Carbon emission integrated prediction method considering driving factor fluctuation, equipment and medium

Abstract

The invention discloses a carbon emission integrated prediction method considering driving factor fluctuation, equipment and a medium, and relates to the technical field of carbon emission prediction. The prediction method comprises the following steps: establishing an integrated prediction model of the carbon emission based on a time sequence model and a multi-factor model; a Monte Carlo method is adopted to carry out dynamic scene simulation on the driving factors influencing the carbon emission, and the fluctuation range of the driving factors is obtained; performing random sampling on the driving factors in the fluctuation range to obtain a plurality of groups of sampling results; the multiple sets of sampling results are input into an integrated prediction model for prediction, and carbon emission prediction results corresponding to the multiple sets of sampling results are obtained; and obtaining a predicted carbon emission interval based on the confidence interval of the carbon emission prediction result. According to the method, the carbon emission integrated prediction model combining different time sequence methods and a multi-factor prediction method is constructed, the uncertainty of future carbon emission influence factors is considered through Monte Carlo simulation, and the accuracy and stability of carbon emission prediction are improved.

Description

Technical Field

[0001] This invention relates to the field of carbon emission prediction technology, and in particular to an integrated carbon emission prediction method, device and medium that takes into account fluctuations in driving factors. Background Technology

[0002] Existing carbon emission prediction methods can be categorized into three types: mathematical statistical methods, artificial intelligence methods, and hybrid methods. Mathematical statistical methods, such as regression methods and grey models, were the primary tools for early carbon emission prediction. The core of these methods is establishing a relationship function between multiple carbon emission influencing factors and carbon emissions. Artificial intelligence methods leverage big data and intelligent models, such as shallow intelligent models like Support Vector Machines (SVM) and Gradient Boosting Decision Trees, and deep neural network models like Backpropagation, Extreme Learning Machines, and Long Short-Term Memory (LSTM). The prediction accuracy of machine learning models is typically improved by optimizing algorithms and adjusting parameters, such as common hybrid prediction methods combining particle swarm optimization and genetic algorithms with machine learning models.

[0004] In practical applications, carbon emission prediction generally employs two methods: univariate time series forecasting and multivariate influencing factor forecasting. Time series forecasting methods can explore the autocorrelation characteristics of carbon emissions, reducing prediction uncertainty caused by model assumptions and the selection of multiple factors. Influencing factor forecasting methods can study the correlation between carbon emissions and other factors such as energy structure. However, univariate methods rely solely on past and current observational time series data, failing to consider relevant factors, leading to poor prediction results. Multivariate methods struggle to predict influencing factors; due to the diversity of datasets, no single model can outperform others in all situations. Different prediction models are applicable to different regions based on scenario settings, easily leading to error accumulation, all of which affect the accuracy and stability of the prediction. Summary of the Invention

[0005] This invention provides an integrated carbon emission prediction method, device, and medium that takes into account fluctuations in driving factors, in order to solve the problems of low prediction accuracy, one-sided prediction results, and inability to adapt to different scenarios in existing methods based on univariate time series prediction or multivariate influencing factors prediction.

[0006] This invention is achieved through the following technical solution:

[0007] A first aspect of the present invention provides an integrated carbon emission prediction method that considers fluctuations in driving factors, comprising:

[0008] An integrated prediction model for carbon emissions was established based on time series models and multi-factor models.

[0009] The Monte Carlo method was used to conduct dynamic scenario simulations of the driving factors affecting carbon emissions, and the fluctuation range of the driving factors was obtained.

[0010] Random sampling is performed on the driving factors within the fluctuation range to obtain multiple sets of sampling results;

[0011] Multiple sets of sampling results are input into the integrated prediction model for prediction, and carbon emission prediction results corresponding to each set of sampling results are obtained.

[0012] The predicted carbon emission range is obtained based on the confidence interval of the carbon emission prediction results.

[0013] This invention employs an ensemble approach to construct a carbon emission prediction model. It uses time-series and multi-factor prediction models as sub-models within the ensemble model, leveraging the information extraction capabilities of each sub-model to build an artificial intelligence model integrating these two models for carbon emission prediction. This model improves prediction accuracy and stability. Furthermore, Monte Carlo simulations are used to account for the uncertainties of future carbon emission influencing factors, enabling bottom-up prediction of regional carbon emission peaks and peak times under dynamic scenarios. This effectively addresses the problems of low prediction accuracy and biased results encountered in previous studies using econometric models and static scenario analysis. Finally, the prediction results of this invention can propose optimal development paths for regions, providing a basis and guidance for carbon emission reduction policy formulation.

