Prediction and control method of building carbon emissions based on machine learning and multi-objective optimization

Abstract

The present invention discloses a method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization. The method comprises the following steps: collecting building design parameters and operating parameters, and constructing a data set for training a machine learning model; training a machine learning model based on the data set, and predicting air conditioning carbon emission indicators, lighting carbon emission indicators, equipment carbon emission indicators, and carbon emission reduction indicators; establishing a multi-objective optimization model, and setting minimization of air conditioning carbon emission indicators, lighting carbon emission indicators, equipment carbon emission indicators, and maximization of carbon emission reduction indicators as optimization objectives; employing a multi-objective particle swarm optimization algorithm to solve the multi-objective optimization model, and obtaining a set of Pareto optimal solutions that meet the compromise between different carbon emission targets, as a set of candidate solutions for subsequent selection using an approximate ideal solution sorting method; based on the candidate solution set, applying the approximate ideal solution sorting method to select the optimal solution, and using the optimal solution for building carbon emission control. The present invention can efficiently find the optimal solution in a complex building parameter space.

Description

Technical Field

[0001] The present application relates to the field of building energy and environmental technology, and in particular to a method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization. Background Art

[0002] With global attention, controlling carbon emissions in the building sector has become a key focus. Carbon emissions generated during building operations primarily come from air conditioning and refrigeration systems, lighting systems, and various energy-consuming equipment. Renewable energy sources such as photovoltaic power generation can offset some of these emissions. In traditional design and operation, the design parameters of each subsystem (energy-saving system, power generation system, and energy consumption system) are often optimized independently, making it difficult to minimize overall carbon emissions. Existing technologies often use single-objective or linear models to predict building carbon emissions, making it difficult to accurately predict multiple carbon emission indicators simultaneously. Optimizing building design parameters often relies on manual trial and error or simple optimization algorithms, making it difficult to find a globally optimal solution within the vast parameter space. Furthermore, different carbon emission targets vary in importance and optimization direction (some prefer smaller values, others prefer larger values). Traditional single-objective optimization methods cannot effectively address the trade-offs between these multiple objectives, making it difficult to obtain a comprehensive optimal solution that meets the requirements for zero-carbon building operation. Summary of the Invention

[0003] The purpose of this invention is to provide a method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization. This method addresses the existing technical challenges of accurately predicting multiple building carbon emission indicators and comprehensively optimizing multiple design parameters. This method can efficiently predict building carbon emissions and optimize building design and operating parameters while considering multiple carbon emission targets, thereby effectively controlling building carbon emissions and contributing to the realization of zero-carbon buildings.

[0004] To achieve the above objectives, the present invention provides the following technical solutions: a method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization, comprising:

[0005] Collect building design and operation parameters to construct a data set for machine learning model training;

[0006] Training a machine learning model based on the data set to predict air conditioning carbon emissions, lighting carbon emissions, equipment carbon emissions, and carbon reduction indicators, wherein the machine learning model uses a gradient boosting decision tree algorithm (XGBoost);

[0007] Establish a multi-objective optimization model, with minimizing air conditioning carbon emissions, lighting carbon emissions, equipment carbon emissions, and maximizing carbon reduction as optimization goals;

[0008] The multi-objective particle swarm optimization algorithm (MOPSO) is used to solve the multi-objective optimization model to obtain a set of Pareto optimal solutions that meet the trade-offs between different carbon emission targets, which are used as candidate solutions for subsequent selection using the TOPSIS method;

[0009] Based on the candidate solution set, the TOPSIS method is applied to select the optimal solution, and the optimal solution is used for building carbon emission control;

[0010] The multi-objective particle swarm optimization algorithm (MOPSO) is used to solve the multi-objective optimization model. The solution process includes the following steps:

[0011] 1) Initialize the particle position within the constraints of the design parameters X i and speed V i , generate the initial population;

[0012] 2) Calculate the air conditioning carbon emission index, lighting carbon emission index, equipment carbon emission index, and carbon emission reduction index for each particle based on the XGBoost prediction model, and use these indexes for multi-objective fitness evaluation;

[0013] 3) Compare dominance relationships and maintain an external archive (ParetoArchive) to save non-dominated solutions;

[0014] 4) Calculate the crowding degree of the solution set to characterize the density distribution of the solution, and use the crowding degree as the basis for selecting the global optimal leader particle;

[0015] 5) Update the particle speed according to the speed update formula V i :

[0016] ;

[0017] in, i : Used to identify the number of the current particle in the population; t : The number of iterations in the current optimization process; : No. t Generation Particles i The position vector of : No. t Generation Particles i Velocity vector; ω: inertia weight; c 1: Individual learning factor; r 1: individual random number; : Individual optimal position vector; c 2: Group learning factor; r 2: Group random number; : global optimal position vector; : No. t Generation Particles i Updated velocity vector;

[0018] 6) Update the particle's position according to the position update formula X i :

[0019] ;

[0020] in, : No. t Generation Particles i Updated position vector;

[0021] 7) Repeat steps 1 to 6 until the preset maximum number of iterations is reached or the convergence criterion is met.

