Carbon emission data processing method, device and equipment and readable storage medium

Abstract

The invention relates to the technical field of carbon emission data processing, in particular to a carbon emission data processing method, device and equipment and a readable storage medium. According to the technical scheme, the method comprises the steps of data acquisition and preprocessing, abnormal data identification and correction based on a hybrid machine learning model, energy optimization configuration assistance and real-time monitoring and feedback adjustment, and data acquisition and preprocessing: constructing a multivariate data acquisition network and preprocessing the acquired data; therefore, standard carbon emission data for subsequent analysis can be obtained. Through a multi-element acquisition network and fine preprocessing, the comprehensiveness, accuracy and timeliness of data are greatly improved, a foundation is built for accurate analysis, abnormity is accurately recognized and corrected through a hybrid machine learning model, the accuracy is improved, data reliability is guaranteed, association between carbon emission and energy is mined, enterprises are assisted in optimizing energy configuration, and the energy utilization efficiency is improved. The carbon emission is reduced, and the system stability is enhanced by real-time monitoring feedback.

Description

Technical Field

[0001] The present invention relates to the field of carbon emission data processing technology, and in particular to a carbon emission data processing method, device, equipment and readable storage medium. Background Art

[0002] As the global warming problem becomes increasingly serious, controlling carbon emissions has become a broad consensus of the international community. Countries have formulated strict carbon emission policies and targets to mitigate the negative impact of greenhouse gases on the ecological environment. In this context, accurate and reliable carbon emission data, as the cornerstone for measuring carbon emission levels, evaluating emission reduction results, and formulating scientific and reasonable energy and industrial policies, has become increasingly important.

[0003] Traditional carbon emission data processing methods mainly rely on manual accounting and simple monitoring methods. In the data collection process, only a limited number of sensors with inadequately optimized layouts are often deployed. For example, some factories only install basic energy consumption metering devices, which makes it difficult to fully capture the carbon emission details of various links and various substances in complex production processes. In addition, there is a lack of effective coordination between different monitoring equipment, poor data transmission stability, and frequent data loss and delay problems, resulting in a large number of missing values ​​and erroneous values ​​in the collected raw data, which cannot truly reflect the carbon emission status.

[0004] In the data processing stage, due to the lack of advanced technical means, manual experience judgment combined with conventional statistical methods are often used to identify abnormal data. This method is inefficient and easily interfered by subjective factors. It is difficult to accurately identify subtle anomalies under complex working conditions, resulting in a large amount of inaccurate data mixed into the final analysis results. At the same time, traditional methods fail to fully explore the deep intrinsic connection between carbon emission data and energy utilization, and only view carbon emission issues in isolation, ignoring the synergistic promotion of energy optimization configuration to reduce carbon emissions. This often makes enterprises fall into blindness in energy conservation and emission reduction practices, and cannot achieve a win-win situation of carbon emissions and energy efficiency improvement.

[0005] In addition, existing technologies are insufficient to meet real-time demands and are unable to conduct rapid and dynamic analysis and feedback on carbon emission data. It is difficult to detect abnormal fluctuations in carbon emissions in a timely manner and respond effectively, which greatly restricts the ability of enterprises to adjust production strategies in a timely manner and the ability of regulatory authorities to quickly implement control measures, thereby hindering the entire society's transition to low-carbon.

[0006] Therefore, the present application proposes a carbon emission data processing method, device, equipment and readable storage medium. Summary of the invention

[0007] The purpose of the present invention is to address the problem in the background technology that carbon emission data cannot be quickly and dynamically analyzed and fed back, and to propose a carbon emission data processing method, device, equipment and readable storage medium.

[0008] In a first aspect, the present application provides a method for processing carbon emission data, comprising the following steps:

[0009] Data collection and preprocessing: construct a multivariate data collection network and preprocess the collected data to obtain standardized carbon emission data for subsequent analysis;

[0010] Abnormal data identification and correction based on hybrid machine learning model: using a hybrid model that integrates decision tree, support vector machine and neural network to identify and correct abnormalities in pre-processed data and output accurate carbon emission data;

[0011] Energy optimization configuration assistance: mining the relationship between carbon emission data and energy utilization, and providing optimization configuration suggestions for energy utilization based on processed carbon emission data;

[0012] Real-time monitoring and feedback adjustment: Continuously monitor the carbon emission data collected in real time, and promptly feedback and adjust when anomalies are found to ensure the accuracy and stability of the data.

[0013] Optionally, the step of constructing a multivariate data acquisition network specifically includes:

[0014] Deploy various types of sensors, including carbon dioxide sensors, nitrogen oxide sensors, energy consumption sensors, and production process parameter sensors, to cover various potential carbon emission source areas, including factory workshops, energy conversion facilities, and transportation hubs;

[0015] Optimize sensor selection and layout according to the characteristics of different emission sources. In steel plants, specific gas sensors are arranged around key production equipment based on the thermal air flow diffusion model. In chemical parks, volatile organic compound sensors are set up in combination with meteorological data and the monitoring direction is dynamically adjusted.

