Carbon emission monitoring method and system fusing decomposition regression, electronic equipment and medium

Abstract

The invention relates to the technical field of carbon emission monitoring, and provides a decomposition regression fused carbon emission monitoring method and system, electronic equipment and a medium, and the method comprises the steps: obtaining real-time power load data, carrying out the empirical mode decomposition, and removing an air conditioner load to obtain a residual IMF component; reconstructing the residual IMF component and the residual term, and outputting a real-time residual power load sequence after the influence of the air conditioner is eliminated; obtaining an electricity-carbon correlation coefficient of a previous time period and an exogenous variable value in a set historical time period, and inputting the electricity-carbon correlation coefficient and the exogenous variable value into the constructed autoregressive moving average regression model for calculation to obtain a current electricity-carbon correlation coefficient; and utilizing the real-time residual power load sequence and the current electricity-carbon correlation coefficient to estimate and obtain a near-real-time terminal fuel carbon emission estimation value of the target area as a monitoring result. According to the method, through power load data reconstruction and machine learning modeling, the regional carbon emission estimation precision is improved under the small sample condition.

Description

Technical Field

[0001] This invention relates to the field of carbon emission monitoring technology, specifically to carbon emission monitoring methods, systems, electronic devices, and media that integrate decomposition and regression. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Global climate change poses a significant threat to human society. Carbon emissions from end-use fuels account for approximately 40% of global emissions, making them a key focus of carbon reduction efforts. Accurate and timely measurement of regional end-use fuel carbon emissions is crucial for developing carbon emission control strategies and assessing the effectiveness of emission reduction efforts. More importantly, carbon emission data can also serve as input for operational optimization, used to adjust the operational strategies of urban or industrial systems. This enables the linkage between dynamic carbon emission monitoring and intelligent control, effectively promoting energy management and low-carbon transformation.

[0004] The current mainstream terminal carbon emission accounting method mainly relies on the fuel emission factor method to calculate carbon emissions based on annual fuel statistics. However, this method has poor real-time performance and is difficult to meet the needs of dynamic monitoring. Direct monitoring methods such as CEMS have high-frequency capabilities, but their deployment costs are high and their coverage is limited, making them difficult to adapt to regional monitoring scenarios. In recent years, some studies have attempted to construct an electricity-carbon coupling model based on electricity load data to estimate regional carbon emissions. Although its feasibility has been initially verified, the following main problems still exist: (1) The model has poor anti-interference ability and is easily affected by weakly correlated loads such as air conditioning, especially during climate-sensitive periods. For example, air conditioning load fluctuates significantly due to weather changes and has a low correlation with fuel carbon emissions. When its proportion is large, it can easily interfere with the model and reduce the estimation accuracy. (2) Terminal fuel carbon emission data are usually statistically measured in annual or monthly units, and the number of historical samples is limited. Under small sample conditions, model training is difficult, and the estimation accuracy and generalization ability are insufficient; (3) Poor model interpretability: Existing methods rely on historical data of electricity and carbon emissions in variable selection, and do not adequately consider exogenous factors such as time, space, and user behavior. The model interpretability is poor, and it is difficult to accurately reflect the internal driving mechanism of regional carbon emission dynamic changes. These problems limit the accuracy and reliability of existing methods in practical applications, and make it difficult to effectively support high-frequency carbon emission supervision and system operation control. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a carbon emission monitoring method, system, electronic device, and medium that integrates decomposition and regression. By reconstructing power load data and using machine learning modeling, the accuracy of regional carbon emission estimation is improved under small sample conditions.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides a carbon emission monitoring method based on fusion decomposition and regression, comprising the following steps: For the target area, obtain real-time power load data; Based on the empirical mode decomposition algorithm, the real-time power load signal is decomposed into several IMF components and residual terms, the IMF components related to air conditioning load are identified, and the remaining IMF components are obtained. Reconstruct the remaining IMF components and residual terms after removing air conditioning load, and output the real-time remaining power load sequence after removing the influence of air conditioning. Obtain the electricity-carbon correlation coefficient of the previous period and the exogenous variable values ​​within the set historical period, input them into the constructed autoregressive moving average regression model to calculate the current electricity-carbon correlation coefficient; By using the real-time remaining power load sequence and the current electricity-carbon correlation coefficient, the near-real-time estimated value of terminal fuel carbon emissions in the target area is obtained as the monitoring result.

[0007] A second aspect of the present invention provides a carbon emission monitoring system based on fusion decomposition and regression, comprising: The acquisition module is configured to acquire real-time power load data for a target area; The decomposition and elimination module is configured to decompose the real-time power load signal into several IMF components and residual terms based on the empirical mode decomposition algorithm, identify the IMF components related to the air conditioning load, and obtain the remaining IMF components. The reconstruction module is configured to reconstruct the remaining IMF components and residual terms after removing the air conditioning load, and output the real-time remaining power load sequence after removing the air conditioning effect. The regression solution module is configured to obtain the electricity-carbon correlation coefficient of the previous period and the exogenous variable values ​​within the set historical period, and input them into the constructed autoregressive moving average regression model to calculate the current electricity-carbon correlation coefficient. The estimation module is configured to use the real-time remaining power load sequence and the current electricity-carbon correlation coefficient to estimate the near-real-time terminal fuel carbon emissions of the target area as the monitoring result.

[0008] A third aspect of the present invention provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, complete the steps in the above-described fusion decomposition and regression carbon emission monitoring method.