[0022] The building design parameters include thermal performance parameters of the building envelope, building shape parameters and renewable energy system configuration parameters; the operating parameters include air conditioning system operating parameters, lighting system operating parameters and equipment operating parameters.

[0023] The data set is obtained by simulating building energy consumption and carbon emissions for different combinations of building design parameters and operating parameters. Each parameter combination corresponds to an air conditioning carbon emission index, a lighting carbon emission index, an equipment carbon emission index, and a carbon emission reduction index.

[0024] The machine learning model is a multi-output regression model that simultaneously predicts air conditioning carbon emissions, lighting carbon emissions, equipment carbon emissions, and carbon reduction indicators. The machine learning model is a multi-output regression model trained using the gradient boosting decision tree algorithm (XGBoost) that simultaneously predicts air conditioning carbon emissions, lighting carbon emissions, equipment carbon emissions, and carbon reduction indicators. The training process includes the following steps:

[0025] Feature construction: Using building design parameters and operation parameters as input variables, construct a feature matrix;

[0026] Label construction: The air conditioning carbon emission index, lighting carbon emission index, equipment carbon emission index, and carbon emission reduction index in the building simulation results or measured data are used as output labels;

[0027] Data preprocessing: normalize or standardize input features and remove or replace outliers;

[0028] Model structure: Four parallel XGBoost sub-models are used, each sub-model corresponds to predicting one of the output indicators;

[0029] Training optimization: Using mean squared error (MSE) as the loss function, cross-validation was used to adjust the key parameters of XGBoost, including learning rate, number of subtrees, and tree depth;

[0030] Multi-output integration: Combine the outputs of each sub-model into a multi-dimensional output vector for subsequent optimization and evaluation.

[0031] The optimal solution is implemented by adjusting the operating parameters of the building's air-conditioning system, lighting system and equipment, thereby achieving optimized control of the building's carbon emissions.

[0032] The TOPSIS method ranks solutions by calculating the closeness of each Pareto optimal solution to the ideal solution, and determines the solution with the largest closeness as the optimal solution.

[0033] The TOPSIS method ranks solutions by calculating the closeness of each Pareto optimal solution to the ideal solution, and determines the solution with the largest closeness as the optimal solution. The process is as follows:

[0034] 1) Perform dimensionless normalization on the values ​​of each of the schemes under each of the carbon emission targets to obtain normalized values:

[0035] ;

[0036] in, X ij : No. i The plan in j The original value of each carbon emission target; Z ij : normalized value; M : total number of solutions;

[0037] 2) Multiplying the normalized value by the weight of each carbon emission target to obtain a weighted normalized matrix:

[0038] ;

[0039] in, W j : No. j The weight coefficient of each carbon emission target; V ij : The value after weighted normalization;

[0040] 3) Construct positive ideal solutions and negative ideal solutions:

[0041] ;

[0042] ;

[0043] in, : No. j A positive ideal solution to a target; : No. j Negative ideal solution of a target;

[0044] 4) Calculate the Euclidean distance of each solution to the positive ideal solution and the negative ideal solution respectively:

[0045] ;

[0046] ;

[0047] in, :plan i The distance to the positive ideal solution; :plan i The distance to the negative ideal solution; N : number of target indicators;

[0048] 5) Calculate the closeness of each solution:

[0049] ;

[0050] in, C i :plan i The closeness value, 0≤ C i ≤≤1;

[0051] 6) All solutions are sorted by closeness C i Sort by size, and the solution with the highest degree of similarity is determined as the optimal solution.

[0052] The air conditioning carbon emission index, lighting carbon emission index and equipment carbon emission index are calculated based on the electricity consumption of the building's air-conditioning system, lighting system and equipment and their corresponding carbon emission factors respectively; the carbon emission reduction index is calculated based on the carbon emissions reduced by replacing conventional energy with energy generated by renewable energy in the building.

