A building carbon emission multi-objective optimization method and system based on a proxy model

Abstract

The application provides a building carbon emission multi-objective optimization method and system based on an agent model, relates to the technical field of building carbon emission optimization, and comprises the following steps: collecting building multi-dimensional data and performing preprocessing; constructing and training a building multi-objective prediction model based on the preprocessed building multi-dimensional data; defining a building carbon emission multi-objective optimization problem, solving the building carbon emission multi-objective optimization problem by combining a multi-objective evolutionary algorithm with the trained building multi-objective prediction model, and obtaining a Pareto optimal solution set; performing standardization and continuous fitting on the Pareto optimal solution set to obtain a continuous Pareto frontier; extracting multi-dimensional feature data of each solution based on the continuous Pareto frontier; screening the Pareto optimal solution set based on the multi-dimensional feature data set of the solution set to obtain a final carbon emission optimization solution; and updating building design parameters based on the final carbon emission optimization solution. The application realizes the integration of efficient optimization and intelligent decision-making of building low-carbon design.

Description

Technical Field

[0001] This invention relates to the field of building carbon emission optimization technology, and in particular to a multi-objective optimization method and system for building carbon emissions based on a surrogate model. Background Technology

[0002] Low-carbon building design has become a core development direction in the industry. Multi-objective optimization technology, which can balance carbon emissions, operation and maintenance costs, and indoor comfort, has been widely used in the building design phase. Current technologies often employ purely data-driven models such as artificial neural networks and Kriging as surrogate tools, combined with multi-objective evolutionary algorithms such as NSGA-II and NSGA-III to solve building performance optimization problems. By generating Pareto optimal solution sets, multiple alternatives are provided for reference, which to some extent improves the scientific rigor and efficiency of low-carbon building design.

[0003] However, existing technologies still face many critical issues in practical engineering applications, failing to meet the core requirements of efficient optimization and intelligent decision-making. Firstly, purely data-driven proxy models lack constraints from building thermal physics laws, relying solely on data fitting relationships. This can easily lead to abnormal predictions that violate thermal principles under sparse data or extreme conditions, making it difficult to guarantee the reliability of the optimization foundation. Secondly, traditional optimization methods often directly call full-process simulation software such as EnergyPlus to calculate target values. Even when coupled with proxy models, insufficient model accuracy and generalization result in excessively long optimization iteration cycles, making it difficult to balance the distribution, convergence, and optimization efficiency of the solution set. Thirdly, the selection of Pareto optimal solutions still relies mainly on manual subjective weighting or simple single-objective optimization, without in-depth feature mining of discrete solution sets. This makes it impossible to accurately identify cost-effective inflection point solutions, resulting in highly subjective and poorly reproducible selection results, severely impacting the engineering feasibility of optimization schemes.

[0004] How to solve the above-mentioned technical problems is the challenge facing this invention. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a multi-objective optimization method and system for building carbon emissions based on a proxy model, which integrates efficient optimization and intelligent decision-making in building low-carbon design.

[0006] The technical solution adopted by this invention to solve its technical problem is: This invention provides a multi-objective optimization method for building carbon emissions based on a surrogate model, comprising the following steps: Collect and preprocess multidimensional building data to obtain preprocessed multidimensional building data; the multidimensional building data includes building design parameters, environmental parameters, carbon emission values, operation and maintenance cost values, and uncomfortable hours. A multi-objective prediction model for buildings is constructed, and the multi-objective prediction model for buildings is trained based on preprocessed multi-dimensional building data to obtain the trained multi-objective prediction model for buildings. A multi-objective optimization problem for building carbon emissions is defined. A multi-objective evolutionary algorithm combined with a trained multi-objective prediction model for buildings is used to solve the multi-objective optimization problem for building carbon emissions, and the Pareto optimal solution set is obtained. The Pareto optimal solution set is standardized and fitted with a continuous model to obtain a continuous Pareto front. Based on the continuous Pareto front, multidimensional feature data of each solution is extracted to obtain a multidimensional feature dataset of the solution set. The multidimensional feature data includes curvature, marginal sensitivity coefficient, and crowding degree. The Pareto optimal solution set is filtered based on the multidimensional feature dataset of the solution set to obtain the final solution for carbon emission optimization; The building design parameters are updated based on the final solution of carbon emission optimization to obtain the building with optimized carbon emissions.

[0007] Preferably, the preprocessing includes data cleaning, data standardization, and data partitioning.

[0008] Preferably, the building multi-objective prediction model adopts the Physical Information Neural Network (PINN) model, which takes building design parameters and environmental parameters as inputs and carbon emission values, operation and maintenance cost values ​​and uncomfortable hours as outputs; The training of the multi-objective building prediction model based on preprocessed multi-dimensional building data includes: A composite loss function is constructed. The preprocessed multidimensional building data is used as training samples. The multi-objective prediction model of buildings is trained under supervision with the goal of minimizing the composite loss function until the model converges. The trained multi-objective prediction model of buildings is then obtained. The composite loss function includes data fitting loss, residual loss of building thermal physics equations, and boundary condition loss.