[0009] A fourth aspect of the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps in the above-described fusion decomposition and regression carbon emission monitoring method.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: The method of this invention exhibits stronger robustness in eliminating weakly correlated loads. By extracting mid-to-low frequency IMF components related to air conditioning load through EMD, it effectively reduces the interference of air conditioning operation on carbon emission estimation models during weather-sensitive periods, thereby improving the accuracy of the estimation. Furthermore, based on a dynamic modeling mechanism between electricity load and carbon emissions, this method is suitable for real-time carbon emission monitoring needs in regional energy systems. In addition, employing the ARMAX model enhances model interpretability while considering system lag characteristics, helping to identify the main driving factors of carbon emission changes and providing a scientific basis for scheduling optimization and intervention strategy formulation.

[0011] The advantages of the present invention, as well as its additional advantages, will be described in detail in the following specific embodiments. Attached Figure Description

[0012] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute a limitation thereof.

[0013] Figure 1 This is a schematic flowchart of the carbon emission monitoring method based on fusion decomposition and regression according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the terminal air conditioning load breakdown process based on the EMD-LSTM algorithm in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the principle of small sample data augmentation for carbon emission estimation based on the BHT algorithm in Embodiment 1 of the present invention; Figure 4 This is a comparison chart of the estimation errors of terminal fuel carbon emissions before and after removing the air conditioning load in the simulation example of Embodiment 1 of the present invention; Figure 5 This is a simulation example of the present invention, showing the estimation error of terminal fuel carbon emissions under different air conditioning ratios; Figure 6 This is a simulation example of the hourly terminal fuel carbon emission estimation results for a provincial region in Embodiment 1 of the present invention. Detailed Implementation

[0014] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0015] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0016] It should be noted that the terminology used herein is for describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. It should be noted that, without conflict, the various embodiments and features within those embodiments can be combined with each other. The embodiments will now be described in detail with reference to the accompanying drawings.

[0017] Example 1 In one or more of the technical solutions disclosed in the embodiments, such as Figures 1 to 6 As shown, a carbon emission monitoring method based on fusion decomposition and regression includes the following steps: Step 1: Obtain real-time power load data for the target area; Step 2: Based on the empirical mode decomposition algorithm, the real-time power load signal is decomposed into several IMF components and residual terms, and the IMF components related to the air conditioning load are identified to obtain the remaining IMF components. Step 3: Reconstruct the remaining IMF components and residual terms after removing the air conditioning load, and output the real-time remaining power load sequence after removing the air conditioning effect; Step 4: Obtain the electricity-carbon correlation coefficient of the previous period and the exogenous variable values ​​within the set historical period, input them into the constructed autoregressive moving average regression model to calculate the current electricity-carbon correlation coefficient; Step 5: Using the real-time remaining power load sequence and the current electricity-carbon correlation coefficient, estimate the near-real-time terminal fuel carbon emissions of the target area as the monitoring result.

[0018] In this embodiment, electricity load data of the target area is first collected to obtain a dynamic change signal reflecting the region's energy use. Then, Empirical Mode Decomposition (EMD) is used to decompose the load signal into several Intrinsic Mode Function (IMF) components and a residual term. The EMD method has good adaptive characteristics and can effectively separate components with different frequency characteristics in the signal. By analyzing the frequency characteristics of the IMF components and their correlation with meteorological variables (such as temperature), IMF components related to the load of climate-sensitive equipment such as air conditioners are identified and removed, thereby eliminating these interference factors that are weakly correlated with end-use fuel carbon emissions. The remaining IMF components and residual term are used to reconstruct the remaining electricity load sequence, representing load changes highly correlated with fuel carbon emissions. This reconstructed load sequence is input into a pre-established Autoregressive Moving Average (ARMAX) model. The model outputs a dynamic correlation coefficient between electricity and carbon emissions, and combined with load changes, calculates the estimated carbon emissions for the target area, thus achieving near real-time dynamic carbon emission estimation.

[0019] This method exhibits stronger robustness in eliminating weakly correlated loads. By extracting mid-to-low frequency IMF components related to air conditioning load through EMD, it effectively reduces the interference of air conditioning operation on carbon emission estimation models during weather-sensitive periods, thereby improving the accuracy of the estimation. Furthermore, based on a dynamic modeling mechanism between electricity load and carbon emissions, this method is suitable for real-time carbon emission monitoring needs in regional energy systems. In addition, employing the ARMAX model enhances model interpretability while considering system lag characteristics, helping to identify the main drivers of carbon emission changes and providing a scientific basis for scheduling optimization and intervention strategy formulation.

[0020] In step 1, the real-time power load data obtained can be hourly data, such as setting the time interval for obtaining data to obtain power load data for one hour or several hours; In step 2, the original total power load signal is decomposed into several intrinsic mode function components (IMF components) and residual terms based on the empirical mode decomposition (EMD) algorithm, and the IMF components related to the air conditioning load are identified, including the following steps: Step 21: Based on the Empirical Mode Decomposition (EMD) algorithm, the original total power load signal is decomposed into several intrinsic mode function components (IMF components) and residual terms. The decomposition process is as follows: (1) Perform EMD decomposition on the original signal. The decomposition formula is as follows: (1); In the above formula, This is the original total power load signal; It is the i-th order intrinsic mode function, i.e., the IMF component, generated by the EMD algorithm; n is the total decomposition order. The remaining items are...

