A carbon emission prediction system and method for papermaking process based on BO-GBDT

Abstract

This invention discloses a carbon emission prediction system and method for papermaking processes based on BO-GBDT. The system collects process parameters and energy consumption data during papermaking production, preprocesses the input data, constructs a feature set containing key influencing factors, and obtains a carbon emission dataset for model training. A Bayesian optimization algorithm is introduced to intelligently search for the optimal hyperparameters of four gradient boosting models, constructing a hyperparameter optimization space and setting an optimization objective function. Model performance is evaluated through cross-validation, and the optimal model is updated and selected. BO-GBDT is selected as the final prediction model. The trained model is then used to accurately predict carbon emissions from the papermaking process. By automatically optimizing the hyperparameters of the gradient boosting decision tree model using Bayesian optimization and combining it with multi-source data feature engineering, high-precision and high-efficiency prediction of carbon emissions from the papermaking process is achieved, providing an effective tool for carbon management in the production process.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and machine learning technology, specifically relating to a carbon emission prediction system and method for papermaking processes based on BO-GBDT. Background Technology

[0002] As a vital foundational industry of the national economy, the paper industry is also an energy-intensive sector and a major carbon emitter. Under the "dual-carbon" strategy, it urgently needs to improve the precision of carbon emission management through intelligent means. The paper production process encompasses multiple stages, including pulping, bleaching, papermaking, and drying, involving numerous process parameters such as pulp concentration, temperature, pressure, flow rate, and machine speed, as well as energy consumption variables such as steam consumption and electricity usage. These factors form a highly complex nonlinear mapping relationship with carbon emissions. Traditional carbon emission prediction methods often rely on empirical formulas or linear regression models, which struggle to effectively capture the characteristics of such high-dimensional, dynamic, and nonlinear processes, leading to significant prediction biases and failing to provide reliable support for process optimization and emission reduction decisions.

[0003] Machine learning-based modeling methods have demonstrated significant advantages in industrial process prediction in recent years, especially Gradient Boosting Decision Trees (GBDT) and its derivative algorithms such as XGBoost, LightGBM, and CatBoost. These methods are widely used in regression prediction tasks due to their powerful nonlinear fitting capabilities, automatic learning mechanisms for feature interactions, and good noise resistance. However, the prediction accuracy of these models is highly dependent on the proper configuration of hyperparameters, including key parameters such as learning rate, maximum tree depth, subsampling rate, and number of estimators. Improper settings can easily lead to overfitting or underfitting, affecting the model's generalization ability.

[0004] Existing technologies for predicting carbon emissions in the papermaking process generally suffer from problems such as single-model approach and low efficiency in hyperparameter tuning. Conventional grid search or random search strategies have high computational costs and slow convergence speed in high-dimensional parameter spaces, making it difficult to efficiently explore the global optimum. Although Bayesian optimization, as an intelligent optimization method based on a probabilistic surrogate model, can construct an approximation of the objective function through Gaussian process regression and use a collection function to balance exploration and utilization, approximating the optimal parameter combination with fewer iterations, no research has yet systematically applied it to a joint optimization framework of multiple gradient boosting algorithms to solve the high-precision modeling challenge of carbon emission prediction in the specific scenario of papermaking. Therefore, how to integrate Bayesian optimization with multiple gradient boosting algorithms to construct an efficient, robust, and comparable carbon emission prediction method has become a pressing technical bottleneck that needs to be overcome. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a carbon emission prediction system and method for papermaking processes based on BO-GBDT, thereby solving the aforementioned technical problems in the prior art.

