CN-122021991-A - SHAP-PSO-based papermaking process carbon emission optimization method

CN122021991ACN 122021991 ACN122021991 ACN 122021991ACN-122021991-A

Abstract

The invention discloses a papermaking process carbon emission optimization method based on SHAP-PSO. Firstly, establishing LightGBM carbon emission prediction models to accurately predict the carbon emission in the papermaking production process, dividing the carbon emission data into three layers of high, medium and low according to the emission level, adopting an interpretability analysis method to respectively calculate the contribution degree and importance sequence of each process feature under the three layers to the carbon emission, simultaneously obtaining global feature importance distribution, finally, taking SHAP analysis results as guidance, integrating feature importance information into a particle swarm optimization algorithm, providing feature selection strategies and optimizing direction boundary constraint through SHAP, and realizing the optimization adjustment of process parameters from high carbon to low carbon. The invention organically combines machine learning prediction, interpretability analysis and intelligent optimization algorithm, not only can accurately predict carbon emission, but also can provide interpretable optimization paths and specific technological parameter regulation schemes.

Inventors

CHEN XIAOBIN
CHEN YE
LIAO JIANMING
ZHANG MIN
DONG YUNYUAN
ZHANG GUANQING

Assignees

衢州学院

Dates

Publication Date: 20260512
Application Date: 20251216

Claims (10)

