A method for modeling and multi-scene optimization and regulation of a modular sunshade system oriented to photothermal balance

Abstract

This invention discloses a proxy modeling and multi-scenario optimization control method for modular shading systems oriented towards light and heat balance. The proxy modeling method visualizes the opening and closing state matrix of the shading system and uses a sample set composed of this matrix and performance distribution images obtained from annual light and heat simulations to train a generative adversarial network, constructing a proxy model capable of rapidly predicting the spatial distribution of multiple indicators such as daylighting, glare, and thermal comfort from the facade state. The optimization control method, based on this proxy model and combined with indoor scene zoning and differentiated performance objectives, transforms shading control into a multi-objective optimization problem, automatically searching for the Pareto optimal solution set using an evolutionary algorithm, and finally outputting the optimal shading control command. This invention replaces time-consuming simulations with an efficient data-driven model, achieving rapid performance evaluation of high-dimensional modular shading systems and adaptive, refined control for multi-scenario needs.

Description

Technical Field

[0001] This invention relates to the field of green building technology and operation and maintenance, specifically to a method for proxy modeling and multi-scenario optimization and control of a modular shading system oriented towards light and heat balance. Background Technology

[0002] Natural lighting and thermal comfort are a pair of deeply coupled environmental variables: on the one hand, introducing natural lighting helps reduce energy consumption for artificial lighting and improves the quality of indoor use; on the other hand, excessive lighting is usually accompanied by higher solar radiation heat gain, which can easily lead to increased indoor temperature and increased cooling load, thereby disrupting the balance of comfort. Therefore, from the perspective of the comprehensive goals of indoor comfort and energy conservation, how to achieve a balance between light and heat between lighting benefits and human comfort is a key issue that building envelope and shading design have long faced.

[0003] In current architectural practice, daylighting design and shading / insulation design remain highly decoupled: the former mainly focuses on daylighting area, spatial geometry, and light-transmitting materials, while the latter emphasizes orientation, shading structure, building envelope insulation performance, and control schedules. Because these two types of design variables overlap and are coupled, independent design paths struggle to accurately depict the balance between light and heat, and achieve global optimization at the overall level. Meanwhile, in the context of low-carbon retrofitting of existing buildings, shading systems need greater adaptability; however, existing modular facade solutions still heavily rely on specific engineering scenarios, and performance evaluation heavily depends on numerical simulation, resulting in computational costs constraining solution iteration and scenario-oriented optimization.

[0004] To address the contradictions of light and heat coupling and dynamic changes in demand, existing technologies can be broadly categorized into two types: The first is intelligent shading control, which adjusts the shading state according to changes in the indoor and outdoor environment to achieve target performance. However, this method often requires repeated comprehensive scoring or performance simulation calculations when conditions change, making it difficult to achieve rapid response and adapt to diverse scenarios. The second method involves creating "schedule presets" or control sets through hourly simulations throughout the year to achieve joint optimization of the light and thermal environments. However, simulation-based schedule settings often only reflect the general patterns of typical climates and cannot cover the complex changes in real-world weather and usage behavior. Furthermore, some methods lack continuous evaluation and verification mechanisms for the control effects, and there is a disconnect between shading equipment and architectural design details, making it difficult to achieve design-led comprehensive optimization.

[0005] Therefore, there is an urgent need for a modular shading proxy modeling and multi-scenario optimization and control method oriented towards light and heat balance: to replace time-consuming simulation with a proxy model to achieve rapid multi-index prediction of high-dimensional shading state, and to establish an executable multi-objective optimization and control strategy by combining the thresholds and preferences of different functional scenarios, so as to improve the adaptability and operational efficiency of existing buildings in scene switching. Summary of the Invention

[0006] To address the aforementioned technical problems, the primary objective of this invention is to propose a modular shading system proxy modeling method oriented towards light and heat balance. This method can construct a rapid prediction model from the facade state image of the shading system to the performance distribution image of multiple indoor indicators (lighting, glare, thermal comfort) by training a conditional generative adversarial network, thereby replacing the time-consuming annual physical simulation with extremely low computational cost.

[0007] Another objective of this invention is to provide a multi-scenario optimization and control method based on the above-mentioned proxy model, which can automatically search for a set of Pareto optimal shading control schemes that achieve the best balance among multiple performance objectives according to the functional layout and differentiated performance requirements of the indoor space, and output executable control commands, thereby realizing efficient, adaptive, and scenario-driven intelligent shading control.

