Coke quality parallel prediction method and device for simulating coking mechanism

Abstract

A coke quality parallel prediction method and device simulating coking mechanism, which is based on actual coking production data, simulates coking mechanism by using data mining technology, combines the advantages of each coking mechanism, and establishes a coke quality prediction method and system with strong interpretability, thereby providing effective guidance for coking coal blending. The method comprises the following steps: 1: data collection; 2: data matching; 3: data conversion; 4: data cleaning; 5: feature analysis; 6: data normalization; 7: data mining model construction. Learners simulate the coking process from different angles, and the trained learners are combined in parallel. Each learner has an independent output, so that the final result can take the advantages and make up for the shortcomings, learn a stable model that performs well in all aspects, and generate a more powerful model through combined output, thereby greatly improving the model precision and generalization ability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coking process, and particularly relates to a coke quality parallel prediction method and device simulating coking mechanism. BACKGROUND

[0002] The coking coal blending theory is established on the basis of coal chemistry, coking science, coal petrology and physical chemistry. In long-term production practice, three coking mechanisms, i.e. plastic coking, surface bonding coking and mesophase coking, are formed. The three coking mechanisms represent the coking process of coking coal from different dimensions.

[0003] In the actual coking production process of a coking enterprise, a coal blending engineer needs to select a coal blending technology based on one of the coking mechanisms to predict coke quality according to coke quality requirements, coking coal resource conditions and other related constraints, and to formulate and improve the coking coal blending scheme. However, the coal blending process has the following problems:

[0004] (1) The coking coal sources are complex, the properties are complex, and the existence of mixed coal leads to large fluctuations in the quality of coking coal and extremely complex characterization of the properties of coking coal.

[0005] (2) There are many characterization indexes of coking coal, and the indexes have poor additivity. The coke quality indexes are not all linear responses, which leads to low precision of the coke quality prediction model, and small coke oven experiments are often needed to verify the model to guide industrial production.

[0006] (3) Since there are many factors and constraints involved in coking coal blending, the calculation of coking coal blending is complex and the workload is large. It is difficult to control coke quality by using the traditional coal blending method, and the overall optimization cannot be achieved. The cost of coal blending is high, and the scarce coking and fertilizer coal resources are wasted to a certain extent.

[0007] (4) The verification period of the coal blending process is long, the coal blending efficiency is low, and the coke quality is unstable and the indexes fluctuate greatly, which cannot meet the quality requirements of large blast furnaces for coke. In addition, the supply of coal sources changes dramatically and the quality fluctuates, which leads to frequent changes in the coal blending scheme and slow response of the coal blending to production.

[0008] The Chinese patent with publication number CN112861359B discloses a full-factor intelligent coal blending system. The system includes: a single coal coking prediction module for predicting the thermal strength of coke under independent coking of single coal; a coal source cost-effectiveness sorting module for selecting the optimal solution in terms of cost-effectiveness under current conditions for all coal sources to be sorted under certain quality complementarity and price support on the premise of setting a target quality, and determining the cost-effectiveness ranking thereof according to the proportion of the solution; a coke quality positive prediction module for calculating a given coal blending ratio and accurately predicting coke cold strength and coke thermal strength for different furnace types and different processes; a coal blending optimization module for optimizing the ratio of the lowest coal-in-furnace cost according to the given coke cold strength and coke thermal strength; and a product yield prediction module for predicting the ton of coke unit yield and coke particle size grading ratio. The invention mainly focuses on the coking characteristics of single coal, the cost-effectiveness of coal sources, and the use of specific formulas to predict quality indicators such as M40 and CSR of coke, and does not involve the use of data mining methods.

[0009] The Chinese patent with publication number CN114692986A discloses a neural network model intelligent coal blending system based on historical coal blending data and corresponding coke detection results. The invention mainly focuses on the use of neural network models in coking coal blending.

[0010] The Chinese patent with publication number CN111950854B discloses a coke quality indicator prediction method based on a multi-layer neural network. The method uses industrial actual production data to first clean the data, uses gradient boosting trees to analyze the correlation of factors affecting coke quality indicators, selects the most relevant variables to ash content, sulfur content, M10, M40, CRI, and CSR, and then constructs training samples, establishes a multi-layer neural network prediction model to predict coke quality indicators, and uses intelligent optimization algorithms to optimize variables in the model to give the final coke quality indicator prediction result. The invention uses gradient boosting trees for feature analysis and establishes a multi-layer neural network to predict coke quality, and does not involve the application of coking mechanism.

