Building hidden carbon emission prediction method and device based on machine learning

Abstract

The invention discloses a building hidden carbon emission prediction method based on machine learning, and the method comprises the following steps: S1, building construction activities are comprehensively considered, a building hidden carbon emission structure is qualitatively and quantitatively analyzed, and a main carbon emission source is selected as an output parameter of a target prediction model; s2, the system identifies influence factors of hidden carbon emission of the building, and grey correlation analysis is adopted to select influence factors with large correlation as input features of the target prediction model; s3, making a data collection table, and carrying out data collection for an actual construction project; s4, constructing a construction consumption prediction model by using a machine learning algorithm; s5, building hidden carbon emission prediction is completed in combination with the carbon emission factors; and S6, according to the methods of S1 to S5, using Python to develop a building hidden carbon emission prediction device. According to the method, the hidden carbon emission level of the building can be rapidly and conveniently estimated in the initial design stage of the building, and then low-carbon design optimization, scheme comparison and selection, decision management and the like are supported.

Description

Technical Field

[0001] The present invention belongs to the technical field of building carbon emission assessment, and specifically relates to a method and device for predicting building embodied carbon emissions based on machine learning. Background Art

[0002] In recent years, the scale and output value of the construction industry have developed rapidly, making great contributions to the world's economic growth and social development, but it is also accompanied by a large amount of resource consumption and carbon emissions. According to statistics, nearly 30-40% of the world's carbon emissions come from the construction industry, of which implicit carbon emissions account for a large proportion.

[0003] At present, the prediction of embodied carbon emissions in buildings often obtains building-related information from engineering drawings or building information models, and estimates construction consumption data and calculates embodied carbon emissions in combination with engineering quotas. However, engineering quota data only reflects the industry average and does not take into account the impact of construction technology and management levels of different contractors. Therefore, the accuracy and representativeness of the prediction results are not strong. In addition, the current prediction of embodied carbon emissions in buildings requires a lot of time, manpower and material resources, which is costly and requires participants to have certain professional basic knowledge. There is a lack of convenient, fast, and user-optimized intelligent analysis methods and tools. Summary of the invention

[0004] To solve the above problems, the present invention discloses a method for predicting embodied carbon emissions in buildings based on machine learning. Based on a large amount of actual engineering case data, a machine learning algorithm is used to construct the correlation between influencing factors such as project characteristics, construction characteristics and management level and construction consumption, and a tool for predicting embodied carbon emissions in buildings is developed. The method can conveniently and quickly carry out quantitative prediction of carbon emissions in the preliminary design stage of buildings, guide the low-carbon design and construction of buildings, and is of great significance for promoting the sustainable development of the construction industry.

[0005] To achieve the above object, the technical solution of the present invention is as follows:

[0006] A method for predicting building embodied carbon emissions based on machine learning, comprising the following steps:

[0007] S1: Comprehensively consider construction activities, qualitatively and quantitatively analyze the implicit carbon emission structure of buildings, and select the main carbon emission sources as the output parameters of the target prediction model;

[0008] S2: The system identifies the influencing factors of building embodied carbon emissions, and uses grey correlation analysis to select the influencing factors with greater correlation as the input features of the target prediction model;

[0009] S3: Prepare data collection forms and conduct data collection for actual construction projects;

[0010] S4: Use machine learning algorithms to build a construction consumption prediction model;

[0011] S5: Combine carbon emission factors to complete the prediction of building embodied carbon emissions;

[0012] S6: Based on the methods of S1-S5, use Python to develop a building embodied carbon emission prediction device.

[0013] The functional framework of the building embodied carbon emission prediction device includes a platform layer, a data layer and an application layer. In the platform layer, users provide information data of the evaluated project according to the project being evaluated, and users are also allowed to modify some built-in parameter data according to the actual situation of the project; the data layer stores data and machine learning models; in the application layer, the building embodied carbon emissions are predicted, visualized and analyzed.