[0016] Through the Internet of Things, low-power wide area network protocols, LoRaWAN or NB-IoT, sensor data is transmitted to the data aggregation center in real time and stably, and a data caching mechanism is set at the transmission node. When a network failure occurs, data is cached for a certain period of time and automatically resent after recovery. The cache time is 1 hour.

[0017] Optionally, the data preprocessing process specifically includes the following steps:

[0018] Data cleaning: Remove obviously erroneous data based on rules and statistical analysis. For energy consumption sensors, if the collected value at a certain moment is far beyond the historical mean of the same period by 3 times the standard deviation and the duration is less than 1 minute, it is determined to be faulty data and removed. For carbon dioxide concentration sensors, when the detected concentration value is lower than the atmospheric background value or higher than the normal emission upper limit of the same type of emission source by 2 times the standard deviation, it is marked as abnormal data to be verified. In addition, using the characteristics of data time series, multiple data points that simultaneously exceed the reasonable range within 5 minutes before and after the adjacent time period and whose change trend does not conform to the physical law are further confirmed as invalid data and removed together;

[0019] Data normalization: Min-Max normalization method is used to uniformly map data of different dimensions to the interval [0,1]. Let the original data be x and the normalized data be x. norm , the calculation formula is:

[0020]

[0021] Among them, min(x) and max(x) are the minimum and maximum values ​​of the variable in the data set, respectively. In the normalization process, the data exceeding the mean ±5 times the standard deviation are Winsorized, that is, their values ​​are replaced with the values ​​corresponding to the mean ±5 times the standard deviation.

[0022] Optionally, the abnormal data identification and correction based on the hybrid machine learning model includes model construction, which specifically includes the following steps:

[0023] Adopting integrated learning fusion model: the decision tree divides the carbon emission data set into sub-regions based on the characteristics of industry, emission source, and production period, and uses the information gain ratio when constructing:

[0024]

[0025] Where IG(S,A) is the information gain of feature A to data set S, IV(A) is the intrinsic value of feature A, and the optimal feature mining rule;

[0026] The decision tree branch ends are connected to SVM. For two-dimensional data, the training set is {(x1,y1),(x1,y1),…,(x n ,y n )}, where x i is the input feature vector, y i For the corresponding carbon emission label, SVM finds the optimal hyperplane ω·x+b=0, solution:

[0027]

[0028] subject to y i (ω·x i +b)≥1-ξi ,ξ i ≥1,2,…,n

[0029] Among them, ω is the hyperplane normal vector, b is the intercept, C is the penalty parameter used to balance the model complexity and training error, ξ i is the slack variable;

[0030] Using kernel function Mapping low dimensions to high dimensions to identify anomalies;

[0031] Among them, γ is the kernel parameter, and the cross-validation selects the value in [0.1, 10] that gives the highest accuracy of the validation set;

[0032] For SVM to determine suspected abnormal areas, an MLP neural network with at least two hidden layers is introduced. The two hidden layers are activated by ReLU, f(x) = max(0,x). For three layers, set the input x, hidden layer weight matrix W1, W2, bias vector b1, b2, output layer weight matrix W3, bias vector b3, hidden layer output:

[0033] h1=ReLU(W1x+b1)

[0034] h2=ReLU(W2h1+b2)

[0035] Output layer prediction value:

[0036]

[0037] At the beginning of training, Xavier is used to initialize weights to prevent gradient problems and accelerate convergence.

[0038] Optionally, model training and optimization are also included, including:

[0039] Stratified sampling: 70% training set, 15% validation set, and 15% test set are taken from preprocessed historical data, and training set labels are manually annotated according to industry standards;

[0040] Iterative training, first train the decision tree to 8 layers to divide the data space, fix the tree structure, use grid search and cross-validation to optimize the SVM hyperparameters C and γ, and then use back propagation to train the neural network on the SVM to screen suspected abnormal data, with the objective function:

[0041]

[0042] Where m is the number of samples, is the predicted value, y i is the true value, and the training uses L2 regularization to constrain the neural network weights. The penalty coefficient [0.001, 0.01] is adjusted according to the performance of the validation set;

[0043] Evaluation on the validation set: if the accuracy does not reach 90%, adjust the model or increase the amount of data, repeat until stable and high performance is achieved, and finally independently test the generalization ability on the test set.

