this physical parameter (set according to the light and shadow design standard or reference simulation data, for example), is a regularization hyperparameter used to control the strength of regularization. By adding this regularization term, the system can ensure the stability of physical parameters, making the output results more in line with the requirements of actual light and shadow design.

[0081] Through the combination of the graph neural network and the regularization term, the system can automatically optimize the mapping relationship between semantic features and physical parameters, ensuring that the light and shadow requirements input by each user can be transformed into a set of control parameters that conform to physical laws. These control parameters will be used in subsequent light and shadow simulation and optimization steps to drive the simulation system to generate light and shadow effects that meet the user's requirements. Through the combination of the graph neural network (GNN) and the regularization term, this step not only efficiently transforms semantic features into physical control parameters, but also improves the accuracy and stability of the transformation through automatic learning and regularization. This enables the user's light and shadow requirements to be accurately expressed in physical simulations and provides a solid foundation for subsequent light and shadow simulation and optimization.

[0082] Step 3: Conduct simulation calculations in the digital twin environment

[0083] The task of this step is to conduct simulation calculations on the light and shadow control parameters generated in Step 2 in the digital twin environment and output simulation images and light and shadow physical indicators . The input is the light and shadow control parameters , where each represents a structured light and shadow physical parameter, which may include light source position, color temperature, brightness, direction, divergence angle, ambient reflectivity, etc. All parameters are mapped by the graph neural network in the previous step and already have structural consistency. The system inputs the parameter into the simulation module and performs light simulation in the preset digital twin three-dimensional space. This twin scene is constructed using a physically based rendering engine, supporting path tracing or global illumination calculation methods, and can simulate processes such as light propagation, object material response, and space reflection, thereby generating realistic and reliable simulation images . The simulation image is the rendering result of the scene under the given parameter , and the image size, resolution, and viewing angle are all aligned with the twin scene standard to ensure consistency in subsequent analysis.

[0084] To provide quantitative feedback capabilities, the system will input it into the lighting index calculation module where, combined with scene geometry and material information, key physical quantities in the light and shadow state are extracted to form an index vector . These indexes include the following categories:

[0085] : Global average illuminance (unit: Lux);

[0086] : Lighting uniformity ;

[0087] : Maximum brightness and regional contrast;

[0088] : Main light spot area and distribution range;

[0089] : Shadow boundary sharpness (measurable by edge gradient);

[0090] To enhance the controllability of the real scene restoration ability, we introduce an optional parameter: simulation environment correction term , which is derived from the sampling data of the real physical environment by on-site sensors and is used to correct the possible parameter deviation between the twin model and the physical world. The authenticity weight during the simulation process can be defined as:

[0091] ;

[0092] Among them, represents the corrected simulation image, is the reference image (such as environmental HDR map, reflectivity map) generated from the on-site sensor sampling values, . This item can be enabled when real environment data is accessed to enhance the physical consistency of the simulation.

[0093] The final output includes: (1) the simulation image (or the corrected image ), for human perception; (2) the structured feedback index , as the basic data for subsequent error modeling and adaptive optimization.

[0094] Step 4: Semantic error modeling and evaluation of simulation feedback

[0095] The main task of this step is to establish a semantic error evaluation mechanism in the system: the user intention vector output in Step 1 and the simulation feedback index output in Step 3 Align by constructing a cross-modal mapping relationship Map the physical feedback back to the semantic space to form a simulation semantic vector , and finally calculate the difference between the two . This is a key step for the entire system to achieve "closed-loop feedback". Without this mechanism, the system cannot determine whether the current simulation is consistent with the user's expectations, and thus cannot enter the next step of parameter adaptive correction. In the previous step (Step 3), we obtained the simulation image and the structured feedback metrics . These physical metrics are quantitative descriptions of the simulation light and shadow results. The user intention semantic vector generated in the first step contains the user's subjective light and shadow requirements. The two are in completely different data spaces, and it is difficult to compare them using traditional methods. The present invention designs a semantic mapping mechanism here to convert the physical feedback into a semantic estimate , and then perform error quantification

[0096] First, we construct a non-linear mapper from the physical feedback to the semantic space for learning the following mapping relationship:

[0097] ;

[0098] is essentially a lightweight graph neural network, where nodes represent semantic labels (such as "soft", "bright", "spotlight"), and edges represent the dependencies between them and the physical metrics. Different from traditional linear regression or MLP, this graph structure can express the complex structure that "one semantic concept is jointly determined by multiple physical metrics". For example, "soft" is controlled by factors such as brightness gradient , illuminance uniformity and the diffusion angle of the main light source