[0021] (2) Extracting IMF components : (2); In the above formula, Represents the i-th order input signal. and These are the upper and lower envelopes obtained by fitting the i-th order input signal, respectively. (3) Calculate the remainder of the input signal for the next order. or i-th order remainder It can be represented as: (3); Step 22: Take the IMF target frequency band corresponding to the air conditioning load and the IMF component related to the temperature variable as the IMF component related to the air conditioning load; Since both electricity load and air conditioning usage exhibit stable time-periodic characteristics, the Empirical Mode Decomposition (EMD) algorithm can be used to separate the air conditioning load from the total electricity consumption. In practical applications, the EMD algorithm can decompose the electricity load sequence into multiple Intrinsic Mode Functions (IMFs) and a residual term. These IMF components, by adaptively extracting the intrinsic vibrational components of the original signal, can characterize the fluctuation characteristics from high-frequency to low-frequency scales.

[0022] The periodic components closely related to air conditioning load are typically concentrated in the low-to-mid-frequency IMF components, and these characteristics can serve as a basis for identifying air conditioning load components. Air conditioning load exhibits significant periodic characteristics in electricity consumption, with the following frequency range: (1) The center frequency of the daily cycle is 0.0417 cph, with a tolerance of ±20%, approximately 0.033-0.050 cph.

[0023] (2) The center frequency of the periodic cycle is 0.00595cph, with a tolerance of ±10%, approximately 0.0054-0.0066cph.

[0024] (3) The central frequency of the monthly cycle is approximately 0.00137 cph.

[0025] When the time series spans several months or years, monthly or seasonal components can also be identified. Based on this, the IMF components located in the aforementioned frequency bands and significantly correlated with temperature variables are merged into the IMF set of the air conditioning load.

[0026] The above implementation method, by jointly utilizing the frequency information of IMF components and their correlation with temperature variables, can more accurately identify the electrical load components related to the operation of air conditioning equipment, effectively eliminating non-fuel load factors that introduce noise into carbon emission modeling. This method has a stronger logical basis and adaptability than traditional methods based on mean or rate of change, and is particularly suitable for scenarios with significant changes in air conditioning load in multi-climate zones or urban systems, thus helping to improve the model's generalization ability and applicability.

[0027] Based on the above, step 3 uses a Long Short-Term Memory (LSTM) network to reconstruct the relevant IMF components of the remaining power load. Based on the residual terms, the remaining power load sequence after removing the air conditioning load is reconstructed. Step 3 is the opposite of step 2. Step 2 is to divide the load into multiple modal components, while step 3 is to integrate the remaining components after removing the air conditioning load into a power load sequence. Since EMD only has signal decomposition capabilities and cannot independently identify or predict air conditioning load components, an LSTM network is introduced to model and learn each IMF component. This LSTM neural network is then used to reconstruct the non-air conditioning load signal. Based on a recurrent neural network framework, the algorithm introduces a special gating mechanism, including a forget gate, input gate, and output gate, making it more suitable for processing power time-series signals. The EMD-LSTM algorithm structure is as follows: Figure 2 As shown.

[0028] Furthermore, the number of layers in the LSTM network is consistent with the number of IMF components. To optimize model parameters and structure, the number of LSTM network layers is dynamically adjusted during training based on the EMD processing results, ensuring that the number of LSTM network layers is equal to the number of IMFs, so that each LSTM layer can capture feature information at a specific time scale. Other network parameters are set based on experience.

[0029] Introducing an LSTM model for reconstruction not only improves the accuracy of modeling electricity load change trends but also enhances the model's adaptability to nonlinear relationships and complex fluctuation patterns. Compared to traditional linear modeling methods, LSTM can better handle multi-frequency and multi-scale load components, thus constructing a more accurate and smoother residual load curve. Furthermore, this method reduces information loss due to the simplicity of the model structure, improves the ability to reconstruct the driving factors of carbon emission changes, and significantly improves the accuracy and stability of carbon emission estimation results.

[0030] Alternative implementations, besides LSTM neural networks, can use other types of sequence learning models for reconstruction, such as gated recurrent units (GRUs), bidirectional LSTMs (Bi-LSTMs), or attention-based Transformer models, to improve the model's adaptability to different types of workload data.

[0031] The purpose of the autoregressive moving average regression modeling of electricity-carbon emissions in step 4 is to construct a time series forecasting model driven by exogenous variables, quantify the dynamic relationship between electricity load and end-use fuel carbon emissions, and obtain statistically significant electricity-carbon correlation coefficients, providing theoretical support and parameter basis for real-time carbon emission estimation.

[0032] In step 4, the electricity-carbon correlation coefficient is set as an endogenous variable, while carbon emission growth rate, carbon emission spatial density, electricity-carbon elasticity coefficient, and per capita carbon emissions are used as exogenous variables. The Yule-Walker method and least squares method are used to estimate the model parameters, and an autoregressive moving average regression model is constructed, including the following steps: Step 41: Establish the basic ARIMAX model: Calculate the exogenous variables including carbon emission growth rate, carbon emission spatial density, electric carbon elasticity coefficient, and per capita carbon emissions, and weight them with the autoregressive term and the moving average term to construct a statistical regression model that dynamically models the change of the electric-carbon correlation coefficient over time, which serves as the basic ARIMAX model. The basic ARIMAX model transforms a non-stationary time series into a stationary one using differencing, typically employing first-order differencing in practice. Subsequently, by integrating the autoregressive structure, moving average correction, and the effects of exogenous variables, the complete basic ARIMAX model can be expressed as follows: (4); In the above formula, This represents the electricity-carbon correlation coefficient during time period t. This indicates the change in the current electricity-carbon correlation coefficient compared to the previous electricity-carbon correlation coefficient; Represents a constant term; and These represent the autoregression coefficient and the order of the moving average, respectively. Represents the i-th order autoregressive coefficient; The coefficients of the j-th order moving average are represented; the exogenous variables are defined as follows: This represents the carbon emission growth rate during period t. This represents the spatial density of carbon emissions during time period t. This represents the electro-carbon elastic coefficient during time period t. This represents the average carbon emissions per person during period t. Represents the regression coefficient of exogenous variables; This represents the prediction error, which follows a normal distribution.