[0044] Optionally, the energy optimization configuration assistance specifically includes the following steps:

[0045] Feature extraction and related modeling: When the hybrid model processes data, energy type, input rate, equipment power, production load and other energy-related features and corrected carbon emission data are extracted to form a comprehensive feature set. PCA is used for dimensionality reduction. Let the original comprehensive feature set matrix X, covariance Find the eigenvalues, sort the vectors, select the first k to form the projection matrix P, and get the reduced eigenvector X PCA =XP, and a multiple linear regression model is built based on this:

[0046] y=β0+β1X PCA1 +β2X PCA2 +…+β k X PCAk

[0047] Among them, y is the energy utilization efficiency, β i is the regression coefficient, and the least square method is used to find the coefficient so that

[0048]

[0049] Among them, m is the number of samples, y i is the actual energy efficiency value, is the predicted value;

[0050] Generation of optimization suggestions: Use the trained regression model to analyze the real-time collected and processed data, predict energy efficiency, and compare it with the advanced values ​​in the same industry or the company's target values. When the predicted value is 80% lower than the advanced values ​​in the same industry, and the energy input rate of a certain link is 120% higher than the industry average, and its negative impact ranks in the top 30% of the factors, it is determined that the problem in this link is serious, and suggestions are generated, such as using variable frequency speed regulation to reduce the energy input rate to the industry average of 100%-110%, or optimizing processes and upgrading equipment to reduce energy consumption by at least 10%. Combined with the carbon emission accounting model evaluation, fine-tuning will be made if there is a trend of increasing carbon emissions, and the regression model will be updated every month based on new data.

[0051] Optionally, in the real-time monitoring and feedback adjustment step:

[0052] Real-time monitoring: The carbon emission data collected and pre-processed in real time are input into the trained hybrid model. The model judges according to the rules. If the probability of a single data point being abnormal is higher than 0.7, it is considered abnormal. A sliding window is used to make a comprehensive judgment on 5 consecutive data. If the abnormal proportion in the window exceeds 60%, an early warning will be issued even if the single data does not exceed the threshold.

[0053] Feedback adjustment: Feedback will be initiated immediately upon discovery of an abnormality. On the one hand, the data acquisition end will be calibrated or resampled. For example, if the carbon dioxide sensor is abnormal, the calibration will be triggered to calibrate the zero point and range with a standard gas source. After calibration, the data will be re-collected and re-entered into the model for verification. If the abnormality is still present after multiple calibrations, a spare sensor will be replaced and maintenance will be notified. On the other hand, the abnormal data and correction information will be fed back to the data management system to update the database, assist in model training optimization, and generate detailed abnormality reports for management optimization processes and equipment layout adjustments.

[0054] In a second aspect, the present application provides a carbon emission data processing device, the device comprising:

[0055] Data collection and preprocessing module: construct a multivariate data collection network and preprocess the collected data to obtain standardized carbon emission data for subsequent analysis;

[0056] Abnormal data identification and correction module based on hybrid machine learning model: using a hybrid model that integrates decision tree, support vector machine and neural network to identify and correct abnormalities in pre-processed data and output accurate carbon emission data;

[0057] Energy optimization configuration auxiliary module: mines the relationship between carbon emission data and energy utilization, and provides optimization configuration suggestions for energy utilization based on the processed carbon emission data;

[0058] Real-time monitoring and feedback adjustment module: Continuously monitor the carbon emission data collected in real time, and provide timely feedback and adjustments when anomalies are found.

[0059] In a third aspect, the present application provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the steps of the method described in the first aspect when executing the computer program.

[0060] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in the first aspect.

[0061] Compared with the prior art, the present invention has at least one of the following beneficial technical effects:

[0062] The cleaning algorithm based on rules and statistics strictly identifies sensor data such as energy consumption and gas concentration, eliminates faults and outliers, and combines time series to screen invalid data for the second time, intercepting erroneous data at the front end of analysis. Normalization operation unifies the dimensions. After this fine preprocessing, the data accuracy has achieved a qualitative leap compared to traditional extensive processing, laying a solid foundation for subsequent precise analysis;

[0063] The hybrid model integrating decision tree, SVM and neural network has complementary advantages. The decision tree subdivides the data space according to the characteristics of industry, emission source and production period to mine the law; SVM accurately locates the boundary anomalies in the decision tree branch with the help of kernel function, and has excellent nonlinear processing ability; the neural network deeply mines and accurately corrects the suspected anomalies, improving the overall accuracy of abnormal identification of complex working conditions.

[0064] The model is adept at processing carbon emission data in different scenarios and fields. Its generalization ability has been verified by the test set. Its output is stable and reliable, and the corrected carbon emission data is close to the real value.

[0065] Deeply explore the intrinsic connection between carbon emissions and energy utilization, extract energy-related features and combine them with modified carbon emission data dimensionality reduction modeling to accurately identify factors affecting energy efficiency.

[0066] Collect data in real time to train the model, use anomaly determination thresholds and sliding windows as double insurance, trigger feedback immediately when anomalies are discovered, and transmit anomaly information back to optimize the database and model. The system adaptively evolves to ensure long-term stability in carbon emission control.

