A multi-objective optimization method and apparatus for the regeneration of graphite from waste lithium batteries
By constructing a database of graphite regeneration from waste lithium batteries and combining it with a machine learning proxy model and lifecycle assessment, we achieved multi-objective optimization of the graphite regeneration process for waste lithium-ion batteries. This solved the problems of low efficiency and poor interpretability in existing technologies, and enabled the design of an efficient and low-carbon regeneration process.
Patent Information
- Application Number
- CN202611289732.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-25
- Publication Date
- 2026-11-10
AI Technical Summary
Existing methods for optimizing graphite recycling processes from waste lithium-ion batteries suffer from low R&D efficiency, difficulty in describing complex nonlinear relationships, lack of interpretability, and the independence of material properties from environmental impact assessments, leading to high costs and low efficiency.
A database for the recycling of graphite from waste lithium batteries was constructed, and a machine learning proxy model and a life cycle assessment model were established. Through a reverse multi-objective optimization method, the synergistic optimization of process parameters, graphite performance, and environmental impact was achieved.
The design efficiency of waste graphite recycling process has been improved, the consumption of chemical reagents, energy consumption and carbon emissions have been reduced, and the electrochemical performance and structural performance of graphite have been improved simultaneously.
Smart Images

Figure CN123070943A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of battery recycling, and in particular to a multi-objective optimization method and apparatus for the regeneration of graphite from waste lithium batteries. Background Technology
[0002] With the rapid development of new energy vehicles, energy storage systems, and consumer electronics, the market demand for lithium-ion batteries continues to grow, and a large number of power batteries and energy storage batteries are gradually entering the retirement stage. Waste lithium-ion batteries not only contain high-value metal resources such as lithium, nickel, and cobalt, but their negative electrode graphite material accounts for approximately 15% to 25% of the total battery mass, possessing high resource recycling value. As the number of waste lithium-ion batteries continues to increase, how to achieve high-value recycling of waste graphite materials has become an important research direction in the field of lithium-ion battery recycling.
[0003] Currently, the main methods for regenerating graphite from waste lithium-ion batteries include chemical purification, high-temperature heat treatment, surface repair, and combined chemical-thermal treatment regeneration. Chemical purification primarily utilizes acid to remove the solid electrolyte interphase (SEI), metallic impurities, and residual lithium salts from the surface of waste graphite. Heat treatment repairs the graphite crystal structure through high-temperature calcination, improving the layered order and electrochemical performance of the graphite. The combined regeneration process integrates chemical purification and heat treatment technologies to achieve synergistic restoration of the structure and performance of waste graphite. Compared to direct scrapping or low-value utilization, recycled graphite can significantly reduce the consumption of natural graphite resources, which is of great significance for promoting the green and low-carbon development of the lithium-ion battery industry.
[0004] However, the recycling process of waste graphite involves multiple process parameters, such as acid concentration, pickling time, calcination temperature, calcination time, and protective atmosphere. These parameters exhibit complex nonlinear coupling effects, and different combinations of parameters not only affect material properties such as graphite interlayer spacing, graphitization degree, and initial coulombic efficiency, but also influence environmental indicators such as chemical reagent consumption, energy consumption, and greenhouse gas emissions. Therefore, achieving a reasonable configuration of recycling process parameters is a key issue in improving the recycling performance of waste graphite and reducing its environmental impact.
[0005] In recent years, machine learning technology has been increasingly applied to materials research and development and process optimization. By establishing surrogate models between process parameters and material properties, material properties can be quickly predicted using limited experimental data, reducing the number of experiments and improving process design efficiency. Meanwhile, Life Cycle Assessment (LCA) methods can conduct a comprehensive environmental impact assessment of materials production and recycling processes from the perspectives of resource consumption, energy utilization, and environmental emissions, providing quantitative evaluation basis for green process design.
[0006] Current research typically employs machine learning to predict material properties or life cycle assessment to analyze the environmental impact of processes. Some studies also utilize response surface methodology and genetic algorithms to optimize process parameters. However, most of these studies treat material property prediction, environmental impact assessment, and process optimization as independent research areas, failing to establish an integrated technical framework that organically combines machine learning prediction models, life cycle environmental assessment, and reverse optimization design. This makes it difficult to simultaneously optimize material properties and environmental benefits. Furthermore, existing optimization methods mostly target single performance indicators, lacking technical solutions for the synergistic optimization of multiple performance and environmental indicators in the waste graphite recycling process. Consequently, optimization results often require extensive experimental verification, resulting in long development cycles, high costs, and difficulty in obtaining an optimal recycling process that balances high performance and low environmental impact.
[0007] Existing graphite recycling processes and optimization methods for spent lithium-ion batteries have the following main shortcomings:
[0008] (1) Process optimization mainly relies on empirical experiments, resulting in low R&D efficiency. Currently, graphite regeneration process parameters are usually optimized using methods such as single-factor experiments, orthogonal experiments, or response surface analysis, which require a large number of repeated experiments to obtain optimal process conditions. When the number of process variables increases, the number of experiments increases exponentially, which not only leads to a long R&D cycle but also high experimental costs, making it difficult to meet the needs of rapid optimization of complex regeneration processes.
[0009] (2) It is difficult to accurately describe the complex nonlinear relationship between multiple process parameters and graphite properties. In the process of waste graphite regeneration, there is a significant coupling effect among multiple process parameters such as pickling conditions, calcination conditions and protective atmosphere. Different combinations of parameters jointly affect performance indicators such as graphite interlayer spacing, graphitization degree and initial coulombic efficiency. Traditional statistical analysis methods and empirical models are difficult to establish a high-precision mapping relationship between complex process parameters and graphite properties, resulting in limited predictive and generalization capabilities.
