High-resolution department carbon emission prediction method and system based on hybrid deep learning model

Abstract

The invention discloses a hybrid deep learning model-based high-resolution department carbon emission prediction method and system, and the method comprises the steps: decomposing carbon emission data into subsequences which comprise a low-frequency subsequence and a high-frequency subsequence; constructing a low-frequency prediction model and a high-frequency prediction model, and respectively predicting the low-frequency sub-sequence and the high-frequency sub-sequence through the low-frequency prediction model and the high-frequency prediction model to obtain a low-frequency prediction result and a high-frequency prediction result; and fusing the low-frequency prediction result and the high-frequency prediction result to obtain a final carbon emission prediction result. Compared with a previous statistical learning measurement model and an evaluation model based on environmental influence, the method has a more accurate effect.

Description

Technical Field

[0001] The present invention belongs to the technical field of carbon emission detection, and in particular relates to a high-resolution departmental carbon emission prediction method and system based on a hybrid deep learning model. Background Art

[0002] In the context of global warming and climate change, it is crucial to accurately predict carbon emissions from various sectors. Existing carbon emission prediction models usually have the following problems that need to be solved:

[0003] Insufficient resolution: Many traditional carbon emission forecasting methods cannot provide sufficiently high temporal resolution, which means that the forecast results may only give average emissions over a large area or over a long period of time, but cannot capture the fluctuations in carbon emissions in different sectors.

[0004] Model accuracy and reliability: Existing forecasting models may perform poorly when dealing with complex, nonlinear carbon emissions data.

[0005] Data integration problem: High-resolution carbon emission forecasting requires the integration of data from different sources, such as remote sensing data, ground monitoring data, economic activity data, etc. Existing methods may have limitations in data fusion and processing, resulting in inaccurate forecast results. Summary of the invention

[0006] In order to solve the above technical problems, the present invention proposes a high-resolution sector carbon emissions prediction method and system based on a hybrid deep learning model to solve the problems existing in the above-mentioned prior art.

[0007] To achieve the above object, the present invention provides a high-resolution sector carbon emissions prediction method based on a hybrid deep learning model, comprising the following steps:

[0008] Decomposing the carbon emission data into subsequences, wherein the subsequences include low-frequency subsequences and high-frequency subsequences;

[0009] Constructing a low-frequency prediction model and a high-frequency prediction model, and predicting the low-frequency subsequence and the high-frequency subsequence respectively by using the low-frequency prediction model and the high-frequency prediction model to obtain a low-frequency prediction result and a high-frequency prediction result;

[0010] The low-frequency prediction results and the high-frequency prediction results are integrated to obtain the final carbon emission prediction results.

[0011] Preferably, decomposing the carbon emission data into subsequences comprises:

[0012] The optimal number of modes and penalty factor of the variational mode decomposition model are obtained according to the frost optimization algorithm, and the carbon emission data are decomposed into subsequences through the optimized variational mode decomposition model.

[0013] Preferably, obtaining the low-frequency prediction result includes:

[0014] An initial low-frequency prediction model is constructed according to a convolutional neural network, a long short-term memory network and an attention mechanism network. The initial low-frequency prediction model is optimized by a sparrow search algorithm to obtain a low-frequency prediction model. The low-frequency subsequence is predicted by the low-frequency prediction model to obtain a low-frequency prediction result.

[0015] Preferably, obtaining the high-frequency prediction result includes:

[0016] A high-frequency prediction model is constructed based on a bidirectional recurrent network and a multi-layer perceptron. The high-frequency subsequence is predicted by the high-frequency prediction model to obtain a high-frequency prediction result.

[0017] The present invention also provides a high-resolution sector carbon emission prediction system based on a hybrid deep learning model, comprising:

[0018] A decomposition module, used for decomposing the carbon emission data into subsequences, wherein the subsequences include low-frequency subsequences and high-frequency subsequences;

[0019] A prediction module is used to construct a low-frequency prediction model and a high-frequency prediction model, and predict the low-frequency subsequence and the high-frequency subsequence respectively through the low-frequency prediction model and the high-frequency prediction model to obtain a low-frequency prediction result and a high-frequency prediction result;

[0020] The fusion module fuses the low-frequency prediction results and the high-frequency prediction results to obtain the final carbon emission prediction results.

