Short-term carbon emission prediction method based on CNN-Bi-LSTM neural network model

Through the short-term carbon emission prediction method based on the CNN-Bi-LSTM neural network model, combined with the CNN convolutional neural network and the Bi-LSTM bidirectional long and short-term memory network, the problem of difficult prediction of short-term carbon emissions in the existing technology is solved, and the problem of short-term carbon emission prediction is realized. Accurate prediction under the influence of multiple factors improves the accuracy and reliability of prediction.

CN119090067APending Publication Date: 2024-12-06ANHUI CANBANG ELECTRIC
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411138788.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-19
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

It is difficult for the prior art to accurately predict short-term carbon emissions, especially under the influence of multiple factors, and a single model is difficult to provide accurate and reliable prediction results.

Method used

Using the CNN-Bi-LSTM neural network model, combined with the CNN convolutional neural network and Bi-LSTM bidirectional long and short-term memory network, a short-term carbon emission prediction method is constructed. By comprehensively processing the carbon emission data and temperature data, features are extracted and performed. Timing modeling.

Benefits of technology

Accurate prediction of short-term carbon emissions has been achieved, significantly improved feature extraction and sequence prediction capabilities, able to more accurately capture the short-term trend of carbon emissions, and improve the accuracy and reliability of predictions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090067A_ABST
    Figure CN119090067A_ABST
Patent Text Reader

Abstract

The invention relates to short-term carbon emission prediction, in particular to a short-term carbon emission prediction method based on a CNN-Bi-LSTM neural network model, and the method comprises the steps: constructing the CNN-Bi-LSTM neural network model; collecting carbon emission data and corresponding temperature data to construct a carbon emission data set based on temperature variables, and dividing the carbon emission data set based on the temperature variables into a training set, a verification set and a test set; performing model training on the CNN-Bi-LSTM neural network model by using the training set to obtain a trained CNN-Bi-LSTM neural network model; collecting carbon emission time sequence data in the current time period, and inputting the carbon emission time sequence data into the trained CNN-Bi-LSTM neural network model to obtain a carbon emission prediction result; according to the technical scheme provided by the invention, the defect that short-term carbon emission is difficult to accurately predict in the prior art can be effectively overcome.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to short-term carbon emission prediction, and in particular to a short-term carbon emission prediction method based on a CNN-Bi-LSTM neural network model. Background Art

[0002] Carbon emissions are not only affected by their own sequence fluctuations, but also related to factors such as economy and climate change. Therefore, carbon emission forecasting is very complicated. First, carbon emissions in the power industry are affected by the energy structure and industrial structure. Thermal power generation, as the main power generation method in my country, generates carbon emissions that account for up to 70% of the power industry. Secondly, the level of economic development of a country or region is also closely related to carbon emissions. Finally, short-term carbon emission forecasts also need to consider some external variables related to meteorological factors, among which temperature changes have a great impact on carbon emissions. For example, a large number of cooling equipment will be used in summer, while a large number of heating equipment will be used in winter. In addition, the arrival of holidays will also have a more obvious impact on short-term carbon emissions. Many influencing factors make carbon emissions highly nonlinear and volatile, which increases the difficulty of short-term carbon emission forecasts in the power sector.

[0003] Existing carbon emission prediction methods mainly include statistical models, gray models and machine learning models. Although statistical models are simple and feasible, they are difficult to fully learn the nonlinear characteristics of carbon emission data. Gray models gradually show disadvantages in accuracy as the number of statistical samples increases. Carbon emission data is easily affected by factors such as weather and season, resulting in strong volatility in the short term, making prediction difficult. Although machine learning models have achieved certain results in carbon emission prediction, when faced with complex changes in carbon emission data, a single model often cannot provide accurate and reliable prediction results. Therefore, combined models with complementary advantages have gradually attracted attention. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the above-mentioned shortcomings of the prior art, the present invention provides a short-term carbon emissions prediction method based on the CNN-Bi-LSTM neural network model, which can effectively overcome the defect of the prior art that it is difficult to accurately predict short-term carbon emissions.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0008] The short-term carbon emission prediction method based on the CNN-Bi-LSTM neural network model includes the following steps:

[0009] S1. Build a CNN-Bi-LSTM neural network model;

[0010] S2. Collect carbon emission data and corresponding temperature data to construct a carbon emission data set based on temperature variables, and divide the carbon emission data set based on temperature variables into a training set, a validation set, and a test set;

[0011] S3, using the training set to train the CNN-Bi-LSTM neural network model to obtain a trained CNN-Bi-LSTM neural network model;

[0012] S4. Collect the carbon emission time series data in the current period, and input the carbon emission time series data into the trained CNN-Bi-LSTM neural network model to obtain the carbon emission prediction results.

[0013] Preferably, a CNN-Bi-LSTM neural network model is constructed in S1, including:

[0014] Based on the CNN convolutional neural network, the Bi-LSTM bidirectional long short-term memory network is added between the pooling layer and the fully connected layer in the CNN convolutional neural network, and an activation layer is set between the pooling layer and the Bi-LSTM bidirectional long short-term memory network to construct a CNN-Bi-LSTM neural network model.

[0015] Preferably, the CNN convolutional neural network comprises an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer connected in sequence;

[0016] Input layer, preprocessing the input data;

[0017] Convolution layer, which uses convolution kernel to perform convolution operation on input data to extract the features of input data;

[0018] Pooling layer, which reduces the spatial size of feature maps, reduces model parameters and computational complexity while retaining important feature information;

[0019] The fully connected layer converts the feature map output by the pooling layer into the final result;

[0020] The output layer outputs the model’s prediction results for the target variable.

[0021] Preferably, the Bi-LSTM bidirectional long short-term memory network is a network that adds a backward LSTM layer on the basis of the LSTM long short-term memory network, that is, the Bi-LSTM bidirectional long short-term memory network includes a forward LSTM layer and a backward LSTM layer;

[0022] The forward LSTM layer processes the input in the order of time steps and outputs a hidden state at each time step. These hidden states contain the forward information from the beginning of the sequence to the current time step.

[0023] The backward LSTM layer processes the input in reverse order of the time steps and outputs a hidden state at each time step. These hidden states contain backward information from the end of the sequence to the current time step.

[0024] The Bi-LSTM bidirectional long short-term memory network concatenates the hidden states of the forward LSTM layer and the backward LSTM layer at each time step to form the final output of the time step, which contains both forward and backward context information.

[0025] Preferably, the LSTM long short-term memory network introduces a gating mechanism to adjust the weight of the network during the cycle, and dynamically manages and updates the unit state according to the gating function, and the LSTM long short-term memory network includes a forget gate, an input gate and an output gate;

[0026] The forget gate determines which information is discarded from the cell state. The forget gate reads the hidden state of the previous time step and the input of the current time step, and then outputs a value between 0 and 1, which determines which information in the cell state should be retained or forgotten.

[0027] The input gate determines which new information should be stored in the cell state. The input gate also reads the hidden state of the previous time step and the input of the current time step, and then generates two values: one is the candidate value vector representing the new information, and the other is the input gate weight that determines which new information should be stored in the cell state.

[0028] The output gate reads the cell state and the hidden state of the previous time step, and then outputs a hidden state value that will be used as the output of the current time step and as the input of the next time step.

[0029] Preferably, in S2, carbon emission data and corresponding temperature data are collected to construct a carbon emission data set based on temperature variables, and the carbon emission data set based on temperature variables is divided into a training set, a validation set and a test set, including:

[0030] S21. Collect carbon emission data and the corresponding maximum and minimum temperature values ​​in chronological order, use the min-max normalization method to process the data, obtain normalized data, and construct a carbon emission data set based on temperature variables;

[0031] S22. Divide the carbon emission data set based on temperature variables into a training set, a validation set, and a test set according to a preset ratio.

[0032] Preferably, in S3, the CNN-Bi-LSTM neural network model is trained using the training set to obtain a trained CNN-Bi-LSTM neural network model, including:

[0033] S31, setting the loss function and optimizer of the CNN-Bi-LSTM neural network model, and inputting the training set into the CNN-Bi-LSTM neural network model for model training;

[0034] S32, calculating the loss value based on the loss function, and the optimizer updates the model parameters according to the loss value and the network gradient information;

[0035] S33. If the loss value is less than the preset threshold, the model training ends, and the current CNN-Bi-LSTM neural network model is the trained CNN-Bi-LSTM neural network model. Otherwise, return to S31 and continue to use the training set for model training.

[0036] Preferably, in S31, the training set is input into the CNN-Bi-LSTM neural network model for model training, including:

[0037] S311, the convolution layer uses the convolution kernel to perform a convolution operation on the input data to extract the features of the input data;

[0038] S312, the pooling layer further extracts features by selecting the mean or maximum value, and sends the extracted features to the Bi-LSTM bidirectional long short-term memory network;

[0039] The output of S313 and Bi-LSTM bidirectional long short-term memory networks is sent to the fully connected layer to extract higher-level feature representations. The fully connected layer further integrates the information of each time step and generates the final result.

[0040] S314, the output layer outputs the prediction results of the model for the target variable.

[0041] Preferably, when the CNN-Bi-LSTM neural network model is trained on the training set, the generalization ability of the model is evaluated by observing the performance on the validation set, and the hyperparameters and structure of the model are tuned;

[0042] After obtaining the trained CNN-Bi-LSTM neural network model, the performance of the model is evaluated using a test set.

[0043] (III) Beneficial effects

[0044] Compared with the prior art, the short-term carbon emission prediction method based on the CNN-Bi-LSTM neural network model provided by the present invention has the following beneficial effects:

[0045] 1) Use the combined model to predict short-term carbon emissions. The CNN convolutional neural network can automatically extract the characteristic information of carbon emission data. The Bi-LSTM bidirectional long short-term memory network can model and learn the time series of the features extracted by the CNN convolutional neural network, so that the combined model can accurately predict short-term carbon emissions.

[0046] 2) By comprehensively considering the factors affecting carbon emission data, a carbon emission dataset based on temperature variables was constructed, which enabled the model to achieve significant improvements in feature extraction and sequence prediction, and to more accurately capture the short-term trend of carbon emissions. Through comparative experiments, not only the performance of the CNN-Bi-LSTM neural network model for short-term carbon emissions was verified, but also the superiority of the model in processing multi-factor and multi-dimensional datasets was further demonstrated. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0048] Figure 1 It is a schematic diagram of the process of the present invention;

[0049] Figure 2 It is a schematic diagram of the structure of CNN convolutional neural network;

[0050] Figure 3 This is a schematic diagram of the structure of the Bi-LSTM bidirectional long short-term memory network;

[0051] Figure 4 This is a schematic diagram of the structure of the LSTM long short-term memory network;

[0052] Figure 5 A graph for collecting carbon emission data;

[0053] Figure 6 This is a comparison chart of carbon emission prediction results of the CNN-Bi-LSTM neural network model under different data sets;

[0054] Figure 7 This is a comparison chart of carbon emission prediction results of different models under the same data set. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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.

[0056] LSTM long short-term memory network performs well in time series prediction. Its gating mechanism helps capture long-term dependencies, but gradient explosion may occur when processing very long sequences, thus affecting model performance. A single LSTM long short-term memory network has limited flexibility and may not be well adapted to various sequence data. The training process requires a lot of memory and time, and overfitting is also a concern when the data set is small.

[0057] CNN convolutional neural network can significantly reduce model parameters by adopting local connection, weight sharing and downsampling technology, thereby solving the parameter explosion problem of traditional neural networks, enabling CNN convolutional neural network to effectively cope with large-scale data sets, avoid parameter explosion, and significantly improve big data processing capabilities.

[0058] The Bi-LSTM bidirectional long short-term memory network improves the model's modeling and representation capabilities for sequence data by taking into account both forward and backward information. It is suitable for a variety of sequence data-related tasks and has achieved good results in many practical applications.

[0059] The short-term carbon emission prediction method based on CNN-Bi-LSTM neural network model, such as Figure 1 As shown, S1, construct a CNN-Bi-LSTM neural network model, specifically including:

[0060] Based on the CNN convolutional neural network, the Bi-LSTM bidirectional long short-term memory network is added between the pooling layer and the fully connected layer in the CNN convolutional neural network, and an activation layer is set between the pooling layer and the Bi-LSTM bidirectional long short-term memory network to construct a CNN-Bi-LSTM neural network model.

[0061] like Figure 2 As shown in Figure 2, the CNN convolutional neural network includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer ( Figure 2 Alternating convolutional layers and pooling layers can improve model performance);

[0062] Input layer, preprocessing the input data;

[0063] Convolution layer, which uses convolution kernel to perform convolution operation on input data to extract the features of input data;

[0064] Pooling layer, which reduces the spatial size of feature maps, reduces model parameters and computational complexity while retaining important feature information;

[0065] The fully connected layer converts the feature map output by the pooling layer into the final result;

[0066] The output layer outputs the model’s prediction results for the target variable.

[0067] like Figure 3 As shown in the figure, the Bi-LSTM bidirectional long short-term memory network adds a backward LSTM layer on the basis of the LSTM long short-term memory network, that is, the Bi-LSTM bidirectional long short-term memory network includes a forward LSTM layer and a backward LSTM layer;

[0068] The forward LSTM layer processes the input in the order of time steps and outputs a hidden state at each time step. These hidden states contain the forward information from the beginning of the sequence to the current time step.

[0069] The backward LSTM layer processes the input in reverse order of the time steps and outputs a hidden state at each time step. These hidden states contain backward information from the end of the sequence to the current time step.

[0070] The Bi-LSTM bidirectional long short-term memory network concatenates the hidden states of the forward LSTM layer and the backward LSTM layer at each time step to form the final output of that time step, which contains both forward and backward context information.

[0071] like Figure 4 As shown in the figure, the LSTM long short-term memory network introduces a gating mechanism to adjust the weight of the network during the cycle, and dynamically manages and updates the unit state according to the gating function. The LSTM long short-term memory network includes a forget gate, an input gate, and an output gate.

[0072] The forget gate determines which information is discarded from the cell state. The forget gate reads the hidden state of the previous time step and the input of the current time step, and then outputs a value between 0 and 1, which determines which information in the cell state should be retained or forgotten.

[0073] The input gate determines which new information should be stored in the cell state. The input gate also reads the hidden state of the previous time step and the input of the current time step, and then generates two values: one is the candidate value vector representing the new information, and the other is the input gate weight that determines which new information should be stored in the cell state.

[0074] The output gate reads the cell state and the hidden state of the previous time step, and then outputs a hidden state value that will be used as the output of the current time step and as the input of the next time step.

[0075] In the technical solution of this application, the Deep Learning Toolbox (version 14.4) provided by Matlab is used to define (see Table 1) and train the CNN-Bi-LSTM neural network model:

[0076] Table 1 Structural design of CNN-Bi-LSTM neural network model

[0077]

[0078] Note: The bidirectional LSTM layer in Table 1 is the Bi-LSTM bidirectional long short-term memory network.

[0079] S2. Collect carbon emission data and corresponding temperature data to construct a carbon emission data set based on temperature variables, and divide the carbon emission data set based on temperature variables into a training set, a validation set, and a test set, specifically including:

[0080] S21. Collect carbon emission data and the corresponding maximum and minimum temperature values ​​in chronological order, use the min-max normalization method to process the data, obtain normalized data, and construct a carbon emission data set based on temperature variables;

[0081] S22. Divide the carbon emission data set based on temperature variables into a training set, a validation set, and a test set according to a preset ratio.

[0082] S3. Use the training set to train the CNN-Bi-LSTM neural network model to obtain a trained CNN-Bi-LSTM neural network model, which specifically includes:

[0083] S31, setting the loss function and optimizer of the CNN-Bi-LSTM neural network model, and inputting the training set into the CNN-Bi-LSTM neural network model for model training;

[0084] S32, calculating the loss value based on the loss function, and the optimizer updates the model parameters according to the loss value and the network gradient information;

[0085] S33. If the loss value is less than the preset threshold, the model training ends, and the current CNN-Bi-LSTM neural network model is the trained CNN-Bi-LSTM neural network model. Otherwise, return to S31 and continue to use the training set for model training.

[0086] Specifically, in S31, the training set is input into the CNN-Bi-LSTM neural network model for model training, including:

[0087] S311, the convolution layer uses the convolution kernel to perform a convolution operation on the input data to extract the features of the input data;

[0088] S312, the pooling layer further extracts features by selecting the mean or maximum value, and sends the extracted features to the Bi-LSTM bidirectional long short-term memory network;

[0089] The output of S313 and Bi-LSTM bidirectional long short-term memory networks is sent to the fully connected layer to extract higher-level feature representations. The fully connected layer further integrates the information of each time step and generates the final result.

[0090] S314, the output layer outputs the prediction results of the model for the target variable.

[0091] Specifically, when the CNN-Bi-LSTM neural network model is trained on the training set, the generalization ability of the model is evaluated by observing the performance on the validation set, and the hyperparameters and structure of the model are tuned;

[0092] After obtaining the trained CNN-Bi-LSTM neural network model, the test set is used to evaluate the performance of the model.

[0093] S4. Collect the carbon emission time series data in the current period, and input the carbon emission time series data into the trained CNN-Bi-LSTM neural network model to obtain the carbon emission prediction results.

[0094] In order to better verify the technical solution of this application, Figure 5 As shown, the carbon emission data of the power sector in Anhui Province and the corresponding maximum and minimum temperature values ​​were collected from January 1, 2019 to October 31, 2022. The sampling interval was 1 day, and there were 1,400 carbon emission data in total.

[0095] First, the carbon emission dataset with only carbon emission data and the carbon emission dataset based on temperature variables were respectively input into the CNN-Bi-LSTM neural network model for model training and carbon emission prediction. Figure 6 The comparative experimental results are shown.

[0096] from Figure 6 It can be seen that the prediction results obtained by using the carbon emission dataset based on temperature variables for model training and carbon emission prediction are closer to the true value and have better prediction performance.

[0097] Secondly, the carbon emission data set based on temperature variables was input into the CNN convolutional neural network, LSTM long short-term memory network, CNN-LSTM neural network model, Bi-LSTM bidirectional long short-term memory network, and CNN-Bi-LSTM neural network model for model training and carbon emission prediction, and the following results were obtained: Figure 7 The comparative experimental results are shown.

[0098] from Figure 7 It can be seen that the fitting effects of CNN convolutional neural network, LSTM long short-term memory network, and CNN-LSTM neural network models are poor, and the overall prediction effect of Bi-LSTM bidirectional long short-term memory network is relatively good, but the effect is not good at the peaks and troughs with large changes, while the CNN-Bi-LSTM neural network model not only has a better degree of fitting in this area, but also the overall prediction result is closest to the true value. Therefore, these two comparative experiments can prove that on the basis of constructing a carbon emission data set based on temperature variables, the use of CNN-Bi-LSTM neural network model can make more accurate predictions of short-term carbon emissions.

[0099] Finally, through RMSE, R 2 , MAPE evaluation indicators are used to evaluate the performance of CNN convolutional neural network, LSTM long short-term memory network, CNN-LSTM neural network model, Bi-LSTM bidirectional long short-term memory network and CNN-Bi-LSTM neural network model, see the table below:

[0100] Table 2 Comparison of evaluation indicators of different models

[0101]

[0102] As can be seen from Table 2, the RMSE and MAPE error values ​​of the CNN-Bi-LSTM neural network model are the smallest. 2 It is closest to 1. Compared with the CNN convolutional neural network, LSTM long short-term memory network, CNN-LSTM neural network model, and Bi-LSTM bidirectional long short-term memory network, the MAPE of the CNN-Bi-LSTM neural network model is reduced to 0.3994, which means that the overall percentage error of the model is much smaller, indicating that the model can effectively improve the accuracy of short-term carbon emission forecasts.

[0103] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. Short-term carbon emission prediction method based on CNN-Bi-LSTM neural network model, Features: The following steps are involved: S1. Build a CNN-Bi-LSTM neural network model; S2. Collect carbon emission data and corresponding temperature data to construct a carbon emission data set based on temperature variables, and divide the carbon emission data set based on temperature variables into a training set, a validation set, and a test set; S3, using the training set to train the CNN-Bi-LSTM neural network model to obtain a trained CNN-Bi-LSTM neural network model; S4. Collect the carbon emission time series data in the current period, and input the carbon emission time series data into the trained CNN-Bi-LSTM neural network model to obtain the carbon emission prediction results.

2. According to the short-term carbon emission prediction method based on the CNN-Bi-LSTM neural network model according to claim 1, Features: The CNN-Bi-LSTM neural network model is constructed in S1, including: Based on the CNN convolutional neural network, the Bi-LSTM bidirectional long short-term memory network is added between the pooling layer and the fully connected layer in the CNN convolutional neural network, and an activation layer is set between the pooling layer and the Bi-LSTM bidirectional long short-term memory network to construct a CNN-Bi-LSTM neural network model.

3. According to the short-term carbon emission prediction method based on the CNN-Bi-LSTM neural network model according to claim 2, Features: The CNN convolutional neural network includes an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer connected in sequence; Input layer, preprocessing the input data; Convolution layer, which uses convolution kernel to perform convolution operation on input data to extract the features of input data; Pooling layer, which reduces the spatial size of feature maps, reduces model parameters and computational complexity while retaining important feature information; The fully connected layer converts the feature map output by the pooling layer into the final result; The output layer outputs the model’s prediction results for the target variable.

