PCA-SVR-based carbon emission prediction system and method

Abstract

The invention provides a PCA-SVR-based carbon emission prediction system and method. The PCA-SVR-based carbon emission prediction system comprises an acquisition module, an analysis module and a prediction module, the acquisition module is used for acquiring carbon emission influence factors and constructing a data set; the analysis module is used for performing principal component analysis based on the data set to obtain generalized carbon emission influence factors; and the prediction module is used for adopting a preset PCA-SVR model to complete carbon emission prediction of the urban office building based on the generalized carbon emission influence factors. According to the method, 11 kinds of carbon emission influence factors are collected, abnormal values are eliminated through principal component analysis, a principal component score matrix is determined, generalized carbon emission influence factors are obtained to serve as independent variables of the SVR prediction model, the function f is introduced to improve the SVR prediction model, and the office building carbon emission can be accurately predicted.

Description

Technical Field

[0001] The present invention belongs to the technical field of carbon emission prediction, and in particular relates to a carbon emission prediction system and method based on PCA-SVR. Background Art

[0002] With the development of industrialization and urbanization, carbon emissions from public buildings rank first among buildings, and office buildings are a major contributor to carbon emissions from public buildings. Therefore, studying carbon emissions from office buildings is of great significance for achieving the dual carbon goals. At present, research on carbon emissions from office buildings mainly focuses on the study of carbon emission influencing factors and the prediction of carbon emissions, or the establishment of various carbon emission prediction models based on the factors affecting building carbon emissions. However, the existing technology only selects about four factors affecting carbon emissions, and the number of factors selected is small, which cannot fully analyze the impact of various carbon emission factors on office building carbon emissions. Studying the types and values ​​of factors affecting office building carbon emissions, the types and values ​​of carbon emission factors, and constructing a suitable carbon emission prediction model for office buildings can more accurately predict office building carbon emissions.

[0003] Among the current carbon emission prediction models such as SVR and neural networks, support vector machine prediction can effectively handle high-dimensional data, nonlinear separable problems, and has strong generalization ability, but it is slow to train for large-scale data, is sensitive to parameter selection and kernel function design, and is not suitable for multi-category problems; neural networks have nonlinear mapping capabilities, self-learning and self-adaptation capabilities, generalization capabilities and fault tolerance capabilities, but have slow convergence speed and inconsistent structures; SVR prediction can handle high-dimensional data and nonlinear problems, has strong robustness to outliers, but is very sensitive to parameter selection.

[0004] Therefore, how to select carbon emission influencing factors to construct a data set and choose a suitable office building carbon emission prediction model to more accurately predict office building carbon emissions is an urgent problem to be solved. Summary of the Invention

[0005] In order to solve the problems existing in the prior art, the present invention provides a carbon emission prediction system and method based on PCA-SVR, which aims to more accurately predict the carbon emissions of office buildings.

[0006] To achieve the above object, the present invention provides the following solution: a carbon emission prediction system based on PCA-SVR, comprising: an acquisition module, an analysis module and a prediction module;

[0007] The acquisition module is used to collect carbon emission influencing factors and construct a data set;

[0008] The analysis module is used to perform principal component analysis based on the data set to obtain generalized carbon emission influencing factors;

[0009] The prediction module is configured to predict the carbon emission of the urban office building based on the generalized carbon emission influencing factors by using a preset PCA-SVR model.

[0010] Preferably, the carbon emission influencing factors include land area, building area, number of energy users, number of staff, total electricity consumption, total water consumption, total natural gas consumption, window-wall ratio, shape coefficient, external wall heat transfer coefficient and external window heat transfer coefficient.

[0011] Preferably, the analysis module is configured to perform principal component analysis based on the data set to obtain the generalized carbon emission influencing factors, including:

[0012] F ij = ZX ij × A ij , i = 1, 2,..., 11, j = 1, 2, 3, 4

[0013] wherein F ij is a principal component score matrix of the jth principal component of the ith carbon emission influencing factor, ZX ij is a standardized matrix of sample data, and A ij is a principal component score coefficient matrix of the jth principal component of the ith carbon emission influencing factor.