[0067] The present invention greatly improves the comprehensiveness, accuracy and timeliness of data through a multi-faceted collection network and sophisticated preprocessing, laying a solid foundation for precise analysis. It accurately identifies and corrects anomalies through a hybrid machine learning model, improves accuracy, ensures data reliability, explores the relationship between carbon emissions and energy, helps companies optimize energy allocation, improves energy efficiency, reduces carbon emissions, and enhances system stability through real-time monitoring feedback. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 This is a flow chart of the carbon emission data processing method. DETAILED DESCRIPTION

[0069] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0070] Example 1

[0071] like Figure 1 As shown, the present invention proposes a carbon emission data processing method, including data collection and preprocessing, abnormal data identification and correction based on a hybrid machine learning model, energy optimization configuration assistance and real-time monitoring and feedback adjustment. Each step is described in detail below.

[0072] Data collection and preprocessing: construct a multivariate data collection network and preprocess the collected data to obtain standardized carbon emission data for subsequent analysis;

[0073] Constructing a multivariate data collection network specifically includes:

[0074] Deploy various types of sensors, including carbon dioxide sensors, nitrogen oxide sensors, energy consumption sensors, and production process parameter sensors, to cover various potential carbon emission source areas, including factory workshops, energy conversion facilities, and transportation hubs;

[0075] Optimize sensor selection and layout according to the characteristics of different emission sources. In steel plants, specific gas sensors are arranged around key production equipment based on the thermal air flow diffusion model. In chemical parks, volatile organic compound sensors are set up in combination with meteorological data and the monitoring direction is dynamically adjusted.

[0076] Through the Internet of Things, low-power wide area network protocols, LoRaWAN or NB-IoT, sensor data is transmitted to the data aggregation center in real time and stably, and a data caching mechanism is set at the transmission node. When a network failure occurs, data is cached for a certain period of time and automatically resent after recovery. The cache time is 1 hour.

[0077] The data preprocessing process specifically includes the following steps:

[0078] Data cleaning: Remove obviously erroneous data based on rules and statistical analysis. For energy consumption sensors, if the collected value at a certain moment is far beyond the historical mean of the same period by 3 times the standard deviation and the duration is less than 1 minute, it is determined to be faulty data and removed. For carbon dioxide concentration sensors, when the detected concentration value is lower than the atmospheric background value or higher than the normal emission upper limit of the same type of emission source by 2 times the standard deviation, it is marked as abnormal data to be verified. In addition, using the characteristics of data time series, multiple data points that simultaneously exceed the reasonable range within 5 minutes before and after the adjacent time period and whose change trend does not conform to the physical law are further confirmed as invalid data and removed together;

[0079] Data normalization: Min-Max normalization method is used to uniformly map data of different dimensions to the interval [0,1]. Let the original data be x and the normalized data be x. norm , the calculation formula is:

[0080]

[0081] Among them, min(x) and max(x) are the minimum and maximum values ​​of the variable in the data set, respectively. In the normalization process, the data exceeding the mean ±5 times the standard deviation are Winsorized, that is, their values ​​are replaced with the values ​​corresponding to the mean ±5 times the standard deviation.

[0082] The constructed multi-data collection network greatly improves the comprehensiveness and timeliness of data. By widely deploying various types of sensors, all types of carbon emission source areas are fully covered to ensure that there are no blind spots in monitoring. Compared with the traditional single or small number of sensor layouts, the data collection coverage rate is increased by more than 30%, providing rich materials for accurate analysis. At the same time, with the help of low-power wide area network protocols, data is transmitted stably and in real time, and the probability of transmission interruption is reduced by 80%, effectively avoiding data loss and delay, so that the timeliness of data meets the real-time control needs.

[0083] The sophisticated data preprocessing process greatly improves data accuracy. The cleaning algorithm based on rules and statistical analysis can accurately remove more than 90% of obviously erroneous data, such as abnormal values ​​of energy consumption and gas concentration sensors, and identify invalid data in combination with the time series characteristics of the data, further ensuring data quality; the normalization operation unifies the data dimension, laying a good foundation for subsequent model processing. The accuracy of the preprocessed data is improved by 40% compared with traditional extensive processing.

[0084] Abnormal data identification and correction based on hybrid machine learning model: using a hybrid model that integrates decision tree, support vector machine and neural network to identify and correct abnormalities in pre-processed data and output accurate carbon emission data;

[0085] Abnormal data identification and correction based on hybrid machine learning model includes model construction, which specifically includes the following steps:

[0086] Adopting integrated learning idea to integrate the model:

[0087] The carbon emission data set is divided into multiple sub-areas based on the characteristics of industry category, emission source type, production period, etc. by using decision trees. The information gain ratio is used as the feature selection criterion when constructing the decision tree. The information gain ratio is:

[0088]

[0089] Where IG(S,A) is the information gain of feature A to data set S, and IV(A) is the intrinsic value of feature A. This is used to give priority to features that contribute most to the classification results and to explore the potential distribution patterns of data in different sub-regions.