[0009] Preferably, defining the multi-objective optimization problem of building carbon emissions includes defining the multi-objective optimization objective function and constraints for building carbon emissions; The multi-objective optimization objectives for building carbon emissions include minimizing building carbon emissions, minimizing building operation and maintenance costs, and minimizing the number of uncomfortable hours indoors. The constraints include feasibility constraints on building design parameters, carbon emission budget constraints, and indoor comfort compliance constraints.

[0010] Preferably, the step of solving the multi-objective optimization problem of building carbon emissions using a multi-objective evolutionary algorithm combined with a trained building multi-objective prediction model includes: The architectural design parameters are used as decision variables, and the feasible range of values ​​for the architectural design parameters constitutes the decision space. Under constraints, a multi-objective evolutionary algorithm is used to iteratively generate an optimal solution set in the decision space, and a trained multi-objective prediction model for buildings is used to predict the optimal solution set to obtain the target values ​​of the optimal solution set; the target values ​​of the optimal solution set include target values ​​for carbon emissions, operation and maintenance costs, and uncomfortable hours. Based on the objective value of the optimization scheme set, the optimization scheme set is sorted by non-dominated order and the constraint compliance is checked to select non-dominated solutions that meet the constraints. When the preset iteration termination condition is reached, all non-dominated solutions that satisfy the constraints are output, and the Pareto optimal solution set is obtained.

[0011] Preferably, the extraction of multidimensional feature data for each solution based on the continuous Pareto front includes: Define a parameterized vector function, and calculate the curvature of each solution with respect to the continuous Pareto front by taking the derivative of the parameterized vector function; The marginal sensitivity coefficient of each solution is calculated based on the trained multi-objective prediction model for buildings. Calculate the congestion degree for each solution.

[0012] Preferably, the filtering of the Pareto optimal solution set based on the multidimensional feature dataset of the solution set includes: Define a curvature threshold, filter Pareto optimal solutions with curvature greater than the curvature threshold, and obtain a set of candidate solutions; Using curvature, marginal sensitivity coefficient and crowding degree as evaluation indicators, the entropy weight method is used to calculate the weight coefficient of each evaluation indicator. Based on the weight coefficient, the TOPSIS method is used to calculate the comprehensive proximity score of each solution in the candidate solution set, and the comprehensive proximity scores are sorted in descending order from high to low. The solution with the highest overall similarity score is selected as the final solution for carbon emission optimization.

[0013] This invention also provides a multi-objective optimization system for building carbon emissions based on a surrogate model, including... The data acquisition and preprocessing module is used to collect multidimensional building data and preprocess it to obtain preprocessed multidimensional building data. The model building and training module is used to build a multi-objective prediction model for buildings and train the multi-objective prediction model for buildings based on preprocessed multi-dimensional building data to obtain the trained multi-objective prediction model for buildings. The multi-objective optimization module is used to define the multi-objective optimization problem of building carbon emissions. It uses a multi-objective evolutionary algorithm combined with a trained building multi-objective prediction model to solve the multi-objective optimization problem of building carbon emissions and obtain the Pareto optimal solution set. The solution set processing and feature extraction module is used to standardize and continuously fit the Pareto optimal solution set to obtain a continuous Pareto front; based on the continuous Pareto front, multidimensional feature data of each solution is extracted to obtain a multidimensional feature dataset of the solution set. The solution set filtering module is used to filter the Pareto optimal solution set based on the multidimensional feature dataset of the solution set to obtain the final solution for carbon emission optimization. The optimization execution module is used to update the building design parameters based on the final solution of carbon emission optimization, so as to obtain the building with optimized carbon emissions.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described multi-objective optimization method for building carbon emissions based on a proxy model.

[0015] The present invention also provides a computer storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described multi-objective optimization method for building carbon emissions based on a proxy model.

[0016] The beneficial effects of this invention are as follows: it integrates efficient optimization and intelligent decision-making in low-carbon building design. The PINN model, embedded with building thermal physics constraints, is used as a proxy model for multi-objective prediction of buildings. By constructing a composite loss function of "data fitting loss + building thermal physics equation residual loss + boundary condition loss," the model's prediction results are ensured to meet both high accuracy requirements and strict adherence to building thermal physics laws, avoiding abnormal predictions. Simultaneously, the successfully trained PINN model is coupled with the NSGA-III multi-objective evolutionary algorithm, significantly shortening the optimization cycle compared to traditional optimization methods that rely on simulation software, while ensuring the distribution and convergence of the solution set. The discrete solution set is fitted to a second-order continuously differentiable Pareto front, and three core features—curvature, marginal sensitivity coefficient, and crowding—are extracted. A two-stage objective screening system is constructed: "high curvature inflection point initial screening + entropy weight-TOPSIS multi-feature comprehensive ranking." This system can accurately identify high-cost-performance inflection point solutions while ensuring the scientific validity of the screening results, solving the problems of large subjective bias and weak engineering applicability in traditional screening methods. Attached Figure Description

[0017] Figure 1 This is a diagram illustrating the method steps of the present invention.

