Method and system for predicting carbon emission in transportation industry

Abstract

The invention discloses a transportation industry carbon emission prediction method and system, and relates to the field of transportation industry carbon emission prediction. In order to solve the defects that an existing model cannot cope with a multi-dimensional and multi-variable complex scene and is low in prediction precision, narrow in application range and poor in generalization ability, the method comprises the steps that influence factors are analyzed on the basis of an STIRPAT model; screening influence factors through a stepwise regression analysis method, and taking the influence factors meeting screening conditions as variables; the total amount of carbon emission generated by energy consumption of the provincial traffic transportation industry is measured and calculated in combination with a top-to-bottom method; a prediction model is constructed based on a CNN-BiLSTM-Attention model, the prediction model is optimized through historical carbon emission data, and hyper-parameters of the model are automatically adjusted by using a PSO algorithm; and predicting the carbon emission data of the preset year. The method is mainly used for carbon emission prediction in the transportation industry.

Description

Technical Field

[0001] The present invention relates to the field of carbon emission prediction in the transportation industry, and in particular to a method and system for predicting carbon emission in the transportation industry. Background Art

[0002] As one of the main sources of energy consumption and carbon emissions, the proportion of carbon emissions in the global total cannot be ignored. Due to the unique nature of the transportation industry, including high regional differences, diversity of transportation modes, and dynamic changes in carbon emission data, they together constitute a series of difficult problems in the prediction process. Although traditional models have demonstrated a certain degree of predictive ability in their respective fields, they often fail to fully capture the full picture of carbon emissions under the interweaving of these complex factors.

[0003] Specifically, the economic structure, transportation infrastructure, energy consumption patterns and policy orientations of different provinces are different, which directly affect the intensity and distribution of carbon emissions. At the same time, multiple modes of transportation such as road, rail, aviation and water transportation coexist, and each mode has different carbon emission characteristics, operating efficiency and development trends, which further increases the difficulty of prediction. In addition, as time goes by, factors such as technological progress, energy structure adjustment, and changes in transportation demand continue to act on the transportation system, making carbon emission data highly time-varying.

[0004] In the face of these challenges, traditional forecasting models often seem to be stretched. They may perform well under certain conditions, but when faced with complex scenarios with multiple dimensions and variables such as the transportation industry, it is often difficult to maintain stable forecasting accuracy and wide applicability. Therefore, we urgently need to introduce more advanced and flexible forecasting models and methods to meet these challenges.

[0005] Therefore, in order to address the carbon emission problem in the provincial transportation industry, we need a transportation industry carbon emission prediction method and system that can predict carbon emissions, analyze current carbon emissions, be applicable to complex and diverse carbon emission scenarios, adaptively adjust hyperparameters, and have high prediction accuracy and generalization capabilities. Summary of the invention

[0006] The present invention aims to solve the defects that existing models cannot cope with complex scenarios with multiple dimensions and variables, and the existing models have low prediction accuracy, narrow scope of application, and poor generalization ability. It provides a method and system for predicting carbon emissions in the transportation industry, which can predict carbon emissions, analyze current carbon emissions, be applicable to complex and diverse carbon emission scenarios, adaptively adjust hyperparameters, and have high prediction accuracy and generalization ability.

[0007] A method for predicting carbon emissions in the transportation industry according to the present invention comprises the following steps: S1. Analyze influencing factors based on the STIRPAT model; S2. Screen the influencing factors through stepwise regression analysis, and use the influencing factors that meet the screening conditions as variables; S3. Combine the top-down approach to calculate the total carbon emissions generated by energy consumption in the provincial transportation industry; S4. Build a prediction model based on the CNN-BiLSTM-Attention model, optimize the prediction model through historical carbon emissions data, and use the PSO algorithm to automatically adjust the hyperparameters of the model; S5. Forecast the carbon emission data for the preset year.

[0008] Further: In S1, the formula of the STIRPAT model is: ; Among them, I represents carbon emissions, P represents population size, A represents wealth, T represents technological level, k is the model coefficient, a, b and c are parameters to be estimated, and e is the error term.

[0009] Furthermore: the influencing factors include per capita GDP, population at the end of the year, gasoline energy consumption ratio, natural gas energy consumption ratio, urbanization rate, passenger turnover, cargo turnover, motor vehicle ownership and / or new energy vehicle ownership.

[0010] Further: In S2, the stepwise regression analysis method constructs the best model by adding or deleting independent variables in each step, and selects or excludes variables based on statistical significance, so as to screen out the influencing factors that contribute most to carbon emissions from all influencing factors of the STIRPAT model.

