Carbon emission prediction method for cigarette production enterprise
By applying the LSTM-Attention model to predict carbon emissions in cigarette manufacturers, the problem of enterprises lacking specific prediction methods is solved, and higher prediction accuracy and more accurate carbon emission reduction measures are achieved.
Patent Information
- Application Number
- CN202510020736.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-07
- Publication Date
- 2025-05-06
AI Technical Summary
Cigarette manufacturers lack specific carbon emission forecast methods, which leads to the inability of enterprises to be targeted and accurate when implementing emission reduction measures.
Using the carbon emission prediction method based on LSTM-Attention, we will determine the carbon emission boundary, collect and preprocess carbon emission activity data, and build an LSTM-Attention model for training, and finally combine carbon emission factors to predict carbon emissions.
It has improved the accuracy of carbon emission forecasting for cigarette manufacturers and provided a systematic and operational carbon emission forecasting method to help the industry achieve more accurate carbon emission reduction and green development goals.
Smart Images
Figure CN119940627A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a carbon emission prediction method for a cigarette production enterprise, and belongs to the field of carbon emission prediction in cigarette production. Background Art
[0002] As a major energy consumer, the tobacco industry produces a large amount of carbon dioxide during its production process. In recent years, the country has introduced a number of energy-saving and carbon-reduction policies, explicitly requiring tobacco companies to initially establish a green, low-carbon, circular development system by 2025 to reduce energy consumption and carbon emissions.
[0003] According to the technical solution disclosed in Chinese invention patent CN117035167B, a method for predicting corporate carbon emissions is proposed. This method uses dynamic time warping (DTW) to screen historical energy consumption data that is highly correlated with corporate carbon emissions as a training set, and combines the long short-term memory network (LSTM) model to predict corporate carbon emissions. Although LSTM can alleviate the gradient vanishing problem, it still has certain limitations in extracting features from long time series, resulting in insufficient prediction accuracy.
[0004] In addition, my country's research on carbon emissions prediction for cigarette manufacturers is still relatively scarce, resulting in a general lack of scientific and specific carbon emissions prediction methods in the industry, making it difficult for companies to be targeted and accurate when implementing emission reduction measures. Summary of the invention
[0005] The purpose of the present invention is to provide a carbon emission prediction method for a cigarette production enterprise, so as to solve the problem of lack of a specific carbon emission prediction method in the cigarette industry.
[0006] In view of the above problems, the present invention provides a method for predicting carbon emissions of cigarette manufacturers based on LSTM-Attention to achieve carbon emissions prediction for cigarette manufacturers and improve the problem of insufficient prediction accuracy of existing carbon emissions prediction methods. Specifically, the method includes the following steps:
[0007] Step S1: Determine the carbon emission boundary according to the entire process of cigarette production;
[0008] Step S2: clarify the data types and sources of carbon emission activity data in cigarette production;
[0009] Step S3: Collect basic data based on the source of carbon emission activity data, and pre-process the data according to its characteristics to obtain data features;
[0010] Step S4: construct an LSTM-Attention carbon emission activity data prediction model based on the long short-term memory network LSTM and the self-attention mechanism;
[0011] Step S5: Input data features into LSTM-Attention for model training and optimize model parameters;
[0012] Step S6: input the data to be predicted into the trained model to obtain the predicted value of carbon emission activity data;
[0013] Step S7: The predicted value of the activity data is combined with the carbon emission factor to predict the carbon emission.
[0014] Furthermore, the types of carbon emission activity data in cigarette production in step S2 include fossil fuel combustion, production process emissions, and net purchased electricity and heat; the sources of the carbon emission activity data are the net purchased electricity used by refrigeration stations, air compressor stations, and production equipment, the natural gas used by steam boilers, the amount of gasoline and diesel used in production and transportation, power generation, and the amount of purchased steam used by tobacco drying machines and rehumidification machines.
[0015] Further, the basic data in step S3 include net purchased electricity, electricity consumption of refrigeration station, electricity consumption of air compression station, electricity consumption of production line AB, electricity consumption of production line C, purchased steam, natural gas, gasoline, diesel, actual output, planned output, product type (coarse / fine), silk-making workshop temperature, silk-making workshop humidity, smoke moisture content, local temperature, local humidity, local weather, local season;
[0016] The preprocessing in step S3 includes: filling missing data by interpolation; performing one-hot encoding on categorical variables, converting product type and season into an independent binary vector; and performing data normalization using the minimum-maximum normalization method to scale the range of numerical features to the range of [0,1].