[0033] In the above basic ARIMAX model: The autoregressive term (AR) is It is used to characterize the current change in the electricity-carbon correlation coefficient compared to past changes. Relationship; The moving average (MA) is This is used to characterize the hysteresis correlation between current errors and historical errors; The exogenous variable term (X) is , representing the degree of influence of external factors on the current change, consists of four items, each corresponding to an exogenous variable; In the above implementation, by introducing exogenous variables such as carbon emission growth rate, carbon emission spatial density, electricity-carbon elasticity coefficient and per capita carbon emissions, an autoregressive moving average model (ARIMAX) with exogenous variables is constructed. This model comprehensively considers the historical trend of the electricity-carbon correlation coefficient, the autoregressive structure, the influence of the error term, and external driving factors, thereby achieving the modeling of the dynamic changes of the electricity-carbon correlation coefficient over time.

[0034] Step 42: Establish an ARIMAX model based on MDT and BHT: Extract core variables using the multidimensional tensor decomposition method. By combining the current core tensor change with the core tensor of the historical continuous period and the historical error term, and introducing the exogenous variable of carbon emissions, the basic ARIMAX model is updated to an ARIMAX model with the core tensor as the regression object, which is the final autoregressive moving average regression model. The core variables are extracted using a multidimensional tensor decomposition method, which is the same as the method in the subsequent step S4. The Tucker decomposition method is used to perform tensor decomposition under orthogonal constraints, resulting in the product of the core tensor and several orthogonal factor matrices.

[0035] To improve model prediction accuracy and reduce computational complexity, this embodiment proposes using the core tensor directly instead of the full high-dimensional tensor for model training. This is achieved by using the current core tensor... By combining the core tensor of historical continuous p periods and historical error terms, the constructed ARIMAX model describes the correlation between historical electricity data and end-use fuel carbon emission data. The specific expression is as follows: (5); in, The tensor form of the current prediction error should be optimized to a minimum. The symbols in formula (5) are the same as in formula (4), with the addition of a sharp angle to indicate the tensor form. The optimization objective function for the above ARIMAX model training and construction process is: (6); In the above formula, This is the total time period; To predict the start time period; Let be the tensor form of the electro-carbon correlation coefficient for time period t; and These represent the autoregression coefficient and the order of the moving average, respectively. express i Autoregressive coefficients; Represents the regression coefficient of exogenous variables; express j Moving average coefficient; Represents the tensor form of the k-th exogenous variable; The tensor representing the prediction error follows a normal distribution; Denotes the m-th order factor matrix. m =1,…, M ; superscript m Indicates existence m A different matrix, index m Indicates the index of each column vector in the matrix; This indicates the search for the L2 norm.

[0036] During the training of the ARIMAX model, the values ​​of the following coefficients are determined, including: autoregressive coefficients. Regression coefficients of exogenous variables The moving average coefficient of the j-th order ; After training, the ARIMAX model uses the obtained electricity-carbon correlation coefficient from the previous period and the values ​​of exogenous variables in the set historical period, including carbon emission growth rate, carbon emission spatial density, electricity-carbon elasticity coefficient, and per capita carbon emissions, to calculate the current electricity-carbon correlation coefficient through the trained autoregressive moving average regression model. Since user energy consumption behavior and energy structure are not easily changed over a period of time, the electricity-carbon correlation coefficient is calculated using the ARIMAX model with the coefficients obtained in step 4. Based on the electricity-carbon correlation coefficient and the real-time remaining electricity load sequence, the carbon emission values ​​for future periods are estimated and predicted. Furthermore, it also includes the training process of the autoregressive moving average regression model (ARIMAX), which includes the following steps: Step S1: Obtain historical data. At the set time granularity, obtain historical power load data, terminal fuel carbon emission statistics, environmental data, and exogenous variable data. Data Acquisition: Collect the following data within a preset time granularity in the target area: (1) Monthly statistics of terminal fuel carbon emissions in the region; (2) Total monthly social electricity consumption and air conditioning electricity consumption in the region; (3) Daily temperature variation curve of the region; (4) Data on four types of exogenous variables: regional carbon emission growth rate, carbon emission spatial density, electricity-carbon elasticity coefficient, and per capita carbon emissions.

[0037] The carbon emission growth rate refers to the amount of carbon emissions per unit of time; it reflects the impact of time factors on carbon emissions, including adjustments to carbon emission policies, the stage of industrial development, and energy price cycles. By introducing this exogenous variable, the trend deviation in the electricity-carbon relationship can be corrected.

[0038] Carbon emission spatial density refers to the amount of carbon emissions per unit area. Because carbon emissions exhibit spatial clustering with industrial layout, population distribution, and transportation modes, using only electricity and carbon emission data as endogenous variables cannot reflect the spatial differences in carbon emissions. Introducing carbon emission spatial density can reduce errors caused by spatial autocorrelation.

[0039] The carbon emission elasticity coefficient is the ratio of the carbon emission growth rate to the electricity consumption growth rate. Due to the non-linear relationship between electricity and carbon emissions, the corresponding carbon emission growth rate will vary depending on seasonal or regional differences for the same electricity consumption growth rate. Introducing this exogenous variable allows for correlation analysis between changes in electricity consumption and fluctuations in carbon emissions.

