A BIM building facade dynamic light-heat balance design method and system

Abstract

The application discloses a BIM building facade dynamic light-heat balance design method and system, and belongs to the technical field of building energy-saving design. The facade component information is parsed from a BIM model, azimuth weight factors and floor height attenuation factors are calculated according to azimuth angles and floor heights, and a space-time decomposition vector is constructed to form a facade design parameter set. Latin hypercube sampling is used to generate training samples, and a double-channel time series convolutional neural network proxy model is constructed, which includes an energy consumption prediction channel and a light environment prediction channel. Then, a multi-objective optimization solution is carried out to minimize the annual energy consumption of the building and maximize the daylighting effective time ratio as the optimization target, and a Pareto optimal solution set is generated. Based on the light-heat coupling index, the dynamic angle of the sunshade board at each time is calculated, and an annual hourly sunshade control strategy is generated. The application realizes the dynamic balance of the daylighting demand and the heat insulation demand of the building facade, and improves the optimization efficiency and the design quality.

Description

Technical Field

[0001] This invention belongs to the field of building energy-saving design technology, specifically relating to a BIM-based dynamic light and heat balance design method and system for building facades. Background Technology

[0002] Building facades are the primary interface for heat exchange and light transmission between buildings and the external environment, and their design directly impacts energy consumption and indoor lighting quality. The development of Building Information Modeling (BIM) technology has provided a digital data foundation for architectural design, enabling designers to extract geometric information and attribute parameters of building components from 3D models. However, effectively combining BIM with optimization algorithms to automate the optimization of facade design parameters remains a significant technical challenge in the field of energy-efficient building design.

[0003] There is an inherent contradiction between the needs for lighting and insulation in building facade design. Increasing the window-to-wall ratio is beneficial for introducing natural light and reducing artificial lighting energy consumption, but it leads to increased solar radiation heat gain and air conditioning load in summer. Conversely, decreasing the window-to-wall ratio is beneficial for blocking solar radiation and reducing air conditioning energy consumption, but it results in insufficient indoor lighting and increased artificial lighting energy consumption. Existing technologies mostly use empirical methods or single-objective optimization methods for facade design, making it difficult to scientifically quantify the balance between lighting and insulation, resulting in designs that cannot simultaneously achieve energy-saving goals and lighting quality. The amount of solar radiation received by different areas of a building facade varies significantly. Facades facing different directions receive different annual cumulative solar radiation due to the sun's trajectory, and facades on different floors receive different intensities of solar radiation due to differences in surrounding shading conditions. Existing technologies typically treat the building facade as a whole, using the same design parameters to optimize all facade units, ignoring the influence of orientation and spatial location on solar radiation reception, making it difficult to provide differentiated design solutions for the actual radiation conditions of each facade unit. Multi-objective optimization of building facades requires performance evaluation of a large number of candidate design schemes. Existing technologies rely on specialized building energy consumption simulation software and lighting environment simulation software for performance calculations. Each simulation is time-consuming, and multi-objective optimization algorithms require evaluating thousands of candidate solutions, resulting in excessively long overall optimization calculation times and failing to meet the timeliness requirements of engineering designs. While surrogate model technology can improve calculation speed, existing simple neural network models such as multilayer perceptrons struggle to capture the temporal dependence between building energy consumption and meteorological conditions, leading to limited prediction accuracy and affecting the reliability of optimization results. Building shading components are crucial for regulating the indoor light and heat environment, and their control strategies directly impact energy-saving effects and lighting quality. Existing shading control methods either employ fixed-angle strategies, failing to adapt to dynamic changes in solar radiation, or use geometric tracking strategies based on solar altitude angles, considering only the geometric relationship of the sun's position without comprehensively considering the actual needs of the indoor light and heat environment. Existing control strategies struggle to adaptively balance insulation and lighting requirements across different seasons and times of the year, hindering the full realization of building energy-saving potential.

[0004] Multi-objective evolutionary optimization algorithms are effective tools for solving multi-objective optimization problems of building facades. The classic NSGA-II algorithm uses non-dominated sorting and crowding calculation to maintain population diversity, but its crowding calculation only considers the distance between adjacent solutions, resulting in insufficient control over the global distribution of the Pareto front. The fixed mutation probability leads to high-quality individuals potentially losing valuable genes due to excessive mutation, while low-quality individuals may get trapped in local optima due to insufficient mutation. These shortcomings of existing algorithms affect the uniformity of the Pareto front distribution and the convergence speed, limiting the quality of the optimization results. Summary of the Invention

[0005] To address the problems existing in the background technology, the present invention provides a BIM-based dynamic light and heat balance design method for building facades, comprising the following steps:

[0006] S1. Extract the facade component information from the BIM model, construct the spatiotemporal decomposition vector according to the orientation and spatial location of each facade unit, and calculate the orientation weight factor and floor height attenuation factor of each facade unit to obtain the weighted facade design parameter set.

[0007] S2. The Latin hypercube sampling method is used to generate training samples in the design parameter space. The building energy consumption simulation software and the light environment simulation software are called to obtain the sample labeling data. A dual-channel temporal convolutional neural network surrogate model containing energy consumption prediction channel and light environment prediction channel is constructed and trained.

[0008] S3. With the optimization objectives of minimizing the building's annual energy consumption and maximizing the effective daylighting time ratio, the NSGA-II algorithm with adaptive crowding degree and dynamic variation probability is used to solve the multi-objective optimization problem and generate the Pareto optimal solution set.

[0009] S4. Based on the facade design parameters selected from the Pareto optimal solution set, calculate the dynamic sunshade angle of each facade unit at all times throughout the year based on the photothermal coupling index, and generate an hourly sunshade control strategy for the whole year.

[0010] Furthermore, step S1 includes:

[0011] S11. Parse the BIM model file in IFC format, extract the exterior wall area, window area, initial angle of sunshade and visible light transmittance of glass for each facade unit, and calculate the azimuth angle and relative height of the floor for each facade unit based on the spatial coordinates of the components.

[0012] S12. Divide all facade units into eight azimuth zones according to the azimuth angle, with each azimuth zone covering a 45-degree range.

[0013] S13. Obtain the annual cumulative solar radiation for each directional area, and calculate the azimuth weight factor for each facade unit using the following formula:

[0014] ;

[0015] in, The orientation weight factor for the current facade unit is dimensionless. The annual cumulative solar radiation (kWh / m²) of the current facade unit's location area. 2 ); The maximum annual cumulative solar radiation (kWh / m²) across all azimuth regions. 2 );

[0016] S14. Calculate the floor height attenuation factor for each facade unit using the following formula:

[0017] ;

[0018] in, The floor height attenuation factor for the current facade unit is dimensionless. This is an empirical coefficient, with a value ranging from 0.05 to 0.15; The natural logarithm operator; The relative height (m) of the floor where the current facade unit is located; For reference height, the value is 10 (m);

[0019] S15. Combine the geometric parameters of each facade unit with the corresponding orientation weighting factor and floor height attenuation factor to construct a spatiotemporal decomposition vector, forming a facade design parameter set.

[0020] Furthermore, step S2 includes:

[0021] S21. Set the window-to-wall ratio to a range of 0.2 to 0.7, the sunshade angle to a range of 0 degrees to 90 degrees, and the visible light transmittance of the glass to a range of 0.3 to 0.8. Use the Latin hypercube sampling method to generate 1500 to 2500 sampling points.

[0022] S22. For each sampling point, call EnergyPlus software to simulate the annual energy consumption and obtain the cooling and heating load values. Call Radiance software to simulate the light environment and count the percentage of hours when the indoor work surface illuminance is in the range of 300 lux to 3000 lux as the effective lighting time ratio.

[0023] S23. Construct a dual-channel temporal convolutional neural network. The energy consumption prediction channel is set with four dilated convolutional layers with dilation factors of 1, 2, 4, and 8 respectively. The light environment prediction channel is set with three dilated convolutional layers with dilation factors of 1, 2, and 4 respectively.

[0024] S24. Divide the sampling points into training, validation and test sets in an 8:1:1 ratio. Train the network using the Adam optimizer until the loss on the validation set no longer decreases. After verifying on the test set that the decision coefficients of both channels are greater than 0.90, the model training is complete.

[0025] Furthermore, step S3 includes:

[0026] S31. Set the population size to 100 to 200, the maximum number of iterations to 200 to 500, and randomly initialize the facade design parameters of each individual in the population.

[0027] S32. Call the dual-channel temporal convolutional neural network surrogate model to evaluate the fitness of each individual and obtain the annual energy consumption prediction value and the effective lighting time ratio prediction value.

[0028] S33. After performing non-dominated ranking on the population, calculate the adaptive crowding degree of each individual within the same non-dominated level using the following formula:

[0029] ;

[0030] in, The adaptive crowding degree of the current individual is dimensionless. The global crowding degree of the current individual is obtained by normalizing and summing the differences in target values ​​between adjacent individuals, and is dimensionless. For exponential function operators; This is an adaptive factor, with a value range of 0.5 to 2.0; Let this be the current iteration algebra; The maximum number of iterations; The local crowding degree of the current individual is calculated by the K-nearest neighbor method, where K is 5 and is dimensionless.

[0031] S34. Calculate the dynamic mutation probability of each individual using the following formula:

[0032] ;

[0033] in, The mutation probability of the current individual is dimensionless. The minimum mutation probability is set to 0.01. The maximum mutation probability is set to 0.20. The current individual's non-dominant level; Population size; To adjust the index, the value ranges from 1.5 to 2.5;

[0034] S35. Perform tournament selection and simulate binary crossover to generate offspring, perform polynomial mutation according to the dynamic mutation probability, merge the parent and offspring, and perform environment selection according to non-dominance level and adaptive crowding degree.

[0035] S36. Repeat steps S32 to S35 until the maximum number of iterations is reached, and output all non-dominated solutions as the Pareto optimal solution set.

[0036] Furthermore, step S4 includes:

[0037] S41. Select a set of facade design parameters from the Pareto optimal solution set and obtain the annual hourly solar radiation data of the building location;

[0038] S42. For each moment of the year, calculate the photothermal coupling index using the following formula:

[0039] ;

[0040] in, The photothermal coupling index at the current moment is dimensionless. This is a dimensionless coefficient representing the weighting factor for insulation requirements. The heat gain from solar radiation at the current moment (W / m²) 2 ); The design maximum daily solar radiation heat gain (W / m 2 ); This is a dimensionless weighting coefficient for lighting requirements. This is the estimated indoor daylight illuminance (lux) at the current moment. The target daylight illuminance value is between 300 and 500 (lux).

[0041] S43. The thermal insulation requirement weighting coefficient and the lighting requirement weighting coefficient shall be automatically adjusted according to the accumulated days of the year using the following formula:

[0042] ;

[0043] in, This is the weighting coefficient for the insulation demand on that day, and it is dimensionless. This is the sine function operator; Pi; It represents the accumulated days of a year, with a value ranging from 1 to 365. This is a dimensionless weighting coefficient representing the daily lighting demand.

[0044] S44. Calculate the sunshade angle based on the sign of the photothermal coupling index: when hour,

[0045] ;

[0046] when hour,

[0047] ;

[0048] in, The current angle (in degrees) of the sunshade; This is the minimum angle of the sunshade, with a value of 0 (degrees). The maximum angle of the sunshade is 90 degrees. The hyperbolic tangent function operator; The response sensitivity coefficient ranges from 2.0 to 5.0. This represents the absolute value of the photothermal coupling index;

[0049] S45. Repeat steps S42 to S44 for all 8760 hours of the year to generate the hourly sunshade angle sequence for the whole year.

