Intelligent coal blending and coking method based on kernel mapping coupling correction algorithm
The intelligent coal blending and coking method using the kernel mapping coupling correction algorithm solves the problem of unstable coke quality in coking production, achieves stable control of coke quality and cost optimization, improves production efficiency and environmental protection level, and promotes the intelligent transformation of coking production.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- МААНЬШАНЬ АЙРОН ЭНД СТИЛ КО ЛТД
- Filing Date
- 2026-03-17
- Publication Date
- 2026-06-19
AI Technical Summary
In existing technologies, the prediction of coal blending quality and the control of coke performance in coking production suffer from insufficient prediction accuracy and weak adaptability. They are unable to reflect the nonlinear coupling relationship between coal quality indicators, resulting in unstable coke quality, high cost, low energy efficiency, and difficulty in meeting the stable production requirements under complex operating conditions.
An intelligent coal blending and coking method based on kernel mapping and coupling correction algorithm is adopted. Through data acquisition, feature scaling, cleaning, nonlinear support vector regression modeling, hyperparameter optimization and coupling correction algorithm, a highly adaptable and accurate prediction model is constructed to achieve stable control of coke quality and cost optimization.
It has achieved stability in coke quality and improved production efficiency, reduced raw material costs, promoted the intelligent transformation of coking production, improved production efficiency and environmental protection levels, and adapted to the complex working conditions of multi-source coal samples.
Smart Images
Figure CN122242852A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of coal chemical technology, specifically to an intelligent coal blending and coking method based on a kernel mapping coupling correction algorithm. Background Technology
[0002] In coking production, coal blending quality prediction and coke performance control are directly related to production stability and product quality. Existing technologies often use empirical formulas or linear models for coal quality evaluation, but these methods struggle to effectively reflect the nonlinear coupling relationships between coal quality indicators, resulting in limited prediction accuracy. The physicochemical properties of different coal sources vary significantly, and fluctuations in factors such as moisture, ash content, volatile matter, and particle size distribution can easily introduce biases, leading to shortcomings in the adaptability and robustness of prediction models. Kernel mapping algorithms can map coal quality data from a low-dimensional space to a high-dimensional feature space to reveal nonlinear correlation characteristics and improve the fitting ability between coal quality and coke performance. However, in practical applications, using kernel mapping algorithms alone can easily lead to overfitting and bias accumulation, making it difficult to meet the stable prediction requirements under complex operating conditions. By introducing correction coefficients into the kernel mapping algorithm, the model calculation results can be dynamically calibrated against actual production data, thereby enhancing the model's adaptability to multi-source coal samples and improving the stability and generalization ability of the prediction results. Therefore, a modeling method combining kernel mapping algorithms and correction coefficients is needed to address the problems of insufficient prediction accuracy and poor adaptability in existing technologies.
[0003] Regarding coking coal blending methods, patent number CN117070237A discloses an online method for controlling coking coal blending to produce high-quality coke. This technology simply assumes that coke quality is determined by the G and Y values of the blended coal, adjusting them by adding binders and leanening agents, and conducting offline laboratory testing. However, this method lacks system accuracy, robustness, and real-time performance. It also ignores the nonlinear relationships between coal quality parameters, resulting in poor model generalization ability. Currently, coking enterprises generally face pressure from a shortage of high-quality coking coal resources, continuously rising raw material costs, and increasingly stringent coke quality requirements from the downstream steel industry. Existing prediction and blending methods are limited by empirical models and fixed correction parameters, making it difficult to accurately reflect the nonlinear coupling relationships between different coal sources. This leads to unstable coke quality predictions, insufficient energy efficiency, and difficulty in supporting stable production under cost reduction, efficiency improvement, and green low-carbon requirements. Therefore, a real-time, accurate method for predicting coke quality is needed for stable quality and cost reduction. Summary of the Invention
[0004] The purpose of this invention is to provide an intelligent coal blending and coking method based on a kernel mapping coupling correction algorithm to overcome the shortcomings of existing technologies.