[0053] The air conditioning carbon emission index, lighting carbon emission index and equipment carbon emission index are calculated as follows:

[0054] Air conditioning carbon emission index = annual electricity consumption of air conditioning system × electricity carbon emission factor;

[0055] Lighting carbon emission index = annual electricity consumption of lighting system × electricity carbon emission factor;

[0056] Equipment carbon emission index = annual electricity consumption of other energy-consuming equipment in the building × electricity carbon emission factor;

[0057] The electricity carbon emission factor is the carbon emission coefficient corresponding to unit electricity, with the unit being kilograms of carbon dioxide per kilowatt-hour. Its value can be determined according to the latest standards published by the country or region;

[0058] The carbon emission reduction index is calculated according to the following formula:

[0059] Carbon emission reduction index = annual power generation of renewable energy system × electricity carbon emission factor;

[0060] Among them, the renewable energy system is the photovoltaic power generation system installed in the building; the electricity carbon emission factor is the carbon emission coefficient corresponding to unit electricity, and the unit is kilograms of carbon dioxide / kilowatt-hour. The value can be determined based on the latest standards announced by the country or region; the annual power generation is the total power generation of the building photovoltaic system throughout the year, and the unit is kilowatt-hour.

[0061] The thermal performance parameters of the envelope structure include the thermal insulation performance parameters of the building's exterior walls, roof and windows; the building shape parameters include the building window-to-wall area ratio and the building shape coefficient; the renewable energy system configuration parameters include the installed capacity of the distributed photovoltaic power generation system; the air-conditioning system operating parameters include the indoor temperature set value and the air-conditioning operating time setting; the lighting system operating parameters include the lighting start and stop schedule and the daylight lighting control strategy; the equipment operating parameters include the operating time plan of various energy-consuming equipment.

[0062] Compared with the existing technology, the beneficial effects of the present invention are: the present invention combines machine learning multi-output prediction with multi-objective intelligent optimization, and can efficiently find the best in the complex building parameter space: 1. Using the trained machine learning prediction model, multiple carbon emission indicators can be instantly estimated, avoiding the high cost of traditional repeated and detailed simulations and improving the optimization efficiency; 2. Through the multi-objective particle swarm optimization algorithm, it is possible to globally search for parameter combinations that take into account various carbon emission targets, overcoming the defect that manual or single-objective optimization is prone to falling into local optimality; 3. The TOPSIS decision-making method is introduced to fully consider the importance of each target and different optimization directions, and to automatically screen out the best comprehensive performance solution from the Pareto solution set, ensuring that the obtained solution will minimize the building carbon emissions and close to zero carbon operation overall; 4. The optimal parameter solution provided by the present invention can be used to guide building design and for control optimization in the operation stage, constructing a carbon emission control mechanism integrating design and operation, which can significantly improve the energy efficiency and carbon emission reduction effect of the building throughout its life cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The present invention is further described with reference to the accompanying drawings. However, the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative effort.

[0064] Figure 1 Schematic diagram of the workflow of the present invention. DETAILED DESCRIPTION

[0065] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0066] See also Figure 1 The present invention provides a method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization, including

[0067] Collect building design and operation parameters to construct a data set for machine learning model training;

[0068] Training a machine learning model based on the data set to predict air conditioning carbon emissions, lighting carbon emissions, equipment carbon emissions, and carbon reduction indicators, wherein the machine learning model uses a gradient boosting decision tree algorithm (XGBoost);

[0069] Establish a multi-objective optimization model, with minimizing air conditioning carbon emissions, lighting carbon emissions, equipment carbon emissions, and maximizing carbon reduction as optimization goals;

[0070] The multi-objective particle swarm optimization algorithm (MOPSO) is used to solve the multi-objective optimization model to obtain a set of Pareto optimal solutions that meet the trade-offs between different carbon emission targets, which are used as candidate solutions for subsequent selection using the TOPSIS method;

[0071] Based on the candidate solution set, the TOPSIS method is applied to select the optimal solution, and the optimal solution is used for building carbon emission control.

[0072] The building design parameters include thermal performance parameters of the building envelope, building shape parameters and renewable energy system configuration parameters; the operating parameters include air conditioning system operating parameters, lighting system operating parameters and equipment operating parameters.

[0073] The data set is obtained by simulating building energy consumption and carbon emissions for different combinations of building design parameters and operating parameters. Each parameter combination corresponds to an air conditioning carbon emission index, a lighting carbon emission index, an equipment carbon emission index, and a carbon emission reduction index.

