A Machine Learning-Based Method for Predicting Carbon Emissions from Integrated Energy Station Construction

By constructing a machine learning-based carbon emission prediction model for integrated energy station construction, and using the XGBoost algorithm for training and testing, the problems of cumbersome calculations and difficult data collection in existing technologies are solved, achieving efficient carbon emission prediction that is applicable to the construction process of integrated energy stations.

CN116362376BActive Publication Date: 2026-11-10STATE GRID JIANGSU ELECTRIC POWER CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202310130789.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-02-17
Publication Date
2026-11-10
Estimated Expiration
2043-02-17

AI Technical Summary

Technical Problem

Existing technologies lack intelligent prediction methods for carbon emissions from the construction of integrated energy stations. The calculation process is cumbersome and data collection is difficult, resulting in poor practical results and making it difficult to promote widely.

Method used

A carbon emission prediction model for the construction of integrated energy stations based on machine learning is constructed. By collecting and processing carbon emission prediction indicators and source data, the XGBoost algorithm is used for model training and testing to achieve carbon emission prediction for new projects.

Benefits of technology

It provides a simplified method for calculating carbon emissions, with excellent model fit, enabling efficient prediction of carbon emissions during the construction process, reducing human intervention, and is universally applicable to a limited number of projects with many data issues.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116362376B_ABST
    Figure CN116362376B_ABST
Patent Text Reader

Abstract

This invention discloses a machine learning-based method for predicting carbon emissions from the construction of integrated energy stations. The method includes: collecting a carbon emission prediction index system for integrated energy station construction and identifying carbon emission sources; preprocessing prediction index data and carbon emission source data for integrated energy station projects; and constructing a machine learning-based carbon emission prediction model for integrated energy station construction to predict carbon emissions from integrated energy station construction. This invention provides an accurate, fast, and intelligent method for predicting carbon emissions from integrated energy station construction, reducing the complexity and cumbersome nature of traditional carbon emission calculation processes, and providing effective and reliable data support for low-carbon design and construction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of carbon emission prediction technology, and in particular to a method for predicting carbon emissions from the construction of integrated energy stations based on machine learning. Background Technology

[0002] Machine learning methods have significant application potential in predictive research for integrated energy stations and buildings; however, current research on carbon emissions from integrated energy stations is lacking. In related research, the analysis and prediction of carbon emissions during the construction of integrated energy stations are insufficiently emphasized. Furthermore, in the actual work of assessing carbon emissions during the construction of integrated energy stations, current carbon emission calculation methods mostly focus on in-process or post-process calculations, relying on large amounts of design and construction data. The calculation process is cumbersome, and due to the large number of participating units, poor data integrity, high technical complexity, and difficulties in data collection, the calculation of carbon emissions during the construction phase of integrated energy stations is difficult, resulting in poor practical effects and limited widespread implementation. Therefore, there is an urgent need for an intelligent prediction method that combines data preprocessing and machine learning. In machine learning algorithms, XGBoost is a novel ensemble learning algorithm based on Boosting. Compared with traditional machine learning algorithms, it has advantages such as fast running speed, strong generalization ability, high prediction accuracy, and good robustness. In addition, the XGBoost algorithm model has high interpretability and can be used for small sample prediction. It has been widely used in runoff prediction, credit card transaction, project investment prediction, and fault monitoring, but it has not yet been deeply applied in the field of carbon emission prediction. Summary of the Invention

[0003] To address the above problems, this invention proposes a machine learning-based method for predicting carbon emissions from the construction of integrated energy stations. First, it constructs a carbon emission prediction index system for integrated energy station construction from the perspectives of project composition, construction process, and carbon footprint, and identifies carbon emission sources. Second, it collects and processes relevant data on the prediction indicators and carbon emission sources. Then, it constructs a carbon emission prediction model for integrated energy station construction based on machine learning algorithms and trains, tests, and evaluates the model. Finally, it implements a practical application by inputting the prediction index data into the carbon emission prediction model to obtain the predicted carbon emission values ​​for newly constructed integrated energy station projects.

