Method for optimizing photothermoelectric performance of vacuum photovoltaic glass based on XGBoost and NSGA-II

Abstract

The invention discloses an XGBoost and NSGA-II-based vacuum photovoltaic glass photo-thermal electric performance optimization method. The method comprises the following steps: S1, calculating overall photo-thermal performance parameters of PVVG; s2, determining a climate partition, a building orientation, basic parameters of rooms for installing PVVGs and overall photo-thermal performance parameters of different PVVGs, and inputting the information into Energyllus to generate a building performance data set containing a plurality of samples; s3, calling the building performance data set through an XGBoost machine learning algorithm, and respectively training agent models capable of quickly and accurately predicting building refrigeration, heating and lighting energy consumption and photovoltaic power generation capacity for the building with the PVVG; s4, setting a design variable and a target function; by adopting the method provided by the invention, a series of optimal combinations are obtained through continuous optimization and global optimization and balancing conflicts or cooperative relationships among objective functions, so that the subjectivity and limitation of manual parameter adjustment are avoided, and various choices are provided for different design scenes.

Description

Technical Field

[0001] This invention relates to the field of energy-saving optimization technology for building envelopes, specifically a method for optimizing the photothermal and electrical performance of vacuum photovoltaic glass based on XGBoost and NSGA-II. Background Technology

[0002] As transparent components within the building envelope, windows have a profound impact on the building's thermal environment and energy consumption levels. Insufficient window insulation can account for up to 30% of a building's heating and cooling energy consumption. Simultaneously, the optical performance of windows directly affects indoor lighting and visual comfort. High-performance windows with good thermal insulation and suitable lighting are crucial for improving building energy efficiency. With the rapid development of the global clean energy market, the focus of building energy conservation research is accelerating towards new types of windows that combine energy efficiency and environmental control capabilities.

[0003] Photovoltaic glass (PVG), as the only energy-saving window capable of actively capturing and utilizing solar radiation, can directly convert solar energy into electricity and also provides shading. However, PVG has poor thermal insulation performance. In contrast, vacuum glass (VG) has excellent thermal insulation performance and a mature industrial base, but it is prone to glare. The combination of these two technologies, known as vacuum photovoltaic glass (PVVG), achieves complementary advantages, effectively realizing the synergy of shading, power generation, and thermal insulation. With the continuous advancement of active and passive energy-saving technologies, PVVG provides a new path to realizing the architectural vision of "self-generating power, high thermal insulation, and low energy consumption."

[0004] However, current performance optimizations for PVVG still have the following shortcomings: (1) During optimization, parameters are often changed manually to compare multiple solutions. The optimization results are discrete and have poor continuity, making it difficult to find the global optimal solution. (2) There is a clear conflict between the energy consumption for heating, cooling, and lighting and the photovoltaic power generation, and the optimal parameter combination cannot be determined by human experience; (3) Traditional optimization methods call building energy consumption simulation software one by one for calculation. The model iteration efficiency is low and the amount of calculation is large. There is a lack of algorithm framework and energy consumption prediction model that can calculate quickly and accurately and support large-scale multi-objective optimization.

[0005] XGBoost (eXtreme Gradient Boosting) is an optimized distributed gradient boosting library developed by Tianqi Chen et al. Based on gradient boosting decision trees, this library introduces regularization terms and second-order derivative information to improve performance, supporting parallel computing, sparse data processing, and various distributed environments. Through innovative techniques such as weighted histogram algorithms and caching optimization, the system demonstrates superior performance in tasks such as classification and regression, and is widely used in data competitions and industrial applications. As of 2024, its distributed version supports environments such as Hadoop, SGE, and MPI, processing billions of samples. Empirical studies in fields such as medicine have shown that the model's accuracy outperforms traditional methods.

[0006] NSGA-II, proposed by Srinivas and Deb in 2000, is superior to NSGA in several ways: it employs a fast non-dominated sorting algorithm, significantly reducing computational complexity compared to NSGA; it uses crowding degree and crowding degree comparison operators instead of the required shared radius `shareQ`, and uses this as the winning criterion in peer comparisons after quicksort, allowing individuals in the quasi-Pareto domain to expand to the entire Pareto domain and be evenly distributed, thus maintaining population diversity; and it introduces an elitist strategy, expanding the sampling space, preventing the loss of the best individuals, and improving the algorithm's computational speed and robustness. Summary of the Invention

[0007] To address the aforementioned issues, this invention provides a method for optimizing the photothermal and electrical performance of vacuum photovoltaic (PVVG) glass based on XGBoost and NSGA-II. This method applies the XGBoost gradient boosting tree model to achieve rapid and accurate optimization of PVVG photothermal and electrical performance, effectively solving the problem of low computational efficiency in traditional simulation methods. Simultaneously, it employs the NSGA-II algorithm, through continuous optimization and global optimization, balancing the conflicts or synergies between objective functions to obtain a series of optimal combinations. This avoids the subjectivity and limitations of manual parameter tuning and provides diverse options for different design scenarios. This method is highly versatile and applicable to different climate zones, building types, and various PVVG structures.

[0008] To achieve this objective, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for optimizing the photothermal and electrical performance of vacuum photovoltaic glass based on XGBoost and NSGA-II, comprising the following steps: S1: Calculate the overall photothermal performance parameters of PVVG, including the heat transfer coefficient U-value, total solar transmittance SHGC, and visible light transmittance Tvis. S2: Define the climate zone, building orientation, basic parameters of the room with PVVG installed, and the overall light and thermal performance parameters of different PVVGs. Input the above information into Energyplus to generate a building performance dataset containing multiple samples. Each sample in the building performance dataset contains 6 input parameters and 4 target energy consumption parameters. The input parameters include U value, SHGC, Tvis, photovoltaic coverage, low emissivity, and window-to-wall ratio. The target energy consumption parameters include artificial lighting energy consumption, cooling energy consumption, heating energy consumption, and photovoltaic power generation. S3: Using the XGBoost machine learning algorithm to access building performance datasets, surrogate models are trained for buildings equipped with PVVG to quickly and accurately predict building cooling, heating, lighting energy consumption, and photovoltaic power generation. S4: Set the design variables and objective function, couple the XGBoost surrogate model and NSGA-II multi-objective optimization algorithm in MATLAB, and obtain the PVVG parameter combination with good photothermal and electromechanical performance after optimization.

[0009] Preferably, in step S1, The formula for calculating the heat transfer coefficient U is:

[0010] Where he is the external convective heat transfer coefficient; hi is the internal convective heat transfer coefficient. is the sum of the thermal resistances of all glass layers; dj and λj are the thickness and thermal conductivity of the glass layers, respectively; hs,i is the thermal conductivity of the hollow layer; hs,v is the thermal conductivity of the vacuum layer.

[0011] Where hgs,i and hrs,i are the convective and radiative conductivities, respectively; λi is the thermal conductivity of the gas; Di is the thickness of the gas layer; Nui is the Nusselt number, which is taken as 1 when Nui < 1; σ is the Stefan-Boltzmann constant; ε1 and ε2 are the emissivity of the glass on both sides of the gas layer, respectively; and Tm,i is the average temperature of the inner surface of the glass on both sides of the gas layer.

[0012] Where α is the gas overall adaptability coefficient; γ is the gas heat capacity; R is the molar gas constant; P is the residual gas pressure; M is the molar mass of the gas; and r, a, h, and λpillar are the support column radius, spacing, height, and thermal conductivity, respectively.

[0013] Preferably, in step S1, The formula for calculating the total solar transmittance (SHGC) is:

[0014] Where Tsol is the total solar transmittance of the glass system, qin is the net heat flux entering the room under solar radiation conditions, qin(Is=o) is the net heat flux entering the room under no solar radiation conditions, Is is the solar incident irradiance acting on the outer surface of the glass; i is the i-th layer of glass from the outside to the inside; ki is the thermal conductivity of the glass; di is the glass thickness; Ti,o and Ti,i are the outer and inner surface temperatures of the glass, respectively; Ji,o and Ji,i are the outer and inner surface radii, respectively; J, ε, ρ, and τ are the long-wave radii, emissivity, reflectivity, and transmittance, respectively; σ is the Stefan-Boltzmann constant; T is the surface temperature; Jin and Jout are the indoor and outdoor environmental long-wave radii, respectively. The formula for calculating visible light transmittance (Tvis) is:

[0015] Where λwj / j+1 represents the wavelength interval λwj, λwj+1; τsl is the average spectral transmittance; Evs is the average value of the relative spectral distribution; R is the average value of the spectral luminous efficiency function of human visual perception; Δλwj is the wavelength interval; and Nvs is the number of wavelength sampling points in the visible light band.