[0099] To enhance the interpretability and accuracy of this mapping, we introduce a cross-dimensional perception regularization term during network training to penalize the impact when "high-perception-sensitive semantics" are misjudged, so that the error can better reflect the user experience difference. It is defined as follows:

[0100] ;

[0101] where represents the perception weight of the -th semantic dimension, and its value is obtained based on psychophysical research or system data fitting, reflecting the influence intensity of this dimension on human perception. For example, is usually higher than This regularization term does not affect the error output , but it will be used in subsequent optimization priority judgment and gradient scheduling.

[0102] Next, calculate the semantic error vector , that is, the difference between the user's intention and the simulation semantics:

[0103] ;

[0104] Each dimension represents the degree of deviation of the system in the th semantic direction. is a structured vector, where each dimension has independent interpretability and can be directly used for the next step of "semantic scheduling optimization".

[0105] To avoid the influence of error values in different dimensions on system judgment, it is also necessary to perform perceptual normalization on the errors, and finally obtain a global weighted error metric:

[0106] ;

[0107] Among them, is the importance weight of the semantic dimension (from task priority or user preference), is the regularization coefficient, which balances the relationship between error and perceived risk. This error metric will be used as the decision basis for whether to start parameter adjustment and which dimension to prioritize in the subsequent steps.

[0108] In the actual system operation, the semantic mapper can be pre-trained with a small number of user samples and further optimized through online feedback during the long-term operation. The system also supports a local fine-tuning mechanism, that is, when the error of a certain semantic dimension deviates for a long time, the mapping edge weight in this direction can be automatically updated locally.

[0109] Step 5: Generation of optimization strategy based on error feedback

[0110] The goal of this step is to establish an optimization strategy model for the semantic error vector output in step 4, which is used to guide the adjustment direction and amplitude of the light and shadow control parameters .

[0111] The input is the semantic error vector generated in step 4 , and each represents the deviation of the system in the semantic dimension ; the auxiliary input is the current light and shadow control parameters generated in step 2. The output is the update amount of the control parameters 。

[0112] In this solution, there is no one-to-one correspondence between semantic errors and physical parameters. A deviation in one semantic dimension may be caused by multiple parameters. Therefore, in this step, a learnable structured mapping function is introduced , and the following optimization strategy model is constructed:

[0113] ;

[0114] This model maps the error dimension to the gradient adjustment amount of the control parameter . Different from ordinary backpropagation, this network integrates a semantic-physical causal relationship matrix into its structure, making its learning structure more physically relevant.

[0115] To avoid the optimization model from generating conflicting adjustments in multiple semantic directions, an orthogonality regularization term is designed in this step , which suppresses the problem of "direction cancellation" caused by error signals from different semantic directions on the adjustment of the same physical parameter. This regularization term is defined as follows:

[0116] ;

[0117] Where:

[0118] is the parameter gradient vector propagated from the semantic error ;

[0119] is 's influence factor on (given by the structure);

[0120] The first term represents the total impulse after all error directions are combined, and the second term is the independent cumulative amount of each error direction;

[0121] The larger the difference, the more serious the conflict. The stronger the punishment during model training, the more likely the optimization strategy is to tend to semantic separation.

[0122] This optimization model can be implemented using the structure of a graph attention network (GAT). It continuously learns the adjustment mode between errors and parameters through training samples (user evaluation deviation + convergence effect after adjustment). During online operation, 's inference process is completely automatic and only based on the error can output , with high efficiency and real-time performance. The output by the modelIt will be used as the input of the parameter adjustment module in the subsequent system. How to perform the adjustment, whether to perform it, and the adjustment range are determined by the system controller or scheduler.

[0123] Step 6: Optimize Convergence Judgment and Result Output

[0124] The task of this step is to model the convergence judgment function , which is used to determine whether the system has reached the user's light and shadow semantic goal in the current round and whether to terminate further parameter optimization and simulation iteration. This judgment process does not directly control parameter update, simulation, or output, but only provides a boolean judgment value at the system level

[0125] The input includes three categories:

[0126] The semantic error vector of the current round (from Step 4);

[0127] The most recent round error history sequence , where represents the weighted semantic error;

[0128] The control parameter adjustment amount output by the previous Step 5 , which is used to judge whether the current strategy still has effective adjustment ability.