[0004] Technical Solution: This invention proposes a carbon emission prediction method for the construction of integrated energy stations based on machine learning. The method includes the following steps:

[0005] Step 1: Collect carbon emission prediction index system for the construction of integrated energy stations and identify carbon emission sources;

[0006] Step 2: Preprocess carbon emission prediction data and carbon emission source data for the construction of integrated energy stations;

[0007] Step 3: Construct a carbon emission prediction model for the construction of integrated energy stations based on machine learning algorithms, and train the model using the data from Step 2;

[0008] Step 4: Input the predicted index data into the carbon emission prediction model for the construction of the integrated energy station, and the model outputs the carbon emission value for the construction of the integrated energy station, thus completing the prediction of the carbon emission of the newly built integrated energy station.

[0009] Furthermore, in step 1, the carbon emission prediction index system for the construction of the integrated energy station includes the number of floors above ground, foundation depth, total building area, foundation earthwork volume, concrete design volume, and steel reinforcement design weight; carbon emission sources include the steel, concrete, electricity consumption, and water consumption required for the construction of the integrated energy station.

[0010] Furthermore, in step 2, the specific methods for preprocessing the carbon emission prediction index data and carbon emission source data for the construction of the integrated energy station are as follows:

[0011] Step 2-1: Convert the carbon emission source data consumed in the construction of the integrated energy station into carbon emission source data per unit building area;

[0012] Step 2-2: Analyze the carbon emission source data per unit building area using box plots to identify abnormal data as data to be repaired.

[0013] Steps 2-3: Use the K-nearest neighbor algorithm to replace missing or outlier values ​​in the carbon emission source data per unit building area;

[0014] Steps 2-4: Using the replaced carbon emission source data per unit building area, combined with the carbon emission factors of each carbon emission source, calculate the carbon emission value of the integrated energy station construction per unit building area, and use it as the output variable of the integrated energy station construction carbon emission prediction model.

[0015] Steps 2-5: Use the Min-Max normalization method to process the carbon emission prediction indicators for each type of integrated energy station construction, and use them as input variables for the integrated energy station construction carbon emission prediction model.

[0016] Furthermore, in step 2-1, the carbon emission source data of the integrated energy station construction is converted into carbon emission source data per unit building area using formula (1);

[0017] (1)

[0018] In the formula:

[0019] Let i be the carbon emission rate per unit building area of ​​the i-th carbon emission source, where i is the type of carbon emission source.

[0020] Let be the total amount of carbon emissions from the i-th source;

[0021] S represents the total building area of ​​the integrated energy station.

[0022] Furthermore, the method in step 2-2 is as follows: Take the 25% quantile Q1 and 75% quantile Q2 of all samples. The difference between Q1 and Q2 is the length of the box IQR. When the carbon emission source value per unit building area is less than (Q1-1.5IQR) or greater than (Q2+1.5IQR), it is an abnormal sample.

[0023] Furthermore, the method in steps 2-3 is as follows: the K-nearest neighbor algorithm is used to calculate the Euclidean distance between the predicted indicators of normal projects and the predicted indicators of projects to be repaired, and the average value of carbon emission source data per unit building area of ​​the k nearest normal projects is selected to replace the missing or outlier values ​​in the carbon emission source data per unit building area.

[0024] Furthermore, the method in steps 2-4 is as follows: The carbon emission source data per unit building area is multiplied and summed with the carbon emission factor to calculate the carbon emission of the integrated energy station per unit building area. This sum is used as the output variable of the integrated energy station construction carbon emission prediction model. The calculation model is shown in the formula:

[0025] (2)

[0026] In the formula:

[0027] Carbon emissions per unit building area during the construction of an integrated energy station;

[0028] The amount of carbon emitted per unit building area is denoted as i;

[0029] F i Let i be the carbon emission factor of the i-th type of carbon emission source, where i is the type of carbon emission source, and there are N types in total.

[0030] The output variable dataset is obtained through calculation. .

[0031] Furthermore, the methods in steps 2-5 are as follows: Carbon emission prediction indicators for each type of integrated energy station construction are processed separately and used as input variables for the integrated energy station construction carbon emission prediction model, forming an input variable set. ,in, , represents the dataset of the j-th predictive index after normalization, which consists of the normalized values ​​of the j-th predictive index of m items;

[0032] (3)

[0033] In the formula:

[0034] This represents the value of the i-th item after normalization for the j-th prediction index.