[0017] The normalization formula is:
[0018] Among them, x′ is the normalized data; x is the original data; x min is the minimum value of the column in the data set; x max It is the largest value of the column in the data set;
[0019] The processed data are used to construct data features, and the data set is divided into training set, test set, and validation set in the ratio of 70%, 15%, and 15%.
[0020] Furthermore, the carbon emission activity data prediction model in step S4 includes an input layer, an LSTM layer, an Attention layer, a fully connected layer and an output layer;
[0021] In order to capture the short-term and long-term dependencies in the time series, a two-layer LSTM is used to extract the features of the time series; the attention weight is calculated for the hidden state sequence output by the LSTM through the self-attention mechanism, emphasizing the key time points and feature combinations that have an important impact on the prediction of carbon emissions; and the predicted value of carbon emission activity data is output through a fully connected layer;
[0022] The core formula of the Attention layer is:
[0023] Among them, Q is Query, K is Key, V is Value, d k is the dimension of the hidden layer.
[0024] Furthermore, the model training in step S5 includes the following parts: inputting the training set data into the constructed neural network model, using mean square error (MSE) as the loss function, and optimizing the model parameters using the Adam optimization algorithm;
[0025] The formula for calculating the mean square error is:
[0026] where y true,i is the true value of the i-th sample, y pred,i is the predicted value of the i-th sample, and n is the total number of samples;
[0027] After each training round, the validation set is used to evaluate the model performance to prevent the model from overfitting; finally, the best carbon emission activity data prediction model is saved.
[0028] Furthermore, the predicted value of the carbon emission activity data in step S6 is a normalized value, and it is necessary to use inverse normalization to obtain the original dimension, that is, the true predicted value of the net purchased electricity, purchased steam, natural gas consumption, gasoline consumption, and diesel consumption.
[0029] Furthermore, the carbon emissions prediction in step S7 includes the following steps: according to the basic carbon accounting equation provided by the IPCC, greenhouse gas (GHG) emissions = activity data (AD) × emission factor (EF), calculate the carbon emissions generated by the actual predicted value of each activity data in step S6; add up the carbon emissions generated by all activity data to obtain the total carbon emissions of the cigarette manufacturing enterprise.
[0030] The present invention combines multi-layer LSTM with the Attention mechanism to propose a carbon emissions prediction method for cigarette manufacturers. With its powerful time series processing capabilities, LSTM solves the shortcomings of traditional methods in capturing long-term dependent features. At the same time, the introduced Attention mechanism automatically assigns attention weights in the hidden state sequence output by LSTM, highlighting the key time points and feature combinations that are crucial to carbon emissions prediction. Through this innovative method, the accuracy of carbon emissions prediction for cigarette manufacturers is improved, and a systematic and operational carbon emissions prediction method is provided for the cigarette industry, helping the industry achieve more accurate carbon emission reduction and green development goals. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 This is a specific implementation flow chart of the carbon emission prediction method for cigarette production enterprises provided by the present invention;
[0032] Figure 2 This is the cell architecture diagram of LSTM;
[0033] Figure 3 It is a schematic diagram of the neural network structure of the carbon emission prediction method for cigarette production enterprises provided by the present invention. DETAILED DESCRIPTION
[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0035] like Figure 1 The specific implementation flow chart of the carbon emission prediction method for cigarette manufacturing enterprises of the present invention is shown, and the specific implementation steps are as follows:
[0036] Step S1: Determine the carbon emission boundary based on the entire process of cigarette production.
[0037] Cigarette production requires multiple processes such as unpacking, silk making, blending and flavoring, and wrapping, among which the silk making stage emits the largest amount of carbon dioxide. By analyzing the entire process of cigarette production, the emissions generated by emission sources owned or controlled by cigarette manufacturers and the indirect emissions generated by the purchased or acquired electricity, steam, heating or cooling consumed by cigarette manufacturers are determined as carbon emission boundaries.
[0038] Step S2: Clarify the types and sources of carbon emission activity data in cigarette production.
[0039] By analyzing the technological process of cigarette production, it is concluded that the types of carbon emission activity data in cigarette production include fossil fuel combustion, production process emissions, and net purchased electricity and heat; the sources of carbon emission activity data are the net purchased electricity used by refrigeration stations, air compressor stations, and production equipment; natural gas used by steam boilers; the amount of gasoline and diesel used in production, transportation, and power generation; and the amount of purchased steam used by tobacco drying machines and rehumidification machines.
[0040] Step S3: Collect basic data based on the source of carbon emission activity data, and pre-process the data according to its characteristics to obtain data features.
