Regional carbon emission prediction method and device, electronic equipment and storage medium
By processing carbon emission data and adjacency relationships through feature decomposition and graph convolutional networks, the problems of redundant information and feature interference in existing technologies are solved, and more efficient and accurate carbon emission prediction is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INNER MONGOLIA DIGITAL INFORMATION CO LTD
- Filing Date
- 2025-12-02
- Publication Date
- 2026-05-08
AI Technical Summary
Existing carbon emission prediction methods use the same approach when processing spatial and temporal features, which leads to the introduction of redundant information and computational burden, reducing prediction accuracy and efficiency. Furthermore, complex fusion strategies are prone to introducing feature interference.
We employ feature decomposition and graph convolutional networks to process carbon emission data and adjacency relationships respectively. We extract component features at different time scales using ensemble empirical mode decomposition and input them, along with carbon emission features and adjacency relationship features, into a graph convolutional network for prediction.
It improves the accuracy and efficiency of carbon emission forecasting, avoids redundant information and feature interference, and enhances the accuracy of forecast results.
Smart Images
Figure CN121996938A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing, and more particularly to a method, apparatus, electronic device, and storage medium for predicting regional carbon emissions. Background Technology
[0002] Global climate change has become the greatest non-traditional security challenge facing human development. In actively addressing this challenge, carbon peaking and carbon neutrality are of paramount importance. As the world's largest carbon emitter, my country has consistently strived to reduce carbon emissions. Therefore, clarifying my country's current carbon emission status, analyzing the main factors influencing carbon emissions, and predicting future carbon emission trends are of profound significance for my country's carbon reduction efforts. Currently, some carbon emission analysis technologies have made progress in practical applications, such as regional carbon emission monitoring and industry emission contribution assessment. However, fine-grained prediction of carbon emissions, especially techniques for modeling and analyzing complex relationships using multi-dimensional data, has not yet achieved large-scale application. The main reason is that the performance of existing models is insufficient to meet practical needs.
[0003] Carbon emission prediction typically requires modeling from both spatial and temporal dimensions to identify the inter-regional interactions and temporal variations in carbon emissions. Current mainstream methods often treat these two types of features the same way, such as directly concatenating spatial and temporal features as input. However, this approach easily introduces redundant information. In reality, depending on the prediction region, not only the complex dynamic changes of time-series features but also the spatial correlations between regions must be considered. Therefore, this indiscriminate approach not only increases the computational burden but may also hinder the effective extraction of important information and patterns from carbon emission data, reducing prediction accuracy and efficiency. Furthermore, most existing methods employ complex fusion strategies, directly encoding spatial and temporal features uniformly to generate joint feature representations, and then predicting carbon emission values using these joint features. However, in real-world scenarios, the contribution of data features from different samples to the prediction task is non-uniform. For carbon emission prediction in some regions, high prediction accuracy may be achieved using only temporal features. Forcing the use of joint feature representations can easily introduce redundant information, leading to mutual interference between different features and reducing prediction accuracy. Summary of the Invention
[0004] This application provides a regional carbon emission prediction method, apparatus, electronic device, and storage medium to address the shortcomings of existing technologies that use the same processing method for spatial and temporal characteristics, and employ complex fusion strategies for spatial and temporal characteristics, thereby significantly improving the accuracy of carbon emission prediction.
[0005] This application provides a method for predicting regional carbon emissions, including the following steps: The system acquires the first historical carbon emission data corresponding to the first-level administrative objects in the target area, the first geographical adjacency data between each of the first-level administrative objects, the second historical carbon emission data corresponding to the second-level administrative objects under each of the first-level administrative objects, the second geographical adjacency data between each of the second-level administrative objects, and the prediction period. The carbon emission data is time-series data. Based on the first historical carbon emission data and the second historical carbon emission data, the first carbon emission characteristics corresponding to the target area are obtained; Based on the first geographical adjacency data and the second geographical adjacency data, the first adjacency feature corresponding to the target area is obtained; Feature decomposition is performed on the first historical carbon emission data and the second historical carbon emission data to obtain the component features of the first historical carbon emission data and the second historical carbon emission data at multiple different time scales. The second carbon emission feature is obtained by fusing multiple component features with the first carbon emission feature. The predicted time period, the second carbon emission feature, and the first adjacency relationship feature are input into a preset graph convolutional network to obtain the carbon emission prediction results of each of the first-level administrative objects in the target area during the predicted time period.
[0006] According to the regional carbon emission prediction method provided in this application, obtaining the first carbon emission characteristic corresponding to the target region based on the first historical carbon emission data and the second historical carbon emission data includes: The primary carbon emission characteristics corresponding to the target area are obtained based on the first historical carbon emission data. Based on the historical carbon emission data of the secondary administrative objects under each primary administrative object in the second historical carbon emission data, the secondary carbon emission characteristics of each primary administrative object are obtained. The primary carbon emission characteristics are enhanced based on the secondary carbon emission characteristics to obtain the first carbon emission characteristics corresponding to the target region.
[0007] According to the regional carbon emission prediction method provided in this application, the step of enhancing the primary carbon emission characteristics based on the secondary carbon emission characteristics to obtain the first carbon emission characteristics corresponding to the target region includes: For each primary administrative entity in the target area, determine the mean value of its corresponding secondary carbon emission characteristics; The first carbon emission feature is obtained by fusing the mean value corresponding to each of the first-level administrative objects with the feature corresponding to the first-level administrative object in the first-level carbon emission feature.
[0008] According to the regional carbon emission prediction method provided in this application, obtaining the first adjacency characteristics corresponding to the target region based on the first geographical adjacency data and the second geographical adjacency data includes: The first-level adjacency characteristics of the target area are obtained based on the first geographical adjacency data; The secondary adjacency characteristics of the target area are obtained based on the second geographical adjacency data; The first-level adjacency feature is enhanced based on the second-level adjacency feature to obtain the first adjacency feature.
[0009] According to the regional carbon emission prediction method provided in this application, the step of enhancing the first-level adjacency features based on the second-level adjacency features to obtain the first adjacency features includes: The second-level adjacency relationship features are aggregated to obtain aggregated second-level adjacency relationship features; The aggregated second-level adjacency features are fused with the first-level adjacency features to obtain the first adjacency feature.
[0010] According to the regional carbon emission prediction method provided in this application, feature decomposition is performed on the first historical carbon emission data and the second historical carbon emission data to obtain the component features of the first historical carbon emission data and the second historical carbon emission data at multiple different time scales, including: The first historical carbon emission data is decomposed using the ensemble empirical mode decomposition method to obtain at least one first intrinsic mode function component and a first residual component. The component features of the first historical carbon emission data at multiple different time scales are obtained based on the at least one first intrinsic mode function component and the first residual component. The second historical carbon emission data is decomposed using the ensemble empirical mode decomposition method to obtain at least one second intrinsic mode function component and a second residual component. Based on the at least one second intrinsic mode function component and the second residual component, the component features of the second historical carbon emission data at multiple different time scales are obtained.
[0011] According to the regional carbon emission prediction method provided in this application, the graph convolutional network includes fully connected layers and a preset number of graph convolutional layers; the step of inputting the prediction period, the second carbon emission feature, and the first adjacency relationship feature into the preset graph convolutional network to obtain the carbon emission prediction results of each of the first-level administrative objects in the target region during the prediction period includes: The second carbon emission feature and the first adjacency relationship feature are extracted by the preset number of graph convolutional layers to obtain convolutional features; The convolutional features and the prediction period are processed by the fully connected layer to obtain the carbon emission prediction results of each of the first-level administrative objects in the target area during the prediction period; In this process, each graph convolutional layer is used for feature extraction using the following formulas (1)-(3), and the carbon emission prediction result is expressed using the following formula (4): (1) (2) (3) (4) in, , The matrix representing the second carbon emission characteristic, The matrix representing the characteristics of the first adjacency relationship. It is the identity matrix. For the first The input feature matrix of the layer, For degree matrix, For the first The learnable parameter matrix of the layer, For activation function, For the preset quantity, The weights of the fully connected layer, This is the bias for the fully connected layer.
[0012] This application also provides a regional carbon emission prediction device, including the following modules: The first acquisition module is used to acquire the first historical carbon emission data corresponding to the first-level administrative objects in the target area, the first geographical adjacency data between each of the first-level administrative objects, the second historical carbon emission data corresponding to the second-level administrative objects under each of the first-level administrative objects, the second geographical adjacency data between each of the second-level administrative objects, and the prediction period, wherein the carbon emission data is time series data. The second acquisition module is used to obtain the first carbon emission characteristics corresponding to the target area based on the first historical carbon emission data and the second historical carbon emission data. The third acquisition module is used to obtain the first adjacency relationship feature corresponding to the target area based on the first geographic adjacency relationship data and the second geographic adjacency relationship data; The feature decomposition module is used to perform feature decomposition on the first historical carbon emission data and the second historical carbon emission data to obtain the component features of the first historical carbon emission data and the second historical carbon emission data at multiple different time scales. A feature fusion module is used to fuse multiple component features with the first carbon emission feature to obtain a second carbon emission feature; The fourth acquisition module is used to input the prediction period, the second carbon emission feature, and the first adjacency relationship feature into a preset graph convolutional network to obtain the carbon emission prediction results of each of the first-level administrative objects in the target area during the prediction period.
[0013] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a regional carbon emission prediction method as described above.
[0014] This application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a regional carbon emission prediction method as described above.
[0015] This application also provides a computer program product, including a computer program that, when executed by a processor, implements a regional carbon emission prediction method as described above.
[0016] In implementing the regional carbon emission prediction method of this application, carbon emission data corresponds to temporal characteristics, and adjacency relationships correspond to spatial characteristics. For carbon emission data, firstly, based on first and second historical carbon emission data, the first carbon emission characteristic corresponding to the target area is obtained. Then, through feature decomposition, the component characteristics of the first and second historical carbon emission data at multiple different time scales are obtained, and the multiple component characteristics are fused with the first carbon emission characteristic to obtain the second carbon emission characteristic. For adjacency relationships, the first adjacency relationship characteristic corresponding to the target area is obtained based on first and second geographical adjacency relationship data. This application uses different methods to process carbon emission data and adjacency relationships separately, which avoids the situation in related technologies where spatial and temporal characteristics are directly concatenated as input, leading to redundant information, increased computational burden, and hindering the effective mining of important information and patterns in carbon emission data, thereby improving the accuracy and efficiency of prediction. Secondly, this application does not employ a complex feature fusion strategy. Instead of directly encoding spatial and temporal features to generate joint feature representations and predicting carbon emission results through joint features, it inputs the second carbon emission feature and the first adjacency relationship feature into a pre-defined graph convolutional network (i.e., performs a simple feature fusion of the second carbon emission feature and the first adjacency relationship feature and then inputs them into the graph convolutional network) to obtain carbon emission prediction results. This avoids the situation where complex fusion strategies can easily introduce redundant information and cause mutual interference between different features, thereby further improving the accuracy of the prediction results. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a flowchart illustrating a regional carbon emission prediction method according to an embodiment of this application; Figure 2 This is a schematic diagram illustrating a first-level adjacency feature matrix according to an embodiment of this application; Figure 3 This is a schematic diagram illustrating the implementation process of a regional carbon emission prediction method according to an embodiment of this application; Figure 4 This is a structural block diagram of a regional carbon emission prediction device according to an embodiment of this application; Figure 5 This is a schematic diagram of the physical structure of an electronic device according to an embodiment of this application. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0020] Figure 1 This is a flowchart illustrating a regional carbon emission prediction method according to an embodiment of this application. (Refer to...) Figure 1 The regional carbon emission prediction method of this application may include the following steps: Step 101: Obtain the first historical carbon emission data corresponding to the first-level administrative objects in the target area, the first geographical adjacency data between each first-level administrative object, the second historical carbon emission data corresponding to the second-level administrative objects under each first-level administrative object, the second geographical adjacency data between each second-level administrative object, and the prediction period. The carbon emission data is time series data.
[0021] In this application, the target area refers to the region where carbon emission results need to be predicted. The target area can be any geographical region, and can be determined according to actual business needs. For example, the target area can be the Beijing-Tianjin-Hebei region, the Pearl River Delta region, etc.
[0022] In this application, considering that the administrative divisions in my country can include provincial-level administrative regions (including provinces, autonomous regions, municipalities directly under the central government, and special administrative regions), prefecture-level administrative regions (including prefecture-level cities, regions, autonomous prefectures, and leagues), county-level administrative regions (including urban districts, county-level cities, counties, autonomous counties, banners, autonomous banners, special zones, and forest areas), and township-level administrative regions (including subdistricts, townships, ethnic townships, and towns), if the first-level administrative object is a provincial-level administrative region, then the second-level administrative object can be a prefecture-level administrative region; if the first-level administrative object is a prefecture-level administrative region, then the second-level administrative object can be a county-level administrative region, and so on.
[0023] For example, if the target area includes prefecture-level city 1, prefecture-level city 2, and prefecture-level city 3, then the first-level administrative objects are prefecture-level city 1, prefecture-level city 2, and prefecture-level city 3. If prefecture-level city 1 has district 1, district 2, county-level city 1, and county 1 under its jurisdiction, prefecture-level city 2 has district 3, county-level city 2 under its jurisdiction, and prefecture-level city 3 has district 4, county-level city 3 under its jurisdiction, then district 1, district 2, county-level city 1, county 1, district 3, county-level city 2, district 4, and county-level city 3 are the second-level administrative objects under all the first-level administrative objects in the entire target area.
[0024] The first set of historical carbon emission data includes the carbon emissions and statistical characteristics (such as mean, variance, and growth rate) of each primary administrative entity over multiple time periods. For example, if the primary administrative entity is prefecture-level city 1, then the historical carbon emission data for prefecture-level city 1 can include the carbon emissions for each year, each quarter, and each month over the past 10 years, as well as the year-on-year growth rate of carbon emissions compared to the previous year. The statistical characteristics can be set according to actual needs.
[0025] The first geographic adjacency data is obtained by statistically analyzing whether each first-level administrative object in the target area is geographically adjacent to another first-level administrative object. In other words, the first geographic adjacency data can determine which first-level administrative objects each first-level administrative object in the target area is adjacent to and which are not adjacent to another first-level administrative object.
[0026] The second set of historical carbon emission data includes the carbon emissions and statistical characteristics (such as mean, variance, and growth rate) of each secondary administrative entity over multiple time periods. For example, if the secondary administrative entity is county-level city 1 under prefecture-level city 1, then the historical carbon emission data for county-level city 1 can include the carbon emissions of county-level city 1 in each year, the carbon emissions in each quarter of each year, the carbon emissions in each month of each year, and the growth rate of carbon emissions from the previous year to the previous year for each year.
[0027] The second-level geographic adjacency data is obtained by statistically analyzing the relationships between various secondary administrative entities within the target area across multiple preset dimensions. These preset dimensions can include direct adjacency, marginal adjacency, and regional association. Direct adjacency refers to entities belonging to the same primary administrative entity and being geographically adjacent; marginal adjacency refers to entities not belonging to the same primary administrative entity but being geographically adjacent; and regional association refers to relationships involving factors such as economic, social, and environmental factors.
[0028] The forecast period refers to the time period during which carbon emission results need to be predicted.
[0029] Step 102: Based on the first historical carbon emission data and the second historical carbon emission data, obtain the first carbon emission characteristics corresponding to the target area.
[0030] In this application, second historical carbon emission data can be used to enhance first historical carbon emission data to obtain a first carbon emission characteristic. The specific implementation of step 102 will be described in detail later in this application.
[0031] Step 103: Based on the first geographic adjacency data and the second geographic adjacency data, obtain the first adjacency features corresponding to the target area.
[0032] In this application, the first geographic adjacency data can be enhanced using second geographic adjacency data to obtain first adjacency features. The specific implementation of step 103 will be described in detail below.
[0033] Step 104: Perform feature decomposition on the first historical carbon emission data and the second historical carbon emission data to obtain the component features of the first historical carbon emission data and the second historical carbon emission data at multiple different time scales.
[0034] Specifically, step 104 may include: The first historical carbon emission data is decomposed using the Ensemble Empirical Mode Decomposition (EEMD) method to obtain at least one first intrinsic mode function component and a first residual component. Based on the at least one first intrinsic mode function component and the first residual component, the component features of the first historical carbon emission data at multiple different time scales are obtained. The second historical carbon emission data is decomposed using the ensemble empirical mode decomposition method to obtain at least one second intrinsic mode function component and a second residual component. Based on the at least one second intrinsic mode function component and the second residual component, the component features of the second historical carbon emission data at multiple different time scales are obtained.
[0035] In this application, the historical carbon emission data obtained in step 101 are all time-series data. The ensemble empirical mode decomposition method is used to process the historical carbon emission data (assuming it is...). By performing eigenvalue decomposition, multiple intrinsic mode function (IMF) components and remainder terms can be obtained. As shown below: , t=1,2,…,T in, The decomposition yields the first... Each time-series component This indicates the remaining trend term.
[0036] Therefore, by performing eigenvalue decomposition on the first historical carbon emission data using the ensemble empirical mode decomposition method, at least one IMF component (first intrinsic mode function component) and a remainder term can be obtained. (First residual component), then, the obtained first eigenmode function components and first residual components can be used as component features of the first historical carbon emission data at multiple different time scales.
[0037] Similarly, by performing eigenvalue decomposition on the second historical carbon emission data using the ensemble empirical mode decomposition method, at least one IMF component (second intrinsic mode function component) and a remainder term can be obtained. (Second residual component), then, the obtained second intrinsic mode function components and second residual components can be used as component features of the second historical carbon emission data at multiple different time scales.
[0038] In this application, the ensemble empirical mode decomposition method is a method for processing nonlinear, non-stationary time series data, which can automatically separate multiple component features of different time scales suitable for the first historical carbon emission data and multiple component features of different time scales suitable for the second historical carbon emission data.
[0039] In this application, the IMF components typically correspond to the fluctuation and noise components in historical carbon emission data. Specifically, high-frequency IMF components mainly correspond to the noise components, while low-frequency IMF components correspond to the fluctuation components. (Remaining items) This represents the long-term trend of historical carbon emission data, also known as the trend component.
[0040] High-frequency IMF components change rapidly, resembling random fluctuations without a clear periodic pattern, similar to noise components. In carbon emission data, some sudden, accidental events may cause small fluctuations, which will be reflected in the high-frequency IMF components.
[0041] Low-frequency IMF components exhibit relatively stable periodic variations, reflecting regular fluctuations in the data over a short period, consistent with the characteristics of volatile components. For example, in carbon emission data, cyclical increases and decreases in carbon emissions occur every few years due to factors such as industrial policy adjustments and seasonal production changes; these changes are reflected in low-frequency IMF components.
[0042] Remaining items This is the portion remaining after multiple rounds of screening and decomposition. Its changes are relatively slow, reflecting the overall direction of data change over a longer period. Therefore, the trend component can be used to reflect whether carbon emission data is rising, falling, or remaining stable over a longer time. For example, with the continuous optimization of the industrial structure of a prefecture-level city and the long-term implementation of environmental protection policies, carbon emission data generally shows a downward trend, and this downward trend will be reflected in the remaining items. middle.
[0043] Step 105: Fuse multiple component features with the first carbon emission feature to obtain the second carbon emission feature.
[0044] In step 105, the component features obtained in step 104 can be fused with the first carbon emission feature to obtain the second carbon emission feature.
[0045] This application uses a direct splicing method to fuse the various component characteristics with the first carbon emission characteristic. Of course, other methods can also be used to achieve simple feature fusion, and this application does not impose specific restrictions on this.
[0046] Step 106: Input the prediction period, the second carbon emission feature, and the first adjacency relationship feature into the preset graph convolutional network (GCN) to obtain the carbon emission prediction results of each first-level administrative object in the target area during the prediction period.
[0047] In conjunction with the above embodiments, in one implementation, the graph convolutional network includes fully connected layers and a predetermined number of graph convolutional layers. Accordingly, step 106 may specifically include: Step 1061: Extract features from the second carbon emission feature and the first adjacency relationship feature using a preset number of graph convolutional layers to obtain convolutional features; Step 1062: Process the convolutional features and prediction time period through a fully connected layer to obtain the carbon emission prediction results of each first-level administrative object in the target area during the prediction time period; In this process, each graph convolutional layer is used for feature extraction using the following formulas (1)-(3), and the carbon emission prediction result is expressed using the following formula (4): (1) (2) (3) (4) in, , The matrix representing the second carbon emission characteristic, The matrix representing the characteristics of the first adjacency relationship. It is the identity matrix. For the first The input feature matrix of the layer, For degree matrix, For the first The learnable parameter matrix of the layer, For activation function, For the preset quantity, The weights of the fully connected layer, This is the bias for the fully connected layer.
[0048] in, It can be used Activation function.
[0049] In this application, the preset number can be set to 2, which can improve prediction efficiency while ensuring the accuracy of the prediction results. With a preset number of 2, the operation in the first graph convolutional layer is as follows: The operations in the second graph convolutional layer are as follows: .
[0050] In this application, carbon emission data corresponds to temporal features, and adjacency relationships correspond to spatial features. For carbon emission data, firstly, based on first and second historical carbon emission data, the first carbon emission feature corresponding to the target area is obtained. Then, feature decomposition is used to obtain the component features of the first and second historical carbon emission data at multiple different time scales. These component features are then fused with the first carbon emission feature to obtain the second carbon emission feature. For adjacency relationships, the first adjacency relationship feature corresponding to the target area is obtained based on first and second geographical adjacency relationship data. This application uses different methods to process carbon emission data and adjacency relationships separately, avoiding the situation in related technologies where spatial and temporal features are directly concatenated as input, leading to redundant information, increased computational burden, and hindering the effective mining of important information and patterns in carbon emission data. This improves the accuracy and efficiency of prediction. Secondly, this application does not employ a complex fusion strategy. Instead of directly encoding spatial and temporal features to generate joint feature representations and predicting carbon emission results through joint features, it inputs the second carbon emission feature and the first adjacency relationship feature into a pre-defined graph convolutional network (i.e., performs a simple feature fusion of the second carbon emission feature and the first adjacency relationship feature and then inputs them into the graph convolutional network) to obtain carbon emission prediction results. This avoids the situation where complex fusion strategies can easily introduce redundant information and cause mutual interference between different features, thereby further improving the accuracy of the prediction results.
[0051] In conjunction with the above embodiments, in one implementation, step 102 may include: Step 1021: Obtain the primary carbon emission characteristics corresponding to the target area based on the first historical carbon emission data.
[0052] In this application, the primary carbon emission characteristics of the target area are composed of the carbon emission characteristics of each primary administrative entity within the target area.
[0053] In this application, after obtaining the primary carbon emission characteristics, these characteristics can be recorded in a matrix to obtain a primary carbon emission characteristic matrix. For example, each row in the primary carbon emission characteristic matrix represents time information, each column represents a primary administrative object, and each element in the matrix represents the carbon emission characteristics of each primary administrative object under different time information. The time information includes the time information corresponding to statistical characteristics (such as mean, variance, growth rate, etc.).
[0054] Step 1022: Based on the historical carbon emission data of the secondary administrative objects under each primary administrative object in the second historical carbon emission data, obtain the secondary carbon emission characteristics of each primary administrative object.
[0055] In step 1022, the secondary carbon emission characteristics corresponding to each primary administrative object can be recorded in a matrix to obtain the secondary carbon emission characteristic matrix corresponding to each primary administrative object.
[0056] For example, based on the historical carbon emission data of the secondary administrative objects under the jurisdiction of the primary administrative object 1, the secondary carbon emission feature matrix 1 corresponding to the primary administrative object 1 can be obtained, and based on the historical carbon emission data of the secondary administrative objects under the jurisdiction of the primary administrative object 2, the secondary carbon emission feature matrix 2 corresponding to the primary administrative object 2 can be obtained.
[0057] For the secondary carbon emission characteristic matrix 1, for example, each row can represent time information, and each column can represent a secondary administrative object under the jurisdiction of the primary administrative object 1. Each element in the matrix represents the carbon emission characteristics of each secondary administrative object under different time information. The time information includes the time information corresponding to statistical characteristics (such as mean, variance, growth rate, etc.).
[0058] Step 1023: Enhance the primary carbon emission characteristics based on the secondary carbon emission characteristics to obtain the first carbon emission characteristics corresponding to the target region.
[0059] Step 1023 may include: For each primary administrative entity in the target area, determine the mean value of its corresponding secondary carbon emission characteristics; The first carbon emission feature is obtained by fusing the mean value corresponding to each of the first-level administrative objects with the feature corresponding to the first-level administrative object in the first-level carbon emission feature.
[0060] In determining the average of the secondary carbon emission characteristics corresponding to each primary administrative object, one can use... Determined, where M represents the number of secondary administrative entities under the current primary administrative entity. This is the carbon emission characteristic matrix corresponding to the m-th secondary administrative entity under the current primary administrative entity.
[0061] When implementing step 1023, for example, if the target area includes first-level administrative objects 1-3, then the mean value corresponding to first-level administrative object 1 is fused with the feature corresponding to first-level administrative object 1 in the first-level carbon emission feature matrix, the mean value corresponding to first-level administrative object 2 is fused with the feature corresponding to first-level administrative object 2 in the first-level carbon emission feature matrix, and the mean value corresponding to first-level administrative object 3 is fused with the feature corresponding to first-level administrative object 3 in the first-level carbon emission feature matrix, and finally the first carbon emission feature in matrix form is obtained.
[0062] In this application, the fusion operation in step 1023 can be a splicing operation or other fusion operations, which can be set according to actual needs.
[0063] In this application, firstly, the primary carbon emission characteristics corresponding to the target area are obtained based on the first historical carbon emission data, and secondly, the secondary carbon emission characteristics are obtained based on the second historical carbon emission data. Then, the primary carbon emission characteristics are enhanced using the secondary carbon emission characteristics to obtain the enhanced primary carbon emission characteristics. The primary carbon emission characteristics are the enhanced primary carbon emission characteristics. This operation can enrich the representation of the primary carbon emission characteristics, thereby improving the accuracy of the prediction results.
[0064] In conjunction with the above embodiments, in one implementation, step 103 may include: Step 1031: Obtain the first-level adjacency relationship features corresponding to the target area based on the first geographical adjacency relationship data.
[0065] In this application, the first-level adjacency characteristics can be recorded in a matrix to obtain a first-level adjacency characteristic matrix. In the first-level adjacency characteristic matrix, the value of each element indicates whether two different first-level administrative objects are geographically adjacent. Figure 2 This is a schematic diagram illustrating a first-level adjacency feature matrix according to an embodiment of this application. For example... Figure 2 As shown, 1 indicates that two first-level administrative objects are geographically adjacent, and 0 indicates that two first-level administrative objects are not geographically adjacent. First-level administrative objects include objects 1 to 7.
[0066] Step 1032: Obtain the secondary adjacency relationship features corresponding to the target area based on the second geographical adjacency relationship data.
[0067] In this application, the features of second-level adjacency relationships can be recorded in a matrix to obtain a second-level adjacency relationship feature matrix. In the second-level adjacency relationship feature matrix, the value of each element represents the relationship between two different second-level administrative objects in multiple preset dimensions.
[0068] In one implementation, if two secondary administrative objects satisfy any one of the preset dimensions, then the relationship between them can be represented by 1 in the secondary adjacency feature matrix. For example, when the preset dimensions include direct adjacency, marginal adjacency, and regional association, if two secondary administrative objects satisfy any one of these relationships, then the relationship between them can be represented by 1. Conversely, if two secondary administrative objects do not satisfy any one of the preset dimensions, then the relationship between them can be represented by 0 in the district / county-level adjacency feature matrix.
[0069] In another implementation, each preset dimension can correspond to a value. If two secondary administrative objects satisfy at least one of the preset dimensions, then the relationship between them can be represented in the secondary adjacency feature matrix by the sum of the values corresponding to all the preset dimensions. For example, if the preset dimensions include direct adjacency (value 1), marginal adjacency (value 1), and regional association (value 0.5), then if two secondary administrative objects satisfy both marginal adjacency and regional association, the relationship between them can be represented by 1.5. Conversely, if two secondary administrative objects do not satisfy any of the preset dimensions, then the relationship between them can be represented by 0 in the secondary adjacency feature matrix.
[0070] In this application, the relationships between all secondary administrative objects in the target area on multiple preset dimensions can be statistically analyzed to obtain a total secondary adjacency relationship feature matrix. That is, the dimension of the secondary adjacency relationship feature matrix can be M*M, where M is the number of all secondary administrative objects in the target area.
[0071] Step 1033: Enhance the first-level adjacency relationship feature based on the second-level adjacency relationship feature to obtain the first adjacency relationship feature.
[0072] Step 1033 may include: The second-level adjacency relationship features are aggregated to obtain aggregated second-level adjacency relationship features; The aggregated second-level adjacency features are fused with the first-level adjacency features to obtain the first adjacency feature.
[0073] In this application, the aggregation operation can be either averaging or maximizing, and the specific value can be set according to actual needs. This application does not impose specific restrictions on the type of aggregation operation.
[0074] In this application, the first-level adjacency features are enhanced by using the aggregated second-level adjacency features to obtain enhanced first-level adjacency features. The first adjacency feature is the enhanced first-level adjacency feature. In this way, the representation of the first-level adjacency features can be enriched, thereby improving the accuracy of the prediction results.
[0075] Figure 3 This is a schematic diagram illustrating the implementation process of a regional carbon emission prediction method according to an embodiment of this application. The following will be combined with... Figure 3 The regional carbon emission prediction method of this application will be described in detail with a complete embodiment. This embodiment includes the following steps: Step 1: Obtain the primary carbon emission characteristic matrix based on the first historical carbon emission data, represented as follows: ; and obtain the first-level adjacency feature matrix corresponding to the target area based on the first geographic adjacency data. .
[0076] Step 2: Obtain the secondary carbon emission characteristic matrix based on the second historical carbon emission data. ,use right Enhancement will be carried out, specifically through This results in the enhanced first-order carbon emission characteristic matrix. , This is the first carbon emission feature matrix, where the plus sign indicates a feature fusion operation. Simultaneously, based on the second geographic adjacency relationship data, the second-level adjacency relationship feature matrix corresponding to the target region is obtained. ,use right Enhancement, through This is achieved by obtaining the enhanced second-level adjacency feature matrix. , That is, the first adjacency feature matrix, where the plus sign indicates the feature fusion operation. This is an aggregate function. (Enhanced) It contains more detailed information on geographical relationships and spatial dependencies, which can enhance the hierarchy and diversity of spatial feature expression.
[0077] Step 3: Using the ensemble empirical mode decomposition method, feature decomposition is performed on the first and second historical carbon emission data to obtain the component features of each data at multiple different time scales; these component features are then compared with... Feature fusion is performed to obtain the feature matrix X, and at the same time... As the characteristic matrix A.
[0078] Step 4: Input the feature matrix A, feature matrix X, and prediction time period into the preset graph convolutional network. The graph convolutional network will eventually output the carbon emission prediction result Y of each first-level administrative object in the target area during the prediction time period.
[0079] In Step 3, the ensemble empirical mode decomposition method is used to decompose the carbon emission time series data into multiple scales. By extracting trend components, fluctuation components and noise components, the complex time series features are separated into feature representations at different time scales. This decomposition method can better capture the long-term changes and short-term fluctuations of carbon emission data, significantly enhance the expressive power of local time series features, and enhance the ability to identify fine-grained changes while capturing time dependencies.
[0080] In Step 4, only the two input matrices of the graph convolutional network (feature matrix A and feature matrix X) were simply integrated without complex fusion processing. This provides basic data for subsequent graph convolutional network modeling, ensuring the effective transfer of spatiotemporal features. In graph convolutional networks, high-order convolutional structures are used to capture spatial dependencies and regional interactions, effectively learning complex relationships between regions and extracting richer spatial features, further enhancing carbon emission prediction capabilities.
[0081] Specifically, complex feature fusion strategies in existing technologies refer to the deep fusion of multiple heterogeneous data. Common complex feature fusion strategies include shared latent space mapping (mapping data from different modalities to the same high-dimensional space), attention-based weighted fusion (assigning different weights to different features), and feature extraction through multi-layer neural networks. These methods can fully explore the deep relationships between data from different sources, but they also introduce additional computational complexity and redundant information, resulting in long model training times and reduced interpretability.
[0082] This application employs a simple integration strategy. Firstly, it directly concatenates the temporal features after EEMD decomposition without considering temporal weight relationships. That is, in step 105, when fusing multiple component features with the first carbon emission feature, only simple feature concatenation is performed to obtain the second carbon emission feature. Secondly, in the process of enhancing the first historical carbon emission data using the second historical carbon emission data, and enhancing the first geographical adjacency data using the second geographical adjacency data, the feature fusion operation is also only a simple feature concatenation operation, without complex feature fusion. Thirdly, in step 106, when inputting the second carbon emission feature and the first adjacency feature into the preset graph convolutional network, only simple aggregation and organization of the second carbon emission feature and the first adjacency feature is performed, without complex joint feature encoding. By adopting the above simple integration strategy, this application can significantly reduce computational complexity and improve training efficiency while maintaining information representation capabilities.
[0083] The regional carbon emission prediction device provided in this application is described below. The regional carbon emission prediction device described below can be referred to in correspondence with the regional carbon emission prediction method described above.
[0084] This application also provides a regional carbon emission prediction device 400, such as Figure 4 As shown. Figure 4 This is a structural block diagram of a regional carbon emission prediction device according to an embodiment of this application. (Refer to...) Figure 4 The regional carbon emission prediction device 400 of this application may include: The first acquisition module 401 is used to acquire the first historical carbon emission data corresponding to the first-level administrative objects in the target area, the first geographical adjacency data between each of the first-level administrative objects, the second historical carbon emission data corresponding to the second-level administrative objects under each of the first-level administrative objects, the second geographical adjacency data between each of the second-level administrative objects, and the prediction period, wherein the carbon emission data is time series data. The second acquisition module 402 is used to obtain the first carbon emission characteristics corresponding to the target area based on the first historical carbon emission data and the second historical carbon emission data. The third acquisition module 403 is used to obtain the first adjacency relationship feature corresponding to the target area based on the first geographic adjacency relationship data and the second geographic adjacency relationship data; The feature decomposition module 404 is used to perform feature decomposition on the first historical carbon emission data and the second historical carbon emission data to obtain the component features of the first historical carbon emission data and the second historical carbon emission data at multiple different time scales. The feature fusion module 405 is used to fuse multiple component features with the first carbon emission feature to obtain a second carbon emission feature. The fourth acquisition module 406 is used to input the prediction period, the second carbon emission feature, and the first adjacency relationship feature into a preset graph convolutional network to obtain the carbon emission prediction results of each of the first-level administrative objects in the target area during the prediction period.
[0085] According to the regional carbon emission prediction device 400 provided in this application, the second acquisition module 402 includes: The first acquisition submodule is used to obtain the primary carbon emission characteristics corresponding to the target area based on the first historical carbon emission data. The second acquisition submodule is used to obtain the secondary carbon emission characteristics corresponding to each primary administrative object based on the historical carbon emission data corresponding to the secondary administrative objects under each primary administrative object in the second historical carbon emission data. The first feature enhancement submodule is used to enhance the primary carbon emission feature based on the secondary carbon emission feature to obtain the first carbon emission feature corresponding to the target region.
[0086] According to the regional carbon emission prediction device 400 provided in this application, the first feature enhancement submodule includes: The determination submodule is used to determine the mean value of the corresponding secondary carbon emission characteristics for each primary administrative object in the target area; The first feature fusion submodule is used to fuse the mean value corresponding to each of the first-level administrative objects with the feature corresponding to the first-level administrative object in the first-level carbon emission feature to obtain the first carbon emission feature.