[0011] Furthermore: the specific steps of constructing the prediction model based on the CNN-BiLSTM-Attention model include: S41. Collect and pre-process the data of carbon emissions over the years; S42, using a convolutional neural network to extract features from the input time series data; S43, dividing the preprocessed data into a training set and a test set; S44. Select an evaluation indicator to evaluate the prediction performance of the model, and evaluate the performance of the prediction model on the validation set and the test set; S45. Use the trained CNN-BiLSTM-Attention model to predict carbon emissions.

[0012] Further: in S44, the evaluation indicators include mean square error MSE, root mean square error RMSE and / or mean absolute error MAE.

[0013] Further: In S5, the specific prediction steps of the CNN-BiLSTM-Attention model are as follows: S51, normalize the original data to eliminate the dimension, and then use the time sliding window function to process the time series data into a data type for supervised learning; S52, inputting the preprocessed time series data into the CNN network, and capturing carbon emission data features from the output data of the CNN network; S53, inputting the output data into the BiLSTM network, and capturing the dependency information between variables in the time dimension from the output data of the BiLSTM network; S54, taking all hidden states of each BiLSTM layer in the BiLSTM network as input, learning through the attention layer, and then calculating the weight of each feature through the fully connected layer with the activation function of Softmax and implementing the weight distribution through the Multiply layer; S55, automatically adjust the model's hyperparameters through the PSO algorithm; S56. According to the historical carbon emission data and the preset influencing factor indicators, the CNN-BILSTM-Attention model is used to predict the carbon emissions for the preset year to obtain a prediction result.

[0014] Further: in S45, the specific steps of automatically adjusting the hyperparameters of the model by the PSO algorithm are as follows: In the PSO algorithm optimization process, each particle represents a set of hyperparameter configurations. The PSO algorithm searches for the optimal hyperparameter combination by iteratively updating the position and velocity of the particle. In each iteration, the particle updates its position and velocity based on its own historical best position and the global best position. When the PSO algorithm converges or reaches the predetermined number of iterations, the particle with the highest fitness is selected as the optimal solution, and its corresponding hyperparameter configuration is applied to the final CNN-BiLSTM-Attention model training; The CNN-BiLSTM-Attention model optimized by the PSO algorithm is trained on the training set and verified on the verification set to ensure that the generalization ability and prediction performance of the CNN-BiLSTM-Attention model are optimized; Finally, the test set is used to evaluate the prediction performance of the optimized model.

[0015] The carbon emission prediction system for the transportation industry described in the present invention includes a STIRPAT model, a screening module, a CNN-BiLSTM-Attention model, a hyperparameter optimization module and a prediction model. The STIRPAT model is used to analyze the influencing factors; The screening module is used to screen the influencing factors through stepwise regression analysis. The CNN-BiLSTM-Attention model is used to construct a prediction model using the influencing factors that meet the screening conditions as variables, and optimize the prediction model through the national provincial transportation carbon emission data over the years; The hyperparameter optimization module is used to automatically adjust the hyperparameters of the model using the PSO algorithm; The prediction model is used to predict carbon emission data for a preset year.

[0016] Furthermore: the analysis dimensions of the influencing factors include population situation, economic development, technological progress and energy consumption.

[0017] The beneficial effects of the present invention are: The present invention adopts the combination of STIRPAT model and CNN-BiLSTM-Attention deep learning model to predict the carbon emissions of provincial transportation industry, which not only helps to understand the current status of carbon emissions, but also provides a scientific basis for formulating effective emission reduction policies and promotes the sustainable development of transportation industry.

[0018] The STIRPAT model provides us with a macro-analysis framework to help us understand the impact of key factors such as population, economy, and technology on carbon emissions; while the CNN-BiLSTM-Attention deep learning model, with its powerful data processing and feature extraction capabilities, effectively copes with the complexity and diversity of carbon emission data in the transportation industry. The introduction of this innovative method is not only a beneficial supplement to traditional forecasting models, but also a profound change in the field of carbon emission forecasting. By continuously optimizing model parameters and strengthening data fusion and feature extraction, more accurate and comprehensive provincial transportation carbon emission forecasts can be achieved, providing strong support for policy formulation and industry development.