[0040] Per capita carbon emissions, calculated by dividing carbon emissions per unit time by the total population, are closely related to a region's level of industrialization, urbanization, and residents' consumption habits. Introducing this exogenous variable not only reflects user consumption behavior but also reflects the state of economic and social development. In summary, these four exogenous variables provide supplementary information beyond the historical data itself from multiple key dimensions, significantly enhancing the model's ability to explain the dynamic changes of the dependent variable.

[0041] Step S2: Based on the Empirical Mode Decomposition (EMD) algorithm, the historical power load signal is decomposed into several intrinsic mode functions (IMFs) and residual terms, and combined with daily temperature change data, the IMF components containing air conditioning load characteristics are extracted; this step is the same as step 2 and will not be described again. Step S3: Model and reconstruct the remaining IMF components after removing air conditioning load based on a Long Short-Term Memory (LSTM) network, and output the historical remaining power load sequence after removing the influence of air conditioning; this step is the same as step 3. Step S4: Divide the terminal fuel carbon emissions by the remaining electricity load sequence to obtain historical electricity-carbon correlation coefficient data. Then, fuse the multi-way delay embedding transform (MDT) and Tucker tensor decomposition to perform sample augmentation on the historical electricity-carbon correlation coefficient data. Electricity consumption data is typically acquired in high-frequency hourly data formats. However, historical terminal fuel carbon emission data used for estimation is low-frequency data, and the available samples are time series with a limited number of data points. To improve the quality of carbon emission data and thus enhance the efficiency and accuracy of estimation, this embodiment proposes a small-sample data augmentation method based on the concept of Block Hankel Tensor (BHT). This method integrates Multi-way Delay Embedding Transform (MDT) and Tucker tensor decomposition to augment the reconstructed residual electricity load sequence, including the following steps: Step S41: Divide the terminal fuel carbon emissions by the remaining electricity load sequence to obtain historical electricity-carbon correlation coefficient data; perform feature space reconstruction based on MDT to transform the historical electricity-carbon correlation coefficient data into a high-dimensional, low-rank BHT matrix; Among them, the historical electricity-carbon correlation coefficient data is a one-dimensional time series; This step utilizes Multidimensional Data Transformation (MDT) to structurally reorganize existing data points, transforming a one-dimensional time series into a high-dimensional, low-rank BHT matrix. This transformation optimizes the data structure and effectively alleviates challenges such as insufficient sample size, low dimensionality, and the difficulty in capturing the complex dynamic features of carbon emission sequences.

[0042] Step S411: Convert the time series of historical electricity-carbon correlation coefficient data into a Hankel matrix. Construct the Hankel matrix using a sliding window to expand the one-dimensional data into a two-dimensional local structure block. The conversion process is as follows: (7); (8); In the above formula, This represents the original one-dimensional time series. ~ This represents the historical electricity-carbon correlation coefficient data, calculated by dividing the final fuel carbon emissions by the remaining electricity load. This represents the Hankel matrix obtained by the sliding window method. Indicates the length of the sliding window.

[0043] In the above implementation, each sliding window generates a new subsequence, resulting in multiple deformed samples; Step S412: Convert the Hankel matrix to a BHT matrix, using the following formula: (9); In the above formula, This represents an operator that performs delayed embedding and sliding window Hankelization on a block-by-block basis; Tensor form representing the Hankel matrix; Indicates the first m The length of each sliding position dimension m =1,…, M ; This represents the total number of elements in each block.

[0044] The core of this step S412 is to transform the time series into a high-dimensional structure, thereby enhancing the data's feature representation capabilities; Step S42: For the obtained BHT matrix, the Tucker decomposition method is used to perform tensor decomposition under orthogonal constraints to obtain the product form of the core tensor and several orthogonal factor matrices. New combination features are obtained through tensor decomposition to obtain enhanced historical electricity-carbon correlation coefficient data samples. To ensure the uniqueness and stability of the decomposition process, the Tucker decomposition method with orthogonal constraints is used to process tensors. This method decomposes a large original tensor into a product of a smaller core tensor and several orthogonal factor matrices. The orthogonal constraints are used to ensure that the decomposition factors are independent of each other, as described below: (10); (11); Wherein, equation (10) represents tensor pattern product; the superscript M indicates that there are M different matrices, and the subscript M indicates the index of each column vector in the matrix; Represents the low-rank tensor in the BHT matrix; This represents the core tensor, which is much smaller than the original tensor. Let represent the factor matrix. Equation (11) expresses the orthogonality constraint, requiring that all factor matrices must satisfy column orthogonality.

[0045] Furthermore, by using these factor matrices and core tensors, interpolation or sampling is used to generate realistic new data blocks, thereby further enhancing the sample data; In the above process, the original small sample data is transformed into a larger-scale, more structured new sample set through Hankel embedding, high-dimensional tensor construction, and orthogonal tensor decomposition, thereby achieving small-sample augmentation and effectively improving the stability and generalization ability of model training. The reconstructed samples maintain the dynamic characteristics and temporal dependency structure of the original data, while introducing orthogonal constraints to suppress redundant features, ensuring the diversity and rationality of the newly added samples, and significantly enhancing feature learning performance under limited data conditions. By controlling the dimension of the core tensor and the orthogonality of the factor matrix, the degree of feature abstraction and data augmentation intensity can be flexibly adjusted to adapt to the modeling needs of different scenarios. The final augmented samples retain the key patterns of the original temporal series while incorporating the structural diversity brought by tensor decomposition, significantly outperforming traditional interpolation or resampling methods.