[0035] These represent the maximum and minimum values ​​of the j-th predictive indicator within the same category, respectively;

[0036] This represents the original value of the i-th item in the j-th prediction indicator;

[0037] j represents the category of predictive indicators, and i represents different items. The number of representative projects.

[0038] Furthermore, in step 3, the method for constructing a carbon emission prediction model for integrated energy station construction based on machine learning algorithms and training the model is as follows:

[0039] Step 3-1: Divide the processed carbon emission prediction index data for integrated energy station construction and the carbon emission data per unit building area for integrated energy station construction into training and test sets;

[0040] Step 3-2: Based on the training set obtained in Step 3-1, divide the training set into K equal-sized subsets. K-1 subsets are used for model training, and the remaining 1 subset constitutes the validation subset.

[0041] Step 3-3: Construct a carbon emission prediction model for the construction of integrated energy stations based on the XGBoost algorithm. The tree ensemble model used in the XGBoost algorithm is trained by addition and trained by adjusting the model parameters through the validation subset based on K-fold cross-validation to obtain multiple carbon emission prediction models for the construction of integrated energy stations.

[0042] Steps 3-4 involve inputting the test set into the carbon emission prediction model for the construction of integrated energy stations, comparing the prediction results of the model with the actual results, and determining the optimal carbon emission prediction model for the construction of integrated energy stations.

[0043] Furthermore, the method in step 3-3 is as follows:

[0044] ① Using the CART classification tree as the base learner, the loss function, objective function, and XGBoost prediction model are set as shown in formulas (4)-(7):

[0045] Define the loss function:

[0046] (4)

[0047] In the formula, Let n represent the actual carbon emissions from the construction of integrated energy stations and the predicted carbon emissions from the construction of integrated energy stations, respectively, where n is the sample size. The loss function;

[0048] Define the objective function:

[0049] (5)

[0050] In the formula, This is the prediction model for the t-th tree. This involves summing the complexity of all t trees and adding it to the regularization term of the objective function to prevent overfitting.

[0051] Setting up the XGBoost prediction model:

[0052] (6)

[0053] In the formula, Both are the training set input samples x after t iterations. i The predicted loss value, This is the predicted loss value for the first t-1 trees. Let be the prediction model for the t-th tree;

[0054] objective function It can be written as shown in equation (7):

[0055]

[0056] (7)

[0057] In the formula, C is a constant term. Here, t is the regularization term, and t is the number of spanning trees. The objective function is the function after generating the t-th tree;

[0058] ② Based on the Taylor formula, the loss function is as shown in formula (8):

[0059] (8)

[0060] In the formula, The first derivative of the loss function, This is the second derivative of the loss function;

[0061] ③ Based on formulas (7) and (8), the objective function can be written as shown in formula (9):

[0062] (9)

[0063] In the formula, As a constant, the objective function is further optimized, as shown in formula (10):

[0064] (10)

[0065] ④ Based on the objective function of formula (10), calculate the first derivative of the loss function at each step when inputting the training set. and second derivative The value of is then used to optimize the objective function to obtain the value at each step. According to formula (6), the overall model is obtained, which is the carbon emission prediction model for the construction of integrated energy stations;

[0066] ⑤ Adjust the model parameters to obtain different carbon emission prediction models for the construction of integrated energy stations.

[0067] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0068] (1) This invention, through analysis of the composition, construction process, and carbon emission sources of integrated energy station projects, constructs a carbon emission prediction index system for integrated energy station construction, clarifies the sources of carbon emissions from integrated energy station construction, and determines the data processing method and carbon emission calculation model. This invention constructs a carbon emission prediction model for integrated energy station construction based on machine learning algorithms, obtains the optimal model through K-fold cross-validation training, and predicts and evaluates the carbon emissions of integrated energy station construction on the test set, thereby completing the prediction of carbon emissions from newly constructed integrated energy station projects.