[0074] The machine learning model is a multi-output regression model that simultaneously predicts air conditioning carbon emissions, lighting carbon emissions, equipment carbon emissions, and carbon reduction indicators. The machine learning model is a multi-output regression model trained using the gradient boosting decision tree algorithm (XGBoost) that simultaneously predicts air conditioning carbon emissions, lighting carbon emissions, equipment carbon emissions, and carbon reduction indicators. The training process includes the following steps:

[0075] Feature construction: Using building design parameters and operation parameters as input variables, construct a feature matrix;

[0076] Label construction: The air conditioning carbon emission index, lighting carbon emission index, equipment carbon emission index, and carbon emission reduction index in the building simulation results or measured data are used as output labels;

[0077] Data preprocessing: normalize or standardize input features and remove or replace outliers;

[0078] Model structure: Four parallel XGBoost sub-models are used, each sub-model corresponds to predicting one of the output indicators;

[0079] Training optimization: Using mean squared error (MSE) as the loss function, cross-validation was used to adjust the key parameters of XGBoost, including learning rate, number of subtrees, and tree depth;

[0080] Multi-output integration: Combine the outputs of each sub-model into a multi-dimensional output vector for subsequent optimization and evaluation.

[0081] The optimal solution is implemented by adjusting the operating parameters of the building's air-conditioning system, lighting system and equipment, thereby achieving optimized control of the building's carbon emissions.

[0082] The multi-objective particle swarm optimization algorithm (MOPSO) is used to solve the multi-objective optimization model. The solution process includes the following steps:

[0083] 1) Initialize the particle position within the constraints of the design parameters X i and speed V i , generate the initial population;

[0084] 2) Calculate the air conditioning carbon emission index, lighting carbon emission index, equipment carbon emission index, and carbon emission reduction index for each particle based on the XGBoost prediction model, and use these indexes for multi-objective fitness evaluation;

[0085] 3) Compare dominance relationships and maintain an external archive (ParetoArchive) to save non-dominated solutions;

[0086] 4) Calculate the crowding degree of the solution set to characterize the density distribution of the solution, and use the crowding degree as the basis for selecting the global optimal leader particle;

[0087] 5) Update the particle speed according to the speed update formula V i :

[0088] ;

[0089] in, i : Used to identify the number of the current particle in the population; t : The number of iterations in the current optimization process; : No. t Generation Particles i The position vector of : No. t Generation Particles i Velocity vector; ω: inertia weight; c 1: Individual learning factor; r 1: individual random number; : Individual optimal position vector; c 2: Group learning factor; r 2: Group random number; : global optimal position vector; : No. t Generation Particles i Updated velocity vector;

[0090] 6) Update the particle's position according to the position update formula X i :

[0091] ;

[0092] in, : No. t Generation Particles i Updated position vector;

[0093] 7) Repeat steps 1 to 6 until the preset maximum number of iterations is reached or the convergence criterion is met.

[0094] The TOPSIS method ranks solutions by calculating the closeness of each Pareto optimal solution to the ideal solution, and determines the solution with the largest closeness as the optimal solution.

[0095] The TOPSIS method ranks solutions by calculating the closeness of each Pareto optimal solution to the ideal solution, and determines the solution with the largest closeness as the optimal solution. The process is as follows:

[0096] 1) Perform dimensionless normalization on the values ​​of each of the schemes under each of the carbon emission targets to obtain normalized values:

[0097] ;

[0098] in, X ij : No. i The plan in j The original value of each carbon emission target; Z ij : normalized value; M : total number of solutions;

[0099] 2) Multiplying the normalized value by the weight of each carbon emission target to obtain a weighted normalized matrix:

[0100] ;

[0101] in, W j : No. j The weight coefficient of each carbon emission target; V ij : The value after weighted normalization;

[0102] 3) Construct positive ideal solutions and negative ideal solutions:

[0103] ;

[0104] ;

[0105] in, : No. j A positive ideal solution to a target; : No. j Negative ideal solution of a target;

[0106] 4) Calculate the Euclidean distance of each solution to the positive ideal solution and the negative ideal solution respectively:

[0107] ;

[0108] ;

[0109] in, :plan i The distance to the positive ideal solution; :plan i The distance to the negative ideal solution; N : number of target indicators;

[0110] 5) Calculate the closeness of each solution:

[0111] ;

[0112] in, C i :plan i The closeness value, 0≤ C i ≤≤1;

[0113] 6) All solutions are sorted by closeness C i Sort by size, and the solution with the highest degree of similarity is determined as the optimal solution.

[0114] The air conditioning carbon emission index, lighting carbon emission index and equipment carbon emission index are calculated based on the electricity consumption of the building's air-conditioning system, lighting system and equipment and their corresponding carbon emission factors respectively; the carbon emission reduction index is calculated based on the carbon emissions reduced by replacing conventional energy with energy generated by renewable energy in the building.