[0021] Preferably, the decomposition module includes a first optimization unit and a decomposition unit;

[0022] The first optimization unit is used to obtain the optimal mode number and penalty factor of the variational mode decomposition model according to the frost optimization algorithm to obtain the optimized variational mode decomposition model;

[0023] The decomposition unit is used to decompose the carbon emission data into subsequences through an optimized variational mode decomposition model.

[0024] Preferably, the prediction module includes a construction unit and a prediction unit;

[0025] The construction unit is used to construct a low-frequency prediction model based on a convolutional neural network, a long short-term memory network and an attention mechanism network, and to construct a high-frequency prediction model based on a bidirectional recurrent network and a multi-layer perceptron;

[0026] The prediction unit is used to predict the low-frequency subsequence and the high-frequency subsequence respectively through the low-frequency prediction model and the high-frequency prediction model to obtain a low-frequency prediction result and a high-frequency prediction result.

[0027] Preferably, the prediction module further includes a second optimization unit;

[0028] The second optimization unit is used to optimize the low-frequency prediction model by using a sparrow search algorithm.

[0029] The present invention also provides an electronic device, comprising: a processor and a memory storing computer program instructions; the processor implements the above method when executing the computer program instructions.

[0030] The present invention also provides a computer storage medium, on which computer program instructions are stored, and the computer program instructions implement the above method when executed by a processor.

[0031] Compared with the prior art, the present invention has the following advantages and technical effects:

[0032] The present invention discloses a high-resolution departmental carbon emission prediction method and system based on a hybrid deep learning model, including: decomposing carbon emission data into subsequences, wherein the subsequences include low-frequency subsequences and high-frequency subsequences; constructing a low-frequency prediction model and a high-frequency prediction model, respectively predicting the low-frequency subsequence and the high-frequency subsequence through the low-frequency prediction model and the high-frequency prediction model to obtain a low-frequency prediction result and a high-frequency prediction result; fusing the low-frequency prediction result and the high-frequency prediction result to obtain a final carbon emission prediction result. The present invention can solve the data integration problem by decomposing carbon emission data into subsequences, thereby improving the problem of insufficient resolution, which greatly improves the accuracy and reliability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0034] Figure 1 A method logic block diagram of an embodiment of the present invention;

[0035] Figure 2 Drawing the original data of carbon emissions from the power industry according to an embodiment of the present invention;

[0036] Figure 3 A high-frequency subsequence data graph after decomposition of the original data of an embodiment of the present invention;

[0037] Figure 4 A low-frequency subsequence data graph after decomposition of the original data plot according to an embodiment of the present invention;

[0038] Figure 5 is an algorithm flow chart of the RIME-VMD model of an embodiment of the present invention;

[0039] Figure 6 This is a structural diagram of the SSA-CNN-LSTM-ATTEN model of an embodiment of the present invention;

[0040] Figure 7 This is a structural diagram of a CNN layer model according to an embodiment of the present invention;

[0041] Figure 8 This is a structural diagram of the LSTM layer model of an embodiment of the present invention;

[0042] Fig. 9 This is a structural diagram of the ATTEN layer model of the attention mechanism according to an embodiment of the present invention;

[0043] Fig.10 This is a structural diagram of a BIGRU layer model according to an embodiment of the present invention;

[0044] Fig.11 This is a structural diagram of the MLP layer model of an embodiment of the present invention. DETAILED DESCRIPTION

[0045] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0046] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0047] The following is an introduction to the existing technology:

[0048] In the context of addressing climate change, carbon emission forecasting has become an important direction for research and policy making. In recent years, carbon emission forecasting methods at different levels (regional level and sectoral level) have been continuously developed in order to more accurately assess and predict carbon emission trends. At present, there is a lack of fine-grained sectoral carbon emission forecasts, and most of them are concentrated on regional carbon emission forecasts. In addition, the carbon emissions of these sectors are mostly nonlinear, non-periodic and unstable data. This brings challenges to the prediction of carbon emissions. The following are some existing representative studies and methods, which explore the strategies and techniques of carbon emission forecasting from different perspectives.