[0019] First, by predicting carbon emissions in the provincial transportation industry, the present invention can gain an in-depth understanding of the current status and future development trends of carbon emissions in the transportation industry. Through accurate data and model analysis, it can clearly grasp the specific situation of carbon emissions in different regions, different modes of transportation, and different time periods, providing detailed and accurate decision-making references for policymakers; Secondly, the present invention combines the STIRPAT model and the CNN-BiLSTM-Attention deep learning model, and introduces the PSO algorithm (particle swarm optimization algorithm) for hyperparameter tuning. This combination demonstrates unprecedented innovation and significant advantages. Third, the STIRPAT model, with its macro-analysis capabilities, has laid a solid theoretical foundation for carbon emission forecasting, accurately capturing the deep impact of macro factors such as population, economy, and technology on carbon emissions. The CNN-BiLSTM-Attention deep learning model, with its excellent data processing capabilities, automatically extracts complex dynamic features and spatial patterns from time series data, effectively dealing with the high dimensionality, nonlinearity, and uncertainty of carbon emission data in the transportation industry; Finally, since the performance of deep learning models is highly dependent on the settings of their hyperparameters, we innovatively introduced the PSO algorithm to automatically adjust these hyperparameters, such as learning rate, number of hidden layer nodes, regularization coefficient, etc. The PSO algorithm simulates the group intelligence behavior of bird flocks foraging and searches for the optimal hyperparameter combination in the solution space, thereby further optimizing the model performance. This process not only reduces the workload of manual parameter adjustment, but also significantly improves the prediction accuracy and generalization ability of the model.

[0020] In summary, the provincial transportation industry carbon emission prediction method combining the STIRPAT model, CNN-BiLSTM-Attention deep learning model and PSO algorithm not only realizes the organic combination of macro and micro analysis at the theoretical level, but also improves the prediction performance of the model through hyperparameter optimization at the technical level. This innovative combination provides strong technical support for coping with the complex and ever-changing challenges of transportation industry carbon emission prediction, and has broad application prospects and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flowchart of the provincial transportation industry carbon emission prediction method based on the STIRPAT model, CNN-BiLSTM-Attention deep learning model, and PSO algorithm optimization; Figure 2 A logical diagram for predicting provincial transportation industry carbon emissions using the CNN-BiLSTM-Attention deep learning model; Figure 3 The training set and test set data for the transportation carbon emissions of Province S; Figure 4 This is the forecast data of carbon emissions from transportation in Province S from 2023 to 2030. DETAILED DESCRIPTION

[0022] The following are only preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a technician familiar with the technical field within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. The embodiments described below are only used to explain the present invention and cannot be interpreted as limitations on the present invention. The protection scope of the present invention should be based on the protection scope of the claims. The embodiments of the present invention are described in detail below. In order to facilitate the description of the present invention and simplify the description, the technical terms used in the specification of the present invention should be interpreted in a broad sense, including but not limited to conventional replacement schemes not mentioned in this application, and also including direct implementation methods and indirect implementation methods.

[0023] Example 1 Combination Figure 1-Figure 4 This embodiment describes a method for predicting carbon emissions in the transportation industry disclosed in this embodiment. The method for predicting carbon emissions in the provincial transportation industry based on the STIRPAT model, the CNN-BiLSTM-Attention deep learning model, and the PSO algorithm (particle swarm optimization algorithm) optimization includes the following steps: S1. Analyze influencing factors based on the STIRPAT model (Stochastic Impacts by Regression on Population, Affluence, and Technology, a scalable stochastic model for evaluating the impact of human factors on the environment); conduct in-depth analysis from four perspectives: provincial population situation P, affluence A, technological progress T, and carbon emissions I.

[0024] The formula of the STIRPAT model is: ; Among them, I represents carbon emissions, that is, energy consumption, P represents population size, A represents wealth, T represents technological level, k is the model coefficient, a, b and c are all parameters to be estimated, and e is the error term.

[0025] (1) Population situation Population is one of the important factors affecting energy consumption. At the provincial level, the population size P directly determines the total amount of energy demand. Population growth is usually accompanied by an increase in energy demand, because more residents mean more production, consumption and transportation activities, which require energy consumption to support. For example, provinces with large populations such as Fujian and Sichuan tend to have relatively high energy consumption. In addition, the population structure (such as age distribution, gender ratio, etc.) will also affect energy consumption patterns. For example, an increase in the proportion of the elderly population may increase the demand for energy-intensive services such as medical care and heating.

[0026] (2) Wealth Level A Affluence A, usually measured by GDP per capita, is another key driver of energy consumption. As the economy grows, residents' incomes increase and consumption levels rise, driving demand for energy-intensive products and services. For example, economically developed provinces such as Province J and Province G tend to consume more energy than relatively underdeveloped provinces. In addition, economic development has also promoted the development of industrialization, urbanization, and transportation, all of which require a lot of energy support. At the same time, the optimization of economic structure and industrial upgrading may also affect energy consumption patterns. For example, the development of high-tech industries and service industries may relatively reduce dependence on traditional energy.