[0041] Relevant data are collected based on the data sources of carbon emission activities, including time, net purchased electricity, electricity consumption of refrigeration station, electricity consumption of air compressor station, electricity consumption of production line AB, electricity consumption of production line C, purchased steam, natural gas, gasoline, diesel, actual output, planned output, product type (coarse / fine), silk-making workshop temperature, silk-making workshop humidity, water content of smoke, local temperature, local humidity, local weather, and local season.
[0042] The data collected above are used to fill in the missing data through linear interpolation;
[0043] Linear interpolation formula:
[0044] Among them, x i and x i+1 are the values of adjacent independent variables with known data, y i and i+1 Corresponding to x i and x i+1 The dependent variable value of the known data, y is the predicted value after interpolation;
[0045] Product type (coarse / fine) and local weather are categorical variables. This feature is discrete and unordered, so one-hot encoding is used to map product type and local weather into binary vectors, in which only one element is 1 and the rest are 0;
[0046] The minimum-maximum normalization method is used to normalize numerical features and scale the value range to the range of [0,1] to reduce the scale difference between different features and accelerate model convergence;
[0047] The normalization formula is:
[0048] Among them, x′ is the normalized data; x is the original data; x min is the minimum value of the column in the data set; x max It is the largest value of the column in the data set;
[0049] Finally, the processed data is used to construct data features, and the data set is divided into training set, test set and validation set in a ratio of 70%, 15% and 15%. The training set is used for model training and parameter optimization; the test set is used to evaluate the generalization ability and prediction effect of the model; the validation set is used to monitor model performance during training, prevent overfitting, and adjust model hyperparameters after each training cycle to ensure the best performance of the model.
[0050] Step S4: Construct a LSTM-Attention carbon emission activity data prediction model based on the long short-term memory network LSTM and self-attention mechanism.
[0051] LSTM is a deep learning model commonly used to process sequence data. LSTM contains a memory unit for storing and updating the state information of historical moments, and three gate units, input gate, forget gate and output gate, for controlling the flow and forgetting of information. Figure 2 The cell architecture diagram of the LSTM is shown.
[0052] Among them, c t-1 is the cell state at the previous moment, h t-1 is the value of the hidden layer at the previous moment, x t is the current input at time t.
[0053] Forget gate: decides which information should be discarded or retained. Passes the information from the previous hidden state and the current input information to the sigmoid function at the same time. The output value is between [0,1]. The closer to 0, the more it should be discarded, and the closer to 1, the more it should be retained.
[0054] Forget gate formula: f t =σ(W f ·[h t-1 ,x t ]+b f );
[0055] Where σ is the sigmoid function, W f and b f are the forgotten weights and bias terms, respectively, and f t is the output of the forget gate.
[0056] Input gate: used to update cell status;
[0057] i t =σ(W i ·[h t-1 ,x t ]+b i ),
[0058] Among them, W i and Wc is the input gate weight, b i and b c is the bias term of the input gate, tanh() is the activation function, i t is the input gate output, is the candidate cell state.
[0059] Cell state: The cell state of the previous layer is multiplied by the forget vector point by point;
[0060]
[0061] Among them, c t is the cell state at the current moment, and ⊙ is the dot product.
[0062] Output gate: The output gate is used to determine the value of the next hidden state, which contains the information of the previous input;
[0063] o t =σ(W o ·[x t ,h t-1 ]+b o ), h t =o t ⊙tanh(c t );
[0064] Among them, h t is the output at time t, W o and b o are the output gate weights and bias terms.
[0065] This example implements the LSTM-Attention model based on the deep learning framework PyTorch. Figure 3 The schematic diagram of the neural network structure of the method for predicting carbon emissions of cigarette manufacturers according to the present invention is shown, which specifically includes: an input layer, an LSTM layer, an Attention layer, a fully connected layer and an output layer.
[0066] In order to capture the short-term and long-term dependencies in the time series, a two-layer LSTM is used to extract the features of the time series, and the self-attention mechanism is used to enhance the model's attention to different time steps. Finally, the model outputs the predicted value of carbon emission activity data through a fully connected layer.
[0067] The input layer passes the data of each sample in the appropriate shape to the LSTM layer.
[0068] The first layer of LSTM processes recent information in the time series by capturing short-term dependencies. For example, the temperature and humidity in the past few days will affect the amount of purchased steam and the electricity consumption of the refrigeration station in the activity data of the day. Furthermore, the hidden state of each time step is generated and passed to the second layer of LSTM to further model more complex time dependencies.
