CN117574785A

CN117574785A - Zero-carbon building multi-objective optimization method based on machine learning hybrid modeling

Info

Publication number: CN117574785A
Application number: CN202410060171.7A
Authority: CN
Inventors: 康一亭; 吕石磊; 张东杰; 崔钰
Original assignee: Tianjin University; China Academy of Building Research CABR
Current assignee: Tianjin University; China Academy of Building Research CABR
Priority date: 2024-01-16
Filing date: 2024-01-16
Publication date: 2024-02-20

Abstract

The invention provides a zero-carbon building multi-objective optimization method based on machine learning hybrid modeling, which relates to the technical field of computers and comprises the following steps: determining an optimization target and design parameters of the zero-carbon building, and constructing a multi-target optimization model of the zero-carbon building; building a geometric model of a building, and importing the geometric model into building energy consumption simulation software to simulate optimization targets under different design parameter conditions; adopting a PSO-SVM algorithm to construct a mapping relation between design parameters and an optimization target, and taking the mapping relation as a zero-carbon building multi-objective function proxy model; adopting NSGA-III algorithm to optimize zero-carbon building design parameters and optimization targets to obtain Pareto optimization results; according to the actual engineering condition, the design parameters and the optimization targets deviating from the actual engineering are removed from the Pareto optimization result, the passive design parameters and the optimization targets of the zero-carbon building are determined, the design of multi-target optimization of the zero-carbon building is realized, the optimization time is shortened, and the efficiency is improved.

Description

Zero-carbon building multi-objective optimization method based on machine learning hybrid modeling

Technical Field

The invention relates to the technical field of computers, in particular to a zero-carbon building multi-objective optimization method based on machine learning hybrid modeling.

Background

As one of three large energy utilization fields, the construction industry is related to the construction field by statistics of energy consumption and carbon emission of nearly 1/3 of the world. The international energy agency (International Energy Agency, IEA) proposes a development roadmap for each industry in the world 2020-2050, wherein all new buildings in the construction industry 2030 are proposed to realize near-zero carbon buildings, 50% of existing buildings in 2040 reach near-zero carbon buildings, and 85% of buildings in 2050 realize near-zero carbon buildings. Thus, for newly built buildings, how to achieve near zero carbon and zero carbon building goals is critical to future building development.

The traditional building design mainly takes architects as the leading, often takes planning conditions, building elevation and functional modeling as main factors to carry out preliminary design, and correspondingly cooperates with electromechanical professions on the basis. However, conventional building designs typically target building energy consumption to optimize, ignoring multiple goals of building carbon emissions, economy, and comfort.

Therefore, how to optimally design a zero-carbon building under multiple targets is a problem to be solved.

Disclosure of Invention

The invention provides a zero-carbon building multi-target optimization design method based on machine learning hybrid modeling, which is used for solving the problem of how to perform multi-target optimization design on a zero-carbon building under multiple targets.

The invention provides a zero-carbon building multi-objective optimization method based on machine learning hybrid modeling, which comprises the following steps:

step 1, determining an optimization target and design parameters of a zero-carbon building, and constructing a multi-target optimization model of the zero-carbon building; the multi-objective optimization model comprises an optimization objective, design parameters and constraint conditions, wherein the constraint conditions represent the value range of the design parameters; the optimization targets comprise the annual running carbon emission amount of a building unit area, the incremental cost of the unit area and the uncomfortable hours of the building year; the design parameters comprise an outer window heat transfer coefficient, a ground heat transfer coefficient, a roof heat transfer coefficient, an east wall window wall ratio, a west wall window wall ratio, a south wall window wall ratio, a north wall window wall ratio, an outer window sunshade coefficient, a wall heat transfer coefficient, a solar heat gain coefficient SHGC value, ventilation times and building orientation;

step 2, building a geometric model of the building according to the geometric dimension of the building and the building envelope construction parameters, and importing the geometric model into building energy consumption simulation software to simulate optimization targets under different design parameter conditions so as to obtain a data sample set; the data sample set comprises annual running carbon emission amount per unit area, incremental cost per unit area and uncomfortable hours per building year under the condition of different design parameters;

Step 3, constructing a mapping relation between design parameters and an optimization target by adopting a particle swarm optimization support vector machine (PSO-SVM) algorithm according to the data sample set, and taking the mapping relation as a zero-carbon building multi-objective function proxy model;

step 4, taking the agent model as an objective function of a NSGA-III algorithm of a third generation non-dominant ordering genetic algorithm, and adopting the NSGA-III algorithm to optimize zero-carbon building design parameters and an optimization target to obtain a Pareto optimization result; the Pareto optimization result comprises optimized design parameters and optimization targets corresponding to the optimized design parameters;

and 5, eliminating the design parameters and the optimization targets deviating from the engineering reality in the Pareto optimization result according to the engineering reality, and determining the passive design parameters and the optimization targets of the zero-carbon building.

According to the zero-carbon building multi-objective optimization method based on machine learning hybrid modeling, the calculation of the annual running carbon emission of the building unit area is represented by formulas (1) and (2):

（1）

（2）

wherein,E _i represent the firstiThe energy consumption of the building is similar to the annual energy consumption of the building,C _m carbon emission per unit building area (kg. CO) representing building operation stage ₂ /m ² ），EF _i Represent the first iThe carbon emission coefficient of the similar energy source,E _i,j represent the firstjClass system NoiThe energy consumption of the type of year,irepresents the energy consumption type of the building terminal, wherein the energy consumption type comprises electric power or fuel gas,ER _i,j represent the firstjClass systems consume class i annual energy provided by renewable energy systems,jrepresents the type of building energy consumption system, which includes air conditioning or lighting,C _p annual carbon reduction (kg. CO) representing building green land carbon sink system ₂ Y), y represents the number of years of operation,Arepresenting a building area;

the calculation of the zero-carbon building unit area increment cost is represented by a formula (3):

（3）

wherein,dCrepresenting the incremental cost per unit area (yuan/m) of a zero-carbon building ² ），Representing zero-carbon buildingkCost per unit area (yuan/m of design ² ），/>Reference building sheetBit area cost (meta/m) ² ) I.e. the cost of energy-saving materials paid by the minimum requirements of passive design;

the method for calculating the uncomfortable hours of the building year comprises the following steps: when the activity level is 1.0-1.3 metabolism equivalent met and the wind speed is less than 0.2m/s, the comfortable temperature area of the human body is 20-23.6 ℃ when the thermal resistance of the winter clothing is 1.0; the comfort temperature area of the human body is 23-26 ℃ when the thermal resistance of the clothing in summer is 1.0.