[0006] The objective of this invention can be achieved through the following technical solutions: A carbon emission prediction system for papermaking process based on BO-GBDT includes a data acquisition and data processing module, a gradient algorithm model for processing regression tasks module, a Bayesian gradient algorithm model hyperparameter optimization module, and an optimized prediction model module. The data acquisition and data processing module acquires production process data of paper mills, including process parameters, energy consumption data, and carbon emission data; preprocesses the raw data, including missing value handling, outlier detection and removal, constructs a feature set containing key influencing factors, and obtains a carbon emission dataset for model training. The gradient algorithm model for processing regression tasks introduces four gradient boosting algorithm models: GBDT, XGBoost, LightGBM, and CatBoost. These models establish regression prediction models for the nonlinear relationship between process parameters, energy consumption variables, and carbon emissions in the papermaking process. The preprocessed dataset is divided into training and test sets according to a ratio. The training set is used to initially train the four models, forming a nonlinear mapping relationship between process parameters, energy consumption variables, and carbon emissions. The Bayesian gradient optimization algorithm model hyperparameter module defines the hyperparameter search space for four models: GBDT, XGBoost, LightGBM, and CatBoost. It uses the Bayesian optimization algorithm to initialize Gaussian process regression as a probabilistic surrogate model and sets the acquisition function. In each iteration, the acquisition function selects a hyperparameter combination from the hyperparameter space based on the current posterior distribution. This hyperparameter combination is used to train the model, and the objective function value is calculated on the test set. The posterior distribution of the Gaussian process is updated based on new observations. The iteration process is repeated until the preset number of iterations or convergence condition is reached, and the optimal hyperparameter configuration for each model during the iteration process is recorded. Optimized prediction model module: Based on the optimal hyperparameter configuration obtained by Bayesian optimization, GBDT, XGBoost, LightGBM, and CatBoost models are retrained respectively to obtain optimized BO-GBDT, BO-XGBoost, BO-LightGBM, and BO-CatBoost prediction models; the trained optimized models are used to predict carbon emissions in the papermaking process, inputting real-time or historical process parameters and outputting corresponding carbon emission prediction values; the prediction results of the four prediction models are compared and analyzed, and the model with the best performance is selected for final prediction.

[0007] The method for predicting carbon emissions from the papermaking process based on BO-GBDT includes the following steps: S1. Data acquisition and data processing; S2. Construct a gradient boosting algorithm regression prediction model; S3, Bayesian optimization of hyperparameter search; S4. Construct the optimized prediction model and make predictions.

[0008] Further, S1 includes: S101. Data Acquisition: Collect process parameter data, energy consumption data, and carbon emission data during the production process of papermaking enterprises. Among them, process parameters include pulp concentration, temperature, pressure, flow rate, and machine speed; energy consumption data includes steam consumption; and the carbon emission data are carbon dioxide equivalent emission values ​​calculated based on energy consumption or obtained by direct monitoring. S102. Data Preprocessing: The collected raw data is preprocessed to identify and handle missing values ​​using interpolation, mean imputation, or deletion methods; outlier detection is performed, using the 3σ criterion to identify and remove outlier data points, and using a sliding window to smooth the data and reduce noise interference; a feature set containing key influencing factors is constructed; the processed dataset is divided into training and test sets in a 7:3 ratio.

[0009] Furthermore, the 3σ criterion is defined as follows: If a data point xi satisfies |xi-μ|>3σ, it is considered an outlier; where μ is the mean of the feature sequence and σ is the standard deviation; a sliding window is used to smooth the data to reduce noise interference, and the length of the sliding window is 5-10 data points. The feature set includes all collected process parameters, energy consumption variables, and their preprocessed derived features.

[0010] Further, S2 includes: S201. Establish regression prediction models for four gradient boosting algorithms: GBDT, XGBoost, LightGBM, and CatBoost; set the initial hyperparameters for each model. S202. Using the training set data, perform preliminary training on four models to learn the nonlinear mapping relationship between process parameters, energy consumption variables and carbon emissions in the papermaking process. The preliminary training process is as follows: the feature data of the training set is used as input and the corresponding carbon emission data is used as output label, which are respectively input into the four gradient boosting algorithm models. The model parameters are iteratively updated by minimizing the loss function, where the loss function is mean squared error (MSE) or mean absolute error (MAE). S203. Evaluate the initial prediction performance of each model on the test set, and record RMSE, MAE, and R.2 Evaluation indicators.

[0011] Furthermore, the initial hyperparameters in S201 include: learning rate set to 0.1, maximum tree depth set to 6, and number of estimators set to 50.

[0012] The evaluation metrics include root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). 2 .