[0016] Preferably, in step S2, The formula for artificial lighting energy consumption is as follows:

[0017] El is the energy consumption for artificial lighting; fl(t) is the artificial lighting demand factor; Pdes is the power density for zone lighting design; Eset is the design illuminance setpoint; Eday(t) is the hourly indoor natural illuminance. The formulas for cooling energy consumption and heating energy consumption are as follows:

[0018] Qsurf is the convective heat transfer on the surfaces of each building envelope; Qair is the heat exchange between air nodes within the area; Qpeople, Qequip, and Qlight are the sensible heat of indoor personnel, equipment, and lighting, respectively; Qinf and Qvent are the sensible heat of infiltration air and mechanical fresh air, respectively; Eheat and Ecool are the heating and cooling energy consumption, respectively. The formula for photovoltaic power generation is as follows:

[0019] Egen is the PVVG power generation; Pgen is the PVVG AC output power; Asurf is the effective light-receiving area of ​​the PVVG; G is the total solar irradiance received by the PVVG surface; η0 is the photoelectric conversion efficiency of the photovoltaic cell under standard test conditions; ηinvert is the inverter efficiency; β0 is the temperature coefficient of maximum power of the module; T PV is the actual operating temperature of the module, which is replaced by PVVG, the outdoor temperature.

[0020] Preferably, step S3 includes the following steps: S31. Input Data Preparation The dataset is randomly shuffled and divided into training and testing sets. The input parameters and target energy consumption parameters in the dataset are linearly normalized to map all data into the 0-1 range. S32, XGBoost model parameter settings Set the XGBoost training parameters, including maximum tree depth, shrinkage step size, minimum sample weight, subsample ratio, feature sampling ratio, number of parallel trees, objective function type, and boosting method; S33, Training Set Modeling The training set is input into the XGBoost package, the residuals are calculated through multiple rounds of iteration and a regression tree structure is generated, the gain is calculated based on the second-order Taylor expansion, and the optimal split point is determined by a greedy algorithm and a pre-sorting algorithm to establish an XGBoost prediction model for a single target energy consumption parameter. S34, Test Set Prediction The test set is input into the trained XGBoost model, and the input samples are traversed layer by layer according to the generated regression tree structure. The outputs of each tree are accumulated to obtain the prediction results of the target energy consumption parameters. Finally, inverse normalization is performed. S35, Model File Generation The MATLAB environment calls the XGBoost state link library to generate binary model files and MAT files. The binary model file saves the regression tree structure and parameter information of the model, while the MAT file saves the input and output normalization parameters and model path configuration. S36. Take steps S32 to S35 as a training process, train the XGBoost model for each of the four target energy consumption parameters. After completing the training of one model, repeat steps S32 to S35 to complete the training of the remaining three target energy consumption parameters in sequence, and obtain the binary model file and MAT file of the XGBoost building energy consumption and photovoltaic power generation prediction model.

[0021] Preferably, between steps S34 and S35, the following is also included: error evaluation, calculating the performance indicators of the model on the training set and the test set, including RMSE, R², MAE, MBE, MRE and cv-RMSE, to form the error evaluation result of the objective function model; if the evaluation result exceeds the acceptable range, then repeat steps S32 to S35 to conduct a new round of model training and evaluation until the result meets the requirements.

[0022] Preferably, step S4 includes the following steps: S41. Initialize the population: Within the range of design variables, MATLAB assigns a set of building parameter combinations to each individual based on the set running parameters, and obtains multi-objective function values ​​by calling the XGBoost prediction model. S42. Perform non-dominated quicksort and crowding calculation within the population. S43. Selection, Crossover, and Mutation Individuals are selected to enter the mating pool using a tournament selection mechanism. Offspring individuals are generated by simulating binary crossover and polynomial mutation. After setting the range of values ​​for the offspring decision variables, the XGBoost prediction model is called again to calculate the multi-objective function value and form the offspring population. S44, Elite Preservation and Population Renewal The parent and offspring generations are merged, and non-dominated quicksort and crowding calculation are performed again. All individuals are arranged in ascending order according to their rank, and individuals of the same rank are arranged in descending order according to their crowding distance. Individuals with higher rankings are selected to form a new population, and the population size is the same as the originally set population size. S45. Iteration and Result Output Repeat steps S43 to S44 until the maximum number of iterations is reached, at which point the calculation ends, yielding the Pareto optimal solution set. The trade-offs between the objectives are then visualized using a 3D scatter plot.

[0023] Preferably, a method for optimizing the photothermal and electrical performance of vacuum photovoltaic glass based on XGBoost and NSGA-II includes the following steps: Step 1: Calculate the overall photothermal performance parameters of PVVG in Windows. The overall photothermal performance parameters of PVVG include the heat transfer coefficient (U-value), total solar transmittance (SHGC), and visible light transmittance (Tvis).

[0024] The overall photothermal performance parameters of different PVVGs were calculated using Windows. A PVVG consists of a PVG and a VG with a low-emissivity coating. Basic photothermal performance parameters of PVGs with different photovoltaic coverage rates, including thermal conductivity, forward and reverse solar transmittance, forward and reverse solar reflectance, forward and reverse visible light transmittance, and forward and reverse visible light reflectance, were imported into Windows to establish a PVG model. Low-emissivity coated glass was selected from Windows' glass database (LBNL Optical Properties Library) and combined with the PVG model to construct a PVVG model. The overall photothermal performance parameters of PVVGs with different photovoltaic coverage rates were obtained using Windows' system calculation module.

[0025] 1. Calculation of U value: PVVG has a complex structure. In Windows software, according to ISO / EN 10077, the U-value of the glass system is calculated by combining the layered superposition of thermal conductivity and the coupling of boundary heat transfer coefficients, as shown in the following formula:

[0026] Where, he is the external convective heat transfer coefficient, taken as 23 W / (m²·K); hi is the internal convective heat transfer coefficient, taken as 8 W / (m²·K); λj is the sum of the thermal resistances of all glass layers; dj and λj are the thickness (m) and thermal conductivity (W / (m·K)) of the glass layers, respectively. hs,i is the thermal conductivity of the hollow layer, which mainly consists of radiative heat transfer between the glass surfaces on both sides of the cavity, heat conduction of the gas, and convection; hs,v is the thermal conductivity of the vacuum layer, which mainly consists of radiative heat transfer between the glass surfaces on both sides of the vacuum layer, heat conduction of the support columns, and heat conduction of the residual gas, neglecting gas convection.

[0027]

[0028] Where hgs,i and hrs,i are the convective and radiative conductivities, respectively; λi is the thermal conductivity of the gas (W / (m·K)); Di is the thickness of the gas layer (m); Nui is the Nusselt number, which is taken as Nui = 1 when Nui < 1, indicating no significant natural convection and only conduction. σ is the Stefan-Boltzmann constant, taken as 5.6697 × 10⁻⁸ W / (m²·K⁴); ε₁ and ε₂ are the emissivity of the glass on both sides of the gas layer, respectively; and Tm,i is the average temperature (K) of the inner surface of the glass on both sides of the gas layer.

[0029]

[0030] Where α is the gas overall adaptability coefficient; γ is the gas heat capacity J / (kg·K); R is the molar gas constant J / (mol·K); P is the residual gas pressure (Pa); M is the molar mass of the gas kg / mol; r, a, h, and λpillar are the support column radius (m), spacing (m), height (m), and thermal conductivity (W / (m·K), respectively.

[0031] 2. SHGC Calculation: The solar heat gain coefficient (SHGC) measures the proportion of solar radiation energy that enters a building through its glass envelope. It is a key indicator in building energy efficiency analysis and solar thermal performance assessment. The WINDOW software calculates the SHGC according to the ISO 15099 standard. Essentially, it is the ratio of net heat entering the building under solar radiation conditions to the incident solar radiation, i.e., the sum of direct solar transmission and secondary heat transfer. The solution process is as follows:

[0032] Where Tsol is the total solar transmittance of the glass system, qin is the net heat flux entering the room under solar radiation conditions (W / m²), qin(Is=o) is the net heat flux entering the room under no solar radiation conditions (W / m²), and Is is the solar incident irradiance acting on the outer surface of the glass (W / m²).

[0033] i represents the i-th layer of glass (from outside to inside); ki is the thermal conductivity of the glass (W / (m·K)); di is the glass thickness (m); Ti,o and Ti,i are the surface temperatures of the outer and inner sides of the glass, respectively; Ji,o and Ji,i are the surface emissivity of the outer and inner sides, respectively. J, ε, ρ, and τ are the long-wave emissivity, emissivity, reflectivity, and transmittance, respectively; σ is the Stefan-Boltzmann constant; T is the surface temperature; Jin and Jout are the long-wave emissivity of the indoor and outdoor environments, respectively.