[0014] Preferably, the preset PCA-SVR model comprises:

[0015]

[0016] wherein n is the office building number, N is the total number of office buildings, K is the kernel function, represents the generalized carbon emission influencing factor, and ε is the insensitive loss function threshold value, p n is an output value, and s.t. represents "restricted to"; α n and are both Lagrange multipliers, corresponding to the upper and lower boundary errors of the sample points, respectively; and C is a regularization parameter.

[0017] The application further provides a carbon emission prediction method based on PCA-SVR, which is implemented by using the foregoing system, and the method comprises the following steps:

[0018] S1, collecting carbon emission influencing factors to construct a data set;

[0019] S2, performing principal component analysis based on the data set to obtain generalized carbon emission influencing factors;

[0020] S3, predicting the carbon emission of the urban office building based on the generalized carbon emission influencing factors by using a preset PCA-SVR model.

[0021] Preferably, the carbon emission influencing factors include: land area, building area, number of energy users, number of staff, total electricity consumption, total water consumption, total natural gas consumption, window-wall ratio, shape coefficient, external wall heat transfer coefficient and external window heat transfer coefficient.

[0022] Preferably, the S2 performs principal component analysis based on the data set to obtain generalized carbon emission influencing factors, including:

[0023] F ij =ZX ij ×A ij , i = 1, 2,..., 11, j = 1, 2, 3, 4

[0024] Wherein, F ij is the principal component score matrix of the jth principal component of the ith carbon emission influencing factor, ZX ij is the standardized matrix of sample data, A ij is the principal component score coefficient matrix of the jth principal component of the ith carbon emission influencing factor.

[0025] Preferably, the preset PCA-SVR model comprises:

[0026]

[0027] Wherein, n is the office building number, N is the total number of office buildings, K is the kernel function, represents the generalized carbon emission influencing factor, and ε is the insensitive loss function threshold value, p n is the output value, and s.t. represents "restricted to"; α n and are both Lagrange multipliers, corresponding to the upper and lower boundary errors of the sample points; and C is a regularization parameter.

[0028] Compared with the prior art, the present application has the following beneficial effects:

[0029] The present application can more accurately predict the carbon emission of office buildings by collecting 11 carbon emission influencing factors, using principal component analysis to eliminate outliers, determining the principal component score matrix, obtaining the generalized carbon emission influencing factors as the independent variables of the SVR prediction model, and introducing the function f to improve the SVR prediction model. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the present application, the following briefly introduces the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0031] Figure 1 Figure 1 is a schematic diagram of a PCA-SVR-based carbon emission prediction system according to an embodiment of the present application;

[0032] Figure 2 Figure 2 is a scatter plot of 11 carbon emission influencing factors according to an embodiment of the present application;

[0033] Figure 3 Figure 3 is a component diagram of 11 carbon emission influencing factors according to an embodiment of the present application;

[0034] Figure 4 Figure 4 is a schematic diagram of a PCA-SVR model prediction result of carbon emission in T1 according to an embodiment of the present application;

[0035] Figure 5 Figure 5 is a schematic diagram of a PCA-SVR model verification result of carbon emission in T1 according to an embodiment of the present application;

[0036] Figure 6 Figure 6 is a schematic diagram of a PCA-SVR model prediction result of annual carbon emission per unit building in T2 according to an embodiment of the present application;

[0037] Figure 7 Figure 7 is a schematic diagram of a PCA-SVR model verification result of annual carbon emission per unit building in T2 according to an embodiment of the present application;

[0038] Figure 8 Figure 8 is a schematic diagram of a PCA-SVR model prediction result of annual carbon emission per energy user in T3 according to an embodiment of the present application;

[0039] Figure 9 Figure 9 is a schematic diagram of a PCA-SVR model verification result of annual carbon emission per energy user in T3 according to an embodiment of the present application;

[0040] Figure 10 Figure 10 is a schematic diagram of a PCA-SVR model prediction result of annual carbon emission per staff in T4 according to an embodiment of the present application;

[0041] Figure 11 Figure 11 is a schematic diagram of a PCA-SVR model verification result of annual carbon emission per staff in T4 according to an embodiment of the present application. DETAILED DESCRIPTION

[0042] The building area factors (land area, building area) and the office staff factors (energy user number, staff number) are derived from the Construction Standard for Office Buildings of Party and Government Agencies (JB169-2014).