[0010] (3) Existing optimization methods lack interpretability and are difficult to reveal the mechanism of action of key process parameters. Most machine learning models only focus on prediction accuracy and lack explanations for the reasons for the prediction results. It is difficult to analyze the contribution of different process parameters to the improvement of graphite performance and cannot provide a theoretical basis for process improvement.
[0011] (4) Material performance optimization and environmental impact assessment are independent of each other, making it difficult to achieve synergistic optimization. Existing studies usually treat life cycle assessment as an independent analysis tool after the process is completed, without directly incorporating environmental impact indicators into the process optimization process. As a result, the optimal process conditions obtained are often only optimized for material performance, which may be accompanied by high energy consumption, chemical consumption and carbon emissions, which is not conducive to green manufacturing and low-carbon development. Summary of the Invention
[0012] The following is an overview of the topics described in detail in this article.
[0013] The purpose of this application is to at least partially solve one of the technical problems existing in the related technologies. The embodiments of this application provide a multi-objective optimization method and apparatus for the regeneration of graphite from waste lithium batteries.
[0014] An embodiment of the first aspect of this application provides a multi-objective optimization method for the regeneration of graphite from waste lithium batteries, comprising:
[0015] A database for the recycling of graphite from waste lithium batteries will be constructed to establish mapping relationships between process parameters, graphite structural parameters, electrochemical performance, and environmental impact indicators.
[0016] Based on the database, a machine learning proxy model is established between the process parameters of waste graphite recycling and the structural properties of graphite.
[0017] Based on the database, a life cycle assessment model is established to quantitatively calculate the environmental impact of the graphite regeneration process of waste lithium-ion batteries within a given system boundary.
[0018] Using the machine learning agent model as a fast performance predictor and the life cycle assessment model as a fast environmental calculator, candidate combinations of process parameters are searched.
[0019] The optimal combination of process parameters is selected from the candidate combinations to obtain the target recommended scheme for the design of graphite recycling process for waste lithium-ion batteries.
[0020] According to certain embodiments of the first aspect of this application, the construction of the waste lithium battery graphite recycling database includes:
[0021] The input data is uniformly organized, which includes one or more of the following: uniformizing process parameter names, uniformizing test indicators, uniformizing units, and standardizing data formats.
[0022] For missing values of continuous variables in the input data, the missing values are estimated by establishing statistical relationships between the variables;
[0023] For categorical variables, a unified coding process is performed based on existing data and descriptions;
[0024] Expand the number of category samples that are below a preset threshold.
[0025] According to certain embodiments of the first aspect of this application, the machine learning proxy model employs extreme random tree regression, with process parameters as input, including one or more of acid concentration, pickling time, calcination temperature, calcination time, and protective atmosphere, and output as performance indicators, including one or more of initial coulombic efficiency, graphite interlayer spacing, XRD diffraction peak position, and Raman defect ratio.
[0026] According to certain embodiments of the first aspect of this application, the machine learning agent model is trained and tested using a 10-fold cross-validation method, and the model's predictive performance is evaluated by evaluation metrics, including one or more of the coefficient of determination, root mean square error, and mean absolute error.
[0027] According to certain embodiments of the first aspect of this application, the Shapley additive interpretation method (SHAP) is used to obtain the contribution degree and direction of different process parameters to the prediction results based on the Shapley values of the input variables, identify the key process parameters affecting the recycling performance of waste graphite, and analyze the coupling relationship between the process parameters to perform global and local interpretation analysis on the machine learning proxy model.
[0028] According to certain embodiments of the first aspect of this application, the predetermined system boundary is the recycling stage of waste lithium batteries; the quantitative calculation includes one or more of the contributions of chemical reagents, power consumption, and gas phase emission proxy.
[0029] The contribution of the chemical reagent is expressed by the formula: ;
[0030] In the formula, Indicates the contribution of chemical reagents. This indicates the mass amount of the chemical component mapped to the solvent category. Greenhouse gas emission factors representing chemical components;
[0031] The contribution to electricity consumption is expressed by the formula: ;
[0032] In the formula, Indicates the contribution to electricity consumption. Indicates total power consumption. Indicates the electricity emission factor;
[0033] The contribution of gas phase emissions as a proxy is expressed by the following formula: ;
[0034] In the formula, Indicates the contribution of gaseous emission proxy. Represents the normalization coefficient. For filtering parameters.
[0035] According to certain embodiments of the first aspect of this application, searching for candidate combinations of process parameters includes:
[0036] Generate a population of initial process parameters that satisfy the initial constraints;
[0037] Candidate process parameters are updated through mutation, crossover, and selection operations, and the comprehensive evaluation function is used to evaluate the candidate process parameters.
[0038] In each iteration, the machine learning proxy model and the life cycle assessment model simultaneously calculate the predictive performance and environmental impact of candidate process parameters, guiding the differential evolution algorithm to gradually search for the optimal region.
[0039] When the maximum number of iterations is reached or the objective function converges, the optimization calculation stops, and candidate process parameter combinations that satisfy the constraints are output.
[0040] According to certain embodiments of the first aspect of this application, the optimal combination of process parameters is selected from the candidate combinations of process parameters to obtain a target recommended scheme for the design of graphite recycling processes for spent lithium-ion batteries, including:
[0041] The machine learning proxy model was invoked to recalculate the corresponding graphite structure performance and electrochemical performance, and the life cycle assessment model was invoked to recalculate the global warming potential and the environmental impacts of chemical reagent consumption, electricity consumption and gas phase emissions.
[0042] All candidate process parameter combinations are sorted according to the comprehensive objective function value, and the candidate process parameter combination with the largest objective function value is output as the target recommended solution. Alternatively, the Pareto optimal solution set is output, and the target recommended solution is selected from multiple perspectives such as performance priority, environmental priority, or comprehensive balance according to actual application requirements.
