Multi-objective optimization method for building integrated photovoltaics (BIPV) construction setting

Abstract

The invention discloses a multi-objective optimization method for building integrated photovoltaics (BIPV) construction setting, and relates to the technical field of building integrated photovoltaics. By decomposing the BIPV into a plurality of modules and carrying out attribute coding on each module, calculation of building energy consumption of the BIPV is converted into calculation of module energy consumption, the energy consumption calculation amount of a complex building is greatly reduced, meanwhile, the energy consumption of the modules can be calculated through geometrical characteristics and boundary conditions of the modules, and the calculation efficiency is improved. Constraint is carried out through construction setting of the building, that is, the overall design of the BIPV can be decomposed into modules with different attribute codes and the number of the modules with each attribute code, and the optimal optimization target in the maximum generating capacity, the minimum building energy consumption and the optimal life cycle cost is determined according to actual requirements. And then obtaining optimal geometric features, boundary conditions and a structure setting ratio through a multi-objective optimization model.

Description

Technical Field

[0001] This invention relates to the field of building-integrated photovoltaics (BIPV) technology, and in particular to a multi-objective optimization method for BIPV structural design. Background Technology

[0002] The multi-objective nature of BIPV optimization requires simultaneous consideration of maximizing power generation, minimizing building energy consumption, economic feasibility, and building integration constraints. Traditional design methods rely on sequential analysis and the designer's intuition, often leading to suboptimal solutions that fail to fully realize the potential of integrated photovoltaic systems. When considering the possible different module configurations within a modular structure, complexity increases exponentially, with different module types and assembly patterns creating unique optimization landscapes that require systematic evaluation methods. Optimizing modular building BIPV systems presents significant design challenges that are difficult to address effectively using traditional methods.

[0003] The early design phase is the most critical period for BIPV optimization. The fundamental decisions regarding the geometry, boundary conditions, and structural configuration of BIPV modules determine the performance potential throughout the building's lifecycle. However, existing design tools lack comprehensive optimization capabilities and cannot systematically assess the complex interactions between photovoltaic power generation, building energy consumption, and building lifecycle costs. Consequently, they cannot determine the optimal geometric features, boundary conditions, and structural configuration ratios of modules based on the best optimization objectives among photovoltaic power generation, building energy consumption, and building lifecycle costs in the early design phase. Summary of the Invention

[0004] The main objective of this invention is to propose a multi-objective optimization method for BIPV construction settings, which aims to solve the technical problem that existing technologies cannot systematically evaluate the complex interaction between photovoltaic power generation, building energy consumption and building life cycle cost, and obtain the optimal geometric characteristics, boundary conditions and construction settings of modules.

[0005] To achieve the above objectives, this invention proposes a multi-objective optimization method for BIPV construction settings, comprising the following steps: S100, Building decomposition: decomposing the BIPV into several modules and encoding the attributes of the modules, wherein the attribute encoding includes at least: geometric features, boundary conditions, and construction settings; S200, Data acquisition: acquiring the building scheme information and building economic information of the BIPV, and establishing a multi-objective optimization model with power generation, building energy consumption, and life cycle cost as optimization objectives; S300, Objective determination: the multi-objective optimization model determines the optimal optimization objective according to actual needs, wherein the optimal optimization objective is one or more of the maximum power generation, minimum building energy consumption, and optimal life cycle cost; S400, Objective optimization: the multi-objective optimization model calculates the attribute codes of the modules and the required quantity of each attribute code based on the optimal optimization objective, thereby obtaining the optimal ratio of geometric features, boundary conditions, and construction settings based on the optimal optimization objective during the design phase.

[0006] Preferably, step S400 includes the following steps: S401, module classification, classifying modules according to their geometric features, thermal boundary conditions, and construction settings; S402, cluster analysis, using the K-means clustering algorithm to cluster the modules into clusters, thereby reducing computational complexity; S403, representativeness optimization, performing multi-objective optimization on the modules included in each cluster to determine the optimal ratio of geometric features, boundary conditions, and construction settings; S404, solution propagation, applying the results of multi-objective optimization to modules in the same cluster.

[0007] Preferably, the following steps are included between steps S100 and S200: S101, establishing a parameter model, treating the building energy consumption of BIPV as the sum of the energy consumption of several modules, and correcting it through the thermal interaction between modules, thereby establishing a function of building energy consumption and modules; S102, establishing feature engineering, inputting the attribute codes of the modules into the parameter modeling platform, thereby generating feature data readable by machine learning algorithms; S103, model training, training the feature data through machine learning algorithms to achieve prediction of building energy consumption.

[0008] Preferably, geometric features include surface type and aspect ratio, as well as the window-to-wall ratio of the BIPV.

[0009] Preferably, the boundary conditions include the outdoors, the ground, the insulating surface, and the air wall.

[0010] Preferably, the structural configuration includes wall structure, photovoltaic structure, roof structure and ground structure.

[0011] Preferably, the lifecycle cost includes capital cost and operation and maintenance cost; wherein, the capital cost includes BIPV component cost, system cost balance, installation cost and material offset savings; and the operation and maintenance cost includes routine maintenance, component replacement and performance monitoring.

[0012] Preferably, in step S101, the parameter modeling platform is the Grasshopper platform.

[0013] Preferably, the multi-objective optimization model is based on the intensity Pareto evolutionary algorithm2.

[0014] Preferably, the machine learning prediction model is a model trained based on the XGBoost algorithm.

[0015] The beneficial effects of this invention are as follows: BIPV is decomposed into several modules, and each module is coded with attributes. This transforms the calculation of building energy consumption in BIPV into the calculation of module energy consumption, which greatly reduces the amount of energy consumption calculation for complex buildings. At the same time, the energy consumption of a module can be calculated through the geometric features and boundary conditions of the module, and constrained by the building's structural settings. That is, the overall design of BIPV can be decomposed into modules with different attribute codes, as well as the required number of modules for each attribute code. Based on actual needs, the optimal optimization objective among maximum power generation, minimum energy consumption, and optimal life cycle cost is determined. Then, the optimal ratio of geometric features, boundary conditions, and structural settings is obtained through a multi-objective optimization model.

[0016] Compared with existing technologies, this invention decomposes complex buildings into modules and encodes their attributes, transforming the overall building energy consumption, which was originally difficult to calculate directly, into the sum of module energy consumption that can be calculated in parallel. Based on this, a multi-objective optimization model integrating power generation, building energy consumption, and life cycle cost is established, and algorithms such as clustering are used to reduce optimization complexity. This enables the rapid generation and evaluation of a large number of design schemes in the early stages of BIPV design with lower computational costs, and ultimately systematically and efficiently finds the optimal design scheme. This achieves the best optimization objective of satisfying the maximum power generation, minimum building energy consumption, and optimal life cycle cost, while obtaining the optimal BIPV construction scheme. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0018] Figure 1This is a flowchart of a multi-objective optimization model. The modular boxes in the architectural scheme column refer to the modules after BIVP decomposition. Figure 2 A flowchart for machine learning; Figure 3 Optimize the framework flowchart for multiple modules; Figure 4 This is a schematic diagram of the boundary condition types for the module; Figure 5 A schematic diagram illustrating the attribute encoding for the orientation of a module surface; Figure 6 A schematic diagram of the design parameter set input for building energy consumption calculation; Figure 7 This is a schematic diagram illustrating the model relationship between building spacing and irradiance. Figures 8a-8d These are schematic diagrams showing the equivalent fitting curves of building distance and irradiance for different orientations; Figure 9 This is a schematic diagram of irradiation incident on a photovoltaic surface; Figure 10 A diagram showing the positional relationship between the photovoltaic panel and the sun; Figure 11 This is a schematic diagram illustrating the integration of machine learning models in Rhino's Grasshopper platform.