[0115] The air conditioning carbon emission index, lighting carbon emission index and equipment carbon emission index are calculated as follows:

[0116] Air conditioning carbon emission index = annual electricity consumption of air conditioning system × electricity carbon emission factor;

[0117] Lighting carbon emission index = annual electricity consumption of lighting system × electricity carbon emission factor;

[0118] Equipment carbon emission index = annual electricity consumption of other energy-consuming equipment in the building × electricity carbon emission factor;

[0119] The electricity carbon emission factor is the carbon emission coefficient corresponding to unit electricity, with the unit being kilograms of carbon dioxide per kilowatt-hour. Its value can be determined according to the latest standards published by the country or region;

[0120] The carbon emission reduction index is calculated according to the following formula:

[0121] Carbon emission reduction index = annual power generation of renewable energy system × electricity carbon emission factor;

[0122] Among them, the renewable energy system is the photovoltaic power generation system installed in the building; the electricity carbon emission factor is the carbon emission coefficient corresponding to unit electricity, and the unit is kilograms of carbon dioxide / kilowatt-hour. The value can be determined based on the latest standards announced by the country or region; the annual power generation is the total power generation of the building photovoltaic system throughout the year, and the unit is kilowatt-hour.

[0123] The thermal performance parameters of the envelope structure include the thermal insulation performance parameters of the building's exterior walls, roof and windows; the building shape parameters include the building window-to-wall area ratio and the building shape coefficient; the renewable energy system configuration parameters include the installed capacity of the distributed photovoltaic power generation system; the air-conditioning system operating parameters include the indoor temperature set value and the air-conditioning operating time setting; the lighting system operating parameters include the lighting start and stop schedule and the daylight lighting control strategy; the equipment operating parameters include the operating time plan of various energy-consuming equipment.

[0124] Example 1

[0125] A multi-output prediction model was established. A parameterized model was constructed for an office building located in a tropical region, selecting 30 key design parameters for its building envelope, air conditioning system, lighting system, equipment load, and photovoltaic power generation system. These parameters encompass energy conservation, production capacity, and energy use factors, such as building orientation, window-to-wall ratio, exterior envelope heat transfer coefficient, window transmittance and solar heat gain coefficient, indoor occupancy density and per capita floor area, lighting power density, air conditioning temperature setpoint, fresh air volume, and photovoltaic panel coverage. During the design phase, a large number of different parameter combinations were generated using Latin hypercube sampling. Building performance simulation software was used to calculate carbon emission indicators for each scenario in a typical meteorological year, including annual carbon emissions corresponding to air conditioning electricity consumption (F1), lighting electricity consumption (F2), equipment electricity consumption (F3), and the carbon emission reductions (F4) derived from photovoltaic power generation. Tens of thousands of sets of parameter and indicator data were collected for training and validation of the machine learning prediction model. The dataset was split into training and test sets in an 8:2 ratio, and the XGBoost regression model was selected as the base algorithm. For each output metric, an XGBoost sub-model was trained, taking the design parameters as input and 30 independent variables, and outputting the corresponding carbon emission metric. Hyperparameters (such as the maximum tree depth, learning rate, and number of subtrees) were adjusted during training to optimize prediction accuracy. After training, the coefficient of determination (R²) for all four metrics on the test set exceeded 0.95, and the mean squared error (MSE) remained low, demonstrating that the model accurately fits the nonlinear relationship between building parameters and carbon emission metrics. The resulting multi-output prediction model can quickly predict the F1, F2, F3, and F4 values ​​of a given solution for any given set of building parameters, enabling rapid assessment of a building's carbon emission performance.

[0126] Multi-objective optimization and optimal solution decision-making. Based on the trained prediction model, a multi-objective optimization problem for building carbon emissions is established. The optimization objectives are: minimizing F1, F2, and F3, that is, minimizing carbon emissions caused by cooling, lighting, and equipment energy consumption; and maximizing F4, that is, increasing renewable energy production to offset carbon emissions. Since the goal is to achieve zero net carbon emissions for the building throughout the year, the optimization must also ensure that F4 is sufficient to offset F1, F2, and F3. Thirty design parameters are used as decision variables, with variable value ranges set according to engineering specifications and practical constraints (for example, a window-to-wall ratio between 0 and 1, a temperature setting between 24°C and 30°C, etc.). A multi-objective particle swarm optimization algorithm is used to solve the problem. The algorithm initializes the population size to 50, with each individual representing a set of building parameters. During the iterative process, the trained XGBoost model is used to calculate the F1-F4 values ​​corresponding to each individual, and the individual's performance on the four objectives is evaluated based on these values. The Pareto dominance relationship is used to filter non-dominated solutions and store the Pareto frontier. The particle swarm moves through parameter space using speed and position update rules, introducing archived Pareto solutions and global optimal guidance. After a preset 100 iterations, approximately 500 Pareto-optimal design solutions were obtained. These solutions represent design choices with different trade-offs. For example, some solutions focus on significantly reducing air conditioning and equipment energy consumption while slightly reducing photovoltaic configuration, while others may increase photovoltaic capacity to offset higher energy consumption. This results in a near-zero-carbon balance across all strategies.