[0027] (3) Technological progress Technological progress plays an important role in reducing energy consumption and improving energy efficiency. At the provincial level, technological progress can be reflected in many aspects, such as the improvement of industrial production processes, the improvement of energy conversion efficiency, and the promotion of energy-saving technologies. For example, as the forefront of technological innovation, the technological progress of S City and B City has made significant contributions to reducing energy consumption. Technological progress can not only reduce the energy consumption per unit of output, but also promote the development and application of clean energy and renewable energy, thereby further reducing dependence on traditional energy.

[0028] (4) Carbon emissions I Carbon emissions I, as the dependent variable of the STIRPAT model, is the result of the combined effect of the above three independent variables. At the provincial level, changes in carbon emissions are affected by multiple factors such as population, economic development, and technological progress. A comprehensive analysis of these factors can reveal the differences and commonalities in carbon emissions among different provinces. For example, some provinces may face greater carbon emission pressure due to their large population and developed economy; while other provinces may have achieved effective control of carbon emissions through technological progress and optimization of industrial structure.

[0029] S2. Screen the influencing factors through stepwise regression analysis, and use the influencing factors that meet the screening conditions as variables; After determining the basic framework of the STIRPAT model and identifying nine influencing factors under the four main categories of population situation P, affluence A, technological progress T and carbon emissions I at the provincial level, namely, GDP per capita, year-end population, gasoline energy consumption ratio, natural gas energy consumption ratio, urbanization rate, passenger turnover, cargo turnover, motor vehicle ownership, and new energy vehicle ownership, in order to further determine the specific impact and significance of these variables on carbon emissions I, we used stepwise regression analysis to screen and optimize the variables.

[0030] Stepwise regression analysis is an effective statistical method that can automatically select independent variables (i.e., P, A, and T) that have a significant impact on the dependent variable (here, carbon emissions I) through a series of iterative steps while considering all candidate variables, and exclude those variables that have insignificant impact or are multicollinear with other variables.

[0031] The specific steps are as follows: Initial model establishment: First, all candidate independent variables (P, A, T) are included in the model to establish an initial regression model that includes all possible influencing factors.

[0032] Variable significance test: Next, a significance test (such as t-test) is performed on each independent variable to assess whether their respective effects on carbon emissions I are significant.

[0033] Stepwise elimination of insignificant variables: Based on the results of the significance test, gradually eliminate those variables that have no significant impact on carbon emissions I. After each step of elimination, re-evaluate the significance of the remaining variables to ensure the stability and accuracy of the model.

[0034] Check for multicollinearity: When eliminating insignificant variables, you should also check whether there is multicollinearity between the remaining variables. If there is multicollinearity, that is, two or more independent variables are highly correlated, it may cause distortion of the model estimation results. Therefore, it is necessary to detect and deal with multicollinearity problems through methods such as variance inflation factor (VIF).

[0035] Determine the final model: After iterative screening in the above steps, a regression model containing independent variables that have a significant impact on carbon emissions I is finally determined. This model can more accurately reflect the actual impact of population, economic development and technological progress on carbon emissions.

[0036] The variables selected by stepwise regression analysis are not only statistically significant, but also can avoid redundant information and multicollinearity problems in the model, thereby improving the accuracy and reliability of model predictions. This is of great significance for us to deeply understand the influencing factors of carbon emissions at the provincial level and formulate effective energy policies.

[0037] Based on the STIRPAT model, the influencing factors were analyzed. For example, from the nine influencing factors, including per capita GDP, population at the end of the year, proportion of gasoline energy consumption, proportion of natural gas energy consumption, urbanization rate, passenger turnover, cargo turnover, number of motor vehicles, and number of new energy vehicles, six influencing factors were finally selected as variables to be incorporated into the prediction model, including the number of permanent population at the end of the year and per capita GDP, proportion of petroleum energy consumption, proportion of natural gas energy consumption, number of motor vehicles, and number of new energy vehicles.

[0038] ; Among them, I is carbon emissions, P is the permanent population at the end of the year, N is per capita GDP, T is the proportion of oil energy consumption, F is the proportion of natural gas energy consumption, J is the number of motor vehicles, X is the number of new energy vehicles, k is the model coefficient, and a, b, c, f, j, and x respectively represent the indexes of the above influencing factors.

[0039] S3. Combine the top-down method to calculate the total carbon emissions generated by energy consumption in the provincial transportation industry; use stepwise regression analysis to screen influencing factors as variables; combine the top-down method to calculate the total carbon emissions generated by the three types of energy consumption in the provincial transportation industry.