[0069] (2) On the one hand, this invention provides ideas and methods for predicting carbon emissions during the construction of integrated energy stations, and provides a reference for how to process data. It has high reference value for projects with limited quantity and data problems. On the other hand, the carbon emission prediction model for integrated energy station construction based on machine learning algorithm constructed in this invention has excellent fitting and small model error. It can predict carbon emissions from integrated energy station construction well, thereby avoiding the collection of a large amount of data and tedious calculations in the process of calculating carbon emissions from integrated energy station construction. It requires less human intervention, the prediction process is simple, and it can realize the prediction of carbon emissions from integrated energy station construction in advance. It has strong universality. Attached Figure Description

[0070] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention.

[0071] Figure 2This is a schematic diagram illustrating the outlier identification method using box plots in the present invention.

[0072] Figure 3 This is a schematic diagram illustrating the method of the present invention for constructing and training a model using 5-fold cross-validation. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The embodiments provide a carbon emission prediction process for the construction of integrated energy stations.

[0074] Example

[0075] In order to solve the problems existing in the prior art, the present invention provides a method for predicting carbon emissions from the construction of integrated energy stations based on machine learning. This method can predict the carbon emissions from the construction of integrated energy stations by identifying the influencing factors. Figure 1 This is a flowchart illustrating the implementation of the present invention. The present invention discloses a method for predicting carbon emissions from the construction of integrated energy stations based on machine learning, specifically including the following steps:

[0076] Step 1: Collect carbon emission prediction index system for the construction of integrated energy stations and identify carbon emission sources.

[0077] The carbon emission prediction indicators for integrated energy station construction need to reflect characteristics such as building structure type, building material usage, and engineering volume. Therefore, this invention defines the carbon emission prediction indicators for integrated energy station construction as the number of above-ground floors, foundation depth, total building area, foundation earthwork volume, concrete design volume, and steel reinforcement design weight, forming a prediction indicator system composed of six indicators. Carbon emission sources include the steel, concrete, electricity consumption, and water consumption required for integrated energy station construction.

[0078] Step 2: Collect and process carbon emission prediction indicators and carbon emission source data for the construction of integrated energy stations. The data collection methods for carbon emission prediction indicators and carbon emission source data for the construction of integrated energy stations are shown in Table 1 below.

[0079] Table 1 Data Collection Table

[0080]

[0081] The carbon emission source data of the integrated energy station construction is converted into carbon emission source data per unit building area for data analysis and processing, as shown in formula (1):

[0082] (1)

[0083] In the formula:

[0084] Let represent the carbon emission consumption per unit building area of ​​the i-th carbon emission source, where i represents the type of carbon emission source.

[0085] Let be the total amount of carbon emission source i.

[0086] S represents the total building area.

[0087] Taking a certain project as an example, the building area S = 1182.6m² 2 If the steel consumption M1 = 1200t, then the steel consumption per unit building area of ​​this project is... Other types of carbon emission sources are treated in the same way.

[0088] Furthermore, box plots are used to analyze carbon emission source data per unit building area to identify anomalous data as data to be repaired. The 25th percentile (Q1) and 75th percentile (Q2) of all samples are taken, and the difference between Q1 and Q2 is the length of the box, IQR. Samples with carbon emission source values ​​per unit building area less than (Q1 - 1.5IQR) or greater than (Q2 + 1.5IQR) are considered anomalous. The box plot generated by this invention is shown below. Figure 2 As shown, the points outside the box represent outliers, i.e., the item numbers with abnormal values ​​for a certain carbon emission source. Taking concrete consumption per unit area as an example, the concrete consumption per unit building area for items 5, 6, 33, and 46 is abnormal and is therefore subject to repair. Other types of carbon emission source data per unit building area are processed in the same way.

[0089] Furthermore, the K-nearest neighbor algorithm is used to calculate the Euclidean distance between the predicted indicators of normal projects and the predicted indicators of projects to be repaired, and the average carbon emission source per unit building area of ​​the normal projects with the k nearest Euclidean distances is selected to replace the missing or outlier values ​​in the carbon emission source data per unit building area.

[0090] Furthermore, after data repair, the carbon emission source data per unit building area is multiplied and summed with the carbon emission factor to calculate the carbon emission of the integrated energy station per unit building area. This carbon emission is used as the output variable of the carbon emission prediction model for the integrated energy station construction. The calculation model is shown in formula (2):

[0091] (2)

[0092] In the formula:

[0093] Carbon emissions per unit building area during the construction of integrated energy stations (unit: kgCO2e / m²) 2 ).