[0034] 3. Tvis calculation: Visible light transmittance is obtained by weighted integral of the spectral transmittance of the window system over the entire visible spectrum using the human visual function and the spectral distribution of a standard light source, as shown in the following formula:

[0035] λwj / j+1 represents the wavelength interval λwj, λwj+1; τsl is the average spectral transmittance; Evs is the average value of the relative spectral distribution; R is the average value of the spectral luminous efficiency function of human photopic vision; Δλwj is the wavelength interval; Nvs is the number of wavelength sampling points in the visible light band.

[0036] Step 2: Define the climate zone, building orientation, basic parameters for rooms with PVVG installation, and overall light and thermal performance parameters for different PVVG types. Input this information into Energyplus to generate a building performance dataset containing multiple samples.

[0037] The appropriate EPW weather file is selected based on the climate zone. Basic room parameters include room length, width, and height; the geometric center of the windows; window width; and structural parameters of the building envelope other than PVVG. Each sample in the building performance dataset contains six input parameters and four target energy consumption parameters. The input parameters include U-value, SHGC, Tvis, photovoltaic coverage, low emissivity, and window-to-wall ratio. The target energy consumption parameters include annual lighting energy consumption, cooling energy consumption, heating energy consumption, and photovoltaic power generation.

[0038] Import the basic parameters of the room into Energyplus to establish a PVVG building energy consumption simulation model. MATLAB then creates a temporary IDF file based on the PVVG building energy consumption simulation model, automatically iterates through each sample, imports the 6 input parameters of each sample into the temporary IDF file, and calculates the 4 target energy consumption parameters corresponding to each sample.

[0039] 1. Heating and cooling energy consumption calculation: In numerical simulations of building thermal environments, the Ideal LoadsAir System provided by EnergyPlus is a simplified model representing the theoretical loads of indoor heating and cooling. This system assumes that the air handling equipment has unlimited capacity and can provide precise sensible heat or cooling in real time to maintain the zone's air temperature at the set value. Therefore, EnergyPlus obtains the instantaneous sensible heat power required to maintain the set temperature by solving the zone's heat balance equations, and then integrates this over time to ultimately generate the sensible heat heating and cooling energy, neglecting latent heat loads or equipment efficiency losses. The formula is as follows:

[0040] Qsurf is the convective heat transfer (W) of each building envelope surface; Qair is the heat exchange between air nodes within a zone (W); Qpeople, Qequip, and Qlight are the sensible heat (W) of indoor occupants, equipment, and lighting, respectively; Qinf and Qvent are the sensible heat (W) of infiltration air and mechanical fresh air, respectively; Eheat and Ecool are... These are the accumulated sensible heat for heating and cooling (J).

[0041] 2. Calculation of energy consumption for artificial lighting: EnergyPlus employs the DElight daylight-artificial lighting coupling algorithm, which means that when sunlight is insufficient, the artificial lighting system is activated to supplement the light. The calculation formula is as follows:

[0042] El is the energy consumption for artificial lighting (J); fl(t) is the artificial lighting demand coefficient; Pdes is the power density of the area lighting design (W); Eset is the design illuminance setpoint (lx); Eday(t) is the hourly indoor natural daylight illuminance (lx).

[0043] 3. Calculation of photovoltaic power generation: The photovoltaic power generation calculation uses a simple model in Energyplus, which compares the solar radiation incident on the PVVG with the glass. The formula for calculating total power generation using real-time temperature correction of the outer surface is as follows:

[0044] Egen is the PVVG power generation (J); Pgen is the PVVG AC output power (W); Asurf is the effective light-receiving area of ​​the PVVG (m²); G is the total solar irradiance received by the PVVG surface (W·m⁻²); η0 is the photoelectric conversion efficiency of the photovoltaic cell under standard test conditions (1000 W·m⁻², 25 ℃, AM 1.5); ηinvert is the inverter efficiency; β0 is the module's maximum power temperature coefficient (°C⁻¹); TPV is the module's actual operating temperature (°C), replaced by the outdoor temperature of the PVVG.

[0045] Step 3: In MATLAB, use the XGBoost machine learning algorithm to call the building performance dataset and train surrogate models for buildings equipped with PVVG to quickly and accurately predict building cooling, heating, lighting energy consumption and photovoltaic power generation.

[0046] The specific steps are as follows: Step 3.1: Input Data Preparation: After randomly shuffling the dataset, it is divided into a training set and a test set. The input parameters and target energy consumption parameters in the dataset are linearly normalized respectively, and all data are mapped to the interval [0,1]. Step 3.2: XGBoost model parameter settings: Set the XGBoost training parameters, including maximum tree depth, shrinkage step size, minimum sample weight, subsample ratio, feature sampling ratio, number of parallel trees, objective function type, and boosting method; Step 3.3: Training set modeling: The training set is input into the XGBoost package, the residuals are calculated through multiple rounds of iteration and a regression tree structure is generated, the gain is calculated based on the second-order Taylor expansion, and the optimal split point is determined by a greedy algorithm and a pre-sorting algorithm to establish an XGBoost prediction model for a single target energy consumption parameter. Step 3.4: Test set prediction: The test set is input into the trained XGBoost model, and the input samples are traversed layer by layer according to the generated regression tree structure. The outputs of each tree are accumulated to obtain the prediction results of the target energy consumption parameters. Finally, inverse normalization is performed. Step 3.5: Error Evaluation: The performance metrics of the model on the training and test sets are calculated, including RMSE, R², MAE, MBE, MRE, and cv-RMSE, to form the error evaluation results of the objective function model. If the evaluation results are outside the acceptable range, steps 3.2 to 3.5 are repeated for a new round of model training and evaluation until the results meet the requirements.

[0047] Step 3.6: Model file generation: Once the model accuracy meets the requirements, the MATLAB environment calls the XGBoost state link library to generate a binary model file (.bin) and a MAT file (.mat). The binary model file stores the regression tree structure and parameter information of the model, while the MAT file stores the input and output normalized parameters and model path configuration. Step 3.7: Treat steps 3.2 to 3.6 as a training process, and train the XGBoost model for each of the four target energy consumption parameters. After completing the training of one model, repeat steps 3.2 to 3.6 to complete the training of the remaining three target energy consumption parameters in sequence, and obtain the binary model file and MAT file of the XGBoost building energy consumption and photovoltaic power generation prediction model.

[0048] Step 4: Before starting multi-objective optimization, set the design variables and objective function. Couple the XGBoost surrogate model and the NSGA-II multi-objective optimization algorithm in MATLAB. After optimization, a PVVG parameter combination with good photothermal and electrostatic performance is obtained. The design variables are U-value, SHGC, Tvis, photovoltaic coverage, low emissivity, and window-to-wall ratio. The objective function is the annual cooling energy consumption, heating energy consumption, lighting energy consumption, and photovoltaic power generation of the PVVG-equipped building. The running parameters of the optimization algorithm include population size, number of iterations, mutation index, crossover index, crossover probability, and tournament size. Call the binary model file and MAT file of the XGBoost building energy consumption and photovoltaic power generation prediction model in the MATLAB environment to perform multi-objective optimization.

[0049] The specific optimization process includes the following steps: Step 4.1: Initialize the population: Within the range of design variables, MATLAB assigns a set of building parameter combinations to each individual based on the set running parameters, and obtains multi-objective function values ​​by calling the XGBoost prediction model. Step 4.2: Perform non-dominated quicksort and crowding calculation within the population.

[0050] Step 4.3: Selection, Crossover, and Mutation: Individuals are selected for the mating pool using a tournament selection mechanism. Offspring individuals are generated through simulated binary crossover and polynomial mutation. After setting the range of values ​​for the offspring decision variables, the XGBoost prediction model is called again to calculate the multi-objective function values, forming the offspring population.

[0051] Step 4.4: Elite Preservation and Population Renewal: The parent and offspring generations are merged, and a new non-dominated quicksort and crowding calculation are performed. All individuals are sorted in ascending order of rank, and individuals of the same rank are sorted in descending order of crowding distance. Individuals with higher rankings are selected to form a new population, with the same population size as the initially set population size.

[0052] Step 4.5: Iteration and Result Output: Repeat steps 4.3 to 4.4 until the maximum number of iterations is reached, at which point the calculation ends. The Pareto optimal solution set is obtained, and the trade-offs between the objectives are visualized using a 3D scatter plot.