[0069] The second layer of LSTM further processes the hidden state based on the first layer, mainly capturing the dependencies in the time series over a longer time range, such as seasonal fluctuations, annual periodicity, etc. These are information that are not easy to capture in the short term, and the output is the hidden state of each time step, which contains the characterization of the long-term trend in the sequence.
[0070] The Attention layer enhances the model's attention to different time steps in the sequence through the self-attention mechanism, enabling the model to automatically decide which time steps are more important for the current prediction task.
[0071] The core formula of the Attention layer is:
[0072] Among them, Q is Query, K is Key, V is Value, d k is the dimension of the hidden layer.
[0073] The self-attention process is as follows:
[0074] First, the output of the second layer LSTM is linearly transformed to obtain Query, Key and Value. The attention weight is calculated through Query, Key and Value. Then, the dot product of the query vector and the key vector is used to obtain the correlation between each time step (i.e., the attention score).
[0075] Finally, the Softmax function is used to calculate the normalized weights of the scores, and these weights are used to perform weighted summation on the Values to obtain the enhanced output.
[0076] The fully connected layer is used to map the final hidden state to the predicted output space and output the target variable, namely, five carbon emission activity data: net purchased electricity, purchased steam, natural gas usage, gasoline usage, and diesel usage.
[0077] The number of samples in each batch during training is set to 64; the number of training rounds is set to 1000; the hidden layer dimensions of the LSTM and self-attention layers are set to 128; the learning rate of the Adam optimizer is set to 0.001; and the length of the input sequence is set to 7.
[0078] Step S5: Input data features into LSTM-Attention for model training and optimize model parameters;
[0079] The data in the training set is used as input features and input into the LSTM-Attention model. The specific training process is as follows:
[0080] The input data passes through the model input layer, LSTM layer, attention layer and fully connected layer to calculate the predicted value y pred .
[0081] Calculate the predicted value y for the current batch pred and the true value y true The mean square error MSE of the current batch is obtained, and the loss value loss of the current batch is obtained. The smaller the error, the better the model performance.
[0082] The formula for calculating the mean square error is:
[0083] where y true,i is the true value of the i-th sample, y pred,i is the predicted value of the i-th sample, and n is the total number of samples;
[0084] After calculating the loss value, the Adam optimizer (adaptive learning rate optimizer) is used to dynamically adjust the learning rate of each parameter to improve the convergence speed.
[0085] To avoid model overfitting, after each training round, the validation set is used to evaluate the model performance. Whenever the validation loss reaches a new minimum value, the current model parameters are saved. When the loss on the validation set no longer improves, the training is stopped early.
[0086] At the end of training, load the model parameters that gave the lowest validation loss and save the best model.
[0087] Step S6: Input the data to be predicted into the trained model to obtain the predicted value of carbon emission activity data.
[0088] The data in the test set is converted into tensors and input into the trained model, which outputs five carbon emission activity data: predicted values of net purchased electricity, purchased steam, natural gas consumption, gasoline consumption, and diesel consumption.
[0089] Further denormalize the model predictions from the normalized state to the original dimension;
[0090] The formula for denormalization is: x = x' (x max -x min )+x min ;
[0091] Where x′ is the normalized carbon emission activity data output by the model, x min With x maxis the minimum and maximum value of each target variable.
[0092] Step S7: Combine the predicted value of the activity data with the carbon emission factor to predict the carbon emissions.
[0093] The actual predicted value of carbon emission activity data obtained after denormalization, i.e. AD 净购入电力 , A.D. 净购入蒸汽 , A.D. 天然气用量 , A.D. 汽油用量 , A.D. 柴油用量 According to the basic carbon accounting equation provided by the Intergovernmental Panel on Climate Change (IPCC), the carbon emissions generated by the true predicted value of each activity data are calculated and the carbon emissions generated by all activity data are added together to obtain the total carbon emissions of cigarette manufacturers.
[0094] Basic equation for carbon accounting: Greenhouse gas (GHG) emissions = activity data (AD) × emission factor (EF);
[0095] Among them, AD is the amount of production or consumption activities that lead to greenhouse gas emissions, such as the consumption of each fossil fuel, the consumption of limestone raw materials, the net purchased electricity, the net purchased steam, etc.;
[0096] EF is a coefficient corresponding to the activity level data, including carbon content per unit calorific value or elemental carbon content, oxidation rate, etc., which characterizes the greenhouse gas emission coefficient per unit of production or consumption activity. EF can be directly based on known data (i.e. default values) provided by IPCC, the U.S. Environmental Protection Agency, the European Environment Agency, etc., or it can be estimated based on representative measurement data. my country has set national parameters based on actual conditions. For example, Appendix 2 of the "Guidelines for the Calculation and Reporting of Greenhouse Gas Emissions from Other Industrial Industries (Trial)" provides default value data for common fossil fuel characteristic parameters.