[0094] Let be the consumption per unit building area of ​​the i-th carbon emission source.

[0095] F i Let be the carbon emission factor of the i-th type of carbon emission source, where i represents the type of carbon emission source, and there are a total of 4 types.

[0096] The output variable dataset is obtained through a computational model. .

[0097] The carbon emission sources per unit building area of ​​the i-th integrated energy station project are shown in Table 2 below, and the carbon emission factors of each carbon emission source are shown in Table 3. The carbon emission per unit building area of ​​the i-th integrated energy station project is calculated using formula (2).

[0098] Table 2 Carbon emission consumption per unit building area

[0099]

[0100] Table 3 Carbon emission factors of various carbon emission sources

[0101]

[0102] Furthermore, the Min-Max normalization method is used to process the carbon emission prediction indicators for the construction of various integrated energy stations, as shown in formula (3). Each type of integrated energy station construction carbon emission prediction indicator is processed separately and used as the input variable of the integrated energy station construction carbon emission prediction model to form an input variable set. ,in .

[0103] (3)

[0104] In the formula:

[0105] This represents the value of the i-th item after normalization for the j-th prediction index.

[0106] These represent the maximum and minimum values ​​of the j-th predictive indicator, respectively.

[0107] This represents the original value of the i-th item in the j-th prediction indicator.

[0108] j represents the category of predictive indicators, and i represents different items. The number of representative projects.

[0109] The results of the normalization process are shown in Table 4.

[0110] Table 4. Results of Min-Max Normalization

[0111]

[0112] Step 3: Construct a carbon emission prediction model for the construction of integrated energy stations based on machine learning algorithms.

[0113] Step 3-1: Divide the processed carbon emission prediction index data for integrated energy station construction and the carbon emission data per unit building area for integrated energy station construction into training and test sets;

[0114] Before constructing a predictive model for the construction of integrated energy stations based on machine learning algorithms, the input dataset obtained in step 2 is used as a basis. and output dataset We used the sklearn package in Python to divide the dataset into training and testing sets, and set the ratio to 7:3.

[0115] Step 3-2: Based on the training set obtained in step 3-1, the dataset is divided into K equal-sized subsets. K-1 subsets are used for model training, and the remaining 1 subset constitutes the validation subset.

[0116] To eliminate the impact of randomness in sample partitioning and sorting on prediction results, the dataset is further divided into K equal-sized subsets based on the training set obtained after data partitioning. K-1 subsets are used for model building, and the remaining subset is used for model validation. The mean of the evaluation metrics from these K trials is used as an estimate of model accuracy. In practice, K is typically chosen to be between 5 and 10; in this invention, K=5 is used. Figure 3 As shown.

[0117] Step 3-3: Construct a carbon emission prediction model for the construction of integrated energy stations based on the XGBoost algorithm. The tree ensemble model used in the XGBoost algorithm is trained by addition and trained by adjusting the model parameters through the validation subset based on K-fold cross-validation to obtain multiple carbon emission prediction models for the construction of integrated energy stations.

[0118] Steps 3-4 involve inputting the test set into the carbon emission prediction model for the construction of integrated energy stations, comparing the prediction results of the model with the actual results, and determining the optimal carbon emission prediction model for the construction of integrated energy stations.

[0119] A carbon emission prediction model for integrated energy station construction based on the XGBoost algorithm is constructed. The tree ensemble model used in the XGBoost algorithm is trained by addition until the tree depth threshold is reached, at which point splitting stops. The carbon emission prediction model for integrated energy station construction is then saved. This invention constructs a carbon emission prediction model for integrated energy station construction based on machine learning algorithms in Python, including the following steps:

[0120] (1) Using the CART classification tree as the base learner, the loss function, objective function and XGBoost prediction model are set as shown in formulas (4)-(7):

[0121] Define the loss function:

[0122] (4)

[0123] In the formula, Let n represent the actual carbon emissions from the construction of integrated energy stations and the predicted carbon emissions from the construction of integrated energy stations, respectively, where n is the sample size. The loss function;

[0124] Define the objective function:

[0125] (5)

[0126] In the formula, This is the prediction model for the t-th tree. This involves summing the complexity of all t trees and adding it to the regularization term of the objective function to prevent overfitting.