[0050] The present invention also provides a BIM building facade dynamic light and heat balance design system, including a BIM data parsing module, a proxy model module, a multi-objective optimization module, a dynamic shading calculation module and a central control module;

[0051] The BIM data parsing module includes an IFC parsing unit, a geometric calculation unit, and a parameter encoding unit connected in sequence. The IFC parsing unit is used to read the BIM model file and extract the attributes of the facade components. The geometric calculation unit is used to calculate the azimuth angle and floor height and perform azimuth zoning. The parameter encoding unit is used to calculate the azimuth weight factor and floor height attenuation factor and output the facade design parameter set.

[0052] The proxy model module includes a data preprocessing unit and an energy consumption prediction network unit and a light environment prediction network unit connected in parallel with the output of the data preprocessing unit; the energy consumption prediction network unit adopts a four-layer dilated convolution structure; the light environment prediction network unit adopts a three-layer dilated convolution structure.

[0053] The multi-objective optimization module includes a population management unit, a genetic operation unit, and a fitness evaluation unit; the population management unit and the genetic operation unit are bidirectionally connected; the input of the fitness evaluation unit is connected to the population management unit, and the output of the fitness evaluation unit is connected to the surrogate model module to invoke the surrogate model for performance prediction; the population management unit is used to perform non-dominated sorting and calculate adaptive crowding; the genetic operation unit is used to perform selection, crossover, and dynamic probabilistic mutation operations;

[0054] The dynamic shading calculation module includes a meteorological data interface unit, a photothermal coupling calculation unit, and an angle solving unit connected in sequence; the photothermal coupling calculation unit is used to calculate the photothermal coupling index and adjust the seasonal weighting coefficient according to the annual accumulated days; the angle solving unit is used to calculate the shading angle according to the photothermal coupling index.

[0055] The central control module is communicatively connected to the BIM data parsing module, the proxy model module, the multi-objective optimization module, and the dynamic shading calculation module, respectively, and is used to call each module in sequence and transmit data.

[0056] In a preferred embodiment, the output of the parameter encoding unit is connected to the first input of the central control module, and the first output of the central control module is connected to the input of the data preprocessing unit; the outputs of the energy consumption prediction network unit and the light environment prediction network unit are both connected to the input of the fitness evaluation unit.

[0057] In a preferred embodiment, the output of the population management unit is connected to the input of the fitness evaluation unit to transmit the individuals to be evaluated, and the output of the fitness evaluation unit is connected to the input of the population management unit to return the fitness value; the genetic operation unit receives the population data output by the population management unit after non-dominated sorting and crowding calculation, and returns the mutated offspring population to the population management unit.

[0058] In a preferred embodiment, the input end of the meteorological data interface unit is connected to an external meteorological database to obtain hourly solar radiation data and solar position data throughout the year; the output end of the angle solving unit is connected to the second input end of the central control module to output the hourly shading control strategy throughout the year.

[0059] In a preferred embodiment, the central control module further includes a data caching unit, which is connected to the output of the BIM data parsing module, the output of the multi-objective optimization module, and the output of the dynamic shading calculation module, respectively, and is used to store the intermediate calculation results of each module.

[0060] The beneficial effects achieved by this invention are as follows:

[0061] The BIM-based dynamic light and heat balance design method and system for building facades provided by this invention achieves differentiated characterization and processing of each unit of the building facade by introducing a spatiotemporal decomposition vector construction mechanism. This invention divides the facade units into multiple azimuth zones based on their azimuth angles and calculates azimuth weight factors based on the annual cumulative solar radiation of each zone, giving facade units with higher solar radiation a higher processing priority during optimization. This invention also introduces a floor height attenuation factor, using a logarithmic function to describe the nonlinear influence of floor height on solar radiation reception, enabling high-rise facade units to obtain weight coefficients that match their actual radiation conditions. The spatiotemporal decomposition vector combines facade geometric parameters with dual weight factors, providing refined input data for subsequent multi-objective optimization algorithms and overcoming the shortcomings of existing technologies that treat the building facade as a whole while ignoring differences in orientation and spatial location.

[0062] This invention employs a dual-channel temporal convolutional neural network as a surrogate model, constructing separate energy consumption prediction and lighting environment prediction channels to replace traditional physical simulation software for performance evaluation. The energy consumption prediction channel uses a multi-layer dilated convolutional structure, expanding the receptive field by progressively increasing the dilation factor layer by layer, effectively capturing the long-term temporal dependency between building energy consumption and annual meteorological conditions. The lighting environment prediction channel uses a relatively shallow network structure, adapting to the relatively simple mapping relationship between daylighting performance and facade parameters. The dual-channel parallel structure allows each prediction task to utilize a targeted network depth, improving computational efficiency while ensuring prediction accuracy. Compared to traditional physical simulation software, the surrogate model's prediction speed is increased by several orders of magnitude, enabling large-scale multi-objective optimization calculations and solving the problem of low optimization efficiency caused by excessively long simulation calculation times in existing technologies.

[0063] This invention improves the classic NSGA-II algorithm in two aspects, enhancing the solution quality and convergence speed of multi-objective optimization. First, it introduces an adaptive crowding calculation mechanism, dynamically fusing global and local crowding through an exponential decay function. This allows the algorithm to focus on global search in the early stages of optimization to explore a broad parameter space, and on local refinement in the later stages to improve the uniformity of the Pareto front distribution. Second, it introduces a dynamic mutation probability mechanism, adaptively adjusting the mutation probability based on the non-dominated level of an individual. This ensures that high-quality individuals receive a lower mutation probability to protect superior genes, while low-quality individuals receive a higher mutation probability to enhance their exploration ability. This improves both the uniformity of the Pareto front distribution and convergence, overcoming the shortcomings of the standard NSGA-II algorithm, which is prone to getting trapped in local optima and has an uneven Pareto front distribution.

[0064] This invention proposes a photothermal coupling index and its driving method for calculating dynamic shading angles, enabling intelligent responses of shading components to indoor photothermal environment demands. The photothermal coupling index comprehensively considers the current solar radiation gain and indoor lighting illuminance deviation. When insulation demand is dominant, the shading angle is increased to block solar radiation; when lighting demand is dominant, the shading angle is decreased to introduce natural light. This invention employs a seasonal weight adaptive adjustment mechanism, where the weights for insulation and lighting demands automatically change with the accumulated days of the year, increasing the insulation weight in summer and the lighting weight in winter, aligning with the actual needs of building operation throughout the year. This invention uses a hyperbolic tangent function to calculate the shading angle; its saturation characteristics ensure the shading angle stabilizes under extreme conditions, avoiding mechanical losses and control oscillations caused by frequent adjustments. Compared to fixed-angle shading and geometric tracking shading, this invention's dynamic shading control method can improve indoor lighting quality and thermal comfort while reducing overall building energy consumption, achieving a dynamic balance between insulation and lighting demands. Attached Figure Description

[0065] Figure 1 Box plots comparing the Pareto front distribution quality of Example 1 and Comparative Example 1; subplot (a) compares the Spacing index of Example 1 and Comparative Example 1, and subplot (b) compares the Hypervolume index of Example 1 and Comparative Example 1.

[0066] Figure 2 The following is a grouped bar chart comparing the prediction accuracy of the surrogate models of Example 1 and Comparative Example 2; wherein, sub-chart (a) compares the energy consumption prediction accuracy of Example 1 and Comparative Example 2, and sub-chart (b) compares the daylighting prediction accuracy of Example 1 and Comparative Example 2.

[0067] Figure 3 The bar charts show the comparison of the shading control effects of Example 1 with Comparative Examples 3 and 4. Sub-chart (a) compares the annual comprehensive energy consumption of Example 1 with Comparative Examples 3 and 4; sub-chart (b) compares the effective lighting time ratio of Example 1 with Comparative Examples 3 and 4; sub-chart (c) compares the thermal comfort compliance hours of Example 1 with Comparative Examples 3 and 4; and sub-chart (d) compares the comprehensive performance scores of Example 1 with Comparative Examples 3 and 4.

[0068] Figure 4 This is a line graph showing the hourly variation curves of the sunshade angle for the south-facing facade unit of Example 1 on three typical days: the summer solstice, the winter solstice, and the spring equinox.

[0069] Figure 5 The figures show the optimization effects of Examples 1, 2, and 3 in different climate zones. Subfigure (a) compares the annual comprehensive energy consumption and effective daylighting time ratio of Examples 1, 2, and 3, while subfigure (b) compares the Pareto solution set size and optimization time of Examples 1, 2, and 3.

[0070] Figure 6 The graph shows the convergence process curves of the multi-objective optimization in Example 1; subgraph (a) is the convergence curve of the Hypervolume index in the optimization process of Example 1, and subgraph (b) is the convergence curve of the optimal individual energy consumption value and the ratio of effective lighting time in the optimization process of Example 1.

[0071] Figure 7 This is a flowchart of a BIM-based dynamic light and heat balance design method for building facades according to the present invention. Detailed Implementation

[0072] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] This invention provides a BIM-based dynamic light and heat balance design system for building facades. The system includes a BIM data parsing module, a proxy model module, a multi-objective optimization module, a dynamic shading calculation module, and a central control module. These modules communicate with each other via data interfaces, jointly performing functions such as extracting facade design parameters, predicting performance, solving optimization problems, and generating shading control strategies.

[0074] The BIM data parsing module implements the function of step S1, responsible for extracting facade component information from the BIM model and constructing a spatiotemporal decomposition vector. This module includes an IFC parsing unit, a geometric calculation unit, and a parameter encoding unit connected in sequence. The input of the IFC parsing unit receives the BIM model file in IFC format. It integrates an IFC file parser, which can read and parse the entity objects and attribute data in the IFC file. The IFC parsing unit traverses the IfcWall and IfcWindow entities in the model, extracts parameters such as exterior wall area, window area, sunshade attributes, and glass transmittance, and outputs the extraction results to the geometric calculation unit. The geometric calculation unit receives the component data output by the IFC parsing unit, calculates the azimuth angle and relative floor height of each facade unit based on the spatial coordinates of the components, and performs azimuth zoning based on the azimuth angle, dividing all facade units into 8 azimuth zones. The output of the geometric calculation unit is connected to the input of the parameter encoding unit. The parameter encoding unit receives the zoning results and geometric parameters output by the geometric calculation unit, obtains the annual cumulative solar radiation data for each directional zone, calculates the azimuth weight factor and floor height attenuation factor for each facade unit, and combines the geometric parameters and weight factors to construct a spatiotemporal decomposition vector, forming a facade design parameter set which is then output to the central control module.

[0075] The proxy model module implements the function of step S2, responsible for quickly predicting the building's annual energy consumption and effective daylighting time ratio. This module includes a data preprocessing unit and an energy consumption prediction network unit and a light environment prediction network unit connected in parallel with the output of the data preprocessing unit. The input of the data preprocessing unit is connected to the central control module, receiving the facade design parameters to be predicted. After normalizing the input parameters, it outputs them to the energy consumption prediction network unit and the light environment prediction network unit respectively. Normalization maps the value range of each parameter to the interval between 0 and 1, which is beneficial to the training convergence and prediction stability of the neural network. The energy consumption prediction network unit adopts a 4-layer dilated convolutional structure with dilation factors of 1, 2, 4, and 8 respectively. Its output is connected to a fully connected layer, outputting the predicted annual energy consumption value. The light environment prediction network unit adopts a 3-layer dilated convolutional structure with dilation factors of 1, 2, and 4 respectively. Its output is connected to a fully connected layer, outputting the predicted effective daylighting time ratio. The outputs of both the energy consumption prediction network unit and the light environment prediction network unit are connected to the fitness evaluation unit of the multi-objective optimization module, which is used to provide rapid performance evaluation for candidate solutions during the optimization process.