[0018] Figure 2 This is a system module diagram of the present invention.

[0019] Figure 3 This is a diagram of the internal structure of a computer device according to Embodiment 3 of the present invention. Detailed Implementation

[0020] To clearly illustrate the technical features of this solution, the following detailed implementation method will be used to explain the solution.

[0021] Example 1: See Figure 1 As shown, this embodiment is a multi-objective optimization method for building carbon emissions based on a surrogate model, including the following steps. S1. Collect and preprocess multidimensional building data to obtain preprocessed multidimensional building data; the multidimensional building data includes building design parameters, environmental parameters, carbon emission values, operation and maintenance cost values, and uncomfortable hours. Preprocessing includes data cleaning, data standardization, and data partitioning.

[0022] It should be noted that building design parameters may include window-to-wall ratio, thickness of external wall insulation layer, shading coefficient of external windows, heat transfer coefficient of building envelope, COP of air conditioning unit, lighting power density, etc.; environmental parameters may include hourly data of typical meteorological years in the project location (dry bulb temperature, wet bulb temperature, total solar radiation intensity on the horizontal plane, wind speed, relative humidity), grid carbon emission factor, local time-of-use electricity price, etc.; carbon emission value is the building life cycle carbon emission value calculated based on GB / T51366 "Standard for Calculation of Building Carbon Emissions"; operation and maintenance cost data is the construction and operation and maintenance life cycle cost value calculated based on the full cycle cost specification of building engineering; comfort data: the number of indoor uncomfortable hours calculated based on GB / T 50785 "Evaluation Standard for Indoor Thermal and Humidity Environment of Civil Buildings".

[0023] Data cleaning: This involves processing the collected multidimensional building data for outliers, missing values, and duplicate values. ① Outlier handling: The quartile method is used to identify outliers in numerical data such as building design parameters, environmental parameters, and carbon emission data. Continuous outliers are corrected by linear interpolation of the nearest valid value, while discrete outliers are directly removed. ② Missing value handling: If the percentage of missing fields in a single sample is ≤10%, the K-nearest neighbor (KNN) algorithm is used to fill in the missing values; if the percentage of missing fields is >10%, the sample is removed. ③ Duplicate value handling: Duplicate samples are identified based on a combination of features including building number, collection time, and core design parameters, and only the first valid sample is retained; Data standardization: Dimensionless processing is performed on the cleaned building multidimensional data to eliminate the impact of dimensional differences on subsequent model training. For benefit-oriented indicators such as carbon emission data, operation and maintenance costs, and uncomfortable hours, the extreme value standardization method (Min-Max standardization) is used to map the data to the [0,1] interval; For interval-type indicators such as building design parameters and environmental parameters, the Z-score standardization method is used to transform them into standard normal distribution data with a mean of 0 and a standard deviation of 1. Data partitioning: The standardized multidimensional building data is divided into training set, validation set and test set according to a preset ratio (preferably 7:2:1). The training set is used for parameter fitting of the multi-objective prediction model of buildings, the validation set is used for model hyperparameter tuning, and the test set is used for model generalization ability verification.

[0024] S2. Construct a multi-objective prediction model for buildings, and train the multi-objective prediction model for buildings based on the preprocessed multi-dimensional building data to obtain the trained multi-objective prediction model for buildings. The building multi-objective prediction model adopts the physical information neural network (PINN) model, which takes building design parameters and environmental parameters as inputs and carbon emission values, operation and maintenance cost values ​​and uncomfortable hours as outputs. Training a multi-objective building prediction model based on preprocessed multi-dimensional building data includes: A composite loss function is constructed. The preprocessed multidimensional building data is used as training samples. The multi-objective prediction model of buildings is trained under supervision with the goal of minimizing the composite loss function until the model converges. The trained multi-objective prediction model of buildings is then obtained. The composite loss function includes data fitting loss, residual loss from building thermal physics equations, and boundary condition loss.

[0025] It should be noted that constructing a white-box proxy model that embeds the physical mechanism of building thermal systems solves the industry pain points of traditional pure data-driven black-box models (ANN, Kriging) such as poor generalization, easy violation of physical laws, and reliance on massive training data. This provides a high-precision, millisecond-level fast predictor of objective functions for subsequent multi-objective optimization.

[0026] A fully connected feedforward neural network is used as the basic backbone network, with an end-to-end structure of "input layer - hidden layer - output layer". Specific parameters are as follows: Input layer: 9 neurons, corresponding one-to-one with the dimensions of 6 architectural design parameters + 3 core environmental parameters, with no bias term; Hidden layers: 5 fully connected layers, each with 128 neurons, with batch normalization used between layers. The activation function is the Swish function (which is more suitable for the physical constraints of PINN training than the ReLU function and can effectively alleviate gradient vanishing; the formula is expressed as:). ; Output layer: Number of neurons = 3, corresponding to the building's total life cycle carbon emissions, operation and maintenance costs, and indoor discomfort hours, respectively, which are completely consistent with the target output data of S1. The activation function is a linear function.