[0005] To achieve the above objectives, the present invention provides the following technical solution:
[0006] A smart coal blending and coking method based on a kernel mapping coupling correction algorithm includes the following steps:
[0007] S1: Collect data from the entire coking process to form a dataset, including: industrial analysis data for single coal types, G-value, Y-value, average maximum reflectance of vitrinite, active-inert ratio, proportion of each coal type, blended coal indicators, fineness of blended coal, target quality indicators, and production costs.
[0008] S2: Divide the dataset in S1 into a training set (70%), a test set (15%), and a validation set (15%). The training set is used to fit the parameters of the support vector regression model, allowing the model to learn from the data and capture the inherent mapping between input features and target variables, which is used for hyperparameter tuning. The test set is used to tune the coupling correction algorithm, which is established based on the model's prediction bias on unknown data. The validation set is used for the final, one-time performance evaluation.
[0009] S3: RobustScaler standardizes the continuous features of the training set in S2, using the median and quartile ranges for feature scaling; the same parameters are used to transform the test and validation sets to ensure data distribution consistency.
[0010] S4: Clean the basic data in S3, including: using KNN interpolation to handle abnormal and missing data; directly deleting feature fields or sample records with missing rates exceeding a preset threshold to ensure dataset quality;
[0011] S5: Nonlinear Support Vector Regression Modeling, including: constructing a nonlinear SVR model on the training set, establishing and training a nonlinear SVR model for each target quality indicator and production cost; using the Gaussian radial basis function (RBF) to map the data to a high-dimensional space, and finding a linear regression model in the high-dimensional space;
[0012] S6: Hyperparameter optimization, including: directly calling the GridSearchCV module in the scikit-learn library on the training set to use grid search and cross-validation to fine-tune the combination of the core hyperparameters of each SVR model, such as the penalty coefficient C, the insensitive loss parameter ε, and the kernel coefficient γ, and determine the optimal parameter configuration;
[0013] S7: Model testing and algorithm coupling correction, including: for the test set, making predictions based on the optimal hyperparameter model and comparing them with actual production data; constructing a coupling correction algorithm based on the comparison results to compensate for system errors and improve the model's prediction accuracy and adaptability to operating conditions;
[0014] S8: Model validation and optimal interval determination, including: performing final validation of the optimal parameter model after calibration by the coupling correction algorithm on the validation set, evaluating its prediction accuracy and robustness, and then using the multi-objective optimization algorithm NSGA-II to determine the production interval where coke quality indicators and production costs are optimally coordinated, providing quantitative decision support for coal blending schemes.
[0015] Furthermore, the target quality indicators in S1 include coke reactivity, coke post-reaction strength, coke sulfur content, coke crush resistance index, coke abrasion resistance index, and lump coke ratio; production costs include raw coal cost, heating fuel cost, and environmental protection and carbon emission cost.
[0016] Furthermore, the specific calculation formula for RobustScaler standardization in S3 is as follows:
[0017] X [x1,x2,x3,...,x n ] R l×n
[0018]
[0019]
[0020]
[0021] Y [y1,y2,y3,...,y n ] R l×n
[0022] In the formula, X is the set of raw coal characteristics and coal blending data. The raw coal characteristics are obtained based on the coal quality analysis data of each individual coal, and the coal blending data and process parameters are all derived from actual production historical data; n is the number of indicators contained in x, and l is the number of individual coals in the sets X and Y. It is the position of the median in the dataset. It is the median of dataset X. It is the interquartile range. It is the lower quantile. It is the upper quantile, y i Y is the RobustScaler-standardized data for the i-th indicator, and Y is the RobustScaler-standardized analysis dataset; where the lower quantile is the position of the variable values arranged in ascending order. The number, the upper quantile is the position. The number.