[0014] Furthermore, the driving factors include population factors, economic factors, industrial factors, energy factors, trade factors, and electricity factors; the population factors include population size, the economic factors include GDP, the industrial factors include the proportion of the secondary industry, the energy factors include energy intensity, the trade factors include the total value of import and export products, and the electricity factors include electricity consumption and electrification rate.

[0015] Furthermore, the integrated prediction model for carbon emissions based on time series models and multi-factor models includes:

[0016] Historical carbon emission data was acquired and divided into training, testing, and validation sets according to the timeline.

[0017] The training set was used to train multiple candidate time series models and multiple candidate multi-factor models respectively;

[0018] The test set is input into the trained candidate time series model and candidate multi-factor model for prediction. The model performance evaluation index is calculated based on the prediction results. The best time series model and the best multi-factor model are selected based on the model performance evaluation index.

[0019] The ensemble weights of the optimal time series model and the optimal multi-factor model are calculated on the test set using the error inversion method. The optimal time series model and the optimal multi-factor model are then integrated based on the ensemble weights to obtain an ensemble prediction model. The validation set is used to verify the accuracy and stability of the ensemble prediction model.

[0020] Furthermore, the time series model includes a convolutional neural network (CNN), a fractional gray prediction model (FAGM(1,1)), and a long short-term memory network (LSTM); the multi-factor model includes a genetic algorithm-optimized neural network (GA-BP), a particle swarm optimization long short-term memory network (PSO-LSTM), and a particle swarm optimization support vector machine (PSO-SVM).

[0021] Furthermore, the optimal time series model is a convolutional neural network (CNN), and the optimal multi-factor model is a genetic algorithm-optimized neural network (GA-BP).

[0022] Furthermore, the calculation expressions for the ensemble weights of the optimal time series model and the optimal multi-factor model on the test set using the error inversion method are as follows:

[0023]

[0024] Among them, w t The ensemble weights for model t are represented by t=1, where t=1 indicates the optimal time series model and t=1 indicates the optimal multi-factor model. MAPE t The negative index represents model t. The negative index is calculated based on the model's prediction results on the test set and is used to measure the deviation between the model's predicted value and the actual value.

[0025] The output expression of the integrated prediction model is:

[0026] CE = w1CE P1 +w2CE P2 ;

[0027] Where CE represents the carbon emission prediction result of the ensemble prediction model, w1 represents the ensemble weights of the optimal time series model, and CE P1 This represents the carbon emission prediction result of the optimal time series model, where w2 represents the ensemble weights of the optimal multi-factor model, and CE P2 This represents the carbon emission prediction result of the optimal multi-factor model.

[0028] Furthermore, the model performance evaluation index is as follows:

[0029]

[0030] Among them, WEI iR represents the comprehensive performance evaluation index of candidate model i. 2 i MAPE i KGE i U1 i These represent the four performance evaluation index values ​​of candidate model i;

[0031] R 2 This is a positive indicator representing the goodness of fit of the model. The closer the value is to 1, the higher the explanatory power of the model. It is calculated as follows:

[0032]

[0033] Among them, y k This represents the predicted value for sample k. Let k represent the actual value of sample k, and n represent the number of samples. Indicates y k The average value;

[0034] MAPE is a negative indicator that uses a percentage to measure the deviation between model predictions and actual values. It is calculated as follows:

[0035]

[0036] Among them, y k This represents the predicted value for sample k. This represents the actual value of sample k, and n represents the number of samples.

[0037] KGE is a positive evaluation index that takes into account correlation, variability, and bias. Its value is close to 1, and the better the model performance, the better. It is calculated as follows:

[0038]

[0039] Where ρ is the correlation coefficient between the predicted value and the actual value, α is the variability ratio of the predicted value relative to the actual value, and β represents the deviation ratio between the average predicted value and the average actual value.

