Carbon emission data simulation development prediction system and method thereof
By integrating multi-source data fusion and carbon flow structure modeling, combined with a multi-stage prediction engine and policy scenario simulation, the problem of multi-source data processing and dynamic response in existing carbon emission prediction technologies has been solved, achieving high-precision, interpretable multi-level carbon emission prediction and policy simulation.
Patent Information
- Application Number
- CN202511301465.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-12
- Publication Date
- 2026-01-23
AI Technical Summary
Existing carbon emission prediction technologies are insufficient in processing multi-source heterogeneous data, lack multi-level segmentation prediction capabilities, lack dynamic response mechanisms, have limited model robustness, and are difficult to adapt to complex data environments and policy changes.
It employs a multi-source data fusion module, a carbon flow structure modeling module, a multi-stage prediction engine, and a policy scenario simulation module, combined with graph neural networks, Bayesian trend models, recurrent neural networks, and compressed sensing optimization, to achieve unified data processing and real-time policy simulation.
It improves the robustness and adaptability of data processing, enhances the interpretability and accuracy of forecasts, can dynamically respond to policy changes, and provides multi-level segmented forecasts and visualization outputs.
Smart Images
Figure CN121389710A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of carbon emissions, in particular to a carbon emission data simulation development prediction system and method thereof. BACKGROUND
[0002] In the context of increasing global climate change, carbon emission prediction and simulation has become a key factor for countries to formulate energy strategies. Various research institutions and enterprises generally rely on carbon emission prediction models to assess the development path under different policy and economic scenarios, thereby supporting government decision-making and enterprise emission reduction management. However, carbon emission data is characterized by wide sources, diverse formats, and varying granularities, covering energy consumption, industrial output, transportation, building energy consumption, and other factors. How to achieve high-precision and interpretable prediction in a complex data environment has become a technical problem that needs to be solved.
[0003] Existing carbon emission prediction methods mainly include three categories: first, methods based on statistical regression models, such as multiple linear regression, time series ARIMA, etc., which can be used for short-term or medium-term prediction, but are difficult to reflect long-term dynamic factors such as policy changes, industrial adjustment, and energy structure transformation; second, methods based on machine learning, such as neural networks, support vector machines, random forests, etc., which can improve prediction accuracy, but lack interpretability and are highly dependent on data integrity, with significant performance degradation in the presence of missing values or noise; third, methods based on system dynamics or energy consumption parameter models, which can simulate the causal mechanism between macroeconomics and carbon emissions, but rely on artificial parameter setting, are highly subjective, and are difficult to automatically adapt to dynamic changes in real-world scenarios.
[0004] In summary, existing technologies generally have the following shortcomings: first, weak data processing capabilities, lack of efficient fusion mechanism for multi-source heterogeneous data; second, single prediction dimension, most models can only perform total quantity prediction, lacking the ability to subdivide industries, regions, and multiple levels; third, lack of dynamic response mechanism, unable to quickly adapt to changes in policy, industrial structure, and climate conditions; fourth, limited model robustness, with decreased stability and credibility of prediction results when faced with high noise or missing data. These shortcomings seriously restrict the application effect of existing carbon emission prediction technology in macro-policy planning, enterprise emission reduction management, and regional energy optimization. Therefore, a carbon emission data simulation development prediction system and method thereof are proposed. SUMMARY
[0005] In view of the shortcomings of the prior art, the present application provides a carbon emission data simulation development prediction system and method thereof to solve the technical problems existing in the prior art.
[0006] The above technical purposes of the present application are achieved through the following technical solutions:
[0007] A carbon emission data simulation development prediction system, comprising the following steps:
[0008] A multi-source data fusion module for receiving and uniformly processing carbon emission data from the fields of energy, industry, transportation, construction, etc., standardizing, filling in missing values and removing outliers of the data, and forming a unified data matrix;
[0009] A carbon flow structure modeling module for constructing a carbon flow atlas with industry, region and energy type as nodes, performing node embedding and path reasoning based on graph neural network to reveal the causal relationship of carbon emission and inter-industry dependence;
[0010] A multi-stage prediction engine including long-term trend prediction, short-term disturbance prediction and sparse back-propagation unit, wherein the long-term trend prediction adopts Bayesian trend model or Gaussian process modeling, the short-term disturbance prediction adopts recurrent neural network, and the sparse back-propagation adopts compressed sensing optimization;
[0011] A strategy scenario simulation module for receiving policy parameters such as carbon tax rate and energy consumption limit, adjusting the edge weight and emission intensity function of the carbon flow atlas in real time to generate predicted paths under different policy backgrounds;
[0012] A result output module for providing the prediction results in the form of trend curve, heat map and API interface to the outside.