[0023] Furthermore, the KNN imputation method in S4 includes: based on the assumption that "similar samples have similar attributes," it intelligently estimates missing values by utilizing the overall similarity between data points. The specific calculation formula is as follows:
[0024]
[0025]
[0026]
[0027]
[0028]
[0029]
[0030]
[0031]
[0032]
[0033]
[0034] In the formula, It is the sample missing rate. It is the number of features in the data. It is the first The number of missing features in the nth sample, the number of missing features in the dataset. Features and the j-th feature Mutual information between them is defined as , It is an indicator that measures the importance of different features among all features. It is a weight matrix. These are the weights obtained after feature normalization. Defined as Euclidean distance, It is the K samples obtained and the samples to be filled. The matrix of time distances in ascending order. To utilize K samples and the sample to be filled The distance between these K samples reflects the relationship between them and the K samples. The weights used to measure sample similarity are defined by the degree of similarity. This is to prevent the weight from being a very small number with a denominator of 0. It is the corrected weight. The sample to be filled is obtained by weighted summation based on weights. The imputed value of the k-th feature.
[0035] Furthermore, the specific calculation formulas for nonlinear support vector regression and hyperparameter optimization in S5-S6 are as follows:
[0036]
[0037]
[0038]
[0039]
[0040]
[0041]
[0042]
[0043] In the formula, There are l training set sample pairs. It is the input column vector of the i-th training sample. That is the corresponding output value. It is the predicted value returned by the regression function. It is a nonlinear mapping function, w is the weight vector, b is a constant, and y is... The corresponding actual value, It is the error requirement of the regression function. This is a relaxation variable introduced to find the optimal hyperplane; c is the penalty factor. It is a Lagrange multiplier. It's a kernel function. It is a radial basis kernel function. is the width of the radial basis function kernel, and g is a parameter in the radial basis function kernel.
[0044] Furthermore, the specific calculation formula for the coupling correction algorithm in S7 is as follows:
[0045]
[0046]
[0047] In the formula, It is the coupling correction coefficient for the i-th sample. It is the post-reaction strength of the i-th coke sample from the actual coke production. It is the predicted value of the intensity nonlinear support vector regression after the reaction of the i-th coke oven sample. These are the coefficients of the regression equation. These are the weights of the indicators corresponding to the blended coal in the regression equation. It is a constant in the regression equation. It is the predicted value of the intensity of coke after reaction, which is the final output of the model.
[0048] Compared with the prior art, the beneficial effects of the present invention are:
[0049] 1. The intelligent coal blending and coking method based on the kernel mapping and coupling correction algorithm of the present invention effectively solves the problems of large quality fluctuations, high costs and low energy efficiency caused by the long-term reliance on manual experience in coal blending in the coking industry.
[0050] 2. The intelligent coal blending and coking method based on the kernel mapping coupling correction algorithm of the present invention uses a nonlinear SVR model to accurately characterize the complex mapping relationship between coal quality characteristics and coke quality and production cost, and realizes quantitative decision-making and optimization control of the coal blending process.
[0051] 3. The intelligent coal blending and coking method based on the kernel mapping coupling correction algorithm of the present invention significantly reduces raw material costs, improves production efficiency and environmental protection level while ensuring stable and qualified coke quality. It promotes the intelligent transformation of coking production from "experience-driven" to "data-driven" and has important engineering application value and industry promotion significance. Attached Figure Description
[0052] Figure 1 This is a logical framework diagram of the intelligent coal blending and coking method of the present invention. Detailed Implementation
[0053] The technical solutions of 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.
[0054] Please see Figure 1 The present invention provides an intelligent coal blending and coking method based on a kernel mapping coupling correction algorithm, comprising the following steps:
[0055] S1: Collect data from the entire coking process to form a dataset, including: industrial analysis data for single coal types, G-value, Y-value, average maximum reflectance of vitrinite, active-inert ratio, proportion of each coal type, blending coal indicators, blending coal fineness, target quality indicators, and production costs. Target quality indicators include coke reactivity, post-reaction strength, sulfur content, coke shatter resistance, abrasion resistance, and lump coke ratio. Production costs include raw coal costs, heating fuel costs, and environmental and carbon emission costs. Specifically, historical data covering the entire coking process will be collected from the enterprise production management system (MES), laboratory information management system (LIMS), and distributed control system (DCS).