[0013] Further, S3 includes: S301. For the four models GBDT, XGBoost, LightGBM and CatBoost, define the hyperparameter search space, including the learning rate (range [0.01, 0.3]), the maximum tree depth (range [3, 10]), the subsample rate (range [0.5, 1.0]), and the number of estimators (n_estimators) (range [50, 500]). S302. Using the Bayesian optimization algorithm, Gaussian process regression is initialized as a probabilistic surrogate model, and a data acquisition function is set. In each iteration, the data acquisition function selects the most promising combination of hyperparameters in the hyperparameter space based on the current posterior distribution. The model is trained using this hyperparameter combination, and R is calculated on the test set. 2 As the value of the objective function; S303. Update the posterior distribution of the Gaussian process based on the new observations, and update the surrogate model's understanding of the hyperparameter space; repeat the iteration process, setting the number of iterations to 50-100, or stopping when the objective function value improves by less than 0.001 after 10 consecutive iterations, which is considered a convergence condition; record the optimal hyperparameter configuration and the corresponding maximum R-squared for each model during the iteration process. 2 value.

[0014] Furthermore, the Gaussian process regression uses a squared exponential kernel function as the covariance function; The objective function value is the coefficient of determination R calculated by the model on the test set. 2 .

[0015] Further, S4 includes: S401. Based on the optimal hyperparameters obtained by Bayesian optimization, retrain GBDT, XGBoost, LightGBM and CatBoost models respectively to obtain optimized BO-GBDT, BO-XGBoost, BO-LightGBM and BO-CatBoost prediction models. S402. Use the trained optimization model to predict carbon emissions in the papermaking process. Input real-time or historical process parameters and output the corresponding predicted carbon emissions. S403. Compare and analyze the prediction results of the four prediction models, select the model with the best performance, and visualize the prediction results to provide data support for carbon emission management, energy conservation and emission reduction decisions and production optimization for paper manufacturing enterprises.

[0016] Furthermore, when retraining the model with the optimal hyperparameters, the complete training set data is used for training; the comparative analysis includes comparing the RMSE, MAE, and R of each model on the independent validation set. 2 The metrics are used to calculate the average relative error between predicted and actual values; the criteria for selecting the best-performing model are: having the minimum RMSE and MAE on both the test and validation sets, and the maximum R-value. 2 value.

[0017] The beneficial effects of this invention are: 1. This invention employs four advanced gradient boosting algorithms, fully leveraging their nonlinear fitting capabilities to accurately capture the complex coupling relationships between multiple process parameters, energy consumption, and carbon emissions during papermaking, thus solving the problem of insufficient prediction accuracy of traditional linear models and empirical formulas. After Bayesian optimization, the BO-GBDT model R... 2 It can reach above 0.9067 and RMSE as low as 0.0022, showing a significant performance improvement compared to the initial model. Its prediction accuracy far exceeds that of existing technologies, which can meet the high-precision carbon emission monitoring needs of enterprises.

[0018] 2. This invention introduces a Bayesian optimization algorithm to replace traditional grid search and random search. Through a Gaussian process regression surrogate model and an intelligent acquisition function, it explores the optimal solution in the hyperparameter space in a targeted manner, without traversing all parameter combinations. Convergence is achieved in only 50-100 iterations, reducing the number of iterations by more than 60% compared to grid search, significantly shortening the parameter tuning time, reducing computational resource consumption, and avoiding the subjective errors of manual parameter tuning, thus ensuring the optimality of hyperparameter configuration.

[0019] 3. This invention systematically compares the performance of four gradient boosting algorithms in the papermaking carbon emission prediction task, using RMSE, MAE, and R... 2 The model employs a comprehensive evaluation of multiple indicators, including average relative error, to select the optimal model based on different production conditions and data scales, thus addressing the issue of insufficient applicability of a single model. Furthermore, the model supports real-time data input and prediction, making it adaptable to the dynamic conditions of continuous production in paper mills and demonstrating strong engineering practicality.

[0020] 4. This invention can be directly integrated into the existing production monitoring systems of paper manufacturing enterprises, outputting real-time carbon emission prediction results. This provides accurate data support for enterprises to optimize process parameters (such as adjusting machine speed and steam flow) and formulate energy-saving and emission-reduction plans, helping them reduce carbon emission intensity and energy consumption, and lower carbon trading costs. Simultaneously, the prediction results can assist enterprises in completing carbon accounting and carbon report preparation, meeting policy compliance requirements, thus achieving both environmental and economic benefits.

