Long-term prediction method for enterprise carbon factor

Abstract

This invention relates to a long-term prediction method for enterprise carbon emission factors, belonging to the field of long-term time series forecasting. The invention mainly comprises two parts: calculation of dynamic enterprise carbon emission factors and long-term prediction of dynamic enterprise carbon emission factors. The calculation process of dynamic enterprise carbon emission factors is as follows: first, the 220kV level carbon emission factor is calculated at different time points using a real-time multi-level power flow tracing framework; second, the 110kV level carbon emission factor is calculated; and finally, the enterprise-level carbon emission factor is calculated. The long-term prediction process of dynamic enterprise carbon emission factors is as follows: the Spearman correlation coefficients of all data features and carbon emission factors are calculated; a correlation threshold is set, and features with correlations greater than the threshold are added to the carbon emission factor prediction dataset; a long-term prediction model for carbon emission factors is designed based on autocorrelation mechanisms and empirical mode decomposition; multivariate prediction of carbon emission factors is performed at 96 steps; and finally, the prediction results are visualized. This invention accurately calculates and predicts enterprise carbon emission factor sequences, providing effective guidance for enterprises' low-carbon electricity use.

Description

Technical Field

[0001] This invention belongs to the field of long-term time series forecasting, specifically relating to a method for long-term forecasting of enterprise carbon factors. Background Technology

[0002] As "dual carbon" practices continue to evolve towards precise carbon control, how to effectively reduce corporate carbon emissions from electricity consumption is a current hot research topic. Accurately measuring corporate carbon emissions from electricity consumption and rationally optimizing corporate electricity consumption behavior are two key approaches to solving this problem. Currently, measuring corporate carbon emissions from electricity consumption mainly relies on the carbon factor method. This method is suitable for large-scale data calculations, but its accuracy is easily affected by the carbon factor value. For a particular company, its electricity cleanliness changes constantly. Using a single constant factor to characterize its carbon emissions per unit of electricity consumption over a considerable period is inappropriate. Therefore, there is an urgent need to study personalized, enterprise-level dynamic carbon factors. Based on this, we can explore the potential trends in carbon factor changes and guide corporate electricity consumption behavior towards low-carbon practices.

[0003] The challenge in calculating enterprise-level dynamic carbon emission factors lies in modeling time-sharing power flow tracking. The structure from the power grid to the enterprise is multi-layered, and each level's power flow tracking model is dynamically evolving. The model needs to cover various typical scenarios and ensure high efficiency throughout the calculation process. For predicting enterprise carbon emission factor sequences, to allow sufficient production scheduling time, the prediction model must be able to characterize the long-term trend of carbon emission factors. Based on the above analysis, this invention proposes a long-term prediction method for enterprise carbon emission factors. Through a real-time multi-layered power flow tracking framework, personalized enterprise dynamic carbon emission factors are calculated, accurately characterizing the carbon emission characteristics of electricity consumption at different times. Based on this, and considering the characteristics of the carbon emission factor sequence, a long-term prediction model is proposed to predict enterprise carbon emission factors over the long term, accurately reflecting the changing trends of enterprise carbon emission factors and providing scientific and effective guidance for low-carbon electricity consumption timing for discrete manufacturing enterprises. Summary of the Invention

[0004] This invention aims to dynamically calculate and predict enterprise-level carbon emissions over the long term, providing guidance for enterprises with discretizable production processes on low-carbon electricity usage, promoting the use of clean electricity as much as possible in enterprise production, reducing carbon emissions from electricity consumption, improving carbon efficiency, and effectively driving enterprises to achieve carbon peak as soon as possible.

