Carbon emission prediction method and device with multiple emission sources

By combining principal component analysis, LSTM, and ARDL models, an energy consumption prediction model was constructed, which solved the problems of single model and low accuracy in carbon emission prediction of industrial parks, and achieved accurate real-time prediction of multiple emission sources, supporting carbon emission management.

CN118917482APending Publication Date: 2024-11-08STATE GRID JIANGSU ELECTRIC POWER CO LTD TAIZHOU POWER SUPPLY BRANCH +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411027233.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-30
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In existing technologies, carbon emission prediction methods for industrial parks suffer from problems such as single models, low prediction accuracy, and inability to adapt to changing energy consumption in real time, especially with the lag in the statistical analysis of consumption from emission sources other than electricity.

Method used

An energy consumption prediction model is constructed by combining principal component analysis, long short-term memory network (LSTM) model and autoregressive distributed lag model (ARDL). By analyzing the historical consumption of electricity emission sources and other emission sources, the consumption of each emission source is predicted. The total carbon emissions are calculated by combining the carbon emission factor, and the model parameters are dynamically adjusted to improve the prediction accuracy.

Benefits of technology

It enables accurate and real-time prediction of carbon emissions from multiple emission sources in industrial parks, improving prediction accuracy and reliability, and supporting carbon emission management in target areas.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917482A_ABST
    Figure CN118917482A_ABST
Patent Text Reader

Abstract

The invention discloses a carbon emission prediction method and device with multiple emission sources. The method comprises the following steps: acquiring the consumption of an electric power emission source in a target area at a current time node and the consumption of all emission sources in a certain time period before the current time node; on the basis of a pre-constructed energy consumption prediction model, the consumption of the power emission source at the current time node and the consumption of all emission sources at a certain time period before the current time node are analyzed and processed, and the consumption of other emission sources except the power emission source at the current time node is predicted; and performing carbon emission analysis processing on the predicted consumption amounts of the electric power emission source at the current time node and other emission sources except the electric power emission source at the current time node to obtain the total carbon emission amount of all the emission sources at the current time node. The method can solve the problem that the consumption statistics of other emission sources except the electric power emission source lags behind, and finally achieves the accurate prediction of the carbon emission of the target region.
Need to check novelty before this filing date? Find Prior Art

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 method and device with multiple emission sources. Background Art

[0002] As the global climate change problem becomes increasingly serious, different regions have formulated and implemented a series of carbon emission control policies to reduce greenhouse gas emissions and alleviate environmental pressure. As one of the important sources of energy consumption and carbon emissions, the carbon emission prediction and management of industrial parks are of great significance to achieving carbon emission reduction goals. However, traditional carbon emission prediction methods often have problems such as single model, low prediction accuracy, and inability to adapt to changing energy consumption in real time. For example, patent application document CN117973587A discloses a prediction system for industrial park carbon emissions. The invention comprises a statistical processing unit, an environmental monitoring unit, an analysis and accounting unit, an offset correction unit and a prediction and output unit. The statistical processing unit is used to collect various energy consumption-related data such as energy consumed in the industrial production process of the park and energy consumption of transportation vehicles, and respectively establishes formulas to calculate the carbon output, so as to simply and accurately calculate the total carbon output prediction value of the park. The actual environment of the park is monitored by the environmental monitoring unit, so as to correct the carbon storage and carbon sink data of the park to adapt to the changes in conditions and ensure the accuracy of the calculation process. The analysis and accounting unit processes the total carbon output prediction value according to the specific carbon capacity of the park, so as to facilitate the offset correction unit to process the offsettable data according to the actual carbon sink of the park. The prediction and output unit collects the results to simply and quickly obtain the carbon emission prediction value of the park.

[0003] Therefore, how to provide a method that can accurately calculate and predict carbon emissions from multiple emission sources at different time points in a park is an urgent problem to be solved by those skilled in the art. Summary of the invention

[0004] In view of the defects in the above-mentioned prior art, the present invention provides a carbon emission prediction method and device with multiple emission sources, which can solve the problem of lagging statistics on consumption of emission sources other than power emission sources, and ultimately achieve accurate prediction of carbon emissions in the target area.

[0005] In a first aspect, the present invention provides a carbon emission prediction method with multiple emission sources, which specifically comprises the following steps:

[0006] Obtain the consumption of electricity emission sources in the target area at the current time node and all emission sources in a certain period of time before the current time node;

[0007] Based on the pre-built energy consumption forecasting model, the consumption of the power emission source at the current time node and all emission sources in a certain period of time before the current time node are analyzed and processed, and the consumption of other emission sources except the power emission source at the current time node is predicted;

[0008] Carbon emissions analysis is performed on the electricity emission sources at the current time node and the predicted consumption of other emission sources except the electricity emission sources at the current time node to obtain the total carbon emissions of all emission sources at the current time node.

[0009] Furthermore, other emission sources besides electricity emission sources include at least one of raw coal, crude oil, and natural gas, wherein the electricity emission sources include electricity purchased from the power grid and electricity generated by distributed photovoltaics in the target area.