[0040] The top-down approach is a method of measurement from macro to micro. It is usually based on overall data at the national or regional level, and decomposes overall carbon emissions into specific industries, sectors or products through allocation coefficients or activity level data. In the measurement of carbon emissions in the provincial transportation industry, the following steps are adopted: Determine the total amount of energy consumption in the transportation industry at the provincial level: Collect energy consumption data for the transportation industry at the provincial level, including the consumption of major energy sources such as gasoline, coal, and electricity.

[0041] Obtain carbon emission factors for various energy sources: Carbon emission factors refer to the amount of carbon emissions generated per unit of energy consumption. For each energy source, we need to obtain its corresponding carbon emission factor. These factors may vary depending on factors such as energy type, combustion efficiency, emission control technology, etc., and need to refer to internationally or domestically recognized databases or research reports.

[0042] Calculate the carbon emissions of each type of energy: Use the product of energy consumption and carbon emission factor to calculate the carbon emissions generated by each energy in the transportation industry. That is, for each energy i, its carbon emissions Equal to its consumption Carbon emission factor The product of = × .

[0043] Summarize the total carbon emissions of the provincial transportation industry: Add the carbon emissions from the three energy sources to get the total carbon emissions of the provincial transportation industry. Equal to all energy carbon emissions The sum of .

[0044] Verification and adjustment: After obtaining the preliminary calculation results, they should be compared with existing statistical data, industry reports or international experience to ensure the rationality and accuracy of the calculation results. If necessary, parameters such as energy consumption and carbon emission factors can be adjusted according to actual conditions.

[0045] Analysis and application: Based on the calculation results, we can conduct an in-depth analysis of the carbon emissions of the provincial transportation industry, identify the main emission sources and emission reduction potential. At the same time, these results can also provide a scientific basis for formulating low-carbon development policies for the transportation industry and for enterprises to implement energy-saving and emission reduction measures.

[0046] By combining the top-down method, the total carbon emissions generated by the three energy consumption in the provincial transportation industry can be calculated more accurately, providing strong support for the formulation of effective emission reduction strategies.

[0047] S4. Construct a prediction model based on the CNN-BiLSTM-Attention model (a convolutional neural network based on the attention mechanism combined with a bidirectional long short-term memory neural network). Optimize the prediction model through the national provincial transportation carbon emissions data over the years, and use the advanced PSO algorithm to automatically adjust the model's hyperparameters, such as learning rate, number of hidden layer nodes, regularization coefficient, etc., to improve the prediction performance of the model, so that it can accurately reflect and predict the overall trend of national carbon emissions and specific changes at the provincial level. The specific process is as follows: (1) Data collection and preprocessing Data collection: Collect data on carbon emissions from transportation at provincial level across the country from authoritative institutions. These data should cover a sufficiently long time span to ensure that the model can learn long-term trends and seasonal changes in carbon emissions. Data cleaning: Clean the collected data, including processing missing values ​​(such as filling them through interpolation, regression prediction, etc.), outlier detection and correction, data format unification, etc.; Feature engineering: construct or select appropriate auxiliary features based on the factors affecting carbon emissions (such as economic development level, population size, transportation infrastructure conditions, energy structure, etc.) to enhance the model's predictive ability; (2) Model architecture design CNN layer: Use convolutional neural network (CNN) to extract features from input time series data. CNN can automatically learn local features in the data and extract high-level features useful for prediction through sliding and pooling operations of convolution kernels.

[0048] enter: , where n represents the sequence length (or the number of words in the text) and d represents the feature dimension (such as the dimension of word embedding).

[0049] Output: , where k is the kernel size and h is the number of kernels (or the number of output channels of the convolution layer). After the convolution operation, the sequence length may be reduced due to edge effects, depending on the kernel size and whether padding is used.

[0050] BiLSTM layer: BiLSTM is used to capture long-term dependencies in data. BiLSTM can consider both past and future information at the same time, thus more comprehensively understanding the contextual information of time series data.

[0051] enter: , which is the output of the CNN layer.

[0052] Output: , where 2h represents the hidden state dimension of BiLSTM. Since BiLSTM contains two LSTMs, the forward and backward LSTMs, the hidden state dimension is twice that of a single LSTM.

[0053] Attention layer: The attention mechanism is introduced after the BiLSTM layer, so that the model can automatically learn the importance of different time steps to the prediction results, and weight the output of BiLSTM accordingly to improve the accuracy of the prediction.

[0054] Input: O∈R(n−k+1)×2h, which is the output of the BiLSTM layer.

[0055] Output: , the Attention layer calculates the weights of each time step and applies these weights to the output of the BiLSTM layer to obtain the weighted summed feature vector M. This vector represents the important information of the entire sequence.