[0005] The long-term prediction method for enterprise carbon factor includes the following steps:

[0006] S1. Collect power grid flow data, enterprise purchased electricity data, and enterprise self-generated and self-consumed electricity data, and calculate the dynamic carbon factor at the enterprise level based on a real-time multi-level power flow tracking framework;

[0007] S2. Through correlation analysis, screen the features closely related to the carbon factor in power grid flow data, enterprise load data and local meteorological data, and construct an enterprise carbon factor prediction dataset;

[0008] S3. Based on autocorrelation mechanism and empirical mode decomposition, design a long-term prediction model for the electric carbon factor;

[0009] S4. Divide the electric carbon factor prediction dataset into training set, validation set, and test set to train, validate, and test the long-term prediction model for the electric carbon factor.

[0010] S5. Visualize the long-term prediction results of enterprise carbon factor, providing guidance for enterprises to rationally arrange production plans and adjust electricity consumption timing.

[0011] Furthermore, the calculation of the enterprise-level dynamic carbon factor based on the real-time multi-level power flow tracking framework in S1 includes the following steps:

[0012] S1.1. Process power grid flow data, enterprise purchased electricity data, and enterprise self-generated and self-consumed electricity data to the same 15-minute time granularity, and perform data preprocessing. Abstract the carbon flow path from the power grid to the enterprise into a three-layer structure, including a 220KV layer. 220KV 110KV Layer 110KV and Enterprise Layer E .

[0013] S1.2. Building Layers 220KV All power generation nodes G 220KV,i , 1 < i < m 220KV Power consumption node L 220KV,j , 1 < j < n 220KV Network loss node 220KV,k , 1 < k < p 220KV , transmission line 220KV,s , 1 < s < q 220KV The topological relationship model, where m 220KV ,n 220KV ,p 220KV ,q 220KV Layer 220KV The number of power generation nodes, power consumption nodes, network loss nodes, and transmission lines is determined based on the actual output of each power generation node. The decision to designate a power generation node as a power consumption node is made, and the Layer is continuously updated. 220KV The topological relationships are established and a standard models are formed. The Layer is then analyzed using a complex power proportional power flow tracing algorithm. 220KV The power sources of all power consumption nodes, combined with Layer 220KV Basic carbon emission coefficient C at each power generation node 220KV,iAnd calculate the 15-minute dynamic carbon dioxide factor F for each electricity consumption node. 220KV,j , 1 < j < n 220KV ;

[0014] S1.3. Conduct in-depth investigations of the 220KV power consumption nodes belonging to the enterprise. Building a Layer 110KV All power generation nodes G 110KV,i , 1 < i < m 110KV Power consumption node L 110KV,j , 1 < j < n 110KV Network loss node 110KV,k , 1 < k < p 110KV , transmission line 110KV,s , 1 < s < q 110KV The topological relationship model, where m 110KV ,n 110KV ,p 110KV ,q 110KV Layer 110KV The number of power generation nodes, power consumption nodes, network loss nodes, and transmission lines is determined based on the actual output of each power generation node. The decision to designate a power generation node as a power consumption node is made, and the Layer is continuously updated. 110KV The topological relationships are established and b standard models are formed. The Layer is then analyzed using a complex power proportional power flow tracing algorithm. 110KV The power sources of all power consumption nodes, combined with Layer 110KV Basic carbon emission coefficient C at each power generation node 110KV,i Calculate the 15-minute dynamic carbon dioxide factor F for each electricity consumption node. 110KV,j , 1 < j < n 110KV Among them, Layer 110KV Purchased electricity base carbon emission coefficient express;

[0015] S1.4. Conduct in-depth investigations of the 110KV power consumption nodes belonging to the enterprise. Viewing a company as an electricity consumption node L E The electricity purchased by enterprises and the electricity generated and consumed by enterprises are regarded as generation node G. E,i , 1 < i < m E Based on the actual output of each power generation node, decide whether to designate that power generation node as a power consumption node, update and build the Layer. E Loss of all power generation nodes, power consumption nodes, and network loss nodes E,k , 1 < k < p E , transmission line E,s , 1 < s < q E c topological relationship models, where m E ,p E ,qE Layer E The number of generating nodes, network loss nodes, and transmission lines is analyzed using a complex power proportional flow tracing algorithm to determine the source of enterprise electricity consumption, combined with Layer... E Basic carbon emission coefficient C at each power generation node E,i Calculate the dynamic electrocarbon factor F of the enterprise at the 15-minute level. E Among them, the basic carbon emission coefficient of purchased electricity by enterprises is based on express.