[0011] In the process of industrial coal blending practice, based on the coking theory of coking coal blending, the coke quality is quickly and accurately predicted, and the appropriate coal blending scheme is optimized, which not only greatly improves the work efficiency, but also realizes the purpose of stable coke quality.

[0012] The rapid development of computer technology and algorithms provides a new idea for coal blending optimization and accurate prediction of coke quality. Data mining coal blending is a combination of computer technology, artificial intelligence, statistics and modern coking technology. The mechanism of coal blending for coking, the experience of coal blending experts and the artificial intelligence technology are combined. The optimal coal blending scheme is calculated by using the storage capacity, data processing and advanced algorithm of the computer through data mining. Under the premise of ensuring the quality of coke, the coal resources are reasonably utilized, the high-value coking coal is saved, and the coking coal resources are expanded, which is beneficial to the coking enterprises to save production cost, improve production efficiency, realize the quality and cost reduction of coke production, and is the development direction of the optimization of coal blending technology. SUMMARY

[0013] In order to solve the technical problems in the background art, the present application provides a coke quality parallel prediction method and device simulating coking mechanism, which starts from actual coking production data, simulates coking mechanism by data mining technology, combines the advantages of each coking mechanism, establishes a coke quality prediction method and system with strong interpretability, provides effective guidance for coking coal blending, improves the accuracy and speed of coke quality prediction, and improves the adaptability of the model.

[0014] In order to achieve the above purpose, the present application adopts the following technical scheme:

[0015] A coke quality parallel prediction method simulating coking mechanism, comprising the following steps:

[0016] Step 1: data acquisition

[0017] According to the characteristics of coking coal blending and coking production data, the actual coking production data of the coking enterprise is collected, the coking production process data affecting the coke quality is analyzed and extracted from the quality report and production process report of the coking enterprise, including single coal, blended coal, coke and coke oven thermal data.

[0018] Step 2: data matching, constructing initial coal coke data set

[0019] Based on the time sequence relationship of coking production, the single coal quality characteristics, blended coal quality characteristics, coal blending ratio, coke quality index and coke oven thermal data are matched, and the initial coal coke data set is constructed.

[0020] Step 3: data conversion

[0021] The analysis basis of various detection data of coal and coke is different, the coal coke detection data is converted into a relatively stable analysis basis, including: total moisture is expressed by received basis, ash and sulfur are expressed by dry basis, volatile matter yield is expressed by dry ash-free basis.

[0022] Step 4: data cleaning

[0023] Based on the requirement of coke quality prediction on data quality, the data cleaning rules and data cleaning scheme are established by analyzing the data category and data characteristics of industrial production. The original data is comprehensively detected to detect missing data, irrelevant / redundant data, abnormal / error data, confirm the nature and location of the problem data, and obtain comprehensive data quality information. The data is corrected by filling in missing data, merging duplicate data, removing redundant data, and detecting abnormal / error data.

[0024] Step 5: Feature analysis

[0025] Feature analysis eliminates unnecessary or redundant features, selects features that best fit the modeling problem, reduces model size, and prevents overfitting. Feature analysis methods include stepwise regression, sequential feature selection, regularization, and nearest neighbor analysis.

[0026] Step 6: Data normalization

[0027] Partially based on bias machine learning algorithms, the coke quality indicators such as ash content, sulfur content, volatile matter, G value, Y value, and coke quality indicators have high dimension and large range. In order to eliminate the dimensional influence between different features and ensure the comparability of data, normalization is used for feature transformation.

[0028] Step 7: Data mining model construction

[0029] A high-quality coal coke data set that meets the requirements of data mining is established, and a coke quality prediction model simulating the coking mechanism is established by using the rule sampling-learner construction-model combination method. It includes the following:

[0030] 7.1 Rule sampling

[0031] The initial coal coke data set D is randomly divided into k data sets D1, D2, …, Di, Di+1, …, Dk of the same size by resampling method. k j and D (-j) are the training set and test set corresponding to the jth execution.