[0021] 5. The technical framework of this invention is universal, and the core logic is not limited to the paper industry. It can be extended to carbon emission prediction tasks in other high-energy-consuming and high-emission industries such as steel, cement, and chemical industries by simply adjusting the parameter types and ranges of data collection. It provides technical support for the low-carbon transformation of the entire industry and has a wide range of application scenarios. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0023] Figure 1 This is a schematic diagram of the overall process structure of an embodiment of the present invention; Figure 2 This is a scatter plot of carbon emission prediction results for four gradient algorithm models in this embodiment of the invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0025] like Figure 1 As shown, this invention provides a method for predicting carbon emissions in a papermaking process based on BO-GBDT, comprising the following steps: S1. Data Acquisition and Data Processing S101. Data Acquisition: Through the enterprise's production monitoring system, energy management system, and carbon emission monitoring equipment, 20 process parameters (paper machine speed 1000-1400 m / min, steam flow rate 14-22 m³ / min) are simultaneously collected during the papermaking process. 3 / min, application rate 60-130 g / m 2 Data on steam consumption and energy consumption (unit: tons / ton of paper), and carbon dioxide equivalent emission data calculated based on energy consumption and carbon emission factors (unit: tons of CO2 / ton of paper).

[0026] S102. Data Preprocessing: Missing Value Handling: For continuous variables, missing values ​​are filled using the forward imputation method, and samples with a missing value ratio exceeding 10% are deleted; Outlier Handling: Outlier data points are identified and removed using the 3σ criterion, and the mean μ and standard deviation σ of each feature are calculated, removing data that exceed the range [μ-3σ, μ+3σ]; Data Smoothing: A sliding window method with a window size of 5 is used to replace the original data with the mean of the data within the window to reduce noise; Dataset Partitioning: The preprocessed effective data is divided into training and test sets according to a preset ratio (7:3 or 8:2), and a random seed of 42 is set to ensure repeatability. In this embodiment, 4105 effective samples are finally obtained.

[0027] S2. Construct a gradient boosting algorithm regression prediction model. S201. Model Initialization: Establish four regression prediction models: GBDT, XGBoost, LightGBM, and CatBoost. Set the initial hyperparameters: learning rate = 0.1, maximum tree depth = 6, number of estimators = 50, and other parameters use the default values ​​of the algorithm.

[0028] S202. Initial Model Training: The feature data of the training set and the corresponding carbon emission labels were input into the four models. The model parameters were iteratively updated using MSE as the loss function to learn the nonlinear mapping relationship between variables. The training times for each model were: GBDT 0.889s, XGBoost 0.074s, LightGBM 0.292s, and CatBoost 1.591s. LightGBM and XGBoost were significantly more efficient than the other two models.

[0029] S203. Initial Performance Evaluation: Calculate the RMSE, MAE, and R-value for each model on the test set. 2 The indicators and evaluation formulas are as follows: Root Mean Square Error (RMSE): ; Mean Absolute Error (MAE): ; Coefficient of determination R 2 (R-squared): ; Where n is the number of samples. This represents actual carbon emissions. To predict carbon emissions, This represents the average of actual carbon emissions.

[0030] The initial model performance evaluation results are shown in Table 1 below. CatBoost, XGBoost, and LightGBM have better initial performance than GBDT. 2 All exceeded 0.83, with CatBoost performing the best.

[0031] Table 1 Initial Model Performance Evaluation Results

[0032] S3, Bayesian optimization of hyperparameter search S301. Hyperparameter Space Definition: Based on the range set in the above modules, define the hyperparameter search space for the four models respectively, covering common and unique hyperparameters, to ensure the rationality and comprehensiveness of the search range.

[0033] S302, Bayesian Optimization Execution: Initialize the Gaussian process regression surrogate model, select the quadratic exponential kernel function, and set the expected improvement EI as the acquisition function. In each iteration, the acquisition function selects a combination of hyperparameters based on the current posterior distribution. After training the model, calculate the test set R. 2 As the objective function value, the posterior distribution is updated to optimize the search direction.