[0013] Preferably, the multi-source data fusion module standardizes the original data d i,t by using the formula:
[0014]
[0015] Wherein, μ i is the field mean, σ i is the standard deviation, is the standardization result.
[0016] Preferably, the missing value filling adopts exponential moving average or K-nearest neighbor interpolation, and the exponential moving average satisfies:
[0017]
[0018] The K-nearest neighbor interpolation satisfies:
[0019]
[0020] Wherein, K is the number of neighbors, and u t is the context vector.
[0021] Preferably, the carbon flow structure modeling module updates the node embedding based on the graph neural network, satisfying:
[0022]
[0023] where A is the weighted adjacency matrix, I is the identity matrix, D is the degree matrix, W (k) is the k-th layer weight matrix, and σ is the activation function.
[0024] Preferably, the long-term trend prediction employs a Gaussian process model:
[0025]
[0026] where the kernel function is:
[0027]
[0028] where l is the length scale, ω is the period parameter, is the variance parameter.
[0029] Preferably, the short-term disturbance prediction employs a long short-term memory network whose gating mechanism satisfies:
[0030] i t = σ(W i x t + U i h t-1 + b i ), f t = σ(W f x t + U f h t-1 + b f ),
[0031] o t = σ(W o x t + U o h t-1 + b o ), c t
[0032] = f t ⊙ c t-1 + i t ⊙ tanh(W c x t + U c h t-1 + b c )
[0033] where x t is the input feature, and h t is the hidden state.
[0034] Preferably, the sparse backpropagation unit employs compressed sensing optimization, satisfying:
[0035]
[0036] where X is the design matrix, z is the sparse coefficient, and y is the observed emission data;
[0037] The policy scenario simulation module adjusts the emission intensity I i,t to satisfy:
[0038] I i,t (p) = I i,t-1 ·exp(-η i p)
[0039] where p is the carbon tax rate, and η i is the industry elasticity parameter.
[0040] Preferably, the policy scenario simulation module adjusts the carbon flow edge weight a′ ij to satisfy:
[0041] a′ ij = a ij ·(1-κp)
[0042] where κ is the policy modulation factor.
[0043] The output results of the multi-stage prediction engine are weighted and fused to satisfy:
[0044]
[0045] where each weight is normalized by the inverse error:
[0046]
[0047] where RMSE m is the root mean square error of the sub-model on the validation set.
[0048] A prediction method of a carbon emission data simulation development prediction system, comprising the following steps:
[0049] S1 Data acquisition and fusion: carbon emission data is collected from multiple fields such as energy, industry, transportation, and construction, the data is standardized and a unified data matrix is formed by missing value filling and outlier removal;
[0050] S2 Carbon flow map modeling: a carbon flow map is constructed with industries, regions, or energy types as nodes, and a graph neural network propagation rule is used
[0051]
[0052] to calculate node embeddings and obtain inter-industry carbon emission dependency relationships;
[0053] S3 Multi-Stage Prediction: A prediction engine that combines long-term trend prediction, short-term disturbance prediction and sparse back-inference. The long-term trend is modeled using a Gaussian process, the short-term disturbance is modeled using a recurrent neural network, and the sparse back-inference is optimized using compressed sensing.
[0054] S4 Scenario Simulation and Output: Input policy parameters such as carbon tax rate and energy consumption limit, and adjust the emission intensity function I. i,t (p)=I i,t-1 ·exp(-η i p) and the edge weight function a′ ij =a ij ·(1-κp) and output the prediction results as trend curves, heatmaps or API interfaces.
[0055] Preferably, the results of the multi-stage prediction are obtained through weighted fusion, satisfying...
[0056]
[0057] Among them, each weight w m Calculated using inverse error normalization:
[0058]
[0059] In summary, the present invention has the following main beneficial effects:
[0060] This invention enhances the fusion and processing capabilities of multi-source heterogeneous data. By introducing standardization, missing value imputation, and robust outlier detection mechanisms, the system can be compatible with structured and unstructured, real-time and offline carbon emission data, ensuring data consistency and time scale uniformity. Under conditions of significant missing data or high noise, the use of exponential moving average, K-nearest neighbor interpolation, and robust Z-score determination effectively guarantees the integrity and reliability of data input, thereby significantly improving the adaptability and stability of the prediction model in complex data environments.