[0032] 7.2 Learner construction

[0033] i independent learners base learner are constructed, represented by learner 1, learner 2, …, learner i respectively, simulating i coking mechanisms; for a given input x, each learner h i tries to approximate the distribution of the original data y and obtains a local output h i (y|x; θ i ); where θ i is the parameter of the i-th learner h i . ​

[0034] Each subset trains its own learner, that is, learns a model on each dataset; by changing the weights of similar samples, the learner learns the optimal model on each dataset.

[0035] 7.3 Model Combination

[0036] The learners simulate the focusing process from different angles and are combined in parallel. Each learner has an independent output, so that the final result can complement each other and learn a stable model that performs well in all aspects. By combining the outputs, a more powerful model is generated, and the model accuracy and generalization ability are greatly improved.

[0037] The model combination method of parallel learners is as follows: the learners are arranged in parallel as plastic focusing learners, surface bonding focusing learners and mesophase focusing learners.

[0038] Furthermore, in step 4, the data correction includes detecting abnormal / erroneous data, as detailed below:

[0039] Statistical methods, distance-based methods, and machine learning models are used to find outlier data that is outside the range of the vast majority of data, and to determine whether outlier data is erroneous or an anomaly.

[0040] Outliers in coking coal blending production data can affect model construction. It is necessary to distinguish whether outliers are due to data anomalies or coal quality anomalies. Outlier analysis methods in coking coal blending production data are divided into two categories of criteria: physical and statistical. Physical criteria remove outliers from the original data that do not meet the criteria; statistical criteria remove data that are not statistically normal.

[0041] ①Statistical criteria for exclusion

[0042] When the operating conditions of the coking process are fixed, the production data generally follow a normal distribution. According to the Schauville criterion, abnormal data are processed, that is, data with a deviation greater than σ are removed, where σ is the standard deviation of the production data. The formula for calculating σ is shown in equation (1).

[0043]

[0044] ② Physical criteria elimination

[0045] In general coking plants, the volatile matter V in coke is... d ≤1.5%, volatile matter V of blended coal daf Generally, it is controlled at 23% to 29%; the formula for calculating the coking rate K is shown in equation (2):

[0046]

[0047] Wherein: a is the correction coefficient of coking rate K, which is related to the volatile matter of coal, furnace structure and operation system, and is usually 1.5%;

[0048] Under normal production conditions, no green coke appears, and the coking rate K is between 73.2% and 79.3%. As the elimination criterion of the coal coke data, a is between 74% and 80%, and the data is eliminated.

[0049] Further, the step 5 specifically includes the following:

[0050] Pearson correlation coefficient is used to select the features with good data independence and large correlation with the coke quality indicators; according to the Pearson correlation calculation formula, the correlation P j,X between two continuous variables j and X is:

[0051]

[0052] Wherein, P j,X is the Pearson correlation coefficient, σ j , σ X are the standard deviations of the characteristic variables j and the coke quality X, μ j , μ K are the expected values of j and X, and cov(j, X) is the covariance of j and X.

[0053] The value range of Pearson correlation coefficient P is (-1, 1), P>0 indicates positive correlation, P<0 indicates negative correlation, and P=0 indicates zero correlation; the greater the absolute value of P is, the higher the correlation degree is, |P| = 0.8-1.0 is extremely strong correlation, 0.6-0.8 is strong correlation, 0.4-0.6 is moderate correlation, 0.2-0.4 is weak correlation, and 0.0-0.2 is extremely weak correlation or no correlation.

[0054] Further, the step 6 is specifically: normalizing the data to [0, 1], and normalizing the variables by using formula (4):

[0055]

[0056] Wherein: V is the characteristic value of the data, V max and V min are the maximum value and the minimum value of each feature in each group of samples, respectively, and V i represents the normalized characteristic value.

[0057] Further, the step 7 further includes the following:

[0058] A weight function π i (x; a) is defined to measure each learner h iThe contribution of each learner. Here, α is the parameter of the weight function; the final output of the model is represented by the contribution of each learner h. i The weighted output is shown in equation (5):

[0059] H(y|x)=π1(x;α)×h1(y|x;θ1)+π2(x;α)×h2(y|x;θ2)+……+π i (x;α)×h i (y|x;θ i ... Equation (5).

[0060] Furthermore, step 7 also includes the following:

[0061] The input indicators for the learner are the coal quality characteristics of a single type of coal, the coal quality characteristics of a blended coal, and the thermal data of a coke oven. The output indicator is the quality index of coke.