[0046] Step S5: Based on the core tensor and exogenous variables, the enhanced sample data is obtained. The autoregressive moving average regression model (ARIMAX) with exogenous variables is trained based on the core tensor and exogenous variables. During the training process, the Yule-Walker method and the least squares method are used to jointly estimate the model parameters to obtain the trained autoregressive moving average regression model (ARIMAX). The optimization objective function for the above ARIMAX model training and construction process is Equation (6), as follows: ; To verify the effectiveness of the proposed EMD-LSTM-BHT-ARIMAX algorithm in estimating regional end-of-life fuel carbon emissions, this embodiment employs multiple algorithms for comparative analysis. The specific process is as follows: First, regional end-of-life fuel carbon emissions for April 2023 are estimated using data from January to March 2023, and then compared with the actual data for that month. Finally, carbon emissions from May to October 2023 are estimated month by month and compared with the actual data.

[0047] This embodiment uses the following four error metrics to reflect the accuracy of the proposed method: root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and minimum-maximum error (MME).

[0048] Table 1 shows the performance of the proposed method compared with three other methods in estimating regional terminal fuel carbon emissions. Specifically: the ARIMAX method does not consider air conditioning load removal and small sample data enhancement; the BHT-ARIMAX method considers small sample data enhancement but not air conditioning load removal; and the EMD-LSTM-ARIMAX method considers air conditioning load removal but not small sample data enhancement.

[0049] Table 1. Comparison of carbon emission estimation results using four types of algorithms;

[0050] As shown in Table 1, by removing the air conditioning load and using the remaining electricity load data for regression prediction, more accurate estimates of end-use fuel carbon emissions can be obtained. Compared with the ARIMAX algorithm, the EMD-LSTM-ARIMAX algorithm reduced the four types of errors (MAPE, MAE, RMSE, and MME) by 58.3%, 57.3%, 60.6%, and 57.1%, respectively. Similarly, compared with the BHT-ARIMAX algorithm, the EMD-LSTM-BHT-ARIMAX algorithm reduced the four types of error indicators by 60.7%, 61.5%, 56.0%, and 62.5%, respectively.

[0051] The results in Table 1 also show that data augmentation using the BHT algorithm significantly improves estimation accuracy. Compared to the ARIMAX algorithm, the four error metrics of the BHT-ARIMAX algorithm decreased by 46.2%, 45.3%, 48.6%, and 42.9%, respectively. Similarly, compared to the EMD-LSTM-ARIMAX algorithm, the four error metrics of the EMD-LSTM-BHT-ARIMAX algorithm decreased by 49.4%, 50.7%, 42.6%, and 50%, respectively.

[0052] In summary, when air conditioning loads, which are weakly correlated with carbon emissions, are removed and estimation is performed using residual electricity load data, the estimation error is reduced by approximately 60% compared to estimation using total electricity load data. Furthermore, when data augmentation methods are applied, the estimation error is reduced by 45% to 50% compared to estimation using the unaugmented raw data.

[0053] To further analyze the impact of air conditioning load on the estimation results of regional terminal fuel carbon emissions, this embodiment compares two estimation scenarios before and after excluding air conditioning load. The proposed algorithm was used to estimate regional terminal fuel carbon emissions from April 2023 to October 2023. The comparison results of the MAPE values ​​under the two scenarios are as follows: Figure 4 As shown.

[0054] according to Figure 4 It is evident that the MAPE value estimated based on carbon emissions excluding air conditioning load is relatively low, typically fluctuating between 2.7% and 3.4%. However, if the entire electricity load is used for estimation, the MAPE value increases significantly and fluctuates considerably, ranging from 6.6% to 10%. Comparative analysis shows that estimating regional end-use fuel carbon emissions based on electricity consumption excluding air conditioning load not only effectively reduces errors but also achieves more stable measurement accuracy.

[0055] Figure 4The study also showed a correlation between the estimation error of end-use fuel carbon emissions and the proportion of air conditioning load when air conditioning load is included. Generally, the higher the proportion of air conditioning load, the larger the estimation error. For example, during June to August, the proportion of air conditioning load was 15% to 21%, and the estimation error was approximately 8.5% to 10%. Conversely, when the proportion of air conditioning load is low, the estimation error is relatively small. For example, in April and October, the proportion of air conditioning load was approximately 7%, and the estimation error was 6.5% to 7%.

[0056] To further analyze the impact of air conditioning load electricity ratio on the calculation error, this embodiment plots the terminal fuel carbon emission estimation error diagram under different air conditioning load ratios, as shown below. Figure 5 As shown.

[0057] Depend on Figure 5 It can be seen that after the air conditioning load is disassembled, the calculation error of the BHT-ARIMAX method is no longer related to the proportion of air conditioning load power consumption. However, when the air conditioning load is not disassembled, the calculation error of the BHT-ARIMAX method is roughly positively correlated with the proportion of air conditioning load power consumption.

[0058] In addition, by Figure 5 It can also be seen that, even when the proportion of air conditioning is relatively low, there is a significant difference in the calculation error before and after considering the dismantling of the air conditioning load. For example, when the proportion of air conditioning load is about 6%, the MAPE before considering the dismantling of the air conditioning load is still about 3.5% higher than the MAPE after the dismantling of the air conditioning load. This is mainly because the inclusion of air conditioning load electricity in historical data makes it difficult for the algorithm to accurately establish the correlation between terminal fuel carbon emissions and electricity consumption, thus leading to a significant calculation error.

[0059] By using the correlation coefficient of electricity carbon emissions obtained by the method proposed in this embodiment, and combining it with real-time electricity load data, the hourly terminal fuel carbon emissions in this region can be estimated. Figure 6 The paper presents the estimated carbon emissions from end-fuel fuels in the province from January to July 2023, obtained using the method proposed in this embodiment.