[0016] Furthermore, the construction of the enterprise electric carbon factor prediction dataset in S2 includes the following steps:

[0017] S2.1. Through Spearman correlation analysis, calculate all characteristics of power grid flow data, enterprise load data, and local meteorological data, and their correlation with the enterprise's carbon factor F. E The correlations form a Z×1 correlation matrix Corr. SP Each row represents the data feature of each dimension and F. E The degree of correlation between them;

[0018] S2.2.Corr SP A negative value in F indicates that the feature dimension is related to F. E Negative correlation; a positive value indicates that this feature dimension is related to F. E Positive correlation, for Corr SP The absolute value of the value in the middle is used to form the Corr. SP,abs And calculate the average value as the correlation threshold T. Corr , will Corr SP,abs Median greater than T Corr The features were selected and added to the electric carbon factor prediction dataset.

[0019] Furthermore, the long-term prediction model for the electric carbon factor in S3 consists of two identical encoders and one decoder. The encoder of the long-term prediction model for the electric carbon factor is composed of an autocorrelation module, a sequence decomposition and recombination module 1, a feedforward network module, and a sequence decomposition and recombination module 2 connected in sequence. The decoder of the long-term prediction model for the electric carbon factor is composed of an autocorrelation module 1, a sequence decomposition and recombination module 1, an autocorrelation module 2, a sequence decomposition and recombination module 2, a feedforward network module, and a sequence decomposition and recombination module 3 connected in sequence.

[0020] Furthermore, the design of the autocorrelation module follows the design of the Auto-Correlation module in Autoformer.

[0021] Furthermore, the sequence decomposition and recombination module uses the empirical mode decomposition method to decompose the input sequence X into 5 IMFs and 1 residual component. The residual component is discarded, and the odd-numbered IMFs and even-numbered IMFs are recombined into two new subsequences X1 and X2, respectively.

[0022] Furthermore, in S4, the ratio of the training set, validation set, and test set is 7:2:1, with each step being 15 minutes. The model input is the current consecutive 96 steps, and the model output is the subsequent consecutive 96 steps.

[0023] Furthermore, the visualization of the long-term prediction results of the enterprise's electrocarbon factor in S5 includes the following steps:

[0024] S5.1. Visualize the prediction results of the enterprise's carbon factor in the form of time series curves;

[0025] S5.2. Capture and label the peak and trough values ​​of the enterprise's carbon factor in every consecutive 96 steps.

[0026] Compared with existing technologies, the beneficial effects of this invention are as follows: It proposes for the first time a method for calculating enterprise-level carbon emission factors. Through a real-time multi-level power flow tracking framework, the power flow tracking model is dynamically updated. The hierarchical progressive approach is used to accurately characterize the 15-minute granularity of enterprise carbon emission factors. Based on autocorrelation mechanism and empirical mode decomposition technology, a long-term prediction model for carbon emission factors is designed to accurately reflect the changing trend of enterprise carbon emission factors, providing scientific and effective guidance for low-carbon electricity consumption timing for discrete manufacturing enterprises. Attached Figure Description

[0027] Figure 1 This is a flowchart of the long-term prediction method for the enterprise's electrocarbon factor in this invention;

[0028] Figure 2 This is a schematic diagram of a multi-level power flow tracing topology model at a certain moment;

[0029] Figure 3 This is a schematic diagram of the calculation results of the enterprise's dynamic carbon factor sequence;

[0030] Figure 4 This is a schematic diagram of the long-term prediction model structure for the electrocarbon factor;

[0031] Figure 5 This is a schematic diagram illustrating the long-term forecast results of the dynamic carbon factor change trend of an enterprise.

[0032] Specific implementation steps

[0033] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The described embodiments are only one of the embodiments of the present invention.