[0127] After obtaining the Pareto solution set, the solution with the best overall performance needs to be selected as the final recommended design and control solution. This embodiment uses the TOPSIS method for decision-making. First, the weights of each objective are determined based on the focus of building carbon emission control. Considering that in the initial building baseline solution, air conditioning and equipment electricity consumption account for a large proportion, followed by lighting, and photovoltaic production capacity depends on the installed capacity, F1, F3, and F4 are assigned high and similar weights, while F2 is given a lower weight. Specifically, based on the aforementioned simulation data, the average values ​​of the four indicators F1, F2, F3, and F4 account for approximately 30%, 7%, 32%, and 31%, respectively. Based on this, the weight vector is determined to be [0.30, 0.07, 0.32, 0.31]. Next, the three cost-based indicators F1, F2, and F3 are normalized and converted into benefit-based indicators. For example, the difference between the indicator value of each solution and the worst value in the solution set is calculated and divided by the range, so that the larger the value, the better the solution. F4 remains unchanged as a benefit-based indicator. The four processed indicators for all Pareto scenarios were then normalized and multiplied by the corresponding weights to form a weighted matrix. The ideal optimal solution was determined as the weighted normalized maximum vector of each indicator, while the worst solution was determined as the minimum vector. The Euclidean distance between each scenario and the ideal optimal solution and the worst solution was calculated. The closeness C value (distance from the worst solution / total distance) was further calculated. Finally, the C values ​​of all scenarios were compared; the closer the C value to 1, the better the scenario. Calculations revealed that among the 500 Pareto scenarios, scenario 192 had the highest closeness, exceeding 0.9, while scenario 383 had the lowest closeness. Scenario 192 was selected as the optimal scenario, with the following predicted indicators: F1 approximately 53,000 kgCO2, F2 approximately 12,000 kgCO2, F3 approximately 55,000 kgCO2, and F4 approximately 54,000 kgCO2. Under this scenario, the sum of F1, F2, and F3 was slightly less than F4, achieving net zero carbon emissions for the entire year. Furthermore, compared to other Pareto set options, this one performs best in terms of comprehensively balancing various objectives. An analysis of a set of building parameters corresponding to Option 192 revealed that, compared to the inferior Option 383, this option significantly improved the performance of the building envelope (e.g., enhanced roof insulation, reducing thermal insulation by approximately 34%), optimized the orientation and window-to-wall ratio (e.g., the building orientation was rotated by approximately 332%, and the east-west window-to-wall ratio was adjusted), and reduced lighting and equipment load parameters. The air conditioning temperature setting was slightly increased, but remained within a comfortable range. This demonstrates that the optimal solution proposed by the proposed method does not rely solely on reducing air conditioning energy consumption to achieve carbon reduction, but rather achieves overall carbon reduction through the comprehensive optimization of various subsystem parameters (e.g., enhanced passive energy conservation in the building envelope, intelligent lighting and equipment management, and sufficient photovoltaic capacity).

[0128] Application of the optimal solution and feedback control. The selected optimal parameter solution was applied to the actual design and operation of the office building. During the design phase, the building was configured according to the solution parameters: building orientation was adjusted to reduce direct solar load, high-resistance wall and roof materials were used to reduce cooling load, high-efficiency lighting fixtures were introduced and lighting controls were arranged appropriately, and sufficient photovoltaic panels were installed to meet energy production requirements. During the operation phase, the solution's operating parameters were imported into the Building Energy Management System (BEMS). Air conditioning temperatures were set, fresh air volume was controlled to the solution's specified values, the lighting system was adaptively adjusted based on natural light and occupancy sensors to ensure lighting usage met the solution's expectations, and the activation strategy for non-essential equipment was implemented according to the optimized results. The building's actual energy consumption and photovoltaic power generation were monitored in real time through the energy monitoring system, and actual carbon emissions data was summarized daily. Operational results showed that the building's actual annual total electricity consumption corresponded to carbon emissions close to the predicted values, and that renewable energy generation was sufficient to offset the consumption, achieving net-zero carbon operation. If monitoring revealed that carbon emissions exceeded the predicted values ​​during a specific period, such as due to increased air conditioning energy consumption caused by unusually high temperatures, this deviation was fed back to the optimization module, triggering a new round of optimization calculations. While maintaining the original design, a new round of optimization can adjust operational control parameters (e.g., further increasing air conditioning temperature settings, increasing photovoltaic output, or activating energy storage devices) to generate a new optimal control solution for management's reference and implementation. Through this iterative feedback mechanism, the method can continuously adjust optimization results to adapt to changes in actual use and maintain the building's low-carbon operation over the long term.