[0076] The multi-objective optimization module implements the function of step S3, responsible for solving the Pareto optimal solution set using the improved NSGA-II algorithm. This module includes a population management unit, a genetic operation unit, and a fitness evaluation unit. The population management unit and the genetic operation unit are bidirectionally connected for the exchange and transfer of population data. The population management unit is responsible for initializing the population, performing non-dominated sorting, and calculating adaptive crowding. At the beginning of each generation, the population management unit performs non-dominated sorting on all individuals in the population, dividing individuals into different non-dominated levels according to dominance relationships; then, it calculates the crowding value of each individual within the same non-dominated level according to the adaptive crowding formula; finally, it outputs the sorting and crowding calculation results to the genetic operation unit. The genetic operation unit is responsible for performing selection, crossover, and dynamic probability mutation operations. The genetic operation unit calculates the dynamic mutation probability based on the non-dominated level of the individuals, performs tournament selection and simulated binary crossover to generate offspring, and performs polynomial mutation on the offspring according to the dynamic mutation probability, returning the mutated offspring population to the population management unit. The fitness evaluation unit's input is connected to the population management unit, receiving the parameters of the individuals to be evaluated. Its output is connected to the surrogate model module, which calls the surrogate model to obtain the predicted energy consumption and effective light-time ratio for each individual. The evaluation results are returned to the population management unit for non-dominated ranking and environmental selection. When the maximum number of iterations is reached, the population management unit outputs all individuals at the first non-dominated level in the current population to the central control module as the Pareto optimal solution set.

[0077] The dynamic shading calculation module implements the function of step S4, responsible for calculating the solar-thermal coupling index and generating an hourly shading control strategy for the entire year. This module includes a meteorological data interface unit, a solar-thermal coupling calculation unit, and an angle calculation unit connected in sequence. The input of the meteorological data interface unit is connected to an external meteorological database to obtain hourly solar radiation data and solar position data for the building's location throughout the year. Meteorological data can be obtained from a typical meteorological year database or a real-time meteorological monitoring system. The meteorological data interface unit outputs the acquired meteorological data to the solar-thermal coupling calculation unit. The solar-thermal coupling calculation unit simultaneously receives the facade design parameters transmitted from the central control module, calculates the solar-thermal coupling index based on the current solar radiation gain and estimated indoor daylighting, and automatically adjusts the seasonal weighting coefficient based on the accumulated days of the year. The output of the solar-thermal coupling calculation unit is connected to the input of the angle calculation unit. The angle calculation unit selects the appropriate formula based on the sign of the solar-thermal coupling index to calculate the shading angle and outputs the calculation result. The output of the angle calculation unit is connected to the second input of the central control module, transmitting the hourly shading angle sequence for the entire year to the central control module for integrated output.

[0078] The central control module is the core scheduling unit of the entire system. It communicates with the BIM data parsing module, the proxy model module, the multi-objective optimization module, and the dynamic shading calculation module, respectively, to sequentially call each module and transmit data. The first input of the central control module is connected to the output of the parameter encoding unit of the BIM data parsing module, receiving the facade design parameter set. The first output of the central control module is connected to the input of the data preprocessing unit of the proxy model module, transmitting the design parameters to be predicted to the proxy model module during the optimization process. The second input of the central control module is connected to the output of the angle solution unit of the dynamic shading calculation module, receiving the year-round hourly shading control strategy. The central control module also includes a data caching unit, which is connected to the outputs of the BIM data parsing module, the multi-objective optimization module, and the dynamic shading calculation module, respectively. This unit stores the intermediate calculation results of each module, including the facade design parameter set, the Pareto optimal solution set, and the year-round hourly shading angle sequence. The data caching unit uses a first-in, first-out (FIFO) or least recently used (LRU) cache replacement strategy to manage storage space, ensuring data consistency and access efficiency during long-term system operation.

[0079] The system described in this invention communicates with its modules via standardized data interfaces. Each module can be deployed and upgraded independently, exhibiting excellent scalability and maintainability. The BIM data parsing module supports different versions of IFC format files, the network structure of the proxy model module can be adjusted according to prediction accuracy requirements, the algorithm parameters of the multi-objective optimization module can be configured according to the problem scale, and the dynamic shading calculation module can access different types of meteorological data sources. Through the collaborative work of these modules, the system of this invention can achieve fully automated processing from BIM model input to shading control strategy output, providing efficient technical support for the light and heat balance design of building facades.

[0080] Reference Figure 7 This invention provides a BIM-based dynamic light and heat balance design method for building facades. This method combines building information modeling technology with multi-objective optimization algorithms. It replaces traditional physical simulation calculations by constructing a dual-channel temporal convolutional neural network proxy model, which greatly improves optimization efficiency while ensuring prediction accuracy. At the same time, it introduces a light and heat coupling index to achieve dynamic control of the sunshade angle, thereby taking into account both heat insulation and lighting needs throughout the building's annual operating cycle.

[0081] The BIM-based dynamic light and heat balance design method for building facades described in this invention includes the following steps S1 to S4. Step S1 is the process of parsing facade component information from the BIM model and constructing a spatiotemporal decomposition vector. Building Information Modeling (BIM) is an engineering data model based on three-dimensional digital technology, which uses the IFC format to store the geometric and attribute information of building components. IFC is an internationally recognized building data exchange standard, short for Industrial Foundation Class. In this invention, the BIM model serves as the source of facade design parameters, providing basic data for subsequent optimization calculations. Because the amount of solar radiation received by different areas of a building facade varies, the south-facing facade receives the most solar radiation throughout the year in the Northern Hemisphere, while the north-facing facade receives the least. East- and west-facing facades mainly receive stronger radiation in the morning and afternoon. Furthermore, the upper floors of high-rise buildings typically receive more solar radiation than lower floors due to fewer obstructions. Therefore, it is unreasonable to optimize all facade units using uniform design parameters; it is necessary to differentiate the facade units according to their orientation and spatial location. This invention introduces an orientation weight factor and a floor height attenuation factor to construct a spatiotemporal decomposition vector, enabling the optimization algorithm to provide differentiated design schemes for facade units in different locations.

[0082] Step S1 specifically includes sub-steps S11 to S15. Sub-step S11 involves parsing the IFC format BIM model file and extracting the parameters of the facade components. This invention first reads the IFC format BIM model file and traverses the IfcWall and IfcWindow entities in the model. The IfcWall entity represents wall components in the building, and the IfcWindow entity represents window components. For each facade unit, this invention extracts its attribute parameters such as exterior wall area, window area, initial angle of the sunshade, and visible light transmittance of the glass. The ratio of exterior wall area to window area is the window-to-wall ratio, a key parameter affecting building energy consumption and lighting performance. Visible light transmittance of the glass represents the proportion of visible light passing through the glass, and its value range is typically 0.3 to 0.8. Simultaneously, this invention calculates the azimuth angle and relative height of the floor for each facade unit based on the spatial coordinates of the components. The azimuth angle is the angle between the facade normal vector and the due north direction, starting from due north as 0 degrees and increasing clockwise, with a value range of 0 degrees to 360 degrees. The relative height of a floor refers to the vertical distance from the geometric center of a facade unit to the ground. Through the above analytical process, this invention obtains a complete set of facade geometric parameters, laying the foundation for subsequent orientation zoning and weight calculation.

[0083] Sub-step S12 involves dividing the facade units into azimuth zones based on their azimuth angles. This invention divides all facade units into 8 azimuth zones based on the azimuth angle of each facade unit, with each azimuth zone covering a 45-degree range. Specifically, facade units with an azimuth angle of 337.5 degrees to 22.5 degrees are classified as the due north azimuth zone, facade units with an azimuth angle of 22.5 degrees to 67.5 degrees are classified as the northeast azimuth zone, facade units with an azimuth angle of 67.5 degrees to 112.5 degrees are classified as the due east azimuth zone, facade units with an azimuth angle of 112.5 degrees to 157.5 degrees are classified as the southeast azimuth zone, facade units with an azimuth angle of 157.5 degrees to 202.5 degrees are classified as the due south azimuth zone, facade units with an azimuth angle of 202.5 degrees to 247.5 degrees are classified as the southwest azimuth zone, facade units with an azimuth angle of 247.5 degrees to 292.5 degrees are classified as the due west azimuth zone, and facade units with an azimuth angle of 292.5 degrees to 337.5 degrees are classified as the northwest azimuth zone. By using orientation partitioning, this invention groups facade units with similar orientations into one category, which facilitates subsequent differentiated processing based on solar radiation characteristics.

[0084] Sub-step S13 calculates the azimuth weighting factor for each facade unit. This invention first obtains typical annual meteorological data for the building's location and calculates the annual cumulative solar radiation for each azimuth region. Annual cumulative solar radiation refers to the total solar radiation energy received per unit area of ​​the facade throughout the year, and its value is closely related to the facade's orientation. In the Northern Hemisphere, the annual cumulative solar radiation is typically highest in the due south direction and lowest in the due north direction. This invention calculates the azimuth weighting factor for each facade unit using the following formula: ;

[0085] in, The orientation weight factor for the current facade unit is dimensionless. The annual cumulative solar radiation (kWh / m²) of the current facade unit's location area. 2 ); The maximum annual cumulative solar radiation (kWh / m²) across all azimuth regions. 2 Using the above formula, the azimuth weight factor for the azimuth region with the highest annual cumulative solar radiation is 1, while the azimuth weight factors for other azimuth regions are values ​​between 0 and 1. The azimuth weight factor reflects the relative importance of each facade unit in the optimization of light and heat balance; facade units with high solar radiation should be given higher priority in the optimization process.

[0086] Sub-step S14 involves calculating the floor height attenuation factor for each facade unit. The upper floors of high-rise buildings typically receive more solar radiation due to their distance from ground-level obstructions. This invention uses a logarithmic function to describe the impact of floor height on solar radiation reception, and calculates the floor height attenuation factor for each facade unit using the following formula:

[0087] ;

[0088] in, The floor height attenuation factor for the current facade unit is dimensionless. This is an empirical coefficient, ranging from 0.05 to 0.15, which can be adjusted according to the climate characteristics of the building's location and the surrounding obstruction. The natural logarithm operator; The relative height (m) of the floor where the current facade unit is located; The reference height is set to 10 (m). The reason for using a logarithmic function is that the effect of building height on solar radiation reception is not linear, but rather gradually levels off with increasing height. The difference between low-rise and mid-rise buildings is larger, while the difference between high-rise and super high-rise buildings is relatively smaller. This characteristic of the logarithmic function is consistent with actual physical laws.

[0089] Sub-step S15 involves constructing a spatiotemporal decomposition vector and forming a facade design parameter set. This invention combines the geometric parameters of each facade unit with corresponding azimuth weighting factors and floor height attenuation factors to construct a spatiotemporal decomposition vector. The spatiotemporal decomposition vector includes information such as the window-to-wall ratio, sunshade angle, visible light transmittance of the glass, azimuth weighting factors, and floor height attenuation factors for each facade unit. Through the spatiotemporal decomposition vector, this invention achieves differentiated characterization of facade units at different locations, enabling subsequent multi-objective optimization algorithms to provide targeted design schemes based on the actual solar radiation conditions of each facade unit. The spatiotemporal decomposition vectors of all facade units together form the facade design parameter set, serving as input for subsequent surrogate model training and optimization calculations.