According to the zero-carbon building multi-objective optimization method based on machine learning hybrid modeling, the construction flow of the agent model comprises the following steps:

acquiring the data sample set, wherein the data sample set comprises a training set and a testing set;

training the agent model by using the training set to obtain a trained agent model;

testing the trained agent model by using the test set to obtain performance evaluation indexes of the test set; wherein the performance evaluation index comprises root mean square difference (RMSE), average absolute error (MAE) and R ² An index;

and evaluating the trained agent model based on the performance evaluation index to finally obtain the agent model.

According to the zero-carbon building multi-objective optimization method based on machine learning hybrid modeling, the mapping relation between design parameters and optimization targets is constructed by adopting a particle swarm optimization support vector machine (PSO-SVM) algorithm, and the mapping relation is used as a zero-carbon building multi-objective function proxy model, and the method comprises the following steps:

initializing a particle swarm, randomly generating a group of particles, wherein each particle represents a group of SVM parameters, and randomly distributing speed and position to each particle;

For each particle, performing model training by using a group of SVM parameters represented by the particle, calculating a performance index of the SVM by a performance evaluation method, and taking the performance index as the adaptability of the particle;

updating global optimum and individual optimum, taking the position of the particle with the maximum adaptability in the current particle swarm as global optimum, and taking the individual optimum position of each particle as the current position of the particle;

updating the speed, the position and the fitness of the particles by using an updating rule of a PSO algorithm, and repeating the process until the preset iteration times are reached or a stopping condition is met;

taking a group of SVM parameters represented by the final returned global optimal particles as model parameters;

and obtaining the proxy model based on the model parameters.

According to the zero-carbon building multi-objective optimization method based on machine learning hybrid modeling, the optimization flow of the NSGA-III algorithm comprises the following steps:

optimizing model super parameters including competition scale, population size, crossover and mutation probability and maximum evolution iteration number;

setting design parameters and ranges, and setting the proxy model as an adaptability function of an NSGA-III algorithm;

Randomly generating a first generation population, said population being a zero-carbon architectural design, each individual in said population representing a potential solution to a problem;

calculating the fitness function value of each individual according to the fitness function;

according to the fitness function value of each individual, non-dominant ordering is carried out on the individuals in the population, and the individuals are divided into a plurality of layers; wherein the first layer comprises individuals not subject to other individuals;

calculating a dominant set and a dominant number of times of each individual;

calculating the crowding degree among individuals in each hierarchy;

selecting a target number of individuals according to the non-dominant ranking and crowding degree, performing genetic operator operation on the selected individuals by using crossover and mutation operation, generating new individuals, merging the newly generated individuals into the current population to form a next generation population, and repeatedly executing the steps until a preset maximum evolution iteration number or convergence condition is met; preferably selecting an individual with a high non-dominant level, and selecting an individual with a low crowding level in the same level;

and taking the individuals in the final population as a group of Pareto front solutions found by the NSGA-III algorithm to obtain a Pareto optimization result.

According to the zero-carbon building multi-objective optimization method based on machine learning hybrid modeling, according to the actual engineering situation, the design parameters and optimization objectives deviating from the actual engineering are removed from the Pareto optimization result, and the passive design parameters and optimization objectives of the zero-carbon building are determined, including:

forming an ideal point by the optimal value of the optimization target at the Pareto front;

calculating the distance from the ideal point to each solution point in the Pareto optimization result by adopting a formula (4);

（4）

wherein,、/>、/>coordinate values for each solution point in the Pareto optimization result, +.>、/>、/>Coordinate values of ideal points;

and taking the solution point with the minimum distance as an optimal solution, and determining the passive design parameters and the optimization targets of the zero-carbon building based on the optimal solution.

The invention also provides a zero-carbon building multi-objective optimization device based on machine learning hybrid modeling, which comprises:

the optimization model module is used for determining an optimization target and design parameters of the zero-carbon building and constructing a multi-target optimization model of the zero-carbon building; the multi-objective optimization model comprises an optimization objective, design parameters and constraint conditions, wherein the constraint conditions represent the value range of the design parameters; the optimization targets comprise the annual running carbon emission amount of a building unit area, the incremental cost of the unit area and the uncomfortable hours of the building year; the design parameters comprise an outer window heat transfer coefficient, a ground heat transfer coefficient, a roof heat transfer coefficient, an east wall window wall ratio, a west wall window wall ratio, a south wall window wall ratio, a north wall window wall ratio, an outer window sunshade coefficient, a wall heat transfer coefficient, a solar heat gain coefficient SHGC value, ventilation times and building orientation;

The data sample set determining module is used for establishing a geometric model of a building according to the geometric dimension of the building and the building envelope construction parameters, and importing the geometric model into building energy consumption simulation software to simulate optimization targets under different design parameter conditions so as to obtain a data sample set; the data sample set comprises annual running carbon emission amount per unit area, incremental cost per unit area and uncomfortable hours per building year under the condition of different design parameters;

the agent model module is determined and used for constructing a mapping relation between design parameters and optimization targets by adopting a particle swarm optimization support vector machine (PSO-SVM) algorithm according to the data sample set, and taking the mapping relation as a zero-carbon building multi-objective function agent model;

the optimization result determining module is used for taking the agent model as an objective function of a NSGA-III algorithm of a third-generation non-dominant ranking genetic algorithm, and optimizing zero-carbon building design parameters and optimization targets by adopting the NSGA-III algorithm to obtain a Pareto optimization result; the Pareto optimization result comprises optimized design parameters and optimization targets corresponding to the optimized design parameters;

and the optimization module is used for eliminating the design parameters and the optimization targets deviating from the engineering reality in the Pareto optimization result according to the engineering reality, and determining the passive design parameters and the optimization targets of the zero-carbon building.