[0129] To demonstrate the emission reduction effects of the present method in practical applications, we present annual comparative data for the baseline and optimized solutions for three carbon emission indicators (air conditioning, lighting, and equipment) and carbon reduction indicators, using a medium-sized office building in a tropical region as an example. The absolute differences and relative rates of change for each indicator are then calculated, along with the overall reduction in the building's annual net carbon emissions. Specific data is shown in the table below:

[0130]

[0131] The comparison results show that the implementation of a collaborative control approach based on machine learning prediction and multi-objective optimization significantly reduced carbon emissions across all building components: air conditioning emissions dropped from 6,000 kgCO2 / year to 4,500 kgCO2 / year, a 25% decrease; lighting emissions, thanks to intelligent dimming and high-efficiency light sources, decreased from 3,400 kgCO2 / year to 1,700 kgCO2 / year, a 50% decrease, making it the most significant component in terms of relative emission reduction; and equipment emissions, thanks to energy scheduling and energy-saving terminals, dropped from 10,600 kgCO2 / year to 7,420 kgCO2 / year, a decrease of approximately 30%. Furthermore, the introduction of a photovoltaic system offsets 1,000 kgCO2 emissions annually, making the building a negative emission contributor on the "source side." Combined with these changes, the building's total annual net carbon emissions dropped from 20,000 kgCO2 / year to approximately 12,620 kgCO2 / year, achieving an overall reduction rate of approximately 36.9%, fully demonstrating the potential and feasibility of this method for high-efficiency carbon reduction in office buildings in tropical regions. The above simulation data show that the optimization scheme significantly reduced the carbon emissions of each subsystem, and the total annual net carbon emissions decreased by 36.9%, verifying the emission reduction effect based on machine learning and multi-objective optimization methods.

[0132] The above are only specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be covered by the scope of protection of the present invention. Matters not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization, characterized in that: include: Collect building design and operation parameters to construct a data set for machine learning model training; Training a machine learning model based on the data set to predict air conditioning carbon emissions, lighting carbon emissions, equipment carbon emissions, and carbon emission reduction indicators, wherein the machine learning model uses a gradient boosting decision tree algorithm; Establish a multi-objective optimization model, with minimizing air conditioning carbon emissions, lighting carbon emissions, equipment carbon emissions, and maximizing carbon reduction as optimization goals; A multi-objective particle swarm optimization algorithm is used to solve the multi-objective optimization model to obtain a set of Pareto optimal solutions that meet the trade-offs between different carbon emission targets, which are used as candidate solutions for subsequent selection using the approximate ideal solution ranking method; Based on the candidate solution set, an optimal solution is selected by applying an approximate ideal solution ranking method, and the optimal solution is used for building carbon emission control; The multi-objective particle swarm optimization algorithm is used to solve the multi-objective optimization model, and the solution process includes the following steps: 1) Initialize the particle position within the constraints of the design parameters X i and speed V i , generate the initial population; 2) Calculate the air conditioning carbon emission index, lighting carbon emission index, equipment carbon emission index, and carbon emission reduction index of each particle based on the gradient boosting decision tree algorithm prediction model, and use these indicators for multi-objective fitness evaluation; 3) Compare dominance relationships and maintain an external archive to save non-dominated solutions; 4) Calculate the crowding degree of the solution set to characterize the density distribution of the solution, and use the crowding degree as the basis for selecting the global optimal leader particle; 5) Update the particle speed according to the speed update formula V i : ； in, i : Used to identify the number of the current particle in the population; t : The number of iterations in the current optimization process; : No. t Generation Particles i The position vector of : No. t Generation Particles i Velocity vector; ω: inertia weight; c 1: Individual learning factor; r 1: individual random number; : Individual optimal position vector; c 2: Group learning factor; r 2: Group random number; : global optimal position vector; : No. t Generation Particles i Updated velocity vector; 6) Update the particle's position according to the position update formula X i : ； in, : No. t Generation Particles i Updated position vector; 7) Repeat steps 1 to 6 until the preset maximum number of iterations is reached or the convergence criterion is met.