[0019] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0022] Example 1: This invention proposes a multi-objective optimization method for BIPV (Building Integrated Photovoltaics) structural design.

[0023] In this embodiment, as Figure 1 and Figure 2 As shown, a multi-objective optimization method for BIPV construction settings includes the following steps: S100, Building Decomposition: Decompose the BIPV into several modules and encode the attributes of each module. The attribute codes include at least geometric features, boundary conditions, and structural settings. S200, Data Acquisition: Acquire the building scheme information and building economic information of the BIPV, and establish a multi-objective optimization model with power generation, building energy consumption, and life-cycle cost as optimization objectives. S300, Objective Determination: The multi-objective optimization model determines the optimal optimization objective based on actual needs. The optimal optimization objective is one or more of the following: maximum power generation, minimum building energy consumption, and optimal life-cycle cost. S400, Objective Optimization: The multi-objective optimization model calculates the attribute codes of the modules and the required quantity of each attribute code based on the optimal optimization objective, thereby obtaining the optimal ratio of geometric features, boundary conditions, and structural settings based on the optimal optimization objective during the design phase.

[0024] In this embodiment, geometric features include surface type and aspect ratio, as well as the window-to-wall ratio of the BIPV; wherein, surface type includes exterior, floor, insulation surface, and air wall. Structural configuration includes wall structure, photovoltaic structure, roof, and floor.

[0025] Specifically, BIPV is decomposed into several modules, and each module is coded with attributes. This transforms the calculation of building energy consumption in BIPV into the calculation of module energy consumption, greatly reducing the amount of energy consumption calculation for complex buildings. At the same time, the energy consumption of a module can be calculated through the geometric features and boundary conditions of the module, and constrained by the building's structural settings. That is, the overall design of BIPV can be decomposed into modules with different attribute codes, as well as the required number of modules for each attribute code. Based on actual needs, the optimal optimization objective among maximum power generation, minimum building energy consumption, and optimal life cycle cost is determined. Then, the optimal ratio of geometric features, boundary conditions, and structural settings is obtained through a multi-objective optimization model.

[0026] In this embodiment, the total energy consumption of a modular building can be expressed as a function of the performance characteristics of individual modules, with the thermal interactions between modules modified through boundary conditions. The mathematical basis of the decomposition strategy can be expressed as: Formula (1) In the formula: This represents the building's total energy consumption; , representing the energy consumption of a single module i; n represents the total number of modules; , indicating the thermal interaction effect between modules.

[0027] In this embodiment, each module is treated as an independent thermal zone, and its modified boundary conditions reflect its position within the larger building components, thereby enabling parallel calculation of thermal interaction effects between them while maintaining the ability to aggregate the results of the overall building analysis; the modification of boundary conditions is achieved by classifying the module surfaces with thermal interface characteristics.

[0028] In this embodiment, as Figure 4 As shown, the boundary conditions include the external surface exposed to environmental conditions (outdoors), the internal surface interfacing with adjacent modules (insulating surfaces), and the ground contact surface (ground) with specific thermal boundary conditions. Meanwhile, Figure 4 The diagram illustrates various module assembly patterns, including horizontal and vertical combinations between two modules, three-module configurations, and four-module configurations. Each configuration exhibits different thermal interface characteristics, which must be accurately represented in the decomposition modeling approach.

[0029] For outdoor applications, the boundary conditions are implemented using standard external surface thermal modeling methods, combining convective and radiative heat transfer with ambient air and sky conditions. The heat transfer coefficient of the outdoor surface is calculated as follows: Formula (2) In the formula: , which represents the heat transfer coefficient of the outdoor surface, typically 15–25 W / (m²·K); , representing the convective heat transfer coefficient, with a value ranging from 3 to 5 W / (m²·K) for natural convection and from 10 to 100 W / (m²·K) for forced convection (affected by wind speed) (the higher the wind speed, the higher the value). , is the radiative heat transfer coefficient, which mainly depends on the surface temperature and emissivity. For the exterior surface of a typical building (emissivity of about 0.9), its value is usually between 4–6 W / (m²·K) (refer to ENISO 6946).

[0030] The surface is modeled using specific boundary conditions that consider soil thermal properties and seasonal geothermal variations. Surface heat transfer modeling uses: Formula (3) In the formula: , representing the ground coupling heat transfer coefficient, which is typically low (0.5–2.5 W / (m²·K)) and depends on soil type, moisture content, and burial depth; , representing the ground contact area; , which is the indoor air temperature; , indicating the ground temperature.

[0031] In this embodiment, as Figure 5 As shown, the geometric features of the module include orientation and surface attributes. The orientation of the module can be obtained by encoding the attributes of the six faces of the module. The attribute codes corresponding to the six faces of the module are E (east), W (west), S (south), N (north), T (top), and B (bottom).

[0032] In this embodiment, as Figure 4 As shown, for modular buildings with a determined design scheme, BIPV structural design is carried out. Regarding characteristic values, when predicting building energy consumption in a specific area, the outdoor meteorological conditions (except for building surface irradiance), indoor environmental conditions, building materials, and other universal conditions are the same, so they are not used as characteristic values.

[0033] In this embodiment, as Figure 4 and Figure 5 As shown, for each module, its characteristics in the overall building and the boundary conditions of each module are different. The east facades of modules A and B are different. AE is the building's exterior wall, and BE is an insulation wall or air wall combined with module A to form a space. The floor DB of module D is spliced ​​with the top surface of module C, while the floor CB of module C is in direct contact with the ground. It can be seen that the six surface types change as the position of the spliced ​​modules change. Therefore, the six surface attributes are used as the main variables to study the energy consumption differences between different modules.

[0034] In this embodiment, as Figure 4 and Figure 5 As shown, attribute codes (surface type, boundary conditions, structural settings, window-to-wall ratio) for 6 faces of 12 design-related building units were extracted, along with building aspect ratio (2 variables), building orientation (6 variables), and irradiance received by the facade (4 variables). To ensure compatibility with machine learning energy consumption prediction models, the data was preprocessed, such as surface type (0, 1, 2), and One-Hot encoding was performed.

[0035] In this embodiment, Table 1 presents the comprehensive machine learning variables used in the decomposition strategy, systematically classifying surface type (T0-T3), boundary conditions (B0-B2), construction settings (C0-C3), window-to-wall ratio (WWR0-7), irradiance value (D0-9), and overall form parameters including aspect ratio and rooftop indices. This systematic coding accurately represents the module's thermal characteristics while maintaining compatibility with machine learning algorithms.

[0036] Table 1 Machine Learning Variables

[0037] In this embodiment, to demonstrate the feasibility of the building decomposition strategy, a comprehensive analysis of four representative climate zones—severe cold (Harbin), cold (Beijing), hot summer and cold winter (Shanghai), and hot summer and warm winter (Shenzhen)—was conducted to verify the feasibility of the modular building decomposition strategy. The verification process involved a systematic comparison between the decomposition modeling results across different module configurations and the integrated simulation results of the entire building, as shown in Table 2, which provides the range of variable values ​​defining the scope of the verification study.