[0049] 1. Carbon emissions forecast for a single sector:

[0050] In the study of single-sector carbon emission forecasting, Bai et al. used the meta-frontier analysis method and linear programming algorithm to calculate the carbon emission efficiency of the transportation sector in various provinces in China from 2005 to 2015. The study evaluated the efficiency performance of the transportation sector in various provinces in terms of carbon emissions by establishing a meta-frontier model, and further calculated the carbon emission reduction potential of each province. This method can effectively identify areas with low carbon emission efficiency, thereby providing a scientific basis for the formulation of carbon emission reduction measures. Li et al. used the Monte Carlo simulation method to study the carbon emission trajectory in China's building operations. They analyzed historical CO2 emission data and used a stochastic process model to predict the future carbon emission trend of building operations. Monte Carlo simulation generates a large number of possible future scenarios and takes into account multiple uncertainties, thereby providing a comprehensive forecast of future carbon emissions. This method is of great reference value for the formulation of long-term building energy conservation and emission reduction policies. Ran et al. proposed a comprehensive carbon emission forecasting method. They first used a support vector machine (SVM) model to predict the industrial carbon emission trend from 2021 to 2040. The support vector machine model effectively captures the complex patterns of industrial carbon emissions by constructing classification boundaries in high-dimensional space. In order to further improve the prediction accuracy, they also used random forest technology to filter out the noise data that affects industrial carbon emissions. Random forest technology effectively reduces the overfitting phenomenon of the model by constructing multiple decision trees and integrating their results, thereby improving the stability and reliability of the prediction.

[0051] 2. Carbon emission forecast at the regional level:

[0052] In the study of carbon emission prediction at the regional level, Chen et al. proposed a method that combines the autoregressive integrated moving average (ARIMA) model with the NAR neural network (NAR-NN). This method uses the ARIMA model to process the linear components in the time series data, and captures the nonlinear components through the NAR neural network, thereby improving the accuracy of the prediction. The results show that China's low-carbon demonstration areas are mainly concentrated in Beijing and Shanghai, low-carbon potential areas are distributed in the southeast coastal areas, high-carbon developed areas are mainly located in the northeast, and the western region is basically a high-carbon area. This regional classification helps to understand the carbon emission characteristics of different regions and provide guidance for the formulation of regional carbon emission reduction policies.

[0053] Huang proposed a genetic algorithm extreme learning machine (GA-ELM) algorithm to predict carbon emissions in the Beijing-Tianjin-Hebei region. The genetic algorithm was used to optimize the parameter settings of the extreme learning machine, thereby improving the prediction performance of the model. The method predicted carbon emissions in the Beijing-Tianjin-Hebei region with and without considering the impact of the coordinated development strategy. The study found that the implementation of the coordinated development strategy can effectively slow down the growth of carbon emissions, thus providing an effective strategy for regional carbon management.

[0054] Zhou used an input-output based Bayesian neural network (IO-BNN) model to study the long-term carbon emission path, carbon peak and carbon reduction potential of Guangdong Province. The IO-BNN model combines the advantages of input-output analysis and Bayesian neural network, and can more accurately capture the dynamic changes of carbon emissions and their influencing factors. The study shows that Guangdong Province can achieve significant emission reduction potential through effective carbon reduction measures during the long-term planning period from 2021 to 2050, providing a scientific basis for local carbon management and policy making.

[0055] The special terms of the present invention are explained as follows:

[0056]

[0057] Embodiment 1:

[0058] like Figure 1 As shown, this embodiment provides a high-resolution sector carbon emissions prediction method based on a hybrid deep learning model, comprising the following steps:

[0059] Decomposing the carbon emission data into subsequences, wherein the subsequences include low-frequency subsequences and high-frequency subsequences;

[0060] Constructing a low-frequency prediction model and a high-frequency prediction model, respectively predicting the low-frequency subsequence and the high-frequency subsequence through the low-frequency prediction model and the high-frequency prediction model, and obtaining a low-frequency prediction result and a high-frequency prediction result;

[0061] The low-frequency prediction results and high-frequency prediction results are integrated to obtain the final carbon emission prediction results.

[0062] The specific method is:

[0063] In the first step, the data set is processed and divided into a training set and a test set. The carbon emission data set is decomposed into high-frequency subsequences and low-frequency subsequences using the RIME-VMD algorithm. Figure 2-4 Conduct examples;

[0064] The specific steps are as follows: VMD realizes the function of decomposing the given real input signal f(t) into sub-signals μ k , where each mode μ kFocused on a center frequency ω k Nearby, f(t) is decomposed into k modes as follows:

[0065] Step 1: Use Hilbert transform to construct each mode μ k Spectrum

[0066] Step 2: Transfer each modal spectrum to the corresponding baseband according to the center frequency of each mode

[0067] Step 3: Determine the bandwidth by Gaussian smoothing of the demodulated signal.