[0014] The device includes four main functional modules. Project basic information collection module: users input basic project information such as project name, project location, construction time, number of floors above ground, etc.; data management module: fill in the data information of influencing factors at three levels: project characteristics, construction characteristics and management level, and can also modify the values ​​of built-in parameters such as carbon emission factors; construction consumption prediction module: display the construction consumption results predicted by the machine learning model in the form of charts; carbon emission result display module: display the construction carbon emissions of the evaluated buildings in the form of charts, and can perform proportional analysis according to sources (building materials, energy and water).

[0015] A building embodied carbon emission prediction system comprises a processor and a memory, wherein the memory contains a program and runs a pre.bat script installation program operating environment, and the program is loaded and executed by the processor to implement the steps of the building embodied carbon emission prediction method.

[0016] The beneficial effects of the present invention are:

[0017] The method for predicting building embodied carbon emissions based on machine learning described in the present invention can quickly predict embodied carbon emissions in the preliminary design stage of a building, and provide data support for design optimization, scheme comparison, and decision-making management. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flow chart of a method for predicting embodied carbon emissions in buildings based on machine learning;

[0019] Figure 2 This is the flow chart of grey relational degree calculation;

[0020] Figure 3 A flowchart for the construction of a building consumption prediction model;

[0021] Figure 4It is the flow chart of particle swarm algorithm in SVR model;

[0022] Figure 5 This is the functional framework diagram of the tool;

[0023] Figure 6 Use a flowchart for the tool;

[0024] Figure 7 For the user interface;

[0025] in Figure 7 (a) User entry interface; Figure 7 (b) shows the basic information collection interface of the project; Figure 7 (c) Data management interface; Figure 7 (d) shows the carbon emission results display interface. DETAILED DESCRIPTION

[0026] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0027] The method for predicting building embodied carbon emissions based on machine learning described in the present invention takes the main building project and foundation project as the research objects, adopts machine learning algorithm to explore the correlation between engineering characteristic factors and consumption, constructs a carbon emission prediction tool, realizes quantitative prediction quickly and conveniently in the building design stage, and guides the low-carbon design and construction of buildings.

[0028] like Figure 1 As shown, the present invention provides a method for predicting building embodied carbon emissions based on machine learning, which specifically includes the following steps:

[0029] (1) Comprehensively consider construction activities, qualitatively and quantitatively analyze the implicit carbon emission structure of buildings, and select the main sources of carbon emissions as the output parameters of the target prediction model. Focus on several key materials that are used in large quantities and have a large proportion of carbon emissions in the foundation and main structure of construction projects, including steel, concrete, bricks, sand and gravel, and wood; energy types include electricity, gasoline, diesel and coal, as well as water.

[0030] (2) Systematically identify 40 factors affecting building embodied carbon emissions from three aspects: project characteristics, construction characteristics, and management level. Grey correlation analysis is performed on the identified factors, and grey correlation calculation is performed, such as Figure 2 As shown in the figure, the top 30 results with the largest correlation are selected as the input features of the building embodied carbon emission prediction model. The calculation process of the grey correlation degree of the influencing factors is as follows:

[0031] Step 21: Select the total embodied carbon emissions of the building as the reference series x0 = {x0(1), x0(2), ..., x0(u)}, and the i-th influencing factor of the embodied carbon emissions of the building as the comparison series x i ={x i (1),x i (2),...,x i (u)}, where i = 1, 2, ..., m, u is the number of buildings collected; m is the number of influencing factors identified;

[0032] Step 22: Use maximum and minimum normalization to make the data dimensionless and obtain x'0 = {x'0(1), x'0(2), ..., x'0(u)} and x' i ={x′ i (1),x′ i (2),...,x′ i (u)}, the normalized calculation formula is as follows:

[0033]

[0034] Where: x j (k) represents the kth data in the jth sequence; x′ j (k) represents x j (k) Normalized result; represent the maximum and minimum values ​​in the jth sequence respectively.

[0035] Step 23: Calculate the grey correlation coefficient for the normalized data, and record x0(k) and x i The correlation coefficient at point k is γ(x′0(k),x′ i (k)), the grey correlation coefficient calculation formula is as follows:

[0036]

[0037] Where: Δ 0i (k) = |x′ i (k)-x′0(k)|, ζ is the resolution coefficient, ζ=0.5

[0038] Step 24: Calculate the correlation γ between the comparison sequence and the reference sequence i , the correlation calculation formula is as follows:

[0039]

[0040] Step 25: Sort the comparison series according to the degree of correlation, and select the top 30 results with the largest correlation as the input features of the building embodied carbon emission prediction model.