[0034] The technical solution adopted in the long-term prediction method for enterprise electrocarbon factor of the present invention is as follows:

[0035] First, power flow data, enterprise purchased electricity data, and enterprise self-generated and self-consumed electricity data are collected. Based on a real-time multi-level power flow tracing framework, dynamic electricity carbon factor at the enterprise level is calculated. This method targets high-power-consuming enterprises with discretizable production processes, such as textile enterprises.

[0036] Furthermore, power grid flow data, enterprise purchased electricity data, and enterprise self-generated and self-consumed electricity data are processed to the same 15-minute time granularity and preprocessed; the carbon flow path from the power grid to the enterprise is abstracted into a three-layer structure, including a 220KV layer. 220KV ), 110KV layer 110KV ) and Enterprise Layer E ).

[0037] First, build the Layer 220KV All power generation nodes G 220KV,i , 1 < i < m 220KV Power consumption node L 220KV,j , 1 < j < n 220KV Network loss node 220KV,k , 1 < k < p 220KV , transmission line 220KV,s , 1 < s < q 220KV The topological relationship model, where m 220KV ,n 220KV ,p 220KV ,q 220KV Layer 220KV The number of power generation nodes, power consumption nodes, network loss nodes, and transmission lines. Since some nodes exhibit different characteristics of power generation and consumption at different times, the Layer is continuously updated based on the actual output of each power generation node to determine whether to classify it as a power consumption node. 220KV The topological relationships are established and a standard model is formed. Layers are analyzed using a complex power proportional power flow tracing algorithm. 220KV The power sources of all power consumption nodes, combined with Layer 220KV Basic carbon emission coefficient C at each power generation node 220KV,i And calculate the 15-minute dynamic carbon dioxide factor F for each electricity consumption node. 220KV,j , 1 < j < n 220KV .

[0038] Secondly, we went deep into the 220KV power consumption nodes belonging to the enterprises. Building a Layer 110KV All power generation nodes G 110KV,i , 1 < i < m110KV Power consumption node L 110KV,j , 1 < j < n 110KV Network loss node 110KV,k , 1 < k < p 110KV , transmission line 110KV,s , 1 < s < q 110KV The topological relationship model, where m 110KV ,n 110KV ,p 110KV ,q 110KV Layer 110KV The number of power generation nodes, power consumption nodes, network loss nodes, and transmission lines. Based on the actual output of each power generation node, decide whether to designate it as a power consumption node, and continuously update the Layer. 110KV The topological relationships are established and b standard models are formed. Layers are analyzed using a complex power proportional power flow tracing algorithm. 110KV The power sources of all power consumption nodes, combined with Layer 110KV Basic carbon emission coefficient C at each power generation node 110KV,i Among them, Layer 110KV Purchased electricity base carbon emission coefficient This indicates that the dynamic carbon dioxide factor F at the 15-minute level is calculated for each electricity consumption node. 110KV,j , 1 < j < n 110KV .

[0039] Finally, we went deep into the 110KV power consumption nodes belonging to the enterprise. Viewing a company as an electricity consumption node L E The electricity purchased by enterprises and the electricity generated and consumed by enterprises are regarded as generation node G. E,i , 1 < i < m E Based on the actual output of each power generation node, decide whether to designate that power generation node as a power consumption node, update and build the Layer. E Loss of all power generation nodes, power consumption nodes, and network loss nodes E,k , 1 < k < p E , transmission line E,s , 1 < s < q E c topological relationship models, where m E ,p E ,q E Layer E The number of generating nodes, network loss nodes, and transmission lines. The source of enterprise electricity consumption is analyzed using a complex power proportional flow tracing algorithm, combined with Layer... E Basic carbon emission coefficient C at each power generation node E,i Among them, the basic carbon emission coefficient of purchased electricity by enterprises is based on This indicates that the dynamic electrocarbon factor F of the enterprise is calculated at the 15-minute level. E .