[0060] like Figure 6 As shown, the time-series characteristics of end-use fuel carbon emissions from January to July 2023 exhibit the following pattern: carbon emissions from end-use fuel consumption in late January to early February were significantly lower than in other months, approximately 50% of those in other periods. This phenomenon is mainly because this period coincides with the Spring Festival holiday, during which a large number of industrial production facilities suspended operations, and industrial users are one of the main sources of carbon emissions from end-use fuel consumption.

[0061] Depend on Figure 6It is evident that end-use fuel consumption at different times leads to significant differences in regional carbon emissions. The highest hourly carbon emissions in this region can reach 91,000 tons, while the lowest are 37,000 tons. Furthermore, the fluctuation range of carbon emissions exhibits a clear monthly variation, with smaller fluctuations during holidays and significantly larger fluctuations on weekdays. These findings reveal the levels and fluctuation patterns of carbon emissions, enabling more targeted carbon emission control measures, such as regulating equipment operation in target areas and implementing environmental improvement plans.

[0062] Example 2 Based on Example 1, this example provides a carbon emission monitoring system based on fusion decomposition and regression, comprising: The acquisition module is configured to acquire real-time power load data for a target area; The decomposition and elimination module is configured to decompose the real-time power load signal into several IMF components and residual terms based on the empirical mode decomposition algorithm, identify the IMF components related to the air conditioning load, and obtain the remaining IMF components. The reconstruction module is configured to reconstruct the remaining IMF components and residual terms after removing the air conditioning load, and output the real-time remaining power load sequence after removing the air conditioning effect. The regression solution module is configured to obtain the electricity-carbon correlation coefficient of the previous period and the exogenous variable values ​​within the set historical period, and input them into the constructed autoregressive moving average regression model to calculate the current electricity-carbon correlation coefficient. The estimation module is configured to use the real-time remaining power load sequence and the current electricity-carbon correlation coefficient to estimate the near-real-time terminal fuel carbon emissions of the target area as the monitoring result.

[0063] The target frequency band of the IMF corresponding to the air conditioning load and the IMF component related to the temperature variable are taken as the IMF component related to the air conditioning load.

[0064] Furthermore, the reconstruction module uses a long short-term memory network algorithm to reconstruct the relevant IMF components of the remaining power load, thereby reconstructing the real-time remaining power load sequence after removing the air conditioning load.

[0065] Furthermore, the regression solution module is configured to set the electricity-carbon correlation coefficient as an endogenous variable, and the carbon emission growth rate, carbon emission spatial density, electricity-carbon elasticity coefficient, and per capita carbon emissions as exogenous variables. The model parameters are estimated using the Yule-Walker method and the least squares method to construct an autoregressive moving average regression model.

[0066] Furthermore, the autoregressive moving average regression model is constructed, including the following steps: We calculate exogenous variables including carbon emission growth rate, carbon emission spatial density, electricity-carbon elasticity coefficient, and per capita carbon emissions, and weight them with autoregressive and moving average terms to construct a statistical regression model that dynamically models the electricity-carbon correlation coefficient over time, which serves as the basic ARIMAX model. The core variables are extracted using the multidimensional tensor decomposition method. By combining the current change in the core tensor with the core tensor of the historical continuous period and the historical error term, and introducing the exogenous variable of carbon emissions, the basic ARIMAX model is updated to an ARIMAX model with the core tensor as the regression object, which is the final autoregressive moving average regression model.

[0067] Furthermore, it also includes a training module configured to train the constructed autoregressive moving average regression model, comprising the following steps: Acquire historical data, including historical power load data, terminal fuel carbon emission statistics, environmental data, and exogenous variable data, within a set time granularity. The historical power load signal is decomposed into several intrinsic mode functions and residual terms, and combined with daily temperature change data, the IMF component containing air conditioning load characteristics is extracted. Based on the Long Short-Term Memory Network, the remaining IMF component after removing the air conditioning load is modeled and reconstructed, and the historical remaining power load sequence after removing the air conditioning effect is output. By dividing the terminal fuel carbon emissions by the historical remaining electricity load sequence, historical electricity-carbon correlation coefficient data is obtained. Multidimensional delayed embedding transformation and Tucker tensor decomposition are then used to enhance the historical electricity-carbon correlation coefficient data. Based on the core tensor and exogenous variables, and using the enhanced sample data, the autoregressive moving average regression model with exogenous variables is trained. During the training process, the Yule-Walker method and the least squares method are used to jointly estimate the model parameters, and the trained autoregressive moving average regression model is obtained.

[0068] Furthermore, sample augmentation is performed on the historical electricity-carbon correlation coefficient data, including the following steps: The historical electricity-carbon correlation coefficient data is obtained by dividing the terminal fuel carbon emissions by the remaining electricity load sequence; the feature space is reconstructed based on MDT, and the historical electricity-carbon correlation coefficient data is transformed into a high-dimensional, low-rank BHT matrix. For the obtained BHT matrix, the Tucker decomposition method is used to perform tensor decomposition under orthogonal constraints, resulting in the product form of the core tensor and several orthogonal factor matrices. New combination features are obtained through tensor decomposition, resulting in enhanced historical electricity-carbon correlation coefficient data samples.

[0069] It should be noted that each module in this embodiment corresponds one-to-one with each step in embodiment 1, and their specific implementation process is the same, so it will not be repeated here.

[0070] Example 3 Based on Embodiment 1, this embodiment provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps in the carbon emission monitoring method of fusion decomposition and regression described in Embodiment 1.