[0041] (3) Based on the main carbon emission sources and influencing factors determined in steps (1) and (2), the collected data includes 41 data collection sub-items (including 11 types of consumption and 30 influencing factors) in four categories: engineering characteristics, construction characteristics, project management, and consumption information. Each type of consumption can be obtained based on construction records and resource statistics, and the influencing factor data can be obtained from the project design documents and construction plan.

[0042] (4) Build a machine learning model for building consumption prediction. The process is as follows: Figure 3 shown.

[0043] Step 41: Preprocess the collected data

[0044] (1) The quartile method was used for data screening. First, the first quartile (Q1), the second quartile (median Q2), and the third quartile (Q3) of the data set were calculated using SPSS software. Second, the distribution range of each value in the data set was determined by calculating the interquartile range. Furthermore, based on the IQR value, the upper and lower limits of the outliers were calculated to screen out outliers that exceeded the range.

[0045] (2) The K nearest neighbor algorithm is used to repair missing values ​​and outliers. By comparing the state vector of the data to be repaired with the state vector of the historical data, the K neighboring values ​​that are most similar to the state vector of the data to be repaired are found, and their weighted average is calculated as the estimated value of the data to be repaired.

[0046] (3) Normalize the preprocessed data and convert the input data set (influencing factors) into a range of 0-1 using linear scaling;

[0047] Step 42: Use the correlation coefficient method to select feature factors. First, remove variables with a small correlation with the target variable (|r|<0.2); second, if there are multiple variables that all satisfy |r|>0.8, only retain the variable with a larger correlation coefficient with the target variable.

[0048] Step 43: Use support vector regression (SVR), artificial neural network (ANN) and XGBoost decision tree machine learning algorithms to build relationship models between influencing factors and various types of construction consumption. The consumption of building materials, water and energy is selected as the prediction model target, and the preprocessed data is used to train and test each prediction model. The data uses ten-fold cross validation, and the training set and test set are divided into 9:1 ratio.

[0049] The SVR algorithm process includes:

[0050] Given a data set (x i ,y i),i=1,2,...,N,x i ∈R n ,y i ∈R, where x i is the i-th dimension input in the n-dimensional space, representing the influencing factors of building embodied carbon emissions, y i Should i The actual output represents the construction consumption of this type. The objective function of the construction consumption prediction model is:

[0051]

[0052] The constraints are:

[0053] y i -(ω T x i +b)≤ε+ξ i

[0054]

[0055] In the formula, ω is the weight vector, b is the bias term, C is the penalty parameter, ξ i and is the slack variable, ε is the tolerance;

[0056] The particle swarm algorithm is used to determine the optimal parameters of the SVR model. Figure 4 shown) include:

[0057] Step 1: Initialize the particle position s i and speed v i , set the best position pbest for each particle i As its initial position, set the global best position gbest as the best position of all particles;

[0058] Step 2: Calculate the fitness of each particle and update pbest i and gbest,

[0059] Step 3: Update the position s of each particle i (t+1) and speed v i (t+1),

[0060] The particle velocity update formula is:

[0061] v i (t+1)=ω·v i (t)+c1·r1·(pbest i -s i (t))+c2·r2·(gbest-s i (t))

[0062] The particle position update formula is:

[0063] s i (t+1)=s i (t)+v i (t+1)

[0064] In the formula, c1 is the individual learning factor; c2 is the social learning factor; r1 and r2 are random numbers between 0 and 1;

[0065] Step 4: Repeat steps 2 and 3 until the maximum number of iterations or the fitness threshold is reached, and output the global best position as the optimal parameter of the prediction model.

[0066] The ANN algorithm process includes: first, initializing the weights and bias parameters; second, passing the input data to the output layer layer by layer after being processed by the weights and activation functions; then updating the weights and biases according to the errors of the output results; and finally, performing forward propagation and back propagation multiple times, continuously adjusting the weights and biases until the network converges.