4. According to the short-term carbon emission prediction method based on the CNN-Bi-LSTM neural network model of claim 3, Features: The Bi-LSTM bidirectional long short-term memory network adds a backward LSTM layer on the basis of the LSTM long short-term memory network, that is, the Bi-LSTM bidirectional long short-term memory network includes a forward LSTM layer and a backward LSTM layer; The forward LSTM layer processes the input in the order of time steps and outputs a hidden state at each time step. These hidden states contain the forward information from the beginning of the sequence to the current time step. The backward LSTM layer processes the input in reverse order of the time steps and outputs a hidden state at each time step. These hidden states contain backward information from the end of the sequence to the current time step. The Bi-LSTM bidirectional long short-term memory network concatenates the hidden states of the forward LSTM layer and the backward LSTM layer at each time step to form the final output of the time step, which contains both forward and backward context information.

5. According to the short-term carbon emission prediction method based on the CNN-Bi-LSTM neural network model according to claim 4, Features: The LSTM long short-term memory network introduces a gating mechanism to adjust the weight of the network during the cycle, and dynamically manages and updates the unit state according to the gating function. The LSTM long short-term memory network includes a forget gate, an input gate, and an output gate; The forget gate determines which information is discarded from the cell state. The forget gate reads the hidden state of the previous time step and the input of the current time step, and then outputs a value between 0 and 1, which determines which information in the cell state should be retained or forgotten. The input gate determines which new information should be stored in the cell state. The input gate also reads the hidden state of the previous time step and the input of the current time step, and then generates two values: one is the candidate value vector representing the new information, and the other is the input gate weight that determines which new information should be stored in the cell state. The output gate reads the cell state and the hidden state of the previous time step, and then outputs a hidden state value that will be used as the output of the current time step and as the input of the next time step.

6. According to the short-term carbon emission prediction method based on the CNN-Bi-LSTM neural network model of claim 2, Features: In S2, carbon emission data and corresponding temperature data are collected to construct a carbon emission data set based on temperature variables, and the carbon emission data set based on temperature variables is divided into a training set, a validation set, and a test set, including: S21. Collect carbon emission data and the corresponding maximum and minimum temperature values ​​in chronological order, use the min-max normalization method to process the data, obtain normalized data, and construct a carbon emission data set based on temperature variables; S22. Divide the carbon emission data set based on temperature variables into a training set, a validation set, and a test set according to a preset ratio.

7. According to the short-term carbon emission prediction method based on the CNN-Bi-LSTM neural network model of claim 6, Features: In S3, the CNN-Bi-LSTM neural network model is trained using the training set to obtain a trained CNN-Bi-LSTM neural network model, including: S31, setting the loss function and optimizer of the CNN-Bi-LSTM neural network model, and inputting the training set into the CNN-Bi-LSTM neural network model for model training; S32, calculating the loss value based on the loss function, and the optimizer updates the model parameters according to the loss value and the network gradient information; S33. If the loss value is less than the preset threshold, the model training ends, and the current CNN-Bi-LSTM neural network model is the trained CNN-Bi-LSTM neural network model. Otherwise, return to S31 and continue to use the training set for model training.

8. According to the short-term carbon emission prediction method based on the CNN-Bi-LSTM neural network model of claim 7, Features: In S31, the training set is input into the CNN-Bi-LSTM neural network model for model training, including: S311, the convolution layer uses the convolution kernel to perform a convolution operation on the input data to extract the features of the input data; S312, the pooling layer further extracts features by selecting the mean or maximum value, and sends the extracted features to the Bi-LSTM bidirectional long short-term memory network; The output of S313 and Bi-LSTM bidirectional long short-term memory networks is sent to the fully connected layer to extract higher-level feature representations. The fully connected layer further integrates the information of each time step and generates the final result. S314, the output layer outputs the prediction results of the model for the target variable.

9. According to the short-term carbon emission prediction method based on the CNN-Bi-LSTM neural network model of claim 7, Features: When the CNN-Bi-LSTM neural network model is trained on the training set, the generalization ability of the model is evaluated by observing the performance on the validation set, and the hyperparameters and structure of the model are tuned; After obtaining the trained CNN-Bi-LSTM neural network model, the performance of the model is evaluated using a test set.

Citation Information

Cited By

  • Carbon emission prediction method and device based on data enhancement and time-frequency fusion comparison

    CN119941267A

  • Intelligent carbon emission prediction system based on neural network

    CN120106307A