[0071] Example 4 Based on Embodiment 1, this embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they complete the steps in the carbon emission monitoring method of fusion decomposition and regression described in Embodiment 1.

[0072] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0073] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method of carbon emission monitoring that fuses disaggregated regression, characterized in that, The method comprises the following steps: For the target area, real-time power load data is obtained; Based on the empirical mode decomposition algorithm, the real-time power load signal is decomposed into several IMF components and residual terms, the IMF components related to air conditioner load are identified, and the remaining IMF components are obtained; The remaining IMF components after excluding the air conditioner load and the residual terms are reconstructed, and the real-time remaining power load sequence after excluding the influence of the air conditioner is output; The electric-carbon correlation coefficient of the previous period and the exogenous variable value in the set historical period are obtained and input into the constructed autoregressive moving average regression model to calculate the current electric-carbon correlation coefficient; The near-real-time terminal fuel carbon emission estimation value of the target area is estimated as the monitoring result.

2. The method of claim 1, wherein: The IMF target frequency band corresponding to the air conditioner load and the IMF component related to the temperature variable are used as the IMF component related to the air conditioner load.

3. The method of claim 1, wherein: The long short-term memory network algorithm is used to reconstruct the relevant IMF components of the remaining power load, and the real-time remaining power load sequence after excluding the air conditioner load is reconstructed.

4. The carbon emission monitoring method of claim 1, wherein: The electric-carbon correlation coefficient is set as an endogenous variable, the carbon emission growth rate, the carbon emission spatial density, the electric-carbon elasticity coefficient, and the carbon emission per capita are set as exogenous variables, the Yule-Walker and least squares methods are used to estimate the model parameters, and an autoregressive moving average regression model is constructed.

5. The carbon emission monitoring method of claim 4, wherein: The autoregressive moving average regression model is constructed, comprising the following steps: The exogenous variable terms including the carbon emission growth rate, the carbon emission spatial density, the electric-carbon elasticity coefficient, and the carbon emission per capita are calculated, weighted with the autoregressive terms and the moving average terms, a dynamic statistical regression model of the electric-carbon correlation coefficient changing over time is constructed as a basic ARIMAX model; The multi-dimensional tensor decomposition method is used to extract the core variables, the current core tensor change amount is combined with the historical continuous period core tensor and the historical error term, and the carbon emission exogenous variable is introduced, the basic ARIMAX model is updated to an ARIMAX model with the core tensor as the regression object, and the autoregressive moving average regression model is finally constructed.

6. The carbon emission monitoring method of claim 1, wherein: The training process of the constructed autoregressive moving average regression model comprises the following steps: Historical data is obtained, historical power load data, terminal fuel carbon emission statistical values, environmental data, and exogenous variable data are obtained at a set time granularity; The historical power load signal is decomposed into several intrinsic mode functions and residual terms, and the IMF components containing air conditioner load characteristics are extracted in combination with the daily temperature change data; The remaining IMF components after excluding the air conditioner load are modeled and reconstructed based on the long short-term memory network, and the historical remaining power load sequence after excluding the influence of the air conditioner is output. The terminal fuel carbon emission is divided by the historical residual power load sequence to obtain historical electricity-carbon correlation coefficient data, and a multi-dimensional delay embedding transformation and Tucker tensor decomposition are fused to perform sample enhancement on the historical electricity-carbon correlation coefficient data; Based on the enhanced sample data, an autoregressive moving average regression model with exogenous variables constructed based on the core tensor and the exogenous variables is trained, and the model parameters are estimated by the Yule-Walker method and the least squares method during the training process to obtain the trained autoregressive moving average regression model.

7. The carbon emission monitoring method of claim 6, wherein: The sample enhancement on the historical electricity-carbon correlation coefficient data comprises the following steps: The terminal fuel carbon emission is divided by the residual power load sequence to obtain historical electricity-carbon correlation coefficient data, and a multi-dimensional delay embedding transformation and Tucker tensor decomposition are fused to perform sample enhancement on the historical electricity-carbon correlation coefficient data; For the obtained BHT matrix, a Tucker decomposition method is used to perform tensor decomposition under orthogonal constraint to obtain the product form of the core tensor and a plurality of orthogonal factor matrices, new combined features are obtained through tensor decomposition, and enhanced historical electricity-carbon correlation coefficient data samples are obtained.

8. A carbon emission monitoring system fusing disaggregated regression, characterized in that, Comprise: The acquisition module is configured to acquire real-time power load data for a target area; The decomposition and removal module is configured to decompose the real-time power load signal into a plurality of IMF components and a residual term based on an empirical mode decomposition algorithm, identify an IMF component related to air conditioning load, and obtain residual IMF components; The reconstruction module is configured to reconstruct the residual IMF components after removing the air conditioning load and the residual term, and output a real-time residual power load sequence after removing the influence of air conditioning; The regression solving module is configured to acquire an electricity-carbon correlation coefficient of the previous period and exogenous variable values in a set historical period, and input them into the constructed autoregressive moving average regression model for calculation to obtain a current electricity-carbon correlation coefficient; The estimation module is configured to estimate a near-real-time terminal fuel carbon emission estimation value of the target area as a monitoring result using the real-time residual power load sequence and the current electricity-carbon correlation coefficient.

9. An electronic device, comprising: A computer program product comprising a memory and a processor, and computer instructions stored on the memory and running on the processor, when the computer instructions are run by the processor, the steps in the carbon emission monitoring method of claim 1-7 are completed.

10. A computer-readable storage medium, characterized in that, A computer program product for storing computer instructions, when the computer instructions are executed by a processor, the steps in the carbon emission monitoring method of claim 1-7 are completed.

Images