Steel slag carbon sequestration potential and economic benefit evaluation system based on machine learning

Abstract

The invention relates to the field of chemical detection and machine learning, and discloses a steel slag carbon sequestration potential and economic benefit evaluation system based on machine learning, which comprises a multi-dimensional characteristic data construction subsystem, a carbon sequestration capability prediction model construction subsystem, an energy consumption and carbon sequestration double-target game optimization subsystem and an economic feasibility evaluation subsystem. The multi-dimensional feature data construction subsystem is used for generating standardized feature data containing chemical, physical and technological parameters; the carbon sequestration capability prediction model construction subsystem establishes a carbon sequestration quantity nonlinear mapping relation by using a machine learning model; the energy consumption and carbon sequestration dual-objective game optimization subsystem outputs a Pareto optimal solution set considering the maximum carbon sequestration amount and the minimum energy consumption cost; and the economic feasibility evaluation subsystem is combined with a carbon transaction price parameter to execute cost income verification on the optimal solution set and screen a recommendation scheme with the maximum net income, so that rapid prediction and technical economic collaborative optimization are realized, and decision support is provided for the steel slag carbon sequestration project.

Description

Technical Field

[0001] This invention relates to the fields of chemical detection and machine learning, specifically to a machine learning-based system for evaluating the carbon sequestration potential and economic benefits of steel slag. Background Technology

[0002] Steel slag contains components such as calcium oxide and magnesium oxide, and carbon dioxide can be fixed through mineral carbonization. The mineral carbonization of steel slag is affected by a variety of physicochemical factors. The chemical composition and physical properties of steel slag from different sources vary, and the carbonization efficiency is related to process parameters such as reaction temperature, pressure, and stirring rate.

[0003] Current methods for assessing the carbon sequestration potential of steel slag primarily employ physical experiments, testing specific steel slag or process conditions. These physical experiments are time-consuming and reagent-intensive, resulting in limited data sample sizes and difficulty in covering multidimensional parameter spaces. Furthermore, when dealing with steel slag raw materials exhibiting fluctuating composition, these experimental methods struggle to analyze the nonlinear influence of raw material characteristics and process conditions on carbon sequestration effectiveness, thus failing to achieve rapid quantitative prediction of carbon sequestration capacity.

[0004] In terms of process optimization, existing studies often aim to increase carbon sequestration rates without fully considering the energy costs required to maintain high temperature, high pressure, and stirring conditions. Solely pursuing carbon sequestration targets leads to increased operating costs for process solutions, lacking a balance between technological output and energy input. Furthermore, technical indicator studies and economic feasibility analyses are usually conducted independently, failing to incorporate carbon emission trading prices, social carbon cost avoidance values, and the overall life-cycle input-output ratio into comprehensive considerations early in the project. This results in process solutions lacking commercial profitability support, increasing the uncertainty of engineering decisions. Summary of the Invention

[0005] This invention provides a machine learning-based system for evaluating the carbon sequestration potential and economic benefits of steel slag, aiming to solve the technical problems of high experimental dependence, lack of multi-objective collaborative optimization, and disconnect between technical and economic evaluation in the current evaluation process of steel slag carbon sequestration projects.

[0006] The machine learning-based steel slag carbon sequestration potential and economic benefit assessment system provided by this invention mainly includes a multi-dimensional feature data construction subsystem, a carbon sequestration capacity prediction model construction subsystem, an energy consumption and carbon sequestration dual-objective game optimization subsystem, and an economic feasibility assessment subsystem.

[0007] In terms of constructing multidimensional feature data, the system uses a multidimensional feature data construction subsystem to collect and structure unstructured data in the field of steel slag mineral carbonization. This subsystem analyzes the oxide composition, physical morphology data, and operating conditions during the carbonization reaction of steel slag raw materials through chemical composition feature extraction, physical property feature extraction, and process parameter feature extraction units. Specifically, the system extracts the mass percentage of various oxides, including calcium oxide and magnesium oxide, to construct chemical composition feature data; extracts the average particle size to construct physical property feature data; and extracts key operating conditions such as reaction temperature, carbon dioxide partial pressure, and fluid dynamics to construct process parameter feature data. To eliminate dimensional differences in heterogeneous data, the system uses a data preprocessing unit to perform mean imputation on missing data and performs standardization transformation on the feature data to make it conform to a standard normal distribution, thus providing a high-quality data foundation for the stable training of subsequent models.

[0008] In terms of carbon sequestration capacity prediction, the system utilizes a carbon sequestration capacity prediction model to construct a subsystem that establishes a nonlinear mapping relationship between the characteristic properties of steel slag and its carbon sequestration effect. This subsystem employs an ensemble learning strategy, utilizing a machine learning model to construct a flexible nonlinear additive model. Through iterative training in a multidimensional feature space, the model can capture the influence of complex nonlinear interactions between chemical composition, physical properties, and process conditions on the amount of carbon sequestration. The system introduces cross-validation logic, dividing the training sample set into non-overlapping subsets for iterative training and validation, evaluating and optimizing the model's hyperparameter combination, ensuring that the prediction model has good generalization performance under unknown process conditions, and achieving accurate prediction of the carbon sequestration amount per unit mass of steel slag.

[0009] In terms of process parameter optimization, the system addresses the conflict between carbon sequestration technical indicators and operating energy costs through a dual-objective game optimization subsystem of energy consumption and carbon sequestration. This subsystem constructs two mutually interdependent optimization objectives: on the one hand, it calls a predictive model to calculate and maximize the predicted carbon sequestration per unit of steel slag; on the other hand, based on reaction kinetics principles and equipment power consumption characteristics, it identifies key factors driving energy consumption, calculates variable energy costs, and minimizes them. The system employs a non-dominated sorting genetic algorithm with an elitist strategy to perform multi-objective optimization within the physically feasible region. Through fast non-dominated sorting and crowding distance calculation, it generates a Pareto optimal solution set containing multiple non-dominated process parameter combinations. This mechanism ensures that the selected process scheme increases carbon sequestration without causing a non-linear surge in energy costs.

[0010] In terms of economic feasibility assessment, the system utilizes an economic feasibility assessment subsystem to transform the optimization results at the technical level into economic decision-making basis. The system constructs a cost model based on the investment and operating costs of steel slag carbon sequestration equipment and conducts a comprehensive analysis in conjunction with carbon emission reduction benefit parameters. The system integrates cost accounting and benefit accounting logic, quantifying the plant's fixed costs, variable energy consumption costs, and the comprehensive economic benefit comprised of carbon emission trading market prices and social carbon cost avoidance values. The system performs rigorous feasibility logic verification, determining the profit-loss relationship between the comprehensive economic value of a unit of carbon sequestration and the unit treatment cost, and selecting economically feasible solutions. The system further calculates the net benefit value, selecting the process parameter combination with the highest net benefit from the Pareto optimal solution set as the recommended solution. The assessment results are then presented intuitively in the form of structured reports and Pareto front scatter plots through a decision report generation module and a visualization unit.

[0011] This invention provides a machine learning-based system for evaluating the carbon sequestration potential and economic benefits of steel slag. It offers the following advantages: 1. This invention constructs standardized feature data including chemical composition, physical properties, and process parameters, and utilizes machine learning models to establish nonlinear mapping relationships, enabling precise capture of the complex interactive effects of multidimensional variables on carbon fixation efficiency. This data-driven prediction method effectively replaces traditional evaluation methods that rely on numerous orthogonal experiments, achieving rapid quantitative prediction of the carbon fixation capacity of steel slag from different sources without consuming physical reagents or incurring time costs.

[0012] 2. This invention constructs a dual-objective game-theoretic optimization model that maximizes carbon sequestration and minimizes energy consumption costs, and employs a non-dominated sorting genetic algorithm with an elitist strategy to find the optimal solution within the physically feasible region. This mechanism effectively resolves the engineering contradiction that solely pursuing carbon sequestration technical indicators often leads to a non-linear surge in energy consumption costs. By outputting the Pareto optimal solution set, the optimal balance combination of process parameters such as reaction temperature, pressure, and time is determined, ensuring that the process scheme improves output while also considering energy utilization efficiency.

[0013] 3. This invention introduces carbon emission trading prices and social carbon cost parameters to establish a full life-cycle economic feasibility assessment logic covering fixed investment amortization and variable operating costs. The system performs rigorous cost-benefit comparison verification on the Pareto optimal solution set to select the process parameter combination with the highest net benefit. This evaluation method, which deeply couples technological potential prediction with economic benefit accounting, directly outputs recommended solutions with commercial profitability, providing a quantitative basis for the industrialization and investment decisions of steel slag carbon sequestration projects. Attached Figure Description

[0014] Figure 1 This is a diagram showing the overall logical architecture of the machine learning-based steel slag carbon sequestration potential and economic benefit assessment system of the present invention. Figure 2 This is a flowchart of the dual-objective game optimization algorithm for energy consumption and carbon sequestration of the present invention; Figure 3 This is a logic diagram for determining the economic feasibility of the entire life cycle of this invention; Figure 4 This is a schematic diagram of the interface display logic of the visualization unit of the present invention; Figure 5 This is a distribution diagram of the ten-fold cross-validation of the present invention. Detailed Implementation

[0015] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Please see the appendix Figure 1 This invention provides a machine learning-based system for evaluating the carbon sequestration potential and economic benefits of steel slag, including a multi-dimensional feature data construction subsystem, a carbon sequestration capacity prediction model construction subsystem, a dual-objective game optimization subsystem for energy consumption and carbon sequestration, and an economic feasibility assessment subsystem.

[0017] Multidimensional feature data construction subsystem: configured to collect and process literature data and experimental records in the field of steel slag mineral carbonization, and generate standardized feature data.

[0018] Carbon sequestration capacity prediction model construction subsystem: connected to the multidimensional feature data construction subsystem, configured to establish a nonlinear mapping relationship between steel slag features and carbon sequestration amount based on standardized feature data and using extreme gradient boosting tree (XGBoost).

[0019] Energy consumption and carbon sequestration dual-objective game optimization subsystem: connected to the carbon sequestration capacity prediction model construction subsystem, configured to execute a multi-objective genetic algorithm (NSGA-II) to find the Pareto optimal solution set between maximizing carbon sequestration and minimizing energy consumption cost.

[0020] Economic Feasibility Assessment Subsystem: Connected to the dual-objective game optimization subsystem of energy consumption and carbon sequestration, it is configured to incorporate carbon trading prices and social carbon costs, perform economic assessment on the Pareto optimal solution set, and output the final recommended solution.

[0021] The multidimensional feature data construction subsystem serves as the basic data input for the entire evaluation system. It is responsible for transforming unstructured literature records or discrete experimental records into structured feature data to meet the input requirements of subsequent machine learning models.

[0022] The multidimensional feature data construction subsystem integrates a chemical composition feature extraction unit. This unit is equipped with a data parsing interface for the chemical composition of steel slag raw materials. The unit identifies and extracts the percentage of calcium oxide by mass from the input raw data source. magnesium oxide mass percentage Silica by mass percentage alumina mass percentage Iron oxide mass percentage Manganese oxide mass percentage Titanium dioxide mass percentage and the mass percentage of phosphorus pentoxide The chemical composition feature extraction unit classifies the above eight oxide components into chemical composition feature data. This chemical composition feature data not only digitizes the material composition of steel slag, but also implicitly contains the basicity information and potential mineral phase composition information of steel slag, providing a feature expression of the reaction material basis for the prediction model.

[0023] The multidimensional feature data construction subsystem integrates a physical property feature extraction unit. This unit is configured to process the physical morphological description data of the steel slag raw material. It extracts the average size of the steel slag particles, which quantifies the fineness of the steel slag powder and directly correlates with the specific surface area during the reaction process. The physical property feature extraction unit constructs physical property feature data from the average size of the steel slag particles. This data provides the prediction model with physical constraints regarding the contact characteristics of the reaction interface.

[0024] The multidimensional feature data construction subsystem integrates a process parameter feature extraction unit. This unit is configured to standardize the operating conditions data during the carbonization reaction process. Specifically, it extracts the reaction temperature, which characterizes the thermodynamic state of the reaction system; the partial pressure of carbon dioxide, which characterizes the driving force of the gas-liquid mass transfer process; the carbon dioxide gas flow rate and concentration, used to quantify the carbon source supply rate; the carbonization time, used to define the kinetic duration of the reaction; the stirring rate, used to quantify the fluid dynamics and mixing efficiency within the reactor; the liquid-to-solid ratio, used to define the proportional relationship between the mass of the liquid phase and the mass of the solid phase in the reaction system; and the sample processing pressure, used to characterize the pre-treatment compaction state of the raw materials before the reaction. This unit aggregates all the above operating parameters into process parameter feature data. This process parameter feature data provides boundary constraints for the prediction model based on external reaction conditions.

[0025] The data preprocessing unit is configured to clean and reconstruct the raw feature data output by the chemical composition feature extraction unit, physical property feature extraction unit, and process parameter feature extraction unit, so as to eliminate the impact of data noise and dimensional differences on the stability of subsequent machine learning model training.

[0026] The data preprocessing unit integrates missing value identification and imputation logic. In actual collected literature data or experimental records, some samples may lack non-critical process parameters or physical property data. The data preprocessing unit is configured to traverse and scan chemical composition feature data, physical property feature data, and process parameter feature data to locate outlier data points that are empty, marked as non-numeric symbols, or exceed physical boundaries. For the located missing data, the data preprocessing unit performs column mean imputation. The data preprocessing unit calculates the arithmetic mean of all valid values ​​in the corresponding feature dimension in the training sample set and assigns the arithmetic mean to the missing data points in that feature dimension. By using mean imputation, the data preprocessing unit can retain valid information in other dimensions of incomplete samples while maintaining the expected overall statistical distribution of the features, avoiding information loss in small sample datasets caused by directly removing incomplete samples.

[0027] The data preprocessing unit integrates Z-score normalization calculation logic. Due to the significant differences in dimensions and orders of magnitude between percentage values ​​in chemical composition feature data, micrometer-level size values ​​in physical property feature data, and megapascal-level pressure values ​​in process parameter feature data, directly inputting heterogeneous data can cause oscillations or convergence lag in the model's loss function during gradient descent. The data preprocessing unit is configured to perform normalization transformation operations independently for each feature dimension. For any feature dimension… The data preprocessing unit first calculates the mean of this feature dimension across the entire training sample set. and standard deviation .

[0028] Subsequently, the data preprocessing unit uses the standardized transformation formula: ; in, This represents the feature value after standardization. Represents the original feature values; This represents the arithmetic mean of the feature dimension in the training set; This represents the standard deviation of the feature dimension in the training set.

[0029] Original eigenvalues Mapped to standardized eigenvalues After Z-score standardization, the feature data output by the data preprocessing unit follows a standard normal distribution with a mean of 0 and a standard deviation of 1 in the numerical space. This distribution transformation ensures that all input features contribute to the prediction model weights on the same order of magnitude, eliminates the dominant influence of large numerical features on model training, and improves the efficiency of the Extreme Gradient Boosting Tree (XGBoost) algorithm in finding the optimal split point in the feature space.

[0030] The carbon fixation capacity prediction model construction subsystem is configured to establish a numerical mapping relationship between the characteristic properties of steel slag and the carbon fixation effect, so as to realize the digital prediction of the carbon fixation potential of steel slag under unknown process conditions.

[0031] The carbon sequestration capacity prediction model construction subsystem establishes a data transmission connection with the data preprocessing unit in the multidimensional feature data construction subsystem. The carbon sequestration capacity prediction model construction subsystem is configured to receive standardized chemical composition feature data. Physical property characteristic data and process parameter characteristic data As input variables to the model, the carbon sequestration capacity prediction model construction subsystem uses the carbon dioxide carbon sequestration per unit mass of steel slag as the model's output target variable.

[0032] The carbon sequestration capacity prediction model construction subsystem integrates a regression prediction engine based on the Extreme Gradient Boosting Tree (XGBoost) algorithm. Multiple machine learning models are configured using an ensemble learning strategy to construct an additive model composed of multiple decision trees. Through iterative training, the subsystem continuously generates new regression trees based on the current residuals and performs a weighted sum of the prediction results from all regression trees to approximate the actual carbon sequestration amount. Leveraging the machine learning model's ability to process high-dimensional sparse data and capture nonlinear feature interactions, the subsystem fits complex carbon sequestration reaction patterns into a multi-dimensional feature space composed of chemical composition, physical properties, and process parameters.

[0033] The carbon sequestration capacity prediction model construction subsystem is equipped with a ten-fold cross-validation logic component. This component randomly divides the training sample set into ten non-overlapping subsets. During model training, the subsystem sequentially selects nine subsets as training data and the remaining subset as validation data, repeating this training and validation process ten times. The subsystem then calculates the average coefficient of determination from the ten validation results. The root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate the generalization performance of the current model under the combination of hyperparameters. Figure 5 As shown.

[0034] After training, validation, and optimization, the carbon sequestration capacity prediction model construction subsystem generates and stores a defined nonlinear prediction function. This nonlinear prediction function characterizes the mathematical mapping relationship between the input feature variables and the output carbon sequestration amount, and its specific mathematical expression is as follows: ; in, This represents the predicted carbon sequestration per unit of steel slag. This represents the structure of the trained XGBoost model. Represents chemical composition characteristic data, Data representing physical properties This represents characteristic data of process parameters.

[0035] The predictive model defined by this formula forms the core of the calculation for subsequent energy consumption and carbon sequestration dual-objective game optimization subsystem to optimize process parameters.

[0036] The Bayesian hyperparameter optimization module is integrated into the carbon sequestration capacity prediction model construction subsystem. It is configured to automatically search for and lock the extreme gradient boosting tree XGBoost algorithm hyperparameter combinations that maximize the performance of the prediction model.

[0037] The Bayesian hyperparameter optimization module is configured to define and initialize the search space for the hyperparameters to be optimized. The search space encompasses the key control parameters of the XGBoost algorithm, including the learning rate controlling the model's learning step size, the maximum tree depth limiting the complexity of the tree structure, the subsample ratio controlling the random sampling ratio of each tree, the column sampling ratio controlling the feature sampling ratio of each tree, and the minimum subweight controlling the minimum sum of weights of the leaf nodes. The Bayesian hyperparameter optimization module sets continuous or discrete value boundaries for each of these hyperparameters, forming a multidimensional hyperparameter optimization space.

[0038] The Bayesian hyperparameter optimization module is configured to construct a surrogate model based on a Gaussian process. The surrogate model is used to fit the probability distribution relationship between the hyperparameter combination and the model's predictive performance metrics. The Bayesian hyperparameter optimization module calculates the average coefficient of determination using 10-fold cross-validation. The objective function is used for optimization. In each iteration, the Bayesian hyperparameter optimization module uses a sampling function (e.g., an expectation boosting strategy) to select the next most promising hyperparameter combination from the search space for evaluation. The selected hyperparameter combination is then fed into the XGBoost algorithm engine for actual training and validation, and the corresponding average coefficient of determination is obtained. As a feedback observation.

[0039] The Bayesian hyperparameter optimization module is configured to update the posterior distribution of the Gaussian process surrogate model based on new observations, thereby correcting the estimate of the shape of the objective function in the hyperparameter space. The module repeatedly executes an iterative loop of selecting hyperparameters, evaluating performance, and updating the surrogate model until a preset upper limit for the number of iterations is reached or the improvement in the objective function value falls below a preset threshold. The Bayesian hyperparameter optimization module ultimately outputs a set of average determination coefficients. The optimal hyperparameter combination that reaches the global maximum value is obtained and passed to the carbon sequestration capacity prediction model construction subsystem for training on the full training sample set to generate the final nonlinear prediction function.

[0040] The prediction execution module, as the system's online inference computing engine, is configured to apply the trained carbon fixation capacity prediction model to the steel slag sample to be tested or the process scheme to be evaluated, and output a quantitative carbon fixation capacity prediction value.

[0041] The prediction execution module is configured with a model parameter loading interface. Through this interface, the module reads and loads the structure file of the trained extreme gradient boosting tree (XGBoost) model from the carbon sequestration capability prediction model construction subsystem. This structure file contains the feature indices of all decision tree split nodes, split thresholds, and weight values ​​of the leaf nodes. The prediction execution module then reconstructs the nonlinear prediction function structure in memory that is completely identical to that used during the training phase.

[0042] The prediction execution module is configured to receive feature data of the sample to be tested. When the system is in single prediction mode, the prediction execution module receives the chemical composition feature data, physical property feature data, and set process parameter feature data of the steel slag to be tested, input by the user. When the system is in process optimization mode, the prediction execution module receives the combination of candidate process parameter feature data generated by the dual-objective game optimization subsystem of energy consumption and carbon sequestration.

[0043] The prediction execution module integrates a consistency preprocessing logic component. To ensure that the data distribution of the input model is consistent with the training data distribution, the consistency preprocessing logic component is configured to retrieve the mean and standard deviation parameters calculated and stored by the data preprocessing unit during the training phase. Using the stored mean and standard deviation parameters, the consistency preprocessing logic component performs Z-score normalization on the feature data of the test samples, rather than recalculating the statistical parameters of the test samples themselves. This mechanism ensures the accuracy of the input data for model inference in the feature space.

[0044] The prediction execution module is configured to perform forward inference computation. It inputs standardized feature data into the loaded extreme gradient boosting tree (XGBoost) model structure. Each regression tree in the model plans a path from the root node to a leaf node based on the input feature values ​​and returns the corresponding leaf node weights. The prediction execution module performs a weighted sum of the leaf node weights returned by all regression trees to calculate the final predicted carbon sequestration amount per unit of steel slag. This predicted value is output to the user interface or fed back as the result of the fitness function calculation to the energy consumption and carbon sequestration dual-objective game optimization subsystem.

[0045] Please see the appendix Figure 2 The objective function construction module is embedded within the dual-objective game optimization subsystem of energy consumption and carbon sequestration. It is configured to transform the engineering problem of process parameter selection into a mathematical optimization problem with explicit numerical objectives. The objective function construction module establishes two conflicting evaluation dimensions: maximizing carbon sequestration technical indicators and minimizing process operating energy costs.

[0046] The self-defined function construction module integrates a technology output calculation unit. This unit is configured to define a first optimization objective function aimed at maximizing carbon sequestration. The technology output calculation unit establishes a data call connection with the prediction execution module in the aforementioned embodiment. During the optimization process, the technology output calculation unit uses chemical composition characteristic data... and physical property characteristic data Locked as a constant, it characterizes the inherent properties of the specific batch of steel slag raw material to be processed. The technical output calculation unit will use the process parameter characteristic data. Data is used as decision variables. The technology output calculation unit calls the nonlinear prediction function stored in the prediction execution module. The predicted carbon sequestration amount is calculated under the current combination of decision variables. The technology output calculation unit sets maximizing this predicted value as the fitness function of the first dimension, mathematically expressed as maximizing...

[0047] The objective function construction module integrates an energy consumption cost calculation unit. This unit is configured to define a second optimization objective function oriented towards minimizing variable operating costs. Based on reaction kinetics principles and equipment power consumption characteristics, the unit constructs a physical energy consumption calculation model. It identifies reaction temperature, carbon dioxide partial pressure, stirring rate, and carbonization time as key factors driving energy consumption. The unit then maps these physical parameters to economic costs using energy conversion coefficients.

[0048] The energy cost calculation unit stores a variable energy cost function. The variable energy cost function is expressed as: ; in, This represents the variable energy cost in a single processing step. This indicates the reaction temperature, corresponding to the heat energy consumption required to maintain the temperature of the reaction vessel. This indicates the partial pressure of carbon dioxide, corresponding to the compressor power consumption required to maintain the pressure environment; This indicates the stirring rate, corresponding to the motor power consumption required for mechanical mixing. Represents carbonization time, serving as a time-domain multiplier for all continuously energy-consuming devices; This represents the energy consumption cost coefficient per unit temperature, and its value is calculated from the specific heat capacity of the reactor and the unit price of heat energy. This represents the unit pressure energy consumption cost coefficient, the value of which is calculated from the adiabatic compression work formula and the unit price of electricity. This represents the energy cost coefficient per unit stirring rate, calculated from the motor efficiency curve and the unit price of electricity. The energy cost calculation unit minimizes the calculated result. The fitness function is set as the second dimension, mathematically expressed as minimizing... .

[0049] The objective function construction module outputs the defined first and second optimization objective functions to the Pareto optimization unit, which builds a mathematical environment for the subsequent multi-objective genetic algorithm to perform game analysis between output and cost, ensuring that the optimized process parameters are not only theoretically optimal for carbon sequestration, but also reasonable in economic operation.

[0050] The Pareto optimization module, as the core solution engine of the dual-objective game optimization subsystem of energy consumption and carbon sequestration, is configured to perform multi-objective evolutionary calculations in a multi-dimensional process parameter space to resolve the mathematical conflict between the two objective functions of maximizing carbon sequestration and minimizing energy consumption cost.

[0051] The Pareto optimization module integrates a constraint setting unit. This constraint setting unit is configured to define process parameter characteristic data. The physical feasible region of each decision variable. The constraint setting unit stores the hardware limit parameters of the reactor and its supporting equipment, including the reaction temperature. Permissible range, partial pressure of carbon dioxide Safe operating range, stirring rate Motor speed range and carbonization time The time window limitation. The constraint setting unit constructs a multi-dimensional rectangular constraint space, which forces all candidate process parameter combinations subsequently generated by the Pareto optimization module to strictly fall within this physical feasible region, preventing the algorithm from outputting an invalid solution that is theoretically optimal but cannot be implemented in engineering.

[0052] The Pareto optimization module integrates the NSGA-II non-dominated sorting genetic algorithm computational engine with an elitist strategy. This engine is configured to initialize and generate a population containing multiple random combinations of process parameters. The NSGA-II engine then invokes the technology output computation unit and the energy cost computation unit to calculate the first optimization objective function value for each individual in the population. Second optimization objective function value

[0053] The Pareto optimization module is configured to perform a fast non-dominated sort operation. The Pareto optimization module performs a sort based on each individual's... Value and The Pareto optimization module determines the dominance relationship between individuals. If individual A has a carbon sequestration capacity no lower than individual B and an energy cost no higher than individual B, and is superior to individual B in at least one dimension, then individual A is considered to dominate individual B. The Pareto optimization module classifies individuals in the population that are not dominated by any other individual into the first non-dominated level, i.e., the current Pareto front. The Pareto optimization module then classifies the remaining individuals into subsequent levels according to their degree of dominance.

[0054] The Pareto optimization module is configured to calculate crowding distance. To maintain the diversity of the solution set distribution in the objective space and prevent the solution set from getting trapped in local convergence, the Pareto optimization module calculates the Euclidean distance between adjacent individuals in the same non-dominated level in the objective function space. The Pareto optimization module prioritizes retaining individuals with larger crowding distances to ensure that the final Pareto optimal solution set can uniformly cover the entire tradeoff range from high cost and high output to low cost and low output.

[0055] The Pareto optimization module is configured to perform genetic evolution operations. It selects parent individuals from the current population using a tournament selection operator. The module then uses simulated binary crossover and polynomial mutation operators to recombine and perturb the process parameter characteristics of the parent individuals, generating a offspring population containing new combinations of process parameters. Finally, employing an elite retention strategy, the module merges the parent and offspring populations and performs non-dominated sorting and crowding distance calculations again, selecting the top-ranked individuals to form the next generation population.

[0056] The Pareto optimization module repeats the above evolutionary cycle until a preset number of iterations is reached. The Pareto optimization module ultimately outputs a set of non-dominant process parameter combinations, constituting the Pareto optimal solution set. Each element in the Pareto optimal solution set represents a specific process operation scheme that cannot further increase carbon sequestration without sacrificing energy costs, or cannot further reduce energy costs without reducing carbon sequestration. The Pareto optimization module then passes this Pareto optimal solution set to the economic feasibility assessment subsystem.

[0057] Please see the appendix Figure 3 The cost accounting unit serves as the basic data support component of the economic feasibility assessment subsystem. It quantifies the comprehensive economic input of the steel slag carbonization project under different process schemes, providing numerical basis for cost-side analysis in subsequent break-even analysis.

[0058] The cost accounting unit integrates a fixed cost data storage logic component. This component is configured to maintain and manage the basic investment data for carbon sequestration plants of different construction scales. The fixed cost data storage logic component is based on the plant's annual steel slag processing volume. Based on the size, carbon sequestration projects are divided into three tiers: large, medium, and small. For each tier, the fixed cost data storage logic component stores the corresponding plant fixed costs. Factory fixed costs This is a comprehensive financial indicator, its components including amortization of the purchase cost of the reactor and supporting equipment, amortization of plant construction costs, amortization of land use rights, fixed management personnel salaries, and non-productive routine maintenance costs. The fixed cost data storage logic component is configured to store the expected annual steel slag processing volume of the plant based on user input. Automatically index and match the corresponding factory fixed costs The numerical value enables static locking of non-process-sensitive costs.

[0059] The cost accounting unit is equipped with a dynamic cost data receiving interface. This interface establishes a numerical transmission connection with the energy consumption and carbon sequestration dual-objective game optimization subsystem. Through this interface, the cost accounting unit receives the corresponding variable energy consumption cost for each combination of process parameters in the Pareto optimal solution set. Variable energy cost The energy consumption cost calculation channel is based on the formula The calculated monetary value of energy consumption in a single processing step. The cost accounting unit will receive the variable energy cost. Corresponding prediction unit of carbon sequestration in steel slag Perform association tagging to ensure a logical one-to-one correspondence between cost data and output data.

[0060] The cost accounting unit integrates a unit comprehensive cost calculation logic component. This component is configured to map static fixed costs and dynamic variable costs to the unit mass of steel slag processing cost. The unit comprehensive cost calculation logic component first performs an amortization calculation, allocating the plant's fixed costs... Divide by the factory's annual steel slag processing volume This yields the fixed cost per unit mass of steel slag. Subsequently, the unit comprehensive cost calculation logic component combines the fixed cost per unit mass of steel slag with the variable energy cost. By performing addition, we construct the cost term representation on the right-hand side of the decision inequality: .

[0061] This calculation process unifies the dimensions from macro-level investment data to micro-level process cost data, enabling the data output by the cost accounting unit to be directly compared with carbon trading revenue. The cost accounting unit transmits the calculated comprehensive cost data for each set of process schemes to the decision-making unit.

[0062] The revenue accounting unit and the decision-making unit together constitute the core decision-making logic of the economic feasibility assessment subsystem. It is configured to transform the carbon sequestration potential at the technical level into a profit and loss judgment at the economic level, ensuring that the final output process solution is commercially feasible.

[0063] The revenue accounting unit is configured to manage economic parameters related to the value of carbon emission reductions. The revenue accounting unit internally stores carbon emission trading market prices. Carbon emission trading market prices This data represents the direct market trading revenue a company can obtain for reducing one ton of carbon dioxide emissions under the carbon trading system. This data can be configured to be obtained in real-time via an external API interface or manually set by the user based on current market conditions. The revenue calculation unit also stores the unit's social carbon cost avoidance value. Unit social carbon cost avoidance value This represents the monetary value of the climate change damage avoided by society for every ton of carbon dioxide emissions reduced, and is typically used as a quantitative basis for government subsidies or corporate social responsibility assessments. The revenue accounting unit is configured to incorporate carbon emissions trading market prices. With unit social carbon cost avoidance value By performing summation operations, a comprehensive economic value factor per unit of carbon sequestration is constructed for subsequent revenue calculations.

[0064] The decision-making unit establishes a bidirectional data connection with the revenue calculation unit and the cost calculation unit. The decision-making unit is configured to receive the Pareto optimal solution set output from the Pareto optimization module. This solution set contains multiple sets of non-dominant process parameter characteristic data. and its corresponding predicted unit of steel slag carbon fixation The decision-making unit simultaneously receives the comprehensive cost data corresponding to each set of process schemes calculated by the cost accounting unit.

[0065] The decision-making unit is configured to perform rigorous economic feasibility logic checks. For each combination of process parameters in the Pareto optimal solution set, the decision-making unit calls the stored feasibility inequalities for verification one by one. The feasibility inequalities stored in the decision-making unit are expressed as follows: ; in, The predicted carbon sequestration per unit of steel slag is provided by the technology output calculation channel; This indicates the price in the carbon emissions trading market; This represents the unit social carbon cost avoidance value; This represents the factory's fixed costs; This indicates the factory's annual steel slag processing volume; This represents the variable energy cost, calculated from the energy cost calculation channel based on the formula. The calculation yielded the result.

[0066] Left side of the inequality This quantifies the total economic benefit generated per unit mass of steel slag through carbon fixation under specific process conditions. (Right side of the inequality) The total economic input required to process a unit mass of steel slag was quantified. The decision-making unit was configured to mark the combination of process parameters that satisfies the above inequalities as an economically feasible solution.

[0067] The decision-making unit also integrates an optimal recommendation and screening logic component. When multiple economically feasible solutions satisfy the feasibility inequality, the optimal recommendation and screening logic component is configured to calculate the net benefit value of each economically feasible solution. The net benefit value is calculated as total economic benefit minus total economic input. The optimal recommendation and screening logic component sorts the net benefit values ​​of all economically feasible solutions and selects the single process parameter combination with the largest net benefit value. The decision-making unit generates a final decision report from this process parameter combination with the largest net benefit value, along with its corresponding predicted carbon sequestration, energy consumption cost, and expected net benefit, and outputs it through a human-computer interaction interface, thereby completing the decision convergence from multi-objective optimization results to a single optimal implementation scheme.

[0068] The decision report generation module, serving as the data organization hub for the result output and human-computer interaction interface subsystem, is configured to transform the discrete numerical results output by the economic feasibility assessment subsystem into structured decision support information. The decision report generation module establishes a one-way data receiving connection with the decision-making unit within the economic feasibility assessment subsystem.

[0069] The decision report generation module is equipped with a feasibility status identification logic component. This component is configured to receive the feasibility status markers of the Pareto optimal solution set output by the decision-making unit. When the feasibility markers output by the decision-making unit indicate no solution or complete infeasibility, the feasibility status identification logic component triggers the generation of a negative decision report. This report includes current market price parameters and fixed cost parameters, and marks the conclusion that profitability cannot be achieved under the current technological conditions. When the decision-making unit outputs that at least one economically feasible solution exists, the feasibility status identification logic component triggers the generation of a positive decision report.

[0070] The decision report generation module integrates an optimal solution parameter encapsulation logic component. For affirmative decision reports, this component is configured to receive the single process parameter combination with the highest net benefit value from the decision-making unit. This component is configured to construct data fields containing detailed process settings, specifically encapsulating the recommended reaction temperature. Recommended partial pressure of carbon dioxide Recommended carbon dioxide gas flow rate Recommended carbon dioxide concentration Recommended carbonization time Recommended stirring rate Recommended liquid-to-solid ratio and recommended sample processing pressure .

[0071] The decision report generation module integrates a key performance indicator (KPI) calculation logic component. This component is configured to generate quantified technical and economic evaluation indicators based on the received optimal solution data. Specifically, it encapsulates the predicted carbon sequestration per unit of steel slag. As a technical output indicator, the key performance indicator (KPI) calculation logic component is configured to encapsulate variable energy consumption cost. With allocated fixed costs The sum is used as a unit comprehensive cost indicator. The key performance indicator calculation logic component is configured to encapsulate total economic benefits. As an expected revenue indicator, the key performance indicator (KPI) calculation logic component is configured to encapsulate the difference between total economic returns and total economic input as an expected net return indicator.

[0072] The decision report generation module is equipped with a data formatting output interface. This interface is configured to serialize the encapsulated process parameter fields and key performance indicator fields according to a preset communication protocol, generating a standardized electronic report data stream. This electronic report data stream is used to drive subsequent visualization units for graphic rendering or to transmit it to external storage devices for archiving.

[0073] Please see the appendix Figure 4 The visualization unit is configured to transform the standardized electronic report data stream output by the decision report generation module into a graphical human-computer interaction interface, providing operators with intuitive data insights and decision support.

[0074] The visualization unit integrates a Pareto front scatter plot rendering engine. This engine is configured to map the Pareto optimal solution set calculated by the Pareto optimization unit to a two-dimensional Cartesian coordinate system. The Pareto front scatter plot rendering engine defines the horizontal axis of the two-dimensional Cartesian coordinate system as the variable energy cost. The vertical axis of the two-dimensional rectangular coordinate system is defined as the predicted unit carbon fixation amount of steel slag. The Pareto front scatter plot rendering engine plots each combination of process parameters in the optimal solution set as a discrete data point. The visualization unit uses a specific color or marker shape to highlight the data point corresponding to the single process parameter combination with the largest net benefit value selected by the decision-making unit, intuitively presenting the position of the recommended solution in the technical and economic trade-off.

[0075] The visualization unit integrates a process parameter details list display area. This area is configured to display the specific values ​​of the optimal process parameter combination, encapsulated by the optimal solution parameter encapsulation logic component. The process parameter details list display area presents the recommended reaction temperature in a table or grid layout. Recommended partial pressure of carbon dioxide Recommended carbon dioxide gas flow rate Recommended carbon dioxide concentration Recommended carbonization time Recommended stirring rate Recommended liquid-to-solid ratio and recommended sample processing pressure This display method ensures that operators can directly read the control commands required by the production equipment settings.

[0076] The visualization unit integrates an economic feasibility comparison chart rendering engine. This engine is configured to calculate key performance indicators (KPIs) based on the KPI calculation logic component, generating a cost-benefit comparison bar chart. The engine then draws the first bar representing the total economic input; the height of this first bar corresponds to the variable energy cost. With allocated fixed costs The sum. The economic feasibility comparison chart rendering engine draws a second bar representing the total economic benefit; the height of the second bar corresponds to the predicted carbon sequestration per unit of steel slag. With carbon emission trading market prices and unit social carbon cost avoidance value The product of the sums. The visualization unit intuitively indicates the expected net income indicator through the height difference between the second and first bar charts, helping managers quickly assess the project's profitability.

[0077] The visualization unit is equipped with an interactive response logic component. When the visualization unit detects that a user has selected a specific data point that is not a recommended point in the Pareto front scatter plot, the visualization unit triggers a synchronous refresh of the process parameter details list display area and the economic feasibility comparison chart rendering engine, displaying the detailed parameters and economic analysis data corresponding to that specific data point. This interactive mechanism allows users to explore other suboptimal solutions on the Pareto front to address potential non-economic constraints in specific production scenarios.

Claims

1. A machine learning-based system for evaluating the carbon sequestration potential and economic benefits of steel slag, characterized in that, include: The multidimensional feature data construction subsystem is used to collect literature data and experimental records in the field of steel slag mineral carbonization and generate standardized feature data containing chemical composition, physical properties and process parameters. A carbon sequestration capacity prediction model construction subsystem is connected to the multidimensional feature data construction subsystem. It is used to establish a nonlinear mapping relationship between the characteristic attributes of steel slag and the carbon dioxide sequestration amount per unit mass of steel slag based on the standardized feature data and a machine learning model. A dual-objective game optimization subsystem for energy consumption and carbon sequestration is connected to the carbon sequestration capacity prediction model construction subsystem. It is used to establish optimization objectives of maximizing carbon sequestration and minimizing energy consumption costs, and executes a multi-objective genetic algorithm to output a Pareto optimal solution set containing multiple non-dominant process parameter combinations. An economic feasibility assessment subsystem, connected to the energy consumption and carbon sequestration dual-objective game optimization subsystem, is used to construct a cost model based on the investment and operating costs of steel slag carbon sequestration equipment, combine carbon emission reduction benefit parameters, perform cost-benefit analysis on the process parameter combinations in the Pareto optimal solution set, and output recommended carbon sequestration process schemes.

2. The machine learning-based steel slag carbon sequestration potential and economic benefit evaluation system according to claim 1, characterized in that, The multidimensional feature data construction subsystem includes: The chemical composition feature extraction unit is used to analyze the oxide composition data of steel slag raw materials to generate chemical composition feature data; The physical property feature extraction unit is used to analyze the physical morphology data of steel slag raw materials to generate physical property feature data; The process parameter feature extraction unit is used to analyze the operating condition data during the carbonization reaction process to generate process parameter feature data. The data preprocessing unit is used to identify missing data in the chemical composition feature data, the physical property feature data, and the process parameter feature data and perform column mean imputation operation, and perform Z-score normalization processing on the imputed feature data.

3. The machine learning-based steel slag carbon sequestration potential and economic benefit evaluation system according to claim 1, characterized in that, The carbon sequestration capacity prediction model construction subsystem integrates multiple machine learning models and ten-fold cross-validation logic components. The various machine learning models are used to construct a flexible nonlinear additive model using an ensemble learning strategy, with the chemical composition, physical properties and process parameters as input variables and the carbon dioxide carbon sequestration amount per unit mass of steel slag as the output target variable for iterative training. The ten-fold cross-validation logic component is used to divide the training sample set into non-overlapping subsets, perform the training and validation process in a loop, and evaluate the generalization performance of the model hyperparameter combination based on the validation results.

4. The machine learning-based steel slag carbon sequestration potential and economic benefit evaluation system according to claim 1, characterized in that, The energy consumption and carbon sequestration dual-objective game optimization subsystem integrates an objective function construction module. The objective function construction module includes a technology output calculation unit and an energy consumption cost calculation unit; The technology output calculation unit is used to call the nonlinear mapping relationship in the subsystem of the carbon fixation capacity prediction model, calculate the predicted carbon fixation amount per unit of steel slag under different combinations of process parameters, and set the maximum predicted carbon fixation amount per unit of steel slag as the fitness function of the first dimension. The energy consumption cost calculation unit is used to identify reaction temperature, carbon dioxide partial pressure, stirring rate and carbonization time as key factors driving energy consumption. Based on the principle of reaction kinetics and the power consumption characteristics of the equipment, it calculates the variable energy consumption cost and sets the minimization of the variable energy consumption cost as the fitness function of the second dimension.

5. The machine learning-based steel slag carbon sequestration potential and economic benefit evaluation system according to claim 4, characterized in that, The dual-objective game optimization subsystem for energy consumption and carbon sequestration also includes a Pareto optimization module. The Pareto optimization module integrates a constraint setting unit and a non-dominated sorting genetic algorithm calculation engine with an elite strategy. The constraint setting unit is used to define the physical feasible domain of each decision variable in the process parameters and construct a multi-dimensional rectangular constraint space. The non-dominated sorting genetic algorithm computation engine with elitist strategy is used to calculate the fitness function values ​​of the first dimension and the second dimension for individuals in the population, perform fast non-dominated sorting operations and crowding distance calculations, and generate the Pareto optimal solution set through genetic evolution operations.

6. The machine learning-based steel slag carbon sequestration potential and economic benefit evaluation system according to claim 1, characterized in that, The economic feasibility assessment subsystem includes a cost accounting unit; The cost accounting unit integrates a fixed cost data storage logic component and a dynamic cost data receiving interface. The fixed cost data storage logic component is used to match the corresponding fixed cost of the factory based on the size of the factory's annual steel slag processing volume. The dynamic cost data receiving interface is used to receive the variable energy cost calculated for each combination of process parameters in the Pareto optimal solution set. The cost accounting unit also integrates a unit comprehensive cost calculation logic component, which is used to calculate the sum of the fixed cost allocated per unit mass of steel slag and the variable energy consumption cost.

7. The machine learning-based steel slag carbon sequestration potential and economic benefit evaluation system according to claim 6, characterized in that, The economic feasibility assessment subsystem also includes a revenue accounting unit and a decision-making unit; The revenue accounting unit is used to store the carbon emission trading market price and the unit social carbon cost avoidance value; The decision-making unit is used to perform feasibility logic verification on each combination of process parameters in the Pareto optimal solution set. When performing feasibility logic verification, the decision-making unit determines whether the product of the predicted unit carbon sequestration of steel slag and the comprehensive economic value factor is greater than the sum of the fixed cost allocated to the unit mass of steel slag and the variable energy consumption cost; wherein, the comprehensive economic value factor is composed of the carbon emission trading market price and the unit social carbon cost avoidance value. The decision-making unit is also used to mark the process parameter combinations that meet the judgment conditions as economically feasible solutions, and to select the process parameter combinations with the largest net benefit value as recommended solutions.

8. The machine learning-based steel slag carbon sequestration potential and economic benefit evaluation system according to claim 7, characterized in that, It also includes a decision report generation module; The decision report generation module is connected to the decision judgment unit and is used to generate a positive decision report based on the selected combination of process parameters with the largest net benefit value. The decision report generation module integrates an optimal solution parameter encapsulation logic component and a key performance indicator calculation logic component. The optimal solution parameter encapsulation logic component is used to construct data fields containing recommended process settings; The key performance indicator calculation logic component is used to generate evaluation indicators including technical output indicators, unit comprehensive cost indicators, expected revenue indicators, and expected net profit indicators.

9. The machine learning-based steel slag carbon sequestration potential and economic benefit evaluation system according to claim 8, characterized in that, It also includes a visualization display unit; The visualization unit is used to convert the data output by the decision report generation module into a graphical human-computer interaction interface; The visualization unit integrates a Pareto front scatter plot rendering engine, which maps the Pareto optimal solution set to a two-dimensional rectangular coordinate system with variable energy consumption cost as the horizontal axis and the predicted unit carbon fixation of steel slag as the vertical axis, and highlights the data points corresponding to the process parameter combination with the largest net benefit value. The visualization unit also integrates a process parameter detail list display area and an economic feasibility comparison chart rendering engine.

10. The machine learning-based steel slag carbon sequestration potential and economic benefit assessment system according to claim 2, characterized in that, The chemical composition characteristics data include the mass percentages of calcium oxide, magnesium oxide, silicon dioxide, aluminum oxide, iron oxide, manganese oxide, titanium dioxide, and phosphorus pentoxide. The physical property characteristic data includes the average size of steel slag particles; The process parameter characteristic data includes reaction temperature, carbon dioxide partial pressure, carbon dioxide gas flow rate, carbon dioxide concentration, carbonization time, stirring rate, liquid-solid ratio, and sample processing pressure.

Images