[0053] Compared with the prior art, the beneficial effects and significant progress of applying the technical solution of the present invention are as follows: 1. The method of the present invention can quickly and accurately predict the annual energy consumption and photovoltaic power generation of buildings equipped with PVVG, and the calculation efficiency is greatly improved; 2. The NSGA-II algorithm used in this invention is adept at handling the conflict between heating, cooling, and lighting energy consumption and photovoltaic power generation. It can obtain a series of optimal trade-off combinations between different objective functions, providing a variety of optimization schemes, which can facilitate people to make flexible choices according to their actual requirements. At the same time, the algorithm ensures the comprehensiveness of the solution set through continuous optimization and global optimization. 3. This invention has strong scalability and versatility. It is not only applicable to different building types in different climate zones, but can also simulate PVVG with various different structures. Attached Figure Description

[0054] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments of the present invention will be briefly described below.

[0055] Figure 1 This is a PVVG construction diagram from Example 2; Figure 2 It is the Energyplus building model in Example 3; Figure 3 This is the prediction result of the target energy consumption parameter in the model testing phase of Example 4; Figure 4 This is a flowchart of NSGA-II calling the XGBoost prediction model to optimize PVVG photothermal performance in Example 5; Figure 5 This is a three-dimensional scatter plot of the Pareto front obtained from the multi-objective optimization in Example 5; The attached diagram shows the following labels: 1. Glass layer; 2. Low-emissivity coating; 3. Support column; 4. Cadmium telluride thin-film photovoltaic cell; 5. Adhesive material; 6. Vacuum layer. Detailed Implementation

[0056] The present invention will be further illustrated below with reference to specific embodiments. These embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading this invention, those skilled in the art can make various alterations and modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0057] The terms "this embodiment," "an embodiment of the present invention," "as shown," "further," and "further improved technical solutions," etc., indicate that the specific features, structures, materials, or characteristics described in the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms are not necessarily directed at the same embodiment or example, and the specific features, structures, materials, or characteristics described can be combined or combined in any suitable manner in one or more embodiments or examples. Furthermore, without causing contradiction, those skilled in the art can combine or combine the different embodiments or examples described in this specification and the features of the different embodiments or examples. Example 1

[0058] This embodiment of a method for optimizing the photothermal and electrical performance of vacuum photovoltaic glass based on XGBoost and NSGA-II includes the following steps: S1: Calculate the overall photothermal performance parameters of PVVG, including the heat transfer coefficient U-value, total solar transmittance SHGC, and visible light transmittance Tvis. S2: Define the climate zone, building orientation, basic parameters of the room with PVVG installed, and the overall light and thermal performance parameters of different PVVGs. Input the above information into Energyplus to generate a building performance dataset containing multiple samples. Each sample in the building performance dataset contains 6 input parameters and 4 target energy consumption parameters. The input parameters include U value, SHGC, Tvis, photovoltaic coverage, low emissivity, and window-to-wall ratio. The target energy consumption parameters include artificial lighting energy consumption, cooling energy consumption, heating energy consumption, and photovoltaic power generation. S3: Using the XGBoost machine learning algorithm to access building performance datasets, surrogate models are trained for buildings equipped with PVVG to quickly and accurately predict building cooling, heating, lighting energy consumption, and photovoltaic power generation. S4: Set the design variables and objective function, couple the XGBoost surrogate model and NSGA-II multi-objective optimization algorithm in MATLAB, and obtain the PVVG parameter combination with good photothermal and electromechanical performance after optimization.

[0059] In this embodiment, in step S1, The formula for calculating the heat transfer coefficient U is:

[0060] Where he is the external convective heat transfer coefficient; hi is the internal convective heat transfer coefficient. is the sum of the thermal resistances of all glass layers; dj and λj are the thickness and thermal conductivity of the glass layers, respectively; hs,i is the thermal conductivity of the hollow layer; hs,v is the thermal conductivity of the vacuum layer.

[0061] Where hgs,i and hrs,i are the convective and radiative conductivities, respectively; λi is the thermal conductivity of the gas; Di is the thickness of the gas layer; Nui is the Nusselt number, which is taken as 1 when Nui < 1; σ is the Stefan-Boltzmann constant; ε1 and ε2 are the emissivity of the glass on both sides of the gas layer, respectively; and Tm,i is the average temperature of the inner surface of the glass on both sides of the gas layer.

[0062] Where α is the gas overall adaptability coefficient; γ is the gas heat capacity; R is the molar gas constant; P is the residual gas pressure; M is the molar mass of the gas; and r, a, h, and λpillar are the support column radius, spacing, height, and thermal conductivity, respectively.

[0063] In this embodiment, in step S1, The formula for calculating the total solar transmittance (SHGC) is:

[0064] Where Tsol is the total solar transmittance of the glass system, qin is the net heat flux entering the room under solar radiation conditions, qin(Is=o) is the net heat flux entering the room under no solar radiation conditions, Is is the solar incident irradiance acting on the outer surface of the glass; i is the i-th layer of glass from the outside to the inside; ki is the thermal conductivity of the glass; di is the glass thickness; Ti,o and Ti,i are the outer and inner surface temperatures of the glass, respectively; Ji,o and Ji,i are the outer and inner surface radii, respectively; J, ε, ρ, and τ are the long-wave radii, emissivity, reflectivity, and transmittance, respectively; σ is the Stefan-Boltzmann constant; T is the surface temperature; Jin and Jout are the indoor and outdoor environmental long-wave radii, respectively. The formula for calculating visible light transmittance (Tvis) is:

[0065] Where λwj / j+1 represents the wavelength interval λwj, λwj+1; τsl is the average spectral transmittance; Evs is the average value of the relative spectral distribution; R is the average value of the spectral luminous efficiency function of human visual perception; Δλwj is the wavelength interval; and Nvs is the number of wavelength sampling points in the visible light band.

[0066] In this embodiment, in step S2, The formula for artificial lighting energy consumption is as follows:

[0067] El is the energy consumption for artificial lighting; fl(t) is the artificial lighting demand factor; Pdes is the power density for zone lighting design; Eset is the design illuminance setpoint; Eday(t) is the hourly indoor natural illuminance. The formulas for cooling energy consumption and heating energy consumption are as follows:

[0068] Qsurf is the convective heat transfer on the surfaces of each building envelope; Qair is the heat exchange between air nodes within the area; Qpeople, Qequip, and Qlight are the sensible heat of indoor personnel, equipment, and lighting, respectively; Qinf and Qvent are the sensible heat of infiltration air and mechanical fresh air, respectively; Eheat and Ecool are the heating and cooling energy consumption, respectively. The formula for photovoltaic power generation is as follows:

[0069] Egen is the PVVG power generation; Pgen is the PVVG AC output power; Asurf is the effective light-receiving area of ​​the PVVG; G is the total solar irradiance received by the PVVG surface; η0 is the photoelectric conversion efficiency of the photovoltaic cell under standard test conditions; ηinvert is the inverter efficiency; β0 is the temperature coefficient of maximum power of the module; T PV is the actual operating temperature of the module, which is replaced by PVVG, the outdoor temperature.

[0070] In this embodiment, step S3 includes the following steps: S31. Input Data Preparation The dataset is randomly shuffled and divided into training and testing sets. The input parameters and target energy consumption parameters in the dataset are linearly normalized to map all data into the 0-1 range. S32, XGBoost model parameter settings Set the XGBoost training parameters, including maximum tree depth, shrinkage step size, minimum sample weight, subsample ratio, feature sampling ratio, number of parallel trees, objective function type, and boosting method; S33, Training Set Modeling The training set is input into the XGBoost package, the residuals are calculated through multiple rounds of iteration and a regression tree structure is generated, the gain is calculated based on the second-order Taylor expansion, and the optimal split point is determined by a greedy algorithm and a pre-sorting algorithm to establish an XGBoost prediction model for a single target energy consumption parameter. S34, Test Set Prediction The test set is input into the trained XGBoost model, and the input samples are traversed layer by layer according to the generated regression tree structure. The outputs of each tree are accumulated to obtain the prediction results of the target energy consumption parameters. Finally, inverse normalization is performed. S35, Model File Generation The MATLAB environment calls the XGBoost state link library to generate binary model files and MAT files. The binary model file saves the regression tree structure and parameter information of the model, while the MAT file saves the input and output normalization parameters and model path configuration. S36. Take steps S32 to S35 as a training process, train the XGBoost model for each of the four target energy consumption parameters. After completing the training of one model, repeat steps S32 to S35 to complete the training of the remaining three target energy consumption parameters in sequence, and obtain the binary model file and MAT file of the XGBoost building energy consumption and photovoltaic power generation prediction model.

[0071] In this embodiment, between steps S34 and S35, the following is also included: error evaluation, calculating the performance indicators of the model on the training set and the test set, including RMSE, R², MAE, MBE, MRE and cv-RMSE, to form the error evaluation result of the objective function model; if the evaluation result exceeds the acceptable range, then steps S32 to S35 are repeated to conduct a new round of model training and evaluation until the result meets the requirements.