[0043] According to certain embodiments of the first aspect of this application, after obtaining the target recommended solution, the process parameters and corresponding test results are added to the waste lithium battery graphite recycling database. After data consistency check and preprocessing, the machine learning proxy model is retrained, and the parameter database in the life cycle assessment model is updated synchronously.
[0044] A second aspect of this application provides a multi-objective optimization device for the regeneration of graphite from waste lithium batteries, employing the multi-objective optimization method for the regeneration of graphite from waste lithium batteries as described in the first aspect of this application, including:
[0045] The waste graphite recycling dataset construction module is used to build a waste lithium battery graphite recycling database and to establish the mapping relationship between process parameters, graphite structural parameters, electrochemical performance and environmental impact indicators.
[0046] An interpretable machine learning proxy model building module is used to establish a machine learning proxy model between waste graphite recycling process parameters and graphite structural properties based on the database.
[0047] The life cycle assessment model construction module is used to establish a life cycle assessment model based on the database and to quantify the environmental impact of the graphite recycling process of waste lithium-ion batteries within a given system boundary.
[0048] The inverse multi-objective optimization module is used to search for candidate combinations of process parameters by using the machine learning agent model as a fast performance predictor and the life cycle assessment model as a fast environmental calculator.
[0049] The optimal process output module is used to select the optimal process parameter combination from the candidate process parameter combinations to obtain the target recommended scheme for the design of graphite recycling process for waste lithium-ion batteries.
[0050] The above-mentioned scheme has at least the following beneficial effects: it establishes a waste graphite recycling process design method that integrates data-driven modeling, performance prediction, mechanism explanation, environmental assessment and reverse optimization. On the basis of ensuring the electrochemical and structural performance of recycled graphite, it achieves a synergistic reduction in chemical reagent consumption, energy consumption and carbon emissions, improves the efficiency of waste graphite recycling process design, reduces experimental costs, and provides an intelligent, multi-objective synergistic optimization method for the green and high-value recycling of waste lithium-ion battery anode materials. Attached Figure Description
[0051] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.
[0052] Figure 1 This is a flowchart of a multi-objective optimization method for the regeneration of graphite from waste lithium batteries;
[0053] Figure 2 This is a comparison chart of the total global warming potential between traditional processes and optimized regeneration processes;
[0054] Figure 3 This is a breakdown of the contributions to the total GWP.
[0055] Figure 4 This is a comprehensive comparison chart of carbon footprint and initial coulombic efficiency;
[0056] Figure 5 It is a radar chart comparing normalized sustainability and material performance comprehensive indicators;
[0057] Figure 6This is a comparison chart of the multi-objective optimization method for graphite recycling of waste lithium batteries and the benchmark process (the percentage change in relative process parameters and performance indicators).
[0058] Figure 7 It corresponds to XRD2 Consistency scatter plot of peak positions;
[0059] Figure 8 It corresponds to the interlayer spacing d 002 Consistent scatter plot;
[0060] Figure 9 It corresponds to Raman I D / I G A scatter plot of the consistency of the ratios;
[0061] Figure 10 It is a scatter plot corresponding to the initial coulomb efficiency;
[0062] Figure 11 It corresponds to XRD2 Comparison of kernel density distribution at peak positions;
[0063] Figure 12 It corresponds to the interlayer spacing d 002 Comparison of kernel density distribution;
[0064] Figure 13 It corresponds to Raman I D / I G Comparison of kernel density distribution of ratios;
[0065] Figure 14 This is a comparison chart of the kernel density distribution corresponding to the initial coulomb efficiency;
[0066] Figure 15 It corresponds to XRD2 The violin diagram at the peak;
[0067] Figure 16 It corresponds to the interlayer spacing d 002 A picture of a violin;
[0068] Figure 17 It corresponds to Raman I D / I G violin diagram of the ratio;
[0069] Figure 18 This is a violin plot corresponding to the initial Coulomb efficiency;
[0070] Figure 19 It corresponds to XRD2 Relationship between residuals and predicted values at peak positions;
[0071] Figure 20 It corresponds to the interlayer spacing d 002A graph showing the relationship between residuals and predicted values;
[0072] Figure 21 It corresponds to Raman I D / I G Relationship between residuals and predicted values of the ratio;
[0073] Figure 22 This is a graph showing the relationship between the residual and the predicted value corresponding to the initial coulomb efficiency;
[0074] Figure 23 It is a global feature importance map based on the average absolute SHAP value;
[0075] Figure 24 It corresponds to XRD2 Heatmap of SHAP interaction at peak position;
[0076] Figure 25 It corresponds to the interlayer spacing d 002 SHAP interaction heatmap;
[0077] Figure 26 It corresponds to Raman I D / I G SHAP interaction heatmap of ratios;
[0078] Figure 27 This is a heatmap of the SHAP interaction corresponding to the initial coulomb efficiency. Detailed Implementation
[0079] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0080] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, or the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.
[0081] This application provides a multi-objective optimization method and apparatus for the recycling of graphite from waste lithium-ion batteries. By establishing a waste graphite recycling database, constructing an interpretable machine learning surrogate model, establishing a lifecycle assessment model, and a reverse multi-objective optimization model, intelligent design and green optimization of waste graphite recycling process parameters are achieved. The overall technical process of this invention includes five functional modules: dataset construction, surrogate model establishment, lifecycle assessment model construction, reverse multi-objective optimization, and optimal process output. Data is transferred between these modules through a unified data interface, forming a closed-loop optimization process of "data construction—performance prediction—environmental assessment—reverse optimization—process output".
[0082] The embodiments of this application will be further described below with reference to the accompanying drawings.
[0083] Reference Figure 1 A multi-objective optimization method for the regeneration of graphite from waste lithium batteries includes the following steps:
[0084] Step S100: Construct a database for the recycling of graphite from waste lithium batteries to establish a mapping relationship between process parameters, graphite structural parameters, electrochemical performance and environmental impact indicators.