[0043] The total energy consumption factor (total electricity consumption, total water consumption, total natural gas consumption) is derived from the energy consumption of more than 500 office buildings in Xiangtan City (established by the State Administration of State Affairs, approved by the State Statistics Bureau, and approved by the State Statistics Bureau with the approval number of Guo Tongzhi

[2015] 63).

[0044] The office building thermal performance coefficient factor (window-to-wall ratio, shape coefficient, external wall heat transfer coefficient, external window heat transfer coefficient) is derived from the current national standard "Building Lighting Design Standard" GB50033; the external wall heat transfer coefficient and the external window heat transfer coefficient are derived from "Public Building Energy Saving Design Standard"; the shape coefficient is derived from the research of Ren Binbin et al. (Ren Binbin, Wang Yiping, Xiao Shaoying, etc. Tianjin Low Energy Consumption Office Building Shape Research [J]. Building Energy Saving, 2015, 43(04): 66-68).

[0045] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0046] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail in combination with the drawings and specific embodiments.

[0047] Embodiment one

[0048] As shown in the Figure 1 , the present application provides a carbon emission prediction system based on PCA-SVR, comprising: a collection module, an analysis module and a prediction module.

[0049] The collection module is used for collecting carbon emission influencing factors and constructing a data set;

[0050] The analysis module is used for principal component analysis based on the data set to obtain generalized carbon emission influencing factors;

[0051] The prediction module is used for predicting the carbon emission of urban office buildings based on the generalized carbon emission influencing factors by using the pre-set PCA-SVR model.

[0052] Further, taking Xiangtan City as an example, a data set of carbon emission influencing factors of 514 office buildings in public institutions in each district of Xiangtan City is collected, including: building area factor A: land area A1 and building area A2, office number factor B: energy-consuming number B1 and staff number B2, total energy consumption factor C: total electricity consumption C1, total water consumption C2 and total natural gas consumption C3, office building thermal performance coefficient factor D: window-wall ratio D1, shape coefficient D2, external wall heat transfer coefficient D3 and external window heat transfer coefficient D4.

[0053] The data set is standardized to eliminate the units and dimensions of different variables in the data set, so that they have the same order of magnitude, and thus the result of the covariance matrix is more accurate. In this embodiment, Z-score standardization is adopted:

[0054]

[0055] wherein x i is the sample data of the i-th carbon emission influencing factor in the data set, mean(x) is the mean of the data, sd(x) is the standard deviation of the data, ZX i is the standardized value of the sample data of the i-th carbon emission influencing factor.

[0056] The sample data in the data set is standardized to obtain a 514 by 11 matrix Zx ij At this time, the vertical coordinates are ZA1, ZA2, ZB1, ZB2, ZC1, ZC2, ZC3, ZD1, ZD2, ZD3 and ZD4, respectively, representing the standardized values of the above-mentioned 11 carbon emission influencing factors, as shown in Table 1:

[0057] Table 1

[0058]

[0059] KMO test and Bartlett sphericity test are performed on the 514 groups of standardized sample data to determine whether the sample data in the data set is suitable for factor analysis, and the test results are shown in Table 2:

[0060] Table 2

[0061]

[0062] As shown in Table 2, the KMO value is 0.801, which is greater than 0.6 and close to 0.9, and the significance level is less than 0.01, indicating that the correlation between the variables of the carbon emission influencing factors in the data set is strong, and the factor analysis is suitable.

[0063] There is a very important concept called eigenvalue or eigenroot in principal component analysis. According to the eigenroot, the variance contribution rate (variance percentage) of each principal component can be calculated. The eigenroot and variance percentage are the basis for determining the number of principal components. The components with eigenroot greater than 1 are extracted as principal components. When the cumulative contribution rate of these principal components is high enough, reaching 70%, it is considered that the original data information can be well represented. The sample data of 514 office buildings in Xiangtan City are subjected to principal component analysis. According to the data analysis of total variance explanation, the results show that the 514 original data samples are not suitable for principal component analysis. Therefore, the sample data is considered to be reduced. After repeated operation of principal component analysis by SPSS software and data analysis of total variance explanation table, the cumulative contribution rate exceeds 70% after a series of data reduction until the data sample is screened to 57 groups. The total variance explanation table is generated as shown in Table 3:

[0064] Table 3

[0065]

[0066] As shown in Table 3, the number of principal components extracted from 11 carbon emission influencing factors is 4, and the cumulative contribution rate reaches 73.44%, indicating that the sample data is suitable for prediction.