[0090] Step S2 involves constructing and training a dual-channel temporal convolutional neural network surrogate model. In the multi-objective optimization problem of building facades, each evaluation of the performance of candidate design schemes requires calculations using building energy consumption simulation software and lighting environment simulation software. While traditional physical simulation software such as EnergyPlus and Radiance offer high accuracy, a single simulation typically takes several minutes to tens of minutes. Multi-objective optimization algorithms need to evaluate thousands to tens of thousands of candidate schemes, resulting in excessively long overall optimization times, making it difficult to meet the timeliness requirements of engineering design. A surrogate model is a mathematical model used to approximate the input-output relationship of complex physical models. Its computational speed is much faster than the original physical model, completing predictions in milliseconds. This invention uses a dual-channel temporal convolutional neural network as the surrogate model. This network includes an energy consumption prediction channel and a lighting environment prediction channel, used to predict the building's annual energy consumption and effective daylighting time ratio, respectively. A temporal convolutional neural network is a deep learning model specifically designed for processing sequential data. It uses a dilated convolutional structure to expand the receptive field, effectively capturing long-range dependencies between input parameters.

[0091] Step S2 specifically includes sub-steps S21 to S24. Sub-step S21 generates training samples using the Latin hypercube sampling method. Latin hypercube sampling is a hierarchical random sampling method that divides the value range of each parameter into several equally probable intervals, ensuring that each interval is sampled exactly once, thus making the samples more evenly distributed in the parameter space. Compared with simple random sampling, Latin hypercube sampling can cover a larger parameter space with fewer samples, effectively reducing the training cost of the surrogate model. In this invention, the window-to-wall ratio is set to a range of 0.2 to 0.7, the sunshade angle to a range of 0 degrees to 90 degrees, and the visible light transmittance of the glass to a range of 0.3 to 0.8. A window-to-wall ratio that is too low will lead to insufficient indoor lighting, while a window-to-wall ratio that is too high will increase air conditioning energy consumption; therefore, 0.2 to 0.7 is a commonly used reasonable range in engineering. A sunshade angle of 0 degrees indicates that the sunshade is completely horizontal, and 90 degrees indicates that the sunshade is completely vertical. This invention uses the Latin hypercube sampling method to generate sampling points in the above parameter space. The number of sampling points can be determined according to the parameter dimension and accuracy requirements. The preferred range of the number of sampling points is 1500 to 2500.

[0092] Sub-step S22 involves calling simulation software to obtain labeled data for the training samples. This invention uses EnergyPlus software to simulate hourly energy consumption throughout the year for each sampling point. EnergyPlus is a building energy consumption simulation software developed by the U.S. Department of Energy, capable of simulating the energy consumption of building heating, cooling, and lighting systems. Through EnergyPlus simulation, this invention obtains the annual heating and cooling load value corresponding to each sampling point, i.e., the total energy consumption of the building for cooling and heating throughout the year. Simultaneously, this invention uses Radiance software to simulate the lighting environment for each sampling point. Radiance is a lighting environment simulation software developed by Lawrence Berkeley National Laboratory, capable of accurately simulating indoor light distribution. This invention statistically analyzes the percentage of hours during the year when the indoor work surface illuminance is between 300 lux and 3000 lux, defining this percentage as the effective lighting time ratio. 300 lux is the minimum illuminance requirement for office building work surfaces, and 3000 lux is the upper limit to prevent glare; illuminance within this range indicates good lighting quality. A higher effective lighting time ratio indicates better indoor natural lighting conditions and lower artificial lighting energy consumption. Through the above simulation process, the present invention labels two target values ​​for each sampling point: annual energy consumption and effective daylighting time ratio, forming a complete training dataset.

[0093] Sub-step S23 involves constructing a dual-channel temporal convolutional neural network. The dual-channel temporal convolutional neural network constructed in this invention comprises two parallel prediction channels. The energy consumption prediction channel uses four dilated convolutional layers with dilation factors of 1, 2, 4, and 8, respectively. Dilated convolution is a convolution operation that inserts holes between elements of a standard convolutional kernel; the dilation factor represents the size of the hole. By increasing the dilation factor layer by layer, the receptive field of the energy consumption prediction channel expands exponentially, enabling it to capture long-range dependencies between input parameters. The light environment prediction channel uses three dilated convolutional layers with dilation factors of 1, 2, and 4, respectively. Since the effective light-gathering time ratio is mainly directly affected by the window-to-wall ratio and glass transmittance, its input-output relationship is relatively simple; therefore, the light environment prediction channel uses a shallower network structure. Each dilated convolutional layer is followed by a batch normalization layer and a ReLU activation function to accelerate training convergence and introduce nonlinearity. ReLU is a commonly used activation function that sets negative values ​​to zero and keeps positive values ​​unchanged. Both channels are ultimately connected to a fully connected output layer, which outputs the predicted energy consumption and the predicted effective light-gathering time ratio, respectively.

[0094] Sub-step S24 involves training a dual-channel temporal convolutional neural network and verifying its accuracy. This invention divides the sampling points into training, validation, and test sets in an 8:1:1 ratio. The training set is used to update network weights, the validation set is used to monitor the training process and prevent overfitting, and the test set is used to finally evaluate the model's accuracy. This invention uses the Adam optimizer to train the network. Adam is an adaptive learning rate optimization algorithm that combines the advantages of momentum and RMSProp, automatically adjusting the learning rate across different parameters and achieving fast convergence. This invention sets hyperparameters such as the initial learning rate, batch size, and maximum number of iterations. Training is terminated early when the validation set loss no longer decreases after several consecutive iterations to prevent overfitting. After training, this invention verifies the determination coefficients of the two channels on the test set. The determination coefficient is an indicator of the goodness of fit of the regression model, ranging from 0 to 1; a value closer to 1 indicates higher prediction accuracy. When the determination coefficients of both channels are greater than 0.90, the surrogate model training is complete and can be used for rapid performance evaluation in subsequent optimization processes. Compared to directly calling EnergyPlus and Radiance simulations, the surrogate model improves prediction speed by about three orders of magnitude, enabling large-scale multi-objective optimization.

[0095] Step S3 involves using the improved NSGA-II algorithm to solve a multi-objective optimization problem. Building facade design is a typical multi-objective optimization problem, requiring simultaneous consideration of two conflicting objectives: minimizing energy consumption and maximizing daylighting quality. Increasing the window-to-wall ratio improves the effective daylighting time ratio but increases summer air conditioning and winter heating loads; decreasing the window-to-wall ratio reduces energy consumption but leads to insufficient indoor lighting. Therefore, there is no single solution that simultaneously optimizes both objectives; rather, there exists a set of Pareto optimal solutions. A Pareto optimal solution is one where no other solution can improve the other objective without worsening either. NSGA-II is a classic multi-objective evolutionary optimization algorithm, short for Non-Dominated Sorting Genetic Algorithm II with Elite Strategy. It uses non-dominated sorting and crowding calculation to maintain population diversity and guide the search towards the Pareto front. This invention improves the NSGA-II algorithm in two aspects: First, it introduces an adaptive crowding mechanism, focusing on global search in the early stages of optimization and local refinement in the later stages. Second, a dynamic mutation probability mechanism is introduced, which uses a larger mutation probability for inferior individuals with a higher non-dominance level to enhance their exploration ability, and uses a smaller mutation probability for superior individuals with a lower non-dominance level to protect their superior genes.

[0096] Step S3 specifically includes sub-steps S31 to S36. Sub-step S31 involves setting the optimization algorithm parameters and initializing the population. This invention sets parameters such as population size, maximum number of iterations, crossover probability, and mutation probability range. The population size represents the number of individuals in each generation, which can be determined based on the problem size and computational resources, with a preferred value range of 100 to 200. The maximum number of iterations represents the maximum number of generations the algorithm can run, which can be determined based on convergence speed and accuracy requirements, with a preferred value range of 200 to 500. This invention randomly initializes the facade design parameters of each individual in the population. Each individual represents a complete facade design scheme, including parameters such as the window-to-wall ratio, sunshade angle, and visible light transmittance of each facade unit. The individuals in the initial population are randomly distributed within the parameter space, providing diverse starting points for subsequent evolutionary searches.

[0097] Sub-step S32 involves calling a surrogate model to evaluate the fitness of individuals in the population. This invention calls the dual-channel temporal convolutional neural network surrogate model trained in step S2 to evaluate the fitness of each individual in the population. For each individual, the surrogate model receives its facade design parameters as input, outputs the predicted annual energy consumption value through the energy consumption prediction channel, and outputs the predicted effective daylighting time ratio value through the light environment prediction channel. These two predicted values ​​serve as the fitness values ​​of that individual on the two optimization objectives. Because the prediction speed of the surrogate model is much faster than that of physical simulation software, this invention can complete the fitness evaluation of a large number of individuals in a short time, providing support for the efficient operation of multi-objective optimization algorithms.

[0098] Sub-step S33 involves performing non-dominated ranking and calculating adaptive crowding. This invention first performs non-dominated ranking on all individuals in the population. The basic idea of ​​non-dominated ranking is that if individual A is not inferior to individual B in all objectives and is superior to individual B in at least one objective, then individual A is said to dominate individual B. Individuals not dominated by any other individual constitute the first non-dominated rank. After removing these individuals, the remaining non-dominated individuals constitute the second non-dominated rank, and so on. Individuals with lower non-dominated ranks are more superior and have a higher retention priority in environmental selection. Within the same non-dominated rank, this invention calculates the adaptive crowding of each individual using the following formula:

[0099] ;

[0100] in, The adaptive crowding degree of the current individual is dimensionless. The global crowding degree of the current individual is obtained by normalizing and summing the differences in target values ​​between adjacent individuals, and is dimensionless. For exponential function operators; This is an adaptive factor, with a value range of 0.5 to 2.0; Let this be the current iteration algebra; The maximum number of iterations; The local crowding degree of the current individual is calculated using the K-nearest neighbor method, where K is 5 and dimensionless. Global crowding degree measures the sparsity of an individual across the entire Pareto front, while local crowding degree measures the sparsity of an individual within its neighborhood. In the early stages of optimization, an exponential decay term is used. When the value is close to 1, the adaptive crowding degree is mainly determined by the global crowding degree, and the algorithm focuses on global search to explore a wider parameter space. As iterations proceed, the exponential decay term gradually decreases to 0, and the adaptive crowding degree gradually becomes dominated by the local crowding degree. The algorithm then shifts to local refinement to improve the uniformity of the Pareto front. This adaptive mechanism allows the algorithm to automatically balance global search and local refinement during the optimization process without requiring manual parameter adjustment.

[0101] Sub-step S34 calculates the dynamic mutation probability of each individual. The traditional NSGA-II algorithm uses the same mutation probability for all individuals, which may lead to high-quality individuals losing superior genes due to excessive mutation, or low-quality individuals getting stuck in local optima due to insufficient mutation. This invention introduces a dynamic mutation probability mechanism, adaptively adjusting the mutation probability based on the individual's non-dominated level. This invention calculates the dynamic mutation probability of each individual using the following formula:

[0102] ;

[0103] in, The mutation probability of the current individual is dimensionless. The minimum mutation probability is set to 0.01. The maximum mutation probability is set to 0.20. The current individual's non-dominant level; Population size; The moderating index ranges from 1.5 to 2.5. According to the formula above, superior individuals with lower non-dominant levels have a lower probability of mutation, thus protecting their superior genes; inferior individuals with higher non-dominant levels have a higher probability of mutation, enhancing their ability to explore new regions. Moderating Index The shape of the curve controlling the mutation probability as a function of non-dominance level is determined by the value; the larger the value, the greater the difference in mutation probability between superior and inferior individuals.