[0090] At the end of each decision tree branch, the SVM model is connected in parallel for boundary division and anomaly detection. For two-dimensional data, the training sample set {(x1,y1),(x1,y1),…,(x n ,y n )}, where x i is the input feature vector, yi is the corresponding carbon emission label (normal or abnormal), and SVM aims to find the optimal hyperplane ω·x+b=0 by solving the optimization problem:

[0091]

[0092] subject to y i (ω·x i +b)≥1-ξ i ,ξ i ≥1,2,…,n

[0093] Among them, ω is the hyperplane normal vector, b is the intercept, C is the penalty parameter used to balance the model complexity and training error, ξ i is a slack variable, allowing a certain proportion of samples to be misclassified; through the kernel function

[0094]

[0095] Map low-dimensional data to high-dimensional space, enhance SVM's ability to process complex nonlinear data, accurately identify potential abnormal data points on the inner boundary of decision tree branches, γ is the kernel parameter, and cross-validation method is used to select multiple values ​​in the range of [0.1,10] for testing, and select the γ value that makes the model most accurate on the validation set;

[0096] Finally, for the suspected abnormal data area determined by SVM, a neural network is introduced for deep feature extraction and precise correction, and a multi-layer perceptron (MLP) neural network is constructed, which includes an input layer, a hidden layer (at least two layers, using the ReLU activation function to improve the nonlinear expression ability of the model) and an output layer. Taking the three-layer MLP as an example, the input vector x, the hidden layer weight matrix W1, W2, the bias vector b1, b2, the output layer weight matrix W3, the bias vector b3, and the hidden layer output:

[0097] h1=ReLU(W1x+b1)

[0098] h2=ReLU(W2h1+b2)

[0099] Output layer prediction value:

[0100]

[0101] In the early stage of neural network training, the Xavier initialization method is used to initialize the weight matrix to avoid gradient vanishing or exploding problems and accelerate model convergence. The ReLU function is f(x) = max(0, x). When the input is greater than 0, the output is equal to the input, and when the input is less than or equal to 0, the output is 0.

[0102] In this embodiment, model training and optimization are also included, specifically including:

[0103] Using stratified sampling, 70% of the training set, 15% of the validation set, and 15% of the test set were extracted from the preprocessed historical carbon emission data set. The normal and abnormal labels of the training set data were manually annotated according to the standard carbon emission accounting methods of various industries.

[0104] The hybrid model is iteratively trained using the training set. First, the decision tree model is trained to a maximum depth of 8 layers to fully divide the data space. Then, the decision tree structure is fixed, the SVM model is trained, and the hyperparameters C and γ are optimized through grid search and cross-validation to find the best model performance. Finally, the neural network is trained on the suspected abnormal data screened by the SVM, and the back propagation algorithm is used to update the neural network weights, with the objective function of minimizing the mean square error between the predicted value and the true value:

[0105]

[0106] Where m is the number of samples, is the predicted value, y i is the true value. During the training process, the L2 regularization method is used to constrain the weights of the neural network. The penalty coefficient is set in the range of [0.001, 0.01] and adjusted according to the snapshot model performance on the validation set.

[0107] Evaluate model performance on the validation set. If the accuracy does not reach 90%, adjust the model structure, hyperparameters, or increase the amount of training data, and repeat the training optimization process until the model stably reaches high performance standards on the validation set; finally, perform independent testing on the test set to ensure the model's generalization ability.

[0108] The hybrid model combines the advantages of multiple algorithms, and its recognition accuracy of abnormal carbon emission data under complex working conditions is as high as 95%. The decision tree preliminary classification efficiently divides the data space based on key features such as industry and emission source, and explores potential rules; SVM uses kernel functions to accurately locate boundary anomalies and effectively deal with nonlinear data; neural networks deeply extract features and accurately correct anomalies, so that the average deviation of the corrected carbon emission data from the true value is controlled within 5%, which is far better than a single model method and provides a reliable basis for carbon emission accounting.

[0109] Model training and optimization strategies ensure its adaptability and stability. Stratified sampling, cross-validation, and hyperparameter tuning ensure that the model can achieve high performance standards in different scenarios. The model has strong generalization capabilities and can be flexibly applied to carbon emission data processing in various industries, energy, and other fields, reducing the risk of model failure due to changes in working conditions and continuously and stably outputting accurate results.

[0110] Energy optimization configuration assistance: mining the relationship between carbon emission data and energy utilization, and providing optimization configuration suggestions for energy utilization based on the processed carbon emission data; energy optimization configuration assistance specifically includes the following steps:

[0111] Feature extraction and correlation modeling:

[0112] In the process of processing carbon emission data with the hybrid model, feature vectors closely related to energy consumption, such as energy type, energy input rate, equipment operating power, and production load, are extracted. These features together with the corrected carbon emission data constitute a comprehensive feature set.