According to the zero-carbon building multi-objective optimization method based on machine learning hybrid modeling, the optimization objective and design parameters of the zero-carbon building are determined through the steps 1, and a zero-carbon building multi-objective optimization model is constructed; the multi-objective optimization model comprises an optimization objective, design parameters and constraint conditions, wherein the constraint conditions represent the value range of the design parameters; the optimization targets comprise the annual running carbon emission amount of a building unit area, the incremental cost of the unit area and the uncomfortable hours of the building year; the design parameters comprise an outer window heat transfer coefficient, a ground heat transfer coefficient, a roof heat transfer coefficient, an east wall window wall ratio, a west wall window wall ratio, a south wall window wall ratio, a north wall window wall ratio, an outer window sunshade coefficient, a wall heat transfer coefficient, a solar heat gain coefficient SHGC value, ventilation times and building orientation; step 2, building a geometric model of the building according to the geometric dimension of the building and the building envelope construction parameters, and importing the geometric model into building energy consumption simulation software to simulate optimization targets under different design parameter conditions so as to obtain a data sample set; the data sample set comprises annual running carbon emission amount per unit area, incremental cost per unit area and uncomfortable hours per building year under the condition of different design parameters; step 3, constructing a mapping relation between design parameters and an optimization target by adopting a particle swarm optimization support vector machine (PSO-SVM) algorithm according to the data sample set, and taking the mapping relation as a zero-carbon building multi-objective function proxy model; step 4, taking the agent model as an objective function of an NSGA-III algorithm, and adopting the NSGA-III algorithm to optimize zero-carbon building design parameters and an optimization target so as to obtain a Pareto optimization result; the Pareto optimization result comprises optimized design parameters and optimization targets corresponding to the optimized design parameters; and 5, eliminating the design parameters and the optimization targets deviating from the engineering reality in the Pareto optimization result according to the engineering reality, and determining the passive design parameters and the optimization targets of the zero-carbon building. Through the combination of the PSO-SVM model and the NSGA-III algorithm, the design of the multi-objective optimization of the zero-carbon building is realized, so that the multi-objective design process of the zero-carbon building is more scientific and standard, meanwhile, the call to energy plus software in the optimization process is reduced, the optimization time is shortened, and the collaborative optimization and design of the running carbon emission, the increment cost of unit area and the uncomfortable hours of the building are effectively improved.

Drawings

In order to more clearly illustrate the invention or the technical solutions of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the invention, and other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic flow diagram of a zero-carbon building multi-objective optimization method based on machine learning hybrid modeling provided by the invention;

FIG. 2 is a schematic illustration of a geometric model of a building provided by the present invention;

FIG. 3 is a schematic flow chart of the PSO-SVM algorithm provided by the present invention;

FIG. 4 is a schematic flow chart of the NSGA-III algorithm provided by the invention;

FIG. 5 is a schematic diagram of a Pareto front solution provided by the present invention;

FIG. 6 is a comparative schematic of HV measure values for a zero carbon building provided by the present invention;

FIG. 7 is a second schematic flow chart of the zero-carbon building multi-objective optimization method based on machine learning hybrid modeling provided by the invention;

fig. 8 is a schematic structural diagram of a zero-carbon building multi-objective optimization device based on machine learning hybrid modeling.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The machine learning hybrid modeling-based zero-carbon building multi-objective optimization method of the present invention is described below with reference to fig. 1-7.

FIG. 1 is a schematic flow chart of a zero-carbon building multi-objective optimization method based on machine learning hybrid modeling, which is provided by the invention, and as shown in FIG. 1, the method comprises steps 101-105; wherein,

step 101, determining an optimization target and design parameters of a zero-carbon building, and constructing a multi-target optimization model of the zero-carbon building; the multi-objective optimization model comprises an optimization objective, design parameters and constraint conditions, wherein the constraint conditions represent the value range of the design parameters; the optimization targets comprise the annual running carbon emission amount of a building unit area, the incremental cost of the unit area and the uncomfortable hours of the building year; the design parameters comprise an outer window heat transfer coefficient, a ground heat transfer coefficient, a roof heat transfer coefficient, an east wall window wall ratio, a west wall window wall ratio, a south wall window wall ratio, a north wall window wall ratio, an outer window sunshade coefficient, a wall heat transfer coefficient, a solar heat gain coefficient SHGC value, ventilation times and building orientation.

Specifically, the optimization targets include the annual operating carbon emissions per unit area of the building, the incremental cost per unit area, and the number of uncomfortable hours per year of the building; the calculation of the annual running carbon emission of the building unit area is represented by formulas (1) and (2):

（1）

（2）

wherein,E _i represent the firstiThe energy consumption of the building is similar to the annual energy consumption of the building,C _m carbon emission per unit building area (kg. CO) representing building operation stage ₂ /m ² ），EF _i Represent the firstiThe carbon emission coefficient of the similar energy source,E _i,j represent the firstjClass system NoiThe energy consumption of the type of year,irepresents the energy consumption type of the building terminal, wherein the energy consumption type comprises electric power or fuel gas,ER _i,j represent the firstjClass systems consume class i annual energy provided by renewable energy systems,jrepresenting the type of building energy consumption system, and the type of energy consumption systemIncluding an air conditioner or a lighting device,C _p annual carbon reduction (kg. CO) representing building green land carbon sink system ₂ Y), y represents the number of years of operation,Arepresenting building area (m) ² ）；

The incremental cost of the unit area of the zero-carbon building mainly occurs in the aspects of energy-saving materials and energy-saving technology selection, so that the initial investment increment limit caused by the energy-saving technology is calculated to measure the economical efficiency. It should be noted that, the energy-saving material economical index such as the building envelope heat preservation and the outer window only considers the cost increased by changing the energy-saving component, and does not consider the building basic component cost, and the initial investment is calculated by adopting the following formula (5):

（5）

Wherein,C _I representing initial investment costs (yuan/m ² ），C _w Represents the price (yuan/m) of the external wall insulation board ³ ），V _w Represents the volume (m) ³ ），C _r Representing the price (Yuan/m) of the roof insulation board ³ ），V _r Representing the volume (m) of a roof insulation board ³ ），C _g Representing the price (Yuan/m) of the ground insulation board ³ ），V _g Represents the volume (m) of the ground heat-insulating board ³ ），C _wd Indicating the price (Yuan/m) of energy-saving door and window ² ），A _wd Indicating the area (m) of energy-saving door and window ² ），C _oi Representing other initial investment costs (primary) such as fresh air heat recovery, renewable energy systems, airtightness, etc.