[0040] U1 is a negative indicator; the closer the value is to 0, the better the predictive performance. It is calculated as follows:

[0041]

[0042] Among them, y k This represents the predicted value for sample k. This represents the actual value of sample k, and n represents the number of samples.

[0043] Furthermore, the Monte Carlo method is used to perform dynamic scenario simulations of the driving factors affecting carbon emissions to obtain the fluctuation range of the driving factors, including:

[0044] The year to be predicted is divided into multiple prediction stages in chronological order, assuming that the average annual growth rate of the driving factor remains constant in each prediction stage.

[0045] The triangular distribution is used to describe the range of growth rate variation. The three key parameters of the triangular distribution are defined as the minimum growth rate, the maximum growth rate, and the initial set growth rate, which yields the fluctuation range of the driving factors.

[0046] A second aspect of the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the integrated carbon emission prediction method considering fluctuations in driving factors as described in any one of the first aspects of the present invention.

[0047] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the integrated carbon emission prediction method considering the fluctuations of driving factors as described in any one of the first aspects of the present invention.

[0048] Compared with existing technologies, this invention has the following advantages and beneficial effects: The integrated carbon emission prediction model constructed in this invention combines metaheuristic algorithms and neural networks, maximizing the accuracy and stability of predictions. Simultaneously, Monte Carlo simulation is utilized in the carbon emission prediction process from 2022 to 2035, and future uncertainties are considered by setting adjustment factors for influencing factors, achieving reliable predictions of peak time and peak value. Compared with single models, the integrated model constructed in this invention simultaneously considers historical time series data and the influencing factors of carbon emissions. Attached Figure Description

[0049] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the examples. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be considered as limiting the scope. A person of ordinary skill in the art can also derive other relevant drawings based on these drawings without inventive effort. In the drawings:

[0050] Figure 1 This is a flowchart of an integrated carbon emission prediction method considering the fluctuations of driving factors according to an embodiment of the present invention;

[0051] Figure 2 This is a flowchart of a method for establishing an integrated prediction model for carbon emissions according to an embodiment of the present invention;

[0052] Figure 3 This is a comparison chart of evaluation index values ​​of a candidate model according to an embodiment of the present invention;

[0053] Figure 4 This is a comparison chart of the relative error results between the integrated prediction model CNN-GA-BP and the best sub-models CNN and GA-BP;

[0054] Figure 5 This is a graph showing the predicted carbon emissions (CE) for 2022-2035, generated through an integrated model combined with Monte Carlo simulation. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0056] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims, and accompanying drawings of this invention are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to other steps or units inherent in the device.

[0057] The terminology used in the various embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the various embodiments of this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. 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 the various embodiments of this application pertain. The terms (such as those defined in a generally used dictionary) are to be interpreted as having the same meaning as in the context of the relevant technical field and are not to be interpreted as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of this application.

[0058] The embodiments of the present invention provide an integrated carbon emission prediction method, device and medium that takes into account the fluctuation of driving factors. It is applicable to the prediction of future carbon emission peaks and peak times by provincial / administrative region division, which helps to improve the accuracy of prediction, propose the optimal development path for the region, and provide corresponding basis and guidance for the formulation of carbon emission reduction policies.

[0059] like Figure 1 The diagram shows a flowchart of an integrated carbon emission prediction method that takes into account fluctuations in driving factors, according to an embodiment of the present invention, which includes the following steps.

[0060] S1, an integrated prediction model for carbon emissions is established based on time series models and multi-factor models.

[0061] S2 uses the Monte Carlo method to perform dynamic scenario simulations of the driving factors affecting carbon emissions, and obtains the fluctuation range of the driving factors.

[0062] S3, random sampling of driving factors within the fluctuation range to obtain multiple sets of sampling results.

[0063] S4. Input the multiple sampling results into the integrated prediction model for prediction to obtain the carbon emission prediction results corresponding to each sampling result.

[0064] S5, based on the confidence interval of the carbon emission prediction results, obtains the predicted carbon emission range.

[0065] To adapt to the diverse carbon emission characteristics influenced by regional economic and policy differences, this invention proposes an integrated carbon emission prediction model that can comprehensively capture the dynamic changes in time series data and the combined influence of multiple factors.

[0066] like Figure 2 The diagram shown is a flowchart of a method for establishing an integrated prediction model for carbon emissions according to this embodiment, which includes the following steps.