[0010] Furthermore, the energy consumption forecasting model is pre-built, including:

[0011] Obtain the historical consumption of all emission sources in the target area at different time points;

[0012] Analyze and process the historical consumption of all emission sources at adjacent time nodes, and the historical consumption of power emission sources and other emission sources except power emission sources at continuous time nodes, and obtain the first prediction model and the second prediction model respectively;

[0013] The first prediction model and the second prediction model are combined to obtain an energy consumption prediction model.

[0014] Furthermore, the historical consumption of all emission sources at adjacent time nodes is analyzed and processed to obtain the first prediction model, including:

[0015] Based on the historical consumption of all emission sources at adjacent time nodes and combined with the weights of different emission sources relative to carbon emissions, the long short-term memory network model is iteratively trained to give the first prediction model, which satisfies the following relationship:

[0016] Y t =f LSTM (X t-1 )

[0017] X t-1 =[w M *M(t-1),w O *O(t-1),w Q *Q(t-1),w D *D(t-1)]

[0018] Where Y t is the first predicted consumption of each emission source in the target area at time node t, f LSTM (Xt ) is the target area in the input sample X t-1 The long short-term memory network model under X t-1 is the input sample, that is, the historical consumption of each emission source at the time node t-1, w M 、w O 、w Q 、w D are the weights of each emission source, M(t-1), O(t-1), Q(t-1), and D(t-1) are the historical consumption of each emission source at the time node t-1.

[0019] Furthermore, the weights of different emission sources relative to carbon emissions include:

[0020] Normalize the historical consumption of different emission sources at different time points in the target area;

[0021] The normalized historical consumption is standardized;

[0022] Calculate the covariance matrix and perform eigenvalue decomposition on the standardized historical consumption;

[0023] Based on the eigenvalue decomposition results, principal component analysis is performed, and the weight of each emission source relative to carbon emissions is given.

[0024] Furthermore, the historical consumption of the power emission sources and other emission sources other than the power emission sources at continuous time nodes is analyzed and processed to obtain a second prediction model, including:

[0025] The power emission source is paired with each other emission source to obtain multiple emission source groups; the autoregressive distribution time lag model is trained with the historical consumption of each emission source group at different time nodes, and the parameters of the autoregressive distribution time lag model are updated by the least squares method to obtain multiple second prediction models corresponding to each emission source group, satisfying the following relationship:

[0026]

[0027] In the formula, y' i,t is the second predicted consumption of the ith emission source at time node t, y i,t-p is the historical consumption of the ith emission source at time tp, x t-q is the historical consumption of the power emission source at the time node tq; δ0 is a constant term, P is the number of time nodes of the i-th emission source, Q is the number of time nodes of the power emission source, φ i,t-p is the fitting parameter of the ith emission source at the tp time node, β t-q is the fitting parameter of the power emission source at the time node tq, ε tis the noise term.

[0028] Furthermore, the first prediction model and the second prediction model are combined to obtain an energy consumption prediction model, including:

[0029] Determine the prediction errors of the first prediction model and the second prediction model according to the historical consumption of different emission sources in the target area at different time points;

[0030] Determining weights of the first prediction model and the second prediction model based on prediction errors of the first prediction model and the second prediction model;

[0031] According to the weights of the first prediction model and the second prediction model, the first prediction model and the second prediction model are combined to obtain an energy consumption prediction model that satisfies the following relationship:

[0032]

[0033] Where σ1 and σ2 are the prediction errors of the first prediction model and the second prediction model respectively, and Y t is the consumption of all emission sources at time node t predicted by the first prediction model, y t is the consumption of all emission sources at time node t predicted by the second prediction model, It is the consumption forecast result of all emission sources at time node t output by the energy consumption forecast model.

[0034] Furthermore, the prediction errors of the first prediction model and the second prediction model include:

[0035] The historical consumption of different emission sources in the target area at different time nodes is processed by the first prediction model and the second prediction model, and the first predicted consumption and the second predicted consumption at multiple time nodes are respectively given;

[0036] Based on the first predicted consumption and the second predicted consumption at multiple time nodes and the historical consumption of different emission sources in the target area at different time nodes, the prediction errors of the first prediction model and the second prediction model are given respectively.

[0037] Furthermore, the prediction error may be one of model accuracy, F1 score, confidence, and the difference between the true value and the predicted value.

[0038] Furthermore, the carbon emission analysis is performed on the power emission source at the current time node and the predicted consumption of other emission sources at the current time node except the power emission source to obtain the total carbon emission of all emission sources at the current time node, including:

[0039] Obtain the carbon emission factors of all emission sources at the current time node;

[0040] The predicted consumption of electricity emission sources at the current time node and other emission sources except electricity emission sources at the current time node are respectively used with the corresponding carbon emission factors to calculate carbon emissions, and the carbon emissions of all emission sources at the current time node are obtained;

[0041] The carbon emissions of all emission sources at the current time node are accumulated to obtain the total carbon emissions of all emission sources at the current time node, which satisfies the following relationship:

[0042] C(t)=M(t)*f m (t)+O(t)*f o (t)+Q(t)*f Q (t)+D(t)*f D (t)

[0043] Where C(t) is the total carbon emissions of the target area at time node t, M(t), O(t), Q(t), and D(t) are the consumption of each emission source at time node t, and f m (t), f o (t), f Q (t), f D (t) are the unit energy consumption carbon emission factors of each emission source at time node t. Specifically, M(t), O(t), Q(t), and D(t) are the consumption of raw coal, crude oil, electricity, and natural gas at time node t, respectively. m (t), f o (t), f Q (t), f D (t) are the carbon emission factors per unit energy consumption of raw coal, crude oil, electricity and natural gas at time node t.