The calculation of the incremental cost of the unit area of the zero-carbon building is represented by the formula (3):

（3）

wherein,dCrepresenting the incremental cost per unit area (yuan/m) of a zero-carbon building ² ），Representing zero-carbon buildingkCost per unit area (yuan/m of design ² ），/>Representing the cost per unit area of a reference building (yuan/m) ² ) I.e. the cost of energy-saving materials paid by the minimum requirements of passive design;

the method for calculating the uncomfortable hours of the building year is as follows: the definition of the indoor thermal comfort zone is derived from the standard of the society of heating, refrigeration and air conditioning, namely, the comfort temperature zone of a human body when the thermal resistance of winter clothing is 1.0 heat resistance is 20.0-23.6 ℃ under the conditions that the activity level is 1.0-1.3 metabolic equivalent met and the wind speed is less than 0.2 m/s; the comfort temperature area of the human body is 23-26 ℃ when the thermal resistance of the clothing in summer is 1.0.

Design parameters include the heat transfer coefficient of the external windowx ₁ ) Heat transfer coefficient of groundx ₂ ) Heat transfer coefficient of roofx ₃ ) Window wall ratio of east wallx ₄ ) Wall-to-wall ratio of western wallx ₅ ) South wall window wall ratio%x ₆ ) Window wall ratio of north wallx ₇ ) Sunshade coefficient of external windowx ₈ ) Heat transfer coefficient of wallx ₉ ) SHGC value%x ₁₀ ) The ventilation times arex ₁₁ ) And the building orientation isx ₁₂ ) Etc. 12 items.

The constraint condition represents the value range of the design parameters, which is mainly limited by the building design standard to be satisfied, and the value range of the design parameters is mainly determined by the building technical standard of near zero energy consumption and the general specification of building energy conservation and renewable energy utilization, and the calculation formula is represented as follows:

（6）

wherein,x _i the parameters of the building design are represented by,z _imax andz _imin the upper and lower limits of the design parameters of the passive zero-carbon building are represented, respectively.

The upper and lower limits of the design parameters of the zero-carbon building are respectively as follows:

the heat transfer coefficient of the outer window is less than or equal to 1.0x ₁ ≤1.5 W/m ² K) and the ground heat transfer coefficient (0.25 +.x ₂ ≤0.4W/m ² K) and heat transfer coefficient of roof (0.1 +.x ₃ ≤0.3W/m ² K) and east wall window wall ratio (0.2 +.x ₄ Less than or equal to 0.4), western wall window wall ratio (less than or equal to 0.2)x ₅ Less than or equal to 0.4), and a south wall window wall ratio (less than or equal to 0.2)x ₆ Less than or equal to 0.5) and a north wall window wall ratio (less than or equal to 0.2)x ₇ Less than or equal to 0.5), and an outer window sunshade coefficient (200 less than or equal to x ₈ Less than or equal to 1000), and the heat transfer coefficient of the wall body (0.1 less than or equal tox ₉ ≤0.3W/m ² K), solar heat gain coefficient (Solar Heat Gain Coefficient, SHGC) value (0.45.ltoreq.x ₁₀ Less than or equal to 0.52), and ventilation times (less than or equal to 0.4)x ₁₁ Less than or equal to 1.0 times/hour) and building orientation (0 less than or equal tox ₁₂ Not more than 360 degrees).

102, building a geometric model of a building according to the geometric dimension of the building and the building envelope construction parameters, and importing the geometric model into building energy consumption simulation software to simulate optimization targets under different design parameter conditions so as to obtain a data sample set; the data sample set includes annual operating carbon emissions per unit area, incremental cost per unit area, and number of annual uncomfortable hours of construction under different design parameter conditions.

Specifically, the step 102 specifically includes the construction of a zero-carbon building design parameter sample matrix, the construction of a building geometric model and the construction of a building optimization target matrix; wherein,

the construction of the zero-carbon building optimization parameter sample matrix comprises the steps of determining an optimization parameter distribution form, determining a sampling method of a sample, determining the size of a sample capacity, executing a sampling process and forming a matrix. The sampling method of the sample is SOBOL sampling by using Simlab software. And adopting SOBOL sampling 1300 group design schemes, wherein design parameters and optimization targets are in one-to-one correspondence, and calculating the annual running carbon emission amount, the incremental cost of unit area and the uncomfortable hours of the building under different schemes.

The constructed design parameter sample matrix is shown as follows:

（7）

wherein,m、nthe design parameter number and the number of samples, respectively.

Construction of a building geometric model in step 102, building a geometric model of a building in skchup, and fig. 2 is a schematic diagram of the geometric model of the building provided by the present invention. And saving the geometric model as a file format which can be identified by energy plus of the building, namely IDF, setting boundary conditions in the energy plus software, wherein the boundary conditions mainly comprise non-optimized parameter information such as building places, climate conditions of the places, personnel room rate, illumination power density and the like, importing IDF files and meteorological data in the energy plus software, inputting optimized parameters, and calculating an optimized target value. The meteorological data is CSWD data of a standard meteorological database downloaded from an energy plus official network in Beijing area.

Building envelope construction parameters include wall construction, roof construction, floor construction, lighting power density, lighting schedule, equipment density, number of people, activity level, personnel occupancy, refrigeration control, heating ventilation and air conditioning (Heating Ventilation and Air Conditioning, HVAC) systems of the building. Table 1 is a partial building envelope configuration parameter as shown in table 1.

TABLE 1 partial building envelope construction parameters

The design parameter sample matrix is imported into the JeGlus+energy plus to calculate a target value corresponding to the JeGlus+energy plus, and an optimization target matrix is formed, wherein the optimization target matrix is shown in the following formula:

（8）

wherein,y _1n 、y _2n andy _3n the carbon emission amount, the incremental cost and the uncomfortable hours of the building per unit area are respectively calculated.

And step 103, constructing a mapping relation between design parameters and an optimization target by adopting a particle swarm optimization support vector machine (PSO-SVM) algorithm according to the data sample set, and taking the mapping relation as a zero-carbon building multi-objective function proxy model.

Specifically, the zero-carbon building multi-objective function proxy model comprises a proxy model of the annual running carbon emission amount of a building unit area, the incremental cost of the unit area and the uncomfortable hours of the building year; wherein F is _carbon (x) Is a functional proxy model of annual carbon emission of building unit area, F _time (x) Is a model of a yearly uncomfortable hour number function proxy, F _cost (x) Is a cost function proxy model per unit area increment.