[0061] This invention proposes a carbon flow structure graph modeling mechanism. By constructing a carbon flow graph with industry, region, and energy type as nodes, and performing node embedding and path reasoning based on graph neural networks, it can characterize the causal relationships and dependency structures of carbon emissions between industries. This modeling approach not only improves the interpretability of prediction results but also avoids the shortcomings of traditional black-box models in explaining causal paths. Especially in multi-industry and multi-regional collaborative scenarios, this mechanism can reveal emission hotspots and transmission relationships, providing a structured basis for industry regulation and cross-departmental collaborative emission reduction.
[0062] The present application designs a multi-stage prediction engine, which organically combines long-term trend prediction, short-term disturbance prediction and sparse back-propagation. The long-term trend prediction adopts Gaussian process or Bayesian trend model, which can capture the macro changes of economic development and energy structure adjustment; the short-term disturbance prediction uses recurrent neural network to dynamically respond to sudden factors such as extreme weather and market fluctuations; the sparse back-propagation adopts compressed sensing optimization, which helps to identify key driving factors and maintain the sparsity and generalization ability of the model. The combination of the three not only guarantees the accuracy and robustness of the prediction, but also realizes the balance between long-term stability and short-term flexibility.
[0063] The present application has the ability of real-time simulation and visualization of policy scenarios. By substituting policy parameters such as carbon tax rate and energy consumption limit into the functions of emission intensity and carbon flow edge weight, the system can dynamically adjust the prediction path and generate carbon emission development curves under different policy scenarios in real time. Combined with visual display and API interface output, the present application can be widely used in government carbon neutral path planning, enterprise carbon footprint management, regional energy structure optimization and other scenarios, which not only improves the practicality of the prediction results, but also enhances its application value in decision support. BRIEF DESCRIPTION OF DRAWINGS
[0064] Figure 1 is a system flowchart of the present application. DETAILED DESCRIPTION
[0065] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0066] Embodiment one
[0067] Reference Figure 1 A carbon emission data simulation development prediction system and method, comprising the following steps:
[0068] A multi-source data fusion module is used to receive and uniformly process carbon emission data from the fields of energy, industry, transportation and construction, standardize the data, fill in missing values and remove outliers, and form a unified data matrix;
[0069] A carbon flow structure modeling module is used to construct a carbon flow atlas with industry, region and energy type as nodes, and perform node embedding and path reasoning based on graph neural network to reveal the causal relationship of carbon emission and the dependence between industries;
[0070] A multi-stage prediction engine, including long-term trend prediction, short-term disturbance prediction and sparse back-propagation unit, wherein the long-term trend prediction adopts a Bayesian trend model or a Gaussian process modeling, the short-term disturbance prediction adopts a recurrent neural network, and the sparse back-propagation adopts compressed sensing optimization;
[0071] A policy scenario simulation module for receiving policy parameters such as carbon tax rate and energy consumption limit, and adjusting the edge weight and emission intensity function of the carbon flow atlas in real time to generate a predicted path under different policy backgrounds;
[0072] A result output module for providing the prediction results in the form of trend curve, heat map and API interface to the outside;
[0073] The multi-source data fusion module performs standardization processing on the original data d i,t , adopts the formula:
[0074]
[0075] Wherein, μ i is the field mean, σ i is the standard deviation, is the standardization result;
[0076] The missing value filling adopts exponential moving average or K-nearest neighbor interpolation, and the exponential moving average satisfies:
[0077]
[0078] The K-nearest neighbor interpolation satisfies:
[0079]
[0080] Wherein, K is the number of neighbors, u t is the context vector;
[0081] The carbon flow structure modeling module updates the node embedding based on the graph neural network, satisfying:
[0082]
[0083] Wherein, A is the weighted adjacency matrix, I is the unit matrix, D is the degree matrix, W (k) is the k-th layer weight matrix, and σ is the activation function;
[0084] The long-term trend prediction adopts a Gaussian process model:
[0085]
[0086] Wherein, the kernel function is:
[0087]
[0088] where l is the length scale, ω is the periodic parameter, is the variance parameter;
[0089] The short-term disturbance prediction adopts a long short-term memory network, and its gating mechanism satisfies:
[0090] i t = σ(W i x t + U i h t-1 + b i ), f t = σ(W f x t + U f h t-1 + b f ),
[0091] o t = σ(W o x t + U o h t-1 + b o ), c t
[0092] = f t ⊙ c t-1 + i t ⊙ tanh(W c x t + U c h t-1 + b c )
[0093] where x t is the input feature, h t is the hidden state;
[0094] The sparse backstepping unit adopts compressed sensing optimization, satisfying:
[0095]
[0096] where X is the design matrix, z is the sparse coefficient, and y is the observed emission data;
[0097] The policy scenario simulation module modulates the emission intensity I i,t , satisfying:
[0098] I i,t (p) = I i,t-1 · exp(-η i p)
[0099] where p is the carbon tax rate, and η i is the industry elasticity parameter;
[0100] The policy scenario simulation module adjusts the carbon flow edge weight a' ij to satisfy:
[0101] a' ij = a ij ·(1-κp)
[0102] where κ is the policy modulation factor;
[0103] The output results of the multi-stage prediction engine are weighted and fused to satisfy:
[0104]
[0105] where each weight is normalized by the inverse error:
[0106]
[0107] where RMSE m is the root mean square error of the sub-model on the validation set.