[0104] Sub-step S35 involves performing genetic operations to generate the offspring population. This invention employs a tournament selection method to choose parent individuals from the current population. The basic process of tournament selection is to randomly select several individuals to form a tournament pool, and then select the individual with the lowest non-dominant level or the highest crowding as the parent. This invention performs a simulated binary crossover operation on the selected parent individuals to generate offspring individuals. Simulated binary crossover is a crossover operator suitable for real-number encoded genetic algorithms, simulating the search characteristics of single-point crossover under binary encoding. Subsequently, this invention performs a polynomial mutation operation on the offspring individuals based on the dynamic mutation probability calculated in sub-step S34. Polynomial mutation is a mutation operator suitable for real-number encoding, which generates mutations in a polynomial distribution around the current gene value. Through crossover and mutation operations, this invention generates an offspring population of the same size as the parent population. The offspring population inherits the superior genes of the parent while introducing new genetic diversity.

[0105] Sub-step S36 involves performing environment selection and determining the termination condition. This invention merges the parent and offspring populations to form a temporary population twice the size of the original population. After performing non-dominated sorting and adaptive crowding calculation on the temporary population, this invention performs environment selection according to the principle of prioritizing non-dominated level and then crowding, retaining a number of excellent individuals from the original population to enter the next generation. Individuals with lower non-dominated levels are given priority for retention, and individuals with higher crowding within the same non-dominated level are given priority for retention. Through environment selection, this invention continuously improves the overall quality of the population while maintaining its size. This invention determines whether the current iteration generation has reached the maximum iteration generation. If the maximum iteration generation has not been reached, it returns to sub-step S32 to continue iterating. If the maximum iteration generation has been reached, it outputs all individuals in the current population at the first non-dominated level as the Pareto optimal solution set. Each solution in the Pareto optimal solution set is an excellent solution that cannot be dominated by other solutions under the current constraints, and designers can select the most suitable solution from them according to actual needs.

[0106] Step S4 involves calculating the dynamic shading angle based on the photothermal coupling index and generating an hourly shading control strategy for the entire year. Shading panels on building facades are crucial components for regulating the indoor light and heat environment, and their angle directly affects the amount of solar radiation entering the room. During summer cooling operations, the shading angle should be increased to block solar radiation and reduce air conditioning load; during winter heating operations, the shading angle should be decreased to allow solar radiation to enter, utilizing solar energy for passive heating. Simultaneously, indoor lighting needs must be considered to avoid increased energy consumption for artificial lighting due to excessive shading. Traditional shading control methods either use a fixed angle or rely on geometric tracking based on the solar altitude angle, making it difficult to simultaneously address both insulation and lighting requirements. This invention introduces a photothermal coupling index, comprehensively considering the current solar radiation gain and indoor illuminance to achieve dynamic optimization control of the shading angle.

[0107] Step S4 specifically includes sub-steps S41 to S45. Sub-step S41 involves selecting facade design parameters from the Pareto optimal solution set and obtaining meteorological data. This invention selects a set of facade design parameters from the Pareto optimal solution set obtained in step S3 as the implementation scheme. Designers can choose according to the priorities of the actual project; if energy conservation is more important, a scheme with lower energy consumption is selected; if lighting is more important, a scheme with a longer effective lighting time is selected; or a compromise scheme that balances the two objectives can be chosen. This invention obtains hourly solar radiation data for the building's location throughout the year, including parameters such as direct solar radiation intensity, diffuse solar radiation intensity, solar altitude angle, and solar azimuth angle. This data can be obtained from a typical meteorological year database. Typical meteorological year data are representative meteorological data obtained from historical meteorological records and can reflect the climate characteristics of the building's location.

[0108] Sub-step S42 involves calculating the photothermal coupling index at each time point. This invention calculates the photothermal coupling index for each time point throughout the year. The photothermal coupling index is a dimensionless index that comprehensively characterizes the relative intensity of insulation and lighting demands at the current moment. This invention calculates the photothermal coupling index using the following formula:

[0109] ;

[0110] in, The photothermal coupling index at the current moment is dimensionless. This is a dimensionless coefficient representing the weighting factor for insulation requirements. The heat gain from solar radiation at the current moment (W / m²) 2 ); The design maximum daily solar radiation heat gain (W / m 2 ); This is a dimensionless weighting coefficient for lighting requirements. This is the estimated indoor daylight illuminance (lux) at the current moment. The target illuminance value ranges from 300 to 500 lux. The first term of the solar-thermal coupling index reflects the intensity of the insulation demand; the greater the heat gain from solar radiation at the current moment, the stronger the insulation demand, and the larger this term's value. The second term of the solar-thermal coupling index reflects the intensity of the lighting demand; the lower the current indoor illuminance is compared to the target value, the stronger the lighting demand, and the more negative this term's value. When the solar-thermal coupling index is positive, it indicates that the current insulation demand is dominant, and the angle of the sunshade should be increased to block solar radiation; when the solar-thermal coupling index is negative or zero, it indicates that the current lighting demand is dominant, and the angle of the sunshade should be decreased to increase indoor lighting.

[0111] Sub-step S43 involves automatically adjusting the thermal insulation and lighting demand weighting coefficients according to the season. The building's thermal insulation and lighting demands vary with the seasons. In summer, outdoor temperatures are high, leading to a stronger demand for thermal insulation; therefore, priority should be given to blocking solar radiation to reduce air conditioning load. In winter, outdoor temperatures are low, allowing for passive heating using solar radiation; simultaneously, due to shorter daylight hours, the demand for lighting is relatively stronger. This invention automatically adjusts the thermal insulation and lighting demand weighting coefficients based on the accumulated days of the year using the following formula:

[0112] ;

[0113] in, This is the weighting coefficient for the insulation demand on that day, and it is dimensionless. This is the sine function operator; Pi; It represents the accumulated days of a year, with a value ranging from 1 to 365. This represents the weighting coefficient for daily daylighting demand, which is dimensionless. Through sine function modulation, the weighting coefficient for thermal insulation demand reaches its maximum value of 0.7 around the summer solstice and its minimum value of 0.3 around the winter solstice, which aligns with the actual needs of building energy management throughout the year. The 80th day of the year corresponds to around the spring equinox in the Northern Hemisphere, at which time the weighting coefficient for thermal insulation demand is 0.5, indicating a relative balance between thermal insulation and daylighting demand.

[0114] Sub-step S44 involves calculating the sunshade angle based on the photothermal coupling index. This invention determines the dominant demand at any given moment based on the sign of the photothermal coupling index and uses different formulas to calculate the sunshade angle. When the photothermal coupling index is greater than zero, it indicates that heat insulation demand takes precedence. This invention calculates the sunshade angle using the following formula:

[0115] ;

[0116] When the photothermal coupling index is less than or equal to zero, it indicates that the need for lighting is prioritized. The present invention calculates the angle of the sunshade according to the following formula: ;

[0117] in, The current angle (in degrees) of the sunshade; This is the minimum angle of the sunshade, with a value of 0 (degrees). The maximum angle of the sunshade is 90 degrees. The hyperbolic tangent function operator; The response sensitivity coefficient ranges from 2.0 to 5.0. This represents the absolute value of the photothermal coupling index. Hyperbolic tangent function. The value ranges from -1 to 1, changing rapidly near the origin and tending to saturate far from the origin. The advantage of using the hyperbolic tangent function is that when the absolute value of the photothermal coupling index is small, the sunshade angle is sensitive to changes in the index, enabling rapid response to minor changes in the photothermal environment; when the absolute value of the photothermal coupling index is large, the sunshade angle tends to its limit, avoiding frequent adjustments to the sunshade due to drastic fluctuations in the index. Response sensitivity coefficient The response speed of the sunshade angle to the photothermal coupling index is controlled. The larger the value, the more sensitive the response. It can be adjusted according to the actual performance of the sunshade drive system.

[0118] Sub-step S45 generates a year-round hourly sunshade angle sequence. This invention iterates through 8760 hours throughout the year, repeating sub-steps S42 to S44 for each hour to calculate the corresponding sunshade angle. The sunshade angle values ​​for all 8760 hours constitute a complete year-round hourly sunshade angle sequence. This sequence can be used as input to an automatic sunshade control system, enabling dynamic adjustment of the sunshade angle. In practical applications, the angle sequence can be smoothed based on the response time of the sunshade drive system to avoid excessively frequent adjustments to the sunshade. Through dynamic shading control, this invention can adaptively balance heat insulation and lighting requirements at different times throughout the year, further reducing overall building energy consumption and improving indoor light environment quality compared to fixed-angle shading and geometric tracking shading.

[0119] Example 1: This example uses a 12-story office building in a city as the application object. The total building height is 48m, and the standard floor area is 1200m². 2 The total area of ​​the exterior facade is 4800m². 2 The total window area is 1680m² 2 The building is located in a hot-summer, cold-winter climate zone, and typical meteorological data from a given year are used as the meteorological input. This embodiment employs the BIM-based dynamic light and heat balance design method for building facades provided by this invention, and the specific implementation process is as follows.

[0120] Step S1 involves parsing the facade component information from the BIM model and constructing a spatiotemporal decomposition vector. In step S11, the IFC parsing unit is used to read the IFC format BIM model file of the office building, traversing the IfcWall and IfcWindow entities in the model, and extracting the exterior wall area, window area, initial angle of the sunshade, and visible light transmittance of the glass for each facade unit. In this embodiment, a total of 96 facade units were identified, including 32 facing south, 32 facing north, 16 facing east, and 16 facing west. The azimuth angle and relative height of the floor to which each facade unit is located are calculated based on the spatial coordinates of the components.

[0121] In step S12, all facade units are divided into 8 azimuth zones according to their azimuth angles, with each azimuth zone covering a 45-degree range. In this embodiment, the south azimuth zone contains 24 facade units, the north azimuth zone contains 24 facade units, the east azimuth zone contains 12 facade units, the west azimuth zone contains 12 facade units, and the remaining 24 facade units are distributed in 4 oblique azimuth zones.

[0122] In step S13, the annual cumulative solar radiation data for each direction of the building's location is obtained. The annual cumulative solar radiation for the due south direction is 1180 kWh / m². 2 The area directly to the north has a density of 420 kWh / m³. 2 The area due east has a density of 780 kWh / m³. 2 The westernmost region has a capacity of 810 kWh / m³. 2 According to the formula Calculate the orientation weight factor for each facade unit, where the orientation weight factor for the due south area is 1.00, for the due west area it is 0.69, for the due east area it is 0.66, and for the due north area it is 0.36.

[0123] In step S14, according to the formula Calculate the floor height attenuation factor for each facade unit, where β is taken as 0.10 and h0 is taken as 10m. The floor height attenuation factor for a 1-story facade unit is 1.10, for a 6-story unit it is 1.19, and for a 12-story unit it is 1.26.

[0124] In step S15, the geometric parameters of each facade unit are combined with the corresponding orientation weight factor and floor height attenuation factor to construct a spatiotemporal decomposition vector, forming a facade design parameter set containing information of 96 facade units.

[0125] Step S2 involves constructing and training a dual-channel temporal convolutional neural network surrogate model.

[0126] In step S21, the window-to-wall ratio is set to a range of 0.2 to 0.7, the sunshade angle is set to a range of 0 degrees to 90 degrees, and the visible light transmittance of the glass is set to a range of 0.3 to 0.8. 2000 sampling points are generated using the Latin hypercube sampling method.

[0127] In step S22, EnergyPlus software was used to simulate annual energy consumption for each sampling point to obtain heating and cooling load values, and Radiance software was used to simulate the light environment and calculate the effective lighting time ratio. The simulation and annotation of 2000 sampling points took approximately 72 hours.