The construction flow of the agent model comprises the following steps:

(1) The data sample set is obtained, and the data sample set comprises a training set and a testing set.

Specifically, the data sample set consisting of the optimized target data samples is segmented according to the proportion of 70% training set and 30% test set, namely 910 groups of data are used for training a model, and 390 groups of data are used for testing the validity of a verification model. And 5-fold cross validation is adopted on the sample data, namely the data sample set is equally divided into 5 groups, 1 group is taken as a test sample each time, and the rest 4 groups are taken as training samples.

And training the proxy model by using the training set to obtain a trained proxy model.

Specifically, training the proxy model by using a training set to obtain a trained proxy model, namely training the PSO-SVM neural network model to obtain a trained PSO-SVM neural network model.

The training process of the PSO-SVM neural network model comprises the following steps:

initializing a particle swarm, randomly generating a group of particles, wherein each particle represents a group of SVM parameters, and randomly distributing speed and position to each particle; wherein, SVM parameters are penalty factor C and kernel function parameters.

For each particle, performing model training by using a group of SVM parameters represented by the particle, calculating a performance index of the SVM by a performance evaluation method, and taking the performance index as the adaptability of the particle; the performance indexes are accuracy, precision, recall rate and the like.

And updating global optimum and individual optimum, taking the position of the particle with the maximum adaptability in the current particle group as global optimum, and taking the individual optimum position of each particle as the current position of the particle.

The speed, position and fitness of the particles are updated using the update rules of the PSO algorithm, and the above process is repeated until a predetermined number of iterations is reached or a stop condition is met.

And taking a group of SVM parameters represented by the final returned global optimal particles as model parameters.

And obtaining the proxy model based on the model parameters.

Optionally, a group of SVM parameters represented by global optimal particles is used as model parameters, and the SVM model can be retrained to obtain a final proxy model, namely a final PSO-SVM neural network model.

(2) Testing the trained agent model by using the test set to obtain performance evaluation indexes of the test set; wherein the performance evaluation index comprises root mean square difference (RMSE), average absolute error (MAE) and R ² The index and the performance index are expressed by the following specific formulas:

（9）

（10）

（11）

wherein,representing the i-th optimization objective simulation value,f _i representing the i-th optimization objective predictor,nrepresenting the total number of samples->Representing the average value of the optimization target simulation values.

(3) And evaluating the trained agent model based on the performance evaluation index to finally obtain the agent model.

Specifically, the trained agent model is evaluated according to the performance evaluation index, the agent model is finally obtained, and the agent model of the annual running carbon emission of the unit area, the incremental cost of the unit area and the uncomfortable hours of the building are respectively built.

For example, a proxy model is established by using a PSO-SVM algorithm, i.e. particle swarm PSO is adopted to perform super-parameter optimization on the SVM algorithm. In the super-parameters, the number of particles in the particle swarm of PSO is 50, the maximum particle swarm velocity is 0.6, the acceleration coefficients are 1.5 and 1.7 respectively, the maximum iteration number is 200, and the fitness function is Root Mean Square Error (RMSE). The kernel function of the SVM is set to a radial basis function (RBF kernel), the hyper-parameters of the kernel function are 10, the regularization factor C is 4, the loss factor is 1, and the cross-validation is 3.

Rmse=0.039, r for annual operating carbon emissions per building unit area ² =0.977, mae=0.030; rmse=0.023, r for incremental cost per unit area ² =0.930, mae=0.060; rmse=0.065, r for the number of uncomfortable hours of construction year ² =0.903, mae=0.048, and the accuracy of the prediction result of the established PSO-SVM model satisfies the requirement.

FIG. 3 is a schematic flow chart of the PSO-SVM algorithm provided by the invention, as shown in FIG. 3, comprising:

step 301, selection of training samples and test samples. A data sample set is obtained, the data sample set comprising a training set and a test set.

Step 302, initializing a particle swarm, penalty factor C and kernel function parameter selection, speed and position of each particle.

Step 303, calculate fitness of each particle. For each particle, model training is carried out by using a group of SVM parameters represented by the particle, and the performance index of the SVM is calculated by a performance evaluation method and is taken as the fitness of the particle.

Step 304, a global optimum and an individual optimum are determined. And taking the position of the particle with the maximum adaptability in the current particle swarm as global optimum, and taking the individual optimum position of each particle as the current position of the particle.

Step 305, updating the speed, position and fitness of the particles, repeating the above process until a predetermined number of iterations is reached or a stop condition is met.

Step 306, test sample verification. Testing the trained PSO-SVM model by using a test set to obtain performance evaluation indexes of the test set; wherein the performance evaluation index comprises root mean square difference (RMSE), average absolute error (MAE) and R ² And (5) an index.

Step 307, determining whether the model meets the accuracy. If the model meets the accuracy, go to step 308; in the case where the model does not meet the accuracy, the process proceeds to step 302.

Step 308, PSO-SVM model. Taking a group of SVM parameters represented by the final returned global optimal particles as model parameters; and obtaining a PSO-SVM model based on the model parameters.

104, taking the agent model as an objective function of a third generation Non-dominant order genetic algorithm (Non-dominated Sorting Genetic Algorithm III, NSGA III), and adopting the NSGA-III algorithm to optimize zero-carbon building design parameters and an optimization target to obtain a Pareto optimization result; the Pareto optimization result comprises an optimized design parameter and an optimization target corresponding to the optimized design parameter.

Specifically, the proxy model is used as an objective function of the NSGA-iii algorithm, and the objective function is used to calculate the fitness function value.

The optimization flow of the NSGA-III algorithm comprises the following steps:

the optimization model super-parameters comprise competition scale, population size, crossover and mutation probability and maximum evolution iteration number. The super parameters of NSGA-III algorithm are set to include competition scale, population size, chromosome size, maximum evolution iteration number, crossover probability and mutation probability. And compiling an NSGA-III multi-objective optimization algorithm program through a MATLAB platform. The basic parameters set the population size as 50, the maximum evolutionary iteration number as 100, the crossover probability as 0.8 and the variation probability as 0.1, and the population initialization is carried out.

Setting design parameters and ranges, and setting the proxy model as an adaptability function of an NSGA-III algorithm. Setting maximum and minimum values of design parameters and iteration termination rules, and setting the proxy model as a fitness function of an NSGA-III algorithm. And calling an optimized target agent model established by the PSO-SVM algorithm as an fitness function.