[0044] Furthermore, after obtaining the total carbon emissions of all emission sources at the current time node, it also includes:

[0045] Obtain the actual consumption of other emission sources except electricity emission sources at the current time node;

[0046] Based on the actual consumption of all emission sources at the current time node and the predicted consumption of emission sources other than electricity emission sources at the current time node, the model parameters of the pre-built energy consumption prediction model are adjusted to obtain an optimized energy consumption prediction model.

[0047] Furthermore, based on the actual consumption of all emission sources at the current time node and the predicted consumption of other emission sources except the power emission source at the current time node, the model parameters of the pre-built energy consumption prediction model are adjusted to obtain an optimized energy consumption prediction model, including:

[0048] According to the actual consumption of all emission sources at the current time node and the predicted consumption of other emission sources except the power emission source at the current time node, the absolute value of the difference between the actual consumption and the predicted consumption of the first prediction model and the second prediction model in the energy consumption prediction model is obtained;

[0049] According to the weights of the first prediction model and the second prediction model, and the corresponding absolute values ​​of the differences, a weighted absolute value of the difference of the energy consumption prediction model is obtained;

[0050] According to the absolute value of the weighted difference of the energy consumption prediction model, the average value of the absolute value of the weighted difference is given;

[0051] With the minimum absolute value of the weighted difference as the goal, the model parameters of the energy consumption prediction model are adjusted through an optimization algorithm; wherein the optimization algorithm can be a gradient descent or a genetic algorithm.

[0052] In a second aspect, the present invention further provides a carbon emission prediction device with multiple emission sources, using the above-mentioned carbon emission prediction method with multiple emission sources, the carbon emission prediction device comprises:

[0053] A data acquisition module, which is used to obtain the consumption of electricity emission sources in the target area at the current time node and all emission sources in a certain period of time before the current time node;

[0054] The consumption prediction module is used to analyze and process the consumption of the power emission source at the current time node and all emission sources in a certain period of time before the current time node based on the pre-built energy consumption prediction model, and predict the consumption of other emission sources except the power emission source at the current time node;

[0055] The carbon emission prediction module is used to perform carbon emission analysis on the predicted consumption of the power emission source at the current time node and other emission sources except the power emission source at the current time node to obtain the total carbon emissions of all emission sources at the current time node.

[0056] The present invention provides a method and device for predicting carbon emissions with multiple emission sources, which at least have the following beneficial effects:

[0057] (1) The consumption of electricity emission sources can be used to predict the consumption of other emission sources (energy), thereby solving the problem of lagging statistics on the consumption of emission sources other than electricity emission sources, and ultimately achieving accurate prediction of carbon emissions in the target area.

[0058] (2) The present invention utilizes a variety of models to explore the correlation between energy consumption and electricity consumption, estimates the energy consumption, and uses the carbon emission factor method to estimate the corresponding carbon emissions based on the energy consumption, ultimately achieving real-time carbon emission assessment.

[0059] (3) After predicting the consumption through the energy consumption prediction model, the energy consumption prediction model can be updated and optimized when the actual consumption is obtained to ensure the accuracy of the energy consumption prediction at future time nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A flowchart of a carbon emission prediction method with multiple emission sources provided by the present invention;

[0061] Figure 2 A flowchart of building an energy consumption prediction model provided by an embodiment of the present invention;

[0062] Figure 3 A flow chart for obtaining total carbon emissions provided in an embodiment of the present invention;

[0063] Figure 4 A schematic diagram of a carbon emission prediction device with multiple emission sources provided by the present invention. DETAILED DESCRIPTION

[0064] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0065] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two.

[0066] It should also be noted that the term "includes", "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, so that a commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprising a ..." do not exclude the existence of other identical elements in the commodity or device including the elements.

[0067] In the prior art, various forms of energy consumption such as coal, oil, electricity, and gas are independently calculated for carbon emissions, and the comprehensive impact of each emission source on the overall carbon emissions is not fully considered, resulting in large deviations in the prediction results. Principal component analysis (PCA), as a data dimension reduction and feature extraction technology, can effectively extract the main features of each emission source and calculate its weight in carbon emissions. The long short-term memory network (LSTM) and the autoregressive distributed lag model (ARDL), as advanced time series analysis methods, can use historical data to predict carbon emissions, making up for the shortcomings of traditional methods in dealing with nonlinearity and time correlation. However, a single LSTM or ARDL model has its own limitations in application. The present invention can further improve the accuracy and reliability of the prediction by combining these two models and selecting the optimal model.

[0068] like Figure 1 As shown, the present invention provides a carbon emission prediction method with multiple emission sources, which specifically includes the following steps:

[0069] Obtain the consumption of electricity emission sources in the target area at the current time node and all emission sources in a certain period of time before the current time node;

[0070] Based on the pre-built energy consumption forecasting model, the consumption of the power emission source at the current time node and all emission sources in a certain period of time before the current time node are analyzed and processed, and the consumption of other emission sources except the power emission source at the current time node is predicted;

[0071] Carbon emissions analysis is performed on the electricity emission sources at the current time node and the predicted consumption of other emission sources except the electricity emission sources at the current time node to obtain the total carbon emissions of all emission sources at the current time node.