[0128] In step S23, a dual-channel temporal convolutional neural network is constructed. The energy consumption prediction channel is set with 4 dilated convolutional layers with dilation factors of 1, 2, 4 and 8 respectively. The light environment prediction channel is set with 3 dilated convolutional layers with dilation factors of 1, 2 and 4 respectively.

[0129] In step S24, the sampling points are divided into training, validation, and test sets in an 8:1:1 ratio, and the network is trained using the Adam optimizer. After training, the determination coefficients of the energy consumption prediction channel are calculated. The coefficient of determination for the light environment prediction channel is 0.953. It is 0.941.

[0130] Step S3 involves using the improved NSGA-II algorithm to perform multi-objective optimization.

[0131] In step S31, the population size is set to 150, the maximum number of iterations is 300, and the facade design parameters of each individual in the population are randomly initialized.

[0132] In step S32, a dual-channel temporal convolutional neural network surrogate model is invoked to evaluate the fitness of each individual.

[0133] In step S33, after performing a non-dominated sort on the population, the population is then sorted according to the formula. Calculate the adaptive crowding degree, where γ is set to 1.0.

[0134] In step S34, according to the formula Calculate the dynamic mutation probability, where Take 0.01, Set 0.20, and η to 2.0.

[0135] In step S35, tournament selection and simulated binary crossover are performed to generate offspring. Polynomial mutation is performed based on the dynamic mutation probability. After merging the parent and offspring, environment selection is performed.

[0136] In step S36, steps S32 to S35 are repeated for a total of 300 iterations, outputting a Pareto optimal solution set containing 87 non-dominated solutions.

[0137] Step S4 is to calculate the dynamic sunshade angle based on the photothermal coupling index.

[0138] In step S41, a set of facade design parameters is selected from the Pareto optimal solution set as the implementation scheme. The window-to-wall ratio of the south-facing scheme is 0.45, the window-to-wall ratio of the north-facing scheme is 0.35, and the window-to-wall ratio of the east-west-facing scheme is 0.40.

[0139] In step S42, according to the formula Calculate the photothermal coupling index, where E t Take 400 lux.

[0140] In step S43, according to the formula and Automatically adjust seasonal weighting coefficients.

[0141] In step S44, the angle of the sunshade is calculated according to the positive and negative signs of the photothermal coupling index using the corresponding formula, and the response sensitivity coefficient k is taken as 3.0.

[0142] In step S45, the entire year's 8760 hours are traversed to generate a year-round hourly sunshade angle sequence.

[0143] After optimization using the method of this invention, the building's annual comprehensive energy consumption in this embodiment is 78.6 kWh / m². 2 The effective lighting time ratio is 72.4%.

[0144] Example 2: This example uses an 18-story commercial complex in a certain city as the application object. The total building height is 72m and the total exterior area is 7200m². 2 The building is located in a cold climate zone. Unlike Example 1, this example uses an empirical coefficient β of 0.12, an adaptive factor γ of 1.5, a regulation exponent η of 2.5, and a response sensitivity coefficient k of 4.0. This example uses Latin hypercube sampling to generate 2500 sampling points. After training, the determination coefficient of the energy consumption prediction channel is... The coefficient of determination for the light environment prediction channel is 0.948. The value is 0.936. The multi-objective optimization settings include a population size of 200, a maximum number of iterations of 400, and output a Pareto optimal solution set containing 112 non-dominated solutions. The optimized building's annual comprehensive energy consumption is 82.3 kWh / m². 2 The effective lighting time ratio is 68.7%.

[0145] Example 3: This example uses an 8-story teaching building in a city as the application object. The total building height is 32m and the total exterior area is 3200m². 2The building is located in a hot-summer, warm-winter climate zone. Unlike Example 1, this example uses an empirical coefficient β of 0.08, an adaptive factor γ of 0.8, a regulation exponent η of 1.8, and a response sensitivity coefficient k of 2.5. This example uses Latin hypercube sampling to generate 1800 sampling points. The multi-objective optimization settings include a population size of 120 and a maximum number of iterations of 250. The optimized building's annual comprehensive energy consumption is 65.2 kWh / m². 2 The effective lighting time ratio is 76.8%.

[0146] Comparative Example 1 uses the standard NSGA-II algorithm instead of the adaptive crowding and dynamic mutation probability improved NSGA-II algorithm of this invention. The other steps of Comparative Example 1 are the same as in Example 1, except that a fixed crowding calculation method and a fixed mutation probability of 0.10 are used in step S3. The multi-objective optimization setting is a population size of 150 and a maximum number of iterations of 300. The Pareto optimal solution set output by Comparative Example 1 contains 79 non-dominated solutions. The optimized building's annual comprehensive energy consumption is 82.1 kWh / m². 2 The effective lighting time ratio is 69.5%.

[0147] Comparative Example 2 uses a multilayer perceptron neural network instead of the dual-channel temporal convolutional neural network of this invention as a surrogate model. The multilayer perceptron contains three hidden layers with 128, 64, and 32 neurons respectively. The other steps in Comparative Example 2 are the same as in Example 1. After training, the determination coefficient of energy consumption prediction is... The coefficient of determination for predicting the light environment is 0.891. The value is 0.873. Due to the low accuracy of the surrogate model, the final optimized result shows a building's annual comprehensive energy consumption of 85.4 kWh / m². 2 The effective lighting time ratio is 66.2%.

[0148] Comparative Example 3 uses a fixed shading angle control strategy instead of the dynamic shading control method of the photothermal coupling index of the present invention. Comparative Example 3 fixes the shading angle at 45 degrees, which does not change over time. The other steps of Comparative Example 3 are the same as in Example 1. After adopting fixed shading, the building's annual comprehensive energy consumption is 86.7 kWh / m². 2 The effective lighting time ratio is 64.8%.

[0149] Comparative Example 4 employs a geometric tracking shading control strategy based on the solar altitude angle. The shading angle varies linearly with the solar altitude angle, as shown in the formula: , where h txn The solar altitude angle is shown. The other steps in Comparative Example 4 are the same as in Example 1. After adopting geometric tracking shading, the building's annual comprehensive energy consumption is 83.2 kWh / m². 2The effective lighting time ratio is 67.1%.

[0150] Experiment 1 compares and verifies the Pareto front distribution quality of Example 1 and Comparative Example 1. The Spacing and Hypervolume indices are used to evaluate the uniformity and convergence of the Pareto front distribution. A smaller Spacing index indicates a more uniform distance between adjacent solutions on the Pareto front, and a more regular distribution; a larger Hypervolume index indicates a larger target space volume covered by the Pareto front, and better convergence and diversity of the algorithm.

[0151] During the experiment, Example 1 and Comparative Example 1 were run independently 30 times each. The Spacing and Hypervolume metrics were recorded for each run, and the mean and standard deviation were calculated. The experimental results are as follows: Figure 1 As shown. From Figure 1 As shown in subplot (a), the box plot of the Spacing index for Example 1 is generally below that of Comparative Example 1, and the box width is narrower, indicating that the Pareto front distribution of Example 1 is more uniform and has better stability. The mean value of the Spacing index for Example 1 is 0.0234, and the standard deviation is 0.0018; the mean value of the Spacing index for Comparative Example 1 is 0.0352, and the standard deviation is 0.0031. From... Figure 1 As shown in subplot (b), the Hypervolume index box plot of Example 1 is generally above that of Comparative Example 1, indicating that the Pareto front of Example 1 has better convergence and diversity. The mean of the Hypervolume index of Example 1 is 0.8672, and the standard deviation is 0.0089; the mean of the Hypervolume index of Comparative Example 1 is 0.7834, and the standard deviation is 0.0156.

[0152] from Figure 1 As can be seen from the two sub-figures, the Spacing index of Example 1 is reduced by 33.5% and the Hypervolume index is increased by 10.7% compared with Comparative Example 1, indicating that the improved NSGA-II algorithm of the present invention is superior to the standard NSGA-II algorithm in terms of Pareto front distribution uniformity and convergence.

[0153] The adaptive congestion calculation formula introduced in this invention is as follows:

[0154] ;

[0155] in, The adaptive crowding level for the current individual; Global congestion level; Local congestion level; As an adaptive factor; Let this be the current iteration algebra; The maximum iteration algebra is used. A dynamic fusion of global and local crowding is achieved through an exponential decay function. In the early stages of optimization, the global crowding weight is relatively large, and the algorithm focuses on exploring the entire parameter space, which is beneficial for discovering widely distributed non-dominated solutions. In the later stages of optimization, the local crowding weight gradually increases, and the algorithm focuses on locally refining the discovered Pareto front, which is beneficial for improving the uniform distribution of solutions. Simultaneously, the dynamic mutation probability formula introduced in this invention is:

[0156] ;

[0157] in, This represents the mutation probability of the current individual. The minimum mutation probability; The maximum mutation probability; The current individual's non-dominant level; Population size; The formula introduces a moderating index. This index grants lower-ranking, inferior individuals with higher non-dominant levels a greater probability of mutation, enhancing their ability to escape local optima. Conversely, it grants lower-ranking, superior individuals with lower non-dominant levels a smaller probability of mutation, protecting desirable genes from destruction. These two mechanisms work synergistically to improve both the uniformity and convergence of the Pareto front.

[0158] Experiment 2 compares and verifies the prediction accuracy of the surrogate models in Example 1 and Comparative Example 2. The coefficient of determination is used. The root mean square error (RMSE) and mean absolute percentage error (MAPE) are used as evaluation indicators. Coefficient of determination. The root mean square error (RMSE) reflects the model's ability to explain data variation; the closer the value is to 1, the higher the prediction accuracy. The mean square error (RMSE) reflects the average deviation between the predicted and the true values. The mean absolute percentage error (MAPE) reflects the average relative error of the predicted values ​​relative to the true values.

[0159] During the experiment, the same test set was used to evaluate the prediction accuracy of the two surrogate models. The test set contained 200 sample points, all from a dataset generated by Latin hypercube sampling. The experimental results are as follows: Figure 2 As shown. From Figure 2 As can be seen from subgraph (a), in the energy consumption prediction task, Example 1 The bar height is significantly higher than that of Comparative Example 2, while the RMSE and MAPE bar heights are significantly lower than those of Comparative Example 2, indicating that the dual-channel temporal convolutional neural network of this invention has higher accuracy in energy consumption prediction. Specific values ​​are as follows: Example 1 The value was 0.953, and the RMSE was 3.42 kWh / m³. 2MAPE was 3.87%; in Comparative Example 2... The value is 0.891, and the RMSE is 5.21 kWh / m³. 2 MAPE was 5.92%. From Figure 2 As shown in subgraph (b), Example 1 also exhibits superior prediction performance in the daylight prediction task. Specifically, the values ​​for Example 1 are: The values ​​were 0.941, RMSE was 2.15%, and MAPE was 2.89%; Comparative Example 2... The values ​​were 0.873, RMSE was 3.56%, and MAPE was 4.78%.

[0160] The dual-channel temporal convolutional neural network of this invention demonstrates its effectiveness in both energy consumption prediction and daylighting prediction tasks. Both are about 6 percentage points higher than the multilayer perceptron, and the prediction error is reduced by about 35%. The improvement in the accuracy of the surrogate model provides a more reliable fitness evaluation basis for subsequent multi-objective optimization.