[0056] S2: Divide the above dataset into a 70:15:15 ratio: 70% training set, 15% test set, and 15% validation set. The training set is used to fit the parameters of the support vector regression model, allowing the model to learn from the data and capture the inherent mapping between input features and the target variable; it is also used for tuning hyperparameters (penalty coefficient, insensitive loss parameter, kernel coefficient). The test set is used solely for tuning the coupling correction algorithm, establishing the correction algorithm based on the model's prediction bias on unknown data to avoid data leakage. The validation set is used only for the final, one-time performance evaluation to confirm the generalization ability of the entire system and prevent overfitting. It is strictly forbidden to use the validation set to establish the correction algorithm, and it is also forbidden to use the test set or validation set for training or tuning in any form.
[0057] S3: RobustScaler standardization is applied to continuous features selected from the training set, using the median and quartile ranges for feature scaling. This makes the feature highly robust to outliers and extreme values in the data, resulting in a more reliable and representative feature distribution. The same parameters are used to transform the test and validation sets to ensure data distribution consistency. The specific calculation formula for RobustScaler standardization is as follows:
[0058] X [x1,x2,x3,...,x n ] R l×n
[0059]
[0060]
[0061]
[0062] Y [y1,y2,y3,...,y n ] R l×n
[0063] In the formula, X is the set of raw coal characteristics and coal blending data. The raw coal characteristics are obtained based on the coal quality analysis data of each individual coal, and the coal blending data and process parameters are all derived from actual production historical data; n is the number of indicators contained in x, and l is the number of individual coals in the sets X and Y. It is the position of the median in the dataset. It is the median of dataset X. It is the interquartile range. It is the lower quantile. It is the upper quantile, y i Y is the RobustScaler-standardized data for the i-th indicator, and Y is the RobustScaler-standardized analysis dataset; where the lower quantile is the position of the variable values arranged in ascending order. The number, the upper quantile is the position. The number.
[0064] S4: Clean the basic data in S3, including: using KNN imputation to handle outliers and missing data; KNN imputation, based on the assumption that "similar samples have similar attributes," intelligently estimates missing values using the overall similarity between data points, thereby preserving the distribution pattern and correlation information between variables in the original dataset to the greatest extent. Feature fields or sample records with missing rates exceeding a preset threshold are directly deleted to ensure dataset quality; the specific calculation formula is as follows:
[0065]
[0066]
[0067]
[0068]
[0069]
[0070]
[0071]
[0072]
[0073]
[0074]
[0075] In the formula, It is the sample missing rate. It is the number of features in the data. It is the first The number of missing features in the nth sample, the number of missing features in the dataset. Features and the j-th feature Mutual information between them is defined as , It is an indicator that measures the importance of different features among all features. It is a weight matrix. These are the weights obtained after feature normalization. Defined as Euclidean distance, It is the K samples obtained and the samples to be filled. The matrix of time distances in ascending order. To utilize K samples and the sample to be filled The distance between these K samples reflects the relationship between them and the K samples. The weights used to measure sample similarity are defined by the degree of similarity. This is to prevent the weight from being a very small number with a denominator of 0. It is the corrected weight. The sample to be filled is obtained by weighted summation based on weights. The imputed value of the k-th feature.