[0129] This scheme proposes a joint convergence criterion function that combines "error magnitude", "error trend stability", and "policy residual ability" . The judgment mechanism is as follows:

[0130] If the current weighted semantic error is less than the global threshold , it is considered that the user goal has been achieved;

[0131] If the error change in the past rounds tends to be stable, that is , it means that the optimization has fallen into a stable fluctuation area;

[0132] If the parameter update amplitude has continuously decreased, indicating that the strategy is approaching "ineffective adjustment", it can also converge.

[0133] The above criteria are uniformly expressed as:

[0134] ;

[0135] Where:

[0136] is the average error of the most recent rounds;

[0137] The variance represents the error trend;

[0138] is the average parameter adjustment amplitude of the current round;

[0139] : Error convergence threshold;

[0140] : Trend stability threshold;

[0141] : minimum effective update amplitude threshold;

[0142] : Perceptual sensitivity weight from step 4.

[0143] This joint criterion ensures that the system can automatically stop after the user's intention is achieved ( Small enough), and can also actively converge when optimization is difficult to advance ( close to zero), while avoiding falling into low-frequency oscillation (through error fluctuation variance Suppression). This takes into account convergence, security and iteration efficiency. Judgment, this step also constructs an optional learning-based convergence judgment model , which is used for intelligent judgment in more complex situations. It is based on the historical error sequence input and uses the label of whether the user terminates the operation as the supervision signal for training. The formula is as follows:

[0144] ;

[0145] The model adopts a lightweight recurrent neural network (such as GRU) structure, which can learn the error evolution pattern under different semantic goals, thereby improving the flexibility and adaptability of judgment.

[0146] This step is to determine whether the system should be terminated. Once the convergence mark is confirmed, , the system controller decides whether to call modules such as result encapsulation and interactive confirmation.

[0147] The technical solution in the above-mentioned embodiment of the present application has at least the following technical effects or advantages: This solution constructs a linkage mechanism from semantic understanding, parameter reasoning to feedback correction, so that users can express their lighting needs only by language or images without professional knowledge, and the system can automatically convert them into high-quality, personalized lighting design results.

[0148] The present application also provides a light and shadow scheme optimization system based on digital twins and artificial intelligence, such as Figure 2 As shown, including:

[0149] An input processing module, which is configured to convert the input light and shadow requirements into a structured semantic vector. The input light and shadow requirements include natural language input and / or image input; and obtain a user intention vector by weighted averaging the semantic vectors generated by different categories of inputs.

[0150] A parameter mapping module, which is configured to convert the user semantic vector into specific light and shadow control parameters.

[0151] A simulation calculation module, which is configured to perform simulation calculations on the light and shadow control parameters, output a simulation image and light and shadow physical indicators, extract key physical quantities in the light and shadow state from the simulation image, and integrate them to form a simulation feedback indicator.

[0152] An error evaluation module, which is configured to align the user intention vector with the simulation feedback indicator, map the simulation feedback indicator to obtain a simulation semantic vector by constructing a cross-modal mapping relationship, calculate the difference between the simulation semantic vector and the user intention vector to obtain a semantic error vector, and perform perceptual normalization processing on the semantic error vector to obtain a global weighted error.

[0153] A strategy optimization module, which is configured to input the semantic error vector and the light and shadow control parameters and output an update amount of the light and shadow control parameters.

[0154] An optimization control and output module, which is configured to make a judgment based on a convergence judgment function by inputting the semantic error vector, the global weighted error, and the adjustment amount of the light and shadow control parameters. When a preset scenario is obtained, the operation is terminated and the final result is output.

[0155] The working principle is as follows: Starting from the user's subjective expression, automatically generate structured light and shadow parameters, verify them through a digital twin environment, and reverse correct the internal model of the system through the simulation results, gradually optimizing the output scheme to achieve a closed-loop adaptive optimization process.