[0108] Embodiment Two
[0109] A prediction method of a carbon emission data simulation development prediction system, comprising the following steps:
[0110] S1 Data acquisition and fusion: carbon emission data is collected from multiple fields such as energy, industry, transportation, and construction, and the data is standardized and a unified data matrix is formed by missing value filling and outlier removal;
[0111] S2 Carbon flow map modeling: industry, region, or energy type is taken as a node to build a carbon flow map, and a graph neural network propagation rule is used
[0112]
[0113] to calculate node embedding and obtain the carbon emission dependency relationship between industries;
[0114] S3 Multi-stage prediction: a prediction engine combining long-term trend prediction, short-term disturbance prediction, and sparse backpropagation is used, where the long-term trend is modeled by a Gaussian process, the short-term disturbance is a recurrent neural network, and the sparse backpropagation is optimized by compressive sensing;
[0115] S4 Scenario simulation and output: input policy parameters such as carbon tax rate and energy consumption limit, adjust the emission intensity function I i,t (p) = I i,t-1 ·exp(-η i p) and the edge weight function a' ij = a ij• (1-κp) and output the prediction results in the form of trend curve, heat map or API interface;
[0116] The results of the multi-stage prediction are obtained by weighted fusion, satisfying
[0117]
[0118] where each weight w m The inverse error normalization is calculated as:
[0119] Embodiment Three
[0120] The system is used for the simulation and prediction of national macro carbon emission trends.
[0121] First, through the multi-source data fusion module, the power production data (unit: GWh) disclosed by the National Energy Administration, the steel production data (unit: ten thousand tons) announced by the Ministry of Industry, the fuel consumption data (unit: ten thousand tons of oil equivalent) announced by the Ministry of Transport, and the average temperature data provided by the meteorological department are accessed. The original data d i,t After standardization:
[0122]
[0123] For example, the mean value μ steel of steel production is 10 million tons, the standard deviation σ steel is 1.5 million tons, the current production is 12 million tons, and the standardized value is (1200-1000) / 150=1.33 (1200-1000) / 150=1.33 (1200-1000) / 150=1.33;
[0124] When some monthly data is missing, exponential moving average interpolation is used:
[0125]
[0126] Thus ensuring the continuity and integrity of the input data;
[0127] After that, the carbon flow structure modeling module constructs a graph G=(V,E), where the node V represents the industry sector, and the edge weight a ij represents the cross-industry carbon flow intensity;
[0128] For example, the carbon flow weight between the power industry and the transportation industry is set as:
[0129] a ij =0.5·IO ij +0.3·Trans ij +0.2·Corr ij
[0130] If the input-output weight of the power industry to the transportation industry is 0.8, the energy transmission weight is 0.6, and the emission-related coefficient is 0.7, then a ij = 0.5 x 0.8 + 0.3 x 0.6 + 0.2 x 0.7 = 0.73;
[0131] In the propagation process of the graph neural network, the node embedding is updated as:
[0132]
[0133] This can capture the dependency between different industries;
[0134] In the prediction phase, the multi-stage prediction engine uses a Gaussian process to fit the long-term trend:
[0135]
[0136] For example, set the length scale l = 24 months and the period w = 12 months to fit the economic cycle and seasonal emission fluctuations. Short-term disturbances are captured by LSTM, such as short-term energy consumption surges caused by extreme weather;
[0137] Finally, the prediction result shows a downward trend in total carbon emissions over the next 10 years, and outputs the trend curve and 95% confidence interval, providing a scientific basis for national policy-making;
[0138] This embodiment uses a graph neural network to model the carbon flow relationship between industries, and through the fusion of long-term and short-term prediction models, it reflects both macro trends and responds to sudden disturbances, achieving dynamic prediction at the national level.