[0072] The target area referred to in the present invention can be a park. Other emission sources besides electricity emission sources include at least one of raw coal, crude oil, and natural gas. Among them, raw coal, crude oil, and natural gas are energy emission sources. Electricity emission sources include electricity purchased from the power grid and electricity generated by distributed photovoltaics in the target area.

[0073] After obtaining the consumption of the power emission sources in the target area at the current time node and all emission sources in a certain period of time before the current time node, the present invention also performs denoising and outlier removal processing on the obtained consumption, thereby reducing noise interference. The certain period of time before the current time node refers to a plurality of continuous historical time nodes immediately adjacent to the current time node.

[0074] When predicting the consumption of emission sources other than electricity (i.e. predicting energy consumption), the energy consumption prediction model is mainly used to process the consumption of emission sources in a certain period of time before the current time node, and finally realizes the consumption prediction of the current time node. Figure 2 As shown, it may include:

[0075] Obtain the historical consumption of all emission sources in the target area at different time points;

[0076] Analyze and process the historical consumption of all emission sources at adjacent time nodes, and the historical consumption of power emission sources and other emission sources except power emission sources at continuous time nodes, and obtain the first prediction model and the second prediction model respectively;

[0077] The first prediction model and the second prediction model are combined to obtain an energy consumption prediction model.

[0078] The analysis and processing of the historical consumption of all emission sources at adjacent time nodes to obtain the first prediction model may include:

[0079] Based on the historical consumption of all emission sources at adjacent time nodes and combined with the weights of different emission sources relative to carbon emissions, the long short-term memory network model is iteratively trained to give the first prediction model, which satisfies the following relationship:

[0080] Y t =f LSTM (X t-1 )

[0081] X t-1 =[w M *M(t-1),w O *O(t-1),w Q *Q(t-1),w D *D(t-1)]

[0082] Where Y t is the first predicted consumption of each emission source in the target area at time node t, f LSTM (X t ) is the target area in the input sample X t-1 The long short-term memory network model under X t-1 is the input sample, that is, the historical consumption of each emission source at the time node t-1, w M 、w O 、w Q 、w Dare the weights of each emission source, M(t-1), O(t-1), Q(t-1), and D(t-1) are the historical consumption of each emission source at the time node t-1. Specifically, w M 、w O 、w Q 、w D are the weights of raw coal, crude oil, electricity and natural gas respectively. M(t-1), O(t-1), Q(t-1) and D(t-1) are the historical consumption of raw coal, crude oil, electricity and natural gas at the time node t-1 respectively.

[0083] In actual application scenarios, the consumption data of all emission sources at the t-1 time node is obtained as input samples, and the corresponding energy consumption data is input into the hidden layer of the long short-term memory network model. Then, based on the input sample data and cell state information, the long short-term memory network model is trained, and finally the first prediction model of energy consumption in the target area at the t time node is output.

[0084] By using historical energy consumption data to iteratively train the long short-term memory network model multiple times, the prediction ability of the model can be improved. Finally, the prediction accuracy of the first prediction model can be verified by cross-validation, and the model parameters of the first prediction model can be adjusted to achieve the best effect.

[0085] When iteratively training the LSTM network model, the weights of different emission sources relative to carbon emissions may include:

[0086] Normalizing the historical consumption of different emission sources at different time points in the target area can eliminate the impact of different dimensions;

[0087] The normalized historical consumption is standardized to ensure the comparability of the data;

[0088] Calculate the covariance matrix and perform eigenvalue decomposition on the standardized historical consumption;

[0089] Based on the results of eigenvalue decomposition, principal component analysis is performed, and the weight of each emission source relative to carbon emissions is given to determine the contribution of different emission sources to the overall carbon emissions.

[0090] In addition, the historical consumption of the power emission source and other emission sources other than the power emission source at continuous time nodes is analyzed and processed to obtain a second prediction model, which may include:

[0091] The power emission source is paired with each other emission source to obtain multiple emission source groups; the autoregressive distribution time lag model is trained with the historical consumption of each emission source group at different time nodes, and the parameters of the autoregressive distribution time lag model are updated by the least squares method to obtain multiple second prediction models corresponding to each emission source group, satisfying the following relationship:

[0092]

[0093] In the formula, y' i,t is the second predicted consumption of the ith emission source at time node t, y i,t-p is the historical consumption of the ith emission source at time tp, x t-q is the historical consumption of the power emission source at the time node tq; δ0 is a constant term, P is the number of time nodes of the i-th emission source, Q is the number of time nodes of the power emission source, φ i,t-p is the fitting parameter of the ith emission source at the tp time node, β t-q is the fitting parameter of the power emission source at the time node tq, ε t is a noise term, which represents the error between the predicted result and the actual result at the current time node t.