[0067] S1-1: Obtain historical carbon emission data and divide it into training set, test set and validation set according to timeline.

[0068] Taking the prediction of provincial carbon emissions as an example, the panel dataset of XX province from 2000 to 2021 is divided into a training set (2000-2015), a test set (2016-2020), and a validation set (2021) according to the time range. Therefore, the training set, test set, and validation set are independent data. When the model is trained, tested, and validated using the training set, test set, and validation set respectively, the datasets used by each other will not have cross-influence on the results, thus improving the accuracy of model testing and validation.

[0069] S1-2, training multiple candidate time series models and multiple candidate multi-factor models using the training set.

[0070] This embodiment selects multiple candidate models for each type of model for training, and determines the prediction model best suited to the characteristics of the current dataset. Taking the selection of three candidate models for each type of model as an example, for time series models, three types are selected: Convolutional Neural Network (CNN), Fractional Grey Prediction Model (FAGM(1,1), and Long Short-Term Memory Network (LSTM). For multi-factor models, three types are selected: Genetic Algorithm-Optimized Neural Network (GA-BP), Particle Swarm Optimized Long Short-Term Memory Network (PSO-LSTM), and Particle Swarm Optimized Support Vector Machine (PSO-SVM). Based on previous research and trial-and-error testing, the specific parameter values ​​for the six candidate models are shown in Table 1.

[0071] Table 1. Main parameter settings for the algorithm

[0072]

[0073] The candidate models are explained as follows:

[0074] CNN (Convolutional Neural Networks) is a type of feedforward neural network that includes convolutional computation and has a deep structure. It is one of the representative algorithms of deep learning.

[0075] LSTM (Long Short-Term Memory) is a type of recurrent neural network designed to address the long-term dependency problem inherent in general recurrent neural networks. All recurrent neural networks have a chain-like structure of repeating neural network modules.

[0076] GA-BP (Genetic Algorithm-Back Propagation) – This algorithm combines genetic algorithms with backpropagation neural networks to address the problems of backpropagation algorithms getting trapped in local minima and slow convergence.

[0077] PSO (Particle Swarm Optimization) is an algorithm that simulates the predation processes of flocks of birds and schools of fish in nature. It finds the global optimum of a problem through cooperation within the group.

[0078] SVM (Support Vector Machines) is a type of generalized linear classifier that performs binary classification of data using supervised learning. It incorporates a regularization term in the solution system to optimize structural risk, and its decision boundary is the hyperplane with the maximum margin calculated from the learning samples.

[0079] S1-3: Input the test set into the trained candidate time series model and candidate multifactor model for prediction, calculate the model performance evaluation index based on the prediction results, and select the best time series model and the best multifactor model based on the model performance evaluation index.

[0080] Four common model performance evaluation metrics were used to evaluate each candidate model on the test set. From each class of candidate models, one evaluation metric was selected to reflect the best model as the sub-model for ensemble learning. The specific expressions for the five metrics for model performance evaluation are shown in Table 2.

[0081] Table 2 Model Performance Evaluation Indicators

[0082]

[0083] The explanations for each indicator are as follows:

[0084] R 2 Goodness of Fit: This refers to how well the regression line fits the observed values. The statistic for measuring goodness of fit is the coefficient of determination (also known as the R-squared). 2 The maximum value is 1, R 2 The closer the R value is to 1, the better the regression line fits the observed values; conversely, the closer the R value is to 1, the better the regression line fits the observed values. 2 The smaller the value, the worse the regression line fits the observed values;

[0085] MAPE (Mean Absolute Percentage Error) refers to the average percentage of the absolute error between the predicted and actual values. It is used to evaluate the accuracy and error level of a prediction model.

[0086] In the table formula, y k These are the actual values ​​of the sample. Here, n is the predicted value, and ρ is the sample size. ρ is the correlation coefficient between the predicted and actual values, measuring the strength of their linear relationship. The formula for calculating ρ is: α represents the ratio of the predicted value to the actual value, reflecting the magnitude of the change in the predicted value relative to the change in the actual value. Its formula is: β represents the ratio of the average predicted value to the average actual value, measuring the degree of deviation of the overall predicted value from the actual value. Its formula is: σ y and These are the standard deviations of the actual value and the predicted value, respectively. and It is their average. R 2 The positive metric represents goodness of fit; the closer the value is to 1, the better the model's interpretability. MAPE is a negative metric, using a percentage to measure the magnitude of deviation. KGE is a positive evaluation metric that considers correlation, variability, and bias; a value closer to 1 indicates better model performance. U1 is a negative metric; a value closer to 0 indicates better predictive performance.