A first generation population is randomly generated, the population being a zero-carbon building design, each individual in the population representing one potential solution to the problem. Each individual contains decision variables and objective function values for the problem, i.e., design parameters and optimization objectives.

And calculating the fitness function value of each individual according to the fitness function.

According to the fitness function value of each individual, non-dominant ordering is carried out on the individuals in the population, and the individuals are divided into a plurality of layers; wherein the first layer comprises individuals not subject to other individuals.

Calculating the count of each individual album and dominant times.

The degree of congestion between individuals in each hierarchy is calculated.

Selecting a target number of individuals according to the non-dominant ranking and crowding degree, performing genetic operator operation on the selected individuals by using crossover and mutation operation, generating new individuals, merging the newly generated individuals into the current population to form a next generation population, and repeatedly executing the steps until a preset maximum evolution iteration number or convergence condition is met; among them, it is preferable to select individuals with a high non-dominant hierarchy and to select individuals with a low degree of congestion in the same hierarchy.

The target value of the optimization target for each scheme is evaluated, and the optimization target is stopped until the maximum evolution iteration number or convergence condition (average change is less than 0.00001) is met through iterative operation through selection, crossover and mutation.

And taking the individuals in the final population as a group of Pareto front solutions found by the NSGA-III algorithm to obtain a Pareto optimization result. The Pareto non-dominant solution after the iteration is completed is assembled to comprise 70 groups of design parameters and optimization targets.

Fig. 4 is a schematic flow chart of NSGA-III algorithm provided in the present invention, as shown in fig. 4, including:

in step 401, a total reference point is calculated.

Step 402, an initial population is generated. A first generation population is randomly generated, the population being a zero-carbon architectural design, each individual in the population representing one potential solution to the problem. Setting super parameters of an optimization model, setting design parameters, a range and a fitness function.

Step 403, non-dominant ordering and non-dominant solution set. Calculating the fitness function value of each individual according to the fitness function; according to the fitness function value of each individual, non-dominant ordering is carried out on the individuals in the population, and the individuals are divided into a plurality of layers; wherein the first layer comprises individuals not subject to other individuals; calculating a dominant set and a dominant number of times of each individual; the degree of congestion between individuals in each hierarchy is calculated.

Step 404, the competitive game selects an individual. A target number of individuals is selected based on the non-dominant ranking and the degree of congestion.

Step 405, crossover and mutation. And performing genetic operator operation on the selected individuals by using crossover and mutation operation according to the selected target number of individuals to generate new individuals.

At step 406, a population of offspring is generated. And combining the newly generated individuals into the current population to form a next generation population.

Step 407, normalize Pareto non-dominant solution set.

Step 408, a next generation is determined based on the reference point.

Step 409, next generation obtained solution.

Step 410, it is determined whether the maximum number of evolutionary iterations is satisfied. In case the maximum number of evolutions is satisfied, go to step 411; if the maximum number of evolution iterations is not satisfied, go to step 404.

And step 411, outputting a Pareto optimization result. And taking the individuals in the final population as a group of Pareto front solutions found by an NSGA-III algorithm to obtain a Pareto optimization result.

The NSGA-III algorithm is adopted to optimize zero-carbon building design parameters and optimization targets, the obtained Pareto optimization result is a group of Pareto front solution, and FIG. 5 is a schematic diagram of the Pareto front solution provided by the invention. The set of solutions is non-dominant under multiple objective functions, i.e., none of the objective functions' solutions are superior to it, and not inferior to it under other objective functions.

And 105, eliminating the design parameters and the optimization targets deviating from the engineering reality in the Pareto optimization result according to the engineering reality, and determining the passive design parameters and the optimization targets of the zero-carbon building.

Specifically, after the NSGA-III algorithm is calculated, a Pareto optimization result obtained by the algorithm is output, and design parameters and optimization targets deviating from engineering reality in the Pareto optimization result are removed according to engineering reality, so that the passive design parameters and optimization targets of the zero-carbon building are finally determined.

（4）

Searching a solution point with the minimum distance by calculating the distance between each solution point and the ideal point, taking the solution point with the minimum distance as an optimal solution, and adopting the following formula:

（12）

according to different engineering actual conditions and decision bases, 4 working conditions are adopted, namely the minimum carbon emission amount of annual operation of unit area of the building, the minimum incremental cost of unit area and the minimum uncomfortable hours of the building are calculated, and the calculated results are shown in table 2.

TABLE 2 scheme solutions for 4 conditions

Optionally, the advantages and disadvantages of the NSGA-iii algorithm are evaluated by using an ultra-volume (HV) measure, and fig. 6 is a schematic diagram showing the comparison of HV measure values of the zero-carbon building provided by the present invention, and as shown in fig. 6, the HV value of the multi-objective optimization method of the zero-carbon building is determined by using a conventional NSGA-II algorithm. The HV measure may be used to evaluate the uniformity and convergence of the distribution of a set of optimal solutions, and if a set of Pareto optimal solutions, the convergence of the uniformity of the distribution will be better, and the HV value will be increased. As can be derived from fig. 6, the HV measure value of the zero-carbon building multi-objective optimization method provided by the invention is higher than the HV measure determined based on NSGA-II algorithm. Therefore, the method provided by the invention can obtain the solution set with better convergence and uniform distribution.