[0139] Embodiment Four
[0140] In this embodiment, the system is applied to carbon footprint management and policy scenario analysis of a large steel enterprise;
[0141] The enterprise uploads production data (coke consumption, iron ore consumption, power usage) and corresponding carbon emissions every month. After data fusion module, first standardization, then robust Z-score to remove outliers:
[0142]
[0143] If the coke consumption value deviates from the median by 4.2 times the MAD, it is determined to be abnormal and is removed or given a lower weight;
[0144] In the carbon flow structure map, the nodes include "ironmaking", "steelmaking", and "steel rolling", and the edge weight is calculated through input-output relationship and historical emission correlation. For example, the edge weight from ironmaking to steelmaking is as high as 0.9, indicating that changes in ironmaking capacity have a significant impact on steelmaking emissions;
[0145] In the prediction stage, short-term perturbation prediction is modeled by LSTM, capturing the emission changes caused by capacity adjustment and energy price fluctuation; long-term trend prediction is completed by Gaussian process;
[0146] When the enterprise inputs policy parameters such as carbon tax rate p = 50 yuan / ton CO2e, the system modulates the emission intensity:
[0147] I i,t (p)=I i,t-1 ·exp(-η i p)
[0148] Assuming the carbon price elasticity of the ironmaking link is η = 0.01, the emission intensity decreases to exp(-0.5) ≈ 0.61 of the original value, i.e. 39% reduction in emissions;
[0149] At the same time, the carbon flow edge weight adjusts with the policy:
[0150] a′ ij =a ij ·(1-κp)
[0151] If κ = 10 -3 , the edge weight decreases by 5%, indicating that the policy has reduced the high-carbon flow between links;
[0152] Finally, the system outputs the emission path curve under different carbon tax levels, showing that the enterprise's carbon emissions decrease faster under a high carbon tax background, providing a basis for the enterprise to develop peak strategy;
[0153] By substituting policy parameters into the emission intensity and carbon flow atlas, the system can dynamically simulate the impact of policy on the enterprise's carbon emission path, embodying the real-time and practicality of scenario simulation.
[0154] Example Five
[0155] In this embodiment, the system is applied to energy structure adjustment and carbon emission prediction in a certain region;
[0156] The regional carbon emission data includes thermal power generation, wind power generation, photovoltaic power generation and corresponding emission factors. The data fusion module unifies the time scale to "month" and fills in the missing wind power data, using K-nearest neighbor interpolation:
[0157]
[0158] Among them, the similarity vector contains wind speed and load rate;
[0159] When constructing the carbon flow structure atlas, "thermal power", "wind power" and "photovoltaic" are taken as nodes, and the carbon flow edge weight reflects the complementarity in power grid dispatching. For example, the edge weight between thermal power and wind power is 0.4, indicating that wind power fluctuation needs to be bottomed out by thermal power;
[0160] In the prediction stage, the long-term trend prediction captures the growth trend of new energy installed capacity by Gaussian process, and the short-term disturbance prediction identifies the fluctuation influence of meteorological conditions on wind and light output by LSTM;
[0161] If the regional planning sets the new energy proportion target, and makes the thermal power gradually decline, the system can reverse the key driving factors through compressive sensing:
[0162]
[0163] The optimization results show that the new energy installed capacity growth coefficient is highly correlated with the carbon emission reduction, and the factor weight in the sparse coefficient z is greater than 0.6; the output results show the carbon emission path under the new energy proportion of 50%, 70%, and 90%; the prediction shows that when the new energy proportion reaches 70%, the regional carbon emission will achieve a turning point of decline in about ten years;
[0164] By identifying the key driving factors through compressive sensing technology, combining graph structure modeling and multi-stage prediction, the system can provide decision basis for regional energy structure optimization, and realize the linkage of macro planning and micro data.
[0165] Although the embodiments of the present application have been shown and described, it can be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.