[0087] Based on the calculation results of each indicator, the candidate model with better performance across all indicators is selected as the sub-model for ensemble learning.

[0088] Furthermore, the comprehensive evaluation index WEI is obtained by combining the above four evaluation indicators, and the candidate models are evaluated.

[0089] Before calculating WEI, four basic indicators need to be standardized. For positive indicators, the standardization formula is:

[0090]

[0091] For negative indicators, the standardized formula is:

[0092]

[0093] In the above formula, Index i * The standardized evaluation index representing candidate model i is substituted into the four standardized indices R. 2* i MAPE i * KGE i * U1 i * They correspond to the index R respectively. 2 Standardized values ​​of MAPE, KGE, and U1, Index i Then substitute the four indicators R 2 i MAPE i KGE i U1 i The value of . The expression for the comprehensive evaluation index is:

[0094]

[0095] WEI i WEI represents the comprehensive performance evaluation index of candidate model i. Based on the performance of the trained candidate models on the test set, the best time series model and multi-factor model are selected for each category. The smaller the WEI value, the better the model performance.

[0096] S1-4. Using the error inversion method, calculate the ensemble weights of the best time series model and the best multi-factor model on the test set respectively. Integrate the best time series model and the best multi-factor model according to the ensemble weights to obtain the ensemble prediction model.

[0097] On the test set data, the weights of the two best models for each category are calculated using the error weight inversion method, as shown in the following formula:

[0098]

[0099] Among them, w t The ensemble weights for model t are represented by t=1, where t=1 indicates the optimal time series model and t=1 indicates the optimal multi-factor model. MAPE t This represents the negative index of model t. The negative index is calculated based on the model's prediction results on the test set and is used to measure the deviation between the model's predicted value and the actual value.

[0100] The final prediction result obtained by constructing the ensemble model is shown in the following formula:

[0101] CE = w1CE P1 +w2CE P2

[0102] Where CE represents the carbon emission prediction result of the ensemble prediction model, w1 represents the ensemble weights of the optimal time series model, and CE P1 This represents the carbon emission prediction result of the optimal time series model, where w2 represents the ensemble weights of the optimal multi-factor model, and CE P2 This represents the carbon emission prediction result of the optimal multi-factor model.

[0103] The validation set described above is used to verify the accuracy and stability of the ensemble prediction model. The optimal sub-model selected in steps S1-3 and the ensemble model obtained in S1-4 are used to predict the validation set data. The evaluation metric R is used to... 2 The accuracy and stability of the ensemble model built using MAPE, Kling-Gupta efficiency, Theil U1, and relative error validation outside of the sample data were evaluated.

[0104] Previous studies on carbon emission prediction have mostly used static scenarios, predicting future trends by setting fixed ratios, but neglecting the uncertainty of driving factors in the simulation. To address this, this invention utilizes Monte Carlo simulation to incorporate the volatility risk of driving factors into the prediction, establishing a dynamic scenario to obtain dynamic simulation results of carbon emissions.

[0105] This invention references the STIRPAT model and the Kaya equation decomposition model to expand the driving factors affecting carbon emissions into five categories: population, economy, technology, energy, and trade. Representative factor indicators for each category and their related explanations are listed in Table 3.

[0106] Table 3 Factors affecting carbon emissions

[0107]

[0108] This invention analyzes carbon emissions from multiple dimensions. The population factor is studied through the year-end resident population size, reflecting the "population effect"—that is, the larger the population, the higher the energy demand and carbon emissions. Based on the Environmental Kuznets Curve hypothesis, the technology effect is expressed as the ratio of secondary industry output to local GDP, highlighting its enormous energy consumption and the resulting carbon emissions. The energy effect focuses on energy intensity, calculated as the ratio of regional energy consumption to GDP in tons of standard coal equivalent, reflecting energy efficiency and its impact on emissions. Furthermore, the trade effect considers total import and export volumes to assess the scale of trade activities and their impact on emissions, as trade and transportation consume significant amounts of fossil fuels. Finally, the type and efficiency of fuel used in electricity production directly affect regional carbon emissions.