2. The method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization according to claim 1, wherein: The building design parameters include thermal performance parameters of the building envelope, building shape parameters and renewable energy system configuration parameters; the operating parameters include air conditioning system operating parameters, lighting system operating parameters and equipment operating parameters.

3. The method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization according to claim 1, wherein: The data set is obtained by simulating building energy consumption and carbon emissions for different combinations of building design parameters and operating parameters. Each parameter combination corresponds to an air conditioning carbon emission index, a lighting carbon emission index, an equipment carbon emission index, and a carbon emission reduction index.

4. The method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization according to claim 1, wherein: The optimal solution is implemented by adjusting the operating parameters of the building's air-conditioning system, lighting system and equipment, thereby achieving optimized control of the building's carbon emissions.

5. The method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization according to claim 1, wherein: The approach-to-ideal solution ranking method ranks solutions by calculating the closeness of each Pareto optimal solution to the ideal solution, and determines the solution with the greatest closeness as the optimal solution.

6. The method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization according to claim 5, characterized in that: The approach to ideal solution ranking method ranks the solutions by calculating the closeness of each Pareto optimal solution to the ideal solution, and determines the solution with the largest closeness as the optimal solution. The process is as follows: 1) Perform dimensionless normalization on the values ​​of each of the schemes under each of the carbon emission targets to obtain normalized values: ； in, : No. i The plan in j The original value of each carbon emission target; : normalized value; M : total number of solutions; 2) Multiplying the normalized value by the weight of each carbon emission target to obtain a weighted normalized matrix: ； in, W j : No. j The weight coefficient of each carbon emission target; V ij : The value after weighted normalization; 3) Construct positive ideal solutions and negative ideal solutions: ； ； in, : No. j A positive ideal solution to a target; : No. j Negative ideal solution of a target; 4) Calculate the Euclidean distance of each solution to the positive ideal solution and the negative ideal solution respectively: ； ； in, :plan i The distance to the positive ideal solution; :plan i The distance to the negative ideal solution; N : number of target indicators; 5) Calculate the closeness of each solution: ； in, C i :plan i The closeness value, 0≤ C i ≤≤1; 6) All solutions are sorted by closeness C i Sort by size, and the solution with the highest degree of similarity is determined as the optimal solution.

7. The method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization according to claim 1, wherein: The air conditioning carbon emission index, lighting carbon emission index and equipment carbon emission index are calculated based on the electricity consumption of the building's air-conditioning system, lighting system and equipment and their corresponding carbon emission factors respectively; the carbon emission reduction index is calculated based on the carbon emissions reduced by replacing conventional energy with energy generated by renewable energy in the building.

8. The method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization according to claim 7, wherein: The air conditioning carbon emission index, lighting carbon emission index and equipment carbon emission index are calculated as follows: Air conditioning carbon emission index = annual electricity consumption of air conditioning system × electricity carbon emission factor; Lighting carbon emission index = annual electricity consumption of lighting system × electricity carbon emission factor; Equipment carbon emission index = annual electricity consumption of other energy-consuming equipment in the building × electricity carbon emission factor; The electricity carbon emission factor is the carbon emission coefficient corresponding to unit electricity, with the unit being kilograms of carbon dioxide per kilowatt-hour. Its value can be determined according to the latest standards published by the country or region; The carbon emission reduction index is calculated according to the following formula: Carbon emission reduction index = annual power generation of renewable energy system × electricity carbon emission factor; Among them, the renewable energy system is the photovoltaic power generation system installed in the building; the electricity carbon emission factor is the carbon emission coefficient corresponding to unit electricity, and the unit is kilograms of carbon dioxide / kilowatt-hour. The value can be determined based on the latest standards announced by the country or region; the annual power generation is the total power generation of the building photovoltaic system throughout the year, and the unit is kilowatt-hour.

9. The method for predicting and controlling building carbon emissions based on machine learning and multi-objective optimization according to claim 2, wherein: The thermal performance parameters of the envelope structure include the thermal insulation performance parameters of the building's exterior walls, roof and windows; the building shape parameters include the building window-to-wall area ratio and the building shape coefficient; the renewable energy system configuration parameters include the installed capacity of the distributed photovoltaic power generation system; the air-conditioning system operating parameters include the indoor temperature set value and the air-conditioning operating time setting; the lighting system operating parameters include the lighting start and stop schedule and the daylight lighting control strategy; the equipment operating parameters include the operating time plan of various energy-consuming equipment.

Images