[0161] Building energy consumption is influenced by the temporal variation of meteorological conditions throughout the year. Solar radiation and outdoor temperature exhibit periodic variations across different months and seasons. Traditional multilayer perceptrons treat input parameters as independent variables, failing to capture this temporal dependence. This invention employs a temporal convolutional neural network that expands the receptive field through dilated convolutional structures. The dilation factors of the four dilated convolutional layers in the energy consumption prediction channel are 1, 2, 4, and 8, respectively, thus covering information across 15 time steps and effectively capturing the long-range temporal dependence of meteorological parameters. The dual-channel parallel structure allows for targeted network depths for energy consumption and daylighting prediction. The deeper energy consumption prediction channel captures complex thermal coupling relationships, while the shallower daylighting prediction channel accommodates relatively simple optical transmission patterns, resulting in high accuracy for both prediction tasks.

[0162] Experiment 3 compares and verifies the dynamic shading control effects of Example 1, Comparative Example 3, and Comparative Example 4. Annual comprehensive energy consumption, effective daylighting time ratio, and thermal comfort compliance hours are used as evaluation indicators. Annual comprehensive energy consumption includes the sum of air conditioning cooling energy consumption, heating energy consumption, and artificial lighting energy consumption; the effective daylighting time ratio is the percentage of hours during the year when the indoor working surface illuminance is between 300 lux and 3000 lux; and the thermal comfort compliance hours are the cumulative number of hours throughout the year when the indoor thermal environment meets thermal comfort standards. The experiment was designed according to GB50189-2015 "Energy Conservation Design Standard for Public Buildings", GB / T50033-2013 "Daylighting Design Standard for Buildings", and ASHRAE Standard 55-2020 Thermal Comfort Standard. During the experiment, hourly simulations were performed throughout the year for the three shading control strategies, and the performance indicators were statistically analyzed.

[0163] Experimental results are as follows Figure 3 As shown. From Figure 3 As shown in subgraph (a), Example 1 has the lowest annual comprehensive energy consumption column height, Comparative Example 3 has the highest, and Comparative Example 4 is in the middle, indicating that the dynamic shading control method of the photothermal coupling index of the present invention has a significant advantage in energy saving. Specifically, the value for Example 1 is 78.6 kWh / m². 2 Comparative Example 3 is 86.7 kWh / m 2 Comparative Example 4 is 83.2 kWh / m 2 .from Figure 3 As shown in subgraph (b), Example 1 has the highest effective lighting time compared to the column height, indicating that the method of the present invention does not sacrifice indoor lighting quality while ensuring energy saving. Specifically, the values ​​are: 72.4% for Example 1, 64.8% for Comparative Example 3, and 67.1% for Comparative Example 4. From... Figure 3 As shown in subplot (c), Example 1 achieved the highest number of hours meeting thermal comfort standards, indicating that the method of the present invention effectively improved the indoor thermal environment. Specifically, the values ​​were: 7125 hours for Example 1, 6542 hours for Comparative Example 3, and 6823 hours for Comparative Example 4. From... Figure 3 As can be seen from subgraph (d), among the comprehensive performance scores calculated after normalizing the three indicators, Example 1 scored the highest, verifying the comprehensive advantages of the method of the present invention.

[0164] comprehensive Figure 3 As can be seen from the four sub-figures, the photothermal coupling index dynamic shading control method of the present invention saves 9.3% energy compared with fixed-angle shading and 5.5% energy compared with geometric tracking shading. At the same time, the effective lighting time is increased by 7.6 percentage points and 5.3 percentage points respectively, and the number of hours meeting thermal comfort standards is increased by 583 hours and 302 hours respectively, achieving a dynamic balance between heat insulation and lighting requirements.

[0165] Comparative Example 3 uses a fixed shading angle of 45 degrees, which cannot be adjusted according to solar radiation intensity and indoor lighting requirements. Insufficient shading in summer leads to increased air conditioning load, while excessive shading in winter results in insufficient passive solar energy utilization and increased artificial lighting energy consumption. Comparative Example 4 uses a geometric tracking strategy based on the solar altitude angle, considering only the geometric relationship of the sun's position and not the actual needs of the indoor light and heat environment. The light-heat coupling index introduced in this invention is:

[0166] ;

[0167] in, The photothermal coupling index; This is the weighting coefficient for insulation requirements; This is the weighting coefficient for lighting requirements; The heat gained from solar radiation at the current moment; The design is based on the maximum solar radiation heat gain per day; This is the estimated indoor illuminance at the current moment; This represents the target daylighting illuminance value. This index comprehensively considers the current solar radiation heat gain and the indoor daylighting illuminance deviation. When the solar radiation is greater than zero, heat insulation is prioritized, and the angle of the sunshade is increased to block solar radiation; when... When the lighting demand is less than or equal to zero, priority is given, and the angle of the sunshade is reduced to increase indoor lighting. Seasonal weighting coefficient. and The system automatically adjusts over time, increasing the weight of heat insulation in summer and natural light in winter to meet the actual needs of the building's year-round operation. (Hyperbolic tangent function) Its saturation characteristics allow the sunshade angle to remain stable under extreme conditions, avoiding frequent adjustments.

[0168] Experiment 4 analyzes the annual dynamic shading angle variation pattern of Example 1. Three typical days—the summer solstice, winter solstice, and spring equinox—were selected to record the hourly shading angle of the south-facing facade unit. The summer solstice is the day with the highest solar altitude angle and strongest solar radiation of the year; the winter solstice is the day with the lowest solar altitude angle and shortest sunshine duration; and the spring equinox is the day with equal day and night and moderate solar radiation intensity. The experiment was designed according to GB / T50176-2016 "Code for Thermal Design of Civil Buildings" and JGJ / T119-2008 "Standard for Terminology of Architectural Lighting".

[0169] Experimental results are as follows Figure 4 As shown. From Figure 4 As can be seen, the three curves exhibit distinct characteristics. The summer solstice curve is located in the upper part of the graph, with the shading angle remaining high between 65 and 78 degrees from 10:00 to 15:00, peaking between 12:00 and 13:00, when solar radiation is strongest, and the shading is almost fully deployed to block solar radiation to the greatest extent. The winter solstice curve is located in the lower part of the graph, with the shading angle remaining low between 15 and 35 degrees throughout the day, not exceeding 35 degrees even at noon, effectively utilizing solar radiation for passive heating and natural lighting. The vernal equinox curve lies between the summer and winter solstices, exhibiting a clear diurnal variation, gradually increasing from 25 degrees to 55 degrees in the morning and gradually decreasing from 55 degrees to 30 degrees in the afternoon, forming a symmetrical inverted U-shaped curve, reflecting the dynamic response of the solar-thermal coupling index to changes in the sun's position.

[0170] from Figure 4 It can be seen that the dynamic shading control method of the photothermal coupling index of the present invention adaptively responds to the photothermal environment requirements of different seasons and times. In summer, it focuses on heat insulation; in winter, it focuses on lighting and passive heating; and in transitional seasons, it takes both into account, thus realizing intelligent regulation of the shading strategy throughout the year.

[0171] This invention employs a seasonal weight adaptive adjustment mechanism, the formula of which is:

[0172] ;

[0173] in, This is the weighting coefficient for insulation requirements; This is the accumulated days of the year. The phase shift of the sine function by 80 degrees corresponds to around the spring equinox in the Northern Hemisphere. The value is set to 0.5, with equal weighting for heat insulation and lighting requirements. (Yearly accumulated days) Approaching 172 Increased to 0.7, the weight of heat insulation demand increases, and the overall angle of the sunshade shifts upward; accumulated days of the year Approaching 355 Reducing the weight to 0.3 increases the weight of daylighting demand, causing the overall angle of the sunshade to shift downwards. The continuity of the sine function ensures a smooth transition between seasons, avoiding drastic fluctuations in the sunshade angle caused by sudden changes in weight.

[0174] Experiment 5 comprehensively compares the optimization effects of Examples 1, 2, and 3 in different climate zones. Annual comprehensive energy consumption, effective daylighting time ratio, Pareto solution set size, and optimization time are used as evaluation indicators. The Pareto solution set size reflects the number of non-dominated solutions discovered by the optimization algorithm; a larger solution set size indicates more options available to designers. Optimization time reflects the total computation time from BIM model input to optimization result output. The experimental design is based on the climate zone classification standards in GB50176-2016 "Code for Thermal Design of Civil Buildings" and JGJ134-2010 "Standard for Energy-Saving Design of Residential Buildings in Hot Summer and Cold Winter Areas." Example 1 is located in a hot summer and cold winter climate zone, Example 2 in a cold climate zone, and Example 3 in a hot summer and warm winter climate zone.

[0175] Experimental results are as follows Figure 5 As shown. From Figure 5 As shown in subgraph (a), the blue bars represent the total annual energy consumption, and the red line represents the effective daylighting time ratio. The lowest total annual energy consumption in Example 3 is 65.2 kWh / m². 2 This is because the demand for heating in winter is relatively small in hot-summer and warm-winter climate zones; the highest annual comprehensive energy consumption in Example 2 is 82.3 kWh / m². 2 This is because the demand for heating is greater in cold climate zones during winter. Example 3 had the highest effective daylighting time ratio at 76.8%, while Example 2 had the lowest at 68.7%, which is related to the distribution of sunshine hours and solar radiation intensity in each climate zone. From... Figure 5As shown in subgraph (b), the green bars represent the Pareto solution set size, and the purple broken line represents the optimization time. The largest Pareto solution set size in Example 2 was 112, and the longest optimization time was 6.8 hours. This is because the contradiction between heating and lighting is more prominent in cold climate zones, requiring a larger population size and more iterations to fully explore the Pareto frontier.

[0176] comprehensive Figure 5 As can be seen from the two sub-figures, the method of the present invention exhibits good applicability in three different climate zones: hot summer and cold winter, cold, and hot summer and warm winter. The optimized energy consumption indicators all meet the energy-saving design standards of the corresponding climate zones, verifying the universality and robustness of the method of the present invention.

[0177] The spatiotemporal decomposition vector construction method of this invention can automatically calculate the annual cumulative solar radiation and azimuth weight factor for each directional region based on meteorological data of the building's location, adapting to the solar radiation characteristics of different climate zones. The dual-channel temporal convolutional neural network surrogate model uses typical meteorological year data as input and can learn the energy consumption and daylighting patterns of different climate zones. The improved NSGA-II algorithm parameters can be flexibly configured according to the problem size, adapting to the complexity of light-heat contradictions in different climate zones. The automatic seasonal weight adjustment mechanism of the light-heat coupling index can adapt to the heating and cooling cycles of different climate zones. The synergistic effect of the above technical means gives the method of this invention broad applicability across climate zones.

[0178] Experiment 6 analyzes the convergence process of the multi-objective optimization in Example 1. The Hypervolume index change curves for each generation of the population, as well as the energy consumption and effective light-gathering time ratio changes of the optimal individual, are recorded during the optimization process. The change in the Hypervolume index with the number of iterations reflects the speed and extent to which the Pareto front approaches the true Pareto front; the change in the performance index of the optimal individual reflects the algorithm's ability to find high-quality solutions.

[0179] Experimental results are as follows Figure 6 As shown. From Figure 6 As shown in subplot (a), the Hypervolume metric curve exhibits a typical S-shaped convergence characteristic. In the first 100 generations, the curve rises rapidly, increasing from an initial 0.42 to 0.78, an increase of 85.7%, indicating that the algorithm has strong global search capabilities in the early stages of optimization, quickly discovering a large number of non-dominated solutions. Between generations 100 and 200, the curve rises steadily, increasing from 0.78 to 0.85, an increase of 9.0%, indicating that the algorithm enters a local refinement stage, gradually improving the quality of the Pareto front. Between generations 200 and 300, the curve flattens out, eventually stabilizing at 0.867, indicating that the algorithm has essentially converged. Figure 6As shown in subplot (b), the red curve represents the energy consumption value of the optimal individual, and the green curve represents the effective lighting time ratio. The energy consumption value starts from an initial 95.3 kWh / m². 2 It continued to decline to 78.6 kWh / m³ 2 The decrease was 17.5%; the effective daylighting time ratio increased continuously from the initial 58.2% to 72.4%, an increase of 14.2 percentage points. The trends of the two curves indicate that the algorithm can simultaneously improve two conflicting objectives during the optimization process.