[0056] Fully connected layer (or output layer): The output M of the Attention layer is passed to one or more fully connected layers (also called dense layers or linear layers). These layers perform linear transformations on the input through learned weight matrices and bias terms, and may increase the nonlinear ability of the model and improve predictability through nonlinear activation functions (such as ReLU, sigmoid, or softmax). The fully connected layer maps the output of the Attention layer to the final prediction result.

[0057] enter: , which is the output of the Attention layer.

[0058] Output: (for regression tasks) or (For classification tasks, where C is the number of categories).

[0059] (3) Model training and optimization Data partitioning: The preprocessed data is divided into a training set and a test set. Data partitioning usually does not involve a specific calculation formula, but rather randomly divides the data set into two parts according to a certain ratio (such as 60% training set and 40% test set) to ensure that the model can show good generalization ability on unseen data.

[0060] Parameter initialization: Initialize the parameters in the CNN-BiLSTM-Attention model, including the convolution kernel size, step size, pooling method, number of LSTM units, and specific implementation of the attention mechanism.

[0061] Hyperparameter tuning: Through the PSO algorithm, the model's hyperparameters are tuned, such as learning rate, batch size, training rounds, etc., to improve the model's predictive performance.

[0062] Training process: Use the training set data to train the model and optimize the model parameters through the back-propagation algorithm and gradient descent method (or its variants) to minimize the prediction error.

[0063] Backpropagation: Calculate the gradient of the loss function with respect to the model parameters through the chain rule.

[0064] Gradient descent: Update the model parameters to minimize the loss function.

[0065] The parameter update formula is: ; Among them, θ is the model parameter, η is the learning rate, The loss function is The gradient at .

[0066] (4) Model evaluation and verification Evaluation indicators: Select appropriate evaluation indicators to evaluate the prediction performance of the model, including mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), etc.

[0067] Mean Square Error (MSE): ; Where N is the number of samples, is the true value of the i-th sample, is the model's predicted value for the i-th sample.

[0068] Root mean square error RMSE: ; Among them, RMSE is the square root of MSE. Compared with MSE, the dimension of RMSE is the same as that of the true value and the predicted value, which is easier to understand.

[0069] Mean absolute error (MAE): ; Among them, MAE measures the average absolute error between the predicted value and the true value, and is not as sensitive to outliers as MSE.

[0070] Validation set and test set evaluation: Evaluate the performance of the model on the validation set and test set to ensure that the model maintains high prediction accuracy on unseen data.

[0071] Comparative analysis: The prediction results of the CNN-BiLSTM-Attention model are compared with other benchmark models (such as single CNN, LSTM, etc.) to verify the effectiveness of the model.

[0072] (5) Model application and interpretation Predicting future carbon emissions: Use the trained CNN-BiLSTM-Attention model to make future predictions on national and provincial transportation carbon emissions, providing a scientific basis for policy making and emission reduction measures.

[0073] Result interpretation: Explain the model's prediction results, analyze the main factors affecting carbon emissions and their changing trends, and provide suggestions for formulating targeted emission reduction strategies.

[0074] Through the above steps, we can build a national provincial transportation carbon emissions prediction model based on the CNN-BiLSTM-Attention model, which can accurately reflect the overall trend of national carbon emissions and specific changes at the provincial level.

[0075] S5. Forecast the carbon emission data for the preset year.

[0076] For example, forecast provincial carbon emissions data from 2024 to 2030 to provide better data reference for carbon peak and carbon neutrality.

[0077] S1. Analyze influencing factors. Based on the STIRPAT model, analyze influencing factors from four perspectives: provincial population situation P, affluence level A, technological progress T, and carbon emissions I.

[0078] Data Collection: As shown in Table 1, data on the year-end permanent population, per capita GDP, technological progress indicators in the transportation industry (such as the proportion of oil energy consumption, the proportion of natural gas energy consumption), urbanization conversion rate, passenger turnover, cargo turnover, motor vehicle ownership, new energy vehicle ownership, and three major energy consumption in the transportation industry in Province S from 2014 to 2022 are collected.

[0079] S2. Convert the STIRPAT model into a linear form, namely:

[0080] Among them, P is the permanent population at the end of the year, N is the per capita GDP, T is the proportion of oil energy consumption, F is the proportion of natural gas energy consumption, E is the urban conversion rate, Z is the passenger turnover, Y is the cargo turnover, J is the number of motor vehicles, X is the number of new energy vehicles, I is carbon emissions, k is the model coefficient, a, b, c, e, f, z, y, j, x respectively represent the indexes of the above influencing factors.