[0067] The component matrix is composed of data corresponding to each original factor and principal component. These data are the coefficients of the factor expression of each original variable, representing the influence degree of the extracted principal component on the original variable. Each data in the component matrix represents the linear relationship between the factor (original variable) and the principal component. According to the component matrix, the mathematical expression of each principal component can be written, which is a multiple linear expression. The component matrix is used to explain the correlation between each component and the principal component, and is the most important part of principal component analysis. The specific data is shown in Table 4:

[0068] Table 4

[0069]

[0070] The correlation of each component with the original index can be more conveniently seen from the scatter plot and component plot corresponding to the component matrix. The scatter plot: since each point is connected like a mountain, it is also called a mountain landslide plot. The higher the position of the point in the plot, the greater the potential energy, and the more important it is in the study of the whole. The scatter plot is drawn according to the total variance explanation (contribution percentage) of each principal component. Its function is to select the number of principal components according to the size of the eigenvalue and the slope of the decline. The larger the eigenvalue, the steeper the slope. In the SPSS software, the eigenvalue greater than 1 is extracted as the principal component by default. The slope of the first principal component is often the steepest. If the cumulative contribution rate of the extracted principal components is low, the number of extracted principal components can be artificially changed, so it is also necessary to select in combination with the variance explanation table. The number of components of the component plot is the number of extracted principal components. The point represented by each index can be projected in the direction of the principal component. The value after projection represents the degree of influence. The larger the value, the greater the influence on the principal component. A positive value indicates a positive correlation, and a negative value indicates a negative correlation.

[0071] As shown in Figure 2 , 3 , the plot shows that the slope of index 1 to 2 is the largest, and there are 4 points with eigenvalues greater than 1, so the principal component analysis extracts 4 principal components. When the number of principal components reaches 4, the component plot cannot directly and intuitively show the projection of each point in the direction of each principal component. At this time, the principal component extraction based on the total variance explanation is the main one.

[0072] The principal component analysis of the 11 carbon emission influencing factors extracts 4 principal components. Principal component 1 (F1) mainly reflects 7 factors: land area, building area, energy-using population, number of staff, electricity consumption, gas consumption, and water consumption. Principal component 2 (F2) mainly reflects the window-wall ratio factor. Principal component 3 (F3) mainly represents the external wall heat transfer coefficient and external window heat transfer coefficient factors. Principal component 4 (F4) mainly reflects the shape coefficient factor. That is, the 11 carbon emission influencing factors are reduced to 4 generalized carbon emission influencing factors: generalized carbon emission influencing factor 1 represents land area, building area, energy-using population, number of staff, electricity consumption, water consumption, and natural gas consumption factors; generalized carbon emission influencing factor 2 represents the window-wall ratio factor; generalized carbon emission influencing factor 3 represents the external wall heat transfer coefficient and external window heat transfer coefficient factors; and generalized carbon emission influencing factor 4 represents the shape coefficient factor.

[0073] 57 office buildings' principal component score coefficient matrix A of 11 carbon emission influencing factors ij As shown in Table 5:

[0074] Table 5

[0075]

[0076] The component score matrix represents the relationship between each indicator variable and the extracted principal component. A high score on a principal component indicates that the linear correlation between the indicator and the principal component is stronger, and the more representative the principal component is. The calculation formula for each principal component is:

[0077] F ij =ZX ij ×A ij ,i=1,2,...,11,j=1,2,3,4

[0078] Among them, F ij is the principal component score matrix of the jth principal component of the i-th carbon emission influencing factor, ZX ij is the standardized matrix of sample data, A ij is the principal component score coefficient matrix of the jth principal component of the i-th carbon emission influencing factor.

[0079] The calculation formula of the principal component score data is as follows:

[0080]

[0081] Among them, f ij is the score of the jth principal component of the i-th carbon emission influencing factor, Zx ij is the standardized value of the jth principal component of the i-th carbon emission influencing factor, a ij is the coefficient of the jth principal component score of the i-th carbon emission influencing factor.