[0027] A multi-component composite loss function is constructed, which combines "data fitting, physical mechanism, and boundary constraints," with minimizing the total loss as the training optimization objective. The formula is as follows:

[0028] The specific implementation of each component in the formula is as follows: The data fitting loss is calculated using the mean squared error (MSE) to constrain the fitting accuracy between the model's predicted values ​​and the actual sample values. The formula is as follows:

[0029] In the formula, This refers to the number of samples within a batch. These are the model's predicted values. This represents the model loss value.

[0030] The residual loss of the building thermal physics equations: The core is the one-dimensional unsteady-state heat conduction control equations of the building envelope as specified in the "Code for Thermal Design of Buildings" GB50176. The residuals are calculated based on PINN's automatic differentiation capability. The output from the underlying constraint model must conform to the laws of building thermal physics to avoid abnormal predictions that violate common sense. The control equations are:

[0031] In the formula, For the density of the enclosure structure material, For the specific heat capacity of the material, The thermal conductivity of the material. In the direction of wall thickness Place, Temperature at any given time; residual loss is the mean square error between the calculated value of the equation and 0. The closer the residual is to 0, the more the model output conforms to physical laws. Boundary condition loss: The third type of convective heat transfer boundary condition, commonly used for building envelopes, is adopted. The constraint model output meets the boundary requirements for outdoor convective heat transfer and indoor design temperature. The formula is: Indoor Boundaries:

[0032] Outdoor boundary:

[0033] In the formula, This is the total thickness of the wall. These are the outdoor and indoor convective heat transfer coefficients, respectively. These represent the hourly outdoor temperature and the indoor design temperature, respectively; the boundary loss is the mean square error between the calculated boundary condition value and 0. Weighting coefficient This is used to balance the priority of data fitting accuracy and physical constraints, ensuring that the model simultaneously meets the requirements of prediction accuracy and physical consistency.

[0034] Training environment and framework: The hardware uses NVIDIA RTX 3090 GPU to accelerate training, and the software framework uses TensorFlow 2.10, which is based on the automatic differentiation module to realize the automatic calculation of the physical equation residuals. Training hyperparameter settings: The optimizer used is the Adam adaptive gradient descent optimizer, and the initial learning rate is set to 1×10. 3. The learning rate decay strategy is to decrease by 10% every 50 epochs, the batch size is 64, and the maximum number of iterations is 1000. Convergence criteria: Training stops when any of the following conditions are met: ① The lumped loss decreases by less than 1×10 for 20 consecutive epochs. 5; ② Reach the maximum number of iterations of 1000; Save the model weights with the lowest loss on the validation set every 10 epochs during training; Model accuracy verification criteria: Two-dimensional verification is performed using a test set. Only models that pass the verification can be used for subsequent optimization solutions. The passing criteria are as follows: Data fitting accuracy: The mean absolute percentage error (MAPE) of all three outputs is ≤8%, and the coefficient of determination (R²) is ≥0.92. Physical consistency accuracy: Residual loss of building thermal equations ≤ 1 × 10 4. No abnormal prediction results that violate the laws of thermodynamics.

[0035] S3. Define a multi-objective optimization problem for building carbon emissions, and use a multi-objective evolutionary algorithm combined with a trained multi-objective prediction model for buildings to solve the multi-objective optimization problem for building carbon emissions and obtain the Pareto optimal solution set. Defining the multi-objective optimization problem of building carbon emissions includes defining the multi-objective optimization objective function and constraints for building carbon emissions; The multi-objective optimization goals for building carbon emissions include minimizing building carbon emissions, minimizing building operation and maintenance costs, and minimizing the number of uncomfortable hours indoors. The constraints include feasibility constraints on building design parameters, carbon emission budget constraints, and indoor comfort compliance constraints.

[0036] The multi-objective optimization problem of building carbon emissions is solved by combining a multi-objective evolutionary algorithm with a trained multi-objective prediction model for buildings, including: The architectural design parameters are used as decision variables, and the feasible range of values ​​for the architectural design parameters constitutes the decision space. Under constraints, a multi-objective evolutionary algorithm is used to iteratively generate an optimal solution set in the decision space, and a trained multi-objective prediction model for buildings is used to predict the optimal solution set to obtain the target values ​​of the optimal solution set. The target values ​​of the optimal solution set include the target values ​​of carbon emissions, operation and maintenance costs, and discomfort hours. Based on the objective value of the optimization scheme set, the optimization scheme set is sorted by non-dominated order and the constraint compliance is checked to select non-dominated solutions that meet the constraints. When the preset iteration termination condition is reached, all non-dominated solutions that satisfy the constraints are output, and the Pareto optimal solution set is obtained.