[0008] To achieve the above-mentioned technical objectives, the present invention adopts the following technical solution:

[0009] Firstly, this invention provides a proxy modeling method for a modular shading system oriented towards light-heat balance, comprising the following steps: S1: Establish a parameterized physical model of the modular shading system, divide the shading system into several independently controllable shading units, and set opening and closing degree parameters to represent the opening state of each shading unit, forming a shading opening and closing degree matrix. S2: Normalize and image-map the shading opening and closing degree matrix to generate facade state images reflecting different shading states; S3: For the opening and closing degree matrix corresponding to multiple sets of facade status images, perform annual light and heat environment simulation to obtain indoor space performance distribution data that includes at least daylighting adaptive index, glare adaptive index and thermal comfort percentage index. Then, normalize and rasterize the indoor space performance distribution data to generate performance distribution images that correspond one-to-one with the facade status images, thereby constructing a pairwise training sample set composed of shading status images and performance distribution images. S4: Divide the paired training sample set into a training set and a test set; S5: The conditional generative adversarial network is used to train the pairwise training sample set, so that the generator of the conditional generative adversarial network learns the mapping relationship from the facade state image to the performance distribution image, and obtains surrogate models for predicting the daylight adaptation index, glare adaptation index and thermal comfort percentage index in step S3 respectively. S6: Quantitatively evaluate the trained surrogate model using the test set and select the optimal model. Determine the parameters of the optimal model and write them into the surrogate model parameter library for subsequent rapid prediction and optimization.

[0010] Furthermore, the modular sunshade unit is a horizontal sunshade unit, and the opening and closing parameter of each unit is its continuously adjustable rotation angle within a preset angle range.

[0011] The shading opening and closing degree matrix is ​​a two-dimensional matrix with multiple rows and columns, corresponding to multiple dynamic shading units;

[0012] The facade status image is specifically generated by: normalizing the opening degree of each dynamic shading unit and converting it into grayscale pixel values ​​according to a preset mapping rule; generating a single-channel facade status image based on the spatial layout of the shading units, so that the image pixel values ​​and the opening degree of the corresponding shading units maintain a monotonic correspondence and a one-to-one spatial correspondence.

[0013] Furthermore, in step S3, the annual solar thermal environment simulation uses typical meteorological year data for year-round simulation analysis; multiple measuring points are arranged at preset intervals and heights in the indoor simulation scene to form a measuring point grid;

[0014] Regular grid measuring points are arranged on the indoor working surface or evaluation plane. The measuring points are located at a preset height, and the spacing between the measuring points is the preset grid spacing. Light environment indicators and thermal environment indicators at each measuring point are collected or calculated to form spatial discrete data for generating a performance distribution image. The index values ​​of each measuring point are mapped to the gray values ​​of the corresponding pixels in the performance distribution image according to their spatial coordinates. The regular grid measuring points are rectangular grids or equidistant grids, and their grid density is preset according to the spatial scale and calculation accuracy requirements.

[0015] Furthermore, in step S3, the paired training sample set contains a predetermined number of sample pairs;

[0016] The sample pairs are divided into a training set and a test set, and the training set and the test set are uniformly distributed within the range of the shading unit opening value.

[0017] Furthermore, in step S5, the conditional generative adversarial network is a Pix2pixHD model, which is trained using a multi-scale discriminator and feature matching loss.

[0018] During training, the input image is a single-channel grayscale image, and no instance mapping is performed; the training hyperparameters include at least a learning rate of 0.0002 and an image input / output size of 256×256.

[0019] Secondly, based on the proxy model trained by the aforementioned proxy modeling method for modular shading systems oriented towards light and heat balance, this invention further discloses a multi-scenario optimization and control method for modular shading systems oriented towards light and heat balance, comprising the following steps:

[0020] A. Load at least three pre-trained surrogate models, corresponding to the daylight adaptation index, glare adaptation index, and thermal comfort percentage index, respectively, to quickly infer the spatial distribution of multiple indices of candidate shading opening and closing schemes.