1. A method for optimizing carbon emissions in a papermaking process based on SHAP-PSO, comprising the steps of: s1, data acquisition and carbon emission prediction modeling, namely acquiring technological parameter data, energy consumption data and carbon emission data in the papermaking production process, dividing a training set and a testing set after preprocessing, constructing and training LightGBM a carbon emission regression prediction model, and ensuring that the model prediction precision meets the requirement; S2, dividing a carbon emission level and SHAP interpretive analysis, namely dividing a carbon emission sample into three levels of high, medium and low according to the number of the levels, calculating SHAP values of process features in the samples of each level by adopting a SHAP interpretive analysis method, acquiring overall and hierarchical feature importance sequences, and identifying key influence factors of different emission levels; S3, PSO intelligent optimization based on SHAP guidance, namely determining optimization variables, optimization directions and boundary constraints of a PSO algorithm according to SHAP analysis results, constructing an fitness function containing a process parameter deviation penalty term, initializing particle swarm parameters, executing PSO optimization, and searching for an optimal process parameter combination; and S4, verifying the optimization effect, analyzing the parameter adjustment amplitude and contribution degree, and outputting the process parameter regulation scheme meeting the production constraint.
2. The SHAP-PSO-based papermaking process carbon emission optimization method according to claim 1, wherein S1 specifically comprises: s101, collecting process parameter data, energy consumption data and carbon emission data in the production process of a paper-making enterprise, wherein the process parameter comprises slurry concentration, temperature, pressure, flow and vehicle speed; S102, preprocessing the collected original data, identifying and processing missing values, processing the missing data by adopting an interpolation method, a mean filling method or a deletion method, detecting the abnormal values, identifying and eliminating abnormal data points by using a 3 sigma criterion, smoothing the data by using a data sliding window to reduce noise interference, constructing a feature set containing key influence factors, dividing the processed data set into a training set and a test set according to the proportion of 7:3, and determining the difference between the training set and the test set; S103, constructing LightGBM a carbon emission regression prediction model, setting initial super parameters including a learning rate, the maximum depth of a tree and the number of leaf nodes, training the model by using training set data, learning a nonlinear mapping relation among process parameters, energy consumption variables and carbon emission, evaluating the model prediction performance on a test set, and recording RMSE, MAE, R 2 evaluation indexes.
3. The SHAP-PSO paper making process carbon emission optimization method according to claim 1, wherein S2 specifically comprises: S201, sorting carbon emission data according to emission levels, and dividing samples into three levels according to the quantile of carbon emission, wherein a high carbon emission level is a sample with the front 30% of carbon emission, a middle carbon emission level is a sample with the middle 45%, and a low carbon emission level is a sample with the rear 25%; S202, constructing LightGBM a carbon emission classification prediction model, and calculating the SHAP value of each process feature in each sample by utilizing TreeExplainer by adopting a SHAP interpretability analysis method; S203, carrying out SHAP value statistical analysis on samples of three carbon emission levels of high, medium and low respectively, calculating an average SHAP value and an absolute value of the SHAP value of each process feature under each level, and obtaining feature importance ranking of each level; S204, calculating the global feature importance, counting the average value of the absolute values of the SHAP values of all the features in all the samples, generating a global feature importance ranking chart, comparing and analyzing the feature importance differences of three levels, identifying key influence factors under different emission levels, and revealing dominant factors of the high-carbon emission samples and dominant features of the low-carbon emission samples.
4. The SHAP-PSO paper making process carbon emission optimization method according to claim 1, wherein S3 includes: s301, selecting N-bit process parameters before global feature importance ranking as optimization variables of a PSO algorithm according to SHAP analysis results, wherein the value range of N is 1-17; S302, determining the influence direction of each optimization variable under the high carbon emission level based on SHAP value analysis results of three levels, namely, positive characteristic of the SHAP value, increase of the SHAP value, decrease of the carbon emission of the SHAP value, and increase of the SHAP value; S303, setting optimization boundary constraint of each process parameter by combining the characteristic distribution of the low-carbon emission level sample, and ensuring that the optimized parameter is within the practical production feasible range; S304, constructing an adaptability function of a PSO algorithm, taking the carbon emission predicted by a LightGBM model as an adaptability value, and optimizing the target to minimize the predicted value; S305, initializing a particle swarm, setting the particle number to 40-60 and the maximum iteration number to 100-200, and setting inertial weight, individual learning factors and social learning factors PSO parameters; s306, aiming at a sample of a high carbon emission level, executing a PSO optimization algorithm, wherein each particle represents a group of process parameter combinations, and searching for optimal process parameters by iteratively updating the particle position and speed; S307, stopping optimizing when the iteration termination condition is met or the predicted carbon emission value is reduced below the minimum carbon emission threshold of the data set, and recording the technological parameter combination corresponding to the global optimal solution.
5. The method for optimizing carbon emissions during a papermaking process of SHAP-PSO as recited in claim 4, wherein said process parameter deviation penalty term is calculated by penalty value = ω x Σ| (x i -x i0 ）/x i0 |; Where ω is the penalty weight, x i is the post-optimization parameter value, and x i0 is the pre-optimization parameter initial value.
6. The SHAP-PSO paper making process carbon emission optimization method of claim 1, wherein S4 includes: S401, inputting LightGBM a prediction model into an optimal process parameter combination obtained by PSO optimization, calculating a carbon emission predicted value after optimization, and verifying an optimization effect; s402, analyzing the change condition of each process parameter before and after optimization, generating a parameter adjustment comparison table, and determining the specific adjustment direction and amplitude of each parameter; S403, combining SHAP analysis results, and explaining the contribution degree of each parameter adjustment to carbon emission reduction to form an interpretable optimization path; S404, outputting a specific technological parameter regulation and control scheme, including target values of optimized slurry concentration, temperature, pressure, flow and vehicle speed parameters; S405, carrying out feasibility assessment on the optimization scheme, verifying whether the optimized parameters meet production process requirements and equipment operation constraints, and carrying out multi-objective optimization if necessary, thereby reducing carbon emission and simultaneously considering production efficiency and product quality.
7. The method of optimizing papermaking process carbon emissions of SHAP-PSO of claim 1, wherein in the carbon emission hierarchy division, the 90 th percentile of carbon emissions of low carbon emission hierarchy samples is used as a minimum carbon emission threshold.
8. The method for optimizing carbon emissions in a papermaking process of SHAP-PSO according to claim 1, wherein a gradient lifting framework is adopted in the LightGBM model training process to learn a nonlinear mapping relationship among process parameters, energy consumption variables and carbon emissions, and the super parameters are determined by grid search or bayesian optimization.
9. The method of optimizing papermaking process carbon emissions of a SHAP-PSO of claim 1, wherein the SHAP interpretability analysis further includes generating a global feature importance ranking map, a hierarchical SHAP value summary map.
10. The method for optimizing carbon emissions in a papermaking process of SHAP-PSO according to claim 1, wherein the inertia weight of the PSO algorithm is linearly decreased by a linear decreasing strategy, an initial value is set to 0.9, and the overall searching capability and the local searching capability of the algorithm are balanced as the number of iterations linearly decreases to 0.65.