[0082] The specific calculation formula for the four principal component scores is:

[0083] F1=0.196ZA1+0.225ZA2+0.205ZB1+0.149ZB2+0.163ZC1+0.199ZC2+0.128ZC3+0.025ZD1-0.011ZD2-0.002ZD3+0.001ZD4

[0084] F2=0.264ZA1+0.106ZA2+0.224ZB1+0.125ZB2-0.339ZC1-0.266ZC2-0.332ZC3+0.293ZD1-0.278ZD2+0.129ZD3+0.113ZD4

[0085] F3=-0.037ZA1-0.052ZA2-0.036ZB1+0.303ZB2-0.019ZC1+0.002ZC2-0.137ZC3+0.078ZD1+0.229ZD2-0.640ZD3+0.509ZD4

[0086] F4 = 0.165ZA1 + 0.042ZA2 + 0.169ZB1 - 0.052ZB2 - 0.322ZC1 - 0.136ZC2 + 0.199ZC3 - 0.365ZD1 + 0.519ZD2 + 0.372ZD3 + 0.391ZD4

[0087] Wherein, F1, F2, F3, F4 are the scores of the first, second, third, fourth principal component, respectively, and ZA1, ZA2, ZB1,..., ZD4 are the standardized values of A1, A2, B1,..., D4, respectively.

[0088] The standardized matrix (57x11) of the sample data of 11 carbon emission influencing factors is multiplied by the principal component score coefficient matrix (11x4) to obtain the principal component score matrix (57x4). The principal component score matrix can be calculated by the formula, and the principal component score matrix is the standardized data matrix of the generalized carbon emission influencing factors, denoted as F ij The data in the principal component score matrix is the score of the four principal components after dimensionality reduction of the 11 carbon emission influencing factors, as shown in Table 6:

[0089] Table 6

[0090]

[0091] The sample data of the PCA-SVR carbon emission prediction model is composed of multiple independent variable vectors and a single dependent variable vector. The generalized carbon emission factors 1, 2, 3, and 4 are used as the independent variables for carbon emission SVR prediction, and the independent variable matrix is F ij , as shown in Table 6. The dependent variable is the data of the relevant annual carbon emission of office buildings, denoted as vector P i . The generalized carbon emission influencing factors obtained after PCA analysis are marked as f, and the pre-set PCA-SVR carbon emission prediction model improves the original SVR prediction model, and the improvement steps are as follows:

[0092] Let The result of PCA can establish a function formula:

[0093]

[0094] Wherein, represents the generalized carbon emission influencing factor.

[0095] The standardized data of the generalized carbon emission influencing factors 1, 2, 3, and 4 are used as the independent variables, and the standardized data of the annual carbon emission of office buildings is used as the dependent variable to form the training sets T1, T2, T3, and T4:

[0096]

[0097] Wherein, pn Y is the output value; n is the office building number, N is the total number of office buildings, N = 57.

[0098] Let the data set R n The linear function expression on the data set R

[0099]

[0100] where ω is the weight vector; b is the offset constant. According to the theory of mathematical statistics, function estimation can be converted to:

[0101] min = || ω || 2

[0102]

[0103] where ε is the insensitive loss function threshold.

[0104] To solve the problem that individual data cannot complete the estimation under the accuracy of ε, a relaxation variable is introduced to convert the optimization process into a dual problem. Both the dual problem and the regression function only involve the inner product operation between sample inputs Therefore, the kernel function is introduced The linear regression problem is converted into a nonlinear regression problem in Hilbert space, and the model is constructed as:

[0105]

[0106]

[0107] where s.t. means "subject to", that is, the following formula is the constraint condition of the problem; α n and are Lagrange multipliers, used to handle constraints in optimization problems, usually in pairs, corresponding to the upper and lower error bounds of the sample points; C is the regularization parameter, used to balance the complexity of the model and the training error. The larger the value, the more the model tends to reduce the error, and the smaller the value, the more errors are allowed to enhance the generalization; N represents the number of samples, the total number of data points, that is, the total number of office buildings.