[0127] Setting up the XGBoost prediction model:

[0128] (6)

[0129] In the formula, Both are the training set input samples x after t iterations. i The predicted loss value, This is the predicted loss value for the first t-1 trees. Let be the prediction model for the t-th tree;

[0130] objective function It can be written as shown in equation (7):

[0131]

[0132] (7)

[0133] In the formula, C is a constant term. Here, t is the regularization term, and t is the number of spanning trees. Let t be the objective function after generating the t-th tree.

[0134] (2) Based on the Taylor formula, the loss function is as shown in formula (8):

[0135] (8)

[0136] In the formula, The first derivative of the loss function, This is the second derivative of the loss function;

[0137] (3) Based on formulas (7) and (8), the objective function can be written as shown in formula (9):

[0138] (9)

[0139] In the formula, As a constant, the objective function is further optimized, as shown in formula (10):

[0140] (10)

[0141] (4) Based on the objective function of formula (10), when inputting the training set, it is only necessary to calculate the first derivative of the loss function at each step. and second derivative The value of is then used to optimize the objective function to obtain the value at each step. According to formula (6), the overall model is obtained, which is the carbon emission prediction model for the construction of integrated energy stations;

[0142] (5) Adjust the model parameters to obtain different carbon emission prediction models for the construction of integrated energy stations.

[0143] This invention adjusts the parameters of a machine learning model and saves the model with the best performance. It builds and adjusts the model based on the XGBRegressor package in Python's machine learning library. Specific parameters include:

[0144] ①Booster is used to select the model for each iteration. There are generally two options: gbtree and gbliner.

[0145] ②learning_rate is the learning rate, ranging from [0,1]. The smaller this parameter is, the slower the calculation speed; the larger this parameter is, the less likely it is to converge.

[0146] ③max_depth is the maximum depth of each tree, ranging from [0,+∞). The larger this parameter is, the easier it is to overfit. The larger the max_depth is, the more specific and local the model will learn.

[0147] ④ n_estimators is the number of trees in machine learning. The more trees there are, the better the model performance. However, when the number reaches a certain level, the improvement in model performance is limited, and it will actually slow down the algorithm.

[0148] ⑤colsample_bytree is the column sampling rate, which is generally the feature sampling rate. It uses column sampling similar to random forest for the features used to generate each tree, with a range of (0,1).

[0149] ⑥ min_child_weight is the minimum sum of weights within each leaf, ranging from [0, +∞). The larger this parameter is, the more conservative the algorithm is, and the less prone it is to overfitting.

[0150] ⑦ lambda is the L2 regularization parameter, used to control the regularization part of machine learning. Its range is [0, +∞). The larger this parameter is, the less likely it is to overfit.

[0151] ⑧ Gamma is the loss threshold, which is a parameter that controls the number of leaves. Gamma specifies the minimum decrease in the loss function required for node splitting, ranging from [0, +∞). The larger this parameter is, the more conservative the algorithm is and the less likely it is to overfit.

[0152] The training results of the model are evaluated, and the test set is input into the carbon emission prediction model for the construction of integrated energy stations for verification. The prediction results of the model are compared with the actual results to evaluate the model. After the evaluation is passed, the best carbon emission prediction model for the construction of integrated energy stations is determined based on the evaluation effect of the model. The model evaluation is based on formulas (11)-(14):

[0153] (11)

[0154] (12)

[0155] (13)

[0156] (14)

[0157] In the formula, It is the actual value of the i-th sample. It is the predicted value of the i-th sample. R is the sample mean, and n is the sample size. In this invention, when R... 2 A model is considered to have acceptable accuracy if its accuracy is greater than or equal to 0.8 and its MAPE is less than or equal to 20%, and the model is then saved. Among all saved models, the accuracy of MSE, MAE, and R is evaluated. 2 The optimal model is selected by comparing the MAPE model indicators. In this invention, the model with the minimum value between MAE and MSE is chosen as the optimal model, provided that R² ≥ 0.9 and MAPE ≤ 10%. Based on the above four model evaluation indicators, this invention preserves the optimal prediction model, and the model parameters are shown in Table 5 below.