[0034] S303, Optimization Termination and Result Recording: Set the iteration count to 80. When R has been continuously iterated for 10 times... 2 Early convergence occurs when the improvement is less than 0.001. The optimal hyperparameter configurations for the four models are obtained as follows: GBDT optimal parameters: OrderedDict([('learning_rate', 0.049988401985468195), ('max_depth', 15), ('max_features', 0.4956954986793187), ('min_samples_leaf',8), ('min_samples_split', 5), ('n_estimators', 477), ('subsample',0.8439286262075258)]); CatBoost optimal parameters: OrderedDict([('bagging_temperature', 0.0), ('depth',8), ('iterations', 180), ('l2_leaf_reg', 1.0), ('learning_rate',0.24825305161186226)]); LightGBM optimal parameters: OrderedDict([('feature_fraction', 1.0), ('learning_rate', 0.01), ('max_depth', 15), ('min_child_samples', 5), ('n_estimators',500), ('num_leaves', 255), ('reg_alpha', 0.0), ('reg_lambda', 1.0), ('subsample', 1.0)]); XGBoost optimal parameters: OrderedDict([('colsample_bytree',0.5859078632290013), ('gamma', 0.0), ('learning_rate', 0.29999999999999993), ('max_depth', 15), ('min_child_weight', 10.0), ('n_estimators', 50), ('reg_alpha', 0.0), ('reg_lambda', 1.0), ('subsample', 1.0)]).

[0035] S4. Construct the optimized prediction model and perform predictions. S401. Model Retraining: Using the complete training set data, retrain the four models based on the above optimal hyperparameters to obtain optimized BO-GBDT, BO-XGBoost, BO-LightGBM and BO-CatBoost models, ensuring that the model performance is fully released.

[0036] S402, Carbon Emission Prediction: Input the test set data and real-time collected process parameters into the optimized model, output the corresponding carbon emission prediction value, and verify the model's generalization ability and real-time prediction performance.

[0037] S403. Model Comparison, Optimization, and Result Application: Calculate the RMSE, MAE, and R of the four optimization models. 2 The average relative error was used to compare and analyze the model performance, and the results are shown in Table 2 below. As can be seen from the table, the BO-GBDT model showed the most significant performance improvement after optimization, with R... 2 The accuracy improved from 0.6234 to 0.9067, while RMSE and MAE decreased to 0.0022 and 0.0016 respectively, making it the best among the four models. The scatter plot of the prediction results versus the actual values ​​for the BO-GBDT model as the final prediction model is shown below. Figure 2As shown, the predicted values ​​fit the actual values ​​very well, with no obvious deviation.

[0038] By integrating this model into the enterprise production management system, it can output carbon emission prediction results in real time, providing accurate data support for process parameter optimization, energy consumption control, and energy conservation and emission reduction decisions. At the same time, it can intuitively display the trend of carbon emission changes through a visual interface, assisting enterprises in carbon accounting and low-carbon planning.

[0039] Table 2 Performance evaluation results of the optimized model

[0040] Reference Figure 2 The x-axis represents actual carbon emissions (tons of CO2 / ton of paper), and the y-axis represents predicted carbon emissions (tons of CO2 / ton of paper). The diagonal line in the graph represents the ideal fit line. As shown in the graph, the scatter points of the BO-GBDT model are most densely distributed near the ideal fit line, with no significant deviations; the BO-LightGBM scatter points are distributed next; the BO-CatBoost and BO-XGBoost scatter points show a small number of deviations, further verifying that the BO-GBDT model has the best prediction accuracy.

[0041] Example 1: A carbon emission prediction method for the papermaking process based on a 7:3 dataset partitioning like Figure 1 As shown in the figure, this invention provides a method for predicting carbon emissions in the papermaking process based on BO-GBDT, and the specific steps are as follows: 1. Data Acquisition and Processing Production data for 30 consecutive days was collected from the kraft paper production line of a large paper manufacturing enterprise. Twenty process parameters, including paper machine speed, pulp concentration, steam flow rate, and sizing amount, were collected through the DCS system. Steam consumption data (unit: tons / ton of paper) was collected through the energy management system. According to the "Carbon Emission Accounting Method for the Paper Industry" and combined with the corresponding energy carbon emission factor (steam carbon emission factor is 0.185 tons CO2 / ton of steam), the carbon dioxide equivalent emission value (unit: tons CO2 / ton of paper) was calculated and used as carbon emission labeling data.