[0072] In this embodiment, step S4 includes the following steps: S41. Initialize the population: Within the range of design variables, MATLAB assigns a set of building parameter combinations to each individual based on the set running parameters, and obtains multi-objective function values ​​by calling the XGBoost prediction model. S42. Perform non-dominated quicksort and crowding calculation within the population. S43. Selection, Crossover, and Mutation Individuals are selected to enter the mating pool using a tournament selection mechanism. Offspring individuals are generated by simulating binary crossover and polynomial mutation. After setting the range of values ​​for the offspring decision variables, the XGBoost prediction model is called again to calculate the multi-objective function value and form the offspring population. S44, Elite Preservation and Population Renewal The parent and offspring generations are merged, and non-dominated quicksort and crowding calculation are performed again. All individuals are arranged in ascending order according to their rank, and individuals of the same rank are arranged in descending order according to their crowding distance. Individuals with higher rankings are selected to form a new population, and the population size is the same as the originally set population size. S45. Iteration and Result Output Repeat steps S43 to S44 until the maximum number of iterations is reached, at which point the calculation ends, yielding the Pareto optimal solution set. The trade-offs between the objectives are then visualized using a 3D scatter plot.

[0073] Example 2: Calculation of overall photothermal performance parameters of PVVG The PVVG structure used in this embodiment is shown in Figure 1. In the figure, 1 is the glass layer, 2 is the low-emissivity coating, 3 is the support column, 4 is the cadmium telluride thin-film photovoltaic cell, 5 is the adhesive material, and 6 is the vacuum layer.

[0074] The overall photothermal performance parameters of PVVG include heat transfer coefficient (U value), total solar transmittance (SHGC), and visible light transmittance (Tvis).

[0075] The overall photothermal performance parameters of different PVVGs were calculated using Windows. PVVG consists of PVG and VG with a low-emissivity coating.

[0076] The specific calculation steps are as follows: 2.1 Import the basic photothermal performance parameters of PVG with different photovoltaic coverage (as shown in Table 1) into Windows, including thermal conductivity, solar forward and reverse transmittance, solar forward and reverse reflectance, visible light forward and reverse transmittance, and visible light forward and reverse reflectance, and establish a PVG model.

[0077] Table 1 Photovoltaic coverage Conductivity Tsol1 Tsol2 Rsol1 Rsol2 Tvis1 Tvis2 Rvis1 Rvis2 90% 0.98 0.06 0.06 0.079 0.159 0.07 0.07 0.063 0.17 80% 0.98 0.102 0.102 0.082 0.154 0.123 0.123 0.066 0.164 70% 0.98 0.145 0.145 0.085 0.149 0.177 0.177 0.069 0.158 60% 0.98 0.21 0.21 0.086 0.139 0.252 0.252 0.071 0.149 50% 0.98 0.275 0.275 0.088 0.128 0.327 0.327 0.074 0.14 40% 0.98 0.34 0.34 0.09 0.117 0.402 0.402 0.077 0.131 In the table, Conductivity is the thermal conductivity, Tsol1 and Tsol2 are the forward and reverse transmittance of solar energy, Rsol1 and Rsol2 are the forward and reverse reflectance of solar energy, Tvis1 and Tvis2 are the forward and reverse transmittance of visible light, and Rvis1 and Rvis2 are the forward and reverse reflectance of visible light.

[0078] 2.2 Select low-emissivity coated glass from the LBNL Optical Properties Library in Windows and combine it with PVG to construct a PVVG model.

[0079] 2.3. Using the system calculation module of the WINDOW software, the overall photothermal performance parameters of PVVG with different photovoltaic coverage rates were obtained (as shown in Table 2). Table 2 U-value SHGC Tvis Photovoltaic coverage % Low emissivity 0.549 0.060 0.055 90 0.01 0.549 0.085 0.097 80 0.01 0.549 0.111 0.139 70 0.01 0.549 0.147 0.198 60 0.01 0.549 0.183 0.257 50 0.01 0.549 0.219 0.315 40 0.01 0.641 0.072 0.057 90 0.04 0.641 0.101 0.101 80 0.04 0.641 0.130 0.145 70 0.04 0.641 0.172 0.206 60 0.04 0.641 0.214 0.267 50 0.04 0.641 0.256 0.328 40 0.04 0.738 0.089 0.061 90 0.07 0.738 0.125 0.107 80 0.07 0.738 0.161 0.154 70 0.07 0.738 0.215 0.219 60 0.07 0.738 0.269 0.283 50 0.07 0.738 0.322 0.348 40 0.07 0.833 0.094 0.060 90 0.10 0.833 0.129 0.106 80 0.10 0.833 0.164 0.152 70 0.10 0.833 0.216 0.216 60 0.10 0.833 0.268 0.280 50 0.10 0.833 0.320 0.345 40 0.10 1.121 0.115 0.062 90 0.20 1.121 0.151 0.108 80 0.20 1.121 0.188 0.155 70 0.20 1.121 0.243 0.221 60 0.20 1.121 0.298 0.287 50 0.20 1.121 0.352 0.352 40 0.20

[0080] PVVG has a complex structure. In Windows software, according to ISO / EN 10077, the U-value of the glass system is calculated by combining the layered superposition of thermal conductivity and the coupling of boundary heat transfer coefficients, as shown in the following formula:

[0081] Where, he is the external convective heat transfer coefficient, taken as 23 W / (m²·K); hi is the internal convective heat transfer coefficient, taken as 8 W / (m²·K); j=1Mdjλj is the sum of the thermal resistances of all glass layers; dj and λj are the thickness (m) and thermal conductivity (W / (m·K)) of the glass layers, respectively. hs,i is the thermal conductivity of the hollow layer, and the heat transfer in the hollow layer mainly consists of radiative heat transfer between the glass surfaces on both sides of the cavity, heat conduction of the gas, and convection; hs,v is the thermal conductivity of the vacuum layer, and the heat transfer in the vacuum layer mainly consists of radiative heat transfer between the glass surfaces on both sides of the vacuum layer, heat conduction of the support columns, and heat conduction of the residual gas, neglecting gas convection.

[0082]

[0083] Where hgs,i and hrs,i are the convective and radiative conductivities, respectively; λi is the thermal conductivity of the gas (W / (m·K)); Di is the thickness of the gas layer (m); Nui is the Nusselt number, which is taken as Nui = 1 when Nui < 1, indicating no significant natural convection and only conduction. σ is the Stefan-Boltzmann constant, taken as 5.6697 × 10⁻⁸ W / (m²·K⁴); ε₁ and ε₂ are the emissivity of the glass on both sides of the gas layer, respectively; and Tm,i is the average temperature (K) of the inner surface of the glass on both sides of the gas layer.

[0084]

[0085] Where α is the gas overall adaptability coefficient; γ is the gas heat capacity J / (kg·K); R is the molar gas constant J / (mol·K); P is the residual gas pressure (Pa); M is the molar mass of the gas kg / mol; r, a, h, and λpillar are the support column radius (m), spacing (m), height (m), and thermal conductivity (W / (m·K), respectively.

[0086] SHGC calculation The solar heat gain coefficient (SHGC) measures the proportion of solar radiation energy that enters a building through its glass envelope. It is a key indicator in building energy efficiency analysis and solar thermal performance assessment. The WINDOW software calculates the SHGC according to the ISO 15099 standard. Essentially, it is the ratio of net heat entering the building under solar radiation conditions to the incident solar radiation, i.e., the sum of direct solar transmission and secondary heat transfer. The solution process is as follows:

[0087] Where Tsol is the total solar transmittance of the glass system, qin is the net heat flux entering the room under solar radiation conditions (W / m²), qin(Is=o) is the net heat flux entering the room under no solar radiation conditions (W / m²), and Is is the solar incident irradiance acting on the outer surface of the glass (W / m²). i is the i-th glass layer (from outside to inside); ki is the thermal conductivity of the glass (W / (m·K)); di is the glass thickness (m); Ti,o and Ti,i are the outer and inner surface temperatures of the glass, respectively; Ji,o and Ji,i are the outer and inner surface radii, respectively. J, ε, ρ, and τ are the long-wave radii, emissivity, reflectivity, and transmittance, respectively; σ is the Stefan-Boltzmann constant; T is the surface temperature; Jin and Jout are the indoor and outdoor environmental long-wave radii, respectively.

[0088] Tvis calculation Visible light transmittance is obtained by weighted integral of the spectral transmittance of the window system over the entire visible spectrum using the human visual function and the spectral distribution of a standard light source, as shown in the following formula:

[0089] λwj / j+1 represents the wavelength interval λwj, λwj+1; τsl is the average spectral transmittance; Evs is the average value of the relative spectral distribution; R is the average value of the spectral luminous efficiency function of human photopic vision; Δλwj is the wavelength interval; Nvs is the number of wavelength sampling points in the visible light band.