[0094] After obtaining the first prediction model and the second prediction model, the present invention can combine them to obtain a final energy consumption prediction model. Specifically, combining the first prediction model and the second prediction model to obtain an energy consumption prediction model can include:

[0095] Determine the prediction errors of the first prediction model and the second prediction model according to the historical consumption of different emission sources in the target area at different time points;

[0096] Determining weights of the first prediction model and the second prediction model based on prediction errors of the first prediction model and the second prediction model;

[0097] According to the weights of the first prediction model and the second prediction model, the first prediction model and the second prediction model are combined to obtain an energy consumption prediction model that satisfies the following relationship:

[0098]

[0099] Where σ1 and σ2 are the prediction errors of the first prediction model and the second prediction model respectively, and Y t is the consumption of all emission sources at time node t predicted by the first prediction model, y t is the consumption of all emission sources at time node t predicted by the second prediction model, It is the consumption forecast result of all emission sources at time node t output by the energy consumption forecast model.

[0100] The determination of the prediction error between the first prediction model and the second prediction model may include:

[0101] The historical consumption of different emission sources in the target area at different time nodes is processed by the first prediction model and the second prediction model, and the first predicted consumption and the second predicted consumption at multiple time nodes are respectively given;

[0102] Based on the first predicted consumption and the second predicted consumption at multiple time nodes, and the historical consumption of different emission sources in the target area at different time nodes, the prediction errors of the first prediction model and the second prediction model are respectively given. The prediction error can be one of the model accuracy, F1 score, confidence, and the difference between the true value and the predicted value.

[0103] After obtaining the energy consumption prediction model, the total carbon emissions of the target area can be predicted by obtaining the consumption of the power emission sources in the target area at the current time node and all emission sources in a certain period of time before the current time node. Figure 3 As shown, the present invention performs carbon emission analysis processing on the power emission source at the current time node and the predicted consumption of other emission sources at the current time node except the power emission source, and obtains the total carbon emissions of all emission sources at the current time node, which may include:

[0104] Obtain the carbon emission factors of all emission sources at the current time node;

[0105] The predicted consumption of electricity emission sources at the current time node and other emission sources except electricity emission sources at the current time node are respectively used with the corresponding carbon emission factors to calculate carbon emissions, and the carbon emissions of all emission sources at the current time node are obtained;

[0106] The carbon emissions of all emission sources at the current time node are accumulated to obtain the total carbon emissions of all emission sources at the current time node, which satisfies the following relationship:

[0107] C(t)=M(t)*f m (t)+O(t)*f o (t)+Q(t)*f Q (t)+D(t)*f D (t)

[0108] Where C(t) is the total carbon emissions of the target area at time node t, M(t), O(t), Q(t), and D(t) are the consumption of each emission source at time node t, and f m (t), f o (t), f Q (t), fD (t) are the unit energy consumption carbon emission factors of each emission source at time node t. Specifically, M(t), O(t), Q(t), and D(t) are the consumption of raw coal, crude oil, electricity, and natural gas at time node t, respectively. m (t), f o (t), f Q (t), f D (t) are the carbon emission factors per unit energy consumption of raw coal, crude oil, electricity and natural gas at time node t.

[0109] The present invention establishes a comprehensive carbon emission prediction system by combining principal component analysis, long short-term memory network and autoregressive distributed lag model. First, the principal component analysis method is used to calculate the weight of each emission source at different time nodes to ensure that the contribution of each energy consumption to the overall carbon emissions is fully reflected. Then, the historical consumption data is trained using the LSTM model (long short-term memory network model) and the ARDL model (autoregressive distributed lag model), and prediction models are established respectively, and the optimal model is selected by comparison for real-time prediction. This method not only improves the accuracy of carbon emission prediction, but also dynamically adjusts the model parameters according to actual conditions, ensures the reliability and applicability of the prediction results, and provides scientific decision-making support for carbon emission management in the target area.

[0110] After the total carbon emissions at the current time node are predicted, in order to ensure the accuracy of energy consumption prediction using the energy consumption prediction model at future time nodes, the energy consumption prediction model can be updated and optimized when the actual consumption is obtained. Specifically, after the total carbon emissions of all emission sources at the current time node are obtained, it can include:

[0111] Obtain the actual consumption of other emission sources except electricity emission sources at the current time node;

[0112] Based on the actual consumption of all emission sources at the current time node and the predicted consumption of other emission sources except the power emission source at the current time node, the model parameters of the pre-built energy consumption prediction model are adjusted to obtain the optimized energy consumption prediction model, including:

[0113] According to the actual consumption of all emission sources at the current time node and the predicted consumption of other emission sources except the power emission source at the current time node, the absolute value of the difference between the actual consumption and the predicted consumption of the first prediction model and the second prediction model in the energy consumption prediction model is obtained;

[0114] According to the weights of the first prediction model and the second prediction model, and the corresponding absolute values ​​of the differences, a weighted absolute value of the difference of the energy consumption prediction model is obtained;

[0115] According to the absolute value of the weighted difference of the energy consumption prediction model, the average value of the absolute value of the weighted difference is given;

[0116] With the minimum absolute value of the weighted difference as the goal, the model parameters of the energy consumption prediction model are adjusted through an optimization algorithm; wherein the optimization algorithm can be a gradient descent or a genetic algorithm.