[0040] Furthermore, through correlation analysis, features closely related to the carbon factor in power grid flow data, enterprise load data, and local meteorological data were screened to construct an enterprise carbon factor prediction dataset.

[0041] First, through Spearman correlation analysis, all characteristics of power grid flow data, enterprise load data, and local meteorological data are calculated and correlated with the enterprise's carbon factor F. E The correlations form a Z×1 correlation matrix Corr. SP Each row represents the data feature of each dimension and F. E The degree of correlation between them.

[0042] Then, Corr SP A negative value in F indicates that the feature dimension is related to F. E Negative correlation; a positive value indicates that this feature dimension is related to F. E Positive correlation. For Corr SP The absolute value of the value in the middle is used to form the Corr. SP,abs And calculate the average value as the correlation threshold T. Corr , will Corr SP,abs Median greater than T Corr The features were selected and added to the electric carbon factor prediction dataset.

[0043] Furthermore, a long-term prediction model for the electric carbon factor is designed based on autocorrelation mechanisms and empirical mode decomposition. The long-term prediction model for the electric carbon factor consists of two identical encoders and one decoder. The encoder of the long-term prediction model for the electric carbon factor is composed of an autocorrelation module, a sequence decomposition and recombination module 1, a feedforward network module, and a sequence decomposition and recombination module 2 connected sequentially. The decoder of the long-term prediction model for the electric carbon factor is composed of an autocorrelation module 1, a sequence decomposition and recombination module 1, an autocorrelation module 2, a sequence decomposition and recombination module 2, a feedforward network module, and a sequence decomposition and recombination module 3 connected sequentially.

[0044] The design of the autocorrelation module follows the design of the Auto-Correlation module in Autoformer. The sequence decomposition and recombination module adopts the empirical mode decomposition method to decompose the input sequence X into 5 IMFs and 1 residual component. The residual component is discarded, and the odd-numbered IMFs and even-numbered IMFs are recombined into two new subsequences X1 and X2, respectively.

[0045] Furthermore, the carbon dioxide factor prediction dataset was divided into training, validation, and test sets to train, validate, and test the long-term carbon dioxide factor prediction model. The ratio of the training, validation, and test sets was 7:2:1. The model operates on a 15-minute timeframe, with the current 96 consecutive timeframes as input and the next 96 consecutive timeframes as output.

[0046] Furthermore, the long-term forecast results of enterprise electricity carbon factor are visualized to provide guidance for enterprises to rationally arrange production plans and adjust electricity consumption timing. First, the forecast results of enterprise electricity carbon factor are visualized in the form of time series curves. Second, the peak and trough values ​​of enterprise electricity carbon factor are captured and labeled in every consecutive 96 steps. Example

[0047] The long-term prediction method for enterprise electrocarbon factor described in this invention is applied to a textile enterprise as an example. Figure 1 As shown, the main steps of this invention are as follows:

[0048] First, power flow data for the 500kV, 220kV, 110kV, 35kV, and 10kV grids to which the company belongs was collected. Data on the company's purchased electricity and its self-generated electricity (mainly distributed photovoltaic power) was also collected, along with meteorological data for the company's location. All data was processed to a uniform 15-minute time granularity; that is, all subsequent calculations were performed in 15-minute increments. Data preprocessing was then performed, with missing values ​​imputed using the mean over a window of 25 steps.

[0049] Furthermore, regarding Layer 220KV Layer 110KV Layer E Perform trend tracing modeling, such as a multi-level trend tracing topology model. Figure 2 As shown.

[0050] First, build the Layer 220KV All power generation nodes G 220KV,i , 1 < i < m 220KV Power consumption node L 220KV,j , 1 < j < n 220KV Network loss node 220KV,k , 1 < k < p 220KV , transmission line 220KV,s , 1 < s < q 220KV Topological relationship model. For example... Figure 2 As shown at this moment, Layer 220KV There are three power generation nodes, ①, ②, and ③, and nine power consumption nodes, ④ to ③. Continuously update Layer 220KVThe topological relationships are established and a standard model is formed. Layers are analyzed using a complex power proportional power flow tracing algorithm. 220KV The power sources of all power consumption nodes, combined with Layer 220KV Basic carbon emission coefficient C at each power generation node 220KV,i And calculate the 15-minute dynamic carbon dioxide factor F for each electricity consumption node. 220KV,j , 1 < j < n 220KV .