[0021] B. Based on the functional layout of the target indoor space, generate scene partition images, which at least include the target area and non-target areas;

[0022] C. Based on the preset performance standards, set differentiated performance thresholds for target areas of different functional types, and generate target performance distribution images corresponding to each performance index based on the performance thresholds;

[0023] D. For a candidate shading opening and closing scheme, after converting it into a facade state image, input it into the three surrogate models respectively, and quickly infer the corresponding prediction performance distribution image;

[0024] E. Within the target area, the predicted performance distribution image is compared pixel by pixel with the corresponding target performance distribution image, and the error rate of each performance index is calculated respectively.

[0025] F. Based on the error rate of each performance indicator, construct a multi-objective evaluation vector;

[0026] G. Employ a multi-objective evolutionary algorithm to iteratively optimize by minimizing the multi-objective evaluation vector, thereby obtaining a set of Pareto-optimal shading opening and closing schemes.

[0027] H. Select the final implementation scheme from the set of Pareto optimal shading opening and closing schemes, and output the corresponding shading unit control command.

[0028] Furthermore, in the scene partitioning image, different grayscale values ​​or color values ​​are used to identify different functional partitions; the differentiated performance threshold includes at least: independent working areas. Adaptive lighting index for public communication areas .

[0029] Furthermore, the error rate The calculation formula is:

[0030] ;

[0031] Where m represents the performance metric category index; The distribution image of the prediction performance output by the surrogate model; A pre-defined target performance distribution image; Let be the normalization scaling factor, representing the th The range of grayscale values ​​for each indicator image is used to normalize the error. The set of all pixels in the entire performance distribution image; This is the weighting function, used to express that different functional partitions are assigned different weights.

[0032] Furthermore, in step F, when constructing the multi-objective evaluation function, the error rates of each performance index are normalized or standardized, and a weighted summation method is used to form a comprehensive target value, with the weight coefficients preset according to the scenario requirements.

[0033] Furthermore, the multi-objective evolutionary algorithm specifically adopts the NSGA-II algorithm.

[0034] Beneficial effects:

[0035] First, this invention normalizes and visualizes the modular shading high-dimensional opening and closing degree matrix, constructs a pairwise sample set of "shading state image - photothermal performance distribution image", and trains a conditional generative adversarial network proxy model to achieve rapid prediction of the spatial distribution of multiple indicators such as daylighting, glare, and thermal comfort. This avoids repeated annual photothermal simulations for each candidate scheme, significantly reducing computational costs and improving the efficiency of scheme iteration and optimization solutions. At the same time, the output format based on the performance distribution image preserves the spatial heterogeneity of indoor performance, making the collaborative evaluation of multiple indicators comparable and interpretable under a unified scale and coordinate system. Furthermore, through quantitative evaluation using a test set and solidification of the optimal model, a reusable proxy model parameter library is formed, providing stable and deployable underlying support for multi-objective optimization control and scene switching adaptability under different functional scenario thresholds and preferences.

[0036] Secondly, this invention characterizes the opening and closing degree of horizontal sunshade units with continuously adjustable rotation angles, and uniformly expresses the high-dimensional sunshade state with a multi-row, multi-column matrix. Then, it normalizes and maps the image to a single-channel grayscale facade state image that corresponds one-to-one with the physical layout (pixel values ​​are monotonically proportional to the opening and closing degree). This not only improves the precision and optimizability of sunshade control, but also preserves the spatial pattern information of facade opening and closing, reduces the dependence on feature engineering, thereby enhancing the stability and interpretability of surrogate model learning, and improving the ability to predict and optimize complex sunshade states.

[0037] Third, by conducting year-round simulations using typical meteorological years and arranging regular grid measurement points (rectangular / equidistant grids, with density adjustable according to scale and accuracy requirements) at preset heights and spacings on the indoor working surface, this invention can obtain spatially discrete data of light and thermal environment indicators under a unified and reproducible climate boundary and spatial sampling framework. This not only fully characterizes the annual light-thermal coupling effect and indoor spatial heterogeneity but also ensures the structure and consistency of the data, facilitating the stable generation of performance distribution images and improving the reliability and scalability of surrogate model training, cross-scheme comparison, and subsequent optimization and control.

[0038] Fourth, by setting a predetermined size of paired sample set and dividing the sample pairs into training and test sets, and ensuring that both are evenly covered within the range of shading unit opening values, this invention can effectively improve the representativeness and coverage of the sample space and avoid prediction distortion caused by training data bias. At the same time, by using an independent test set, it can achieve objective quantitative evaluation and screening of the surrogate model, enhance the model's generalization ability and stability, and thus improve the reliability of subsequent rapid prediction and optimization control under different shading opening combinations.