[0110] Furthermore, the Monte Carlo method was used to conduct dynamic scenario simulations of the driving factors affecting carbon emissions, specifically including:

[0111] S2-1 divides the year to be predicted into multiple prediction stages according to time sequence, assuming that the average annual growth rate of the driving factors remains constant in each prediction stage.

[0112] Taking the five-year plan that aligns with the development of XX Province as an example, the projected targets for 2022-2035 are divided into three phases: 2022-2025, 2026-2030, and 2031-2035, with the assumption that the average annual growth rate of each driving factor remains constant in each phase.

[0113] S2-2 uses a triangular distribution to describe the range of growth rate variation. The three key parameters of the triangular distribution are defined as the minimum growth rate, the maximum growth rate, and the initial set growth rate, thus obtaining the fluctuation range of the driving factors.

[0114] This invention establishes the distribution of each driving factor by defining a prior probability distribution of the uncertainty of the driving factors. It uses a triangular distribution to describe the possible range of growth rate fluctuations. Among the three key parameters of the triangular distribution, the lowest growth rate serves as the lower bound of the fluctuation range, and the highest growth rate serves as the upper bound. The initial growth rate is set as the most probable value. Therefore, based on the distribution of growth rates, the fluctuation range of each driving factor's value can be obtained, fully considering the uncertainty and fluctuation of the driving factors in the prediction, rather than treating them as static data.

[0115] Random sampling is performed on the triangular distribution of each driving factor. Each sampling is equivalent to randomly selecting a growth rate value from the distribution, and a single carbon emission prediction is made based on this value. The specific number of simulations depends on the actual situation, but preferably involves sampling 500 sets of random results, meaning the prediction model is simulated 500 times. Experiments show that further increasing the number of simulations does not significantly improve the prediction effect; on the contrary, it leads to a significant increase in computation time.

[0116] Finally, in step S5, based on the results of multiple simulations, and using the law of large numbers and the central limit theorem, the simulation results tend to follow a normal distribution. The carbon emission interval prediction value is obtained from the 95% confidence interval of the normal distribution.

[0117] Furthermore, the carbon emission integrated prediction method considering the fluctuations of driving factors proposed in this invention was experimentally tested. The best sub-model was selected for the integrated prediction model for four groups of regions using both training and test sets. Based on previous research and trial-and-error testing, the main parameter values ​​of the six candidate models are shown in Table 1. In addition, the general parameter settings for the two optimization algorithms, GA and PSO, are as follows: GA algorithm: population size 50, number of evolutions 10, parameter boundary [-2,2]; PSO algorithm: local and global search strengths 1.5 and 1.7, maximum number of iterations 1000, population size 30, and coefficients for rate and location changes 0.6.

[0118] The WEI (Weathering Indicators) comprehensively evaluates the carbon emission prediction results of the trained candidate models on the test set using four basic indicators and a comprehensive indicator. Figure 3 Table 4 shows a comparison of the evaluation metric values ​​of the candidate models in the dataset. Figure 3 It can be seen that the three sub-models LSTM, CNN, and GA-BP have better performance in carbon emission prediction. Table 4 further shows the comprehensive evaluation index values, indicating that the two best sub-models are CNN (WEI = 0.9918) and GA-BP (WEI = 0.9935).

[0119] Table 4. Comprehensive performance index (WEI) of candidate models

[0120]

[0121] Weights are calculated based on the errors between the predicted and actual values ​​of the two optimal sub-models in the test set, and then the ensemble prediction model is constructed. Using the 2021 dataset as the validation set, the relative errors of the ensemble prediction model CNN-GA-BP are compared with the prediction results of the optimal sub-models CNN and GA-BP. Figure 4 As shown, the RE of the ensemble prediction model is significantly lower than that of the CNN model and similar to that of the GA-BP model. It has a significant advantage in reducing error and can better combine the advantages of the two types of sub-models to improve the overall prediction accuracy.