[0068]

[0069] This results in a constrained variational problem, where f(t) is the data sequence, {uk} is the set of all modes, {wk} is the center frequency sequence, t is the t-th order moment, K is the modal value, and δ(t) is the Dirac distribution. is the convolution operator. In order to transform the constrained variational problem into an unconstrained variational problem, the quadratic penalty factor α and the Lagrange multiplier λ are introduced, and then the multiplier alternation method is used to update uk and wk to solve the transformation problem.

[0070] Step 2: Use the segmented training set as the vector input of CNN-ATTEN-LSTM (CALSTM);

[0071] Step 3: Use the SSA algorithm to optimize the hidden_nodes, alpha, and epoch of the CNN-LSTM-ATTEN model. After determining the optimal parameters, select the optimal parameters to build the model;

[0072] Step 4: Determine the loop termination condition, retrain the optimized CNN-LSTM-ATTEN model and compare the results to determine the effectiveness of the low-frequency subsequence module. If the termination condition is met, exit the loop to end the prediction process and output the results.

[0073] Step 5: Use the BIGRU-MLP model to predict the features of the remaining high-frequency subsequences;

[0074] Step 6: Integrate the prediction results of the low-frequency subsequence and the high-frequency subsequence to obtain the final prediction day result.

[0075] The RIME-VMD-SSA-CNN-LSTM-ATTEN-BIGRU-MLP hybrid model predicts the future carbon emissions of six sectors at a fine-grained level. The carbon emissions are divided into six sectors, namely Domestic Aviation, Ground Transport, Industry, International Aviation, Power, and Residential. The frost optimization algorithm RIME can determine the optimal number of modes k and penalty factor α of the VMD model, decompose the carbon emission data into subsequences, and the decomposition process is as follows: Figure 5 As shown; the sparrow search algorithm SSA algorithm is used to optimize the CNN-LSTM-ATTEN network; in order to avoid the disadvantage of insufficient frequency band feature extraction caused by using only the same prediction model for all data, this embodiment proposes a new hybrid deep learning model for the high- and low-frequency spectrum frequency division prediction method of carbon emissions by sector, and SSA-CNN-LSTM-ATTEN is used to predict the low-frequency subsequence after decomposition. The structure of SSA-CNN-LSTM-ATTEN is as follows Figure 6 As shown, the CNN layer model structure is as follows Figure 7 As shown; the LSTM layer model structure is as follows Figure 8 As shown; the structure of the attention mechanism ATTEN layer model is as follows Fig. 9 As shown, the BIGRU-MLP model is used to predict the high-frequency subsequence after decomposition. The BIGRU layer model structure is as follows Fig.10 As shown; the MLP layer model structure is as follows Fig.11 As shown, better results were achieved, providing a new perspective for related carbon emissions and emission reduction research.

[0076] Models_Power <![CDATA[R 2 ]]> MAE RMSE MAPE LSTM 0.8635 5.3670 6.5042 0.0359 CNN-LSTM 0.8967 4.5544 5.6563 0.0303 SSA-CNN-LSTM 0.9232 3.9513 4.8765 0.0268 SSA-CNN-LSTM-Atten 0.9514 2.9259 3.8804 0.0201 VMD-SSA-CNN-LSTM-Atten 0.9596 2.5710 3.4943 0.0165 VMD-SSA-CNN-LSTM-Atten-MLP 0.9711 2.0617 2.9573 0.0132 RIME-VMD-SSA-CNN-LSTM-Atten-BIGUR-MLP 0.9763 1.9897 2.6591 0.0127

[0077] RMSE, MAE, MAPE and R of the RVMD-SSA-CALSTM-BIGRU-MLP (i.e., RIME-VMD-SSA-CNN-LSTM-Atten-MLP) model proposed in this paper 2The four indicators of the VMD-SSA-CNN-LSTM-Atten-MLP model are 2.6591, 1.9897, 0.0127 and 0.9763 respectively. The four indicators of the VMD-SSA-CNN-LSTM-Atten-MLP model are 2.9573, 2.0617, 0.0132 and 0.9711 respectively. From the data results, it can be seen that before predicting the sequence, the original sequence is decomposed by VMD and optimized with RIME, which can make the prediction result more accurate and the MLP model is indeed more suitable for processing high-frequency time series. The four indicators of the VMD-SSA-CNN-LSTM-Atten model are 3.4943, 2.5710, 0.0165 and 0.9596 respectively. It can be concluded that the use of the adaptive spectrum allocation prediction algorithm proposed in this paper on the decomposed signal can effectively increase the accuracy, proving its effectiveness. The CNN-LSTM model optimized by SSA has a lower error, which confirms the rationality of the model's automatic optimization of parameters using the SSA algorithm. It can also be concluded that after adding the attention mechanism, the attention mechanism can assign variable weights to the features in the time series according to the time step, promoting the indirect connection between the LSTM output and the related input, showing the completeness of the experiment.