[0090] 2.4 To ensure a sufficient, continuous, and uniformly distributed number of glass model samples, linear interpolation was used to augment the data based on the original discrete photovoltaic coverage rates. With a tolerance of 2%, intermediate values ​​were inserted between adjacent groups. For example, for 90% to 80%, values ​​of 88%, 86%, 84%, and 82% were inserted. Simultaneously, the corresponding U-values, SHGC, and Tvis were linearly interpolated according to the proportion of photovoltaic coverage change to obtain smooth and sufficient sample data. Taking a sample with a photovoltaic coverage rate of 90% to 80% and a low emissivity of 0.01 as an example, the overall photothermal performance parameters of the PVVG after data augmentation are shown in Table 3.

[0091] Table 3 U-value SHGC Tvis Photovoltaic coverage Low emissivity 0.549 0.060 0.055 90% 0.01 0.549 0.065 0.063 88% 0.01 0.549 0.070 0.072 86% 0.01 0.549 0.075 0.080 84% 0.01 0.549 0.080 0.088 82% 0.01 0.549 0.085 0.097 80% 0.01

[0092] This embodiment defines the basic parameters of climate zones, building orientation, PVVG room installation, and the overall light and thermal performance parameters of different PVVGs. This information is then input into Energyplus to generate a building performance dataset containing multiple samples.

[0093] The appropriate EPW weather file is selected based on the climate zone. Basic room parameters include room length, width, and height; the geometric center of the windows; window width; and structural parameters of the building envelope other than PVVG. Each sample in the building performance dataset contains six input parameters and four target energy consumption parameters. The input parameters include U-value, SHGC, Tvis, photovoltaic coverage, low emissivity, and window-to-wall ratio. The target energy consumption parameters include annual lighting energy consumption, cooling energy consumption, heating energy consumption, and photovoltaic power generation.

[0094] Import the basic parameters of the room into Energyplus to establish a PVVG building energy consumption simulation model. MATLAB creates a temporary IDF file based on the simulation model and automatically iterates through each sample, importing the 6 input parameters of each sample into the temporary IDF file to calculate the 4 target energy consumption parameters (lighting energy consumption, cooling energy consumption, heating energy consumption, and photovoltaic power generation) corresponding to each sample.

[0095] Heating and cooling energy consumption calculation In numerical simulations of building thermal environments, the Ideal LoadsAir System provided by EnergyPlus is a simplified model representing the theoretical loads of indoor heating and cooling. This system assumes that the air handling equipment has unlimited capacity and can provide precise sensible heat or cooling in real time to maintain the zone's air temperature at the set value. Therefore, EnergyPlus obtains the instantaneous sensible heat power Qsys(t) required to maintain the set temperature by solving the zone's heat balance equation, and then integrates it over time to ultimately generate the sensible heat heating and cooling energy, neglecting latent heat loads or equipment efficiency losses. The formula is as follows:

[0096] Qsurf is the convective heat transfer on the surfaces of each building envelope (W); Qair is the heat exchange between air nodes within the area (W); Qpeople, Qequip, and Qlight are the sensible heat of indoor personnel, equipment, and lighting (W); Qinf and Qvent are the sensible heat of infiltration air and mechanical fresh air (W); Eheat and Ecool are the cumulative sensible heat for heating and cooling (J).

[0097] EnergyPlus employs the DElight daylight-artificial lighting coupling algorithm, which means that when sunlight is insufficient, the artificial lighting system is activated to supplement the light. The calculation formula is as follows:

[0098] El is the energy consumption for artificial lighting (J); fl(t) is the artificial lighting demand factor; Pdes is the power density of the area lighting design (W); Eset is the design illuminance setpoint (lx); Eday(t) is the hourly indoor natural daylight illuminance (lx).

[0099] The photovoltaic power generation is calculated using a simple model in Energyplus. The total power generation is calculated by correcting for the solar radiation incident on the PVVG and the real-time temperature of the outer surface of the glass. The formula is as follows:

[0100] Egen is the PVVG power generation (J); Pgen is the PVVG AC output power (W); Asurf is the effective light-receiving area of ​​the PVVG (m²); G is the total solar irradiance received by the PVVG surface (W·m⁻²); η0 is the photoelectric conversion efficiency of the photovoltaic cell under standard test conditions (1000 W·m⁻², 25 ℃, AM 1.5); ηinvert is the inverter efficiency; β0 is the module's maximum power temperature coefficient (°C⁻¹); TPV is the module's actual operating temperature (°C, replaced by the outdoor temperature of the PVVG).

[0101] The building selected for this embodiment is a small office in Shanghai. Its building model in Energyplus is shown in Figure 2. The office is 3 meters long, 3 meters wide and 3 meters high. The PVVG is installed on the south facade. The geometric center of the window is the geometric center of the south facade. The window width is 3 meters.

[0102] Except for PVVG, the structural parameters of the other building envelopes are shown in Tables 4 and 5.

[0103] Table 4

[0104] Table 5

[0105] In this embodiment, the photovoltaic (PV) coverage rate ranges from 10% to 90%, uniformly distributed with a tolerance of 2%; the low emissivity values ​​are 0.01, 0.04, 0.07, 0.1, and 0.2; and the window-to-wall ratio ranges from 0.3 to 0.9, uniformly distributed with a tolerance of 0.05. For PVVG models with PV coverage rates of 40%, 50%, 60%, 70%, 80%, and 90%, the U-values, SHGC, and Tvis are calculated using Windows; the data for other PV coverage rates are obtained through linear interpolation. By arranging and combining the above PV coverage rate, low emissivity, and window-to-wall ratio, 2665 sets of input parameters are obtained. Using MATLAB and calling Energyplus, these 2665 sets of data are automatically traversed to calculate the corresponding target energy consumption parameters, resulting in a complete building performance dataset.

[0106] Example 4 In MATLAB, the XGBoost machine learning algorithm is used to access a building performance dataset. For buildings equipped with PVVG (Photovoltaic Vulcanization), surrogate models are trained to quickly and accurately predict building cooling, heating, lighting energy consumption, and photovoltaic power generation. The specific steps are as follows: 4.1 Input Data Preparation After randomly shuffling the dataset, 2050 samples were selected as the training set and 615 samples as the test set. The input parameters and target energy consumption parameters in the dataset were linearly normalized to map all data to the interval [0,1].

[0107] 4.2 XGBoost Model Parameter Settings Set the XGBoost training parameters, including maximum tree depth, shrinkage step size, minimum sample weight, subsample ratio, feature sampling ratio, number of parallel trees, objective function type, and boosting method.

[0108] In this embodiment, the maximum tree depth is 6, the shrinkage step size is 0.1, the minimum sample weight is 1, the subsample ratio is 0.8, the feature sampling ratio is 0.9, the number of parallel trees is 1, the objective function type is binary classification logistic regression, and the boosting method is gradient boosting tree.

[0109] 4.3 Training Set Modeling The training set is input into the XGBoost package, the residuals are calculated through multiple rounds of iteration and a regression tree structure is generated, the gain is calculated based on the second-order Taylor expansion, and the optimal split point is determined by a greedy algorithm and a pre-sorting algorithm to establish an XGBoost prediction model for a single target energy consumption parameter.

[0110] 4.4 Test Set Prediction The test set is input into the trained XGBoost model, and the input samples are traversed layer by layer according to the generated regression tree structure. The outputs of each tree are accumulated to obtain the prediction results of the target energy consumption parameters. Finally, inverse normalization is performed, as shown in Figure 3.

[0111] 4.5 Error Evaluation The performance metrics of the model on the training and test sets are calculated, including RMSE, R², MAE, MBE, MRE, and cv-RMSE, to form the error evaluation results of the objective function model. If the evaluation results are outside the acceptable range, steps 4.2 to 4.5 are repeated for a new round of model training and evaluation until the results meet the requirements.

[0112] The error evaluation results of each XGBoost prediction model in this embodiment are shown in Table 6.

[0113] Table 6 RMSE R² MAE MBE MRE (%) cv-RMSE(%) Photovoltaic power generation 1.1935 0.9999 0.8789 0.0422 0.55 0.64 Lighting energy consumption 0.2462 0.9998 0.1577 0.0011 0.36 0.62 Heating energy consumption 0.8337 0.9997 0.5295 −0.0425 0.31 0.5 Cooling energy consumption 9.5457 0.995 3.9865 0.589 0.93 1.83 4.6 Model File Generation Once the model accuracy meets the requirements, the MATLAB environment calls the XGBoost state link library to generate a binary model file (.bin) and a MAT file (.mat). The binary model file stores the regression tree structure and parameter information of the model, while the MAT file stores the input and output normalized parameters and model path configuration.