[0037] It should be noted that by clearly defining the mathematical boundaries and constraint rules of multi-objective optimization, and coupling the trained PINN surrogate model with a multi-objective evolutionary algorithm, the Pareto optimal solution set that satisfies the engineering constraints can be quickly obtained, thus solving the problems of extremely low efficiency and long iteration cycle of traditional methods that directly call simulation software for optimization.

[0038] Construct a constrained three-objective minimization optimization problem, the complete mathematical expression of which is:

[0039] The specific implementation of each parameter in the formula is as follows: Decision variables: 6 architectural design parameters defined in S1. , respectively Window-to-wall ratio Thickness of external wall insulation layer Window shading coefficient heat transfer coefficient of building envelope COP of air conditioning units Lighting power density; Optimize the objective function: 3 conflicting minimization objectives, quickly predicted using a trained PINN model: Minimize carbon emissions throughout the building's life cycle; Minimize the operation and maintenance costs throughout the entire building lifecycle; Minimize the number of hours of indoor thermal discomfort; Constraints: Boundary constraints The feasible range of architectural design parameters conforms to the requirements of building design codes. For example: window-to-wall ratio 0.3~0.7, external wall insulation layer thickness 20~150mm, external window shading coefficient 0.2~0.8, and building envelope heat transfer coefficient 0.3~1.5W / (㎡). K), air conditioning COP≥2.6, lighting power density≤5.0W / ㎡; Performance constraints The constraints include: ① Carbon emission budget constraints: Total life-cycle carbon emissions ≤ the carbon emission limit standard for civil buildings in the project location; ② Indoor comfort compliance constraints: Uncomfortable hours per year ≤ 120h / year (compliant with GB50736).

[0040] The NSGA-III algorithm is used to solve the problem. ① Algorithm hyperparameter initialization: Set population size = 200, crossover probability = 0.9, mutation probability = 0.1, maximum number of iterations = 200 generations, and use the Das-Dennis method to generate uniformly distributed reference points (number of reference points = 91) to adapt to the three-dimensional spatial distribution of the three-objective optimization. ②PINN model coupling: The trained PINN model is used as a fast predictor of the objective function and coupled into the iterative process of the NSGA-III algorithm, replacing the traditional successive simulation software calls, which effectively improves the optimization efficiency; ③ Iterative evolution process: Within the feasible region of the constraints, complete population initialization, simulate binary crossover, and polynomial mutation to generate offspring population; perform non-dominated sorting based on reference point on the merged parent and offspring population, and at the same time complete the constraint compliance check, prioritize retaining non-dominated solutions that satisfy the constraints, and eliminate infeasible solutions; ④ Iteration termination condition: The iteration terminates when any of the following conditions are met: 1. The maximum number of iterations of 200 generations is reached; 2. There are no updates to the non-dominated solution set for 30 consecutive generations; After the iteration terminates, all non-dominated solutions that satisfy the constraints are output to obtain the final Pareto optimal solution set.

[0041] S4. Standardize and continuously fit the Pareto optimal solution set to obtain the continuous Pareto front; extract multidimensional feature data for each solution based on the continuous Pareto front to obtain the multidimensional feature dataset of the solution set; the multidimensional feature data includes curvature, marginal sensitivity coefficient and crowding degree. The multidimensional feature data extracted for each solution based on the continuous Pareto front includes: Define a parameterized vector function, and calculate the curvature of each solution with respect to the continuous Pareto front by taking the derivative of the parameterized vector function; The marginal sensitivity coefficient of each solution is calculated based on the trained multi-objective prediction model for buildings. Calculate the congestion degree for each solution.

[0042] It should be noted that by transforming the discrete Pareto optimal solution set into a second-order continuously differentiable Pareto front, multi-dimensional core features that can characterize the engineering value of the solution are extracted, providing an objective quantitative basis for subsequent screening and solving the problems of discrete solution sets being unable to accurately identify high-performance inflection points and lacking quantitative screening indicators.

[0043] 1. Pareto optimal solution set standardization and continuous fitting implementation Solution set standardization: For the three optimization objective values ​​of the Pareto optimal solution set, the Min-Max extreme value standardization method is used to linearly map each objective value to the interval [0,1] to eliminate the dimensional differences between different objectives and obtain the standardized three-dimensional objective space discrete point set; Continuous fitting: The standardized discrete point set is fitted using cubic B-spline interpolation to obtain a continuous Pareto front in the three-dimensional target space that is second-order continuously differentiable. The core reason for choosing cubic B-splines is that they have local support and second-order continuous differentiability, which can ensure the accuracy and stability of subsequent curvature calculations, while avoiding the Runge phenomenon of high-order polynomial fitting and adapting to the nonlinear characteristics of the Pareto front.