[0117] The present invention takes 2020 as the current time node to be predicted and the consumption of raw coal emission sources in a certain park as the prediction target. It needs to obtain at least the electricity consumption x in 2020. t =422723 (10,000 kWh), electricity consumption in 2019 x t-1 =419440 (10,000 kWh), raw coal consumption in 2019 t-1 =3480271(t). Substituting the above data into the second prediction model for prediction, it can be obtained that the raw coal consumption in 2020 predicted by the second prediction model is 3843066(t).

[0118] Specifically, t =659125.9+0.4475y t-1 +1.3202x t +2.5473x t-1 =3843066(t). At the current time node, the δ0 of the second prediction model is 659125.9, φ1 is 0.4475, and β0 and β1 are 1.3202 and 2.5473 respectively.

[0119] In addition, based on the historical consumption of each emission source in 2019, the first prediction model (LSTM model) was used to predict the consumption in 2020, and the predicted result of raw coal consumption was 3780428 (t).

[0120] After predicting the consumption of raw coal in 2020 according to the first prediction model and the second prediction model respectively, the results of the two prediction models can be combined. For example, the weights of the first prediction model and the second prediction model are equal, and the weighted raw coal consumption prediction value of the first prediction model and the second prediction model is the average of the two: 3811747 (t). For the prediction of other emission sources in 2020, the same steps as the prediction of raw coal can be used, which will not be repeated here.

[0121] Based on the above forecast, energy consumption and carbon emission factors are estimated to calculate the carbon emissions of each emission source. Among them, the carbon emissions of raw coal in 2020 are:

[0122] CE ij =AD ij *NCV i *CCi *O ij

[0123] Among them, CE ij represents the carbon emissions generated by energy consumption i in target area j, AD ij NCV represents the consumption of energy i in target area j. i Indicates the net calorific value generated by consuming unit mass energy i, CC i It represents the amount of CO2 emitted when energy i produces unit net calorific value, O ij represents the oxidation coefficient of target area j when consuming energy i. Specifically, in this embodiment, when the energy type is raw coal, AD ij =3811747(t), NCV i =0.21PJ / (10 4 t), CC i =26.32(tC / TJ),O ij = 84%, the carbon emissions generated by the consumption of raw coal in target area j (park j) in 2020 are calculated as: CE ij =1769736.17(t). Using the same calculation method, the carbon emissions of all emission sources in Park J in 2020 can be calculated. Finally, the carbon emissions of all emission sources in 2020 can be integrated to obtain the total carbon emissions of Park J in 2020.

[0124] like Figure 4 As shown, the present invention also provides a carbon emission prediction device with multiple emission sources, which adopts the above-mentioned carbon emission prediction method with multiple emission sources, specifically including:

[0125] A data acquisition module, which is used to obtain the consumption of electricity emission sources in the target area at the current time node and all emission sources in a certain period of time before the current time node;

[0126] The consumption prediction module is used to analyze and process the consumption of the power emission source at the current time node and all emission sources in a certain period of time before the current time node based on the pre-built energy consumption prediction model, and predict the consumption of other emission sources except the power emission source at the current time node;

[0127] The carbon emission prediction module is used to perform carbon emission analysis on the predicted consumption of the power emission source at the current time node and other emission sources except the power emission source at the current time node to obtain the total carbon emissions of all emission sources at the current time node.

[0128] The carbon emission prediction device also includes a model building module, including:

[0129] Obtain the historical consumption of all emission sources in the target area at different time points;

[0130] Analyze and process the historical consumption of all emission sources at adjacent time nodes, and the historical consumption of power emission sources and other emission sources except power emission sources at continuous time nodes, and obtain the first prediction model and the second prediction model respectively;

[0131] The first prediction model and the second prediction model are combined to obtain an energy consumption prediction model.

[0132] Model building modules, including:

[0133] Based on the historical consumption of all emission sources at adjacent time nodes and combined with the weights of different emission sources relative to carbon emissions, the long short-term memory network model is iteratively trained to give the first prediction model, which satisfies the following relationship:

[0134] Y t =f LSTM (X t-1 )

[0135]

[0136] Where Y t is the first predicted consumption of each emission source in the target area at time node t, f LSTM (X t ) is the target area in the input sample X t-1 The long short-term memory network model under X t-1 is the input sample, that is, the historical consumption of each emission source at the time node t-1, w M 、w O 、w Q 、w D are the weights of each emission source, M(t-1), O(t-1), Q(t-1), and D(t-1) are the historical consumption of each emission source at the time node t-1.

[0137] Model building modules, including:

[0138] Normalize the historical consumption of different emission sources at different time points in the target area;

[0139] The normalized historical consumption is standardized;

[0140] Calculate the covariance matrix and perform eigenvalue decomposition on the standardized historical consumption;

[0141] Based on the eigenvalue decomposition results, principal component analysis is performed, and the weight of each emission source relative to carbon emissions is given.

[0142] Model building modules, including:

[0143] The power emission source is paired with each other emission source to obtain multiple emission source groups; the autoregressive distribution time lag model is trained with the historical consumption of each emission source group at different time nodes, and the parameters of the autoregressive distribution time lag model are updated by the least squares method to obtain multiple second prediction models corresponding to each emission source group, satisfying the following relationship:

[0144]

[0145] In the formula, y' i,t is the second predicted consumption of the ith emission source at time node t, y i,t-p is the historical consumption of the ith emission source at time tp, x t-q is the historical consumption of the power emission source at the time node tq; δ0 is a constant term, P is the number of time nodes of the i-th emission source, Q is the number of time nodes of the power emission source, φ i,t-p is the fitting parameter of the ith emission source at the tp time node, β t-q is the fitting parameter of the power emission source at the time node tq, ε t is the noise term.