[0062] The coal quality characteristics of a single type of coal include ash content, sulfur content, volatile matter, caking index, maximum thickness of the plastic layer (Y value), maximum shrinkage of the plastic layer (X value), maximum Gibbs freeness, plastic range, initial flow temperature, maximum flow temperature, final flow temperature, solidification temperature, maximum expansion (b value), maximum shrinkage (a value), solid-soft range, coal ash composition: SiO2, Al2O3, Fe2O3, TiO2, CaO, MgO, K2O, Na2O, SO3, MnO2, P2O5, and vitrinite reflectance. The standard deviation of reflectance (s), the vitrinite reflectance range (V3-V21), the content of active components, the content of inert components, and other relevant indicators.

[0063] The coal quality characteristics of blended coal include ash content, sulfur content, volatile matter, caking index, plastic layer index, and other relevant indicators.

[0064] The thermal data of a coke oven include coking temperature and coking time.

[0065] The quality index of coke, i.e., the output index, is the ash content (A) of coke. d Total sulfur S t,d , fracture strength M 40 Abrasion resistance M 10 Thermal reactivity (CRI) and thermal post-reaction strength (CSR).

[0066] Furthermore, step 7 also includes the following:

[0067] The data mining algorithms used in the training of the learner include one or more of the following models: least squares linear regression, ridge regression, LASSO regression, support vector machine (SVR), random forest (RF), decision tree (ID3, C4.5, C5.0 and CART), logistic regression, and neural network.

[0068] The data mining algorithm in each learner is completely same or completely different, the learning rate η of the machine learning algorithm is between 0.01-0.5, the maximum depth μ is between 3-10, the penalty function parameter γ is between 0-0.4, the cyclic iteration training number r is between 100-5000, the regularization penalty coefficient λ is between 0-5, and the training process is automatically terminated and the model parameters of each learner are solidified after reaching the maximum iteration period.

[0069] Further, it further comprises:

[0070] Step 8: model evaluation

[0071] The stratified sampling method is adopted to divide the verification set, so that the verification set and the training set have similar category proportions.

[0072] The model evaluation parameters adopt correlation coefficient R 2 , mean square error MSE, mean absolute error MAE, mean absolute percentage error MAPE and the like, which are respectively shown in formula (6)-formula (9).

[0073]

[0074]

[0075]

[0076]

[0077] Wherein, y is an actual value, is a predicted value.

[0078] The application further provides a device for implementing the coke quality parallel prediction method of the simulated coking mechanism, which comprises a processor and a memory connected with the processor, wherein the processor is configured to execute the coke quality parallel prediction method of the simulated coking mechanism, and the memory is used for storing executable instructions of the processor.

[0079] The application further provides a computer storage medium, which stores a computer program, and the computer program is executed by a processor to implement the coke quality parallel prediction method of the simulated coking mechanism.

[0080] Compared with the prior art, the application has the beneficial effects that:

[0081] The present application can better utilize the coal coke professional knowledge, and the historical data of single coal quality, blending coal quality, coke quality, coking process parameters and the like generated in the coking industrial production are established, a coke quality prediction model with high stability and precision is established, which can be used for coke quality online prediction and coal blending monitoring in long-period industrial production, and has important significance for digital transformation of coking enterprises. Different learningers simulate different coal blending mechanisms, each learninger has strong independence, and the difference between learningers is large, so that the final result can complement each other, and the advantages are complementary; a stable and good model in various aspects is learned by using the parallel combination of independent learningers, the final result is outputted through combination, the error can be significantly reduced, and the model precision and generalization ability can be greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0082] Figure 1 The technical flow chart of the coke quality prediction based on the data mining technology simulation coking mechanism of the present application is shown in the figure.

[0083] Figure 2 The specific process framework diagram of the data cleaning of the present application is shown in the figure.

[0084] Figure 3 The combination mode between learningers of the present application is shown in the figure.

[0085] Figure 4 The technical flow chart of the model evaluation of the present application is shown in the figure.

[0086] Figure 5 The coke quality prediction system diagram of the simulation coking mechanism of the present application is shown in the figure.

[0087] Figure 6 The data matching diagram based on the timing relationship of the coking production of the present application is shown in the figure.