[0108] In the constraint condition, indicates that the sum of the difference of the Lagrange multipliers must be 0, ensuring the dual condition of the optimization problem; α n ≥ 0 and respectively indicate that α n is non-negative, the upper bound of indicates that the regularization parameter C is evenly distributed to each sample, ensuring that the total constraint of all samples does not exceed C, for example

[0109] Optimal solution The regression function is:

[0110]

[0111] According to the linear training sets T1, T2, T3 and T4, the independent variables and dependent variables in the model are normalized to ensure that all data are between [0, 1], and the normalization formula is as follows:

[0112]

[0113] The training sets T1, T2, T3 and T4 are standardized respectively to obtain standardized data matrices T1 55×5 , T2 54×5 , T3 55×5 and T4 54×5 , which are used as sample data matrices for predicting annual carbon emissions, carbon emissions per unit area, carbon emissions per capita and carbon emissions per capita for compilation, respectively, denoted as T'1, T'2, T'3 and T'4. According to the split validation method, they are divided into prediction group and validation group, and 70% of the sample data in each group is selected as the training set (prediction set) of the prediction group, and the remaining is used as the test set (validation set) of the validation group. The prediction set is subjected to related carbon emission prediction by the PCA-SVR model to obtain a fitting curve of actual value and predicted value, and the quality of the prediction result is explained by the RMSE value, and the validation set is used to verify whether the prediction result of the training set is accurate. The four groups of prediction set and validation set are subjected to PCA-SVR prediction to obtain Figures 4 to 11 .

[0114] Among them, Figure 4 , Figure 6 , Figure 8 and Figure 10 are the PCA-SVR model prediction results of T'1, T'2, T'3 and T'4, Figure 5 , Figure 7 , Figure 9 and Figure 11 are the PCA-SVR model validation results of data matrices T'1, T'2, T'3 and T'4. Predictionsamples is the sample data, Prediction results is the prediction result of the prediction set, Comparison of training set prediction results is the comparison of the prediction results of the validation set, and RMSE (Root Mean Squared Error) value is the root mean square error of the actual value and the predicted value.

[0115] From Figure 4It can be seen that the RMSE value of the PCA-SVR prediction results of T′1 is 0.14, which is close to 0.1, indicating that the PCA-SVR model has a good prediction of the annual carbon emissions of 39 office buildings in Xiangtan City. Figure 4 The actual value of data point 15 is greater than the predicted value, indicating that the annual carbon emissions of the office building represented by the 15th data point in the prediction set have great carbon reduction potential.

[0116] Depend on Figure 5 It is known that the RMSE value of the PCA-SVR verification result of T′1 is 0.65, which represents the PCA-SVR model prediction result of the annual carbon emissions of office buildings in Xiangtan City ( Figure 4 ) has a good verification result. The actual value of data point 8 deviates significantly from the predicted value, far exceeding the predicted value, indicating that the annual carbon emissions of the office building represented by data point 8 have great carbon reduction potential.

[0117] Depend on Figure 6 The PCA-SVR prediction results for T′2 show an RMSE value of 0.78, indicating a good PCA-SVR prediction of annual carbon emissions per unit building for 38 office buildings in Xiangtan City. The actual value of data point 20 deviates significantly from the predicted value, exceeding the predicted value, indicating significant potential for carbon reduction in annual carbon emissions per unit building for the 20th data point in the prediction set. The values ​​of the generalized carbon emission influencing factors for the 20th data point are -0.48, -0.33, 0.00, and 0.71, respectively. Generalized carbon emission influencing factor 4 has the largest absolute value. The physical meaning of generalized carbon emission influencing factor 4 represents the shape coefficient, indicating that the shape coefficient of this office building is the primary factor affecting its high annual carbon emissions per unit building.

[0118] Figure 7 The RMSE value of the PCA-SVR validation results for T′2 is 0.60, indicating a good validation result for the PCA-SVR prediction model for annual carbon emissions per unit building. The actual value of data point 4 in the figure deviates significantly from the predicted value and is smaller than the predicted value. Analysis of the office building corresponding to data point 4 reveals that this office building has a large floor area, a small number of energy users, and low energy consumption, resulting in lower actual annual carbon emissions per unit building. The actual value of data point 15 in the figure deviates significantly from the predicted value and is larger than the predicted value. According to the PCA analysis results for this data point, the absolute value of generalized carbon emission influencing factor 4 is the largest, and the shape coefficient should be minimized. Analysis of the actual office building data for this data point also reveals that this office building has a large number of energy users and staff, resulting in higher annual carbon emissions, and therefore, higher actual annual carbon emissions per unit building.