Based on the description, the zero-carbon building multi-objective optimization method based on machine learning hybrid modeling constructs a zero-carbon building multi-objective optimization model by determining the optimization objective and design parameters of the zero-carbon building; the multi-objective optimization model comprises an optimization objective, design parameters and constraint conditions, wherein the constraint conditions represent the value range of the design parameters; the optimization targets comprise the annual running carbon emission amount of a building unit area, the incremental cost of the unit area and the uncomfortable hours of the building year; the design parameters comprise an outer window heat transfer coefficient, a ground heat transfer coefficient, a roof heat transfer coefficient, an east wall window wall ratio, a west wall window wall ratio, a south wall window wall ratio, a north wall window wall ratio, an outer window sunshade coefficient, a wall heat transfer coefficient, a solar heat gain coefficient SHGC value, ventilation times and building orientation; building a geometric model of a building according to the geometric dimension of the building and the building envelope construction parameters, and importing the geometric model into building energy consumption simulation software to simulate optimization targets under different design parameter conditions so as to obtain a data sample set; the data sample set comprises annual running carbon emission amount per unit area, incremental cost per unit area and uncomfortable hours per building year under the condition of different design parameters; according to the data sample set, adopting a PSO-SVM algorithm to construct a mapping relation between design parameters and an optimization target, and taking the mapping relation as a zero-carbon building multi-objective function proxy model; taking the agent model as an objective function of an NSGA-III algorithm, and adopting the NSGA-III algorithm to optimize zero-carbon building design parameters and an optimization target to obtain a Pareto optimization result; the Pareto optimization result comprises optimized design parameters and optimization targets corresponding to the optimized design parameters; and removing design parameters and optimization targets deviating from engineering reality from the Pareto optimization result according to engineering reality, and determining passive design parameters and optimization targets of the zero-carbon building. Through the combination of the PSO-SVM model and the NSGA-III algorithm, the design of the multi-objective optimization of the zero-carbon building is realized, so that the multi-objective design process of the zero-carbon building is more scientific and standard, meanwhile, the call to energy plus software in the optimization process is reduced, the optimization time is shortened, and the collaborative optimization and design of the running carbon emission, the increment cost of unit area and the uncomfortable hours of the building are effectively improved.

Fig. 7 is a second flow chart of a zero-carbon building multi-objective optimization method based on machine learning hybrid modeling, as shown in fig. 7, including:

stage one, model establishment and data generation.

And (5) building a passive physical model. Building a geometric model of the building in SketchUp according to the geometric dimension of the building and the building envelope construction parameters, and storing the geometric model as a file format 'IDF' which can be identified by building energy consumption simulation software energy plus.

Passive building parameter variables. Determining an optimization target and design parameters of a zero-carbon building, wherein the optimization target comprises the annual running carbon emission amount of a building unit area, the incremental cost of the unit area and the uncomfortable hours of the building year; the design parameters comprise 12 items of an outer window heat transfer coefficient, a ground heat transfer coefficient, a roof heat transfer coefficient, an east wall window wall ratio, a west wall window wall ratio, a south wall window wall ratio, a north wall window wall ratio, an outer window sunshade coefficient, a wall heat transfer coefficient, a solar heat gain coefficient SHGC value, ventilation times, building orientation and the like.

SOBOL sampling. An IDF file and meteorological data are imported into energy plus software, design parameters are input, and optimization targets are calculated, wherein the distribution form of the optimization parameters is determined, the sampling method of samples is determined, the size of sample capacity is determined, sampling process is performed, and a matrix is formed. The sampling method of the sample is SOBOL sampling by Simlab software. And adopting SOBOL sampling 1300 group design schemes, wherein design parameters and optimization targets are in one-to-one correspondence, and calculating the annual running carbon emission amount, the incremental cost of unit area and the uncomfortable hours of the building under different schemes.

And (5) establishing a simulation sample. And (3) importing the constructed design parameter sample matrix into the JeGlus+energy plus to calculate a target value corresponding to the design parameter sample matrix to form an optimization target matrix.

And step two, data prediction and verification.

PSO-SVM algorithm. And constructing a mapping relation between design parameters and an optimization target by adopting a PSO-SVM algorithm on the MATLAB platform according to the established simulation sample. Including model training, model testing, and model evaluation.

And (5) model training. And training the PSO-SVM model by using the training set to obtain a trained PSO-SVM model.

And (5) model testing. Testing the trained PSO-SVM model by using a test set to obtain performance evaluation indexes of the test set; wherein the performance evaluation index comprises root mean square difference (RMSE), average absolute error (MAE) and R ² And (5) an index.

And (5) evaluating a model. And evaluating the trained PSO-SVM model based on the performance evaluation index to finally obtain the PSO-SVM model, and taking the PSO-SVM model as a zero-carbon building multi-objective function proxy model.

And step three, multi-objective optimization and optimal solution generation.

Multi-objective establishment. Optimization objectives are determined from carbon emissions, economy and thermal comfort, including annual operating carbon emissions per unit area of the building, incremental cost per unit area, number of uncomfortable hours per year of the building.

NSGA-III optimization. Taking the agent model as an objective function of an NSGA-III algorithm, and adopting the NSGA-III algorithm to optimize zero-carbon building design parameters and an optimization target on a MATLAB platform to obtain a Pareto optimal solution set, namely a Pareto optimization result; the Pareto optimization result comprises the optimized design parameters and optimization targets corresponding to the optimized design parameters.

Actual engineering experience. And removing design parameters and optimization targets deviating from engineering reality from the Pareto optimization result according to engineering reality, and determining passive design parameters and optimization targets of the zero-carbon building.

The zero-carbon building multi-target optimization device based on the machine learning mixed modeling provided by the invention is described below, and the zero-carbon building multi-target optimization device based on the machine learning mixed modeling described below and the zero-carbon building multi-target optimization method based on the machine learning mixed modeling described above can be correspondingly referred to each other.

Fig. 8 is a schematic structural diagram of a zero-carbon building multi-objective optimization device based on machine learning hybrid modeling according to the present invention, and as shown in fig. 8, a zero-carbon building multi-objective optimization device 800 based on machine learning hybrid modeling includes: a module 801 for constructing an optimization model, a module 802 for determining a data sample set, a module 803 for determining a proxy model, a module 804 for determining an optimization result and an optimization module 805; wherein,

The construction optimization model module 801 is used for determining an optimization target and design parameters of the zero-carbon building and constructing a multi-target optimization model of the zero-carbon building; the multi-objective optimization model comprises an optimization objective, design parameters and constraint conditions, wherein the constraint conditions represent the value range of the design parameters; the optimization targets comprise the annual running carbon emission amount of a building unit area, the incremental cost of the unit area and the uncomfortable hours of the building year; the design parameters comprise an outer window heat transfer coefficient, a ground heat transfer coefficient, a roof heat transfer coefficient, an east wall window wall ratio, a west wall window wall ratio, a south wall window wall ratio, a north wall window wall ratio, an outer window sunshade coefficient, a wall heat transfer coefficient, a solar heat gain coefficient SHGC value, ventilation times and building orientation;

the data sample set determining module 802 is configured to establish a geometric model of a building according to the geometric dimension of the building and the building envelope construction parameters, and import the geometric model into building energy consumption simulation software to simulate optimization targets under different design parameter conditions, so as to obtain a data sample set; the data sample set comprises annual running carbon emission amount per unit area, incremental cost per unit area and uncomfortable hours per building year under the condition of different design parameters;