[0078] Finally, the proposed model is compared with other model combinations and single models. The RMSE, MAE and MAPE values ​​of the main model are the smallest. 2 The results finally show that the proposed RVMD-SSA-CALSTM-BIGRU-MLP model has the highest prediction accuracy and can reasonably handle multi-sector carbon emission prediction research.

[0079] From the data results of the carbon emission forecast by sector, it can be seen that before the forecast sequence, the original sequence is decomposed by VMD and optimized with RIME, which can make the forecast result more accurate and the MLP model is indeed more suitable for processing high-frequency time series; compared with RIME-VMD-SSA-CNN-LSTM-ATTEN, it can be seen that the hybrid deep learning forecasting method proposed in this embodiment can effectively increase the accuracy of the decomposed signal, which proves the effectiveness of the method. The CNN-LSTM-ATTEN model optimized by SSA has a lower error, which confirms the rationality of the model's automatic optimization of parameters using the SSA algorithm. It can also be concluded that after adding the attention mechanism, the attention mechanism can assign variable weights to features in the time series according to the time step, promote the indirect connection between LSTM output and related inputs, and show the completeness of the experiment. Finally, compared with the proposed model and other model combinations and a single model, the RMSE, MAE and MAPE values ​​of the main model are the smallest, and R 2 Specifically, compared with the baseline model CNN-LSTM, RMSE is reduced by 52.9%, MAE is reduced by 56.53%, MAPE is reduced by 58.1%, and R 2The results finally show that the proposed RIME-VMD-HFM-LFM model has the highest prediction accuracy and can reasonably handle multi-sector carbon emission prediction research.

[0080] Embodiment 2:

[0081] The present invention also discloses a high-resolution sector carbon emission prediction system based on a hybrid deep learning model, comprising:

[0082] A decomposition module, used for decomposing the carbon emission data into subsequences, wherein the subsequences include low-frequency subsequences and high-frequency subsequences;

[0083] A prediction module is used to construct a low-frequency prediction model and a high-frequency prediction model, and predict the low-frequency subsequence and the high-frequency subsequence respectively through the low-frequency prediction model and the high-frequency prediction model to obtain a low-frequency prediction result and a high-frequency prediction result;

[0084] The fusion module integrates the low-frequency prediction results and the high-frequency prediction results to obtain the final carbon emission prediction results.

[0085] Further, the decomposition module includes a first optimization unit and a decomposition unit;

[0086] The first optimization unit is used to obtain the optimal mode number and penalty factor of the variational mode decomposition model according to the frost optimization algorithm, and obtain the optimized variational mode decomposition model;

[0087] The decomposition unit is used to decompose the carbon emission data into subsequences through the optimized variational mode decomposition model.

[0088] Further, the prediction module includes a construction unit and a prediction unit;

[0089] The construction unit is used to construct a low-frequency prediction model based on a convolutional neural network, a long short-term memory network, and an attention mechanism network, and to construct a high-frequency prediction model based on a bidirectional recurrent network and a multi-layer perceptron;

[0090] The prediction unit is used to predict the low-frequency subsequence and the high-frequency subsequence respectively through the low-frequency prediction model and the high-frequency prediction model to obtain the low-frequency prediction result and the high-frequency prediction result.

[0091] Furthermore, the prediction module also includes a second optimization unit;

[0092] The second optimization unit is used to optimize the low-frequency prediction model through a sparrow search algorithm.

[0093] The present invention also discloses an electronic device, which comprises: a processor and a memory storing computer program instructions; the method is implemented when the processor executes the computer program instructions.

[0094] The present invention also discloses a computer storage medium, on which computer program instructions are stored, and the method is implemented when the computer program instructions are executed by a processor.