Description

SHAP-PSO-based papermaking process carbon emission optimization method Technical Field The invention belongs to the technical field of industrial process optimization and intelligent control, and particularly relates to a papermaking process carbon emission optimization method based on SHAP-PSO. Background The paper industry is taken as an important basic raw material industry, and the production process relates to a plurality of high-energy-consumption links such as pulping, papermaking, drying and the like, and is an important field of energy consumption and carbon emission. In the macroscopic context of achieving the "two carbon" goal, how to effectively reduce carbon emissions from papermaking processes has become a core challenge for the industry. The accurate modeling and optimization of the complex relation between the technological parameters and the carbon emission are key technical bases for achieving the aim. The prior art mainly has the following problems that the carbon emission prediction precision is insufficient, the traditional statistical regression method is difficult to accurately capture the complex nonlinear relation between the technological parameters and the carbon emission, meanwhile, the traditional machine learning prediction model is mostly a black box model, the interpretation is lacking, the specific influence mechanism and contribution degree of each technological parameter to the carbon emission cannot be revealed, and clear guidance is difficult to provide for technological optimization. In addition, the traditional optimization method cannot fully consider the difference of key influence factors under different carbon emission levels, and on the premise of ensuring production stability, a plurality of process parameters are difficult to systematically and cooperatively optimize, so that the optimization effect is limited. Therefore, how to construct a method which has high prediction precision and strong interpretability and can realize systematic parameter optimization becomes a technical problem to be solved in the field of carbon emission reduction in the papermaking process. Disclosure of Invention Aiming at the defects of the prior art, the invention aims to provide a SHAP-PSO-based carbon emission optimization method for a papermaking process, which solves the technical problems in the prior art. The aim of the invention can be achieved by the following technical scheme: a SHAP-PSO based papermaking process carbon emission optimization method comprising the steps of: s1, data acquisition and carbon emission prediction modeling, namely acquiring technological parameter data, energy consumption data and carbon emission data in the papermaking production process, dividing a training set and a testing set after preprocessing, constructing and training LightGBM a carbon emission regression prediction model, and ensuring that the model prediction precision meets the requirement; S2, dividing a carbon emission level and SHAP interpretive analysis, namely dividing a carbon emission sample into three levels of high, medium and low according to the number of the levels, calculating SHAP values of process features in the samples of each level by adopting a SHAP interpretive analysis method, acquiring overall and hierarchical feature importance sequences, and identifying key influence factors of different emission levels; S3, PSO intelligent optimization based on SHAP guidance, namely determining optimization variables, optimization directions and boundary constraints of a PSO algorithm according to SHAP analysis results, constructing an fitness function containing a process parameter deviation penalty term, initializing particle swarm parameters, executing PSO optimization, and searching for an optimal process parameter combination; and S4, verifying the optimization effect, analyzing the parameter adjustment amplitude and contribution degree, and outputting the process parameter regulation scheme meeting the production constraint. Further, the step S1 specifically includes: s101, collecting process parameter data, energy consumption data and carbon emission data in the production process of a paper-making enterprise, wherein the process parameter comprises slurry concentration, temperature, pressure, flow and vehicle speed; S102, preprocessing the collected original data, identifying and processing missing values, processing the missing data by adopting an interpolation method, a mean filling method or a deletion method, detecting the abnormal values, identifying and eliminating abnormal data points by using a 3 sigma criterion, smoothing the data by using a data sliding window to reduce noise interference, constructing a feature set containing key influence factors, dividing the processed data set into a training set and a test set according to the proportion of 7:3, and determining the difference between the training set and the test set; S103, constructing LightGBM a carbon