[0085] Step S200: Based on the database, establish a machine learning proxy model between waste graphite recycling process parameters and graphite structural properties;
[0086] Step S300: Based on the database, establish a life cycle assessment model to quantitatively calculate the environmental impact of the graphite recycling process of waste lithium-ion batteries within the defined system boundary.
[0087] Step S400: Using the machine learning agent model as a fast performance predictor and the life cycle assessment model as a fast environmental calculator, search for candidate process parameter combinations.
[0088] Step S500: Select the optimal combination of process parameters from the candidate combinations of process parameters to obtain the target recommended scheme for the design of graphite recycling process for waste lithium-ion batteries.
[0089] For step S100, constructing a database for the recycling of graphite from waste lithium batteries includes the following steps:
[0090] The input data is uniformly organized, which includes one or more of the following: uniformizing process parameter names, uniformizing test indicators, uniformizing units, and standardizing data formats.
[0091] For missing values of continuous variables in the input data, the missing values are estimated by establishing statistical relationships between the variables;
[0092] For categorical variables, a unified coding process is performed based on existing data and descriptions;
[0093] Expand the number of category samples that are below a preset threshold.
[0094] Specifically, a waste lithium-ion battery graphite recycling database is constructed using a waste graphite recycling dataset construction module. This database is used to establish mapping relationships between process parameters, graphite structural parameters, electrochemical performance, and environmental impact indicators, providing basic data support for subsequent machine learning model training and reverse optimization.
[0095] The raw data in the database comes from publicly published research literature related to graphite regeneration from spent lithium-ion batteries. Due to differences in experimental conditions, testing methods, and data representation across different studies, all data was first standardized, including unifying process parameter names, testing indicators, units, and data formats. For continuous variables such as acid concentration, pickling time, calcination temperature, and calcination time, the International System of Units (SI) was used for representation; for categorical variables such as acid type and protective atmosphere, a unified coding system was established to provide a consistent data format for subsequent machine learning model input.
[0096] Considering the common issues of missing parameters, scattered data sources, and significant differences in sample sizes across different process routes in the literature data, a data preprocessing workflow was further established. For missing values of continuous variables, the Multiple Imputation by Chained Equations (MICE) method was used for estimation. By establishing statistical relationships between variables, iterative predictions were performed on the missing data, thereby improving data integrity. For categorical variables, unified coding was performed based on existing data and literature descriptions. After completing the missing value repair, to avoid bias in the machine learning model due to uneven distribution of samples across different process categories, the Synthetic Minority Oversampling Technique (SMOTE) was used to expand the minority class samples, maintaining a relative balance in the number of samples across categories and improving the model's learning and generalization abilities for different process routes.
[0097] After data cleaning, all process parameters are constructed as input variables, mainly including pickling conditions and calcination conditions; graphite performance indicators are used as output variables, ultimately forming a standardized, multi-source heterogeneous waste graphite recycling database with a unified data structure, providing a unified data foundation for subsequent proxy model training, life cycle assessment, and reverse optimization.
[0098] For step S200, an interpretable machine learning proxy model building module is used to establish a machine learning proxy model between waste graphite recycling process parameters and graphite structural properties based on the database, so as to realize rapid prediction of graphite properties under complex process conditions, replace a large number of repetitive experiments, and improve process optimization efficiency.
[0099] Specifically, the machine learning proxy model adopts extreme random tree regression. Compared with traditional linear models and single decision tree models, the extreme random tree algorithm forms an ensemble learning model by constructing multiple random decision trees, which can effectively describe the complex nonlinear relationship between various process parameters in the waste graphite recycling process. It has better prediction accuracy, anti-overfitting ability and generalization performance.
[0100] The input is process parameters, including one or more of acid concentration, pickling time, calcination temperature, calcination time, and protective atmosphere. The output is performance indicators, including one or more of initial coulombic efficiency, graphite interlayer spacing, XRD diffraction peak position, and Raman defect ratio, to achieve simultaneous prediction of the recycling performance of waste graphite.
[0101] The machine learning agent model is trained and tested using a 10-fold cross-validation method. The predictive performance of the model is evaluated using evaluation metrics, which include one or more of the following: coefficient of determination, root mean square error, and mean absolute error.
[0102] Specifically, to ensure model reliability, a 10-fold cross-validation method was used to train and test the surrogate model, and the model's predictive performance was comprehensively evaluated using metrics such as the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). Cross-validation effectively reduces random errors caused by the limited number of samples and improves the model's predictive ability under unknown process parameters.
[0103] By using the Shapley additive interpretation method (SHAP), the contribution and direction of different process parameters to the prediction results are obtained based on the Shapley values of the input variables. Key process parameters affecting the recycling performance of waste graphite are identified, and the coupling relationship between various process parameters is analyzed. Global and local interpretation analyses are then performed on the machine learning surrogate model.
[0104] Specifically, to further improve the interpretability of the model, the SHAP (SHapley Additive exPlanations) interpretation method is introduced on the basis of the surrogate model to perform global and local interpretation analysis of the machine learning model. By calculating the Shapley values corresponding to each input variable, the contribution and direction of different process parameters to the prediction results are obtained, thereby identifying the key process parameters affecting the recycling performance of waste graphite and analyzing the coupling relationship between the parameters. The SHAP analysis results can not only verify the rationality of the surrogate model's prediction results, but also provide a theoretical basis for subsequent process optimization, realizing the combination of machine learning prediction and mechanism analysis.
[0105] For step S300, the life cycle assessment model construction module establishes a life cycle assessment model based on the database to quantify the environmental impact of the graphite regeneration process of waste lithium-ion batteries within the given system boundary, and uses the calculation results as the environmental target input for subsequent reverse multi-objective optimization to realize the "performance-environment" collaborative design of the regeneration process.