[0146] Model building modules also include:

[0147] Determine the prediction errors of the first prediction model and the second prediction model according to the historical consumption of different emission sources in the target area at different time points;

[0148] Determining weights of the first prediction model and the second prediction model based on prediction errors of the first prediction model and the second prediction model;

[0149] According to the weights of the first prediction model and the second prediction model, the first prediction model and the second prediction model are combined to obtain an energy consumption prediction model that satisfies the following relationship:

[0150]

[0151] Where σ1 and σ2 are the prediction errors of the first prediction model and the second prediction model respectively, and Y t is the consumption of all emission sources at time node t predicted by the first prediction model, y t is the consumption of all emission sources at time node t predicted by the second prediction model, and Y(t) is the consumption prediction result of all emission sources at time node t output by the energy consumption prediction model.

[0152] Model building modules also include:

[0153] The historical consumption of different emission sources in the target area at different time nodes is processed by the first prediction model and the second prediction model, and the first predicted consumption and the second predicted consumption at multiple time nodes are respectively given;

[0154] Based on the first predicted consumption and the second predicted consumption at multiple time nodes and the historical consumption of different emission sources in the target area at different time nodes, the prediction errors of the first prediction model and the second prediction model are given respectively.

[0155] Carbon emission prediction module, including:

[0156] Obtain the carbon emission factors of all emission sources at the current time node;

[0157] The predicted consumption of electricity emission sources at the current time node and other emission sources except electricity emission sources at the current time node are respectively used with the corresponding carbon emission factors to calculate carbon emissions, and the carbon emissions of all emission sources at the current time node are obtained;

[0158] The carbon emissions of all emission sources at the current time node are accumulated to obtain the total carbon emissions of all emission sources at the current time node, which satisfies the following relationship:

[0159] C(t)=M(t)*f m (t)+O(t)*f o (t)+Q(t)*f Q (t)+D(t)*f D (t)

[0160] Where C(t) is the total carbon emissions of the target area at time node t, M(t), O(t), Q(t), and D(t) are the consumption of each emission source at time node t, and f m (t), f o (t), f Q (t), f D (t) are the unit energy consumption carbon emission factors of each emission source at time node t. Specifically, M(t), O(t), Q(t), and D(t) are the consumption of raw coal, crude oil, electricity, and natural gas at time node t, respectively. m (t), f o (t), f Q (t), f D (t) are the carbon emission factors per unit energy consumption of raw coal, crude oil, electricity and natural gas at time node t.

[0161] The carbon emission prediction device also includes a model updating module, and further includes:

[0162] Obtain the actual consumption of other emission sources except electricity emission sources at the current time node;

[0163] Based on the actual consumption of all emission sources at the current time node and the predicted consumption of emission sources other than electricity emission sources at the current time node, the model parameters of the pre-built energy consumption prediction model are adjusted to obtain an optimized energy consumption prediction model.

[0164] The model update module also includes:

[0165] According to the actual consumption of all emission sources at the current time node and the predicted consumption of other emission sources except the power emission source at the current time node, the absolute value of the difference between the actual consumption and the predicted consumption of the first prediction model and the second prediction model in the energy consumption prediction model is obtained;

[0166] According to the weights of the first prediction model and the second prediction model, and the corresponding absolute values ​​of the differences, a weighted absolute value of the difference of the energy consumption prediction model is obtained;

[0167] According to the absolute value of the weighted difference of the energy consumption prediction model, the average value of the absolute value of the weighted difference is given;

[0168] With the minimum absolute value of the weighted difference as the goal, the model parameters of the energy consumption prediction model are adjusted through an optimization algorithm; wherein the optimization algorithm can be a gradient descent or a genetic algorithm.

[0169] Although preferred embodiments of the present invention have been described, additional changes and modifications may be made to these embodiments by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A carbon emission prediction method with multiple emission sources, characterized in that: The specific steps include: Obtain the consumption of electricity emission sources in the target area at the current time node and all emission sources in a certain period of time before the current time node; Based on the pre-built energy consumption forecasting model, the consumption of the power emission source at the current time node and all emission sources in a certain period of time before the current time node are analyzed and processed, and the consumption of other emission sources except the power emission source at the current time node is predicted; Carbon emissions analysis is performed on the electricity emission sources at the current time node and the predicted consumption of other emission sources except the electricity emission sources at the current time node to obtain the total carbon emissions of all emission sources at the current time node.

2. The carbon emission prediction method according to claim 1, characterized in that: Other emission sources besides electricity emission sources include at least one of raw coal, crude oil and natural gas.

3. The carbon emission prediction method according to claim 1, characterized in that: After obtaining the total carbon emissions of all emission sources at the current time node, it also includes: Obtain the actual consumption of other emission sources except electricity emission sources at the current time node; Based on the actual consumption of all emission sources at the current time node and the predicted consumption of emission sources other than electricity emission sources at the current time node, the model parameters of the pre-built energy consumption prediction model are adjusted to obtain an optimized energy consumption prediction model.