[0095] The present invention has the following technical advantages:

[0096] 1. The RIME-VMD-SSA-CNN-LSTM-ATTEN-BIGRU-MLP hybrid model is innovatively used in the field of carbon emission forecasting, and expands the traditional single-sector and regional-based carbon emission forecasting to sector-based carbon emissions, which has finer temporal high-resolution granularity and practical significance.

[0097] 2. Using deep learning and deep learning technology in the field of carbon emission prediction has a more accurate effect than previous statistical learning econometric models and environmental impact-based assessment models.

[0098] 3. A carbon emission prediction model structure with high temporal fine-grained resolution, RIME-VMD-SSA-CNN-LSTM-ATTEN-BIGRU-MLP, and an adaptive spectrum division prediction structure are proposed to process complex data into high-frequency and low-frequency data, and use different algorithms for processing.

[0099] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A high-resolution sector carbon emissions prediction method based on a hybrid deep learning model, characterized in that: The following steps are involved: Decomposing the carbon emission data into subsequences, wherein the subsequences include low-frequency subsequences and high-frequency subsequences; Constructing a low-frequency prediction model and a high-frequency prediction model, and predicting the low-frequency subsequence and the high-frequency subsequence respectively by using the low-frequency prediction model and the high-frequency prediction model to obtain a low-frequency prediction result and a high-frequency prediction result; The low-frequency prediction results and the high-frequency prediction results are integrated to obtain the final carbon emission prediction results.

2. The method according to claim 1, characterized in that Decomposing the carbon emission data into subsequences includes: The optimal number of modes and penalty factor of the variational mode decomposition model are obtained according to the frost optimization algorithm, and the carbon emission data are decomposed into subsequences through the optimized variational mode decomposition model.

3. The method according to claim 1, characterized in that The obtaining of the low-frequency prediction result comprises: An initial low-frequency prediction model is constructed according to a convolutional neural network, a long short-term memory network and an attention mechanism network. The initial low-frequency prediction model is optimized by a sparrow search algorithm to obtain a low-frequency prediction model. The low-frequency subsequence is predicted by the low-frequency prediction model to obtain a low-frequency prediction result.

4. The method according to claim 1, characterized in that: The obtaining of high frequency prediction results comprises: A high-frequency prediction model is constructed based on a bidirectional recurrent network and a multi-layer perceptron. The high-frequency subsequence is predicted by the high-frequency prediction model to obtain a high-frequency prediction result.

5. A high-resolution sector carbon emissions prediction system based on a hybrid deep learning model, characterized in that: include: A decomposition module, used for decomposing the carbon emission data into subsequences, wherein the subsequences include low-frequency subsequences and high-frequency subsequences; A prediction module is used to construct a low-frequency prediction model and a high-frequency prediction model, and predict the low-frequency subsequence and the high-frequency subsequence respectively through the low-frequency prediction model and the high-frequency prediction model to obtain a low-frequency prediction result and a high-frequency prediction result; The fusion module fuses the low-frequency prediction results and the high-frequency prediction results to obtain the final carbon emission prediction results.

6. The system according to claim 5, characterized in that The decomposition module includes a first optimization unit and a decomposition unit; The first optimization unit is used to obtain the optimal mode number and penalty factor of the variational mode decomposition model according to the frost optimization algorithm to obtain the optimized variational mode decomposition model; The decomposition unit is used to decompose the carbon emission data into subsequences through an optimized variational mode decomposition model.

7. The system according to claim 5, characterized in that The prediction module includes a construction unit and a prediction unit; The construction unit is used to construct a low-frequency prediction model based on a convolutional neural network, a long short-term memory network and an attention mechanism network, and to construct a high-frequency prediction model based on a bidirectional recurrent network and a multi-layer perceptron; The prediction unit is used to predict the low-frequency subsequence and the high-frequency subsequence respectively through the low-frequency prediction model and the high-frequency prediction model to obtain a low-frequency prediction result and a high-frequency prediction result.

8. The system according to claim 7, characterized in that The prediction module also includes a second optimization unit; The second optimization unit is used to optimize the low-frequency prediction model by using a sparrow search algorithm.

9. An electronic device, characterized in that: The device comprises: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the method according to any one of claims 1 to 4 is implemented.

10. A computer storage medium, characterized in that: The computer storage medium stores computer program instructions, which, when executed by a processor, implement the method according to any one of claims 1 to 4.

Images