[0081] Table 1: Data table of various influencing factors over the years

[0082] Based on the above values, we get b=0.522, c=0.146, d=0.136, f=0.163, e=0.054, z=−0.048, y=0.0001, j=0.154, x=0.138. Therefore, we find that the year-end resident population, per capita GDP, the proportion of oil energy consumption, the proportion of natural gas energy consumption, the number of motor vehicles and the number of new energy vehicles are significant influencing factors, while other factors are not statistically significant and are therefore eliminated.

[0083] S3. Combine the top-down method to calculate the total carbon emissions generated by the three energy consumptions in the transportation industry in Province S.

[0084] Energy consumption data collection: Based on Table 1, the consumption of the three main energy sources in the transportation industry in Province S from 2014 to 2022 is collected.

[0085] Obtaining carbon emission factors: Obtain the carbon emission factors of each energy source from relevant agencies.

[0086] Carbon emission calculation: According to the consumption of each energy source and the carbon emission factor, the carbon emissions of each energy source are calculated, and the sum is obtained to obtain the total carbon emissions of the transportation industry in Province S from 2014 to 2022 as shown in Table 2.

[0087] Table 2: Data of carbon emissions from transportation in S Province from 2014 to 2022 and the influencing factors after screening

[0088] S4. Based on the CNN-BiLSTM-Attention model, a prediction model is constructed through the transportation carbon emissions data of S Province over the years, and the advanced PSO algorithm is used to automatically adjust the model's hyperparameters to improve the model's prediction performance.

[0089] (1) Data preparation: The transportation carbon emissions data of Province S from 2014 to 2022 and the screened influencing factors (population, GDP per capita, gasoline energy consumption ratio, and natural gas energy consumption ratio) are used as input features.

[0090] (2) Model construction and training: Construct a CNN-BiLSTM-Attention model, where the CNN layer is used to extract the spatial features of the input data, the BiLSTM layer is used to capture the long-term dependencies in the time series data, and the Attention layer is used to enhance the model's attention to important time steps. Use the PSO algorithm to perform hyperparameter tuning, such as adjusting the learning rate, convolution kernel size, and number of LSTM units, to improve model performance. Figure 2 .

[0091] Use the training set data to train the model and optimize the model parameters through the back propagation algorithm. Figure 3 .

[0092] (3) Model evaluation and prediction: Evaluate the model performance on the validation set and the test set to ensure that the model has good generalization ability.

[0093] Use the trained model to predict the transportation carbon emissions in Province S in future years.

[0094] S5. Use the prediction model to predict the carbon emissions of Province S from 2023 to 2030.

[0095] By bringing the carbon emission data from 2014 to 2022 into the prediction model, we can get the predicted carbon emission data from 2023 to 2030: Forecasted carbon emissions in 2023: 9.9054 million tons, Forecasted carbon emissions in 2024: 9.8018 million tons, Forecasted carbon emissions in 2025: 9.9065 million tons, Forecasted carbon emissions in 2026: 10.1216 million tons, Forecasted carbon emissions in 2027: 10.2824 million tons, Forecasted carbon emissions in 2028: 10.3947 million tons, Forecasted carbon emissions in 2029: 9.9326 million tons, Forecasted carbon emissions in 2030: 9.7186 million tons, see Figure 4 .

[0096] The advantage of the fusion of the STIRPAT model and the CNN-BiLSTM-Attention deep learning model is that the data required by the STIRPAT model usually include macro data such as population, economy, and technology, while the CNN-BiLSTM-Attention model can process more complex time series data, such as traffic flow and vehicle operation data. The combination of the two realizes multi-dimensional data fusion from macro to micro, from static to dynamic, and provides a more comprehensive information basis for carbon emission prediction. Due to the combination of the macro analysis ability of the STIRPAT model and the complex sequence processing ability of the CNN-BiLSTM-Attention model, the prediction model performs well in capturing the long-term trend and short-term fluctuation of carbon emission changes, and improves the accuracy of the prediction. This prediction model is particularly suitable for carbon emission prediction in the provincial transportation industry. It can comprehensively consider various factors in this field, such as population, economy, technology, and traffic flow, and provide a scientific basis for the formulation of transportation emission reduction policies.

Claims

1. A method for predicting carbon emissions in the transportation industry, characterized in that: The steps include: S1. Analyze influencing factors based on the STIRPAT model; S2. Screen the influencing factors through stepwise regression analysis, and use the influencing factors that meet the screening conditions as variables; S3. Combine the top-down approach to calculate the total carbon emissions generated by energy consumption in the provincial transportation industry; S4. Build a prediction model based on the CNN-BiLSTM-Attention model, optimize the prediction model through historical carbon emissions data, and use the PSO algorithm to automatically adjust the hyperparameters of the model; S5. Forecast the carbon emission data for the preset year.