[0088] Figure 7 The coke crushing strength (M40) schematic diagram of the physical criterion of the present application is shown in the figure.

[0089] Figure 8 The correlation thermodynamic diagram of the blending coal data of the present application is shown in the figure.

[0090] Figure 9 The 5-fold cross-validation schematic diagram of the present application is shown in the figure.

[0091] Figure 10 The model accuracy trend graph with the change of iteration number of the present application is shown in the figure.

[0092] Figure 11 The model accuracy trend graph with the change of sample number of the present application is shown in the figure. DETAILED DESCRIPTION

[0093] The specific embodiments of the present application are further described below with reference to the accompanying drawings:

[0094] As Figure 1 shown, a coke quality parallel prediction method of a simulated coking mechanism of the present application comprises the following steps:

[0095] 1Data acquisition

[0096] According to the characteristics of coking blending coal and coking production data, the actual coking production data of the coking enterprise is collected, and the coking production process data affecting coke quality is analyzed and extracted from the quality report and production process report of the coking enterprise, including single coal, blended coal, coke and coke oven thermal data, etc.

[0097] 2Data matching

[0098] Based on the timing relationship of coking production, the single coal quality characteristics, blended coal quality characteristics, coal blending ratio, coke quality index and coke oven thermal data are matched, and an initial coal coke data set is constructed.

[0099] 3Data conversion

[0100] The analysis basis of various detection data of coal and coke is different, and the coal coke detection data is converted to a relatively stable analysis basis, such as total moisture expressed as received basis, ash and sulfur expressed as dry basis, and volatile matter yield expressed as dry ash-free basis.

[0101] 4Data cleaning

[0102] As Figure 2 shown, the main process of data cleaning is as follows:

[0103] 4.1Analysis data: based on the demand of coke quality prediction for data quality, analyze the category and characteristics of industrial production data, establish data cleaning rules and data cleaning scheme.

[0104] 4.2Detection data: comprehensively detect the original data, detect missing data, irrelevant / redundant data, abnormal / error data, confirm the nature and location of the problem data, and obtain comprehensive data quality information.

[0105] 4.3Correct the data: fill in the missing data, merge the repeated data, remove the redundant data, and detect the abnormal / error data.

[0106] 1)Fill in the missing data

[0107] The missing data is reasonably filled by interpolation method or modeling method.

[0108] The interpolation method includes but is not limited to mean interpolation method, median interpolation method, random interpolation method, multiple interpolation method, Lagrange interpolation method and Newton interpolation method, etc.

[0109] Modeling methods include, but are not limited to, regression, Bayesian, decision tree, probabilistic statistical modeling, and deep learning methods.

[0110] 2) Merge duplicate data

[0111] Duplicate data is handled using a matching and merging method. Records in the dataset are sorted according to certain rules, and duplicate records are detected by comparing their similarity to neighboring records. Duplicate data is then merged.

[0112] 3) Delete redundant data

[0113] Redundant data is deleted directly.

[0114] 4) Detect abnormal / erroneous data

[0115] Statistical methods, distance-based methods, and machine learning models are used to find outlier data that is outside the range of the vast majority of data, and to determine whether outliers are erroneous or isolated data.

[0116] Statistical detection methods include, but are not limited to, those based on normal distribution and multivariate Gaussian distribution. These methods involve setting a confidence level for the data to be identified and using the confidence level to determine whether it is abnormal.

[0117] Distance-based detection methods include, but are not limited to, those based on Mahalanobis distance, Euclidean distance, standardized Euclidean distance, Manhattan distance, Chebyshev distance, Jaccard distance, information entropy, correlation distance, and cosine similarity.

[0118] Detection methods based on machine learning algorithms include, but are not limited to, decision trees, PCA clustering, random forests, etc., which randomly select features from the feature set and choose the segmentation value with the shortest path between the maximum and minimum values ​​of the features.

[0119] Outliers in coking coal blending production data can impact model construction. It's crucial to distinguish whether outliers are due to data anomalies or coal quality anomalies. Outlier analysis methods in coking coal blending production data fall into two categories: physical and statistical criteria. Physical criteria remove outliers from the original data that do not conform to the criteria (data anomalies). Statistical criteria remove data that is statistically abnormal.