[0038] Table 2 Range of Variable Values

[0039] Based on this parameter space, 17×17×17×17×2×2×2×2 = 1,336,336 potential building samples were generated. However, due to physical constraints, some combinations were eliminated to ensure realistic building configurations. The validation method evaluated the prediction accuracy through multiple performance metrics, including mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²). The mean absolute error was calculated as follows: Formula (4) The root mean square error is calculated as follows: Formula (5) In the formula: This indicates the actual energy consumption. This indicates the predicted energy consumption. n represents the number of verification test cases.

[0040] The validation results show that, compared with the comprehensive modeling of all tested climate zones, the prediction accuracy of the decomposition modeling method is within ±10%, with the largest deviation occurring in the cold climate zone where the inter-module thermal bridging effect is most pronounced; the determination coefficients of all climate zones exceed 0.90, indicating a strong correlation between the decomposition modeling results and the comprehensive modeling results.

[0041] In this embodiment, the feature engineering framework for modular building energy prediction is built upon a comprehensive six-sided attribute coding system that captures the thermal and geometric characteristics of each building module. As shown in Table 3, this coding system systematically represents the thermal properties and boundary conditions of all six surfaces of each modular unit: north, east, south, west, top, and bottom. Each surface is characterized by four main attributes: surface type, boundary conditions, construction settings, and window-to-wall ratio.

[0042] Surface type classification employs a classification coding system to distinguish between wall surfaces (T0), roof slab surfaces (T1), floor surfaces (T2), and air wall surfaces (T3). The coding utilizes one-hot encoding to ensure compatibility with machine learning algorithms. Formula (6) in These are binary indicators representing walls, roofs, floors, and air walls, respectively.

[0043] The boundary condition specification captures the thermal interface characteristics of each surface through categorical variables representing the outdoor (B0), ground (B1), and adiabatic surface (B2). The boundary condition coding follows a mathematical representation: Formula (7) Table 3 presents a comprehensive variable combination coding system, which systematically demonstrates how different surface orientations are coded with their corresponding surface types, boundary conditions, structural settings, and window-to-wall ratios.

[0044] Table 3 Variable Combination Coding

[0045] In this embodiment, geometric features play a crucial role in predicting the energy consumption of modular buildings. The aspect ratio of the geometric features directly affects the surface area-to-volume ratio, thermal bridging effect, and solar radiation characteristics of the modules. Geometric features include aspect ratio, surface area, volume characteristics, and spatial orientation parameters, which capture the three-dimensional configuration of individual modules within a larger building component.

[0046] The aspect ratio is calculated across two horizontal dimensions, representing the length-to-width ratio and aspect ratio of a single module. These ratios affect natural ventilation patterns, structural thermal bridge locations, and solar thermal gain distribution. The aspect ratio and aspect ratio are calculated as follows: Formula (8) Formula (9) In the formula, and These represent the module's length and width dimensions, respectively.

[0047] In this embodiment, the following steps are further included between step S100 and step S200: S101. Establish a parametric model, treating the building energy consumption of BIPV as the sum of the energy consumption of several modules, and correcting it through the thermal interactions between modules, thereby establishing a function of building energy consumption and modules. Specifically, the Grasshopper-Octopus integration is used as the basic platform for BIPV optimization. Grasshopper serves as the primary parametric modeling environment, providing visual programming capabilities to support rapid prototyping and iterative design exploration of modular building configurations. The platform's node-based interface helps create complex parametric relationships between geometric parameters, material properties, and performance indicators, while maintaining clear visualization of data flow and computational dependencies. Integration with Rhino 3D modeling software provides powerful geometric modeling capabilities for accurately representing modular building components and photovoltaic system configurations.

[0048] S102. Establish feature engineering by encoding the module's attributes into the parametric modeling platform, thereby generating feature data readable by machine learning algorithms. Establish the parameter model Octopus as a multi-objective optimization engine, implementing the Strength Pareto Evolutionary Algorithm 2 (SPEA2) to systematically explore the design space and determine the optimal trade-offs between competing objectives. The integration between Grasshopper and Octopus enables direct parameter manipulation and objective function evaluation without requiring external data transfer or file-based communication protocols. This seamless integration eliminates the computational overhead associated with data transformation and enables real-time optimization feedback during the design and development process.

[0049] S103. Model training: The feature data is trained using machine learning algorithms to predict building energy consumption. Specifically, the trained XGBoost model is integrated into the Grasshopper parametric design environment through a Python script interface, so that the model can be executed directly within the optimization platform without external software dependencies or complex data transmission protocols, maintaining full compatibility with the existing parametric design workflow, and providing real-time energy feedback during multi-objective optimization.

[0050] Specifically, by collecting all attribute-encoded features from all modules, these features are combined and transformed into a set of digital feature vectors that a machine learning model can read. Specifically, such as Figure 2 As shown, Figure 2 This visualization illustrates the input design parameter set for the building energy model, demonstrating how the variables are systematically organized into manageable categories, thereby facilitating feature set construction and constraint handling. It also shows how a multidimensional parameter space was constructed in this study to enable efficient sampling and model training.

[0051] The constructed feature vector set is input into the trained machine learning prediction model, which outputs the predicted energy consumption of each module (such as one or more of the heat load, cooling load, or total energy consumption). Finally, the predicted energy consumption values ​​of all modules are added together to obtain the total energy consumption prediction value of the entire BIPV building.

[0052] In this embodiment, XGBoost (Extreme Gradient Boosting) is selected as the main machine learning algorithm for building energy prediction, given its superior performance in handling complex, high-dimensional, and mixed variable datasets.

[0053] In other embodiments, RF (Random Forest), Decision Tree, or ANN (Artificial Neural Network) are selected as the main machine learning algorithms for building energy prediction in modular buildings.

[0054] XGBoost employs an ensemble learning approach, using extreme gradient enhancement combined with multiple weak learners (such as decision trees) to iteratively improve prediction accuracy, while incorporating a regularization mechanism to prevent overfitting. The XGBoost algorithm optimizes an objective function that combines prediction error with a regularization term: Formula (10) In the formula, , representing the loss function that measures the difference between the actual value and the predicted value; , represents the regularization term for the Kth tree; This represents the total number of trees in the set.

[0055] The regularization term is defined as: Formula (11) In the formula, This controls the reduction of the minimum loss required for tree splitting; , represents the L2 regularization parameter (regularization strength); , representing the number of leaves in the tree; , representing the leaf weight.

[0056] In this embodiment, the machine learning model is trained as follows: The Grasshopper plugin for Rhino integrates parametric modeling methods with machine learning models, allowing the use of Python scripts (such as...) within Grasshopper. Figure 11 As shown, the trained XGBoost model can be accessed via a remote component interface, allowing direct execution within the parametric design environment without external software dependencies or complex data transmission protocols. Integration maintains full compatibility with existing Grasshopper workflows while adding comprehensive energy prediction capabilities. This creates an automated workflow for building energy simulation, enabling rapid dataset generation and model training, and allowing real-time energy feedback during the design and development process for real-time performance evaluation. As geometric parameters are modified within the Grasshopper interface, energy predictions are updated immediately, allowing designers to observe the impact of design changes on energy performance in real time, thus promoting energy-informed design decisions throughout the conceptual design phase.