[0042] The collected raw data underwent preprocessing as follows: Missing value handling: A total of 128 missing values ​​were detected, all of which were continuous variables. Forward imputation was used to fill the missing values, and no samples were deleted. Outlier handling: 47 outlier data points were identified using the 3σ criterion, mainly abrupt changes in vehicle speed and steam flow, and were removed. Data smoothing: A sliding window mean method with a window size of 5 was used to smooth all feature data to reduce noise interference from equipment monitoring. Dataset partitioning: 4105 valid samples were obtained and randomly divided into a training set (2873 samples) and a test set (1232 samples) in a 7:3 ratio. A random seed of 42 was set to ensure the reproducibility of the experiment.

[0043] 2. Construction and Initial Training of the Basic Gradient Boosting Model Four gradient boosting regression models were built using Python with the Scikit-learn, XGBoost, LightGBM, and CatBoost libraries, respectively. The initial hyperparameters were set as follows: learning_rate=0.1, max_depth=6, n_estimators=50, with other parameters set to default. The models were trained using the training set data with MSE as the loss function. The training times for each model were: GBDT 0.889 seconds, XGBoost 0.074 seconds, LightGBM 0.292 seconds, and CatBoost 1.591 seconds.

[0044] The initial model performance was evaluated on the test set, and the results are shown in Table 1. It can be seen that CatBoost has the best initial performance (R²). 2 =0.8573), GBDT's initial performance is relatively weak (R 2 =0.6234), mainly because the initial hyperparameters did not match the model characteristics, resulting in the GBDT fitting ability not being fully utilized.

[0045] 3. Bayesian optimization of hyperparameter search Bayesian optimization was implemented using the GPyOpt library. A Gaussian process regression surrogate model was initialized, with a squared exponential kernel function selected. EI was set as the acquisition function, the number of iterations was set to 80, and the convergence condition was 10 consecutive iterations of R. 2 The improvement is less than 0.001. Based on the above definition of hyperparameter space, hyperparameter optimization was performed on the four models respectively. Finally, each model converged within 60-75 iterations, significantly shortening the optimization time.

[0046] After optimization, the optimal hyperparameter configurations for each model were obtained. The learning rate of GBDT was optimized to around 0.05, and n_estimators was increased to 477, effectively improving the model's fitting ability. The learning_rate of XGBoost was optimized to 0.3, and max_depth was increased to 15, enhancing both model complexity and fitting accuracy.

[0047] 4. Optimize model training and prediction applications Using the complete training set data, four models were retrained based on the optimal hyperparameters to obtain optimized models. Performance was evaluated on the test set, and the results are shown in Table 2. The BO-GBDT model showed the most significant performance improvement, with R... 2 The performance improved from 0.6234 to 0.9067, RMSE decreased from 0.0044 to 0.0022, and MAE decreased from 0.0034 to 0.0016, significantly outperforming other models in prediction accuracy; BO-LightGBM was the second best performing (R...2 =0.8938), the performance of BO-CatBoost and BO-XGBoost is slightly improved but not significantly changed, mainly because their initial models already have a high degree of fit.

[0048] The BO-GBDT model was used as the final prediction model and integrated into the enterprise's production monitoring platform to collect process parameters in real time and output predicted carbon emissions. Operational verification showed that the model's prediction delay was less than 0.5 seconds, meeting real-time monitoring requirements; the average relative error between the predicted and actual monitored values ​​was 0.82%, meeting the enterprise's requirements for process optimization and carbon accounting. Based on model feedback, the enterprise adjusted the paper machine speed to 1200 m / min and optimized the steam flow rate to 18 m³ / min. 3 / min, reducing carbon emissions per unit product by 3.2%, achieving the energy conservation and emission reduction target.

[0049] Example 2: A carbon emission prediction method for the papermaking process based on an 8:2 dataset partitioning The difference between this embodiment and Embodiment 1 is that the dataset is divided in an 8:2 ratio, with the training set accounting for 80% (3284 samples) and the test set accounting for 20% (821 samples). The other steps (data collection, preprocessing, model building, Bayesian optimization process) are completely consistent with Embodiment 1.

[0050] After the same Bayesian optimization process, the performance of each model improved slightly, with the BO-GBDT model showing the best performance on the test set: R 2 =0.9152, RMSE=0.0020, MAE=0.0015, and the mean relative error decreased to 0.75%, with prediction performance slightly better than Example 1. This is because the increased training set size allows the model to learn more comprehensive feature patterns, further improving fitting accuracy.