[0039] Fifth, this invention uses Pix2pixHD as a conditional generative adversarial network and introduces a multi-scale discriminator and feature matching loss, which can simultaneously enhance the global structural consistency and local detail fidelity of the output performance distribution, and more accurately reproduce the spatial pattern of indoor light and heat performance. At the same time, the input uses a single-channel grayscale image and does not introduce instance mapping, which reduces input redundancy and label dependence, reduces training complexity and improves the model's convergence stability and cross-scene generalization ability, thereby improving the reliability of proxy prediction and engineering deployability.

[0040] Sixth, the multi-scenario optimization and control method of this invention achieves rapid inference of the spatial distribution of lighting, glare, and thermal comfort of candidate shading schemes by loading a multi-index pre-trained surrogate model, avoiding repeated high-time-consuming simulations during the iteration process; it also encodes the planar functional layout into scene partition images and translates differentiated performance standards into target performance distribution images, so that the needs of different functional areas participate in the evaluation in a unified and calculable form; on this basis, it constructs a multi-objective evaluation vector using the pixel-by-pixel error rate within the target area and introduces a multi-objective evolutionary algorithm to search for the Pareto optimal solution set, thereby outputting multiple sets of optional compromise schemes under the trade-off of multiple indicators, and finally transforming the optimal scheme into executable shading control commands, realizing adaptive, interpretable, and highly efficient optimization and control for multiple scenarios.

[0041] Seventh, the multi-scene optimization and control method of the present invention identifies functional areas in scene partition images with different gray levels / colors and sets differentiated performance targets for different areas. It can embed the needs of multi-functional spaces into the optimization evaluation in an intuitive and calculable way, so that the evaluation focuses on key areas and reflects the priority of areas, thereby improving the matching degree and interpretability of optimization results to actual use scenarios and enhancing the adaptability across layout scenarios.

[0042] Eighth, in the multi-scenario optimization and control method of the present invention, the error rate definition introduces regional masking and unified normalization processing, so that the evaluation focuses on the target functional area and can be weighted according to the importance of the area. At the same time, the deviation of different performance indicators is converted into comparable values, which improves the stability, fairness and interpretability of multi-indicator comprehensive evaluation and optimization search.

[0043] Ninth, in the multi-scenario optimization and control method of the present invention, by normalizing / standardizing the error rate of each indicator and summarizing it with adjustable weights, the bias caused by the difference in scale of different indicators can be eliminated, making the comprehensive evaluation more stable and comparable; at the same time, the weights can be adjusted according to the needs of the scenario, which makes it easier to achieve controllable preference expression and compromise decision-making between lighting, glare and thermal comfort, thereby improving the adaptability and interpretability of multi-scenario optimization and control.

[0044] Tenth, in the multi-scenario optimization and control method of this invention, by maintaining the diversity of solution sets through fast non-dominated sorting and crowding distance, it can efficiently approach the Pareto front and output a set of representative compromise solutions among conflicting goals of lighting, glare and thermal comfort; at the same time, it has weak requirements on the form of the objective function, good adaptability to complex nonlinear and high-dimensional decision variables, and the optimization process is stable with good convergence and solution set coverage, thereby improving the reliability and selectivity of multi-scenario shading optimization and control. Attached Figure Description

[0045] Figure 1 This is a flowchart of the modular shading agent modeling and multi-scenario optimization and control method for light and heat balance in this invention.

[0046] Figure 2 This is a schematic diagram of the modular sunshade system and control unit in this invention;

[0047] Figure 3 This is a parameterized model and a construction diagram of the photothermal performance simulation of the modular shading system in this invention;

[0048] Figure 4 This is a spatial scene visualization diagram of the application simulation settings and multi-dimensional photothermal indicators in this invention;

[0049] Figure 5 This is a comparison chart of the optimization effects of the shading strategy under multiple scene spatial layouts in this invention;

[0050] Figure 6 This is a three-dimensional distribution diagram of the Pareto front solution of the photothermal performance error value in multiple scenarios in this invention. Detailed Implementation

[0051] The following detailed explanation, using the attached diagrams of the instruction manual as an example, illustrates the modification of the external shading system in an open office building unit:

[0052] like Figure 1As shown, this invention follows a parameterized modular shading approach—rapid agent prediction—multi-objective optimization—multi-scenario verification: First, the shading facade is discretized into independently controllable modular units, and the opening and closing variables and photothermal synergy evaluation indicators such as DA, GA, and TCP are defined. A dataset of shading opening and closing states and photothermal performance distribution is constructed through batch sample simulation. The Pix2pixHD agent model is trained to achieve rapid inference of performance results, replacing time-consuming simulations. Subsequently, the pre-trained agent model is loaded, a multi-objective function for photothermal synergy is defined, and the Pareto front solution set is output using NSGA-II iterative optimization. Finally, the optimal compromise solution is compared and selected in typical office layout scenarios such as independent partitioning, collaborative centralization, and hybrid sharing, and the optimal compromise solution is transformed into modular opening and closing control commands, thereby achieving adaptive shading optimization and control across scenarios.

[0053] In this invention, a typical open-plan office unit was selected as the verification scenario: the interior space dimensions are 10.8m wide × 8.1m deep × 3.6m high, and the exterior windows face southwest. Figure 2 As shown, the outer interface is set with a horizontal modular sunshade array, with sunshades distributed in 12×18, corresponding to 216 independently controllable units; each sunshade unit is 0.6m×0.3m in size, and its opening and closing parameters are expressed by rotation angle, which is continuously adjustable in the range of 0° to 90°.

[0054] Meanwhile, a mapping rule between the shading opening and closing degree matrix and the image representation is established: First, the rotation angle of each shading unit is linearly normalized to the opening and closing degree OR (with a value range of [0,1]), and then the OR is mapped to a gray value of [0,255] to obtain a single-channel facade grayscale image, where the pixel brightness is proportional to the OR of the corresponding shading unit; then, interpolation resampling is used to uniformly convert the facade image into 256×256 pixels, which is used as the input of the surrogate model.

[0055] The environmental performance indicators described in this invention are used for sample generation and optimization evaluation. The specific indicators are as follows:

[0056] DA (Daylight Autonomy) is a daylight adaptation index that measures the proportion of time within a specific period during which the natural illuminance at a point in an indoor space exceeds a given threshold. It is typically used for annual working hours or annual daytime hours. The calculation formula is as follows:

[0057] ;

[0058] This represents the number of hours the device has been used. This represents the simulation time step; For effective daylight levels; It is the illuminance threshold used to assess the sufficiency of sunlight; It is the first The sunlight satisfaction indicator (satisfaction mark) at each time step means whether the sunlight level range requirement is met at that time step.

[0059] Glare Autonomy (GA) is used to assess the proportion of time spent experiencing visual discomfort within a given workspace and time period. It is typically assessed based on Daylight Glare Probability (DGP). DGP quantifies the probability of discomfort caused by sunlight at a specific viewpoint and time, with a value ranging from 0 to 1 (often expressed as a percentage). Because DGP is highly directional, its value varies significantly with the viewing angle. Therefore, a weighted average of the DGP values ​​from multiple directions at various measurement points is calculated. The specific formula is as follows:

[0060] ;

[0061] Represents the total number of space sensors distributed. The number of simulation time steps for each point; A binary function used to determine a specific sensor point. and time step Does the solar glare probability (DGP) at a given location meet acceptable glare conditions?

[0062] The Thermal Comfort Percentage (TCP) index, based on the adaptive comfort model in ASHRAE-55, measures the proportion of time within a given time period that a space meets thermal comfort conditions (e.g., temperature is within an acceptable range). The specific calculation formula is as follows:

[0063] ;

[0064] To assess the total number of time steps, This represents the indoor working temperature at each step. It is the first Each time step has a thermal comfort satisfaction indicator variable (satisfaction marker), which means whether the thermal comfort temperature range requirement is met at that time step.