[0057] Meanwhile, the training dataset is partitioned using an 80:20 split, with 4000 samples allocated for training and 1000 samples for validation to ensure robust performance evaluation. The training process utilizes an early stopping mechanism to prevent overfitting while maximizing prediction performance. The validation process employs k-fold cross-validation (k=5) to evaluate the model's stability and generalization ability across different data subsets.

[0058] Data validation metrics include several performance metrics, including the coefficient of determination (R²), mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE).

[0059] The formula for calculating the coefficient of determination is: Formula (12) in, , which is the mean of the actual values; , is the actual value (true value) of the i-th observation; , is the predicted value (model prediction) of the i-th observation.

[0060] To determine the optimal configuration for different climate zones and building types, this embodiment combines a grid search method with Bayesian optimization techniques. The optimization process considers several hyperparameters that significantly affect model performance, including learning rate, maximum tree depth, minimum sub-weight, subsample ratio, and regularization parameter; the specific calculation formulas are as follows: Formula (13) In the formula, , representing the learning rate; , representing the prediction from the t-th tree; , representing the prediction after t iterations.

[0061] Meanwhile, generating a comprehensive training dataset for predicting energy consumption in modular buildings requires systematic sampling of the multidimensional parameter space defined by the feature engineering framework.

[0062] In this embodiment, Latin Hypercube Sampling (LHS) is employed to ensure effective and representative parameter space coverage while minimizing the total number of simulations required. LHS provides superior space-filling properties, ensuring that each input variable is sampled uniformly across its entire range.

[0063] For modular building applications, LHS implementation involves 12 main input parameters, including six-sided attribute coding, geometric features, and irradiance. Please refer to Table 3 for specific coding details.

[0064] Training data for the machine learning model was generated using high-precision physical simulation software (such as EnergyPlus and TRNSYS). For each module parameter combination generated through Latin hypercube sampling, simulation calculations were performed using the aforementioned physical simulation software to obtain the actual energy consumption value (such as heat load and cooling load) of the module under specific meteorological conditions, which served as the target label for machine learning. The sampling process generated 5000 unique parameter combinations using EnergyPlus physical simulation software, providing sufficient diversity for robust model training while maintaining the computational feasibility of performing energy simulations using LHS. The mathematical foundation of LHS ensures a uniform distribution throughout the parameter space; the specific calculation formula is as follows: Formula (14) In the formula, , representing the normalized parameter values ​​of sample i and parameter j; , represents a random permutation of the integers j from 1 to n; , represents a random number between 0 and 1.

[0065] In this embodiment, the attribute encoding also includes the construction settings that characterize the construction settings of the module, and each construction type corresponds to a set of defined thermal parameters (such as U value, g value, etc.), which are stored in the model as known conditions in advance.

[0066] Specifically, the attribute coding method is shown in Table 1. C0 indicates that the construction setting type is wall construction; C1 indicates that the construction setting is photovoltaic construction; C2 indicates that the construction setting type is ceiling; and C3 indicates that the construction setting type is floor. Different materials of the modules have different thermal conductivity, so introducing construction settings can further correct the energy consumption prediction results.

[0067] In this embodiment, the attribute encoding also includes the wall-to-window ratio, which is used to characterize the building structure settings that affect the solar thermal gain coefficient. Please refer to Table 1 for the numerical range of the wall-to-window ratio.

[0068] Specifically, the wall-to-window ratio (WWR) reflects solar radiation heat gain, heat gain and heat loss, and natural and artificial lighting. Regarding heat gain and loss, windows typically have a higher heat transfer coefficient (U-value) than walls; therefore, a higher WWR increases heat loss in winter and heat gain in summer, leading to changes in heating and cooling loads. For solar radiation heat gain, the window's solar heat gain coefficient (SHGC) directly affects indoor solar radiation heat gain. A higher WWR can reduce heating demand in cold regions but increase cooling load in hot regions. Regarding natural and artificial lighting, a reasonable WWR can improve indoor lighting and reduce lighting energy consumption, but excessive window opening can lead to glare or thermal comfort problems. Therefore, the wall-to-window ratio is used as an optimization variable input into a machine learning model, combined with other surface features to predict energy consumption, resulting in more accurate energy consumption predictions.

[0069] In this embodiment, the attribute encoding also includes geometric features used to characterize the natural ventilation mode, the location of the structural thermal bridge and the distribution of solar thermal gain. The geometric features include at least the aspect ratio and the width-to-height ratio of the module. The specific attribute encoding method is shown in Table 1, where A0 represents the aspect ratio in the horizontal direction and A1 represents the width-to-height ratio in the vertical direction.

[0070] Specifically, geometric features affect energy consumption prediction in the following ways: In natural ventilation mode, the aspect ratio and width-to-height ratio directly affect the ratio of the module's surface area to its volume. A higher aspect ratio (slender module) may result in a larger outer surface area, thereby increasing heat exchange with natural ventilation and affecting the heating and cooling loads.

[0071] Structural thermal bridges, such as those at module edges and connections, can create thermal bridging effects, increasing heat transfer and leading to higher energy consumption.

[0072] The distribution of solar thermal gain, module geometry, and aspect ratio all influence the distribution of solar irradiance. For example, slender modules may receive uneven irradiance on different facades, thus affecting photovoltaic power generation and building heat load.

[0073] In this embodiment, the attribute encoding also includes irradiance used to characterize the module's irradiance affected by the spatial shading effect. Irradiance represents the annual average irradiance. For a region, the annual average irradiance fluctuates within a certain range and can be regarded as a constant. At the same time, the spatial effect formed by the building spacing also has a significant impact on irradiance.

[0074] like Figure 7 , Figure 8a , Figure 8b , Figure 8c as well as Figure 8d As shown, closer building spacing (0-1m) results in a significant reduction in irradiance due to spatial shading effect, while larger spacing (8-9m) allows for maximum solar radiation. The mapping relationship between building spacing and irradiance can be found in Table 4.

[0075] Table 4. Building Spacing-Irradiance Relationship Mapping

[0076] In this embodiment, solar radiation directly affects building heat load and photovoltaic power generation potential. Within the building complex where the module is located, the building spacing can affect solar irradiance, and thus the thermal effect. Therefore, building spacing, as an optimization parameter for building energy consumption, can further improve the accuracy of energy consumption prediction.

[0077] In this embodiment, as Figure 1 As shown, generating BIPV requires inputting two types of parameters: first, building scheme information, including the building scheme (six-sided coding of modular boxes), the connection method of modular boxes (double-box or triple-box connection), climate zone, surrounding environment, building type, etc.; second, building economic information, mainly involving system lifespan, discount rate, component cost, balance sheet system cost, installation cost, operation and maintenance cost, material offset benefits, and fiscal incentives required for photovoltaic life cycle calculation. This study constructed an economic database in Grasshopper, as shown in Table 5. Users can select and output relevant information according to their actual situation.

[0078] Table 5. BIPV Economic Database Architecture

[0079] Example 2

[0080] This embodiment provides a photovoltaic power generation model to calculate the amount of electricity generated.

