Process enterprise multi-cycle composite carbon emission prediction method based on space-time diagram network
Through a method based on spatiotemporal graph network and combined with the STGCN model framework, the problem that traditional models cannot fully consider the relationship between infrastructure within enterprises is solved, and the accuracy and accuracy of carbon emission prediction are significantly improved.
Patent Information
- Application Number
- CN202510015782.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-06
- Publication Date
- 2025-05-13
AI Technical Summary
Traditional time series models cannot fully consider the interconnection and interaction between infrastructure within the enterprise when predicting corporate carbon emissions, resulting in insufficient prediction accuracy and accuracy.
The multi-cycle composite carbon emission prediction method of process enterprises based on spatiotemporal graph network is adopted to construct a carbon emission time series data set by collecting and modeling the topological relationships of infrastructure, and predicting it using the STGCN model framework.
It significantly improves the accuracy of carbon emission forecasting, can more accurately monitor and predict corporate carbon emission data, and provides more reliable analysis and prediction results.
Smart Images
Figure CN119990415A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of carbon emission prediction, and in particular to a multi-period composite carbon emission prediction method for a process enterprise based on a spatiotemporal graph network. Background Art
[0002] Process-based enterprises are centered on core production processes. During the production and operation process, a large amount of energy and material flows are involved, which are mainly used in the enterprise's infrastructure. Therefore, the carbon emissions of the enterprise's infrastructure are the main source of carbon emissions. In order to accurately calculate and accurately predict the carbon emissions of the production and operation activities of enterprises, it is necessary to conduct detailed statistical analysis and effective application of the complex time series data of multiple types and sources generated by the infrastructure.
[0003] When using traditional time series model forecasting technology to estimate a company's carbon emissions, it usually relies on historical carbon emission data of each independent infrastructure. Although this method can provide a relatively short-term carbon emission forecast for each infrastructure, it ignores the mutual influence and synergy between the infrastructures within the enterprise, resulting in an inability to accurately predict the overall distribution of carbon emissions and their scope of influence. Therefore, this method has inherent limitations in the precision and accuracy of the forecast.
[0004] Under the existing technical framework, traditional time series forecasting models often fail to fully consider the interconnections and interactions between infrastructure within an enterprise, which limits their ability to accurately predict carbon emissions distribution, multi-period composite forecasts and their impact range. Summary of the invention
[0005] In order to solve the problems raised in the above-mentioned background technology, the purpose of the present invention is to provide a multi-period composite carbon emission prediction method for process enterprises based on spatiotemporal graph network, comprising the following steps: S1, collecting enterprise infrastructure data and performing topological relationship modeling: recording the attribute information of the enterprise's specific infrastructure, input or output type data, and recording the connection relationship between the infrastructures according to the causal relationship of the data; S2, constructing a carbon emission time series data set: carbon emission time series data of the enterprise's specific infrastructure at equal time intervals within a period of time, data preprocessing, and constructing a data set; S3, converting the enterprise infrastructure graph structure into an adjacency matrix and a feature matrix; S4, building an STGCN model framework: the model framework consists of an input layer, an ST-ConvBlock layer, a GCN-Block layer and an output layer; S5, training the model; S6, real-time prediction and iterative update: obtaining real-time infrastructure carbon emissions and obtaining carbon emission forecasts for each infrastructure of the enterprise.
[0006] The step S1 specifically includes the following steps:
[0007] S11. Record the attributes of the infrastructure, record the type of input data required by each infrastructure, and record the type of output data generated by each infrastructure.
[0008] S12. Record causal relationships: Analyze and record the flow paths of data between different infrastructures, and identify and record the dependencies between infrastructures.
[0009] S13. Record connection relationships: record direct physical connections between infrastructures and record logical connections between infrastructures.
[0010] S14. Obtain the carbon emission time series data of each enterprise’s infrastructure.
[0011] The step S2 specifically includes the following steps:
[0012] S21. Data recording: Record the carbon emissions of a company’s specific infrastructure over a period of time.
[0013] S22. Data preprocessing: Use the interquartile range statistical method to clean the collected data, process missing values and outliers in the data, and unify the time format and unit of the data.
[0014] S23. Data storage: storing the data in a database.
[0015] S24. Dataset construction: Apply data transformation and data augmentation to improve the generalization ability of the model, and use the dataset construction tool to build it.
[0016] S25. Dataset division: Verify the integrity of the dataset and divide the dataset into training set, validation set and test set.
[0017] The step S3 specifically includes the following steps: representing the enterprise infrastructure graph structure as an adjacency matrix and a feature matrix, wherein the adjacency matrix is a binary matrix indicating whether there are edges connecting the nodes, and the feature matrix converts the time series data into feature vectors according to the nodes corresponding to the infrastructure in the enterprise infrastructure graph structure network, and forms a feature matrix.
[0018] The step S4 specifically comprises the following steps:
[0019] S41, input layer inputs training data;
[0020] S42, using the He initialization method, Gaussian distribution is used to initialize the weight parameters, the variance is Where is the input dimension of the weight matrix, and the bias parameter can be initialized to 0 or any small value. The formula is as follows,
[0021] Among them, W represents the weight parameter and n represents the input dimension of the weight matrix.
[0022] S43, construct the ST-Block layer, which consists of multiple one-dimensional convolutional layers and residual blocks to process time series data. The formula is as follows: X (k) =ReLU(LX (k-1 )W (k-1) ). Among them, X (k-1) is the feature matrix of the k-1th layer, W (k-1) is the weight matrix of the k-1th layer, L is the graph Laplacian matrix, and ReLu is the activation function.
[0023] S44, construct the GCN-Block layer, which consists of multiple graph convolutional layers and residual blocks to process spatial data. The formula is as follows: Among them, X t represents the feature matrix at time t, A represents the adjacency matrix of the graph structure, I is the identity matrix, is the degree matrix, W (1) and W (2) are the weight matrices of the two convolutional layers respectively, and σ represents the activation function.
[0024] S45, the output layer includes a time domain convolution layer and a fully connected layer. The convolution kernel size of the time domain convolution layer is The number is C o , mapping the output to The fully connected layer is in Output
[0025] The step S5 specifically comprises the following steps:
[0026] S51. Obtain the adjacency matrix and the characteristic matrix, and initialize the parameter matrix using the He initialization method.
[0027] S52. Determine the training rounds, batch size, and the number of iterations is sample size / batch size.
[0028] S53, forward propagation is performed to calculate the predicted value, and then the loss function is used to calculate the loss. The loss function uses the RMSE method to calculate the difference between the predicted result and the actual result. The formula is:
[0029]
[0030] Where n represents the number of samples, Y i represents the true value of the i-th sample, Represents the predicted value of the i-th sample.
[0031] S54, perform back propagation and use the gradient descent method to update the model parameters.
[0032] S55. Obtain model parameters.
[0033] The step S6 specifically comprises the following steps:
[0034] S61. Obtain real-time infrastructure carbon emissions, and use a visual real-time monitoring interface to display corporate carbon emissions.
[0035] S62. Using the carbon emission prediction model, the predicted carbon emission of each infrastructure of the enterprise after a certain time interval is obtained.
[0036] Compared with the existing technology, the present invention applies graph neural network technology and combines it with the spatiotemporal graph convolutional network, which can fully consider the spatiotemporal relationship between enterprise infrastructures, thereby significantly improving the accuracy of carbon emission prediction, realizing real-time monitoring of enterprise carbon emission data, and providing enterprises with more accurate carbon emission analysis and prediction results. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 This is a physical topology diagram of enterprise production and operation in this embodiment;
[0038] Figure 2 This is a structural diagram of the STGCN model constructed in this embodiment;
[0039] Figure 3 This is a diagram showing the prediction results in this embodiment. DETAILED DESCRIPTION
[0040] The present invention will be further described below with reference to the accompanying drawings.
[0041] A multi-period composite carbon emission prediction method for process enterprises based on a spatiotemporal graph network includes the following steps: S1. Collecting enterprise infrastructure data and performing topological relationship modeling: recording the attribute information of the enterprise's specific infrastructure, inputting or outputting type data, and recording the connection relationship between the infrastructures based on the causal relationship of the data, and constructing the physical topological structure of the enterprise's production and operation. Figure 1 As shown in , understand the interconnection and dependency of the internal infrastructure of the enterprise; S2, build a carbon emission time series data set: carbon emission time series data of specific infrastructure of the enterprise at equal time intervals within a period of time, data preprocessing, and data set construction; S3, convert the enterprise infrastructure graph structure into an adjacency matrix and a feature matrix; as shown in Figure 2As shown, S4, building the STGCN model framework: the model framework consists of an input layer, an ST-ConvBlock layer, a GCN-Block layer and an output layer; S5, training the model; S6, real-time prediction and iterative update: obtaining real-time infrastructure carbon emissions and obtaining the predicted carbon emissions of each infrastructure of the enterprise.
[0042] Step S1 specifically includes the following steps:
[0043] S11. Record the attributes of infrastructure, the types of input data required for each infrastructure, and the types of output data generated by each infrastructure, such as the types and names of production facilities, office facilities, and network facilities.
[0044] S12. Record causal relationships: Analyze and record the flow paths of data between different infrastructures, and identify and record the dependencies between infrastructures. For example, the operation of some infrastructures may depend on the stable output of other infrastructures.
[0045] S13. Record connection relationships: record direct physical connections between infrastructures, such as pipelines, cables, network connections, etc., and record logical connections between infrastructures.
[0046] S14. Obtain the carbon emission time series data of each enterprise’s infrastructure.
[0047] Step S2 specifically includes the following steps:
[0048] S21. Data recording: recording the carbon emissions of a company’s specific infrastructure over a period of time. The data includes specific year, month, day, hour, minute, and second information.
[0049] S22. Data preprocessing: Use the interquartile range statistical method to clean the collected data, process missing values and outliers in the data, and unify the time format and unit of the data.
[0050] S23. Data storage: Store data in the database to ensure data security and reliability.
[0051] S24. Dataset construction: Apply data transformation and data augmentation to improve the generalization ability of the model, using dataset construction tools such as PyTorch's Dataset class.
[0052] S25. Dataset division: Verify the integrity of the data set to ensure that there is no missing or erroneous data. Divide the data set into training set, validation set and test set to facilitate model training and evaluation.
[0053] Step S3 specifically includes the following steps: representing the enterprise infrastructure graph structure as an adjacency matrix and a feature matrix. The adjacency matrix is a binary matrix that indicates whether there are edges connecting the nodes. The feature matrix converts the time series data into feature vectors according to the nodes corresponding to the infrastructure in the enterprise infrastructure graph structure network, and forms a feature matrix.
[0054] Step S4 specifically includes the following steps:
[0055] S41, input layer inputs training data;
[0056] S42. Using the He initialization method can effectively accelerate the convergence of the model and improve the accuracy of the model. Gaussian distribution is used to initialize the weight parameters with a variance of Where is the input dimension of the weight matrix, and the bias parameter can be initialized to 0 or any small value. The formula is as follows, Among them, W represents the weight parameter and n represents the input dimension of the weight matrix.
[0057] S43, construct the ST-Block layer, which consists of multiple one-dimensional convolutional layers and residual blocks to process time series data. The formula is as follows: X (k) =ReLU(LX (k-1) W (k-1) ). Among them, X (k-1) is the feature matrix of the k-1th layer, W (k-1) is the weight matrix of the k-1th layer, L is the graph Laplacian matrix, and ReLU is the activation function.
[0058] The combination of multiple one-dimensional convolutional layers and residual blocks can improve the feature extraction ability of the model, thereby better predicting time series data. The number and size of one-dimensional convolutional layers and residual blocks can be adjusted according to the characteristics of the data.
[0059] S44, construct the GCN-Block layer, which consists of multiple graph convolutional layers and residual blocks to process spatial data. The formula is as follows: Among them, X t represents the feature matrix at time t, A represents the adjacency matrix of the graph structure, is the identity matrix, is the degree matrix, W (1) and W (2) are the weight matrices of the two convolutional layers respectively, and σ represents the activation function.
[0060] The combination of multiple graph convolutional layers and residual blocks can improve the feature extraction ability of the model, thereby better predicting spatial data. The number and size of graph convolutional layers and residual blocks can be adjusted according to the characteristics of the data.
[0061] S45, the output layer includes a time domain convolution layer and a fully connected layer. The convolution kernel size of the time domain convolution layer is The number is C o , mapping the output to The fully connected layer is in Output
[0062] Step S5 specifically includes the following steps:
[0063] S51. Obtain the adjacency matrix and the characteristic matrix, and initialize the parameter matrix using the He initialization method.
[0064] S52. Determine the training rounds, batch size, and the number of iterations is sample size / batch size.
[0065] S53, forward propagation is performed to calculate the predicted value, and then the loss function is used to calculate the loss. The loss function uses the RMSE method to calculate the difference between the predicted result and the actual result. The formula is:
[0066]
[0067] Where n represents the number of samples, Y i represents the true value of the i-th sample, Represents the predicted value of the i-th sample.
[0068] S54, perform back propagation and use the gradient descent method to update the model parameters.
[0069] S55. Obtain model parameters.
[0070] Step S6 specifically includes the following steps:
[0071] S61. Obtain real-time infrastructure carbon emissions, and visualize the real-time monitoring interface to display the company's carbon emissions. S62. Obtain the predicted carbon emissions of each infrastructure of the company after a certain time interval through the carbon emission prediction model. Figure 3 , and get the prediction result graph.
[0072] After obtaining real-time infrastructure carbon emissions data, the carbon emissions prediction model is used to predict the carbon emissions of each enterprise infrastructure after a certain time interval. These predicted values are used as new inputs to predict the carbon emissions of the next time interval through the model. This process will continue until a preset moment is reached, and this step will be repeated continuously during this period to achieve real-time monitoring and prediction of the enterprise's carbon emissions. Through this iterative update method, the model can continuously optimize itself and improve the accuracy of the prediction.
Claims
1. A multi-period composite carbon emission prediction method for process enterprises based on spatiotemporal graph network, characterized by: The following steps are involved: S1. Collect enterprise infrastructure data and perform topological relationship modeling: record the attribute information of the enterprise's specific infrastructure, input or output type data, and record the connection relationship between infrastructures based on the causal relationship of the data; S2. Construct a carbon emission time series data set: carbon emission time series data of the enterprise's specific infrastructure at equal time intervals over a period of time, data preprocessing, and data set construction; S3. Convert the enterprise infrastructure graph structure into an adjacency matrix and a feature matrix; S4. Build the STGCN model framework: the model framework consists of an input layer, ST-ConvBlock layer, GCN-Block layer, and an output layer; S5. Train the model; S6. Real-time prediction and iterative update: obtain real-time infrastructure carbon emissions and obtain the predicted carbon emissions of each enterprise infrastructure.
2. According to claim 1, a method for predicting multi-period composite carbon emissions of process enterprises based on a spatiotemporal graph network is characterized by: The step S1 specifically includes the following steps: S11. Record the attributes of the infrastructure, record the type of input data required by each infrastructure, and record the type of output data generated by each infrastructure. S12. Record causal relationships: Analyze and record the flow paths of data between different infrastructures, and identify and record the dependencies between infrastructures. S13. Record connection relationships: record direct physical connections between infrastructures and record logical connections between infrastructures. S14. Obtain the carbon emission time series data of each enterprise’s infrastructure.