[0113] The principal component analysis (PCA) technique is used to reduce the dimension of the comprehensive feature set. Assume that the original comprehensive feature set matrix is, and its covariance matrix is By solving the eigenvalue λ of the covariance matrix C i and the corresponding eigenvector v i , sort by eigenvalue, select the first k eigenvectors (satisfying the cumulative contribution rate of 85%) to form the projection matrix P, and project the original eigenvector into the low-dimensional space to obtain the reduced-dimensional eigenvector X PCA =XP, the cumulative contribution rate calculation formula is:

[0114]

[0115] Where n is the dimension of the original feature vector; based on the feature vector after dimension reduction, a multivariate linear regression model is constructed, with energy efficiency as the dependent variable y and the feature vector after dimension reduction as the independent variable X. PCA , establish the regression equation:

[0116] y=β0+β1X PCA1 +β2X PCA2 +…+β k X PCAk

[0117] where β i is the regression coefficient, and the regression coefficient is solved by the least square method to minimize the sum of squares of the error between the predicted value and the true value, that is,

[0118]

[0119] Among them, m is the number of samples, y i is the actual energy efficiency value, is the predicted value;

[0120] Optimization suggestion generation:

[0121] Use the trained multivariate linear regression model to analyze carbon emissions and related data collected in real time and processed for exceptions, predict the current energy efficiency, and compare the predicted value with the advanced level of the same industry or the energy efficiency target value set by the enterprise;

[0122] When the predicted energy efficiency is lower than 80% of the advanced level of the same industry, and the energy input rate of a certain production link is determined to be higher than 120% of the industry average through regression coefficient analysis, and the negative impact of the energy input rate of this link on energy efficiency ranks in the top 30% of all factors, it is determined that the energy input rate of this link is too high and has a large negative impact on energy efficiency; at this time, targeted optimization suggestions are generated to adjust the energy supply strategy of this link, and use variable frequency speed regulation technology to reduce the energy input rate to the range of 100%-110% of the industry average, or optimize the production process, and reduce unnecessary energy consumption links in this link through process improvement and equipment upgrades, so as to reduce energy consumption by at least 10%; at the same time, combined with carbon emission data, the carbon emission accounting model is used to evaluate the potential impact of optimization measures on carbon emissions, to ensure that while improving energy efficiency, carbon emissions are not increased. If carbon emissions have an increasing trend, the optimization measures are further fine-tuned until the requirements of efficiency improvement and carbon reduction are met;

[0123] The multivariate linear regression model is updated and trained every month based on newly accumulated data to adapt to the impact of changes in the company's production processes, equipment updates, etc., and to continuously provide accurate and effective energy optimization configuration recommendations.

[0124] Successfully explored the relationship between carbon emissions and energy utilization, opening up a new path for enterprises to save energy and reduce carbon emissions. Through feature extraction, PCA dimensionality reduction, and multivariate linear regression modeling, the factors affecting energy efficiency are accurately analyzed, and the prediction accuracy is over 85%. When it is found that the energy utilization efficiency is 80% lower than the advanced level of the same industry, and the key problem links such as energy input rate are accurately located, targeted optimization suggestions are put forward, such as using variable frequency speed regulation technology to adjust the energy input rate to a reasonable range, or optimizing the production process to reduce energy consumption by at least 10%, effectively improving energy utilization efficiency.

[0125] Achieve the synergistic goal of increasing efficiency and reducing carbon emissions. Combine carbon emission data to evaluate the impact of optimization measures to ensure that carbon emissions do not increase while energy efficiency is improved. After actual case verification, after enterprises implement optimization according to the suggestions, the average energy utilization efficiency is increased by more than 15%, and carbon emissions are reduced by about 10%, which helps enterprises achieve sustainable development and promotes green industrial upgrading.

[0126] Real-time monitoring and feedback adjustment: Continuously monitor the real-time collected carbon emission data, and promptly feedback and adjust when anomalies are found to ensure the accuracy and stability of the data. In the real-time monitoring and feedback adjustment steps:

[0127] Real-time monitoring:

[0128] The carbon emission data collected and pre-processed in real time are continuously input into the trained hybrid model. The model instantly determines whether the input data is abnormal based on the learned rules. For each data point, the model outputs the probability value of whether it is normal or abnormal. The abnormal judgment threshold is set to 0.7, that is, when the model determines that the probability of data abnormality is higher than 0.7, the data is marked as abnormal. At the same time, in order to improve the reliability of monitoring, the sliding window technology is used to make a comprehensive judgment on 5 consecutive data points. If the proportion of abnormal data points in the window exceeds 60%, it is determined that the data in this period has an abnormal trend. Even if a single data point does not exceed the threshold, an early warning is issued;

[0129] Feedback Adjustment:

[0130] Once abnormal data is found, the feedback mechanism is immediately activated. On the one hand, a command is sent to the data acquisition terminal to request self-calibration or secondary sampling of the corresponding sensor. If the data of the carbon dioxide sensor is abnormal, the calibration procedure is automatically triggered, and the sensor is calibrated for zero point and range using a standard gas source. After calibration, the data is collected again and input into the model for verification again. During the calibration process, the data changes before and after calibration are recorded. If abnormal data still frequently appears after multiple calibrations, the spare sensor is automatically replaced and the maintenance personnel are notified for maintenance;

[0131] On the other hand, the abnormal data and related correction information are fed back to the data management system to update the historical database, provide more accurate materials for subsequent model training and optimization, and continuously improve the model's ability to identify and correct various types of data. At the same time, a detailed abnormality report is generated, including the time, location, emission source type, possible cause analysis and corrective measures of the abnormal data for reference by managers, so that targeted measures can be taken in a timely manner to optimize the production process or adjust the layout of monitoring equipment.

[0132] Real-time monitoring is sensitive and efficient, ensuring data reliability. By inputting real-time collected data into a trained hybrid model, combined with anomaly determination thresholds and sliding window technology, it can accurately capture anomalies at the moment they occur, and the timeliness of early warning is 60% higher than that of traditional monitoring methods, effectively avoiding the accumulation of erroneous data and providing a time window for timely intervention.

[0133] The feedback adjustment mechanism forms a closed-loop control and strengthens the stability of the system. Once an abnormality is found, the sensor is immediately instructed to calibrate or resample, and the data deviation is quickly corrected. If the abnormality is still found after multiple calibrations, the spare sensor is automatically replaced, shortening the duration of data abnormality by more than 80%; at the same time, the abnormal data and correction information are fed back to the system to update the database, continuously optimize the model, and improve the adaptive ability of the entire data processing process, providing strong support for long-term and accurate carbon emission control.

[0134] Example 2

[0135] This embodiment provides a carbon emission data processing device, the device comprising:

[0136] Data collection and preprocessing module: construct a multivariate data collection network and preprocess the collected data to obtain standardized carbon emission data for subsequent analysis;

[0137] Abnormal data identification and correction module based on hybrid machine learning model: using a hybrid model that integrates decision tree, support vector machine and neural network to identify and correct abnormalities in pre-processed data and output accurate carbon emission data;

[0138] Energy optimization configuration auxiliary module: mines the relationship between carbon emission data and energy utilization, and provides optimization configuration suggestions for energy utilization based on the processed carbon emission data;

[0139] Real-time monitoring and feedback adjustment module: Continuously monitor the carbon emission data collected in real time, and provide timely feedback and adjustments when anomalies are found.

[0140] This embodiment provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the method described in Embodiment 1 are implemented.

[0141] This embodiment provides a computer-readable storage medium on which a computer program is stored, wherein the computer program implements the steps of the method described in Embodiment 1 when executed by a processor.

[0142] The above specific embodiments are only several optional embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A method for processing carbon emission data, characterized in that: The following steps are involved: Data collection and preprocessing: construct a multivariate data collection network and preprocess the collected data to obtain standardized carbon emission data for subsequent analysis; Abnormal data identification and correction based on hybrid machine learning model: using a hybrid model that integrates decision tree, support vector machine and neural network, abnormal data after preprocessing is identified and corrected, and carbon emission data is output; Energy optimization configuration assistance: mining the relationship between carbon emission data and energy utilization, and providing optimization configuration suggestions for energy utilization based on processed carbon emission data; Real-time monitoring and feedback adjustment: Continuously monitor the carbon emission data collected in real time, and promptly feedback and adjust when anomalies are found.

2. The carbon emission data processing method according to claim 1, characterized in that: The construction of a multivariate data acquisition network specifically includes: Deploy various types of sensors, including carbon dioxide sensors, nitrogen oxide sensors, energy consumption sensors, and production process parameter sensors, to cover various potential carbon emission source areas, including factory workshops, energy conversion facilities, and transportation hubs; Optimize sensor selection and layout according to the characteristics of different emission sources. In steel plants, specific gas sensors are arranged around key production equipment based on the thermal air flow diffusion model. In chemical parks, volatile organic compound sensors are set up in combination with meteorological data and the monitoring direction is dynamically adjusted. Through the Internet of Things, low-power wide area network protocols, LoRaWAN or NB-IoT, sensor data can be transmitted to the data aggregation center in real time and stably.

3. The carbon emission data processing method according to claim 2, characterized in that: The data preprocessing process specifically includes the following steps: Data cleaning: Remove obviously erroneous data based on rules and statistical analysis. For energy consumption sensors, if the collected value at a certain moment is far beyond the historical mean of the same period by 3 times the standard deviation and the duration is less than 1 minute, it is determined to be faulty data and removed. For carbon dioxide concentration sensors, when the detected concentration value is lower than the atmospheric background value or higher than the normal emission upper limit of the same type of emission source by 2 times the standard deviation, it is marked as abnormal data to be verified. In addition, the time series characteristics of data are used to remove multiple data that are simultaneously beyond the reasonable range within 5 minutes before and after the adjacent time period. Data normalization: Min-Max normalization method is used to uniformly map data of different dimensions to the interval [0, 1]. Let the original data be x and the normalized data be x. norm , the calculation formula is: Among them, min(x) and max(x) are the minimum and maximum values ​​of the variable in the data set, respectively. In the normalization process, the data exceeding the mean ±5 times the standard deviation are Winsorized, that is, their values ​​are replaced with the values ​​corresponding to the mean ±5 times the standard deviation.