[0081] The standard test conditions (STC) for solar photovoltaic cells are specified as follows: irradiance of 1000 W / m², cell temperature of 25°C, and AM1.5 spectral distribution, to ensure performance parameters such as open-circuit voltage. Short-circuit current Maximum power However, in actual operating environments, factors such as irradiance, temperature, spectrum, and atmospheric quality often deviate from standard test conditions, leading to significant changes in output characteristics. For example, for every 1°C increase in temperature, the output characteristics of monocrystalline silicon modules... The decrease is approximately 0.3-0.5% (temperature coefficient), while Because the carrier mobility exhibits a positive temperature coefficient (approximately 0.05% / ℃); for every 100 W / m² decrease in irradiance, It decreases linearly, while The photovoltaic (STC) output exhibits nonlinear degradation due to the series resistance. To achieve accurate modeling under any environment, the STC parameters need to be corrected in multiple dimensions, thereby converting the performance parameter values ​​under standard test conditions into performance parameter values ​​under actual working environment conditions, thus enabling the simulation of the photovoltaic cell output characteristics under any environmental conditions. In this embodiment, the photovoltaic power generation is corrected according to the actual situation using the following photovoltaic performance parameters: Formula (15) In the formula: Indicates the photovoltaic power generation efficiency at time t; This indicates the photovoltaic output under standard test conditions, i.e., the photovoltaic installed capacity. Indicates solar irradiance; unit: W / m². This represents the irradiance under standard test conditions; set to 1000 W / m². This represents the power temperature coefficient, typically taken as -0.43%. This indicates the battery temperature under standard test conditions; the experiment was set at 25℃. This indicates the temperature of the battery when the PV is at its operating point; it can be calculated from the ambient temperature and solar irradiance or backsheet temperature.

[0082] If the ambient temperature, irradiance, and rated operating temperature of the module are known, the temperature of the photovoltaic cell can be calculated as follows: Formula (16) In the formula: Indicates ambient temperature in °C; Indicates the rated operating temperature of the component in °C.

[0083] If the backsheet temperature from the test data is known, and considering the impact of the external environment of the Sodia laboratory on photovoltaic power generation, the temperature of the photovoltaic cell can be calculated as follows: Formula (17) In the formula: This indicates the temperature of the cells in a photovoltaic module, in degrees Celsius (°C). This indicates the measured temperature on the back of the photovoltaic module, in degrees Celsius (°C). This indicates the measured solar irradiance on the photovoltaic module, expressed in watts per square meter (W / m²). The reference solar irradiance for photovoltaic modules is typically 1000 W / m². This generally indicates a temperature of 2-3℃.

[0084] The angle of incident sunlight is also a factor affecting power generation calculations; incident solar energy is the main factor determining the magnitude of photovoltaic power generation. Important characteristics of incident solar energy include: the solar radiant power density; the angle at which incident solar radiation strikes the photovoltaic module; and the radiant energy from the sun to a specific surface over a year or a day. Although solar radiation reaching the Earth's atmosphere is relatively constant, the radiation on the Earth's surface varies greatly due to changes in latitude, longitude, and the time of year. Therefore, calculating the irradiance incident on a photovoltaic surface requires considering numerous factors.

[0085] The amount of solar radiation incident on the tilted module surface is the component of the incident solar radiation perpendicular to the module surface.

[0086] In this embodiment, as Figure 4 As shown, based on the amount of solar radiation measured on a horizontal surface ( ) or solar radiation measured perpendicular to the sun ( ) Calculate the amount of radiation incident on the inclined surface ( The specific calculation formula is as follows: Formula (18) Formula (19) For modules with arbitrary tilt and orientation, the formula is as follows: Formula (20) In the formula: α is the angle of elevation; It is the solar azimuth angle; It is the module tilt angle.

[0087] Modules laid flat on the ground =0°, vertical module =90°. This is the azimuth angle facing the module. Most modules are aligned to face the equator. Modules in the Southern Hemisphere will face north. = 0°, while modules in the Northern Hemisphere typically face directly south. =180°. and These are the light intensity on the module and the intensity of the incident light, respectively (unit: W / m²). It is the direct light component.

[0088] Given the azimuth and tilt angle of a photovoltaic installation at a certain location, the incident light intensity on a module with arbitrary tilt and direction can be calculated using formula (20), such as... Figure 5 As shown.

[0089] Elevation angle is the angular height of the sun above the horizontal line. At sunrise, the elevation angle is 0°, and when the sun is directly overhead (e.g., at the equator during the spring and autumn equinoxes), it is 90°. The elevation angle varies throughout the day. It also depends on the latitude of a given location and the number of days in a year. A key parameter in photovoltaic system design is the maximum elevation angle, which refers to the highest altitude the sun reaches at a specific time of year. This maximum elevation angle typically occurs around noon and is influenced by latitude and the sun's declination.

[0090] in: δ is latitude (positive in the Northern Hemisphere, negative in the Southern Hemisphere); δ is the solar declination, which depends on the day of the year.

[0091] The elevation angle α can be obtained using the following formula: Formula (21) Solar declination (denoted by δ) is a seasonal variation caused by the tilt of the Earth's axis of rotation and the Earth's revolution around the Sun. If the Earth's axis were not tilted, solar declination would always remain at 0°. However, due to the Earth's axial tilt of 23.45°, solar declination varies within a positive and negative range. Only at the spring and autumn equinoxes is solar declination equal to 0°. Solar declination can be calculated using the following formula: Formula (22) In the formula: d represents a day of the year; on January 1st, d = 1. Declination is zero at the spring equinoxes (March 22nd and September 22nd), positive in the Northern Hemisphere summer, and negative in the Northern Hemisphere winter. Declination reaches its maximum of 23.45° on June 22nd (Northern Hemisphere summer solstice) and its minimum of -23.45° on December 21st-22nd (Northern Hemisphere winter solstice). This equation assumes the sun's orbit is a perfect circle, and the factor 360 / 365 converts the number of days into position in the orbit.

[0092] The hour angle is used to convert local solar time (LST) into the angular position of the sun in the sky. By definition, the hour angle is 0° at solar noon. Because the Earth rotates 15° per hour, the sun's angle in the sky changes by 15° every hour from solar noon. The hour angle is negative in the morning and positive in the afternoon. The formula for calculating the hour angle is as follows: Formula (23) Local solar time 12 noon (LST) is defined as the moment when the sun is at its highest point in the sky. Local time (LT) is usually different from LST due to the eccentricity of the Earth's orbit and human adjustments such as time zones and daylight saving time.

[0093] The Time Correction Factor (TC) (in minutes) corrects for variations in local solar time (LST) caused by longitude changes within a time zone, the eccentricity of the Earth's orbit, and the Earth's axial tilt. The 4-minute factor is due to the actual fact that the Earth rotates 1° every 4 minutes. The formulas for calculating local solar time and the Time Correction Factor are as follows: Formula (24) Formula (25) The Local Standard Time Meridian (LSTM) is a reference meridian for a specific time zone, similar to the Prime Meridian used for Greenwich Mean Time. Its calculation formula is as follows: Formula (26) In the formula: It is the hour difference between local time (LT) and Coordinated Universal Time (UTC). This also equals a time zone; 15° = 360° / 24 hours. For example, China is in the East 8 time zone, so the LSTM is 120°E.