Claims
1. A carbon emission data analog development forecasting system, characterized by, The method comprises the following steps: A multi-source data fusion module is configured to receive and uniformly process carbon emission data from the fields of energy, industry, transportation, and construction, standardize the data, fill in missing values, and remove outliers to form a uniform data matrix; A carbon flow structure modeling module is configured to construct a carbon flow atlas with industries, regions, and energy types as nodes, perform node embedding and path reasoning based on a graph neural network to reveal the causal relationship of carbon emissions and the dependence between industries; A multi-stage prediction engine includes long-term trend prediction, short-term disturbance prediction, and sparse back-propagation units, wherein the long-term trend prediction adopts a Bayesian trend model or a Gaussian process model, the short-term disturbance prediction adopts a recurrent neural network, and the sparse back-propagation adopts compressed sensing optimization; A policy scenario simulation module is configured to receive policy parameters such as carbon tax rates and energy consumption limits, adjust the edge weights and emission intensity functions of the carbon flow atlas in real time, and generate predicted paths under different policy backgrounds; A result output module is configured to provide the prediction results in the form of trend curves, heat maps, and API interfaces.
2. The carbon emissions data analog development forecasting system of claim 1, wherein, The multi-source data fusion module performs standardization processing on the original data d i,t using the formula: where μ i is the field mean, σ i is the standard deviation, is the normalized result.
3. A carbon emissions data analog development forecasting system according to claim 2, wherein, The missing value filling adopts exponential moving average or K-nearest neighbor interpolation, and the exponential moving average satisfies: The K-nearest neighbor interpolation satisfies: where K is the number of neighbors, u t is the context vector.
4. The carbon emissions data analog development forecasting system of claim 3, wherein, The carbon flow structure modeling module updates the node embedding based on the graph neural network, satisfying: Wherein, A is a weighted adjacency matrix, I is a unit matrix, D is a degree matrix, W (k) is the weight matrix of the kth layer, and σ is an activation function.
5. A carbon emissions data analog development forecasting system according to claim 4, wherein, The long-term trend prediction adopts a Gaussian process model: The kernel function is: where l is a length scale and ω is a periodic parameter, is a variance parameter.
6. A carbon emissions data analog development forecasting system according to claim 5, wherein, The short-term disturbance prediction adopts a long short-term memory network, and the gating mechanism satisfies: i t = σ(W f x t + U i h t-1 + b i ), f t = σ(W f x t + U f h t-1 + b f ), o t = σ(W o x t + U o h t-1 + b o ),c t = f t ⊙c t-1 + i t ⊙ tanh(W c x t + U c h t-1 + b c ) where x t is the input feature, h t is the hidden state.
7. A carbon emissions data analog development forecasting system according to claim 6, wherein, The sparse back-propagation unit adopts compressed sensing optimization, satisfying: Where X is the design matrix, z is the sparse coefficient, and y is the observed emission data. The policy scenario simulation module performs modulation on the emission intensity I i,t satisfies: I i,t (p) = I i,t-1 • exp(-η i p) where p is the carbon tax rate, η i is the industry elasticity parameter.
8. A carbon emissions data analog development forecasting system according to claim 7, wherein, The policy scenario simulation module adjusts the carbon flow edge weight a' ij to satisfy: a′ ij = a ij • (1 - kp) Where κ is the policy modulation factor. The output results of the multi-stage prediction engine are weighted and fused, satisfying: Where each weight is normalized by inverse error: where RMSE is the root mean square error of the sub-model on the validation set. m where RMSE is the root mean square error of the sub-model on the validation set.
9. A prediction method suitable for use in a carbon emission data analog development prediction system according to any one of claims 1 to 8, characterized in that, The method comprises the following steps: S1 data collection and fusion: collect carbon emission data from energy, industry, transportation, construction and other fields, and standardize the data And through the missing value filling and outlier rejection to form a unified data matrix; S2 Carbon flow atlas modeling: constructing a carbon flow atlas with industries, regions, or energy types as nodes based on graph neural network propagation rules Calculate node embedding to obtain the dependence relationship between industries; S3 Multi-stage prediction: using a prediction engine that combines long-term trend prediction, short-term disturbance prediction, and sparse back-propagation, wherein the long-term trend is modeled by a Gaussian process, the short-term disturbance is modeled by a recurrent neural network, and the sparse back-propagation is optimized by compressed sensing; S4 Scenario simulation and output: input policy parameters such as carbon tax rate, energy consumption limit, etc., adjust the emission intensity function I i,t (p) = I i,t-1 · exp(-η i p) and the edge weight function a' ij = a ij · (1 - κp) and output the prediction results in the form of trend curve, heat map or API interface.
10. The prediction method of a carbon emission data analog development prediction system according to claim 9, wherein, The results of the multi-stage prediction are obtained by weighted fusion, satisfying where each weight w m Inverse error normalization calculation: