Carbon emission accounting and prediction method and system based on multi-source data fusion

Abstract

The invention discloses a carbon emission accounting and prediction method and system based on multi-source data fusion. The method comprises the following steps: acquiring data of a plurality of accounting indexes under a predetermined accounting boundary within a first predetermined time period; fusing static data and dynamic data in the data to obtain fusion features; inputting the fusion features into an accounting predictive coding module to obtain key features; inputting fusion features corresponding to a second predetermined time period in the fusion features into a decomposition module to obtain respective trend components and periodic components of the plurality of accounting indexes; obtaining an input feature based on the trend component, the periodic component, a preset prediction time length and respective data mean values of the plurality of accounting indexes; inputting the input features and the key features into an accounting prediction decoding module to obtain carbon emission features; and inputting the carbon emission characteristics into an accounting prediction module to obtain the carbon emission in the second predetermined time period and the predicted carbon emission in the predicted time length.

Description

Technical Field

[0001] The present disclosure generally relates to the field of artificial intelligence, and more specifically, to a carbon emission accounting and prediction method and system based on multi-source data fusion. Background Art

[0002] In recent years, with the increasingly severe global climate change problem, carbon emission research has become the focus of attention of the scientific community and policymakers. Carbon emissions are not only one of the main driving factors of global warming, but also have an important impact on the ecosystem, public health and the global economy. Excessive greenhouse gas emissions have brought profound challenges to human society and the natural environment. Therefore, by accounting for and predicting carbon emissions, it is not only possible to help researchers comprehensively understand the current situation of carbon emissions, but also to provide key data support for the overall planning and policy formulation of countries in response to climate change and the implementation of environmental governance.

[0003] Currently, for the research on carbon emissions, the accounting and prediction are usually carried out in two parts. The first step is carbon emission accounting, the core of which is to clarify the sources of carbon emissions and quantify the specific emissions. Accounting first needs to determine the accounting boundary; after determining the accounting boundary, it is necessary to analyze the emission sources within the accounting boundary that may cause carbon dioxide emissions, and then use the emission factor method to calculate the carbon emission data. Although this method is intuitive and easy to apply, it has a high dependence on the integrity and accuracy of the data. If there are errors or omissions in the data used in the accounting process, it may lead to large deviations in the accounting results, thus affecting subsequent decision-making and analysis. The second step is carbon emission prediction, the purpose of which is to analyze and predict the future emission trend based on historical carbon emission data on the basis of accounting. Currently, carbon emission prediction is usually completed by using traditional statistical models (such as time series analysis, regression models) or machine learning models (such as support vector machines, neural networks, etc.). However, these models are often difficult to provide high-precision prediction results in a complex and changing environment. Especially when the relationship between variables is complex or the external environment changes greatly, the prediction accuracy of existing models may decrease significantly; in addition, when dealing with long-term predictions, these methods usually need to rely on a large number of additional assumptions, which further reduces the reliability of the results. Generally speaking, although the current carbon emission accounting and prediction methods have constructed a relatively systematic framework, there is still obvious room for improvement in terms of data dependence, model applicability and prediction accuracy, which indicates that more accurate carbon emission accounting and prediction models need to be developed to support the realization of low-carbon transformation and sustainable development goals. Summary of the Invention

[0004] Embodiments of the present disclosure provide a carbon emission accounting and prediction method and system based on multi-source data fusion, which can effectively solve the problem of inaccurate prediction results in the prior art.

[0005] In one general aspect, a carbon emission accounting and prediction method based on multi-source data fusion is provided, including: obtaining data of multiple accounting indicators within a first predetermined time period under a predetermined accounting boundary; fusing static data and dynamic data in the data to obtain fusion features; inputting the fusion features into an accounting prediction coding module to obtain key features, where the key features indicate periodic information of the data; inputting the fusion features corresponding to a second predetermined time period in the fusion features into a decomposition module to obtain a trend component and a periodic component of each of the multiple accounting indicators, where the second predetermined time period is a partial time period within the first predetermined time period and the start time of the second predetermined time period is later than the start time of the first predetermined time period; obtaining input features based on the trend component, the periodic component, a preset prediction time length, and the data mean of each of the multiple accounting indicators; inputting the input features and the key features into an accounting prediction decoding module to obtain carbon emission features; and inputting the carbon emission features into an accounting prediction module to obtain the carbon emissions in the second predetermined time period and the predicted carbon emissions for the prediction time length.

[0006] Optionally, fusing static data and dynamic data in the data to obtain fusion features includes: inputting the static data in the data into a pre-trained static coding module to obtain static features; and fusing the static features and the dynamic data in the data to obtain fusion features.

[0007] Optionally, the static coding module includes a first fully connected network, a second fully connected network, and an activation function, where inputting the static data in the data into a pre-trained static coding module to obtain static features includes: inputting the static data into the first fully connected network to obtain a first feature; inputting the first feature into the activation function to obtain a second feature; and inputting the second feature and the static data into the second fully connected network to obtain static features.

[0008] Optionally, obtaining input features based on the trend component, the periodic component, a preset prediction time length, and the data mean of each of the multiple accounting indicators includes: for each of the multiple accounting indicators, performing the following processing: filling a preset value of the prediction time length into the periodic component of the current accounting indicator to obtain a filled periodic feature; filling the data mean of the current accounting indicator for the prediction time length into the trend component of the current accounting indicator to obtain a filled trend feature; and determining the periodic feature and the trend feature as the input features.

[0009] Optionally, the accounting prediction coding module includes multiple layers of encoders. Each layer of encoder includes an autocorrelation module, a first variational mode decomposition module, a feedforward module, and a second variational mode decomposition module. Among them, the fused features are input into the accounting prediction coding module to obtain key features, including: through each layer of encoder in the multiple layers of encoders, the following processing is performed: the fused features are input into the autocorrelation module of the current layer of encoder to obtain periodic information; the periodic information and the fused features are input into the first variational mode decomposition module of the current layer of encoder to obtain the first periodic feature and the first trend feature; the first periodic feature is input into the feedforward module of the current layer of encoder to obtain feedforward information; the feedforward information and the first periodic feature are input into the second variational mode decomposition module of the current layer of encoder to obtain the second periodic feature and the second trend feature; the second periodic feature is determined as the input of the autocorrelation module of the next layer of encoder; the second periodic feature output by the last layer of encoder is determined as the key feature.

[0010] Optionally, the accounting prediction decoding module includes multiple layers of decoders. Each layer of decoder includes a first autocorrelation module, a third variational mode decomposition module, a second autocorrelation module, a fourth variational mode decomposition module, a feedforward module, and a fifth variational mode decomposition module. Among them, the input features and the key features are input into the accounting prediction decoding module to obtain carbon emission features, including: through each layer of decoder in the multiple layers of decoders, the following processing is performed: the periodic features in the input features are input into the first autocorrelation module of the current layer of decoder to obtain the first periodic information; the first periodic information and the periodic features in the input features are input into the third variational mode decomposition module of the current layer of decoder to obtain the third periodic feature and the third trend feature; the third periodic feature and the key features are input into the second autocorrelation module of the current layer of decoder to obtain the second periodic information; the second periodic information and the third periodic feature are input into the fourth variational mode decomposition module of the current layer of decoder to obtain the fourth periodic feature and the fourth trend feature; the fourth periodic feature is input into the feedforward module of the current layer of decoder to obtain feedforward information; the feedforward information and the fourth periodic feature are input into the fifth variational mode decomposition module of the current layer of decoder to obtain the fifth periodic feature and the fifth trend feature; the fifth periodic feature is determined as the input of the first autocorrelation module of the next layer of decoder; the fifth periodic feature output by the last layer of decoder is determined as the final periodic feature; the trend features in the input features and the third trend feature, the fourth trend feature, and the fifth trend feature of all layers of decoders are weighted and summed to obtain the final trend feature; the final periodic feature and the final trend feature are determined as the carbon emission features.

[0011] In another general aspect, a carbon emission accounting and prediction system based on multi-source data fusion is provided, including: a data selection unit configured to obtain data of multiple accounting indicators within a first predetermined time period under a predetermined accounting boundary; a fusion unit configured to fuse static data and dynamic data in the data to obtain fusion features; an encoding unit configured to input the fusion features into an accounting prediction encoding module to obtain key features, where the key features indicate periodic information of the data; a decomposition unit configured to input the fusion features corresponding to a second predetermined time period in the fusion features into a decomposition module to obtain a trend component and a periodic component of each of the multiple accounting indicators, where the second predetermined time period is a partial time period within the first predetermined time period and the start time of the second predetermined time period is later than the start time of the first predetermined time period; a filling unit configured to obtain input features based on the trend component, the periodic component, a preset prediction time length, and the data mean of each of the multiple accounting indicators; a decoding unit configured to input the input features and the key features into an accounting prediction decoding module to obtain carbon emission features; and a prediction unit configured to input the carbon emission features into an accounting prediction module to obtain the carbon emissions for the second predetermined time period and the predicted carbon emissions for the prediction time length.

[0012] Optionally, the fusion unit is further configured to input the static data in the data into a pre-trained static encoding module to obtain static features; and fuse the static features and the dynamic data in the data to obtain fusion features.

[0013] Optionally, the static encoding module includes a first fully connected network, a second fully connected network, and an activation function, where the fusion unit is further configured to input the static data into the first fully connected network to obtain a first feature; input the first feature into the activation function to obtain a second feature; and input the second feature and the static data into the second fully connected network to obtain static features.

[0014] Optionally, the filling unit is further configured to, for each of the multiple accounting indicators, perform the following processing: fill a preset value for the prediction time length into the periodic component of the current accounting indicator to obtain a filled periodic feature; fill the data mean of the current accounting indicator for the prediction time length into the trend component of the current accounting indicator to obtain a filled trend feature; and determine the periodic feature and the trend feature as the input features.

[0015] Optionally, the accounting prediction encoding module includes multiple layers of encoders. Each layer of encoder includes an autocorrelation module, a first variational mode decomposition module, a feed-forward module, and a second variational mode decomposition module. Among them, the encoding unit is further configured to perform the following processing through each layer of encoder in the multiple layers of encoders: input the fusion feature into the autocorrelation module of the current layer of encoder to obtain periodic information; input the periodic information and the fusion feature into the first variational mode decomposition module of the current layer of encoder to obtain a first periodic feature and a first trend feature; input the first periodic feature into the feed-forward module of the current layer of encoder to obtain feed-forward information; input the feed-forward information and the first periodic feature into the second variational mode decomposition module of the current layer of encoder to obtain a second periodic feature and a second trend feature; determine the second periodic feature as the input of the autocorrelation module of the next layer of encoder; determine the second periodic feature output by the last layer of encoder as the key feature.

[0016] Optionally, the accounting prediction decoding module includes multiple layers of decoders. Each layer of decoder includes a first autocorrelation module, a third variational mode decomposition module, a second autocorrelation module, a fourth variational mode decomposition module, a feed-forward module, and a fifth variational mode decomposition module. Among them, the decoding unit is further configured to perform the following processing through each layer of decoder in the multiple layers of decoders: input the periodic feature in the input feature into the first autocorrelation module of the current layer of decoder to obtain first periodic information; input the first periodic information and the periodic feature in the input feature into the third variational mode decomposition module of the current layer of decoder to obtain a third periodic feature and a third trend feature; input the third periodic feature and the key feature into the second autocorrelation module of the current layer of decoder to obtain second periodic information; input the second periodic information and the third periodic feature into the fourth variational mode decomposition module of the current layer of decoder to obtain a fourth periodic feature and a fourth trend feature; input the fourth periodic feature into the feed-forward module of the current layer of decoder to obtain feed-forward information; input the feed-forward information and the fourth periodic feature into the fifth variational mode decomposition module of the current layer of decoder to obtain a fifth periodic feature and a fifth trend feature; determine the fifth periodic feature as the input of the first autocorrelation module of the next layer of decoder; determine the fifth periodic feature output by the last layer of decoder as the final periodic feature; perform a weighted sum of the trend feature in the input feature and the third trend feature, the fourth trend feature, and the fifth trend feature of all layers of decoders to obtain the final trend feature; determine the final periodic feature and the final trend feature as the carbon emission feature.

[0017] In another general aspect, a computer-readable storage medium storing instructions is provided. When the instructions are run by at least one computing device, at least one computing device is prompted to execute the carbon emission accounting and prediction method based on multi-source data fusion as described in any one of the above.

[0018] In another general aspect, a system is provided that includes at least one computing device and at least one storage device storing instructions, where, when the instructions are run by the at least one computing device, the at least one computing device is prompted to execute the carbon emission accounting and prediction method based on multi-source data fusion as described in any of the above.

[0019] In another general aspect, a computer program product is provided that includes computer instructions which, when executed by a processor, implement the carbon emission accounting and prediction method based on multi-source data fusion as described in any of the above.

[0020] According to the carbon emission accounting and prediction method and system based on multi-source data fusion of the embodiments of the present disclosure, in the data selection stage, the accuracy of accounting and prediction is improved by selecting data of multiple accounting indicators, and the accounting prediction coding module of the present disclosure can mine periodic rule information in historical data and transmit it to the accounting prediction decoding module to optimize the prediction effect. Moreover, the accounting prediction module of the present disclosure realizes the dual functions of accounting and prediction through the carbon emission characteristics obtained previously. Therefore, through the present disclosure, the problem of inaccurate prediction results in the prior art can be effectively solved.

[0021] Additional aspects and / or advantages of the general concept of the present disclosure will be partly set forth in the description that follows, and partly will be obvious from the description, or may be learned by practice of the general concept of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Through the following description with reference to the drawings showing embodiments, the above and other objects and features of the embodiments of the present disclosure will become more apparent, where: Figure 1 is a flowchart showing the carbon emission accounting and prediction method based on multi-source data fusion of the embodiments of the present disclosure; Figure 2 is an architecture diagram showing the carbon emission accounting and prediction model of the embodiments of the present disclosure; Figure 3 is a flowchart showing the carbon emission accounting and prediction method system of the embodiments of the present disclosure; Figure 4 is a block diagram showing the carbon emission accounting and prediction system based on multi-source data fusion of the embodiments of the present disclosure. DETAILED DESCRIPTION

[0023] The following specific embodiments are provided to assist the reader in obtaining a comprehensive understanding of the methods, apparatuses, and / or systems described herein. However, after understanding the disclosure of the present application, various changes, modifications, and equivalents of the methods, apparatuses, and / or systems described herein will be apparent. For example, the order of operations described herein is merely exemplary and is not limited to those set forth herein, but may be changed as will be apparent after understanding the disclosure of the present application, except for operations that must occur in a specific order. Additionally, descriptions of features known in the art may be omitted for greater clarity and conciseness.

[0024] The features described herein may be implemented in different forms and should not be construed as limited to the examples described herein. Instead, the examples described herein are provided only to illustrate some of the many possible ways of implementing the methods, apparatuses, and / or systems described herein, which will be apparent after understanding the disclosure of the present application.

[0025] As used herein, the term "and / or" includes any one of the associated listed items and any combination of any two or more thereof.

[0026] Although terms such as "first", "second", and "third" may be used herein to describe various components, elements, regions, layers, or parts, these components, elements, regions, layers, or parts should not be limited by these terms. Instead, these terms are only used to distinguish one component, element, region, layer, or part from another. Thus, a first component, first element, first region, first layer, or first part referred to in the examples described herein may also be referred to as a second component, second element, second region, second layer, or second part without departing from the teachings of the examples.

[0027] In the specification, when an element (such as a layer, region, or substrate) is described as "on", "connected to", or "coupled to" another element, the element may be directly "on", directly "connected to", or "coupled to" the other element, or there may be one or more other elements therebetween. In contrast, when an element is described as "directly on", "directly connected to", or "directly coupled to" another element, there may be no other elements therebetween.

[0028] The terms used herein are only for describing various examples and are not intended to limit the disclosure. Unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. The terms "comprising", "including", and "having" specify the presence of the recited features, quantities, operations, components, elements, and / or combinations thereof, but do not preclude the presence or addition of one or more other features, quantities, operations, components, elements, and / or combinations thereof.

[0029] Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains after understanding this disclosure. Unless explicitly defined herein, terms (such as those defined in a general dictionary) shall be construed to have a meaning consistent with their meaning in the context of the relevant art and this disclosure, and shall not be interpreted in an idealized or overly formal manner.

[0030] In addition, in the description of the examples, when it is considered that a detailed description of related structures or functions that are well-known will cause an ambiguous interpretation of this disclosure, such detailed descriptions will be omitted.

[0031] The following describes in detail a carbon emission accounting and prediction method and system based on multi-source data fusion of this disclosure with reference to the accompanying drawings.

[0032] This disclosure proposes a carbon emission accounting and prediction method. Figure 1 is a flowchart showing an embodiment of the carbon emission accounting and prediction method based on multi-source data fusion of this disclosure. Refer to Figure 1 , the carbon emission accounting and prediction method based on multi-source data fusion includes the following steps: In step S101, data of multiple accounting indicators within a first predetermined time period under a predetermined accounting boundary is obtained.

[0033] As an example, the above-mentioned predetermined accounting boundary may be a country, region, industry, enterprise, or specific facility and activity, and this disclosure does not limit this. Accounting indicators need to be determined under a specific accounting boundary, that is, different accounting boundaries may include different accounting indicators, and this disclosure does not limit this.

[0034] As an example, taking the ceramic industry as the predetermined accounting boundary and the first predetermined time period as the years 2013 - 2022, relying on the carbon emission accounting reports of more than 100 enterprises in the ceramic industry of a certain province collected from 2013 to 2022, analyze the influencing factors of carbon emissions, and construct a multi-source dataset. This multi-source dataset contains data of multiple accounting indicators in the ceramic industry from 2013 to 2022. Specifically, the primary factors affecting carbon emissions in the ceramic industry are fuel consumption and electricity consumption during the production process of ceramic products. Secondly, some ceramic products will generate carbon dioxide during the production process from raw materials. In addition, different production processes and products will also affect carbon dioxide emissions. All of the above are direct factors affecting carbon emission changes. Furthermore, the industrial structure, population size, energy consumption structure, and technological development of enterprises will also indirectly affect carbon emissions, which are also relatively important factors in the carbon emission prediction process. Therefore, the dataset construction can be selected and constructed based on the above accounting indicators. It should be noted that the above fuel consumption, electricity consumption, carbon dioxide generated from raw materials, the impact of different processes on carbon dioxide, the industrial structure of enterprises, population size, energy consumption structure, etc. all belong to the accounting indicators of the ceramic industry, and other accounting indicators may also be included. This disclosure does not limit this.

[0035] In step S102, fuse the static data and dynamic data in the data to obtain fused features.

[0036] As an example, still taking the ceramic industry as the predetermined accounting boundary, data such as fuel consumption, electricity consumption, and the amount of carbon dioxide generated from raw materials are data that change over time, so they can be regarded as dynamic data. While data corresponding to production techniques, the industrial structure of enterprises, energy consumption structure, etc. are not data that change in real time with time. Therefore, they can be regarded as static data.

[0037] As an example, at the level of selecting emission sources, the monthly consumption data of bituminous coal, petroleum, natural gas, electricity, barium carbonate, and calcium rice can be selected as dynamic emission source data, that is, dynamic data. At the same time, select the location, company size, production scale, number of employees, and operating products as static influencing factors, that is, static data.

[0038] According to the embodiments of the present disclosure, fusing the static data and dynamic data in the data to obtain fused features may include: inputting the static data in the data into a pre-trained static encoding module to obtain static features; fusing the static features and the dynamic data in the data to obtain fused features. According to this embodiment, through the learnable static encoding module, the feature information of the static data can be extracted for subsequent prediction, thereby improving the prediction accuracy and the expression ability of the model.

[0039] As an example, traditional time series prediction processes often ignore the impact of static data on the prediction results. To address this issue, in this embodiment, a learnable static encoding module is used to extract the feature information of the static data. The feature information output by the static encoding module is concatenated with the dynamic data to obtain the final fused feature. Subsequent prediction using this fused feature can improve the prediction accuracy and the expressive power of the model.

[0040] According to an embodiment of the present disclosure, the static encoding module includes a first fully connected network, a second fully connected network, and an activation function. Among them, inputting the static data in the data into the pre-trained static encoding module to obtain static features may include: inputting the static data into the first fully connected network to obtain a first feature; inputting the first feature into the activation function to obtain a second feature; inputting the second feature and the static data into the second fully connected network to obtain static features. Through the structure of the static encoding module in this embodiment, the expressive power in the negative value range can be enhanced while maintaining stable gradients.

[0041] As an example, the above static encoding module can be implemented using a two-layer fully connected network (Dense) and use ELU as the activation function to enhance the expressive power of the overall prediction model in the negative value range while maintaining stable gradients. The output of the static encoding module is a fixed-dimensional embedding vector representing the features of the static data. Specifically, assuming that the static data is transformed into a vector representation, such as the static data is represented as a vector , where n represents the dimension of the static data, the overall encoding process of the static encoding module can be represented as follows:

[0042] where W1, b1 represent the network parameters of the first fully connected network, and W2, b2 represent the network parameters of the second fully connected network, and ELU() represents the activation function.

[0043] After the above static data is encoded by the static encoding module, it can be fused with the dynamic data to obtain the final fused data. Assuming that the dynamic data can also be converted into a vector representation, such as the dynamic data is represented as a vector , where L represents the length of the dynamic data and m is the dimension of the dynamic data (i.e., the types of accounting indicators), the fused data can be expressed by the following formula:

[0044] It should be noted that the fully connected network of the above static encoding module is not limited to 2 layers and can also be other numbers of layers, and the present disclosure does not limit this.

[0045] In step S103, the fused feature is input into the accounting prediction coding module to obtain a key feature, where the key feature indicates the periodic information of the data.

[0046] According to an embodiment of the present disclosure, the accounting prediction coding module includes multiple layers of encoders. Each layer of encoder includes an autocorrelation module, a first variational mode decomposition module, a feed-forward module, and a second variational mode decomposition module. Among them, inputting the fused feature into the accounting prediction coding module to obtain the key feature may include: performing the following processing through each layer of encoder in the multiple layers of encoders: inputting the fused feature into the autocorrelation module of the current layer of encoder to obtain periodic information; inputting the periodic information and the fused feature into the first variational mode decomposition module of the current layer of encoder to obtain a first periodic feature and a first trend feature; inputting the first periodic feature into the feed-forward module of the current layer of encoder to obtain feed-forward information; inputting the feed-forward information and the first periodic feature into the second variational mode decomposition module of the current layer of encoder to obtain a second periodic feature and a second trend feature; determining the second periodic feature as the input of the autocorrelation module of the next layer of encoder; and determining the second periodic feature output by the last layer of encoder as the key feature.

[0047] Through this embodiment, a variational mode decomposition (VMD) module is used to replace the traditional decomposition module based on moving average to solve the problem that the simple moving average method cannot handle the sudden fluctuations of trend features and periodic features. Moreover, this embodiment uses multiple layers of encoders, which can extract relatively pure periodic features.

[0048] As an example, the above-obtained fused feature is input into the accounting prediction coding module to extract the periodic feature of the data, and the extracted periodic feature is used as cross information to help the accounting prediction decoding module improve the prediction result.

[0049] As an example, the above accounting prediction coding module may be composed of N layers of encoders. Each layer of encoder may include an autocorrelation (Auto-Correlation) module, two VMD modules (a first VMD module and a second VMD module), and a feed-forward (Feed-forward) module. Among them, the Auto-Correlation module discovers the period-based dependencies by calculating the sequence autocorrelation and aggregates similar subsequences through time-delay aggregation; the two VMD modules are used to extract the periodic information in the encoding process, and a Feed-forward module is used to deepen the extracted features.

[0050] Specifically, the above Auto-Correlation module is a new attention mechanism mentioned in the Autoformer model, and this mechanism uses the fast Fourier transform to calculate the autocorrelation function. , this function reflects the similarity of time delay, and then based on the selected delay , roll similar sub - processes to the same index position, and through for aggregation; The Auto - Correlation mechanism discovers period - based dependencies by calculating the autocorrelation of a sequence, and aggregates similar subsequences through time - delay aggregation to replace the self - attention mechanism of the original Transformer. The Auto - Correlation calculation process can be shown as follows:

[0051]

[0052] Among them, is used to obtain the autocorrelation parameter, represents the operation of with a time delay . The function is used to generate the probability distribution of the input data. Q, K, and V respectively represent the query vector, key vector, and value vector corresponding to the fused features.

[0053] The VMD module is implemented based on the variational mode decomposition algorithm. This algorithm is a non - recursive signal decomposition method proposed by Dragomiretskiy. The algorithm includes two main steps: constructing a variational problem and solving the variational problem. During the decomposition process, the bandwidth and center frequency of each IMF component can be updated by alternating iteration. Finally, through adaptive decomposition, the VMD algorithm can obtain a series of effective components with limited bandwidth from the original signal. Take the first component as the trend feature of the data, and reconstruct the other components, that is, use the features obtained by superimposing the high - frequency components and intermediate - frequency components respectively as the periodic features. The calculation process of this module can be as follows:

[0054]

[0055] Among them, are the effective components obtained by decomposing through the VMD algorithm, K represents the number of effective components, is the periodic feature, is the trend feature, is used to adjust the ratio of high - frequency components and intermediate - frequency components to reduce the noise in the original data, m represents the number of high - frequency components.

[0056] The feed-forward module adds a non-linear mapping between each layer of the encoder and the decoder to enhance the model's expressive power.

[0057] It should be noted that l the overall equation of the -th layer encoder can be summarized as

[0058]

[0059]

[0060] where represents the periodic features obtained by the l -th layer encoder through the i -th decomposition module. The inputs at the input end of each decomposition module can be connected using residuals. The output by the last layer encoder is the key feature.

[0061] In step S104, the fusion features corresponding to the second predetermined time period in the fusion features are input into the decomposition module to obtain the trend components and periodic components of each of the multiple accounting indicators. Here, the second predetermined time period is a partial time period within the first predetermined time period, and the start time of the second predetermined time period is later than the start time of the first predetermined time period.

[0062] As an example, the above decomposition module can also use a VMD module, and the present disclosure does not limit this. Generally, the above second predetermined time period is the second half of the first predetermined time period, and the present disclosure does not limit this.

[0063] As an example, taking the decomposition module as a VMD module, inputting the fusion features corresponding to the second predetermined time period in the fusion features into the VMD module can output the trend components and periodic components of each of the multiple accounting indicators, as shown in the following formula:

[0064] where L represents the data length, represents the fusion features corresponding to the second predetermined time period in the fusion features, contains the periodic components of each of the multiple accounting indicators, contains the trend components of each of the multiple accounting indicators.

[0065] In step S105, input features are obtained based on the trend components, periodic components, a preset prediction time length, and the data mean values of each of the multiple accounting indicators.

[0066] According to an embodiment of the present disclosure, obtaining input features based on a trend component, a periodic component, a preset prediction time length, and the data mean values of multiple accounting metrics may include: for each of the multiple accounting metrics, performing the following processing: filling the periodic component of the current accounting metric with a preset value for the prediction time length to obtain a filled periodic feature; filling the trend component of the current accounting metric with the data mean value of the current accounting metric for the prediction time length to obtain a filled trend feature; and determining the periodic feature and the trend feature as input features. Through this embodiment, a periodic feature and a trend feature adapted to the prediction time length can be obtained through filling.

[0067] As an example, the periodic component may be filled with "0" based on the prediction time length, and the trend component may be filled with the corresponding data mean value of the corresponding accounting metric based on the prediction time length, as shown in the following formula:

[0068] where includes the data mean values corresponding to the multiple accounting metrics respectively, includes the filled periodic features of the multiple accounting metrics respectively, includes the filled trend features of the multiple accounting metrics respectively.

[0069] Based on and the input features of the accounting prediction decoding module can be obtained. where O is the prediction time length. For example, and can be directly used as input features.

[0070] In step S106, the input features and the key features are input into the accounting prediction decoding module to obtain carbon emission features.

[0071] According to an embodiment of the present disclosure, the accounting prediction decoding module includes multiple layers of decoders. Each layer of decoder includes a first autocorrelation module, a third variational mode decomposition module, a second autocorrelation module, a fourth variational mode decomposition module, a feed-forward module, and a fifth variational mode decomposition module. Wherein, inputting the input features and key features into the accounting prediction decoding module to obtain carbon emission features may include: performing the following processing through each layer of decoder in the multiple layers of decoders: inputting the periodic features in the input features into the first autocorrelation module of the current layer of decoder to obtain the first periodic information; inputting the first periodic information and the periodic features in the input features into the third variational mode decomposition module of the current layer of decoder to obtain the third periodic features and the third trend features; inputting the third periodic features and the key features into the second autocorrelation module of the current layer of decoder to obtain the second periodic information; inputting the second periodic information and the third periodic features into the fourth variational mode decomposition module of the current layer of decoder to obtain the fourth periodic features and the fourth trend features; inputting the fourth periodic features into the feed-forward module of the current layer of decoder to obtain the feed-forward information; inputting the feed-forward information and the fourth periodic features into the fifth variational mode decomposition module of the current layer of decoder to obtain the fifth periodic features and the fifth trend features; determining the fifth periodic features as the input to the first autocorrelation module of the next layer of decoder; determining the fifth periodic features output by the last layer of decoder as the final periodic features; weighted summing the trend features in the input features, the third trend features, the fourth trend features, and the fifth trend features of all layers of decoders to obtain the final trend features; determining the final periodic features and the final trend features as the carbon emission features.

[0072] Through this embodiment, the variational mode decomposition (VMD) module is used to replace the traditional decomposition module based on moving average to solve the problem that the simple moving average method cannot handle the sudden fluctuations of trend features and periodic features. Moreover, this embodiment uses multiple layers of decoders, and accurate prediction of periodic features and trend features can be obtained for subsequent prediction.

[0073] As an example, after obtaining the input features, the periodic features in the input features are input into the accounting prediction decoding module for feature extraction. Assume that the accounting prediction decoding module has M layers of decoders. Each layer of decoding module includes two autocorrelation (Auto-Correlation) modules and a feed-forward (Feed-forward) module. A VMD module is connected behind each module to further obtain the trend component and the periodic component of the intermediate features. Moreover, the second Auto-Correlation module also receives and processes the key features obtained previously. The output part of the last layer of decoder includes the finally extracted periodic features and the superposition of all trend information for final accounting and prediction, as shown in the following formula:

[0074]

[0075] Among them, among them and respectively represent the periodic feature and the trend feature extracted by the l th VMD module in the i th decoder. N is the number of layers of the encoder, and the output by the last layer of the decoder, that is, the final periodic feature, that is, the final trend feature.

[0076] In step S107, the carbon emission feature is input into the accounting and prediction module to obtain the carbon emission amount in the second predetermined time period and the predicted carbon emission amount for the predicted time length.

[0077] As an example, the above-mentioned accounting and prediction module can be a three-layer Multilayer Perceptron (abbreviated as MLP), and the present disclosure does not limit this.

[0078] As an example, the output feature obtained by the last layer of the decoder, that is, the carbon emission feature, can be input into a three-layer MLP for learning. RELU is used as the activation function between each layer, and finally the accounting and prediction values of the carbon emission amount are obtained. Among them, the emission amount obtained based on historical data is used as the accounting part, and the emission amount obtained based on the generated prediction information is used as the prediction part, so as to complete the accounting and prediction of carbon emissions. Specifically, it can be as follows:

[0079] Among them, M is the number of layers of the decoder.

[0080] For the convenience of understanding the above present disclosure, the following is combined with Figure 2 and Figure 3 for a systematic description.

[0081] Based on key technologies such as time series feature decomposition and Transformer architecture, the present disclosure proposes a multi-source data fusion carbon emission accounting and prediction model based on an improved Transformer. By selecting multi-source factors affecting carbon emissions to construct a data set, and then introducing static influencing factors into the model through a learnable static encoder, accurate accounting and prediction of carbon emissions can be realized based on the improved Transformer architecture.

[0082] Figure 2 Shows the architecture diagram of the carbon emission accounting and prediction model, asFigure 2 As shown, the model includes a static encoder (i.e., the above-mentioned static encoding module), an N-layer encoder (i.e., the above-mentioned accounting prediction encoding module), an N-layer decoder (i.e., the above-mentioned accounting prediction decoding module), a VMD module, and an MLP module (i.e., the above-mentioned accounting prediction module). Among them, each layer of the encoder contains an Auto-Correlation module, two VMD modules, and a Feed-forward module. Each layer of the decoder contains two Auto-Correlation modules and a Feed-forward module. A VMD module is connected behind each module.

[0083] In the constructed multi-source dataset, the static data in the multi-source dataset (i.e., the input data) is input into the static encoder to obtain static features, and then the static features are fused with the dynamic data in the multi-source dataset to obtain fused features. Then, the fused features are input into the first layer of the encoder to obtain corresponding second-cycle features, and the second-cycle features are continuously input into the next layer of the encoder to obtain the second-cycle features output by the next layer of the encoder until passing through N layers of the encoder to obtain the second-cycle features output by the Nth layer of the encoder, which are determined as the key features to be used subsequently. Then, part of the data needs to be selected from the multi-source dataset as prediction data. For example, the data corresponding to the second half of the time (assumed to be a preset duration) can be selected as prediction data. The prediction data is input into the VMD module to obtain the trend component and the period component of each accounting index, and then these two components are filled to obtain the filled trend features and period features of each accounting index. Then, the filled period features are input into the first layer of the decoder to obtain fifth-cycle features. Then, the fifth-cycle features are continuously input into the next layer of the decoder to obtain the fifth-cycle features output by the next layer of the decoder until passing through M layers of the decoder to obtain the fifth-cycle features output by the Mth layer of the decoder, which are determined as the final cycle features. At the same time, the filled trend features and all the trend features generated in the M layers of the decoder are weighted and summed to obtain the final trend feature. After inputting the final trend feature and the final cycle feature into the MLP, the output result can be obtained, that is, the carbon emissions for the preset duration and the predicted carbon emissions for the predicted time length.

[0084] It should be noted that the inputs of the encoder and the decoder will be immediately converted into corresponding query vectors Q, key vectors K, and value vectors V, and then subsequent processing is performed through QKV. This belongs to the internal processing method of the encoder and the decoder, and the present disclosure will not expand on this.

[0085] Furthermore, for the carbon emission accounting and prediction model, during the model training process, the mean absolute error (MAE) between the output value and the actual carbon emission value can be calculated as the loss for training the model. This disclosure does not limit this; the actual carbon emission value can be selected as needed, such as the total monthly carbon dioxide emissions data of a company.

[0086] Figure 3 A flowchart showing the carbon emission accounting and prediction method system is as Figure 3 shown, and it includes a total of five steps: data selection, static data encoding and fusion, historical feature extraction, accounting and prediction feature extraction, and carbon emission accounting and prediction. S1: Data selection. Data can be selected with a ceramic enterprise as the accounting boundary, reasonable accounting indicators can be selected, and a multi-source data set can be constructed to provide basic support for subsequent algorithms.

[0087] S2: Static data encoding and fusion. A learnable static encoder is used to encode the static data in the data set, and the encoding result is combined with the dynamic data in the data set to achieve effective fusion of static and dynamic information.

[0088] S3: Historical feature extraction. The fused data is used to deeply extract the key features of historical time series data using the accounting and prediction encoding module.

[0089] S4: Accounting and prediction feature extraction. Input features are constructed based on some historical data in the data set, and the key features and input features are processed through the accounting and prediction decoding module to generate carbon emission features for the final accounting and prediction.

[0090] S5: Carbon emission accounting and prediction. The extracted carbon emission features are input into the accounting and prediction module to calculate historical carbon emissions and predict future carbon emissions.

[0091] It should be noted that in the data selection stage of the above embodiments, the accuracy of accounting and prediction is further improved by selecting data from multiple sources; in the static data encoding and fusion stage, a learnable static encoder is used to extract the feature information of static data, thereby improving the prediction accuracy and the expression ability of the model; in the historical feature extraction and accounting and prediction feature extraction stages, the VMD module is used to solve the problem that the simple moving average method cannot handle sudden fluctuations in trends and cycles.

[0092] In summary, this disclosure focuses on the carbon emission accounting and prediction tasks. By deeply studying relevant algorithms, a multi-source data fusion carbon emission accounting and prediction algorithm based on an improved Transformer is proposed. This algorithm aims to improve the accuracy and efficiency of accounting and prediction. The algorithm adopts an improved Transformer architecture, including: a data selection module, a static data encoding and fusion module, an accounting prediction encoding module, an accounting prediction decoding module, and an accounting prediction module. Specifically, the data selection module is used to construct a multi-source data set; the static data encoding and fusion module is used to encode the static data of carbon emissions and fuse these encodings with dynamic data to form fusion features; the accounting prediction encoding module is used to deeply mine the periodic laws in historical data. This module extracts the key information of historical time series data and uses this information as cross features to be passed to the accounting prediction decoding module to optimize the prediction effect; the accounting prediction decoding module realizes the dual functions of accounting and prediction by integrating trends and periodicity; the accounting prediction module then performs accounting and prediction by integrating the trend and periodic information.

[0093] It should be noted that a random "mask" mechanism can be introduced during the model training process to ensure that the model can obtain relatively accurate accounting and prediction results in a complex and changing data environment. This disclosure does not limit this.

[0094] This disclosure relates to the fields of data analysis, artificial intelligence, deep learning, carbon emission accounting, and carbon emission prediction, and also relates to technologies such as carbon emission data analysis and processing, and long-term prediction of time series data. It can accurately account for carbon emission data by combining various factors affecting carbon emissions and predict future long-term carbon emission data.

[0095] Figure 4 is a block diagram of a carbon emission accounting and prediction system based on multi-source data fusion showing an embodiment of the present disclosure, as Figure 4 shown, the device includes a data selection unit 40, a fusion unit 42, an encoding unit 44, a decomposition unit 46, a filling unit 48, a decoding unit 410, and a prediction unit 412.

[0096] A data selection unit 40 is configured to obtain data of multiple accounting metrics within a first predetermined time period under a predetermined accounting boundary; a fusion unit 42 is configured to fuse static data and dynamic data in the data to obtain a fusion feature; an encoding unit 44 is configured to input the fusion feature into an accounting prediction encoding module to obtain a key feature, where the key feature indicates periodic information of the data; a decomposition unit 46 is configured to input the fusion feature corresponding to a second predetermined time period in the fusion feature into a decomposition module to obtain a trend component and a periodic component of each of the multiple accounting metrics, where the second predetermined time period is a partial time period within the first predetermined time period and the start time of the second predetermined time period is later than the start time of the first predetermined time period; a filling unit 48 is configured to obtain an input feature based on the trend component, the periodic component, a preset prediction time length, and the data mean of each of the multiple accounting metrics; a decoding unit 410 is configured to input the input feature and the key feature into an accounting prediction decoding module to obtain a carbon emission feature; a prediction unit 412 is configured to input the carbon emission feature into an accounting prediction module to obtain the carbon emissions of the second predetermined time period and the predicted carbon emissions of the prediction time length.

[0097] According to an embodiment of the present disclosure, the fusion unit 42 is further configured to input the static data in the data into a pre-trained static encoding module to obtain a static feature; and fuse the static feature and the dynamic data in the data to obtain a fusion feature.

[0098] According to an embodiment of the present disclosure, the static encoding module includes a first fully connected network, a second fully connected network, and an activation function, where the fusion unit 42 is further configured to input the static data into the first fully connected network to obtain a first feature; input the first feature into the activation function to obtain a second feature; and input the second feature and the static data into the second fully connected network to obtain a static feature.

[0099] According to an embodiment of the present disclosure, the filling unit 48 is further configured to perform the following processing for each of the multiple accounting metrics: fill a preset value of the prediction time length into the periodic component of the current accounting metric to obtain a filled periodic feature; fill the data mean of the current accounting metric of the prediction time length into the trend component of the current accounting metric to obtain a filled trend feature; and determine the periodic feature and the trend feature as the input feature.

[0100] According to an embodiment of the present disclosure, the accounting prediction encoding module includes a multi-layer encoder. Each layer of the encoder includes an autocorrelation module, a first variational mode decomposition module, a feed-forward module, and a second variational mode decomposition module. Among them, the encoding unit 44 is further configured to perform the following processing through each layer of the encoder in the multi-layer encoder: input the fusion feature into the autocorrelation module of the current layer encoder to obtain periodic information; input the periodic information and the fusion feature into the first variational mode decomposition module of the current layer encoder to obtain a first periodic feature and a first trend feature; input the first periodic feature into the feed-forward module of the current layer encoder to obtain feed-forward information; input the feed-forward information and the first periodic feature into the second variational mode decomposition module of the current layer encoder to obtain a second periodic feature and a second trend feature; determine the second periodic feature as the input of the autocorrelation module of the next layer encoder; and determine the second periodic feature output by the last layer encoder as the key feature.

[0101] According to an embodiment of the present disclosure, the accounting prediction decoding module includes a multi-layer decoder. Each layer of the decoder includes a first autocorrelation module, a third variational mode decomposition module, a second autocorrelation module, a fourth variational mode decomposition module, a feed-forward module, and a fifth variational mode decomposition module. Among them, the decoding unit 410 is further configured to perform the following processing through each layer of the decoder in the multi-layer decoder: input the periodic feature in the input feature into the first autocorrelation module of the current layer decoder to obtain first periodic information; input the first periodic information and the periodic feature in the input feature into the third variational mode decomposition module of the current layer decoder to obtain a third periodic feature and a third trend feature; input the third periodic feature and the key feature into the second autocorrelation module of the current layer decoder to obtain second periodic information; input the second periodic information and the third periodic feature into the fourth variational mode decomposition module of the current layer decoder to obtain a fourth periodic feature and a fourth trend feature; input the fourth periodic feature into the feed-forward module of the current layer decoder to obtain feed-forward information; input the feed-forward information and the fourth periodic feature into the fifth variational mode decomposition module of the current layer decoder to obtain a fifth periodic feature and a fifth trend feature; determine the fifth periodic feature as the input of the first autocorrelation module of the next layer decoder; determine the fifth periodic feature output by the last layer decoder as the final periodic feature; perform a weighted sum of the trend feature in the input feature and the third trend feature, the fourth trend feature, and the fifth trend feature of all layers of the decoder to obtain the final trend feature; and determine the final periodic feature and the final trend feature as the carbon emission feature.

[0102] According to an embodiment of the present disclosure, there is provided a computer-readable storage medium storing instructions, wherein when the instructions are run by at least one computing device, the at least one computing device is caused to execute the carbon emission accounting and prediction method based on multi-source data fusion as described in any of the above embodiments.

[0103] According to an embodiment of the present disclosure, a system is provided that includes at least one computing device and at least one storage device storing instructions, wherein when the instructions are run by the at least one computing device, the at least one computing device is caused to execute the carbon emission accounting and prediction method based on multi-source data fusion according to any of the above embodiments.

[0104] According to an embodiment of the present disclosure, a computer program product is provided that includes computer instructions that, when executed by a processor, implement the carbon emission accounting and prediction method based on multi-source data fusion according to any of the above.

[0105] Although some embodiments of the present disclosure have been shown and described, those skilled in the art should understand that these embodiments can be modified without departing from the principles and spirit of the present disclosure as defined by the claims and their equivalents.

Claims

1. A carbon emission accounting and prediction method based on multi-source data fusion, characterized in that Including: Obtain data of multiple accounting indicators within a first predetermined time period under a predetermined accounting boundary; Fuse the static data and dynamic data in the said data to obtain a fused feature; Input the fused feature into an accounting prediction coding module to obtain a key feature, where the key feature indicates the periodic information of the said data; Input the fused feature corresponding to a second predetermined time period in the fused feature into a decomposition module to obtain the trend component and periodic component of each of the multiple accounting indicators, where the second predetermined time period is a partial time period within the first predetermined time period and the start time of the second predetermined time period is later than the start time of the first predetermined time period; Based on the trend component, the periodic component, a preset prediction time length, and the data mean of each of the multiple accounting indicators, obtain an input feature; Input the input feature and the key feature into an accounting prediction decoding module to obtain a carbon emission feature; Input the carbon emission feature into an accounting prediction module to obtain the carbon emissions of the second predetermined time period and the predicted carbon emissions for the prediction time length.

2. The carbon emission accounting and prediction method according to claim 1, wherein The fusing the static data and dynamic data in the said data to obtain a fused feature includes: Input the static data in the said data into a pre-trained static coding module to obtain a static feature; Fuse the static feature and the dynamic data in the said data to obtain a fused feature.

3. The carbon emission accounting and prediction method according to claim 2, wherein The static coding module includes a first fully connected network, a second fully connected network, and an activation function. Among them, the inputting the static data in the said data into a pre-trained static coding module to obtain a static feature includes: Input the static data into the first fully connected network to obtain a first feature; Input the first feature into the activation function to obtain a second feature; Input the second feature and the static data into the second fully connected network to obtain a static feature.

4. The carbon emission accounting and prediction method according to claim 1, characterized in that, The obtaining the input feature based on the trend component, the periodic component, a preset prediction time length, and the data mean of each of the multiple accounting indicators includes: For each of the multiple accounting indicators, perform the following processing: fill the periodic component of the current accounting indicator with a preset value for the prediction time length to obtain a filled periodic feature; fill the trend component of the current accounting indicator with the data mean of the current accounting indicator for the prediction time length to obtain a filled trend feature; Determine the periodic feature and the trend feature as the input feature.

5. The carbon emission accounting and prediction method according to claim 1, wherein The accounting prediction coding module includes multiple layers of encoders, and each layer of encoder includes an autocorrelation module, a first variational mode decomposition module, a feed-forward module, and a second variational mode decomposition module. Among them, the inputting the fused feature into the accounting prediction coding module to obtain a key feature includes: Through each layer of encoder in the multiple layers of encoders, perform the following processing: Input the fused feature into the autocorrelation module of the current layer of encoder to obtain periodic information; Input the periodic information and the fused feature into the first variational mode decomposition module of the current layer of encoder to obtain a first periodic feature and a first trend feature; Input the first - cycle feature into the feed - forward module of the current - layer encoder to obtain feed - forward information; Input the feed - forward information and the first - cycle feature into the second variational mode decomposition module of the current - layer encoder to obtain a second - cycle feature and a second - trend feature; Determine the second - cycle feature as the input of the autocorrelation module of the next - layer encoder; Determine the second - cycle feature output by the last - layer encoder as the key feature.

6. The carbon emission accounting and prediction method according to claim 4, wherein, The accounting prediction decoding module includes multiple layers of decoders. Each layer of decoder includes a first autocorrelation module, a third variational mode decomposition module, a second autocorrelation module, a fourth variational mode decomposition module, a feed - forward module, and a fifth variational mode decomposition module. Among them, inputting the input feature and the key feature into the accounting prediction decoding module to obtain a carbon - emission feature includes: Through each layer of decoder in the multiple layers of decoders, perform the following processing: Input the cycle feature in the input feature into the first autocorrelation module of the current - layer decoder to obtain first - cycle information; Input the first - cycle information and the cycle feature in the input feature into the third variational mode decomposition module of the current - layer decoder to obtain a third - cycle feature and a third - trend feature; Input the third - cycle feature and the key feature into the second autocorrelation module of the current - layer decoder to obtain second - cycle information; Input the second - cycle information and the third - cycle feature into the fourth variational mode decomposition module of the current - layer decoder to obtain a fourth - cycle feature and a fourth - trend feature; Input the fourth - cycle feature into the feed - forward module of the current - layer decoder to obtain feed - forward information; Input the feed - forward information and the fourth - cycle feature into the fifth variational mode decomposition module of the current - layer decoder to obtain a fifth - cycle feature and a fifth - trend feature; Determine the fifth - cycle feature as the input of the first autocorrelation module of the next - layer decoder; Determine the fifth - cycle feature output by the last - layer decoder as the final - cycle feature; Perform a weighted sum of the trend feature in the input feature and the third - trend feature, fourth - trend feature, and fifth - trend feature of all layers of decoders to obtain a final - trend feature; Determine the final - cycle feature and the final - trend feature as the carbon - emission feature.

7. A carbon emission accounting and prediction system based on multi-source data fusion, characterized in that, Including: A data selection unit configured to obtain data of multiple accounting indicators within a first predetermined time period under a predetermined accounting boundary; A fusion unit configured to fuse the static data and dynamic data in the data to obtain a fusion feature; An encoding unit configured to input the fusion feature into an accounting prediction encoding module to obtain a key feature, where the key feature indicates the periodic information of the data; A decomposition unit configured to input the fusion feature corresponding to a second predetermined time period in the fusion feature into a decomposition module to obtain the trend component and cycle component of each of the multiple accounting indicators, where the second predetermined time period is a partial time period within the first predetermined time period and the start time of the second predetermined time period is later than the start time of the first predetermined time period; A filling unit, configured to obtain input features based on the trend component, the periodic component, a preset prediction time length, and the data mean values of the multiple accounting indicators; A decoding unit, configured to input the input features and the key features into an accounting prediction decoding module to obtain carbon emission features; A prediction unit, configured to input the carbon emission features into an accounting prediction module to obtain the carbon emissions of the second predetermined time period and the predicted carbon emissions of the prediction time length.

8. A computer-readable storage medium storing instructions, characterized in that, When the instruction is run by at least one computing device, it causes the at least one computing device to execute the carbon emission accounting and prediction method based on multi-source data fusion according to any one of claims 1 to 6.

9. A system comprising at least one computing device and at least one storage device storing instructions, characterized in that, When the instruction is run by the at least one computing device, it causes the at least one computing device to execute the carbon emission accounting and prediction method based on multi-source data fusion according to any one of claims 1 to 6.

10. A computer program product comprising computer instructions, characterized in that, When the computer instruction is executed by a processor, it implements the carbon emission accounting and prediction method based on multi-source data fusion according to any one of claims 1 to 6.

Images