[0119] Depend on Figure 8The RMSE value of the PCA-SVR prediction results for T′3 is 0.59, less than 1, indicating a good validation of the PCA-SVR prediction model for annual carbon emissions per energy user. The actual values ​​for data points 14 and 31 deviate significantly from the predicted values, and are larger than the predicted values. Analysis of the office buildings corresponding to data points 14 and 31 reveals that these buildings have high energy consumption, but few energy users are counted, resulting in a high actual annual carbon emissions per energy user. This may be due to inaccurate counting of energy users during the statistical process.

[0120] Figure 9 The RMSE value of the PCA-SVR validation results for T′3 is 0.7, less than 1, indicating a good validation result for the PCA-SVR prediction model for annual carbon emissions per energy user. However, there is a certain degree of deviation between the actual values ​​and the predicted values ​​for many data points, indicating that there is a certain degree of error in the statistics of the actual number of energy users. This may be due to the reality of the varying number of non-local office workers and varying office hours in office buildings. If annual carbon emissions per energy user are used as a carbon emission metric for office buildings in Xiangtan, the statistical method and accuracy of the number of energy users need to be carefully considered.

[0121] Depend on Figure 10 The PCA-SVR prediction results for T′4 show an RMSE of 0.38, indicating a good PCA-SVR prediction of annual per capita carbon emissions. The actual values ​​for data points 3 and 14 deviate significantly from the predicted values, with the actual values ​​exceeding the predicted values. Considering the nature of the staffing staff's work, the actual per capita carbon emissions for office buildings may be higher due to the staff's limited travel and the high number of external staff.

[0122] Depend on Figure 11 The PCA-SVR validation results for T′4 show an RMSE value of 0.42, indicating a good validation result for the PCA-SVR prediction model for annual carbon emissions per capita based on energy consumption. Some data points in the figure deviate somewhat from the actual values ​​and predicted values. This may be due to the fact that the actual annual carbon emissions per capita are constrained by the absence of the staff members in the office buildings, resulting in a poor fit between the actual and predicted values. However, the deviation is small and within an acceptable range. If annual carbon emissions per capita are used as the carbon emission standard for office buildings in Xiangtan, it is necessary to analyze the energy consumption of each staff member to obtain a more accurate annual carbon emissions per capita.

[0123] Because the SVR prediction is generally applied in low-dimensional prediction, the increase of carbon emission influencing factors will cause difficulty in obtaining the carbon emission prediction curve. As a dimension reduction method, the PCA can retain about 80% or even higher information of original carbon emission influencing factors, reduce the dozens of carbon emission influencing factors to 4 or 5, and then perform the SVR prediction, and the prediction result is relatively accurate. From the perspective of statistics, the retained information is about 70% and is sufficient to obtain the correct prediction result, but the specific situation is analyzed, in the application, the 4 carbon emission influencing factors after the dimension reduction, i.e. the generalized carbon emission influencing factors, are also explained in terms of physical meaning, and the dozens of original carbon emission influencing factors are also relatively relevant general carbon emission influencing factors, so the dimension reduction is relatively successful and can be used for the SVR prediction.

[0124] In conclusion, the application can collect 11 carbon emission influencing factors, remove the abnormal values by using the principal component analysis, determine the principal component score matrix, obtain the generalized carbon emission influencing factors as independent variables of the SVR prediction model, and introduce the function f to improve the SVR prediction model, so that the carbon emission of the office building can be accurately predicted.

[0125] Embodiment two

[0126] The application further provides a carbon emission prediction method based on the PCA-SVR, and the system in the embodiment one is applied, and the method comprises the following steps:

[0127] S1, collecting carbon emission influencing factors and constructing a data set;

[0128] S2, performing principal component analysis based on the data set to obtain generalized carbon emission influencing factors;

[0129] S3, based on the generalized carbon emission influencing factors, using the preset PCA-SVR model to complete the carbon emission prediction of the urban office building.

[0130] The carbon emission influencing factors comprise land area, building area, energy-using population, compiled population, total electric consumption, total water consumption, total natural gas consumption, window-wall ratio, shape coefficient, external wall heat transfer coefficient and external window heat transfer coefficient.