[0067] The XGBoost decision tree algorithm process includes: first, initializing the model and defining the initial prediction value as the average value of the basic model; second, calculating the residual according to the negative gradient of the loss model; third, constructing a regression tree to fit the residual, and using a greedy algorithm to select the best split point; updating the prediction value of the model through the new tree, and repeating the above steps until the maximum number of iterations is reached or the absolute value of the residual is less than the set threshold.

[0068] The mean square error (RMSE), mean absolute error (MAE) and absolute percentage error (MAPE) are selected as evaluation indicators of model error and accuracy, and the coefficient of determination (R2) is selected as the evaluation indicator of interpretability. The evaluation indicators are calculated by the following formula:

[0069]

[0070]

[0071] In the formula, y i is the true value of the test set, is the predicted value, is the mean value and n is the number of samples.

[0072] Step 44: The smaller the RMSE, MAE, and MAPE values, the better the model performance, and the larger the R2 value, the better the fit between the model and the data. By comparing the evaluation indicators of each machine learning model, select the algorithm with the best performance.

[0073] (5) Combine the consumption prediction results with the carbon emission factor to calculate the building embodied carbon emissions and obtain the building embodied carbon emissions prediction model. The calculation formula is as follows:

[0074]

[0075] In the formula, is the total amount of embodied carbon emissions per unit building area; m w is the water consumption per unit building area; EF w is the carbon emission factor of water; m i is the consumption of building material i per unit building area; EF i is the carbon emission factor of building material i; m j is the consumption of energy type j per unit building area; EF j is the carbon emission factor of energy type j.

[0076] (6) Based on the above process, a building embodied carbon emissions prediction tool was developed using Python application software.

[0077] The functional framework of the tool includes platform layer, data layer and application layer, such as Figure 5 As shown in the figure; at the platform level, users are required to provide specific data of influencing factors based on the project being evaluated. In addition, users can freely modify the data of some parameters according to actual conditions. The data layer stores data and trains machine learning models. At the application layer, the building's embodied carbon emissions are calculated, visualized, and analyzed.

[0078] The tool usage process is as follows Figure 6 As shown, the usage process includes five user interfaces such as Figure 7 As shown. Figure 7 (a) shows the user entry interface, where the user clicks “Start Calculation” to start. Figure 7 (b) shows the basic project information collection interface, which requires users to provide project information, including project name, location, construction time, etc. At the bottom of the interface, buttons are provided to jump to different types of activity data interfaces; Figure 7 (c) shows the data management interface. Users can enter data according to the prompts and freely modify the carbon emission factor. The tool uses the data provided by the user to predict the concrete consumption of the building, combines it with the carbon emission factor, and evaluates the related embodied carbon emissions. Figure 7 (d) shows the carbon emission results display interface, which visualizes the specific implicit carbon emission results, including the total amount and proportion of various sources.

[0079] It should be noted that the above content only illustrates the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications all fall within the protection scope of the claims of the present invention.

Claims

1. A method for predicting building embodied carbon emissions based on machine learning, characterized in that: The following steps are involved: S1: Comprehensively consider construction activities, qualitatively and quantitatively analyze the implicit carbon emission structure of buildings, and select the main sources of carbon emissions as the output parameters of the target prediction model. The implicit carbon emission sources of buildings include building materials, water and energy; S2: Systematically identify 40 factors affecting building embodied carbon emissions from three aspects: project characteristics, construction characteristics, and management level; perform grey correlation analysis on the identified factors, calculate grey correlation, and select the top 30 results with the largest correlation as the input features of the building embodied carbon emissions prediction model; S3: Data collection is carried out for actual engineering projects using a data collection matrix, which includes project characteristics, construction features, management level, and multiple dimensions and characteristics of material, water, and energy consumption during construction; S4: Clean and normalize the data collected in S3, and remove redundant and irrelevant feature factors through feature selection; use support vector regression, artificial neural network and XGBoost decision tree algorithm to build a relationship model between the three types of influencing factors and construction consumption, and select the algorithm with the best performance to build a consumption prediction model; S5: Combine the consumption prediction result with the carbon emission factor to calculate the building embodied carbon emissions and obtain the building embodied carbon emissions prediction model. The calculation formula is as follows: In the formula, is the total amount of embodied carbon emissions per unit building area; m w water consumption per unit of building area; EF w is the carbon emission factor of water; m i is the consumption of building material i per unit building area; EF i is the carbon emission factor of building material i; m j is the energy consumption per unit building area of ​​energy type j; EF j is a carbon emission factor of energy type j; S6: According to the method described in steps S1-S5, use Python to develop a building embodied carbon emission prediction device.

2. According to claim 1, a method for predicting building embodied carbon emissions based on machine learning is characterized in that: The calculation process of the grey correlation degree of influencing factors in step S2 is as follows: Step 1: Select the total embodied carbon emissions of buildings as the reference series x0 = {x0(1), x0(2), ..., x0(u)}, and the i-th influencing factor of the embodied carbon emissions of buildings as the comparison series x i ={x i (1),x i (2),...,x i (u)}, where i = 1, 2, ..., m, u is the number of buildings collected; m is the number of influencing factors identified; Step 2: Use maximum and minimum normalization to make the data dimensionless and obtain x'0 = {x'0(1), x'0(2), ..., x'0(u)} and x' i ={x' i (1),x' i (2),...,x' i (u)}, the normalized calculation formula is as follows: Where: x j (k) represents the kth data in the jth sequence; x′ j (k) represents x j (k) Normalized result; Represent the maximum and minimum values ​​in the jth sequence respectively; Step 3: Calculate the grey correlation coefficient for the normalized data, and record x0(k) and x i (k) The correlation number at point k is γ(x0′(k), x i ′(k)), the grey relational coefficient calculation formula is as follows: Where: Δ 0i (k) = x i ′(k)-x0'(k), ζ is the resolution coefficient, ζ=0.5 Step 4: Calculate the correlation γ between the comparison sequence and the reference sequence i , the correlation calculation formula is as follows: Step 5: Sort the comparison series according to the degree of correlation, and select the top 30 results with the largest correlation as the input features of the building embodied carbon emissions prediction model.

3. The method for predicting building embodied carbon emissions based on machine learning according to claim 1 is characterized in that: The steps for building the machine learning prediction model in step S4 are as follows: Step 1: Preprocess the collected data, including: (1) The quartile method was used for data screening; first, the first quartile, second quartile and third quartile of the data set were calculated using SPSS software; second, the distribution range of each value in the data set was determined by calculating the interquartile range; third, the upper and lower limits of the outliers were calculated based on the IQR value, and outliers that exceeded the range were screened out; (2) Use the K nearest neighbor algorithm to repair missing values ​​and outliers. By comparing the state vector of the data to be repaired with the state vector of the historical data, find the K neighboring values ​​that are most similar to the state vector of the data to be repaired, and calculate their weighted average as the estimated value of the data to be repaired; (3) Normalize the preprocessed data and use linear scaling to convert the input data set into a range of 0-1; Step 2: Use the correlation coefficient method to select feature factors; first, remove variables with a small correlation with the target variable; second, if there are multiple variables that all satisfy |r|>0.8, only retain the variable with a large correlation coefficient with the target variable; Step 3: Use the support vector regression SVR algorithm, artificial neural network ANN algorithm, and XGBoost decision tree algorithm to build the relationship model between the influencing factors and the consumption of various types of construction; select the consumption of building materials, water, and energy as the prediction model targets, and use the preprocessed data to train and test each prediction model. The data uses a ten-fold cross validation, and the training set and the test set are divided into a ratio of 9:1; Step 4: Select mean square error RMSE, mean absolute error MAE and absolute percentage error MAPE as evaluation indicators of model error and accuracy, select coefficient of determination R2 as the evaluation indicator of interpretability, and select the model constructed by the algorithm with the best performance as the final consumption prediction model.

4. According to the method for predicting building embodied carbon emissions based on machine learning as described in claim 3, it is characterized in that: The machine learning prediction model in step S4 uses the support vector regression SVR algorithm, the artificial neural network ANN algorithm, and the XGBoost decision tree algorithm. The algorithm processes are as follows: 1) The SVR algorithm process includes: Given dataset (x i ,y i ),i=1,2,...,N,x i ∈R n ,y i ∈R, where x i is the i-th dimension input in the n-dimensional space, representing the influencing factors of building embodied carbon emissions, y i Should x i The actual output represents the construction consumption of this type. The objective function of the construction consumption prediction model is: The constraints are: y i -(oh T x i +b)≤ε+ξ i In the formula, ω is the weight vector, b is the bias term, C is the penalty parameter, ξ i and is the slack variable, ε is the tolerance; The particle swarm algorithm is used to determine the optimal parameters of the SVR model; first, the position and velocity of the particle swarm are initialized, then the fitness of each particle is calculated, and then the position and velocity of the particle are updated according to the fitness of the particle. The above steps are repeated until the maximum number of iterations or the fitness threshold is reached, and the position of the global optimal particle is output as the optimal parameter of the SVR model; 2) The ANN algorithm process includes: first, initializing weights and bias parameters; second, passing input data layer by layer to the output layer after being processed by weights and activation functions; third, updating weights and biases according to the error of the output results; and finally, performing forward propagation and back propagation multiple times to continuously adjust weights and biases until the network converges; 3) The XGBoost decision tree algorithm process includes: first, initializing the model and defining the initial prediction value as the average value of the basic model; second, calculating the residual according to the negative gradient of the loss model; then, building a regression tree to fit the residual, and using a greedy algorithm to select the best split point; updating the prediction value of the model through the new tree, and repeating the above steps until the maximum number of iterations is reached or the absolute value of the residual is less than the set threshold.

5. The method for predicting building embodied carbon emissions based on machine learning according to claim 4 is characterized in that: The process of determining the optimal prediction model parameters by the particle swarm algorithm is as follows: Step 1: Initialize the position of the particle i and speed v i , set the optimal position of each particle pbest i As its initial position, set the global best position gbest as the best position of all particles; Step 2: Calculate the fitness of each particle and update pbest i and gbest, Step 3: Update the position s of each particle i (t+1) and speed v i (t+1), The particle velocity update formula is: v i (t+1)=ω·v i (t)+c1·r1·(pbest i -s i (t))+c2·r2·(gbest-s i (t)) The particle position update formula is: s i (t+1)=s i (t)+v i (t+1) In the formula, c1 is the individual learning factor; c2 is the social learning factor; r1 and r2 are random numbers between 0 and 1; Step 4: Repeat steps 2 and 3 until the maximum number of iterations or the fitness threshold is reached, and output the global best position as the optimal parameter of the prediction model.

6. The method for predicting building embodied carbon emissions based on machine learning according to claim 3 is characterized in that: The machine learning model evaluation index is calculated by the following formula: In the formula, y i is the true value of the test set, is the predicted value, is the mean value, n is the number of samples; The smaller the RMSE, MAE and MAPE values, the better the model performance; the larger the R2 value, the better the fit between the model and the data; by comparing the evaluation indicators of each machine learning model, the algorithm with the best performance is selected.

7. The method for predicting building embodied carbon emissions based on machine learning according to claim 1 is characterized in that: The functional framework of the building embodied carbon emission prediction device described in step S6 includes a platform layer, a data layer and an application layer; at the platform layer, the user provides information data of the project to be evaluated according to the project to be evaluated, and the user is also allowed to modify some built-in parameter data according to the actual situation of the project; the data layer stores data and machine learning models; at the application layer, the building embodied carbon emissions are predicted, visualized and analyzed; The device includes 4 functional modules; Project basic information collection module: users input basic project information such as project name, project location, construction time, and number of floors above ground; Data management module: fill in the data information of influencing factors in three aspects: project characteristics, construction characteristics and management level, and modify the values ​​of built-in parameters; Construction consumption prediction module: displays the construction consumption results predicted by the machine learning model in the form of charts; Carbon emission result display module: Displays the construction carbon emissions of the assessed buildings in a graphical form and conducts proportional analysis by source.

8. A system capable of implementing the method for predicting building embodied carbon emissions based on machine learning as described in claim 1, characterized in that: The system comprises a processor and a memory, wherein the memory contains a program and runs a pre.bat script installation program operating environment, and the program is loaded and executed by the processor to implement the steps of the building embodied carbon emission prediction method.

Images