[0094] Mean time deviation (E_o T) is an empirical equation in minutes used to correct for the effects of Earth's orbital eccentricity and axial tilt. It provides an adjustment value for time to accurately align true solar time with uniform time. Its approximation is accurate to within half a minute, used to adjust for errors in daily observations and time calculations, ensuring consistency with astronomical observations in practical applications. () is an empirical equation measured in minutes, used to correct for the eccentricity of Earth's orbit and the tilt of its axis. An approximation accurate to within half a minute is: Formula (27) in: Formula (28) Azimuth refers to the direction of sunlight, using a compass as a reference. At noon, the sun is always due south in the Northern Hemisphere and due north in the Southern Hemisphere. Azimuth changes throughout the day. At the spring and autumn equinoxes, regardless of latitude, the sun rises due east and sets due west; therefore, the azimuth at sunrise is 90°, and the azimuth at sunset is 270°. However, in general, azimuth varies with latitude and time of year. The complete formula for calculating the sun's position throughout the day is as follows: Formula (29) In the formula: α is the angle of elevation. It's latitude. It's an angle.

[0095] The above equation only gives the correct azimuth angle for the sun in the morning; therefore: when LST < 12 or HRA < 0, the azimuth angle = LST > 12 or HRA > 0, azimuth = 360° The core of photovoltaic surface irradiance calculation lies in converting horizontal surface irradiance data into irradiance data under inclined surfaces or tracking systems.

[0096] Example 3

[0097] This embodiment provides a specific optimization method for multi-objective optimization, based on any of the above embodiments.

[0098] In this embodiment, SPEA2 serves as the primary optimization engine for the BIPV system design, maintaining solution diversity while handling multiple conflicting objectives. SPEA2's fitness allocation mechanism considers solution dominance relationships and population density characteristics to guide the evolutionary search to the optimal region of the design space.

[0099] The SPEA2 algorithm maintains two populations: the primary population and the secondary population. External archives of non-dominant solutions discovered during storage optimization. The fitness assignment procedure calculates the strength value of each individual based on the number of its dominant solutions: Formula (30) In the formula: Indicates the intensity of individual i; This indicates that solution i is superior to solution j.

[0100] The original fitness calculation includes strength values ​​from the dominant solution:

[0101] Formula (31) In the formula: This represents the initial fitness of individual i; a lower value indicates better performance.

[0102] Density estimation provides an additional distinction between solutions with the same original fitness value: Formula (32) In the formula: Let represent the distance to the k-th nearest neighbor in the target space, where ; The final fitness value combines the original fitness and density information: Formula (33) Table 6 shows the optimization parameter settings used in the Octopus implementation to ensure a balance between exploration and utilization throughout the optimization process.

[0103] Table 6 Octopus Parameter Settings

[0104] In this embodiment, as Figure 3 As shown, in step S400, the objective optimization process based on the multi-objective optimization framework specifically includes the following steps: S401. Module classification: Modules are classified according to their surface type, boundary conditions, structural settings, irradiance, wall-to-window ratio, and aspect ratio.

[0105] S402. Cluster analysis: The K-means clustering algorithm is used to cluster the modules into clusters, thereby reducing computational complexity. Specifically, the K-means algorithm is used to identify naturally occurring groups within each module category, reducing the optimization space from 200 individual modules to approximately 15-20 representative clusters.

[0106] S403, Representative Optimization: Perform multi-objective optimization on the modules included in each cluster, and perform comprehensive multi-objective optimization on the representative modules of each cluster to determine the optimal geometric features, boundary conditions and construction settings ratio, thereby determining the optimal BIPV configuration.

[0107] S404, Solution Propagation, applies the results of multi-objective optimization to modules in the same cluster, ensuring consistent performance across similar building components.

[0108] In this embodiment, multi-objective optimization addresses three main objectives for capturing the basic performance characteristics of a BIPV system: minimizing building energy consumption, maximizing photovoltaic power generation, and optimizing lifecycle costs. These three main objectives often conflict with each other, requiring a systematic trade-off analysis to determine the optimal design solution.

[0109] The energy consumption target is achieved by optimizing the configuration of building structure features and photovoltaic system integration to minimize the building's annual energy intensity; Formula (34) In the formula: EUI stands for Building Energy Intensity (kWh / m²·year). Total annual energy consumption; This refers to the building area.

[0110] The goal of photovoltaic power generation is to maximize annual power generation by optimizing the scale and configuration of photovoltaic systems. Formula (35) In the formula: APEG stands for annual photovoltaic power generation (kWh / year). This refers to the instantaneous photovoltaic output power. The time step is 1 hour.

[0111] The lifecycle cost objective minimizes the net present value of all costs and benefits associated with the deployment of the BIPV system during the analysis period.

[0112] Formula (36)

[0113] In the formula: LCC stands for Life Cycle Cost; Indicates the cost of capital; Indicates operating and maintenance costs; To save energy costs; Government subsidies and tax breaks.

[0114] In this embodiment, lifecycle costs include capital costs and operating costs.

[0115] Capital costs include BIPV component costs, system cost balance, installation costs, and material offset savings.

[0116] Operation and maintenance costs include routine maintenance, component replacement, and performance monitoring.

[0117] The capital cost formula takes into account the unique characteristics of modular structures, which affect costs and installation procedures: Formula (37) In the formula: Indicates net cost of capital; Indicates the cost of BIPV components; Indicates the system cost balance; Indicates installation cost; This indicates material savings; This indicates fiscal incentives.

[0118] BIPV component costs vary considerably depending on the technology type, module size, and integration complexity. Cost calculations incorporate unit pricing per installed watt of capacity. PVC = Unit cost (yuan / W) × System capacity (W) Formula (38) The system cost balance includes all electrical components required for grid connection and system operation; the cost scales according to system capacity. BOSc = Inverter cost (yuan / kW) × System capacity (kW) + Integration cost (yuan / m²) × Module area (m²) Formula (39) Installation costs include labor and equipment expenses associated with BIPV system deployment. For modular building applications, factory-based installation can significantly reduce these costs compared to on-site-based installation. Formula (40) Material offset savings recognize that BIPV systems can replace traditional building construction materials, reducing overall project costs: Formula (41) Operation and maintenance costs are recurring expenses throughout the entire BIPV system lifecycle, including routine maintenance, component replacement, and performance monitoring. The present value calculation of operation and maintenance costs incorporates the effects of inflation and the discount rate. Formula (42) In the formula: Indicates annual maintenance cost; This indicates the cost of replacing the inverter; n represents the system lifetime; r represents the discount rate.

[0119] Depending on system complexity and local service availability, annual maintenance costs typically range from 15 to 25 yuan per installed kilowatt. Inverter replacement costs occur approximately every 10 years, representing a significant maintenance expense. Formula (43) Fiscal incentives provide substantial support for BIPV deployment. One-time installation subsidies offer direct cost reductions for BIPV projects. Different regions have different fiscal incentive mechanisms for BIPV, typically calculated based on installed capacity. Formula (44).

[0120] Example 4

[0121] This embodiment provides a constraint processing method based on any of the above embodiments.

[0122] In this embodiment, constraint processing ensures that all generated design schemes meet practical limitations. The constraint framework addresses multiple performance domains; this embodiment, using a school as an example, proposes constraint processing conditions including sufficient lighting, natural ventilation requirements, and factors affecting design feasibility and resident comfort.