[0051] This embodiment verifies the stability of the method of the present invention under different dataset partitioning ratios. When there are sufficient training samples, the model prediction accuracy can be further improved, adapting to application scenarios with different data scales.

[0052] Example 3: Cross-condition applicability verification To verify the cross-condition applicability of the method of this invention, data from the corrugated paper production line of the enterprise (different from the kraft paper production conditions in Example 1) were selected, and a model was constructed and predicted using the method of Example 1. The results show that the R-value of the BO-GBDT model under the corrugated paper production conditions is [missing information]. 2 =0.8876, RMSE=0.0025, MAE=0.0018, average relative error=0.95%. Although the performance is slightly lower than that of kraft paper, it still maintains high precision, indicating that the method of the present invention can be adapted to the production conditions of different paper types and has strong applicability.

[0053] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A carbon emission prediction system for papermaking processes based on BO-GBDT, characterized in that, It includes a data acquisition and data processing module, a gradient algorithm model for processing regression tasks module, a Bayesian gradient algorithm model hyperparameter optimization module, and an optimized prediction model module; The data acquisition and data processing module acquires production process data of paper mills, including process parameters, energy consumption data, and carbon emission data; preprocesses the raw data, including missing value handling, outlier detection and removal, constructs a feature set containing key influencing factors, and obtains a carbon emission dataset for model training. The gradient algorithm model for processing regression tasks introduces four gradient boosting algorithm models: GBDT, XGBoost, LightGBM, and CatBoost. These models establish regression prediction models for the nonlinear relationship between process parameters, energy consumption variables, and carbon emissions in the papermaking process. The preprocessed dataset is divided into training and test sets according to a ratio. The training set is used to initially train the four models, forming a nonlinear mapping relationship between process parameters, energy consumption variables, and carbon emissions. The hyperparameter module for Bayesian gradient optimization algorithm models defines the hyperparameter search space for four models: GBDT, XGBoost, LightGBM, and CatBoost. It uses the Bayesian optimization algorithm to initialize Gaussian process regression as a probabilistic surrogate model and sets the acquisition function. In each iteration, the acquisition function selects a combination of hyperparameters in the hyperparameter space based on the current posterior distribution, trains the model using the hyperparameter combination, calculates the objective function value on the test set, and updates the posterior distribution of the Gaussian process based on the new observations. Repeat the iterative process until the preset number of iterations or convergence condition is reached, and record the optimal hyperparameter configuration for each model during the iterative process; Optimized prediction model module: Based on the optimal hyperparameter configuration obtained by Bayesian optimization, GBDT, XGBoost, LightGBM and CatBoost models are retrained respectively to obtain optimized BO-GBDT, BO-XGBoost, BO-LightGBM and BO-CatBoost prediction models; the trained optimized models are used to predict carbon emissions in the papermaking process, inputting real-time or historical process parameters and outputting the corresponding carbon emission prediction values; The prediction results of the four prediction models were compared and analyzed, and the model with the best performance was selected for the final prediction.

2. The method for predicting carbon emissions in the papermaking process based on BO-GBDT according to claim 1, characterized in that, Includes the following steps: S1. Data acquisition and data processing; S2. Construct a gradient boosting algorithm regression prediction model; S3, Bayesian optimization of hyperparameter search; S4. Construct the optimized prediction model and make predictions.

3. The method for predicting carbon emissions in the papermaking process based on BO-GBDT according to claim 2, characterized in that, S1 includes: S101. Data Acquisition: Collect process parameter data, energy consumption data, and carbon emission data during the production process of papermaking enterprises. Among them, process parameters include pulp concentration, temperature, pressure, flow rate, and machine speed; energy consumption data includes steam consumption; and the carbon emission data are carbon dioxide equivalent emission values ​​calculated based on energy consumption or obtained by direct monitoring. S102. Data Preprocessing: The collected raw data is preprocessed to identify and handle missing values ​​using interpolation, mean imputation, or deletion methods; outlier detection is performed, using the 3σ criterion to identify and remove outlier data points, and using a sliding window to smooth the data and reduce noise interference; a feature set containing key influencing factors is constructed; the processed dataset is divided into training and test sets in a 7:3 ratio.