[0114] 4.7. Treat steps 4.2 to 4.6 as a training process, and train the XGBoost model for each of the four target energy consumption parameters. After completing the training of one model, repeat steps 4.2 to 4.6 to complete the training of the remaining three target energy consumption parameters in sequence, and obtain the binary model file and MAT file of the XGBoost building energy consumption and photovoltaic power generation prediction model.

[0115] Example 5 Before starting the multi-objective optimization, design variables and objective functions are defined. The XGBoost surrogate model and the NSGA-II multi-objective optimization algorithm are coupled in MATLAB. After optimization, a PVVG parameter combination with good photothermal and electrostatic performance is obtained. The design variables are U-value, SHGC, Tvis, photovoltaic coverage, low emissivity, and window-to-wall ratio. The objective functions are the annual cooling energy consumption, heating energy consumption, lighting energy consumption, and photovoltaic power generation of buildings equipped with PVVG. The running parameters of the optimization algorithm include population size, number of iterations, mutation index, crossover index, crossover probability, and tournament size. The binary model file and MAT file of the XGBoost building energy consumption and photovoltaic power generation prediction model are called in the MATLAB environment for multi-objective optimization.

[0116] In this embodiment, the population size is 50, the number of iterations is 200, the mutation and crossover indices are both 20, the crossover probability is 0.9, and the tournament size is 2. The specific optimization process includes the following steps: 5.1 Initializing the Population: Under the range of design variables, MATLAB assigns a set of building parameter combinations to each individual based on the set running parameters, and obtains multi-objective function values ​​by calling the XGBoost prediction model.

[0117] The range of values ​​for the design variables in this embodiment is shown in Table 7.

[0118] Table 7

[0119] 5.2 Perform non-dominated quicksort and crowding calculation within the population.

[0120] 5.3 Selection, Crossover, and Mutation: Individuals are selected for the mating pool using a tournament selection mechanism. Offspring individuals are generated through simulated binary crossover and polynomial mutation. After setting the range of values ​​for the offspring decision variables, the XGBoost prediction model is called again to calculate the multi-objective function values, forming the offspring population.

[0121] 5.4 Elite Preservation and Population Update: Parent and offspring generations are merged, and a new non-dominated quicksort and crowding calculation are performed. All individuals are sorted in ascending order of rank, and individuals of the same rank are sorted in descending order of crowding distance. Individuals with higher rankings are selected to form a new population, with the same population size as the initially set population size.

[0122] 5.5 Iteration and Result Output: Repeat steps 5.3 to 5.4 until the maximum number of iterations is reached, at which point the calculation ends. The Pareto optimal solution set is obtained (Table 8), and the trade-offs between the objectives are visualized through a three-dimensional scatter plot (Figure 4).

[0123] The overall process for optimizing the PVVG photothermal and electrical performance based on the NSGA-II and XGBoost prediction models in this embodiment is shown in Figure 5.

[0124] In this embodiment, the optimization phase took approximately 41.6 minutes. With the population size and number of iterations remaining constant, the traditional method, by iteratively calculating building energy consumption and photovoltaic power generation using Energyplus one call at a time, would take approximately 1130.8 minutes. The organic combination of the XGBoost surrogate model and NSGA-II reduces the optimization process time by approximately 96.3% while maintaining prediction accuracy, significantly improving computational efficiency.

[0125] Table 8 U-value SHGC Tvis CCR Lowe WWR Cooling energy consumption [kW·h] Heating energy consumption [kW·h] Lighting energy consumption [kW·h] Photovoltaic power generation [kW·h] 1.135 0.113 0.061 90% 0.2 0.9 456.77 323.13 69.09 501.96 0.977 0.513 0.546 10% 0.15 0.88 969.33 124.43 19.34 55.53 0.714 0.481 0.539 11% 0.06 0.65 772.2 91.68 20.95 39.93 0.580 0.061 0.058 90% 0.02 0.3 308.77 237.31 98.56 166.73 1.128 0.113 0.056 90% 0.2 0.9 455.51 323.19 71.13 500.73 0.686 0.334 0.536 10% 0.05 0.3 437.92 111.62 28.48 18.44 1.045 0.512 0.540 10% 0.17 0.89 977.37 143.43 19.34 55.72 0.591 0.061 0.059 90% 0.02 0.3 308.77 237.31 98.56 166.73 0.567 0.141 0.096 83% 0.01 0.34 348.04 226.05 77.1 174.72 0.750 0.067 0.064 89% 0.07 0.89 380.66 265.97 67.36 488.47 0.963 0.405 0.482 17% 0.14 0.89 845.25 119.58 20.5 99.91 0.567 0.102 0.079 86% 0.01 0.31 325 230.68 87.98 157.47 1.040 0.249 0.277 49% 0.17 0.81 603.22 206.01 27.67 249.46 1.130 0.141 0.099 83% 0.2 0.9 471.38 290.57 50.73 459.32 1.123 0.087 0.103 83% 0.2 0.89 410.15 303.8 49.47 446.3 0.584 0.123 0.106 82% 0.02 0.69 405.57 218.1 54.27 347.51 0.566 0.087 0.086 85% 0.01 0.31 318.87 231.16 84.75 153.44 0.873 0.196 0.201 66% 0.11 0.86 553.37 187.77 32.61 337.9 0.750 0.094 0.081 87% 0.07 0.9 421.77 249.06 58.13 475.81 0.756 0.086 0.072 87% 0.08 0.79 391.62 251.28 65.36 426.94 1.040 0.339 0.358 37% 0.17 0.81 702.91 171.68 24.4 187.13 0.626 0.079 0.062 90% 0.03 0.3 319.9 236.54 95.9 165.2 0.673 0.111 0.125 78% 0.05 0.9 431.52 213.16 43.52 426 1.059 0.479 0.525 13% 0.18 0.89 944.77 146.36 19.59 66.39 1.056 0.290 0.279 52% 0.18 0.83 667.9 201.22 27.36 272.24 1.125 0.144 0.097 83% 0.2 0.9 489.33 291.72 52.76 459.7 0.674 0.160 0.143 76% 0.05 0.9 510.95 211.37 39.78 408.11 0.997 0.437 0.478 21% 0.15 0.89 882.64 157.61 20.4 121.18 1.124 0.091 0.089 86% 0.2 0.9 427.7 306.36 54.38 472.8 0.966 0.316 0.424 22% 0.15 0.88 733.22 128.22 21.89 128.18 1.045 0.368 0.366 39% 0.17 0.84 756.21 175.21 23.8 198.33 1.075 0.469 0.497 16% 0.18 0.9 918.22 150.38 20 88.54 0.997 0.315 0.326 46% 0.15 0.8 676.06 185.01 25.8 220.21 0.557 0.092 0.096 83% 0.01 0.33 325.5 229.6 77.61 178.98 1.130 0.062 0.056 90% 0.2 0.9 382.37 320.15 72.4 497.94 1.042 0.275 0.369 39% 0.17 0.84 657.22 191.38 24.02 203.69 0.991 0.375 0.521 12% 0.16 0.88 817.35 161.95 20.08 66.19 0.577 0.164 0.146 75% 0.02 0.69 442.48 195.27 43.2 304.63 0.973 0.305 0.433 21% 0.15 0.88 719.04 127.67 21.65 116.46 0.868 0.181 0.194 67% 0.11 0.86 533.47 190.38 33.63 339.16 1.135 0.090 0.101 83% 0.2 0.9 420.98 303.66 50.28 451.54 0.736 0.089 0.069 88% 0.07 0.9 414.76 264.9 63.75 489.53 0.997 0.333 0.300 46% 0.15 0.8 697.15 182.94 26.56 231.84 0.697 0.461 0.540 10% 0.05 0.3 507.67 99.08 27.05 18.68 0.947 0.395 0.514 10% 0.14 0.83 799.38 111.91 20.49 52.16 1.053 0.232 0.244 59% 0.18 0.82 585.79 215.17 30.2 288.97 1.046 0.260 0.349 40% 0.17 0.81 620.66 189.12 24.89 198.59 0.729 0.073 0.074 87% 0.07 0.9 391.08 256.54 61.02 479.09 0.978 0.441 0.497 10% 0.15 0.88 892.96 139.55 20.05 58.55 1.129 0.066 0.064 89% 0.2 0.9 387.7 316.53 67.35 487.93 Where CCR is the photovoltaic coverage rate, Lowe is the low emissivity, and WWR is the window-to-wall ratio. The solutions are arranged in descending order of congestion distance.

[0126] Finally, it should be noted that: The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention. Non-essential improvements, adjustments or substitutions made by those skilled in the art based on the content of this specification are all within the scope of protection claimed by the present invention.