[0076] S5: Nonlinear Support Vector Regression Modeling, including: constructing a nonlinear SVR model on the training set; establishing and training a nonlinear SVR model for each target quality indicator and production cost; using a Gaussian radial basis function (RBF) kernel to map the data to a high-dimensional space; and searching for a linear regression model in the high-dimensional space. The nonlinear SVR model using the RBF kernel involves implicitly mapping the original feature space to a high-dimensional or even infinite-dimensional feature space through support vector machine regression. In this high-dimensional space, the originally complex and inseparable nonlinear relationships become linear and separable. This allows SVR to find an optimal linear regression hyperplane in this new space, which corresponds to a complex nonlinear function in the original space.
[0077] S6: Hyperparameter optimization, including: directly calling the GridSearchCV module in the scikit-learn library to optimize the core hyperparameters of each SVR model, such as the penalty coefficient C, the insensitive loss parameter ε, and the kernel coefficient γ, using grid search and cross-validation to determine the optimal parameter configuration; among which, hyperparameter tuning includes: transforming and customizing theoretically high-potential models (nonlinear SVR) into a high-performance, high-reliability dedicated prediction engine that is highly adapted to specific production data, achieving a leap from "possible" to "optimal" model performance.
[0078] In the above steps, the specific calculation formulas for nonlinear support vector regression and hyperparameter optimization are as follows:
[0079]
[0080]
[0081]
[0082]
[0083]
[0084]
[0085]
[0086] In the formula, There are l training set sample pairs. It is the input column vector of the i-th training sample. That is the corresponding output value. It is the predicted value returned by the regression function. It is a nonlinear mapping function, w is the weight vector, b is a constant, and y is... The corresponding actual value, It is the error requirement of the regression function. This is a relaxation variable introduced to find the optimal hyperplane; c is the penalty factor. It is a Lagrange multiplier. It's a kernel function. It is a radial basis kernel function. is the width of the radial basis function kernel, and g is a parameter in the radial basis function kernel.
[0087] S7: Model testing and algorithm coupling correction, including: For the test set, predictions are made based on the optimal hyperparameter model and compared with actual production data; a coupling correction algorithm is constructed based on the comparison results to compensate for systematic errors; specifically, after hyperparameter tuning, the obtained optimal parameter combination is assigned to the model architecture, and predictions are made on the test set. The prediction results are compared with the corresponding actual production results on the test set to analyze their systematic biases and error distribution. Based on this analysis, a coupling correction algorithm is constructed. The correction algorithm aims to compensate for the inherent biases exhibited by the model on the test set, thereby improving the model's industrial applicability. The specific calculation formula of the coupling correction algorithm is as follows:
[0088]
[0089]
[0090] In the formula, It is the coupling correction coefficient for the i-th sample. It is the post-reaction strength of the i-th coke sample from the actual coke production. It is the predicted value of the intensity nonlinear support vector regression after the reaction of the i-th coke oven sample. These are the coefficients of the regression equation. These are the weights of the indicators corresponding to the blended coal in the regression equation. It is a constant in the regression equation. It is the predicted value of the intensity of coke after reaction, which is the final output of the model.
[0091] S8: Model Validation and Optimal Interval Determination, including: final validation of the optimal parameter model calibrated by the coupling correction algorithm on the validation set to evaluate its prediction accuracy and robustness; then, using the multi-objective optimization algorithm NSGA-II to determine the production interval where coke quality indicators and production costs are optimally synergistically, providing quantitative decision support for coal blending schemes. Specifically, the fixed hyperparameter model is combined with the determined coupling correction algorithm to form a complete prediction system, which is then finally tested on the validation set. The performance indicators obtained on the validation set represent the final performance of the method of this invention, demonstrating its excellent generalization ability and industrial application value.