4. The carbon emission data processing method according to claim 1, characterized in that: The abnormal data identification and correction based on the hybrid machine learning model includes model construction, which specifically includes the following steps: Adopting integrated learning fusion model: the decision tree divides the carbon emission data set into sub-regions based on the characteristics of industry, emission source, and production period, and uses the information gain ratio when constructing: Where IG(S, A) is the information gain of feature A to data set S, IV(A) is the intrinsic value of feature A, and the optimal feature mining rule; The decision tree branch ends are connected to SVM. For two-dimensional data, the training set is {(x1, y1), (x1, y1), …, (x n ,y n )}, where x i is the input feature vector, y i For the corresponding carbon emission label, SVM finds the optimal hyperplane ω·x+b=0, solution: subject to y i (ω·x i +b)≥1-ξ i ,ξ i ≥1,2,…,n Among them, ω is the hyperplane normal vector, b is the intercept, C is the penalty parameter used to balance the model complexity and training error, and ξi is the slack variable; Using kernel function Mapping low dimensions to high dimensions to identify anomalies; Among them, γ is the kernel parameter, and the cross-validation selects the value with the highest accuracy of the validation set within [0.1, 10]; For SVM to determine suspected abnormal areas, an MLP neural network with at least two hidden layers is introduced. The two hidden layers are activated by ReLU, f(x) = max(0, x). When there are three layers, let the input x, the hidden layer weight matrix W1, W2, the bias vector b1, b2, the output layer weight matrix W3, the bias vector b3, and the hidden layer output: h1=ReLU(W1x+b1) h2=ReLU(W2h1+b2) Output layer prediction value: Training with Xavie r Initializing weights prevents gradient problems and accelerates convergence.

5. The carbon emission data processing method according to claim 4, characterized in that: It also includes model training and optimization, including: Stratified sampling: 70% training set, 15% validation set, and 15% test set are taken from preprocessed historical data, and training set labels are manually annotated according to industry standards; Iterative training, first train the decision tree to 8 layers to divide the data space, fix the tree structure, use grid search and cross-validation to optimize the SVM hyperparameters C and γ, and then use back propagation to train the neural network on the SVM to screen suspected abnormal data, with the objective function: Where m is the number of samples, is the predicted value, yi is the true value, the training uses L2 regularization to constrain the neural network weights, and the penalty coefficient [0.001, 0.01] is adjusted according to the performance of the validation set; Evaluation on the validation set: if the accuracy does not reach 90%, adjust the model or increase the amount of data, repeat until stable and high performance is achieved, and finally independently test the generalization ability on the test set.

6. The carbon emission data processing method according to claim 1, characterized in that: The energy optimization configuration assistance specifically includes the following steps: Feature extraction and related modeling: When the hybrid model processes data, energy-related features such as energy type, input rate, equipment power, production load, and corrected carbon emission data are extracted to form a comprehensive feature set. PCA is used for dimensionality reduction. Let the original comprehensive feature set matrix X, covariance Find the eigenvalues, sort the vectors, select the first k to form the projection matrix P, and get the reduced eigenvector X PCA =XP, and a multiple linear regression model is built based on this: y=β0+β1X PCA1 +β2X PCA2 +…+b k X PCAk Among them, y is the energy utilization efficiency, βi is the regression coefficient, and the coefficient is obtained by the least square method. Among them, m is the number of samples, yi is the actual energy efficiency value, is the predicted value.

7. The carbon emission data processing method according to claim 1, characterized in that: In the real-time monitoring and feedback adjustment steps: Real-time monitoring: The carbon emission data collected and pre-processed in real time are input into the trained hybrid model. If the probability of a single data point being abnormal is higher than 0.7, it is considered abnormal. A sliding window is used to comprehensively judge at least 5 consecutive data points. If the abnormal proportion in the window exceeds 60%, an early warning is issued even if the single data point does not exceed the threshold. Feedback adjustment: feedback is initiated immediately when an abnormality is found, so that the data acquisition end is calibrated or resampled. When an abnormality occurs, calibration is triggered and the standard gas source is used to calibrate the zero point and range. After calibration, data is resampled and then entered into the model for verification; Abnormal data and correction information are fed back to the data management system to update the database.

8. A carbon emission data processing device, characterized in that: The device comprises: Data collection and preprocessing module: construct a multivariate data collection network and preprocess the collected data to obtain standardized carbon emission data for subsequent analysis; Abnormal data identification and correction module based on hybrid machine learning model: using a hybrid model that integrates decision tree, support vector machine and neural network to identify and correct abnormalities in pre-processed data and output accurate carbon emission data; Energy optimization configuration auxiliary module: mines the relationship between carbon emission data and energy utilization, and provides optimization configuration suggestions for energy utilization based on the processed carbon emission data; Real-time monitoring and feedback adjustment module: Continuously monitor the carbon emission data collected in real time, and provide timely feedback and adjustments when anomalies are found.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images