[0156] As described above, it is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. An optimization method for a light and shadow scheme based on digital twin and artificial intelligence, characterized in that Including: Input the user requirements, convert the user requirements into semantic vectors, and obtain the user intention vector by weighted average of the semantic vectors generated from different category inputs; Map the user intention vector in the graph neural network to obtain the light and shadow control parameters. Among them, the mapping uses the semantic features in the user intention vector as nodes in the graph neural network, and the light and shadow control parameters as the mapping targets of the nodes, and adjusts the edge weights of the nodes in the graph to ensure the accuracy of the mapping; Perform simulation calculations on the light and shadow control parameters, output the simulation images and light and shadow physical indicators, extract the key physical quantities in the light and shadow state from the simulation images, and integrate them into the simulation feedback indicators. Among them, the simulation calculation is based on the execution of light simulation in the preset digital twin three-dimensional space, and the simulation image is the rendering result of the scene under the given parameters; Align the user intention vector with the simulation feedback indicators. By constructing a cross-modal mapping relationship, map the simulation feedback indicators to obtain the simulation semantic vector, calculate the difference between the simulation semantic vector and the user intention vector to obtain the semantic error vector, and perform perceptual normalization processing on the semantic error vector to obtain the global weighted error; among them, in the cross-modal mapping relationship, the nodes represent semantic labels, and the edges represent the dependence relationship between the semantic labels and the physical indicators; Establish an optimization strategy model, input the semantic error vector and the light and shadow control parameters, and output the update amount of the light and shadow control parameters. Among them, the optimization strategy model adopts the graph attention network structure and is set to map the semantic error vector to the gradient adjustment amount of the light and shadow control parameters; Establish a convergence data function, input the semantic error vector, the global weighted error and the adjustment amount of the light and shadow control parameters for judgment. When the preset scenario is obtained, terminate the operation and output the final result.

2. The method for optimizing the light and shadow scheme based on digital twin and artificial intelligence according to claim 1, wherein, The conversion of the user requirements includes natural language processing and image input processing. The natural language processing converts the user requirements into vector representations through a pre-trained language model; the image input processing extracts the visual features of the image through a convolutional neural network to obtain vector representations.

3. The method for optimizing the light and shadow scheme based on digital twin and artificial intelligence according to claim 2, characterized in that, The features extracted in the natural language processing include color temperature, brightness, light source and time.

4. The method for optimizing the light and shadow scheme based on digital twin and artificial intelligence according to claim 2, wherein, The features extracted in the image input processing include light source position, color temperature and brightness distribution.

5. The method for optimizing the light and shadow scheme based on digital twin and artificial intelligence according to claim 1, wherein The key physical quantities extracted from the simulation images include global average illuminance, lighting uniformity, maximum brightness and regional contrast, main spot area and distribution range, and shadow boundary sharpness.

6. The method for optimizing the light and shadow scheme based on digital twin and artificial intelligence according to claim 1, wherein An orthogonality regularization term is introduced into the optimization strategy model, and the difference between the combined gradient and the independent gradient is used as the regularization term to punish the non-orthogonal gradient directions and suppress the cancellation of the adjustment directions of the error signals from different semantic directions to the same physical parameter.

7. The method for optimizing the light and shadow scheme based on digital twin and artificial intelligence according to claim 1, characterized in that, The preset scenarios include: The current weighted semantic error is less than the global threshold, it is considered that the user's goal has been achieved, and the operation is automatically stopped; The error changes in the past w rounds tend to be stable, that is, the error change trend is less than the trend stability threshold, indicating that the optimization has fallen into the stable fluctuation area and converges actively; The parameter update amplitude continuously shrinks, indicating that the strategy approaches ineffective adjustment and converges actively.

8. An optimization system for a light and shadow scheme based on digital twin and artificial intelligence, characterized in that, Including: An input processing module, which is configured to convert the input light and shadow requirements into a structured semantic vector, where the input light and shadow requirements include natural language input and / or image input; and obtain a user intention vector by weighted averaging the semantic vectors generated from different categories of inputs. A parameter mapping module, which is configured to convert the user semantic vector into specific light and shadow control parameters. A simulation calculation module, which is configured to perform simulation calculations on the light and shadow control parameters, output a simulation image and light and shadow physical indicators, extract key physical quantities in the light and shadow state from the simulation image, and integrate them to form a simulation feedback index. An error evaluation module, which is configured to align the user intention vector with the simulation feedback index, map the simulation feedback index to obtain a simulation semantic vector by constructing a cross-modal mapping relationship, calculate the difference between the simulation semantic vector and the user intention vector to obtain a semantic error vector, and perform perceptual normalization processing on the semantic error vector to obtain a global weighted error. A strategy optimization module, which is configured to input the semantic error vector and the light and shadow control parameters and output an update amount of the light and shadow control parameters. An optimization control and output module, which is configured to make a judgment based on a convergence judgment function, input the semantic error vector, the global weighted error, and the adjustment amount of the light and shadow control parameters. When a preset scenario is obtained, the operation is terminated and the final result is output.

Images