[0092] Based on the above embodiments, this invention provides an intelligent coal blending and coking method based on a kernel mapping coupling correction algorithm. Through RobustScaler standardization, scaling is performed using median and quartile ranges, making it highly robust to outliers and extreme values in the data, resulting in a more reliable and representative feature distribution. KNN interpolation is used to handle outliers and missing data. Based on the assumption that "similar samples have similar attributes," the overall similarity between data points is used to intelligently estimate missing values, thereby preserving the distribution patterns and correlation information between variables in the original dataset to the greatest extent. Furthermore, a nonlinear SVR model with an RBF kernel is selected: Support Vector Machine regression implicitly maps the original feature space to a high-dimensional or even infinite-dimensional feature space. In this high-dimensional space, the originally complex and inseparable nonlinear relationships become linear and separable relationships, enabling SVR to find an optimal linear regression hyperplane in this new space. This hyperplane corresponds to a complex nonlinear function in the original space. This invention performs final validation on a validation set on the optimal parameter model calibrated by the coupling correction algorithm to evaluate its prediction accuracy and robustness. Based on this, the multi-objective optimization algorithm NSGA-II is used to determine the production range in which coke quality indicators and production costs are optimally coordinated, providing quantitative decision support for coal blending schemes.
[0093] In summary, the intelligent coal blending and coking method based on the kernel mapping coupling correction algorithm of this invention establishes a database covering raw coal characteristics, blending costs, and coke quality indicators through full-process data collection, thereby achieving unified management and feature extraction of multi-source data. Secondly, by dividing the data into training, testing, and validation sets, the model is trained and calibrated step-by-step, and the prediction results are dynamically optimized using correction coefficients, ultimately obtaining a highly adaptable and accurate prediction model. Furthermore, based on a large amount of actual production data from enterprises, the model outputs coke quality results, enabling enterprises to adjust coal blending schemes and coke oven process parameters in real time, achieving intelligent coal selection, stable coke quality, and optimized energy utilization efficiency. This effectively reduces production costs, improves coke performance, and supports the green and low-carbon development of coking enterprises.
[0094] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A smart coal blending and coking method based on a kernel mapping coupling correction algorithm, characterized in that, Includes the following steps: S1: Collect data from the entire coking process to form a dataset, including: industrial analysis data for single coal types, G-value, Y-value, average maximum reflectance of vitrinite, active-inert ratio, proportion of each coal type, blended coal indicators, fineness of blended coal, target quality indicators, and production costs. S2: Divide the dataset in S1 into a training set (70%), a test set (15%), and a validation set (15%). The training set is used to fit the parameters of the support vector regression model, allowing the model to learn from the data and capture the inherent mapping between input features and target variables, which is used for hyperparameter tuning. The test set is used to tune the coupling correction algorithm, which is established based on the model's prediction bias on unknown data. The validation set is used for the final, one-time performance evaluation. S3: RobustScaler standardizes the continuous features of the training set in S2, using the median and quartile ranges for feature scaling; the same parameters are used to transform the test and validation sets to ensure data distribution consistency. S4: Clean the basic data in S3, including: using KNN interpolation to handle abnormal and missing data; directly deleting feature fields or sample records with missing rates exceeding a preset threshold to ensure dataset quality; S5: Nonlinear Support Vector Regression Modeling, including: constructing a nonlinear SVR model on the training set, establishing and training a nonlinear SVR model for each target quality indicator and production cost; using the Gaussian radial basis function (RBF) to map the data to a high-dimensional space, and finding a linear regression model in the high-dimensional space; S6: Hyperparameter optimization, including: directly calling the GridSearchCV module in the scikit-learn library on the training set to use grid search and cross-validation to fine-tune the combination of the core hyperparameters of each SVR model, such as the penalty coefficient C, the insensitive loss parameter ε, and the kernel coefficient γ, and determine the optimal parameter configuration; S7: Model testing and algorithm coupling correction, including: for the test set, making predictions based on the optimal hyperparameter model and comparing them with actual production data; constructing a coupling correction algorithm based on the comparison results to compensate for system errors and improve the model's prediction accuracy and adaptability to operating conditions; S8: Model validation and optimal interval determination, including: performing final validation of the optimal parameter model after calibration by the coupling correction algorithm on the validation set, evaluating its prediction accuracy and robustness, and then using the multi-objective optimization algorithm NSGA-II to determine the production interval where coke quality indicators and production costs are optimally coordinated, providing quantitative decision support for coal blending schemes.