[0123] Daylight factor requirements

[0124] Lighting constraints must maintain minimum illuminance levels in accordance with building standards and occupancy requirements. For educational facilities in climatic zones, the daylight factor of classroom desk surfaces must not be less than 2% to ensure sufficient natural lighting and visual comfort for student activities.

[0125] Formula (45)

[0126] In the formula: The average solar radiation coefficient; The minimum required solar radiation coefficient; This is the climate zone correction factor.

[0127] The daylighting calculation incorporates regional differences by taking into account climate-specific correction factors that account for variations in outdoor illuminance levels and sky conditions. The baseline correction factor for climate zone III is 1.0, while for other climate zones, a baseline correction factor of 0.85–1.20 is used based on the characteristics of regional solar energy resources.

[0128] Window area ratio control

[0129] The window-to-floor area ratio restriction ensures adequate natural ventilation and emergency exit capacity, which is crucial for the safety and comfort of educational buildings. Figure 10 This explains the geometric relationship between the window area, wall area, and floor area calculated using the defined constraints.

[0130] Formula (46)

[0131] In the formula: Indicates the total window area; This indicates the area occupied by the classroom.

[0132] The window area ratio requirement for educational buildings is 1:5, which can ensure sufficient natural light penetration while maintaining appropriate thermal performance characteristics, but may lead to excessive solar thermal gain and cooling load, while ensuring sufficient sunlight for educational activities.

[0133] In the process of parametric modeling, constraint handling also includes: evaluating each generated parameter combination according to predefined feasibility criteria and eliminating invalid combinations.

[0134] For example, the system determines whether the "wall-to-window ratio (WWR) of the ground (T2) is coded as greater than 0" (in the predefined case, the ground does not contain windows according to industry standards, and therefore does not have a wall-to-window ratio). The system automatically identifies and removes these invalid feature combinations, ensuring that only reasonable and realistic building configurations are sent into the model for prediction. This ensures the practical significance and reliability of the prediction results and avoids erroneous predictions caused by unreasonable input data.

[0135] The constraints are enforced using a penalty function to prevent infeasible solutions while maintaining population diversity during optimization; solutions that violate the daylighting or window-to-wall ratio constraints receive a fitness value that is proportional to the degree of constraint violation.

[0136] Example 5

[0137] This embodiment provides a specific k-means clustering method based on any of the above embodiments.

[0138] In this embodiment, the k-means clustering algorithm provides a systematic classification based on the module's boundary conditions, geometric features, and construction settings, thereby identifying representative modules that can be optimized individually and then applied to similar modules throughout the BIPV. This leverages the inherent similarities between modular building components while considering variations in boundary conditions, sunlight exposure, and thermal properties. Figure 9 The results of module k-means clustering analysis are presented, demonstrating the grouping of module systems based on geometric features. The k-means clustering method processes multiple input parameters, including surface properties, boundary conditions, solar irradiance levels, and geometric features, to identify naturally occurring populations within a module population; the feature engineering process of the clustering analysis incorporates the attribute encoding of the modules.

[0139] In this embodiment, the number of clusters (K value) in the k-means clustering method is automatically determined using the Elbow Method. The specific process is as follows: For each K value (K=1 to 10), calculate the Within-Cluster Sum of Squares (WCSS). Plot the relationship between K value and WCSS to find the elbow inflection point. The optimal cluster is selected based on the K value that significantly slows down the decline in WCSS. The WCSS calculation formula is as follows:

[0140] in: K is the number of clusters. For the i-th cluster Let i be the centroid of the i-th cluster These are data points within a cluster.

[0141] Cluster analysis is used as an example in a case study of a school building: Because many modules of the campus buildings are similar, and considering the mass production of these modules, different types of building modules are clustered separately. Each identical box in each category uses a different photovoltaic installation scheme. k-means is a popular unsupervised learning technique whose core objective is to assign data points to a pre-defined number of clusters.

[0142] Based on the input building model data, the Grasshopper program automatically extracts the surface type, boundary conditions, and irradiance of each building. This data forms the basis for subsequent cluster analysis. Since the data includes both numerical and categorical data, standardization of the numerical data is necessary. A combined strategy of categorical coding and numerical standardization is employed: one-hot encoding of categorical variables converts the six facade boundary variables into a binary feature matrix. Standardization ensures consistent scale across all features, avoiding adverse effects on clustering results due to large numerical ranges of certain features, thereby improving the accuracy and reliability of clustering. The processed data is shown in Table 7.

[0143] Table 7 Standardization of Single-Box Characteristic Values

[0144] Since each module has the same building surface type, it is not used as a feature value. The feature values ​​of each type of module are shown in Table 8, which are the attributes of each surface and the amount of irradiance received by the four facades.

[0145] Table 8. Examples of feature module values ​​(taking modules A1 to A10 as examples)

[0146] In this embodiment, after processing the data using the K-means clustering algorithm, the clustering results for each module were obtained based on the characteristics of the building modules. For each cluster, a representative module was selected as the center of each class, and different types of modules were grouped together. Since the data contains both numerical and categorical data, data standardization is crucial during the clustering process, ensuring consistency between different features. In this clustering, the number of clusters was automatically set based on numerical features. For example, module D, due to its small number and high similarity, was divided into 2 clusters, while other modules were divided into 3 clusters. The resulting clustering results are shown in Tables 5-6. For ease of understanding, Figure 5-11 Cluster analysis diagrams for different module types are displayed, including clustering results for modules A and AR, showing the distribution of different types of modules in the feature space and clearly indicating the cluster centers.

[0147] Table 9 Cluster Analysis Results

[0148] Cluster analysis revealed distinct patterns within each module type, reflecting variations in solar irradiance, boundary conditions, and thermal properties. Module A exhibited three distinct clusters, with cluster 0 representing the largest group (20 modules) characterized by moderate solar irradiance and mixed boundary conditions. Module AR showed similar clustering patterns but with varying proportions, reflecting a mirror relationship between these module types. Automatic clustering determination based on numerical features resulted in different cluster counts across different module types. Module D showed only two clusters due to its smaller population and higher similarity among instances, while other module types showed three clusters, reflecting greater diversity in thermal and solar irradiance characteristics.

[0149] By selecting representative modules from each cluster, systematic optimization can be performed on all conditions encountered in each module type while maintaining computational efficiency. The selected representative modules serve as optimization targets, and their results are subsequently applied to all modules within their respective clusters to ensure consistent performance across similar building components.

[0150] Example 6

[0151] This embodiment provides a specific calculation method for the optimal optimization target, based on any of the above embodiments.

[0152] In this embodiment, the annual load per unit building area, annual photovoltaic power generation, and BIPV life cycle cost were used to evaluate the energy-saving effect of the building scheme.

[0153] (1) Energy Use Intensity (EUI) per unit building area

[0154] Building load per unit area refers to the ratio of the total energy consumed by a building in one year to the total building area, expressed in units of... This includes building heating energy consumption, cooling energy consumption, lighting energy consumption, and equipment energy consumption. The calculation formula is as follows: Formula (47) In the formula: Indicates heating energy consumption; Indicates cooling energy consumption; Indicates energy consumption for lighting; This indicates the energy consumption of the equipment.

[0155] (2) Annual Photovoltaic Energy Generation (APEG)

[0156] Photovoltaic power generation is an important indicator for evaluating the performance of BIPV systems. It is usually calculated based on factors such as the building's sunlight conditions, photovoltaic module efficiency, and installation angle. By integrating photovoltaic power generation modules, the photovoltaic power generation and overall energy efficiency of a building can be dynamically optimized.