[0106] To ensure comparability among different candidate process conditions, this module uses recycled graphite per unit mass as the functional unit, with a functional unit mass of m = 1 kg. The system boundary is preferably defined as the regeneration stage, including the two core stages of acid washing, chemical pretreatment, and thermal remodeling. Upstream waste graphite generation and downstream battery manufacturing are excluded from the calculation boundary of this module to avoid uncontrollable parameters due to unnecessary extrapolation, thereby ensuring the stability and repeatability of the evaluation function during optimization. This module preferably uses Global Warming Potential (GWP, unit: kg CO2) as the functional unit. 2-eq As a comprehensive environmental indicator, and using a lifecycle accounting approach, the GWP can be updated in real time as process parameters change.
[0107] To achieve computability and embeddability optimization, the quantification calculation includes one or more of the following: chemical reagent contribution, power consumption contribution, and gaseous emission proxy contribution; and is defined as:
[0108] .
[0109] In terms of chemical reagent contribution, regeneration pretreatment often uses acidic solutions or mixed solutions containing oxidizing components. This module denotes the pretreatment solvent as a category variable s, and maps the solvent category s to the chemical component set R(s) through a preset component mapping rule.
[0110] For any component r∈R(s), its mass quantity is: ;in The molecular weight of component r (g·mol) −1 The coefficient 1000 is used for unit conversion.
[0111] When the solvent contains multiple components, the contribution of the chemical reagents is obtained by summing over all components. This contribution is expressed by the formula: ;
[0112] In the formula, Indicates the contribution of chemical reagents. This indicates the mass amount of the chemical component mapped to the solvent category. Greenhouse gas emission factors representing chemical components.
[0113] In a preferred embodiment, emission factors are taken from a unified database or a list of enterprises and are set as constants by the system or configured by the user, thereby ensuring that different candidate schemes are compared under the same evaluation benchmark.
[0114] Regarding the contribution to electricity consumption, electricity consumption is divided into two parts: pickling electricity consumption and roasting electricity consumption. The electricity consumption during the pickling stage is approximated using an engineering approximation of equipment power versus time, assuming the power of the pickling reactor is... (kW), pickling time is (h) has: ;
[0115] The power consumption during the roasting stage considers the combined effect of sensible heat from temperature rise and heat preservation during the holding period. Let the roasting temperature be... (°C), roasting time is (h), furnace power is (kW), furnace efficiency is η, and equivalent specific heat is Cp (kJ·kg). −1 ·K −1 Then, the calcination power consumption is expressed as: ;
[0116] Total power consumption is expressed as: ;
[0117] The contribution to electricity consumption is expressed by the formula: ;
[0118] In the formula, Indicates the contribution to electricity consumption. Indicates total power consumption. The electricity emission factor (kgCO) 2-eq ·kWh −1 ).
[0119] Regarding the contribution of gas-phase emissions as a proxy, since the actual gas-phase emissions during the regeneration process are related to raw material differences, furnace sealing, and side reactions, direct and accurate modeling usually requires online monitoring, which is not conducive to real-time invocation during optimization cycles. Therefore, this invention provides a computable proxy term coupled with electrochemical performance for relative ranking of different candidate process conditions: the initial coulombic efficiency (ICE, %) is considered as a comprehensive characterization reflecting irreversible side reactions and surface reactivity, and a normalization coefficient is defined as follows: ;
[0120] The contribution of gas phase emissions as a proxy is expressed by the following formula: ;
[0121] In the formula, Indicates the contribution of gaseous emission proxy. Represents the normalization coefficient. In a preferred embodiment, to filter parameters, 0.85 is acceptable; is the stoichiometric conversion factor from C to CO2.
[0122] Additionally, this module allows the proxy item to be disabled when needed, allowing... =0, to adapt to different enterprise boundaries or data availability.
[0123] Based on the above formulas, this module can output any candidate process conditions. This allows us to identify whether carbon emissions primarily originate from chemicals or electricity consumption during subsequent optimization, and provides directional guidance for process improvement.
[0124] To support comparative evaluation, a reference operating condition GWP can be defined. For example, the arithmetic mean of the GWP under the same parameter system based on historical data or literature conditions can be used as a benchmark value to measure the relative emission reduction of the optimized scheme. At the same time, scenario perturbations can be applied to key parameters to verify the stability of the optimization conclusions.
[0125] The main parameters of the life cycle assessment model are shown in Table 1.
[0126] Table 1. Parameter Table of Life Cycle Assessment Model
[0127]
[0128] For step S400, the inverse multi-objective optimization module uses the machine learning agent model as a fast performance predictor and the life cycle assessment model as a fast environmental calculator to search for candidate process parameter combinations.
[0129] Searching for candidate combinations of process parameters includes the following steps:
[0130] Generate a population of initial process parameters that satisfy the initial constraints;
[0131] Candidate process parameters are updated through mutation, crossover, and selection operations, and the comprehensive evaluation function is used to evaluate the candidate process parameters.
[0132] In each iteration, the machine learning proxy model and the life cycle assessment model simultaneously calculate the predictive performance and environmental impact of candidate process parameters, guiding the differential evolution algorithm to gradually search for the optimal region.
[0133] When the maximum number of iterations is reached or the objective function converges, the optimization calculation stops, and candidate process parameter combinations that satisfy the constraints are output.
[0134] Specifically, the inverse multi-objective optimization module automatically searches for combinations of process parameters within a given process feasible domain that can simultaneously improve the electrochemical performance of recycled graphite and reduce environmental impact.
[0135] The Differential Evolution (DE) algorithm is preferred for global optimization search. First, an initial population of process parameters satisfying the constraints is randomly generated. Candidate solutions are continuously updated through mutation, crossover, and selection operations, and a comprehensive evaluation function is used to evaluate each candidate process. During each iteration, the surrogate model and the life cycle assessment model simultaneously calculate the predicted performance and environmental impact of each candidate process, guiding the DE algorithm towards the optimal region. When the maximum number of iterations is reached or the objective function converges, the optimization calculation stops, and the optimal combination of process parameters satisfying the constraints is output. Material property predictions and GWP calculation results are obtained simultaneously during each candidate solution evaluation, thus forming an iterative closed-loop optimization. This approach avoids traditional trial-and-error experimental optimization, significantly reducing the number of experiments and improving process design efficiency.
[0136] For step S500, the optimal process parameter combination is selected from the candidate process parameter combinations through the optimal process output module to obtain the target recommended scheme for the graphite recycling process design of waste lithium-ion batteries.
[0137] Specifically, for each candidate optimal solution obtained by the optimization algorithm, the corresponding graphite structural performance and electrochemical performance are recalculated by calling the machine learning proxy model. At the same time, the life cycle assessment model is called to recalculate its global warming potential (GWP) and environmental impacts such as chemical reagent consumption, electricity consumption and gaseous emissions, so as to ensure that the final output results are consistent with the optimization evaluation function.
[0138] For the implementation method that uses a weighted comprehensive objective function for optimization, all candidate solutions are sorted according to the comprehensive objective function value, and the process parameter combination with the largest objective function value is output as the recommended regeneration process. For the implementation method that uses the Pareto multi-objective optimization strategy, the Pareto optimal solution set is output, and the final recommended solution is selected from different perspectives such as performance priority, environmental priority, or comprehensive balance according to the actual application requirements.
[0139] The optimal process output module generates a mapping diagram between process parameters and performance indicators, an environmental contribution analysis diagram, and a Pareto front curve. These are used to assist researchers in analyzing the trade-off between performance and environmental impact under different process conditions, providing a visual basis for decision-making in actual process design.
[0140] Establish an optimization result feedback mechanism. When new experimental data is obtained, process parameters and corresponding test results are added to the waste lithium battery graphite recycling database. After data consistency checks and preprocessing, the machine learning surrogate model is retrained, and the parameter database in the life cycle assessment model is updated synchronously. This allows the surrogate model to continuously learn from new data and improve prediction accuracy and optimization reliability.
[0141] Through the above feedback and update mechanism, a closed-loop optimization process of "data collection - database update - proxy model training - life cycle evaluation - reverse optimization - experimental verification - data feedback" is formed. This makes the solution not only applicable to single process optimization, but also able to continuously improve model performance as data accumulates, providing continuous optimization capabilities for waste graphite from different sources and different recycling process routes.
[0142] This application provides a multi-objective optimization device for the regeneration of graphite from waste lithium batteries.
[0143] A multi-objective optimization device for the regeneration of graphite from waste lithium batteries includes:
[0144] The waste graphite recycling dataset construction module is used to build a waste lithium battery graphite recycling database and to establish the mapping relationship between process parameters, graphite structural parameters, electrochemical performance and environmental impact indicators.
[0145] An interpretable machine learning proxy model building module is used to establish a machine learning proxy model between waste graphite recycling process parameters and graphite structural properties based on the database.
[0146] The life cycle assessment model construction module is used to establish a life cycle assessment model based on the database and to quantify the environmental impact of the graphite recycling process of waste lithium-ion batteries within a given system boundary.
[0147] The inverse multi-objective optimization module is used to search for candidate combinations of process parameters by using the machine learning agent model as a fast performance predictor and the life cycle assessment model as a fast environmental calculator.
[0148] The optimal process output module is used to select the optimal process parameter combination from the candidate process parameter combinations to obtain the target recommended scheme for the design of graphite recycling process for waste lithium-ion batteries.
[0149] It should be noted that the multi-objective optimization device for the recycling of graphite from waste lithium batteries uses the multi-objective optimization method for the recycling of graphite from waste lithium batteries as described above. Each module of the multi-objective optimization device for the recycling of graphite from waste lithium batteries corresponds to each step of the multi-objective optimization method for the recycling of graphite from waste lithium batteries. Both adopt the same technical solution, solve the same technical problems, and bring about the same technical effects.
[0150] In this embodiment, an interpretable machine learning proxy model was constructed and combined with lifecycle assessment to perform reverse multi-objective optimization of the graphite recycling process for spent lithium batteries. The optimal recycling process conditions were finally determined as follows:
[0151] Pickling and purification stage: Hydrochloric acid (HCl) was used as the washing solvent, with an acid concentration of 2.31 mol / L, and the washing time was 1.14 h at 27.9℃.
[0152] Calcination and reconstruction stage: under an argon (Ar) protective atmosphere, heat-treated and calcined at 577℃ for 1.01h.
[0153] Under the aforementioned mild process conditions, this embodiment achieved the following results: the regenerated graphite material achieved near-complete recovery of its crystal structure and electrochemical performance, with an initial coulombic efficiency (ICE) as high as 98.18% and a layer spacing d 002 Reduced to 0.337 nm (close to the standard of natural commercial graphite), Raman defect ratio I D / I G The value is 0.68. Simultaneously, carbon emissions throughout the entire regeneration process are significantly reduced, and the global warming potential within the system boundary is reduced from the average level of 12.26 kg CO₂ of traditional processes. 2-eq Significantly reduced to 5.60 kg CO 2-eq Greenhouse gas emissions were reduced by as much as 54.4%, achieving a win-win situation of high performance and low carbon emissions.
[0154] Reference Figures 2 to 6 , Figures 2 to 6 A performance-sustainability comparison chart of the baseline process and the optimized graphite recycling process; specifically, Figure 2 It is a comparison chart of the total global warming potential of traditional processes and optimized regeneration processes, including three parts: chemical (acid) related emissions, energy consumption related emissions, and gaseous phase emissions; Figure 3 This is a breakdown of the contributions to the total GWP. Figure 4 This is a comprehensive comparison chart of carbon footprint and initial coulombic efficiency, where the size of the data points represents total energy consumption, and the shaded area is the ideal target area for low carbon emissions and high ICE. Figure 5 This is a radar chart comparing normalized sustainability and material performance indicators, covering low carbon footprint, low energy consumption, low chemical consumption, high initial coulombic efficiency, and low Raman defect ratio (I0). D / I G Five dimensions; Figure 6 This is a comparison chart of the multi-objective optimization method for graphite regeneration of waste lithium batteries and the baseline process, showing the relative percentage changes in process parameters and performance indicators, including the improvement in total carbon footprint, energy consumption, chemical consumption, and initial coulombic efficiency.
[0155] Traditional processes often improve graphite electrochemical performance by blindly increasing calcination temperature and extending heat treatment time, leading to high energy consumption and carbon emissions. This embodiment demonstrates that complete graphite reconstruction can be achieved without stringent high-temperature conditions by finely adjusting the coupling relationship between chemical pretreatment and mild heat treatment. At a lower calcination temperature, the performance of recycled graphite approaches commercial benchmarks, while greenhouse gas emissions are significantly reduced by 54.4% over its entire life cycle, achieving low-carbon, high-value recycling and synergistically improving material performance and environmental benefits.
[0156] The waste graphite processed by the process in this embodiment achieves an initial coulombic efficiency of 98.18%, effectively reducing irreversible lithium loss in the first cycle; the crystal order is highly restored, and the interlayer spacing d 002 Shrinking to 0.337 nm, XRD2 The diffraction peak was precisely restored to 26.43°, indicating that the graphite flakes were arranged in an orderly layered manner, restoring a highly ordered graphitized structure; surface defects were significantly repaired, and the Raman defect ratio was lower than that of I. D / I G The optimization to 0.68 indicates that disorder and surface micro-defects have been effectively controlled.
[0157] Reference Figures 7 to 22 , Figures 7 to 22 This demonstrates the predictive performance and distribution characteristics of the Extra Trees model; specifically, Figure 7 It corresponds to XRD2 Consistency scatter plot of peak positions; Figure 8 It corresponds to the interlayer spacing d 002 Consistent scatter plot; Figure 9 It corresponds to Raman I D / I G A scatter plot of the consistency of the ratios; Figure 10 It is a scatter plot corresponding to the initial coulomb efficiency; Figure 11 It corresponds to XRD2 Comparison of kernel density distribution at peak positions; Figure 12 It corresponds to the interlayer spacing d 002 Comparison of kernel density distribution; Figure 13 It corresponds to Raman I D / I G Comparison of kernel density distribution of ratios; Figure 14 This is a comparison chart of the kernel density distribution corresponding to the initial coulomb efficiency; Figure 15 It corresponds to XRD2 The violin diagram at the peak; Figure 16 It corresponds to the interlayer spacing d 002 A picture of a violin; Figure 17 It corresponds to Raman I D / I G violin diagram of the ratio; Figure 18 This is a violin plot corresponding to the initial Coulomb efficiency; Figure 19 It corresponds to XRD2 Relationship between residuals and predicted values at peak positions; Figure 20 It corresponds to the interlayer spacing d 002 A graph showing the relationship between residuals and predicted values; Figure 21 It corresponds to Raman I D / I G Relationship between residuals and predicted values of the ratio; Figure 22 This is a graph showing the relationship between the residual and the predicted value corresponding to the initial coulombic efficiency. The graph illustrates that this model can faithfully reproduce the complex nonlinear relationship between process parameters and material properties, quickly identifying the globally optimal process range without the need for expensive and inefficient traditional single-factor or orthogonal trial-and-error experiments.
[0158] Various tree models were employed, among which the surrogate model built using Extreme Random Tree Regression (Extra Trees) demonstrated superior predictive reliability for crystal structure descriptors (interlayer spacing d). 002 and XRD2 The predictive coefficient of determination R for peak position 2 The predictive coefficient of determination R for complex electrochemical performance parameters (ICE) reached 0.993; 2 The accuracy reached 0.900. The machine learning surrogate model has high prediction accuracy.
[0159] Reference Figures 23 to 27 23 to Figure 27 This demonstrates a SHAP-based analysis of parameter importance and interactions; specifically, Figure 23 It is a global feature importance map based on the average absolute SHAP value; Figure 24 It corresponds to XRD2 Heatmap of SHAP interaction at peak position; Figure 25 It corresponds to the interlayer spacing d002 SHAP interaction heatmap; Figure 26 It corresponds to Raman I D / I G SHAP interaction heatmap of ratios; Figure 27 This is a heatmap of the SHAP interaction corresponding to the initial coulomb efficiency. The intensity of the color indicates the strength of the interaction effect between parameter pairs.
[0160] By introducing interpretability analysis, it was clearly revealed that: chemical pretreatment determines the starting point and difficulty of the subsequent thermal reconstruction stage; while the recovery of electrochemical coulombic efficiency depends on the strong coupling effect of the "chemical-thermal" two-stage parameters.
[0161] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.
Claims
1. A multi-objective optimization method for the regeneration of graphite from waste lithium batteries, characterized in that, include: A database for the recycling of graphite from waste lithium batteries will be constructed to establish mapping relationships between process parameters, graphite structural parameters, electrochemical performance, and environmental impact indicators. Based on the database, a machine learning proxy model is established between the process parameters of waste graphite recycling and the structural properties of graphite. Based on the database, a life cycle assessment model is established to quantitatively calculate the environmental impact of the graphite regeneration process of waste lithium-ion batteries within a given system boundary. Using the machine learning agent model as a fast performance predictor and the life cycle assessment model as a fast environmental calculator, candidate combinations of process parameters are searched. The optimal combination of process parameters is selected from the candidate combinations to obtain the target recommended scheme for the design of graphite recycling process for waste lithium-ion batteries.
2. The multi-objective optimization method for regenerating graphite from waste lithium batteries according to claim 1, characterized in that, The construction of the waste lithium battery graphite recycling database includes: The input data is uniformly organized, which includes one or more of the following: uniformizing process parameter names, uniformizing test indicators, uniformizing units, and standardizing data formats. For missing values of continuous variables in the input data, the missing values are estimated by establishing statistical relationships between the variables; For categorical variables, a unified coding process is performed based on existing data and descriptions; Expand the number of category samples that are below a preset threshold.
3. The multi-objective optimization method for regenerating graphite from waste lithium batteries according to claim 1, characterized in that, The machine learning proxy model employs extreme random tree regression. The input is process parameters, including one or more of acid concentration, pickling time, calcination temperature, calcination time, and protective atmosphere. The output is performance indicators, including one or more of initial coulombic efficiency, graphite interlayer spacing, XRD diffraction peak position, and Raman defect ratio.
4. The multi-objective optimization method for regenerating graphite from waste lithium batteries according to claim 1, characterized in that, The machine learning agent model is trained and tested using a 10-fold cross-validation method. The predictive performance of the model is evaluated using evaluation metrics, which include one or more of the following: coefficient of determination, root mean square error, and mean absolute error.
5. The multi-objective optimization method for regenerating graphite from waste lithium batteries according to claim 1, characterized in that, By using the Shapley additive interpretation method (SHAP), the contribution and direction of different process parameters to the prediction results are obtained based on the Shapley values of the input variables. Key process parameters affecting the recycling performance of waste graphite are identified, and the coupling relationship between various process parameters is analyzed. Global and local interpretation analyses are then performed on the machine learning surrogate model.
6. The multi-objective optimization method for regenerating graphite from waste lithium batteries according to claim 1, characterized in that, The defined system boundary is the recycling stage of waste lithium batteries; the quantitative calculation includes one or more of the contributions of chemical reagents, power consumption, and gas phase emission proxy. The contribution of the chemical reagent is expressed by the formula: ; In the formula, Indicates the contribution of chemical reagents. This indicates the mass amount of the chemical component mapped to the solvent category. Greenhouse gas emission factors representing chemical components; The contribution to electricity consumption is expressed by the formula: ; In the formula, Indicates the contribution to electricity consumption. Indicates total power consumption. Indicates the electricity emission factor; The contribution of gas phase emissions as a proxy is expressed by the following formula: ; In the formula, Indicates the contribution of gaseous emission proxy. Represents the normalization coefficient. For filtering parameters.
7. The multi-objective optimization method for regenerating graphite from waste lithium batteries according to claim 1, characterized in that, Search for candidate process parameter combinations, including: Generate a population of initial process parameters that satisfy the initial constraints; Candidate process parameters are updated through mutation, crossover, and selection operations, and the comprehensive evaluation function is used to evaluate the candidate process parameters. In each iteration, the machine learning proxy model and the life cycle assessment model simultaneously calculate the predictive performance and environmental impact of candidate process parameters, guiding the differential evolution algorithm to gradually search for the optimal region. When the maximum number of iterations is reached or the objective function converges, the optimization calculation stops, and candidate process parameter combinations that satisfy the constraints are output.
8. The multi-objective optimization method for regenerating graphite from waste lithium batteries according to claim 1, characterized in that, The optimal combination of process parameters is selected from the candidate combinations to obtain the target recommended scheme for the design of graphite recycling processes for spent lithium-ion batteries, including: The machine learning proxy model was invoked to recalculate the corresponding graphite structure performance and electrochemical performance, and the life cycle assessment model was invoked to recalculate the global warming potential and the environmental impacts of chemical reagent consumption, electricity consumption and gas phase emissions. All candidate process parameter combinations are sorted according to the comprehensive objective function value, and the candidate process parameter combination with the largest objective function value is output as the target recommended solution. Alternatively, the Pareto optimal solution set is output, and the target recommended solution is selected from multiple perspectives such as performance priority, environmental priority, or comprehensive balance according to actual application requirements.
9. The multi-objective optimization method for regenerating graphite from waste lithium batteries according to claim 1, characterized in that, After obtaining the target recommended solution, the process parameters and corresponding test results are added to the waste lithium battery graphite recycling database. After data consistency check and preprocessing, the machine learning agent model is retrained, and the parameter database in the life cycle assessment model is updated synchronously.
10. A multi-objective optimization device for the regeneration of graphite from waste lithium batteries, characterized in that, The multi-objective optimization method for regenerating graphite from waste lithium batteries as described in any one of claims 1 to 9 includes: The waste graphite recycling dataset construction module is used to build a waste lithium battery graphite recycling database and to establish the mapping relationship between process parameters, graphite structural parameters, electrochemical performance and environmental impact indicators. An interpretable machine learning proxy model building module is used to establish a machine learning proxy model between waste graphite recycling process parameters and graphite structural properties based on the database. The life cycle assessment model construction module is used to establish a life cycle assessment model based on the database and to quantify the environmental impact of the graphite recycling process of waste lithium-ion batteries within a given system boundary. The inverse multi-objective optimization module is used to search for candidate combinations of process parameters by using the machine learning agent model as a fast performance predictor and the life cycle assessment model as a fast environmental calculator. The optimal process output module is used to select the optimal process parameter combination from the candidate process parameter combinations to obtain the target recommended scheme for the design of graphite recycling process for waste lithium-ion batteries.