[0087] According to the regional carbon emission prediction device 400 provided in this application, the third acquisition module 403 includes: The third acquisition submodule is used to obtain the first-level adjacency relationship features corresponding to the target area based on the first geographical adjacency relationship data; The fourth acquisition submodule is used to obtain the secondary adjacency relationship features corresponding to the target area based on the second geographical adjacency relationship data; The second feature enhancement submodule is used to enhance the first-level adjacency feature based on the second-level adjacency feature to obtain the first adjacency feature.
[0088] According to the regional carbon emission prediction device 400 provided in this application, the second feature enhancement submodule includes: The aggregation submodule is used to aggregate the second-level adjacency relationship features to obtain aggregated second-level adjacency relationship features; The second feature fusion submodule is used to fuse the aggregated second-level adjacency relationship features with the first-level adjacency relationship features to obtain the first adjacency relationship feature.
[0089] According to the regional carbon emission prediction device 400 provided in this application, the feature decomposition module 404 includes: The first decomposition submodule is used to perform feature decomposition on the first historical carbon emission data by using the ensemble empirical mode decomposition method to obtain at least one first intrinsic mode function component and a first residual component, and to obtain the component features of the first historical carbon emission data at multiple different time scales based on the at least one first intrinsic mode function component and the first residual component. The second decomposition submodule is used to perform feature decomposition on the second historical carbon emission data using the ensemble empirical mode decomposition method to obtain at least one second intrinsic mode function component and a second residual component, and to obtain the component features of the second historical carbon emission data at multiple different time scales based on the at least one second intrinsic mode function component and the second residual component.
[0090] According to the regional carbon emission prediction device 400 provided in this application, the graph convolutional network includes a fully connected layer and a preset number of graph convolutional layers; the fourth acquisition module 406 includes: The feature extraction submodule is used to extract features from the second carbon emission feature and the first adjacency relationship feature through the preset number of graph convolutional layers to obtain convolutional features; The prediction submodule is used to process the convolutional features and the prediction time period through the fully connected layer to obtain the carbon emission prediction results of each of the first-level administrative objects in the target area during the prediction time period; In this process, each graph convolutional layer is used for feature extraction using the following formulas (1)-(3), and the carbon emission prediction result is expressed using the following formula (4): (1) (2) (3) (4) in, , The matrix representing the second carbon emission characteristic, The matrix representing the characteristics of the first adjacency relationship. It is the identity matrix. For the first The input feature matrix of the layer, For degree matrix, For the first The learnable parameter matrix of the layer, For activation function, For the preset quantity, The weights of the fully connected layer, This is the bias for the fully connected layer.
[0091] Figure 5 This is a schematic diagram of the physical structure of an electronic device according to an embodiment of this application, as shown below. Figure 5 As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a regional carbon emission prediction method, which includes: The system acquires the first historical carbon emission data corresponding to the first-level administrative objects in the target area, the first geographical adjacency data between each of the first-level administrative objects, the second historical carbon emission data corresponding to the second-level administrative objects under each of the first-level administrative objects, the second geographical adjacency data between each of the second-level administrative objects, and the prediction period. The carbon emission data is time-series data. Based on the first historical carbon emission data and the second historical carbon emission data, the first carbon emission characteristics corresponding to the target area are obtained; Based on the first geographical adjacency data and the second geographical adjacency data, the first adjacency feature corresponding to the target area is obtained; Feature decomposition is performed on the first historical carbon emission data and the second historical carbon emission data to obtain the component features of the first historical carbon emission data and the second historical carbon emission data at multiple different time scales. The second carbon emission feature is obtained by fusing multiple component features with the first carbon emission feature. The predicted time period, the second carbon emission feature, and the first adjacency relationship feature are input into a preset graph convolutional network to obtain the carbon emission prediction results of each of the first-level administrative objects in the target area during the predicted time period.
[0092] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0093] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute a regional carbon emission prediction method provided by the above methods, the method comprising: The system acquires the first historical carbon emission data corresponding to the first-level administrative objects in the target area, the first geographical adjacency data between each of the first-level administrative objects, the second historical carbon emission data corresponding to the second-level administrative objects under each of the first-level administrative objects, the second geographical adjacency data between each of the second-level administrative objects, and the prediction period. The carbon emission data is time-series data. Based on the first historical carbon emission data and the second historical carbon emission data, the first carbon emission characteristics corresponding to the target area are obtained; Based on the first geographical adjacency data and the second geographical adjacency data, the first adjacency feature corresponding to the target area is obtained; Feature decomposition is performed on the first historical carbon emission data and the second historical carbon emission data to obtain the component features of the first historical carbon emission data and the second historical carbon emission data at multiple different time scales. The second carbon emission feature is obtained by fusing multiple component features with the first carbon emission feature. The predicted time period, the second carbon emission feature, and the first adjacency relationship feature are input into a preset graph convolutional network to obtain the carbon emission prediction results of each of the first-level administrative objects in the target area during the predicted time period.
[0094] In another aspect, this application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform a regional carbon emission prediction method provided by the methods described above, the method comprising: The system acquires the first historical carbon emission data corresponding to the first-level administrative objects in the target area, the first geographical adjacency data between each of the first-level administrative objects, the second historical carbon emission data corresponding to the second-level administrative objects under each of the first-level administrative objects, the second geographical adjacency data between each of the second-level administrative objects, and the prediction period. The carbon emission data is time-series data. Based on the first historical carbon emission data and the second historical carbon emission data, the first carbon emission characteristics corresponding to the target area are obtained; Based on the first geographical adjacency data and the second geographical adjacency data, the first adjacency feature corresponding to the target area is obtained; Feature decomposition is performed on the first historical carbon emission data and the second historical carbon emission data to obtain the component features of the first historical carbon emission data and the second historical carbon emission data at multiple different time scales. The second carbon emission feature is obtained by fusing multiple component features with the first carbon emission feature. The predicted time period, the second carbon emission feature, and the first adjacency relationship feature are input into a preset graph convolutional network to obtain the carbon emission prediction results of each of the first-level administrative objects in the target area during the predicted time period.