[0158] Table 5 Parameter selection for the optimal machine learning model

[0159]

[0160] The performance metrics of the optimal machine learning model are shown in Table 6:

[0161] Table 6 Performance of the Optimal Machine Learning Model

[0162]

[0163] Step 4, practical application: Input the predicted index data into the prediction model, and the model outputs the carbon emission values ​​of the integrated energy station construction. The prediction results are shown in Table 7. The carbon emission prediction method for integrated energy station construction based on machine learning algorithms proposed in this study achieves high accuracy, with the maximum error within 10% and the average absolute error of 5.8%, demonstrating high practical and predictive value.

[0164] Table 7. Predicted Carbon Emissions from Integrated Energy Station Construction

[0165]

[0166] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting carbon emissions from the construction of integrated energy stations based on machine learning, characterized in that, The method includes the following steps: Step 1: Collect carbon emission prediction index system for the construction of integrated energy stations and identify carbon emission sources; Step 2: Preprocess carbon emission prediction data and carbon emission source data for the construction of integrated energy stations; Step 3: Construct a carbon emission prediction model for the construction of integrated energy stations based on machine learning algorithms, and train the model using the data from Step 2; Step 4: Input the prediction index data into the carbon emission prediction model for the construction of integrated energy stations, and the model outputs the carbon emission value for the construction of integrated energy stations to predict the carbon emission of newly built integrated energy stations. In step 3, the method for constructing a carbon emission prediction model for integrated energy station construction based on machine learning algorithms and training the model is as follows: Step 3-1: Divide the processed carbon emission prediction index data for integrated energy station construction and the carbon emission data per unit building area for integrated energy station construction into training and test sets; Step 3-2: Based on the training set obtained in Step 3-1, divide the training set into K equal-sized subsets. K-1 subsets are used for model training, and the remaining 1 subset constitutes the validation subset. Step 3-3: Construct a carbon emission prediction model for the construction of integrated energy stations based on the XGBoost algorithm. The tree ensemble model used in the XGBoost algorithm is trained by addition and trained by adjusting the model parameters through the validation subset based on K-fold cross-validation to obtain multiple carbon emission prediction models for the construction of integrated energy stations. Steps 3-4: Input the test set into the carbon emission prediction model for the construction of the integrated energy station for prediction, compare the prediction results of the prediction model with the actual results, and determine the best carbon emission prediction model for the construction of the integrated energy station. The method for step 3-3 is as follows: ① Using the CART classification tree as the base learner, the loss function, objective function, and XGBoost prediction model are set as shown in formulas (4)-(7): Define the loss function: (4) In the formula, Let n represent the actual carbon emissions from the construction of integrated energy stations and the predicted carbon emissions from the construction of integrated energy stations, respectively, where n is the sample size. The loss function; Define the objective function: (5) In the formula, This is the prediction model for the t-th tree. This involves summing the complexity of all t trees and adding it to the regularization term of the objective function to prevent overfitting. Setting up the XGBoost prediction model: (6) In the formula, Both are the training set input samples x after t iterations. i The predicted loss value, This is the predicted loss value for the first t-1 trees. Let be the prediction model for the t-th tree; objective function It can be written as shown in equation (7): ; (7) In the formula, C is a constant term. Here, t is the regularization term, and t is the number of spanning trees. The objective function is the function after generating the t-th tree; ② Based on the Taylor formula, the loss function is as shown in formula (8): (8) In the formula, The first derivative of the loss function, This is the second derivative of the loss function; ③ Based on formulas (7) and (8), the objective function can be written as shown in formula (9): (9) In the formula, As a constant, the objective function is further optimized, as shown in formula (10): (10) ④ Based on the objective function of formula (10), calculate the first derivative of the loss function at each step when inputting the training set. and second derivative The value of is then used to optimize the objective function to obtain the value at each step. According to formula (6), the overall model is obtained, which is the carbon emission prediction model for the construction of integrated energy stations; ⑤ Adjust the model parameters to obtain different carbon emission prediction models for the construction of integrated energy stations.