[0051] Secondly, we went deep into the 220KV power consumption nodes belonging to the enterprises. Building a Layer 110KV All power generation nodes G 110KV,i , 1 < i < m 110KV Power consumption node L 110KV,j , 1 < j < n 110KV Network loss node 110KV,k , 1 < k < p 110KV , transmission line 110KV,s , 1 < s < q 110KV The topological relationship model. Continuously updating Layers. 110KV Topological relationships are established and b standard models are formed. For example... Figure 2 As shown at this moment, Layer 110KV There are three power generation nodes: "220kV power transmission," "waste incineration power generation," and "local photovoltaic station," and two power consumption nodes: "10kV line" and "other loads." Layer analysis is performed using a complex power proportional flow tracing algorithm. 110KV The power sources of all power consumption nodes, combined with Layer 110KV Basic carbon emission coefficient C at each power generation node 110KV,i Among them, Layer 110KV Purchased electricity base carbon emission coefficient This indicates that the dynamic carbon dioxide factor F at the 15-minute level is calculated for each electricity consumption node. 110KV,j , 1 < j < n 110KV .

[0052] Finally, we went deep into the 110KV power consumption nodes belonging to the enterprise. Viewing a company as an electricity consumption node L E The electricity purchased by enterprises and the electricity generated and consumed by enterprises are regarded as generation node G. E,i , 1 < i < m E Update and build Layer E Loss of all power generation nodes, power consumption nodes, and network loss nodes E,k , 1 < k < p E , transmission line E,s , 1 < s < q Ec topological relationship models. For example... Figure 2 As shown at this moment, Layer E There are two power generation nodes: a 10kV line and a self-consumption distributed photovoltaic system. The source of the enterprise's electricity consumption is analyzed using a complex power proportional flow tracing algorithm, combined with Layer... E Basic carbon emission coefficient C at each power generation node E,i Among them, the basic carbon emission coefficient of purchased electricity by enterprises is based on This indicates that the dynamic electrocarbon factor F of the enterprise is calculated at the 15-minute level. E .

[0053] Furthermore, the company's 15-minute dynamic electrocarbon factor F E The calculation results are as follows Figure 3 As shown.

[0054] Furthermore, through Spearman correlation analysis, all characteristics of power grid flow data, enterprise load data, and local meteorological data were calculated and correlated with the enterprise's carbon factor F. E The correlations form a Z×1 correlation matrix Corr. SP Each row represents the data feature of each dimension and F. E The degree of correlation between them. Corr SP A negative value in F indicates that the feature dimension is related to F. E Negative correlation; a positive value indicates that this feature dimension is related to F. E Positive correlation. For Corr SP The absolute value of the value in the middle is used to form the Corr. SP,abs And calculate the average value as the correlation threshold T. Corr , will Corr SP,abs Median greater than T Corr The features were selected and added to the electric carbon factor prediction dataset.

[0055] Furthermore, based on autocorrelation mechanisms and empirical mode decomposition, a long-term prediction model for the electric carbon factor is designed. The structure of the long-term prediction model for the electric carbon factor is as follows: Figure 4 As shown. Specifically, the long-term prediction model for the electric carbon factor consists of two identical encoders and one decoder. The encoder of the long-term prediction model for the electric carbon factor is composed of an autocorrelation module, a sequence decomposition and recombination module 1, a feedforward network module, and a sequence decomposition and recombination module 2 connected in sequence. The decoder of the long-term prediction model for the electric carbon factor is composed of an autocorrelation module 1, a sequence decomposition and recombination module 1, an autocorrelation module 2, a sequence decomposition and recombination module 2, a feedforward network module, and a sequence decomposition and recombination module 3 connected in sequence.