2. A method for predicting carbon emissions in the transportation industry according to claim 1, characterized in that: In S1, the formula of the STIRPAT model is: ； Among them, I represents carbon emissions, P represents population size, A represents wealth, T represents technological level, k is the model coefficient, a, b and c are parameters to be estimated, and e is the error term.

3. A method for predicting carbon emissions in the transportation industry according to claim 1, characterized in that: The influencing factors include per capita GDP, population at the end of the year, gasoline energy consumption ratio, natural gas energy consumption ratio, urbanization rate, passenger turnover, cargo turnover, motor vehicle ownership and / or new energy vehicle ownership.

4. A method for predicting carbon emissions in the transportation industry according to claim 1, characterized in that: In S2, the stepwise regression analysis method constructs the best model by adding or deleting independent variables in each step, and selects or excludes variables based on statistical significance, so as to screen out the influencing factors that contribute most to carbon emissions from all influencing factors of the STIRPAT model.

5. A method for predicting carbon emissions in the transportation industry according to claim 1, characterized in that: The specific steps of constructing a prediction model based on the CNN-BiLSTM-Attention model include: S41. Collect and pre-process the data of carbon emissions over the years; S42, using a convolutional neural network to extract features from the input time series data; S43, dividing the preprocessed data into a training set and a test set; S44. Select an evaluation indicator to evaluate the prediction performance of the model, and evaluate the performance of the prediction model on the validation set and the test set; S45. Use the trained CNN-BiLSTM-Attention model to predict carbon emissions.

6. A method for predicting carbon emissions in the transportation industry according to claim 5, characterized in that: In S44, the evaluation index includes mean square error (MSE), root mean square error (RMSE) and / or mean absolute error (MAE).

7. A method for predicting carbon emissions in the transportation industry according to claim 1, characterized in that: In S5, the specific prediction steps of the CNN-BiLSTM-Attention model are as follows: S51, normalize the original data to eliminate the dimension, and then use the time sliding window function to process the time series data into a data type for supervised learning; S52, inputting the preprocessed time series data into the CNN network, and capturing carbon emission data features from the output data of the CNN network; S53, inputting the output data into the BiLSTM network, and capturing the dependency information between variables in the time dimension from the output data of the BiLSTM network; S54, taking all hidden states of each BiLSTM layer in the BiLSTM network as input, learning through the attention layer, and then calculating the weight of each feature through the fully connected layer with the activation function of Softmax and implementing the weight distribution through the Multiply layer; S55, automatically adjust the model's hyperparameters through the PSO algorithm; S56. According to the historical carbon emission data and the preset influencing factor indicators, the CNN-BILSTM-Attention model is used to predict the carbon emissions for the preset year to obtain a prediction result.

8. A method for predicting carbon emissions in the transportation industry according to claim 7, characterized in that: In S55, the specific steps of automatically adjusting the hyperparameters of the model by the PSO algorithm are as follows: In the PSO algorithm optimization process, each particle represents a set of hyperparameter configurations. The PSO algorithm searches for the optimal hyperparameter combination by iteratively updating the position and velocity of the particle. In each iteration, the particle updates its position and velocity based on its own historical best position and the global best position. When the PSO algorithm converges or reaches the predetermined number of iterations, the particle with the highest fitness is selected as the optimal solution, and its corresponding hyperparameter configuration is applied to the final CNN-BiLSTM-Attention model training; The CNN-BiLSTM-Attention model optimized by the PSO algorithm is trained on the training set and verified on the verification set to ensure that the generalization ability and prediction performance of the CNN-BiLSTM-Attention model are optimized; Finally, the test set is used to evaluate the prediction performance of the optimized model.

9. A transportation industry carbon emission prediction system, characterized in that: Including STIRPAT model, screening module, CNN-BiLSTM-Attention model, hyperparameter optimization module and prediction model, The STIRPAT model is used to analyze the influencing factors; The screening module is used to screen the influencing factors through stepwise regression analysis. The CNN-BiLSTM-Attention model is used to construct a prediction model using the influencing factors that meet the screening conditions as variables, and optimize the prediction model through the national provincial transportation carbon emission data over the years; The hyperparameter optimization module is used to automatically adjust the hyperparameters of the model using the PSO algorithm; The prediction model is used to predict carbon emission data for a preset year.

10. A transportation industry carbon emission prediction system according to claim 9, characterized in that: The analysis dimensions of the influencing factors include population situation, economic development, technological progress and energy consumption.

Images