4. The carbon emission prediction method according to claim 1, characterized in that: Pre-built energy consumption forecasting models, including: Obtain the historical consumption of all emission sources in the target area at different time points; Analyze and process the historical consumption of all emission sources at adjacent time nodes, and the historical consumption of power emission sources and other emission sources except power emission sources at continuous time nodes, and obtain the first prediction model and the second prediction model respectively; The first prediction model and the second prediction model are combined to obtain an energy consumption prediction model.

5. The carbon emission prediction method according to claim 4, characterized in that: Combining the first prediction model and the second prediction model, an energy consumption prediction model is obtained, including: Determine the prediction errors of the first prediction model and the second prediction model according to the historical consumption of different emission sources in the target area at different time points; Determining weights of the first prediction model and the second prediction model based on prediction errors of the first prediction model and the second prediction model; According to the weights of the first prediction model and the second prediction model, the first prediction model and the second prediction model are combined to obtain an energy consumption prediction model.

6. The carbon emission prediction method according to claim 4, characterized in that: The historical consumption of all emission sources at adjacent time nodes is analyzed and processed to obtain the first prediction model, including: Based on the historical consumption of all emission sources at adjacent time nodes and combined with the weights of different emission sources relative to carbon emissions, the long short-term memory network model is iteratively trained to give the first prediction model, which satisfies the following relationship: Y t =f LSTM (X t-1 ) X t-1 =[w M *M(t-1),w O *O(t-1),w Q *Q(t-1),w D *D(t-1)] Where Y t is the first predicted consumption of each emission source in the target area at time node t, f LSTM (X t ) is the target area in the input sample X t-1 The long short-term memory network model under X t-1 is the input sample, that is, the historical consumption of each emission source at the time node t-1, w M 、w O 、w Q 、w D are the weights of each emission source, M(t-1), O(t-1), Q(t-1), and D(t-1) are the historical consumption of each emission source at the time node t-1.

7. The carbon emission prediction method according to claim 6, characterized in that: The weight of different emission sources relative to carbon emissions, including: Normalize the historical consumption of different emission sources at different time points in the target area; The normalized historical consumption is standardized; Calculate the covariance matrix and perform eigenvalue decomposition on the standardized historical consumption; Based on the eigenvalue decomposition results, principal component analysis is performed, and the weight of each emission source relative to carbon emissions is given.

8. The carbon emission prediction method according to any one of claims 4 to 7, characterized in that: The historical consumption of power emission sources and other emission sources other than power emission sources at continuous time nodes is analyzed and processed to obtain the second prediction model, including: The power emission source is paired with each other emission source to obtain multiple emission source groups; the autoregressive distribution time lag model is trained with the historical consumption of each emission source group at different time nodes, and the parameters of the autoregressive distribution time lag model are updated by the least squares method to obtain multiple second prediction models corresponding to each emission source group, satisfying the following relationship: In the formula, y' i,t is the second predicted consumption of the ith emission source at time node t, y i,t-p is the historical consumption of the ith emission source at time tp, x t-q is the historical consumption of the power emission source at the time node tq; δ0 is a constant term, P is the number of time nodes of the i-th emission source, Q is the number of time nodes of the power emission source, φ i,t-p is the fitting parameter of the ith emission source at the tp time node, β t-q is the fitting parameter of the power emission source at the time node tq, ε t is the noise term.

9. The carbon emission prediction method according to claim 8, characterized in that: Perform carbon emission analysis on the power emission source at the current time node and the predicted consumption of other emission sources at the current time node except the power emission source to obtain the total carbon emissions of all emission sources at the current time node, including: Obtain the carbon emission factors of all emission sources at the current time node; The predicted consumption of electricity emission sources at the current time node and other emission sources except electricity emission sources at the current time node are respectively used with the corresponding carbon emission factors to calculate carbon emissions, and the carbon emissions of all emission sources at the current time node are obtained; The carbon emissions of all emission sources at the current time node are accumulated to obtain the total carbon emissions of all emission sources at the current time node, which satisfies the following relationship: C(t)=M(t)*f m (t)+O(t)*f o (t)+Q(t)*f Q (t)+D(t)*f D (t) Where C(t) is the total carbon emissions of the target area at time node t, M(t), O(t), Q(t), and D(t) are the consumption of each emission source at time node t, and f m (t), f o (t), f Q (t), f D (t) are the carbon emission factors per unit energy consumption of each emission source at time node t.

10. A carbon emission prediction device with multiple emission sources, characterized in that: Using the carbon emission prediction method with multiple emission sources as described in any one of claims 1 to 9, the carbon emission prediction device includes: A data acquisition module, which is used to obtain the consumption of electricity emission sources in the target area at the current time node and all emission sources in a certain period of time before the current time node; The consumption prediction module is used to analyze and process the consumption of the power emission source at the current time node and all emission sources in a certain period of time before the current time node based on the pre-built energy consumption prediction model, and predict the consumption of other emission sources except the power emission source at the current time node; The carbon emission prediction module is used to perform carbon emission analysis on the predicted consumption of the power emission source at the current time node and other emission sources except the power emission source at the current time node to obtain the total carbon emissions of all emission sources at the current time node.