[0044] 2. Quantization of the entire process for multidimensional feature extraction: Based on the continuous Pareto front, three core features are extracted for each discrete Pareto solution, constructing a multidimensional feature dataset that corresponds one-to-one with the solution set. The specific extraction method is as follows: ① Curvature feature extraction (geometric features, cost-effectiveness in representing inflection points): Construct a three-dimensional parameterized vector function for the continuous Pareto front: ,in For continuous arc length parameters, These are, respectively, standardized carbon emissions, operation and maintenance costs, and hours of discomfort; Find the first derivative of the parameterized vector function Second derivative The local curvature of each solution point is calculated using the curvature formula of the three-dimensional parametric curve. The formula is:

[0045] Physical meaning: The higher the curvature, the more dramatic the marginal substitution rate of the three optimization objectives changes at that point. It is a high-cost-performance inflection point that "slightly sacrifices non-core objectives and significantly improves core objectives", and the higher the engineering value.

[0046] ② Marginal sensitivity coefficient extraction (physical characteristics, characterizing the rationality of the target trade-off): Based on the automatic differentiation capability of the trained PINN model, the gradient data of the optimization objective with respect to the architectural design parameters of each solution is calculated. The dimensionless marginal sensitivity coefficients between targets are calculated using the elasticity coefficient method. The formula is:

[0047] In the formula, For the goal For the target The marginal sensitivity coefficient, in physical terms, is: the target For every 1% change, the target The corresponding magnitude of change directly quantifies the marginal rate of substitution and cost-effectiveness between the two conflicting objectives; Physical meaning: The closer it is to 1, the more balanced the trade-off between the two goals is, with no extreme sacrifices. A value much greater than 1 indicates the target For the target The changes are extremely sensitive, and the cost-effectiveness of the project is very low.

[0048] ③ Crowding feature extraction (distribution features, representativeness of the representation scheme): Using the crowding calculation rules of NSGA-III, each optimization objective dimension is sorted in ascending order of its objective value; For each solution, the sum of the differences between its objective value and the objective values ​​of its two adjacent solutions in each objective dimension is calculated; this is the crowding degree of the solution, given by the formula:

[0049] In the formula, For the first The crowding of each solution The first In terms of the target dimension, the first The objective value of two adjacent solutions; Physical meaning: The greater the crowding, the sparser the distribution of the solution in the target space, and the stronger its representativeness. This can prevent the selected solutions from being concentrated in the same trade-off interval and ensure the balance of the selection results.

[0050] S5. Based on the multidimensional feature dataset of the solution set, the Pareto optimal solution set is filtered to obtain the final solution for carbon emission optimization; Filtering Pareto optimal solution sets based on multidimensional feature datasets of solution sets includes: Define a curvature threshold, filter Pareto optimal solutions with curvature greater than the curvature threshold, and obtain a set of candidate solutions; Using curvature, marginal sensitivity coefficient and crowding degree as evaluation indicators, the entropy weight method is used to calculate the weight coefficient of each evaluation indicator. Based on the weight coefficient, the TOPSIS method is used to calculate the comprehensive proximity score of each solution in the candidate solution set, and the comprehensive proximity scores are sorted in descending order from high to low. The solution with the highest overall similarity score is selected as the final solution for carbon emission optimization.

[0051] It should be noted that through a two-stage screening process of "initial screening of high-value inflection points → comprehensive objective ranking", the single optimal solution with the highest engineering value is locked from a massive number of Pareto solutions, completely solving the pain points of traditional screening methods such as large subjective bias, poor reproducibility, and weak engineering implementation.

[0052] Phase 1: Initial screening at high curvature inflection points (wide-range convergence) Adaptive curvature threshold setting: Calculate the overall arithmetic mean of the curvature values ​​of the entire Pareto solution set, and set the curvature filtering threshold to 1.5 times the arithmetic mean, while forcibly limiting the threshold value range to 0.6~0.8; the setting rules for the upper and lower limits of the threshold are as follows: if the calculated threshold is >0.8, it is forcibly set to 0.8; if the calculated threshold is <0.6, it is forcibly set to 0.6, to avoid the problem of no effective solutions due to the threshold being too high and the filtering range being too large due to the threshold being too low; Initial screening rules: Select Pareto optimal solutions with curvature greater than the curvature screening threshold to obtain a candidate solution set composed of high-value inflection points, complete the first round of large-scale convergence, and filter out low-cost linear solutions.

[0053] The second stage: Multi-feature comprehensive ranking based on entropy weight-TOPSIS (precise optimization). Curvature, marginal sensitivity coefficient, and crowding are used as the sole evaluation indicators. The comprehensive ranking is completed using the completely objective entropy weight-TOPSIS method. The specific steps are as follows: ① Evaluation Dimension Differentiation and Standardization: Clarify the optimization direction of each indicator, use a differentiation and standardization formula to eliminate dimensional differences, and map the standardized values ​​to the [0,1] interval. The closer the value is to 1, the better the performance. Curvature: The larger the value, the better; positive Min-Max normalization is used. Marginal sensitivity coefficient: The closer the value is to 1, the better (representing a reasonable balance between objectives). The optimal standardization is adopted using the center-optimal method, and the formula is: Crowding degree: A moderately high value is better (representing a balanced distribution and strong representativeness). The optimal standardization is achieved using the center-centered method, and the formula is as follows:

[0054] In the formula, The standardized value. The original value, For the first The arithmetic mean of the indicators.