[0122] Figure 5 The results show the carbon emission (CE) projections for 2022–2035 from the integrated model combined with Monte Carlo simulations. (a) shows the CE trend over the 95% confidence interval, and (b) shows the CE probability density function curves and peak times for each year of the simulation results.

[0123] according to Figure 5As shown, different development scenarios lead to significant differences in the carbon emission evolution trend of XX Province. Specifically, under the "Business as Usual (BAU)" scenario, XX Province's carbon emissions will peak in 2032, which is the latest peak time among all scenarios, with a peak of 341.38 million tons (Mt). In contrast, under the "Economic Development Scenario (SEG)" and "Sustainable Development Scenario (MS)," XX Province's carbon emissions will peak in 2031, slightly earlier than under the BAU scenario. The peak under the SEG scenario is 345.34 Mt, slightly higher than the 335.26 Mt under the MS scenario.

[0124] Although the carbon peaking time is the same under both the SEG and MS scenarios, achieving carbon peaking as early as possible is not the only goal from the perspective of XX Province's long-term development strategy. More importantly, while achieving carbon peaking, efforts should also be made to control cumulative carbon emissions to avoid irreversible environmental and climate impacts from long-term excessive emissions. Therefore, in designing future emission reduction pathways, governments at all levels and industries need to simultaneously focus on controlling both carbon peaking time and cumulative carbon emissions to achieve more sustainable green development goals.

[0125] The carbon emission integrated prediction model constructed in this invention combines metaheuristic algorithms and neural networks, maximizing the accuracy and stability of predictions. Furthermore, Monte Carlo simulations are utilized in the carbon emission prediction process from 2022 to 2035, and future uncertainties are considered by setting adjustment factors for influencing factors, achieving reliable predictions of peak time and peak value. Compared to single models, the integrated model constructed in this invention considers both historical time-series data and the influencing factors of carbon emissions. The relative error on the validation set is 0.0056, demonstrating improved prediction accuracy and stability when applied to different regions.

[0126] Embodiments of the present invention also provide an electronic device including a processor and a memory, wherein the number of processors may be one or more. The memory, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules. The processor executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the memory, thereby realizing the carbon emission integrated prediction method considering the fluctuations of driving factors according to any of the above embodiments of the present invention.

[0127] The memory may primarily comprise a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a given function; the data storage area may store data created based on terminal usage. Furthermore, the memory may include high-speed random access memory (RAM) and non-volatile memory, such as at least one disk storage device, flash memory, or other non-volatile solid-state storage device. In some instances, the memory may further include memory remotely located relative to the processor, which can be connected to the electronic device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks (LANs), mobile communication networks, and combinations thereof.

[0128] Embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the carbon emission integrated prediction method considering driving factor fluctuations of any embodiment of the present invention.

[0129] Embodiments of the present invention also provide a computer program product that, when run on a computer, causes the computer to execute the carbon emission integrated prediction method considering the fluctuations of driving factors according to any of the above embodiments of the present invention.

[0130] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An integrated carbon emission prediction method considering fluctuations in driving factors, characterized in that: include: Establish an integrated prediction model for carbon emissions based on time series model and multi-factor model; The Monte Carlo method is used to conduct dynamic scenario simulation on the driving factors that affect carbon emissions, and the fluctuation range of the driving factors is obtained; Randomly sampling the driving factors within the fluctuation range to obtain multiple groups of sampling results; Inputting the multiple groups of sampling results into the integrated prediction model for prediction respectively, and obtaining the carbon emission prediction results corresponding to each group of sampling results; A predicted carbon emission range is obtained based on the confidence interval of the carbon emission prediction result.

2. The integrated carbon emission prediction method considering fluctuations of driving factors according to claim 1 is characterized in that: The driving factors include demographic factors, economic factors, industrial factors, energy factors, trade factors and electricity factors; the demographic factors include population size, the economic factors include GDP, the industrial factors include the proportion of secondary industry, the energy factors include energy intensity, the trade factors include the total value of imports and exports, and the electricity factors include electricity consumption and electrification rate.