Claims

1. A method for optimizing the photothermal and electrical performance of vacuum photovoltaic glass based on XGBoost and NSGA-II, characterized in that, Includes the following steps: S1: Calculate the overall photothermal performance parameters of PVVG, including the heat transfer coefficient U-value, total solar transmittance SHGC, and visible light transmittance Tvis. S2: Define the climate zone, building orientation, basic parameters of the room with PVVG installed, and the overall light and thermal performance parameters of different PVVGs. Input the above information into Energyplus to generate a building performance dataset containing multiple samples. Each sample in the building performance dataset contains 6 input parameters and 4 target energy consumption parameters. The input parameters include U value, SHGC, Tvis, photovoltaic coverage, low emissivity, and window-to-wall ratio. The target energy consumption parameters include artificial lighting energy consumption, cooling energy consumption, heating energy consumption, and photovoltaic power generation. S3: Using the XGBoost machine learning algorithm to access building performance datasets, surrogate models are trained for buildings equipped with PVVG to quickly and accurately predict building cooling, heating, lighting energy consumption, and photovoltaic power generation. S4: Set the design variables and objective function, couple the XGBoost surrogate model and NSGA-II multi-objective optimization algorithm in MATLAB, and obtain the PVVG parameter combination with good photothermal and electromechanical performance after optimization.

2. The method for optimizing the photothermal and electrical performance of vacuum photovoltaic glass based on XGBoost and NSGA-II as described in claim 1, characterized in that, In step S1, The formula for calculating the heat transfer coefficient U is: ; Where he is the external convective heat transfer coefficient; hi is the internal convective heat transfer coefficient. is the sum of the thermal resistances of all glass layers; dj and λj are the thickness and thermal conductivity of the glass layers, respectively; hs,i is the thermal conductivity of the hollow layer; hs,v is the thermal conductivity of the vacuum layer. ; Where hgs,i and hrs,i are the convective and radiative conductivities, respectively; λi is the thermal conductivity of the gas; Di is the thickness of the gas layer; Nui is the Nusselt number, which is taken as Nui = 1 when Nui < 1; σ is the Stefan-Boltzmann constant; ε1 and ε2 are the emissivity of the glass on both sides of the gas layer, respectively; and Tm,i is the average temperature of the inner surface of the glass on both sides of the gas layer. ; Where α is the gas overall adaptability coefficient; γ is the gas heat capacity; R is the molar gas constant; P is the residual gas pressure; M is the molar mass of the gas; and r, a, h, and λpillar are the support column radius, spacing, height, and thermal conductivity, respectively.

3. The method for optimizing the photothermal and electrical performance of vacuum photovoltaic glass based on XGBoost and NSGA-II as described in claim 1, characterized in that, In step S1, The formula for calculating the total solar transmittance (SHGC) is: ; Where Tsol is the total solar transmittance of the glass system, qin is the net heat flux entering the room under solar radiation conditions, qin(Is=o) is the net heat flux entering the room under no solar radiation conditions, Is is the solar incident irradiance acting on the outer surface of the glass; i is the i-th layer of glass from the outside to the inside; ki is the thermal conductivity of the glass; di is the glass thickness; Ti,o and Ti,i are the outer and inner surface temperatures of the glass, respectively; Ji,o and Ji,i are the outer and inner surface radii, respectively; J, ε, ρ, and τ are the long-wave radii, emissivity, reflectivity, and transmittance, respectively; σ is the Stefan-Boltzmann constant; T is the surface temperature; Jin and Jout are the indoor and outdoor environmental long-wave radii, respectively. The formula for calculating visible light transmittance (Tvis) is: ; Where λwj / j+1 represents the wavelength interval λwj, λwj+1; τsl is the average spectral transmittance; Evs is the average value of the relative spectral distribution; R is the average value of the spectral luminous efficiency function of human photopic vision; Δλwj is the wavelength interval; and Nvs is the number of wavelength sampling points in the visible light band.

4. The method for optimizing the photothermal and electrical performance of vacuum photovoltaic glass based on XGBoost and NSGA-II as described in claim 1, characterized in that, In step S2, The formula for artificial lighting energy consumption is as follows: ; El is the energy consumption for artificial lighting; fl(t) is the artificial lighting demand factor; Pdes is the power density for zone lighting design; Eset is the design illuminance setpoint; Eday(t) is the hourly indoor natural illuminance. The formulas for cooling energy consumption and heating energy consumption are as follows: ; Qsurf is the convective heat transfer on the surfaces of each building envelope; Qair is the heat exchange between air nodes within the area; Qpeople, Qequip, and Qlight are the sensible heat of indoor personnel, equipment, and lighting, respectively; Qinf and Qvent are the sensible heat of infiltration air and mechanical fresh air, respectively; Eheat and Ecool are the heating and cooling energy consumption, respectively. The formula for photovoltaic power generation is as follows: ; Egen is the PVVG power generation; Pgen is the PVVG AC output power; Asurf is the effective light-receiving area of ​​the PVVG; G is the total solar radiation irradiance received by the PVVG surface; η0 is the photoelectric conversion efficiency of the photovoltaic cell under standard test conditions; ηinvert is the inverter efficiency; β0 is the temperature coefficient of maximum power of the module. T PV is the actual operating temperature of the module, which is replaced by PVVG, the outdoor temperature.

5. The method for optimizing the photothermal and electrical performance of vacuum photovoltaic glass based on XGBoost and NSGA-II as described in claim 1, characterized in that, Step S3 includes the following steps: S31. Input Data Preparation The dataset is randomly shuffled and divided into training and testing sets. The input parameters and target energy consumption parameters in the dataset are linearly normalized to map all data into the 0-1 range. S32, XGBoost model parameter settings Set the XGBoost training parameters, including maximum tree depth, shrinkage step size, minimum sample weight, subsample ratio, feature sampling ratio, number of parallel trees, objective function type, and boosting method; S33, Training Set Modeling The training set is input into the XGBoost package, the residuals are calculated through multiple rounds of iteration and a regression tree structure is generated, the gain is calculated based on the second-order Taylor expansion, and the optimal split point is determined by a greedy algorithm and a pre-sorting algorithm to establish an XGBoost prediction model for a single target energy consumption parameter. S34, Test Set Prediction The test set is input into the trained XGBoost model, and the input samples are traversed layer by layer according to the generated regression tree structure. The outputs of each tree are accumulated to obtain the prediction results of the target energy consumption parameters. Finally, inverse normalization is performed. S35, Model File Generation The MATLAB environment calls the XGBoost state link library to generate binary model files and MAT files. The binary model file saves the regression tree structure and parameter information of the model, while the MAT file saves the input and output normalization parameters and model path configuration. S36. Take steps S32 to S35 as a training process, train the XGBoost model for each of the four target energy consumption parameters. After completing the training of one model, repeat steps S32 to S35 to complete the training of the remaining three target energy consumption parameters in sequence, and obtain the binary model file and MAT file of the XGBoost building energy consumption and photovoltaic power generation prediction model.

6. The method for optimizing the photothermal and electrical performance of vacuum photovoltaic glass based on XGBoost and NSGA-II as described in claim 5, characterized in that, Between steps S34 and S35, there is also: error evaluation, calculating the performance metrics of the model on the training set and the test set, including RMSE, R², MAE, MBE, MRE and cv-RMSE, to form the error evaluation result of the objective function model; if the evaluation result exceeds the acceptable range, then repeat steps S32 to S35 to conduct a new round of model training and evaluation until the result meets the requirements.

7. The method for optimizing the photothermal and electrical performance of vacuum photovoltaic glass based on XGBoost and NSGA-II as described in claim 1, characterized in that, Step S4 includes the following steps: S41. Initialize the population: Within the range of design variables, MATLAB assigns a set of building parameter combinations to each individual based on the set running parameters, and obtains multi-objective function values ​​by calling the XGBoost prediction model. S42. Perform non-dominated quicksort and crowding calculation within the population. S43. Selection, Crossover, and Mutation Individuals are selected to enter the mating pool using a tournament selection mechanism. Offspring individuals are generated by simulating binary crossover and polynomial mutation. After setting the range of values ​​for the offspring decision variables, the XGBoost prediction model is called again to calculate the multi-objective function value and form the offspring population. S44, Elite Preservation and Population Renewal The parent and offspring generations are merged, and non-dominated quicksort and crowding calculation are performed again. All individuals are arranged in ascending order according to their rank, and individuals of the same rank are arranged in descending order according to their crowding distance. Individuals with higher rankings are selected to form a new population, and the population size is the same as the originally set population size. S45. Iteration and Result Output Repeat steps S43 to S44 until the maximum number of iterations is reached, at which point the calculation ends, yielding the Pareto optimal solution set. The trade-offs between the objectives are then visualized using a 3D scatter plot.

Images