[0065] This invention constructs a paired training sample set and uses EPW data from a typical meteorological year for year-round analysis. For example... Figure 3As shown, the indoor measuring point grid is arranged with 13×18 measuring points, the measuring point height is 0.6m above the ground, the grid spacing is 0.6m×0.6m, and a total of 234 measuring points are output. Annual environmental performance simulations were performed for each OR scheme based on the Rhino-Grasshopper platform: Radiance and EnergyPlus were used for annual simulations; details can be found here. Figure 4 Regarding the simulation threshold settings, this invention adopts the following: DA threshold illuminance of 450 lux; GA threshold of 30%; TCP threshold of 85%; thermal comfort uses ASHRAE 55 combined with an adaptive comfort model, setting a neutral temperature offset of ±2.5℃, and considering mixed ventilation mode conditions. Meanwhile, the indoor personnel and equipment boundary conditions adopt a typical office schedule: weekdays 09:00–17:00.

[0066] In this invention, after completing a year-long simulation, a total of 1000 image pairs were obtained, of which 900 pairs were randomly selected as the training set and 100 pairs were used as the test set. The conditional generative adversarial network Pix2pixHD was used to train three proxy models: DA, GA, and TCP. The optimal training hyperparameters included: input and output image size of 256×256, learning rate of 0.0002, batch size of 1, and the use of a multi-scale discriminator and feature matching.

[0067] This invention employs scenario-based comparative experiments to verify adaptive control in multifunctional spaces. The interior layout is set up as three representative multi-scenario types: independent zoning, collaborative centralized, and hybrid shared. The independent zoning type emphasizes the functional separation of independent workstations from public areas; the collaborative centralized type centers on a public communication area surrounded by independent workstations; and the hybrid shared type interweaves independent and communication functions in the space to accommodate flexible work modes. Simultaneously, to achieve the design optimization of the "multifunctional area differentiation goal," this embodiment encodes the interior floor plan layout as a scenario zoning image: white (255,255,255) represents independent work areas, gray (130,130,130) represents public communication areas, and black (0,0,0) represents non-computing areas. Based on standards such as LEED v4.1 and ASHRAE 55, differentiated performance thresholds were set for different functional areas. For example, the daylight adaptation index for independent work areas is ≥70%, the glare adaptation index is ≤90%, and the thermal comfort percentage index is ≥88%; the daylight adaptation index for public communication areas is ≥50%, the glare adaptation index is ≤75%, and the thermal comfort percentage index is ≥83%.

[0068] In terms of setting target constraints, this invention achieves consistency from visualization to computation through image-based target mapping: using tools such as Adobe Photoshop and Script, the grayscale values ​​of the target performance distribution image are manually or semi-automatically adjusted to the spatial distribution corresponding to the target threshold; then, the output image of the surrogate model is compared pixel by pixel with the target image to obtain the allowable error rate within the target area as the optimization target.

[0069] In the multi-scenario optimization and control stage of this invention, three pre-trained surrogate models (DA, GA, and TCP) are loaded. For any candidate shading scheme individual, its gene encoding is a 216-dimensional vector (corresponding to a 12×18 shading unit OR). This vector is decoded into an OR matrix and mapped to generate a 256×256 grayscale image of the facade status. These images are then input into the three surrogate models respectively to quickly infer the three types of performance distribution images.

[0070] Subsequently, pixel-by-pixel error calculation, normalization, and standardization are performed within the target area to form a multi-objective evaluation vector. This embodiment introduces a comprehensive error function. As a comprehensive target value, it is used to quantify the overall deviation of candidate solutions from multiple indicators such as daylight adaptation, glare adaptation, and thermal comfort percentage. Its calculation method is as follows:

[0071] ;

[0072] in, It is the solution of the candidate scheme, corresponding to the set of opening and closing parameters of the modular shading system; , , Preset weights; The target error in the daylight adaptive index; The target error in the glare adaptive index; This represents the target error in the thermal comfort adaptive index.

[0073] Representation scheme The overall error (overall target value) in the current scenario, where The default shading opening and closing scheme is usually set (e.g., uniform opening, regular opening, or a preset initial state). This represents the optimal solution obtained through optimization. The overall error (overall target value) under the same scenario, where The optimal shading opening / closing degree scheme is usually obtained through multi-objective optimization (such as NSGA-II) and selected according to decision rules.