2. The intelligent coal blending and coking method based on the kernel mapping coupling correction algorithm as described in claim 1, characterized in that: The target quality indicators in S1 include coke reactivity, coke post-reaction strength, coke sulfur content, coke crush resistance index, coke abrasion resistance index, and lump coke ratio; production costs include raw coal cost, heating fuel cost, and environmental protection and carbon emission cost.
3. The intelligent coal blending and coking method based on the kernel mapping coupling correction algorithm as described in claim 1, characterized in that: The specific calculation formula for RobustScaler standardization in S3 is as follows: X [x1,x2,x3,...,x n ] R l×n Y [y1,y2,y3,...,y n ] R l×n In the formula, X is the set of raw coal characteristics and coal blending data. The raw coal characteristics are obtained based on the coal quality analysis data of each individual coal, and the coal blending data and process parameters are all derived from actual production historical data; n is the number of indicators contained in x, and l is the number of individual coals in the sets X and Y. It is the position of the median in the dataset. It is the median of dataset X. It is the interquartile range. It is the lower quantile. It is the upper quantile, y i Y is the RobustScaler-standardized data for the i-th indicator, and Y is the RobustScaler-standardized analysis dataset; where the lower quantile is the position of the variable values arranged in ascending order. The number, the upper quantile is the position. The number.
4. The intelligent coal blending and coking method based on the kernel mapping coupling correction algorithm as described in claim 1, characterized in that: The KNN imputation method in S4 includes: based on the assumption that "similar samples have similar attributes", it intelligently estimates missing values by utilizing the overall similarity between data points. The specific calculation formula is as follows: In the formula, It is the sample missing rate. It is the number of features in the data. It is the first The number of missing features in the nth sample, the number of missing features in the dataset. Features and the j-th feature Mutual information between them is defined as , It is an indicator that measures the importance of different features among all features. It is a weight matrix. These are the weights obtained after feature normalization. Defined as Euclidean distance, It is the K samples obtained and the samples to be filled. The matrix of time distances in ascending order. To utilize K samples and the sample to be filled The distance between these K samples reflects the relationship between them and the K samples. The weights used to measure sample similarity are defined by the degree of similarity. This is to prevent the weight from being a very small number with a denominator of 0. It is the corrected weight. The sample to be filled is obtained by weighted summation based on weights. The imputed value of the k-th feature.
5. The intelligent coal blending and coking method based on the kernel mapping coupling correction algorithm as described in claim 1, characterized in that: The specific calculation formulas for nonlinear support vector regression and hyperparameter optimization in S5-S6 are as follows: In the formula, There are l training set sample pairs. It is the input column vector of the i-th training sample. That is the corresponding output value. It is the predicted value returned by the regression function. It is a nonlinear mapping function, w is the weight vector, b is a constant, and y is... The corresponding actual value, It is the error requirement of the regression function. This is a relaxation variable introduced to find the optimal hyperplane; c is the penalty factor. It is a Lagrange multiplier. It's a kernel function. It is a radial basis kernel function. is the width of the radial basis function kernel, and g is a parameter in the radial basis function kernel.
6. The intelligent coal blending and coking method based on the kernel mapping coupling correction algorithm as described in claim 1, characterized in that: The specific calculation formula for the coupling correction algorithm in S7 is as follows: In the formula, It is the coupling correction coefficient for the i-th sample. It is the post-reaction strength of the i-th coke sample from the actual coke production. It is the predicted value of the intensity nonlinear support vector regression after the reaction of the i-th coke oven sample. These are the coefficients of the regression equation. These are the weights of the indicators corresponding to the blended coal in the regression equation. It is a constant in the regression equation. It is the predicted value of the intensity of coke after reaction, which is the final output of the model.