[0056] The design of the autocorrelation module follows the design of the Auto-Correlation module in Autoformer. The sequence decomposition and recombination module adopts the empirical mode decomposition method to decompose the input sequence X into 5 IMFs and 1 residual component. The residual component is discarded, and the odd-numbered IMFs and even-numbered IMFs are recombined into two new subsequences X1 and X2, respectively.

[0057] Furthermore, the carbon dioxide factor prediction dataset was divided into training, validation, and test sets to train, validate, and test the long-term carbon dioxide factor prediction model. The ratio of the training, validation, and test sets was 7:2:1. The model operates on a 15-minute timeframe, with the current 96 consecutive timeframes as input and the next 96 consecutive timeframes as output.

[0058] Finally, the long-term forecast results of the enterprise's electricity carbon factor are visualized to guide enterprises in rationally scheduling production plans and adjusting electricity consumption timing. First, the forecast results of the enterprise's electricity carbon factor are visualized in the form of a time series curve. Second, the peak and trough values ​​of the enterprise's electricity carbon factor are captured and labeled in each consecutive 96-step interval. The visualization results are as follows: Figure 5 As shown, line 1 represents the true value, line 2 represents the predicted value, and captures the peak and trough values ​​of the sequence.

Claims

1. A long-term prediction method for enterprise carbon factor, characterized in that, Includes the following steps: S1. Collect power grid flow data, enterprise purchased electricity data, and enterprise self-generated and self-consumed electricity data. Based on a real-time multi-level power flow tracing framework, calculate the dynamic electricity carbon factor at the enterprise level, including the following steps: S1.

1. Process power grid flow data, enterprise purchased electricity data, and enterprise self-generated and self-consumed electricity data to the same 15-minute time granularity, and perform data preprocessing. Abstract the carbon flow path from the power grid to the enterprise into a three-layer structure, including a 220KV layer. 220KV 110KV Layer 110KV and Enterprise Layer E； S1.

2. Building Layers 220KV All power generation nodes G 220KV,i , 1 < i < m 220KV Power consumption node L 220KV,j , 1 < j < n 220KV Network loss node 220KV,k , 1 < k < p 220KV , transmission line 220KV,s , 1 < s < q 220KV The topological relationship model, where m 220KV ,n 220KV ,p 220KV ,q 220KV Layer 220KV The number of power generation nodes, power consumption nodes, network loss nodes, and transmission lines is determined based on the actual output of each power generation node. The decision to designate a power generation node as a power consumption node is made, and the Layer is continuously updated. 220KV The topological relationships are established and a standard models are formed. The Layer is then analyzed using a complex power proportional power flow tracing algorithm. 220KV The power sources for all power consumption nodes, combined with Layer 220KV Basic carbon emission coefficient C at each power generation node 220KV,i And calculate the 15-minute dynamic carbon dioxide factor F for each electricity consumption node. 220KV,j , 1 < j < n 220KV ; S1.

3. Conduct in-depth investigations of the 220KV power consumption nodes belonging to the enterprise. Building a Layer 110KV All power generation nodes G 110KV,i , 1 < i < m 110KV Power consumption node L 110KV,j , 1 < j < n 110KV Network loss node 110KV,k , 1 < k < p 110KV , transmission line 110KV,s , 1 < s < q 110KV The topological relationship model, where m 110KV ,n 110KV ,p 110KV ,q 110KV Layer 110KV The number of power generation nodes, power consumption nodes, network loss nodes, and transmission lines is determined based on the actual output of each power generation node. The decision to designate a power generation node as a power consumption node is made, and the Layer is continuously updated. 110KV The topological relationships are established and b standard models are formed. The Layer is then analyzed using a complex power proportional power flow tracing algorithm. 110KV The power sources for all power consumption nodes, combined with Layer 110KV Basic carbon emission coefficient C at each power generation node 110KV,i Calculate the 15-minute dynamic carbon dioxide factor F for each electricity consumption node. 110KV,j , 1 < j < n 110KV Among them, Layer 110KV Purchased electricity base carbon emission coefficient express; S1.