[0055] ② Objective weight calculation based on entropy weight method: Using the candidate solution set as the calculation sample, the objective weights of the three evaluation indicators are calculated using the entropy weight method. The core logic is: the higher the dispersion of the indicator data, the greater its contribution to the scheme discrimination, and the higher its weight. There is no subjective human assignment throughout the process. The specific steps are as follows: Calculate the first The first indicator The characteristic proportion of each scheme :

[0056] In the formula, The total number of alternative solutions; if Then let ; Calculate the first Information entropy of each indicator: ; Calculate the first Coefficient of difference of each indicator : ; Calculate the first Objective weighting coefficients of each indicator : ③ Comprehensive evaluation and ranking based on the TOPSIS method: Construct a weighted standardized evaluation matrix: ; Determine the ideal solution (The set of optimal values ​​for each indicator) and the negative ideal solution (Set of worst values ​​for each indicator):

[0057] Calculate the Euclidean distance from each solution to the positive ideal solution. Euclidean distance to the negative ideal solution :

[0058] Calculate the overall similarity score for each option: The value range is [0,1], and the closer it is to 1, the better the overall performance of the solution; Based on overall relevance score Sort the candidate solution set in descending order from highest to lowest.

[0059] The solution with the highest overall closeness score is selected as the final solution for carbon emission optimization. If multiple solutions have the same highest overall closeness score, the solution with the best carbon emission target performance is selected as the final solution.

[0060] S6. Update the building design parameters based on the final solution of carbon emission optimization to obtain the building with optimized carbon emissions.

[0061] Example 2: See Figure 2 As shown, this embodiment is a multi-objective optimization system for building carbon emissions based on a surrogate model, including... The data acquisition and preprocessing module is used to collect multidimensional building data and preprocess it to obtain preprocessed multidimensional building data. The model building and training module is used to build a multi-objective prediction model for buildings and train the multi-objective prediction model for buildings based on preprocessed multi-dimensional building data to obtain the trained multi-objective prediction model for buildings. The multi-objective optimization module is used to define the multi-objective optimization problem of building carbon emissions. It uses a multi-objective evolutionary algorithm combined with a trained building multi-objective prediction model to solve the multi-objective optimization problem of building carbon emissions and obtain the Pareto optimal solution set. The solution set processing and feature extraction module is used to standardize and continuously fit the Pareto optimal solution set to obtain a continuous Pareto front; based on the continuous Pareto front, multidimensional feature data of each solution is extracted to obtain a multidimensional feature dataset of the solution set. The solution set filtering module is used to filter the Pareto optimal solution set based on the multidimensional feature dataset of the solution set to obtain the final solution for carbon emission optimization. The optimization execution module is used to update the building design parameters based on the final solution of carbon emission optimization, so as to obtain the building with optimized carbon emissions.

[0062] Example 3: This embodiment provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0063] This computer device can be a server, and its internal structure diagram can be as follows: Figure 3 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores server data. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a multi-objective optimization method for building carbon emissions based on a proxy model.

[0064] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0065] Example 4: This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0066] If the functions implemented by the method are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art or the current technical solution, can be embodied in the form of a software product. This current computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0067] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0068] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0069] The technical features of this invention not described can be implemented by or using existing technology, and will not be repeated here. Of course, the above description is not a limitation of this invention, and this invention is not limited to the examples above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of this invention should also be within the protection scope of this invention.

Claims

1. A multi-objective optimization method for building carbon emissions based on a surrogate model, characterized in that, Includes the following steps Collect multidimensional building data and preprocess it to obtain preprocessed multidimensional building data; The building's multidimensional data includes building design parameters, environmental parameters, carbon emission values, operation and maintenance cost values, and hours of discomfort. A multi-objective prediction model for buildings is constructed, and the multi-objective prediction model for buildings is trained based on preprocessed multi-dimensional building data to obtain the trained multi-objective prediction model for buildings. A multi-objective optimization problem for building carbon emissions is defined. A multi-objective evolutionary algorithm combined with a trained multi-objective prediction model for buildings is used to solve the multi-objective optimization problem for building carbon emissions, and the Pareto optimal solution set is obtained. The Pareto optimal solution set is standardized and fitted with a continuous model to obtain a continuous Pareto front. Based on the continuous Pareto front, multidimensional feature data of each solution is extracted to obtain a multidimensional feature dataset of the solution set. The multidimensional feature data includes curvature, marginal sensitivity coefficient, and crowding degree; The Pareto optimal solution set is filtered based on the multidimensional feature dataset of the solution set to obtain the final solution for carbon emission optimization; The building design parameters are updated based on the final solution of carbon emission optimization to obtain the building with optimized carbon emissions.

2. The multi-objective optimization method for building carbon emissions based on a surrogate model according to claim 1, characterized in that, The preprocessing includes data cleaning, data standardization, and data partitioning.