4. The method for predicting carbon emissions from the papermaking process based on BO-GBDT according to claim 3, characterized in that, The 3σ criterion is defined as follows: If a data point xi satisfies |xi-μ|>3σ, it is considered an outlier; where μ is the mean of the feature sequence and σ is the standard deviation; a sliding window is used to smooth the data to reduce noise interference, and the length of the sliding window is 5-10 data points. The feature set includes all collected process parameters, energy consumption variables, and their preprocessed derived features.

5. The method for predicting carbon emissions in the papermaking process based on BO-GBDT according to claim 2, characterized in that, S2 includes: S201. Establish regression prediction models for four gradient boosting algorithms: GBDT, XGBoost, LightGBM, and CatBoost; set the initial hyperparameters for each model. S202. Using the training set data, perform preliminary training on four models to learn the nonlinear mapping relationship between process parameters, energy consumption variables and carbon emissions in the papermaking process. The preliminary training process is as follows: the feature data of the training set is used as input and the corresponding carbon emission data is used as output label, which are respectively input into the four gradient boosting algorithm models. The model parameters are iteratively updated by minimizing the loss function, where the loss function is mean squared error (MSE) or mean absolute error (MAE). S203. Evaluate the initial prediction performance of each model on the test set, and record RMSE, MAE, and R. 2 Evaluation indicators.

6. The method for predicting carbon emissions in the papermaking process based on BO-GBDT according to claim 5, characterized in that, The initial hyperparameters in S201 include: learning rate set to 0.1, maximum tree depth set to 6, and number of estimators set to 50. The evaluation metrics include root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). 2 .

7. The method for predicting carbon emissions in the papermaking process based on BO-GBDT according to claim 2, characterized in that, S3 includes: S301. For the four models GBDT, XGBoost, LightGBM and CatBoost, define the hyperparameter search space, including the learning rate (range [0.01, 0.3]), the maximum tree depth (range [3, 10]), the subsample rate (range [0.5, 1.0]), and the number of estimators (n_estimators) (range [50, 500]). S302. Using the Bayesian optimization algorithm, Gaussian process regression is initialized as a probabilistic surrogate model, and a data acquisition function is set. In each iteration, the data acquisition function selects the most promising combination of hyperparameters in the hyperparameter space based on the current posterior distribution. The model is trained using this hyperparameter combination, and R is calculated on the test set. 2 As the value of the objective function; S303. Update the posterior distribution of the Gaussian process based on the new observations, and update the surrogate model's understanding of the hyperparameter space; repeat the iteration process, setting the number of iterations to 50-100, or stopping when the objective function value improves by less than 0.001 after 10 consecutive iterations, which is considered a convergence condition; record the optimal hyperparameter configuration and the corresponding maximum R-squared for each model during the iteration process. 2 value.

8. The method for predicting carbon emissions in the papermaking process based on BO-GBDT according to claim 7, characterized in that, The Gaussian process regression uses the squared exponential kernel function as the covariance function; The objective function value is the coefficient of determination R calculated by the model on the test set. 2 .

9. The method for predicting carbon emissions from the papermaking process based on BO-GBDT according to claim 2, characterized in that, S4 includes: S401. Based on the optimal hyperparameters obtained by Bayesian optimization, retrain GBDT, XGBoost, LightGBM and CatBoost models respectively to obtain optimized BO-GBDT, BO-XGBoost, BO-LightGBM and BO-CatBoost prediction models. S402. Use the trained optimization model to predict carbon emissions in the papermaking process. Input real-time or historical process parameters and output the corresponding predicted carbon emissions. S403. Compare and analyze the prediction results of the four prediction models, select the model with the best performance, and visualize the prediction results to provide data support for carbon emission management, energy conservation and emission reduction decisions and production optimization for paper manufacturing enterprises.

10. The carbon emission prediction system for papermaking process based on BO-GBDT according to claim 1, characterized in that, When retraining the model with the optimal hyperparameters, the complete training set data is used for training; the comparative analysis includes comparing the RMSE, MAE, and R of each model on the independent validation set. 2 The metrics are used to calculate the average relative error between predicted and actual values; the criteria for selecting the best-performing model are: having the minimum RMSE and MAE on both the test and validation sets, and the maximum R-value. 2 value.

Images