3. According to claim 1, a method for predicting multi-period composite carbon emissions of process enterprises based on a spatiotemporal graph network is characterized by: The step S2 specifically includes the following steps: S21. Data recording: Record the carbon emissions of a company’s specific infrastructure over a period of time. S22. Data preprocessing: Use the interquartile range statistical method to clean the collected data, process missing values and outliers in the data, and unify the time format and unit of the data. S23. Data storage: storing the data in a database. S24. Dataset construction: Apply data transformation and data augmentation to improve the generalization ability of the model, and use the dataset construction tool to build it. S25. Dataset division: Verify the integrity of the dataset and divide the dataset into training set, validation set and test set.
4. According to claim 1, a method for predicting multi-period composite carbon emissions of process enterprises based on a spatiotemporal graph network is characterized by: The step S3 specifically includes the following steps: representing the enterprise infrastructure graph structure as an adjacency matrix and a feature matrix, wherein the adjacency matrix is a binary matrix indicating whether there are edges connecting the nodes, and the feature matrix converts the time series data into feature vectors according to the nodes corresponding to the infrastructure in the enterprise infrastructure graph structure network, and forms a feature matrix.
5. According to claim 1, a method for predicting multi-period composite carbon emissions of process enterprises based on a spatiotemporal graph network is characterized by: The step S4 specifically comprises the following steps: S41, input layer inputs training data; S42, using the He initialization method, Gaussian distribution is used to initialize the weight parameters, the variance is Where is the input dimension of the weight matrix, and the bias parameter can be initialized to 0 or any small value. The formula is as follows, Among them, W represents the weight parameter and n represents the input dimension of the weight matrix. S43, construct the ST-Block layer, which consists of multiple one-dimensional convolutional layers and residual blocks to process time series data. The formula is as follows: X (k) =ReLU(LX (k-1) W (k-1) ). Among them, X (k-1) is the feature matrix of the k-1th layer, W (k-1) is the weight matrix of the k-1th layer, L is the graph Laplacian matrix, and ReLu is the activation function. S44, construct the GCN-Block layer, which consists of multiple graph convolutional layers and residual blocks to process spatial data. The formula is as follows: Among them, X t represents the feature matrix at time t, A represents the adjacency matrix of the graph structure, I is the identity matrix, is the degree matrix, W (1) and W (2) are the weight matrices of the two convolutional layers respectively, and σ represents the activation function. S45, the output layer includes a time domain convolution layer and a fully connected layer. The convolution kernel size of the time domain convolution layer is The number is C o , mapping the output to The fully connected layer is in Output 6. The method for predicting multi-period composite carbon emissions of process enterprises based on a spatiotemporal graph network according to claim 1 is characterized by: The step S5 specifically comprises the following steps: S51. Obtain the adjacency matrix and the characteristic matrix, and initialize the parameter matrix using the He initialization method. S52. Determine the training rounds, batch size, and the number of iterations is sample size / batch size. S53, forward propagation is performed to calculate the predicted value, and then the loss function is used to calculate the loss. The loss function uses the RMSE method to calculate the difference between the predicted result and the actual result. The formula is: Where n represents the number of samples, Y i represents the true value of the i-th sample, Represents the predicted value of the i-th sample. S54, perform back propagation and use the gradient descent method to update the model parameters. S55. Obtain model parameters.
7. The method for predicting multi-period composite carbon emissions of process enterprises based on a spatiotemporal graph network according to claim 1 is characterized by: The step S6 specifically comprises the following steps: S61. Obtain real-time infrastructure carbon emissions, and use a visual real-time monitoring interface to display corporate carbon emissions. S62. Using the carbon emission prediction model, the predicted carbon emission of each infrastructure of the enterprise after a certain time interval is obtained.