[0095] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0096] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A method for predicting regional carbon emissions, characterized in that, include: The system acquires the first historical carbon emission data corresponding to the first-level administrative objects in the target area, the first geographical adjacency data between the first-level administrative objects, the second historical carbon emission data corresponding to the second-level administrative objects under the first-level administrative objects, the second geographical adjacency data between the second-level administrative objects, and the prediction period, wherein the carbon emission data is time series data. Based on the first historical carbon emission data and the second historical carbon emission data, the first carbon emission characteristics corresponding to the target area are obtained; Based on the first geographical adjacency data and the second geographical adjacency data, the first adjacency feature corresponding to the target area is obtained; Feature decomposition is performed on the first historical carbon emission data and the second historical carbon emission data to obtain the component features of the first historical carbon emission data and the second historical carbon emission data at multiple different time scales. The second carbon emission feature is obtained by fusing multiple component features with the first carbon emission feature. The predicted time period, the second carbon emission feature, and the first adjacency relationship feature are input into a preset graph convolutional network to obtain the carbon emission prediction results of each of the first-level administrative objects in the target area during the predicted time period.
2. The regional carbon emission prediction method according to claim 1, characterized in that, The step of obtaining the first carbon emission characteristic corresponding to the target area based on the first historical carbon emission data and the second historical carbon emission data includes: The primary carbon emission characteristics corresponding to the target area are obtained based on the first historical carbon emission data. Based on the historical carbon emission data of the secondary administrative objects under each primary administrative object in the second historical carbon emission data, the secondary carbon emission characteristics of each primary administrative object are obtained. The primary carbon emission characteristics are enhanced based on the secondary carbon emission characteristics to obtain the first carbon emission characteristics corresponding to the target region.
3. The regional carbon emission prediction method according to claim 2, characterized in that, The step of enhancing the primary carbon emission characteristics based on the secondary carbon emission characteristics to obtain the first carbon emission characteristics corresponding to the target region includes: For each primary administrative entity in the target area, determine the mean value of its corresponding secondary carbon emission characteristics; The first carbon emission feature is obtained by fusing the mean value corresponding to each of the first-level administrative objects with the feature corresponding to the first-level administrative object in the first-level carbon emission feature.
4. The regional carbon emission prediction method according to claim 1, characterized in that, The step of obtaining the first adjacency feature corresponding to the target area based on the first geographic adjacency data and the second geographic adjacency data includes: The first-level adjacency characteristics of the target area are obtained based on the first geographical adjacency data; The secondary adjacency characteristics of the target area are obtained based on the second geographical adjacency data; The first-level adjacency feature is enhanced based on the second-level adjacency feature to obtain the first adjacency feature.
5. The regional carbon emission prediction method according to claim 4, characterized in that, The step of enhancing the first-level adjacency features based on the second-level adjacency features to obtain the first adjacency features includes: The second-level adjacency relationship features are aggregated to obtain aggregated second-level adjacency relationship features; The aggregated second-level adjacency features are fused with the first-level adjacency features to obtain the first adjacency feature.
6. The regional carbon emission prediction method according to claim 1, characterized in that, Feature decomposition is performed on the first historical carbon emission data and the second historical carbon emission data to obtain the component features of the first historical carbon emission data and the second historical carbon emission data at multiple different time scales, including: The first historical carbon emission data is decomposed using the ensemble empirical mode decomposition method to obtain at least one first intrinsic mode function component and a first residual component. The component features of the first historical carbon emission data at multiple different time scales are obtained based on the at least one first intrinsic mode function component and the first residual component. The second historical carbon emission data is decomposed using the ensemble empirical mode decomposition method to obtain at least one second intrinsic mode function component and a second residual component. Based on the at least one second intrinsic mode function component and the second residual component, the component features of the second historical carbon emission data at multiple different time scales are obtained.
7. The regional carbon emission prediction method according to claim 1, characterized in that, The graph convolutional network includes fully connected layers and a preset number of graph convolutional layers; the step of inputting the prediction period, the second carbon emission feature, and the first adjacency relationship feature into the preset graph convolutional network to obtain the carbon emission prediction results of each of the first-level administrative objects in the target area during the prediction period includes: The second carbon emission feature and the first adjacency relationship feature are extracted by the preset number of graph convolutional layers to obtain convolutional features; The convolutional features and the prediction period are processed by the fully connected layer to obtain the carbon emission prediction results of each of the first-level administrative objects in the target area during the prediction period; In this process, each graph convolutional layer is used for feature extraction using the following formulas (1)-(3), and the carbon emission prediction result is expressed using the following formula (4): （1） （2） （3） （4） in, , The matrix representing the second carbon emission characteristic, The matrix representing the characteristics of the first adjacency relationship. It is the identity matrix. For the first The input feature matrix of the layer, For degree matrix, For the first The learnable parameter matrix of the layer, For activation function, For the preset quantity, The weights of the fully connected layer, This is the bias for the fully connected layer.
8. A regional carbon emission prediction device, characterized in that, include: The first acquisition module is used to acquire the first historical carbon emission data corresponding to the first-level administrative objects in the target area, the first geographical adjacency data between each of the first-level administrative objects, the second historical carbon emission data corresponding to the second-level administrative objects under each of the first-level administrative objects, the second geographical adjacency data between each of the second-level administrative objects, and the prediction period, wherein the carbon emission data is time series data. The second acquisition module is used to obtain the first carbon emission characteristics corresponding to the target area based on the first historical carbon emission data and the second historical carbon emission data. The third acquisition module is used to obtain the first adjacency relationship feature corresponding to the target area based on the first geographic adjacency relationship data and the second geographic adjacency relationship data; The feature decomposition module is used to perform feature decomposition on the first historical carbon emission data and the second historical carbon emission data to obtain the component features of the first historical carbon emission data and the second historical carbon emission data at multiple different time scales. A feature fusion module is used to fuse multiple component features with the first carbon emission feature to obtain a second carbon emission feature; The fourth acquisition module is used to input the prediction period, the second carbon emission feature, and the first adjacency relationship feature into a preset graph convolutional network to obtain the carbon emission prediction results of each of the first-level administrative objects in the target area during the prediction period.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements a regional carbon emission prediction method as described in any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a regional carbon emission prediction method as described in any one of claims 1 to 7.