[0157] Please refer to Example 2 for the formula for calculating photovoltaic power generation; it will not be repeated here.

[0158] (3) Life Cycle Cost (LCC)

[0159] The life cycle cost (LCC) of BIPV refers to all capital costs and operation and maintenance costs (O&M costs) incurred throughout the entire life cycle of a BIPV system, taking into account the time value of money. The sum after discounting.

[0160] The initial capital cost of a BIPV building envelope mainly includes the following components: The costs include BIPV component costs, installation structure materials and their installation fees, and balance system (BOS) materials and installation costs. The BOS encompasses electrical components such as inverters, cables, and junction boxes.

[0161] The calculation of LCC needs to be achieved through dynamic cost analysis, and its core formula is as follows: Formula (48) In the formula: This indicates the initial investment cost or cost of capital for a BIPV project. The calculation is performed according to formula (37), which will not be elaborated here; The total cost of operation and maintenance (O&M cost) of the BIPV project lifecycle is calculated according to formula (42), which will not be elaborated here.

[0162] Daylight factor (DF) is an important indicator in architectural design for evaluating the level of natural lighting indoors. It refers to the percentage of standard outdoor illuminance (usually horizontal illuminance under a standard clear sky) received at a point inside a building. A high DF reflects the sufficiency of natural light inside a building; a higher DF means better indoor natural lighting conditions, which helps improve indoor visual comfort and reduce the cost of artificial lighting. Different light climate zones (as shown in Figure 11) have different requirements for DF (as shown in Table 10). Table 11 lists the standard design values ​​for daylighting in architectural design for each light climate zone.

[0163] Table 10 Daylight Factor

[0164] Table 11. Standard daylighting values ​​for each daylighting grade (Source: Standard for Daylighting Design of Buildings GB 50033-2013)

[0165] In this embodiment, after importing the machine learning model, design variables generated during the optimization process, including window-to-wall ratio and photovoltaic installation status, can be directly derived. Through visualization on the Rhino & Grasshopper platform, designers can view the changes in these variables in real time. This is based on the light climate zone (Class IV) requirements (e.g., Class I DF ≥ 0.85). (18000lx), use Grasshopper's Cull Pattern component to remove solutions that do not meet the conditions.

[0166] After optimization, this study uses the Pareto optimal solution set for BIPV building envelopes generated by the Rhino-Grasshopper and Octopus platforms, which includes multiple schemes balancing annual load per unit building area, annual photovoltaic power generation, and life cycle cost (LCC). Three priority strategies are designed, and different design schemes are introduced according to actual needs during the results analysis phase.

[0167] When power generation is taken as the best optimization objective, the logic of multi-objective optimization follows Objective2 (photovoltaic power generation) > Objective1 (annual net energy consumption) > Objective3 (life cycle cost). When building energy consumption is taken as the best optimization objective, the logic of multi-objective optimization follows Objective1 (annual net energy consumption) > Objective2 (photovoltaic power generation) > Objective3 (life cycle cost). When lifecycle cost is the optimal optimization objective, the logic of multi-objective optimization follows Objective3 (total lifecycle cost) > Objective1 (annual net energy consumption) > Objective2 (photovoltaic power generation).

[0168] For the selected optimal optimization objective, the results and their visualizations can be exported to various formats such as Excel through the Grasshopper platform for further analysis and report generation. The reports include design parameters, economic assessments, energy consumption calculations, and photovoltaic power generation forecasts, facilitating reporting and decision support for users.

[0169] In this embodiment, the optimization tool not only generates diverse BIPV optimization schemes, but also supports flexible decision-making based on priorities or constraints, ensuring that users' multiple requirements for energy saving, economy and power generation are met.

[0170] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, several equivalent substitutions or obvious modifications can be made without departing from the concept of the present invention, and all such modifications, achieving the same performance or purpose, should be considered within the scope of protection of the present invention.

Claims

1. A multi-objective optimization method for BIPV construction settings, characterized in that, Includes the following steps: S100, Building Decomposition: Decompose the BIPV into several modules and encode the attributes of the modules, wherein the attribute encoding includes at least: geometric features, boundary conditions, and construction settings. S200, Data Acquisition: Acquire the building scheme information and building economic information of the BIPV, and establish a multi-objective optimization model with power generation, building energy consumption and life cycle cost as optimization objectives; S300. Target determination: The multi-objective optimization model determines the optimal optimization target based on actual needs, wherein the optimal optimization target is one or more of the following: maximum power generation, minimum building energy consumption, and optimal life cycle cost. S400, Target Optimization: The multi-objective optimization model calculates the attribute codes of the module and the required quantity of each attribute code based on the optimal optimization objective, thereby obtaining the optimal geometric features, boundary conditions, and construction settings ratio based on the optimal optimization objective during the design phase.

2. The multi-objective optimization method for BIPV construction settings as described in claim 1, characterized in that, Step S400 includes the following steps: S401. Module classification: Classify the modules according to their geometric features, thermal boundary conditions, and structural settings. S402. Cluster analysis: The modules are clustered into clusters using the K-means clustering algorithm, thereby reducing computational complexity. S403, Representative optimization: Perform multi-objective optimization on the modules included in each cluster to determine the optimal geometric features, boundary conditions and construction settings ratio; S404, Solution propagation, applying the results of the multi-objective optimization to the modules in the same cluster.

3. The multi-objective optimization method for BIPV construction settings as described in claim 1, characterized in that, The following steps are also included between step S100 and step S200: S101. Establish a parameter model, regard the building energy consumption of the BIPV as the sum of the energy consumption of several modules, and correct it through the thermal interaction between the modules, thereby establishing a function of the building energy consumption and the modules. S102. Establish feature engineering by inputting the attribute codes of the module into the parameter modeling platform to generate feature data readable by machine learning algorithms. S103. Model training: The feature data is trained using a machine learning algorithm to predict building energy consumption.

4. The multi-objective optimization method for BIPV construction settings as described in claim 1, characterized in that, The geometric features include the surface type and aspect ratio, as well as the window-to-wall ratio of the BIPV.

5. The multi-objective optimization method for BIPV construction settings as described in claim 1, characterized in that, The boundary conditions include the outdoors, the ground, the insulating surface, and the air wall.

6. The multi-objective optimization method for BIPV construction settings as described in claim 1, characterized in that, The structural configuration includes wall structure, photovoltaic structure, roof and floor.

7. The multi-objective optimization method for BIPV construction settings as described in claim 1, characterized in that, The lifecycle cost includes capital cost and operation and maintenance cost; The capital costs include BIPV component costs, system cost balances, installation costs, and material offset savings. The operation and maintenance costs include routine maintenance, component replacement, and performance monitoring.

8. The multi-objective optimization method for BIPV construction settings as described in claim 1, characterized in that, In step S101, the parameter modeling platform is the Grasshopper platform.

9. The multi-objective optimization method for BIPV construction settings as described in claim 1, characterized in that, The multi-objective optimization model is based on the intensity Pareto evolutionary algorithm.

10. The multi-objective optimization method for BIPV construction settings as described in claim 1, characterized in that, The machine learning prediction model is a model trained based on the XGBoost algorithm.

Images