[0180] comprehensive Figure 6 As shown in the two subgraphs, the improved NSGA-II algorithm of this invention exhibits excellent convergence performance, finding a high-quality Pareto optimal solution set within 300 generations. The Hypervolume index reaches 0.867, energy consumption is reduced by 17.5%, and light intake is improved by 14.2 percentage points. The adaptive congestion mechanism assigns a large weight to global congestion in the early stages of optimization, prompting the algorithm to quickly explore the parameter space, manifested as a rapid increase in the Hypervolume index in the first 100 generations. As iterations progress, the weight of local congestion gradually increases, and the algorithm shifts to local refinement of the discovered Pareto front, as shown by a steady increase in the Hypervolume index between generations 100 and 200. The dynamic mutation probability mechanism grants inferior individuals a larger mutation probability, helping them escape local optima and ensuring continuous improvement in energy consumption and effective light intake time ratio throughout the optimization process, rather than premature convergence. The synergistic effect of these two mechanisms enables the algorithm to achieve good convergence within 300 generations, balancing convergence speed and solution quality.

[0181] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. 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 BIM-based dynamic light and heat balance design method for building facades, characterized in that, Includes the following steps: S1. Extract the facade component information from the BIM model, construct the spatiotemporal decomposition vector according to the orientation and spatial location of each facade unit, and calculate the orientation weight factor and floor height attenuation factor of each facade unit to obtain the weighted facade design parameter set. S2. The Latin hypercube sampling method is used to generate training samples in the design parameter space. The building energy consumption simulation software and the light environment simulation software are called to obtain the sample labeling data. A dual-channel temporal convolutional neural network surrogate model containing energy consumption prediction channel and light environment prediction channel is constructed and trained. S3. With the optimization objectives of minimizing the building's annual energy consumption and maximizing the effective daylighting time ratio, the NSGA-II algorithm with adaptive crowding degree and dynamic variation probability is used to solve the multi-objective optimization problem and generate the Pareto optimal solution set. S4. Based on the facade design parameters selected from the Pareto optimal solution set, calculate the dynamic sunshade angle of each facade unit at all times throughout the year based on the photothermal coupling index, and generate an hourly shading control strategy for the whole year. Step S1 includes: S11. Parse the BIM model file in IFC format, extract the exterior wall area, window area, initial angle of sunshade and visible light transmittance of glass for each facade unit, and calculate the azimuth angle and relative height of the floor for each facade unit based on the spatial coordinates of the components. S12. Divide all facade units into eight azimuth zones according to the azimuth angle, with each azimuth zone covering a 45-degree range. S13. Obtain the annual cumulative solar radiation for each directional area, and calculate the azimuth weight factor for each facade unit using the following formula: ； in, The orientation weight factor for the current facade unit is dimensionless. The annual cumulative solar radiation (kWh / m²) of the current facade unit's location area. 2 ); The maximum annual cumulative solar radiation (kWh / m²) across all azimuth regions. 2 ); S14. Calculate the floor height attenuation factor for each facade unit using the following formula: ； in, The floor height attenuation factor for the current facade unit is dimensionless. This is an empirical coefficient, with a value ranging from 0.05 to 0.15; The natural logarithm operator; The relative height (m) of the floor where the current facade unit is located; For reference height, the value is 10 (m); S15. Combine the geometric parameters of each facade unit with the corresponding orientation weighting factor and floor height attenuation factor to construct a spatiotemporal decomposition vector, forming a facade design parameter set.

2. The BIM-based dynamic light and heat balance design method for building facades according to claim 1, characterized in that, Step S2 includes: S21. Set the window-to-wall ratio to a range of 0.2 to 0.7, the sunshade angle to a range of 0 degrees to 90 degrees, and the visible light transmittance of the glass to a range of 0.3 to 0.

8. Use the Latin hypercube sampling method to generate 1500 to 2500 sampling points. S22. Perform annual energy consumption simulation for each sampling point to obtain the heating and cooling load values, and perform light environment simulation to count the percentage of hours in which the indoor work surface illuminance is in the range of 300 lux to 3000 lux as the effective lighting time ratio. S23. Construct a dual-channel temporal convolutional neural network. The energy consumption prediction channel is set with four dilated convolutional layers with dilation factors of 1, 2, 4, and 8 respectively. The light environment prediction channel is set with three dilated convolutional layers with dilation factors of 1, 2, and 4 respectively. S24. Divide the sampling points into training, validation and test sets in an 8:1:1 ratio. Train the network using the Adam optimizer until the loss on the validation set no longer decreases. After verifying on the test set that the decision coefficients of both channels are greater than 0.90, the model training is complete.

3. The BIM-based dynamic light and heat balance design method for building facades according to claim 1, characterized in that, Step S3 includes: S31. Set the population size to 100 to 200, the maximum number of iterations to 200 to 500, and randomly initialize the facade design parameters of each individual in the population. S32. Call the dual-channel temporal convolutional neural network surrogate model to evaluate the fitness of each individual and obtain the annual energy consumption prediction value and the effective lighting time ratio prediction value. S33. After performing non-dominated ranking on the population, calculate the adaptive crowding degree of each individual within the same non-dominated level using the following formula: ； in, The adaptive crowding degree of the current individual is dimensionless. The global crowding degree of the current individual is obtained by normalizing and summing the differences in target values ​​between adjacent individuals, and is dimensionless. For exponential function operators; This is an adaptive factor, with a value range of 0.5 to 2.0; Let this be the current iteration algebra; The maximum number of iterations; The local crowding degree of the current individual is calculated by the K-nearest neighbor method, where K is 5 and is dimensionless. S34. Calculate the dynamic mutation probability of each individual using the following formula: ； in, The mutation probability of the current individual is dimensionless. The minimum mutation probability is set to 0.

01. The maximum mutation probability is set to 0.

20. The current individual's non-dominant level; Population size; To adjust the index, the value ranges from 1.5 to 2.5; S35. Perform tournament selection and simulate binary crossover to generate offspring, perform polynomial mutation according to the dynamic mutation probability, merge the parent and offspring, and perform environment selection according to non-dominance level and adaptive crowding degree. S36. Repeat steps S32 to S35 until the maximum number of iterations is reached, and output all non-dominated solutions as the Pareto optimal solution set.

4. The BIM-based dynamic light and heat balance design method for building facades according to claim 1, characterized in that, Step S4 includes: S41. Select a set of facade design parameters from the Pareto optimal solution set and obtain the annual hourly solar radiation data of the building location; S42. For each moment of the year, calculate the photothermal coupling index using the following formula: ； in, The photothermal coupling index at the current moment is dimensionless. This is a dimensionless coefficient representing the weighting factor for insulation requirements. The heat gain from solar radiation at the current moment (W / m²) 2 ); The design maximum daily solar radiation heat gain (W / m 2 ); This is a dimensionless weighting coefficient for lighting requirements. This is the estimated indoor daylight illuminance (lux) at the current moment. The target daylight illuminance value is between 300 and 500 (lux). S43. The thermal insulation requirement weighting coefficient and the lighting requirement weighting coefficient shall be automatically adjusted according to the accumulated days of the year using the following formula: ； in, This is the weighting coefficient for the insulation demand on that day, and it is dimensionless. This is the sine function operator; Pi; It represents the accumulated days of a year, with a value ranging from 1 to 365. This is a dimensionless weighting coefficient representing the daily lighting demand. S44. Calculate the sunshade angle based on the sign of the photothermal coupling index: when hour, ； when hour, ； in, The current angle (in degrees) of the sunshade; This is the minimum angle of the sunshade, with a value of 0 (degrees). The maximum angle of the sunshade is 90 degrees. The hyperbolic tangent function operator; The response sensitivity coefficient ranges from 2.0 to 5.

0. The absolute value of the photothermal coupling index; S45. Repeat steps S42 to S44 for all 8760 hours of the year to generate the hourly sunshade angle sequence for the whole year.

5. A BIM-based dynamic light and heat balance design system for building facades that implements the method described in any one of claims 1 to 4, characterized in that, It includes a BIM data parsing module, a proxy model module, a multi-objective optimization module, a dynamic shading calculation module, and a central control module; The BIM data parsing module includes an IFC parsing unit, a geometric calculation unit, and a parameter encoding unit connected in sequence; the IFC parsing unit is used to read the BIM model file and extract the attributes of the facade components; the geometric calculation unit is used to calculate the azimuth angle and floor height and perform azimuth zoning. The parameter encoding unit is used to calculate the orientation weight factor and the floor height attenuation factor and output the facade design parameter set; The proxy model module includes a data preprocessing unit and an energy consumption prediction network unit and a light environment prediction network unit connected in parallel with the output of the data preprocessing unit; the energy consumption prediction network unit adopts a four-layer dilated convolution structure; the light environment prediction network unit adopts a three-layer dilated convolution structure. The multi-objective optimization module includes a population management unit, a genetic operation unit, and a fitness evaluation unit; the population management unit and the genetic operation unit are bidirectionally connected; the input of the fitness evaluation unit is connected to the population management unit, and the output of the fitness evaluation unit is connected to the proxy model module to call the proxy model for performance prediction; The population management unit is used to perform non-dominated sorting and calculate adaptive crowding; the genetic operation unit is used to perform selection, crossover, and dynamic probability mutation operations. The dynamic shading calculation module includes a meteorological data interface unit, a photothermal coupling calculation unit, and an angle solving unit connected in sequence; the photothermal coupling calculation unit is used to calculate the photothermal coupling index and adjust the seasonal weighting coefficient according to the annual accumulated days; the angle solving unit is used to calculate the shading angle according to the photothermal coupling index. The central control module is communicatively connected to the BIM data parsing module, the proxy model module, the multi-objective optimization module, and the dynamic shading calculation module, respectively, and is used to call each module in sequence and transmit data.

6. The BIM-based dynamic light and heat balance design system for building facades according to claim 5, characterized in that, The output of the parameter encoding unit is connected to the first input of the central control module, and the first output of the central control module is connected to the input of the data preprocessing unit; the outputs of the energy consumption prediction network unit and the light environment prediction network unit are both connected to the input of the fitness evaluation unit.

7. The BIM-based dynamic light and heat balance design system for building facades according to claim 5, characterized in that, The output of the population management unit is connected to the input of the fitness evaluation unit to transmit the individuals to be evaluated. The output of the fitness evaluation unit is connected to the input of the population management unit to return the fitness value. The genetic operation unit receives the population data output by the population management unit after non-dominated sorting and crowding calculation, and returns the mutated offspring population to the population management unit.

8. The BIM-based dynamic light and heat balance design system for building facades according to claim 5, characterized in that, The input end of the meteorological data interface unit is connected to an external meteorological database to obtain hourly solar radiation data and solar position data throughout the year; the output end of the angle solving unit is connected to the second input end of the central control module to output the hourly shading control strategy throughout the year.

9. The BIM-based dynamic light and heat balance design system for building facades according to claim 5, characterized in that, The central control module also includes a data caching unit, which is connected to the output of the BIM data parsing module, the output of the multi-objective optimization module, and the output of the dynamic shading calculation module, respectively, and is used to store the intermediate calculation results of each module.

Images