[0074] By comparison and The overall error change before and after optimization was analyzed, and then facade input schemes based on the designer's experience with different levels of precision were introduced. Calculate separately Error variation of the control scheme relative to the optimal scheme The sensitivity to input complexity under non-optimal input conditions and the relative merits of different solutions are evaluated as follows:

[0075] ;

[0076] During implementation, if The tendency to increase indicates that in this type of scenario, the empirical input scheme deviates significantly from the model prediction result, and the model prediction scheme has better reliability; if A significant decrease indicates that the empirical input scheme and the model prediction results are close in this type of scenario. By analyzing different... of Comparisons can be used to: ① evaluate the system's sensitivity to input complexity under non-optimal inputs; ② identify the relative advantages and disadvantages of different input schemes and the room for improvement; ③ provide quantitative basis for the final compromise solution selection, robustness verification, and scenario-based decision-making for the Pareto solution set.

[0077] In this invention, the DEAP library is used to implement NSGA-II as a multi-objective evolutionary optimization solver. Its key parameters are preferably set as follows: dimension 216, population size 100, number of iterations 50, individual mutation probability 0.05, and polynomial mutation parameter 20. A global minimization search is performed on three objectives (light adaptation index grayscale error, glare adaptation index grayscale error, and thermal comfort percentage index grayscale error). In each iteration, the process loops through individual decoding → image generation → surrogate inference → error assessment → non-dominated sorting and crowding selection → crossover mutation → population update until the termination condition is met, at which point multiple Pareto optimal solution sets are output.

[0078] Effect verification:

[0079] Figure 5 The performance of the optimal solution (based on specific weight selection) obtained by the method of this invention was compared with that of two benchmark solutions (based on empirical judgment of different control accuracies) in three scenarios. It can be seen that the optimized solution can effectively balance the needs of each area, provide sufficient DA in the independent working area, and suppress potential glare and overheating risks in the strong sunlight area by adjusting the shading, thus achieving a synergistic improvement in photothermal performance.

[0080] In summary, this invention provides a complete technical solution from efficient modeling to intelligent decision-making, offering a powerful tool for the design and operation of modular dynamic shading systems, and significantly improving the overall performance of buildings in terms of comfort, health, and energy conservation.

[0081] The calibrated surrogate model in this invention can accurately reproduce the spatial performance distribution characteristics and reduce the time required for a single evaluation by nearly 90% compared to traditional simulation.

[0082] The results show that data-driven facade configuration is better able to grasp the light and heat trade-off than empirical solutions; the framework can serve as a scalable visual spatial decision support tool for adaptive modular facade renovation and can be extended to more climate, geometric and performance objectives.

[0083] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A proxy modeling method for a modular shading system oriented towards photothermal balance, characterized in that, Includes the following steps: S1. Establish a parameterized physical model of the modular shading system, divide the shading system into several independently controllable shading units, and set opening and closing degree parameters to represent the opening state of each shading unit, forming a shading opening and closing degree matrix. S2. Normalize and image-map the shading opening and closing degree matrix to generate facade state images reflecting different shading states; S3. For the opening and closing degree matrix corresponding to multiple sets of facade state images, perform annual light and heat environment simulation to obtain indoor space performance distribution data that includes at least daylighting adaptive index, glare adaptive index and thermal comfort percentage index. Normalize and rasterize the indoor space performance distribution data to generate performance distribution images that correspond one-to-one with the facade state images, thereby constructing a pairwise training sample set composed of shading state images and performance distribution images. S4. Divide the paired training sample set into a training set and a test set; S5. The conditional generative adversarial network is used to train the pairwise training sample set, so that the generator of the conditional generative adversarial network learns the mapping relationship from the facade state image to the performance distribution image, and obtains surrogate models for predicting the daylight adaptation index, glare adaptation index and thermal comfort percentage index in step S3 respectively. S6. Quantitatively evaluate the trained surrogate model using the test set and select the optimal model. Determine the parameters of the optimal model and write them into the surrogate model parameter library for subsequent rapid prediction and optimization.

2. The proxy modeling method for modular shading systems oriented towards photothermal balance according to claim 1, characterized in that, The modular sunshade unit is a horizontal sunshade unit, and the opening and closing parameter of each unit is its continuously adjustable rotation angle within a preset angle range. The shading opening and closing degree matrix is ​​a two-dimensional matrix with multiple rows and columns, corresponding to multiple dynamic shading units; The facade status image is specifically generated by: normalizing the opening degree of each dynamic shading unit and converting it into grayscale pixel values ​​according to a preset mapping rule; generating a single-channel facade status image based on the spatial layout of the shading units, so that the image pixel values ​​and the opening degree of the corresponding shading units maintain a monotonic correspondence and a one-to-one spatial correspondence.