[0131] S2, performing principal component analysis based on the data set to obtain generalized carbon emission influencing factors, comprising:

[0132] F ij =ZX ij ×A ij ,i=1,2,...,11,j=1,2,3,4

[0133] Wherein, F ij is the principal component score matrix of the jth principal component of the ith carbon emission influencing factor, and ZXij is the standardized matrix of sample data, A ij is the principal component score coefficient matrix of the jth principal component of the i-th carbon emission influencing factor;

[0134] S3 uses a pre-built PCA-SVR model to calculate the carbon emissions of urban office buildings based on broad carbon emission influencing factors. The pre-built PCA-SVR model includes:

[0135]

[0136] Among them, n is the office building number, N is the total number of office buildings, K is the kernel function, represents the generalized carbon emission impact factor, ε is the insensitive loss function threshold, and p n is the output value, st means "limited by"; α n and are Lagrange multipliers, corresponding to the upper and lower bound errors of the sample points respectively; C is the regularization parameter.

[0137] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A carbon emission prediction system based on PCA-SVR, characterized in that: The system includes: acquisition module, analysis module and prediction module; The acquisition module is used to collect carbon emission influencing factors and construct a data set; The analysis module is used to perform principal component analysis based on the data set to obtain generalized carbon emission influencing factors; The prediction module is used to complete the carbon emission prediction of urban office buildings based on the generalized carbon emission influencing factors using a preset PCA-SVR model.

2. The carbon emission prediction system based on PCA-SVR according to claim 1 is characterized in that: The carbon emission influencing factors include: land area, building area, number of energy users, number of staff, total electricity consumption, total water consumption, total natural gas consumption, window-to-wall ratio, shape coefficient, exterior wall heat transfer coefficient and exterior window heat transfer coefficient.

3. The carbon emission prediction system based on PCA-SVR according to claim 1 is characterized in that: The analysis module is used to perform principal component analysis based on the data set to obtain broad carbon emission influencing factors, including: F ij =ZX ij ×A ij ,i=1,2,...,11,j=1,2,3,4 Among them, F ij is the principal component score matrix of the jth principal component of the i-th carbon emission influencing factor, ZX ij is the standardized matrix of sample data, A ij is the principal component score coefficient matrix of the jth principal component of the i-th carbon emission influencing factor.

4. The carbon emission prediction system based on PCA-SVR according to claim 3 is characterized in that: The preset PCA-SVR model includes: Among them, n is the office building number, N is the total number of office buildings, K is the kernel function, represents the generalized carbon emission impact factor, ε is the insensitive loss function threshold, and p n is the output value, st means "limited by"; α n and are Lagrange multipliers, corresponding to the upper and lower bound errors of the sample points respectively; C is the regularization parameter.

5. A carbon emission prediction method based on PCA-SVR, implemented by using the carbon emission prediction system based on PCA-SVR according to any one of claims 1 to 4, characterized in that: The method comprises: S1. Collect factors affecting carbon emissions and build a data set; S2. Performing principal component analysis based on the data set to obtain broad carbon emission influencing factors; S3. Based on the generalized carbon emission influencing factors, a preset PCA-SVR model is used to complete the carbon emission prediction of urban office buildings.

6. The carbon emission prediction method based on PCA-SVR according to claim 5 is characterized in that: The carbon emission influencing factors include: land area, building area, number of energy users, number of staff, total electricity consumption, total water consumption, total natural gas consumption, window-to-wall ratio, shape coefficient, exterior wall heat transfer coefficient and exterior window heat transfer coefficient.

7. The carbon emission prediction method based on PCA-SVR according to claim 5 is characterized in that: The S2 performs principal component analysis based on the data set to obtain broad carbon emission influencing factors, including: F ij =ZX ij ×A ij ,i=1,2,...,11,j=1,2,3,4 Among them, F ij is the principal component score matrix of the jth principal component of the i-th carbon emission influencing factor, ZX ij is the standardized matrix of sample data, A ij is the principal component score coefficient matrix of the jth principal component of the i-th carbon emission influencing factor.

8. The carbon emission prediction method based on PCA-SVR according to claim 7 is characterized in that: The preset PCA-SVR model includes: Among them, n is the office building number, N is the total number of office buildings, K is the kernel function, represents the generalized carbon emission impact factor, ε is the insensitive loss function threshold, and p n is the output value, st means "limited by"; α n and are Lagrange multipliers, corresponding to the upper and lower bound errors of the sample points respectively; C is the regularization parameter.

Images