Claims
1. A method for predicting carbon emissions for a cigarette manufacturing enterprise, characterized in that: The method comprises the following steps: Step S1: Determine the carbon emission boundary according to the entire process of cigarette production; Step S2: clarify the data types and sources of carbon emission activity data in cigarette production; Step S3: Collect basic data based on the source of carbon emission activity data, and pre-process the data according to its characteristics to obtain data features; Step S4: Constructing a LSTM-Attention carbon emission activity data prediction model based on the long short-term memory network LSTM and self-attention mechanism; Step S5: Input data features into LSTM-Attention for model training and optimize model parameters; Step S6: input the data to be predicted into the trained model to obtain the predicted value of carbon emission activity data; Step S7: Combine the predicted value of the activity data with the carbon emission factor to predict the carbon emissions.
2. The carbon emission prediction method for a cigarette manufacturing enterprise according to claim 1, characterized in that: The types of carbon emission activity data in cigarette production in step S2 include fossil fuel combustion, production process emissions, and net purchased electricity and heat; the sources of carbon emission activity data are net purchased electricity used by refrigeration stations, air compressor stations, and production equipment; natural gas used by steam boilers; the amount of gasoline and diesel used in production, transportation, and power generation; and the amount of purchased steam used by tobacco drying machines and rehumidification machines.
3. The carbon emission prediction method for a cigarette manufacturing enterprise according to claim 1, characterized in that: The basic data in step S3 include net purchased electricity, electricity consumption of refrigeration station, electricity consumption of air compressor station, electricity consumption of production line AB, electricity consumption of production line C, purchased steam, natural gas, gasoline, diesel, actual output, planned output, product type (coarse / fine), silk-making workshop temperature, silk-making workshop humidity, smoke moisture content, local temperature, local humidity, local weather, local season; The preprocessing in step S3 includes: filling missing data by interpolation; performing one-hot encoding on categorical variables, converting product type and season into an independent binary vector; and performing data normalization using the minimum-maximum normalization method to scale the range of numerical features to the range of [0,1]. The normalization formula is: Among them, x′ is the normalized data; x is the original data; x min is the smallest value in the data set; x max is the largest value in the data set; The processed data are used to construct data features, and the data set is divided into training set, test set, and validation set in the ratio of 70%, 15%, and 15%.
4. The carbon emission prediction method for a cigarette manufacturing enterprise according to claim 1, characterized in that: The carbon emission activity data prediction model in step S4 includes an input layer, an LSTM layer, an Attention layer, a fully connected layer and an output layer; In order to capture the short-term and long-term dependencies in the time series, a two-layer LSTM is used to extract the features of the time series; The attention weights are calculated for the hidden state sequence output by LSTM through the self-attention mechanism, emphasizing the key time points and feature combinations that have a significant impact on carbon emission prediction; the predicted value of carbon emission activity data is output through a fully connected layer; The core formula of the Attention layer is: Among them, Q is Query, K is Key, V is Value, d k is the dimension of the hidden layer.
5. The carbon emission prediction method for a cigarette manufacturing enterprise according to claim 1, characterized in that: The model training in step S5 includes the following parts: inputting the training set data into the constructed neural network model, using mean square error (MSE) as the loss function and the Adam optimization algorithm to optimize the model parameters; The formula for calculating the mean square error is: where y true,i is the true value of the i-th sample, y pred,i is the predicted value of the i-th sample, and n is the total number of samples; After each training round, the validation set is used to evaluate the model performance to prevent overfitting of the model; finally, the best carbon emission activity data prediction model is saved.
6. The carbon emission prediction method for a cigarette manufacturing enterprise according to claim 1, characterized in that: The predicted value of the carbon emission activity data in step S6 is a normalized value, which needs to be denormalized to obtain the original dimension, that is, the actual predicted value of the net purchased electricity, purchased steam, natural gas, gasoline and diesel.
7. The carbon emission prediction method for a cigarette manufacturing enterprise according to claim 1, characterized in that: The carbon emissions prediction in step S7 includes the following steps: according to the basic carbon accounting equation provided by the IPCC, greenhouse gas (GHG) emissions = activity data (AD) × emission factor (EF), calculate the carbon emissions generated by the actual predicted value of each activity data in step S6; add up the carbon emissions generated by all activity data to obtain the total carbon emissions of the cigarette manufacturing enterprise.
Citation Information
Patent Citations
A method for predicting corporate carbon emissions
CN117035167B