3. The proxy modeling method for a modular shading system oriented towards photothermal balance according to claim 2, characterized in that, In step S3, the annual solar thermal environment simulation uses typical meteorological year data for year-round simulation analysis; multiple measuring points are arranged at preset intervals and heights in the indoor simulation scene to form a measuring point grid; Three sets of regular measurement point grids are arranged on the indoor working surface or evaluation plane. The measurement points are located at a preset height, and the spacing between the measurement points is the preset grid spacing. The light environment index and thermal environment index at each measurement point are collected or calculated to form spatial discrete data for generating a performance distribution image. The index values ​​of each measurement point are mapped to the gray values ​​of the corresponding pixels in the performance distribution image according to their spatial coordinates. The regular grid measurement points are rectangular grids or equidistant grids, and their grid density is preset according to the spatial scale and calculation accuracy requirements.

4. The proxy modeling method for a modular shading system oriented towards photothermal balance according to claim 1, characterized in that, In step S3, the paired training sample set contains a predetermined number of sample pairs; The sample pairs are divided into a training set and a test set, and the training set and the test set are uniformly distributed within the range of the shading unit opening value.

5. The proxy modeling method for a modular shading system oriented towards photothermal balance according to claim 1, characterized in that, In step S5, the conditional adversarial network is a Pix2pixHD model, and the training process uses a multi-scale discriminator and feature matching loss for training; the input to the training process is a single-channel grayscale image, and no instance mapping is performed.

6. A multi-scenario optimization and control method for a modular shading system oriented towards photothermal balance, characterized in that, The proxy model trained based on the proxy modeling method for modular shading systems oriented towards photothermal balance as described in any one of claims 1 to 4 includes the following steps: A. Load at least three pre-trained surrogate models, corresponding to the daylight adaptation index, glare adaptation index, and thermal comfort percentage index, respectively, to quickly infer the spatial distribution of multiple indices of candidate shading opening and closing schemes. B. Based on the functional layout of the target indoor space, generate scene partition images, which at least include the target area and non-target areas; C. Based on the preset performance standards, set differentiated performance thresholds for target areas of different functional types, and generate target performance distribution images corresponding to each performance index based on the performance thresholds; D. For a candidate shading opening and closing scheme, after converting it into a facade state image, input it into the three surrogate models respectively, and quickly infer the corresponding prediction performance distribution image; E. Within the target area, the predicted performance distribution image is compared pixel by pixel with the corresponding target performance distribution image, and the error rate of each performance index is calculated respectively. F. Based on the error rate of each performance indicator, construct a multi-objective evaluation vector; G. Employ a multi-objective evolutionary algorithm to iteratively optimize by minimizing the multi-objective evaluation vector, thereby obtaining a set of Pareto-optimal shading opening and closing schemes. H. Select the final implementation scheme from the set of Pareto optimal shading opening and closing schemes, and output the corresponding shading unit control command.

7. The method for multi-scenario optimization and control of a modular shading system for photothermal balance according to claim 6, characterized in that: In step B, different functional zones are identified using different grayscale values ​​or color values ​​in the scene partitioning image; In step C, the differentiated performance thresholds include at least the thresholds for different daylight adaptation indicators, glare adaptation indicators, and thermal comfort percentage indicators for independent work areas and public communication areas, respectively.

8. The method for multi-scenario optimization and control of a modular shading system for photothermal balance according to claim 7, characterized in that: In step E, the error rate of the performance index The calculation formula is: ； Where m represents the performance metric category index; The distribution image of the prediction performance output by the surrogate model; A pre-defined target performance distribution image; Let be the normalization scaling factor, representing the th The range of grayscale values ​​for each indicator image is used to normalize the error. The set of all pixels in the entire performance distribution image; This is the weighting function, used to express that different functional partitions are assigned different weights.

9. The method for multi-scenario optimization and control of a modular shading system for photothermal balance according to claim 8, characterized in that, In step F, when constructing the multi-objective evaluation function, the error rates of each performance index are normalized or standardized, and a weighted summation method is used to form a comprehensive target value. The weight coefficients are preset according to the scenario requirements.

10. The method for multi-scenario optimization and control of a modular shading system for photothermal balance according to claim 9, characterized in that, In step G, the multi-objective evolutionary algorithm adopts the NSGA-II algorithm.

Images