The agent model determining module 803 is configured to construct a mapping relationship between design parameters and optimization targets by using a particle swarm optimization support vector machine (PSO-SVM) algorithm according to a data sample set, and take the mapping relationship as a zero-carbon building multi-objective function agent model;

the optimization result determining module 804 is configured to use the proxy model as an objective function of a NSGA-iii algorithm of a third generation non-dominant ranking genetic algorithm, and optimize zero-carbon building design parameters and an optimization target by using the NSGA-iii algorithm to obtain a Pareto optimization result; the Pareto optimization result comprises optimized design parameters and optimization targets corresponding to the optimized design parameters;

and the optimization module 805 is configured to reject the design parameters and the optimization targets deviating from the engineering reality in the Pareto optimization result according to the engineering reality, and determine the passive design parameters and the optimization targets of the zero-carbon building.

According to the zero-carbon building multi-objective optimization device based on machine learning hybrid modeling, an optimization objective and design parameters of a zero-carbon building are determined, and a zero-carbon building multi-objective optimization model is constructed; the multi-objective optimization model comprises an optimization objective, design parameters and constraint conditions, wherein the constraint conditions represent the value range of the design parameters; the optimization targets comprise the annual running carbon emission amount of a building unit area, the incremental cost of the unit area and the uncomfortable hours of the building year; the design parameters comprise an outer window heat transfer coefficient, a ground heat transfer coefficient, a roof heat transfer coefficient, an east wall window wall ratio, a west wall window wall ratio, a south wall window wall ratio, a north wall window wall ratio, an outer window sunshade coefficient, a wall heat transfer coefficient, a solar heat gain coefficient SHGC value, ventilation times and building orientation; building a geometric model of a building according to the geometric dimension of the building and the building envelope construction parameters, and importing the geometric model into building energy consumption simulation software to simulate optimization targets under different design parameter conditions so as to obtain a data sample set; the data sample set comprises annual running carbon emission amount per unit area, incremental cost per unit area and uncomfortable hours per building year under the condition of different design parameters; according to the data sample set, adopting a PSO-SVM algorithm to construct a mapping relation between design parameters and an optimization target, and taking the mapping relation as a zero-carbon building multi-objective function proxy model; taking the agent model as an objective function of an NSGA-III algorithm, and adopting the NSGA-III algorithm to optimize zero-carbon building design parameters and an optimization target to obtain a Pareto optimization result; the Pareto optimization result comprises optimized design parameters and optimization targets corresponding to the optimized design parameters; and removing design parameters and optimization targets deviating from engineering reality from the Pareto optimization result according to engineering reality, and determining passive design parameters and optimization targets of the zero-carbon building. Through the combination of the PSO-SVM model and the NSGA-III algorithm, the design of the multi-objective optimization of the zero-carbon building is realized, so that the multi-objective design process of the zero-carbon building is more scientific and standard, meanwhile, the call to energy plus software in the optimization process is reduced, the optimization time is shortened, and the collaborative optimization and design of the running carbon emission, the increment cost of unit area and the uncomfortable hours of the building are effectively improved.

Alternatively, the calculation of annual operating carbon emission of the building unit area is expressed by formulas (1) and (2)

（1）

（2）

Wherein,E _i represent the firstiThe energy consumption of the building is similar to the annual energy consumption of the building,C _m carbon emission per unit building area (kg. CO) representing building operation stage ₂ /m ² ），EF _i Represent the firstiThe carbon emission coefficient of the similar energy source,E _i,j represent the firstjClass system NoiThe energy consumption of the type of year,irepresents the energy consumption type of the building terminal, wherein the energy consumption type comprises electric power or fuel gas,ER _i,j represent the firstjClass systems consume class i annual energy provided by renewable energy systems,jrepresents the type of building energy consumption system, which includes air conditioning or lighting,C _p annual carbon reduction (kg. CO) representing building green land carbon sink system ₂ Y), y represents the number of years of operation,Arepresenting building area (m) ² ）；

（3）

Optionally, the construction flow of the proxy model includes the following steps:

Optionally, the determining optimization result module 804 is specifically configured to:

and obtaining the proxy model based on the model parameters.

Optionally, the optimization flow of the NSGA-iii algorithm includes the following steps:

calculating a dominant set and a dominant number of times of each individual;

calculating the crowding degree among individuals in each hierarchy;

Optionally, the optimizing module 805 is specifically configured to:

（4）

The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.

From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.

Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A zero-carbon building multi-objective optimization method based on machine learning hybrid modeling is characterized by comprising the following steps:

2. The machine learning hybrid modeling-based zero-carbon building multi-objective optimization method according to claim 1, wherein the calculation of annual operating carbon emission per building unit area is represented by formulas (1) and (2):

（1）

（2）

（3）

wherein,dCrepresenting the incremental cost per unit area (yuan/m) of a zero-carbon building ² ），Representing zero-carbon buildingkCost per unit area of the individual design, +.>Representing the cost per unit area of a reference building (yuan/m) ² ) I.e. the cost of energy-saving materials paid by the minimum requirements of passive design;

3. The machine learning hybrid modeling-based zero-carbon building multi-objective optimization method according to claim 1, wherein the construction flow of the proxy model comprises the following steps:

4. The machine learning hybrid modeling-based zero-carbon building multi-objective optimization method according to claim 3, wherein the constructing a mapping relation between design parameters and optimization objectives by using a particle swarm optimization support vector machine (PSO-SVM) algorithm, and using the mapping relation as a zero-carbon building multi-objective function proxy model comprises:

and obtaining the proxy model based on the model parameters.

5. The machine learning hybrid modeling-based zero-carbon building multi-objective optimization method according to claim 1, wherein the optimization flow of the NSGA-iii algorithm comprises the following steps:

calculating a dominant set and a dominant number of times of each individual;

calculating the crowding degree among individuals in each hierarchy;

6. The machine learning hybrid modeling-based zero-carbon building multi-objective optimization method according to claim 1, wherein the step of eliminating the design parameters and the optimization targets deviating from the engineering reality from the Pareto optimization result according to the engineering reality, and determining the passive design parameters and the optimization targets of the zero-carbon building comprises the following steps:

（4）