2. The method for predicting carbon emissions from the construction of integrated energy stations based on machine learning according to claim 1, characterized in that, In step 1, the carbon emission prediction index system for the construction of the integrated energy station includes the number of floors above ground, foundation depth, total building area, foundation earthwork volume, concrete design volume, and steel reinforcement design weight; carbon emission sources include the consumption of steel, concrete, electricity, and water required for the construction of the integrated energy station.

3. The method for predicting carbon emissions from the construction of integrated energy stations based on machine learning according to claim 1, characterized in that, In step 2, the specific methods for preprocessing the carbon emission prediction index data and carbon emission source data for the construction of integrated energy stations are as follows: Step 2-1: Convert the carbon emission source data consumed in the construction of the integrated energy station into carbon emission source data per unit building area; Step 2-2: Analyze the carbon emission source data per unit building area using box plots to identify missing or abnormal data as data to be repaired. Steps 2-3: Use the K-nearest neighbor algorithm to replace missing or outlier values ​​in the carbon emission source data per unit building area; Steps 2-4: Using the replaced carbon emission source data per unit building area, combined with the carbon emission factors of each carbon emission source, calculate the carbon emission value of the integrated energy station construction per unit building area, and use it as the output variable of the integrated energy station construction carbon emission prediction model. Steps 2-5: Use the Min-Max normalization method to process the carbon emission prediction indicators for each type of integrated energy station construction, and use them as input variables for the integrated energy station construction carbon emission prediction model.

4. The method for predicting carbon emissions from the construction of an integrated energy station based on machine learning according to claim 3, characterized in that, In step 2-1, the carbon emission source data of the integrated energy station construction is converted into carbon emission source data per unit building area using formula (1); (1) In the formula: Let i be the carbon emission rate per unit building area of ​​the i-th carbon emission source, where i is the type of carbon emission source. Let be the total amount of carbon emissions from the i-th source; S represents the total building area of ​​the integrated energy station.

5. The method for predicting carbon emissions from the construction of integrated energy stations based on machine learning according to claim 3, characterized in that, The method in step 2-2 is as follows: Take the 25% quantile Q1 and 75% quantile Q2 of all samples. The difference between Q1 and Q2 is the length of the box IQR. When the carbon emission source value per unit building area is less than (Q1-1.5IQR) or greater than (Q2+1.5IQR), it is an abnormal sample.

6. A method for predicting carbon emissions from the construction of integrated energy stations based on machine learning, as described in claim 3 or 5, characterized in that, The method in steps 2-3 is as follows: Use the K-nearest neighbor algorithm to calculate the Euclidean distance between the predicted indicators of normal projects and the predicted indicators of projects to be repaired, and select the average carbon emission source data per unit building area of ​​the K nearest normal projects to replace the missing or outlier values ​​in the carbon emission source data per unit building area.

7. The method for predicting carbon emissions from the construction of integrated energy stations based on machine learning according to claim 3, characterized in that, The method in steps 2-4 is as follows: The carbon emission source data per unit building area is multiplied by the carbon emission factor and summed to calculate the carbon emission of the integrated energy station per unit building area. This sum is used as the output variable of the integrated energy station construction carbon emission prediction model. The calculation model is shown in the formula: (2) In the formula: Carbon emissions per unit building area during the construction of an integrated energy station; The amount of carbon emitted per unit building area is denoted as i; F i Let i be the carbon emission factor of the i-th type of carbon emission source, where i represents the type of carbon emission source, and there are a total of 4 types. The output variable dataset is obtained through calculation. .

8. The method for predicting carbon emissions from the construction of integrated energy stations based on machine learning according to claim 3, characterized in that, The method for steps 2-5 is as follows: The carbon emission prediction indicators for each type of integrated energy station construction are processed separately and used as input variables for the integrated energy station construction carbon emission prediction model, forming an input variable set. ,in, , represents the dataset of the j-th predictive index after normalization, which consists of the normalized values ​​of the j-th predictive index of m items; (3) In the formula: This represents the value of the i-th item after normalization for the j-th prediction index. These represent the maximum and minimum values ​​of the j-th predictive indicator within the same category, respectively; This represents the original value of the i-th item in the j-th prediction indicator; j represents the category of predictive indicators, and i represents different items. The number of representative projects.

Citation Information

Patent Citations

  • Comprehensive energy system operation state monitoring method based on multi-task learning

    CN115146696A