3. The multi-objective optimization method for building carbon emissions based on a surrogate model according to claim 2, characterized in that, The building multi-objective prediction model adopts the physical information neural network (PINN) model, with building design parameters and environmental parameters as inputs, and carbon emission values, operation and maintenance cost values ​​and uncomfortable hours as outputs; The training of the multi-objective building prediction model based on preprocessed multi-dimensional building data includes: A composite loss function is constructed. The preprocessed multidimensional building data is used as training samples. The multi-objective prediction model of buildings is trained under supervision with the goal of minimizing the composite loss function until the model converges. The trained multi-objective prediction model of buildings is then obtained. The composite loss function includes data fitting loss, residual loss of building thermal physics equations, and boundary condition loss.

4. The multi-objective optimization method for building carbon emissions based on a surrogate model according to claim 3, characterized in that, The definition of the multi-objective optimization problem for building carbon emissions includes defining the objective function and constraints for the multi-objective optimization of building carbon emissions. The multi-objective optimization objectives for building carbon emissions include minimizing building carbon emissions, minimizing building operation and maintenance costs, and minimizing the number of uncomfortable hours indoors. The constraints include feasibility constraints on building design parameters, carbon emission budget constraints, and indoor comfort compliance constraints.

5. The multi-objective optimization method for building carbon emissions based on a surrogate model according to claim 4, characterized in that, The method of solving the multi-objective optimization problem of building carbon emissions by combining a multi-objective evolutionary algorithm with a trained multi-objective prediction model for buildings includes: The architectural design parameters are used as decision variables, and the feasible range of values ​​for the architectural design parameters constitutes the decision space. Under constraints, a multi-objective evolutionary algorithm is used to iteratively generate an optimal solution set in the decision space, and a trained multi-objective prediction model for buildings is used to predict the optimal solution set to obtain the target values ​​of the optimal solution set; the target values ​​of the optimal solution set include target values ​​for carbon emissions, operation and maintenance costs, and uncomfortable hours. Based on the objective value of the optimization scheme set, the optimization scheme set is sorted by non-dominated order and the constraint compliance is checked to select non-dominated solutions that meet the constraints. When the preset iteration termination condition is reached, all non-dominated solutions that satisfy the constraints are output, and the Pareto optimal solution set is obtained.

6. The multi-objective optimization method for building carbon emissions based on a surrogate model according to claim 5, characterized in that, The multidimensional feature data extracted for each solution based on the continuous Pareto front includes: Define a parameterized vector function, and calculate the curvature of each solution with respect to the continuous Pareto front by taking the derivative of the parameterized vector function; The marginal sensitivity coefficient of each solution is calculated based on the trained multi-objective prediction model for buildings. Calculate the congestion degree for each solution.

7. The multi-objective optimization method for building carbon emissions based on a surrogate model according to claim 6, characterized in that, The filtering of Pareto optimal solution sets based on the multidimensional feature dataset of solution sets includes: Define a curvature threshold, filter Pareto optimal solutions with curvature greater than the curvature threshold, and obtain a set of candidate solutions; Using curvature, marginal sensitivity coefficient and crowding degree as evaluation indicators, the entropy weight method is used to calculate the weight coefficient of each evaluation indicator. Based on the weight coefficient, the TOPSIS method is used to calculate the comprehensive proximity score of each solution in the candidate solution set, and the comprehensive proximity scores are sorted in descending order from high to low. The solution with the highest overall similarity score is selected as the final solution for carbon emission optimization.

8. A multi-objective optimization system for building carbon emissions based on a surrogate model, characterized in that, The steps of implementing the multi-objective optimization method for building carbon emissions based on the surrogate model as described in any one of claims 1 to 7 during execution include: The data acquisition and preprocessing module is used to collect multidimensional building data and preprocess it to obtain preprocessed multidimensional building data. The model building and training module is used to build a multi-objective prediction model for buildings and train the multi-objective prediction model for buildings based on preprocessed multi-dimensional building data to obtain the trained multi-objective prediction model for buildings. The multi-objective optimization module is used to define the multi-objective optimization problem of building carbon emissions. It uses a multi-objective evolutionary algorithm combined with a trained building multi-objective prediction model to solve the multi-objective optimization problem of building carbon emissions and obtain the Pareto optimal solution set. The solution set processing and feature extraction module is used to standardize and continuously fit the Pareto optimal solution set to obtain a continuous Pareto front; based on the continuous Pareto front, multidimensional feature data of each solution is extracted to obtain a multidimensional feature dataset of the solution set. The solution set filtering module is used to filter the Pareto optimal solution set based on the multidimensional feature dataset of the solution set to obtain the final solution for carbon emission optimization. The optimization execution module is used to update the building design parameters based on the final solution of carbon emission optimization, so as to obtain the building with optimized carbon emissions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the multi-objective optimization method for building carbon emissions based on the surrogate model as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-objective optimization method for building carbon emissions based on the surrogate model as described in any one of claims 1 to 7.

Images