4. Conduct in-depth investigations of the 110KV power consumption nodes belonging to the enterprise. Viewing a company as an electricity consumption node L E The electricity purchased by enterprises and the electricity generated and consumed by enterprises are regarded as generation node G. E,i , 1 < i < m E Based on the actual output of each power generation node, decide whether to designate that power generation node as a power consumption node, update and build the Layer. E Loss of all power generation nodes, power consumption nodes, and network loss nodes E,k , 1 < k < p E , transmission line E,s , 1 < s < q E c topological relationship models, where m E ,p E ,q E Layer E The number of generating nodes, network loss nodes, and transmission lines is analyzed using a complex power proportional flow tracing algorithm to determine the source of enterprise electricity consumption, combined with Layer... E Basic carbon emission coefficient C at each power generation node E,i Calculate the dynamic electrocarbon factor F of the enterprise at the 15-minute level. E Among them, the basic carbon emission coefficient of purchased electricity by enterprises is based on express; S2. Through correlation analysis, features closely related to the carbon factor in power grid flow data, enterprise load data, and local meteorological data are selected to construct an enterprise carbon factor prediction dataset, including the following steps: S2.

1. Through Spearman correlation analysis, calculate all characteristics of power grid flow data, enterprise load data, and local meteorological data, and their correlation with the enterprise's carbon factor F. E The correlations form a Z×1 correlation matrix Corr. SP Each row represents the data feature of each dimension and F. E The degree of correlation between them; S2.2.Corr SP A negative value in F indicates that the feature dimension is related to F. E Negative correlation; a positive value indicates that this feature dimension is related to F. E Positive correlation, for Corr SP The absolute value of the value in the middle is used to form the Corr. SP,abs And calculate the average value as the correlation threshold T. Corr , will Corr SP,abs Median greater than T Corr The features were selected and added to the electric carbon factor prediction dataset; S3. Based on autocorrelation mechanism and empirical mode decomposition, a long-term prediction model for the electric carbon factor is designed. The designed long-term prediction model for the electric carbon factor consists of two identical encoders and one decoder. The encoder of the long-term prediction model for the electric carbon factor is composed of an autocorrelation module, a sequence decomposition and recombination module 1, a feedforward network module, and a sequence decomposition and recombination module 2 connected in sequence. The decoder of the long-term prediction model for the electric carbon factor is composed of an autocorrelation module 1, a sequence decomposition and recombination module 1, an autocorrelation module 2, a sequence decomposition and recombination module 2, a feedforward network module, and a sequence decomposition and recombination module 3 connected in sequence. The design of the autocorrelation module follows the design of the Auto-Correlation module in Autoformer; The sequence decomposition and recombination module uses the empirical mode decomposition method to decompose the input sequence X into 5 IMFs and 1 residual component. The residual component is discarded, and the odd-numbered IMFs and even-numbered IMFs are recombined into two new subsequences X1 and X2, respectively. S4. Divide the electric carbon factor prediction dataset into training set, validation set, and test set to train, validate, and test the long-term prediction model for the electric carbon factor. S5. Visualize the long-term prediction results of enterprise carbon factor, providing guidance for enterprises to rationally arrange production plans and adjust electricity consumption timing.

2. The long-term prediction method for enterprise electrocarbon factor according to claim 1, characterized in that, In S4, the ratio of training set, validation set, and test set is 7:2:1, with each step being 15 minutes. The model input is the current consecutive 96 steps, and the model output is the next consecutive 96 steps.

3. The long-term prediction method for enterprise electrocarbon factor according to claim 1, characterized in that, The visualization of long-term prediction results of enterprise carbon factor in S5 includes the following steps: S5.

1. Visualize the prediction results of the enterprise's carbon factor in the form of time series curves; S5.

2. Capture and label the peak and trough values ​​of the enterprise's carbon factor in every consecutive 96 steps.

Images