3. The integrated carbon emission prediction method considering fluctuations of driving factors according to claim 1 is characterized in that: The integrated prediction model of carbon emissions based on the time series model and the multi-factor model includes: Obtain historical carbon emission data and divide it into training set, test set and validation set according to the timeline; Using the training set, a plurality of candidate time series models and a plurality of candidate multi-factor models are trained respectively; Input the test set into the trained candidate time series model and the candidate multi-factor model for prediction, calculate the model performance evaluation index according to the prediction results, and select the best time series model and the best multi-factor model according to the model performance evaluation index; The error inversion method is used to calculate the integration weights of the best time series model and the best multi-factor model on the test set respectively, and the best time series model and the best multi-factor model are integrated according to the integration weights to obtain an integrated prediction model. The validation set is used to verify the accuracy and stability of the integrated prediction model.

4. The integrated carbon emission prediction method considering fluctuations of driving factors according to claim 3 is characterized in that: The time series model includes a convolutional neural network CNN, a fractional grey prediction model FAGM (1,1) and a long short-term memory network LSTM; the multi-factor model includes a genetic algorithm optimized neural network GA-BP, a particle swarm optimized long short-term memory network PSO-LSTM and a particle swarm algorithm optimized support vector machine PSO-SVM.

5. The integrated carbon emission prediction method considering driving factor fluctuations according to claim 4 is characterized in that: The optimal time series model is a convolutional neural network (CNN), and the optimal multi-factor model is a genetic algorithm optimized neural network (GA-BP).

6. The integrated carbon emission prediction method considering driving factor fluctuations according to claim 3 is characterized in that: The calculation expressions for calculating the integrated weights of the optimal time series model and the optimal multi-factor model respectively on the test set using the error inversion method are: Among them, w t represents the integrated weight of model t, t=1 represents the best time series model, t=1 represents the best multi-factor model, MAPE t Represents the negative indicator of model t. The negative indicator is calculated based on the model's prediction results for the test set and is used to measure the deviation between the model's predicted value and the actual value. The output expression of the integrated prediction model is: CE=w1CE P1 +w2CE P2 ; Among them, CE represents the carbon emission prediction result of the integrated prediction model, w1 represents the integrated weight of the best time series model, and CE P1 represents the carbon emission prediction result of the best time series model, w2 represents the integrated weight of the best multi-factor model, CE P2 Represents the carbon emission prediction results of the best multi-factor model.

7. The integrated carbon emission prediction method considering driving factor fluctuations according to claim 3 is characterized in that: The model performance evaluation index is: Among them, WEI i represents the comprehensive performance evaluation index of candidate model i, R 2* i 、MAPE i * , KGE i * 、U1 i * Represents the four performance evaluation indicators R of candidate model i 2 , MAPE, KGE, and U1 standardized values; R 2 It is a positive indicator, indicating the goodness of model fit. The closer the value is to 1, the higher the degree of model explanation. It is calculated as: Among them, y k represents the predicted value of sample k, represents the actual value of sample k, n represents the number of samples, Represents y k The average value of MAPE is a negative indicator that uses a percentage to measure the deviation between the model prediction value and the actual value. It is calculated as: Among them, y k represents the predicted value of sample k, represents the actual value of sample k, and n represents the number of samples; KGE is a positive evaluation index that takes into account correlation, variability, and bias. The closer its value is to 1, the better the model performance is. It is calculated as: Among them, ρ is the correlation coefficient between the predicted value and the actual value, α is the variability ratio of the predicted value relative to the actual value, and β represents the deviation ratio between the average predicted value and the average actual value; U1 is a negative indicator. The closer the value is to 0, the better the prediction performance is. It is calculated as: Among them, y k represents the predicted value of sample k, represents the actual value of sample k, and n represents the number of samples.

8. The integrated carbon emission prediction method considering driving factor fluctuations according to claim 1 is characterized in that: The Monte Carlo method is used to perform dynamic scenario simulation on the driving factors that affect carbon emissions to obtain the fluctuation range of the driving factors, including: Divide the years to be forecasted into multiple forecast stages in chronological order, assuming that the average annual growth rate of the driving factors in each forecast stage remains unchanged; The triangular distribution is used to describe the variation range of the growth rate. The three key parameters of the triangular distribution are defined as the minimum growth rate, the maximum growth rate and the initial set growth rate, and the fluctuation range of the driving factors is obtained.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the integrated carbon emission prediction method considering the fluctuation of driving factors described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the integrated carbon emission prediction method considering the fluctuation of driving factors described in any one of 1-8 is implemented.

Images