[0120] ③ Statistical criteria elimination

[0121] When the operating conditions of the coking process are fixed, the production data generally follow a normal distribution. Outlier data is handled according to the Chauvenet Criterion, i.e., data with a deviation greater than σ are removed, where σ is the standard deviation of the production data. The formula for calculating σ is shown in equation (1).

[0122]

[0123] ④ Physical criteria elimination

[0124] The volatile matter (V) in coke from general coking plants d ≤1.5%, volatile matter (V) of blended coal daf The coking rate (K) is generally controlled between 23% and 29%. The formula for calculating the coking rate (K) is shown in equation (2).

[0125]

[0126] In the formula: a is the correction coefficient for coking rate K, which is related to coal volatile matter, furnace structure and operating system, and is usually 1.5%.

[0127] Under normal production conditions (without coking), the coking rate (K) is between 73.2% and 79.3%. The data is removed based on the criterion that K ≤ α (α is between 74% and 80%).

[0128] 4.4 Validation Data

[0129] Verify the corrected data. If the verification results do not match the task objectives, further corrections are required.

[0130] 5 Feature Analysis

[0131] Feature analysis can eliminate unnecessary or redundant features, select the features that best fit the modeling problem, reduce model size, and prevent overfitting.

[0132] Feature analysis methods include, but are not limited to, stepwise regression, sequence feature selection, regularization, and nearest neighbor analysis. For example, Pearson correlation coefficients (PCCs) can be used to select features that are highly correlated with coke quality indicators and have good data independence. According to the Pearson correlation formula, the correlation P between two continuous variables j and X is... j,X for:

[0133]

[0134] Among them, P j,X Let σ be the Pearson correlation coefficient. j , σ X Let j be the characteristic variable of the blended coal and μ be the standard deviation of the coke quality X. j μ K Let j and X be the expected values, respectively, and cov(j,X) be the covariance of j and X.

[0135] The Pearson correlation coefficient P ranges from -1 to 1, where P > 0 indicates a positive correlation, P < 0 indicates a negative correlation, and P = 0 indicates zero correlation. A larger absolute value of P indicates a stronger correlation: |P| = 0.8-1.0 indicates a very strong correlation, 0.6-0.8 indicates a strong correlation, 0.4-0.6 indicates a moderate correlation, 0.2-0.4 indicates a weak correlation, and 0.0-0.2 indicates a very weak correlation or no correlation.

[0136] 6. Data Normalization

[0137] Some bias-based machine learning algorithms are sensitive to outliers. Indicators such as ash content, sulfur content, volatile matter, G value, Y value, and coke quality have high dimensionality and large range. In order to eliminate the influence of different dimensions between features and ensure the comparability of data, normalization is used to transform features and normalize the data to the range [0,1]. Equation (4) is used to normalize the variables.

[0138]

[0139] In the formula: V is the characteristic value of the data, V max and V min These represent the maximum and minimum values ​​of each feature in each group of samples. This represents the normalized eigenvalues.

[0140] 7. Data Mining Model Construction

[0141] After performing data transformation, cleaning, feature analysis, and normalization on the initial coal and coke dataset, a high-quality coal and coke dataset that meets the requirements of data mining is established. A coke quality prediction model simulating the coking mechanism is built using a rule-based sampling-learner construction-model combination approach.

[0142] 7.1 Rule Sampling

[0143] The initial coal and char dataset D is randomly divided into k datasets of equal size, D1, D2, ..., D3, using a resampling method (sampling with replacement). k D j and D (-j) The training set and test set are for the j-th execution.

[0144] 7.2 Building a Learner

[0145] Construct i independent base learners, denoted as learner 1, learner 2, ..., learner i, to simulate the coking mechanism. For a given input x, each learner h... i Both attempts aim to approximate the distribution of the original data y and obtain a local output h. i (y|x;θi ), where θ i It is the i-th learner h i The parameters.

[0146] Each subset trains its own learner (learning a model on each dataset). By changing the weights of similar samples, the learner learns the optimal model on each dataset.

[0147] 7.3 Model Combination

[0148] like Figure 3 As shown, the learners simulate the focusing process from different angles. The trained learners are combined in parallel, and each learner has an independent output. This allows the final result to complement each other's strengths and weaknesses, learn a stable model that performs well in all aspects, and generate a more powerful model by combining the outputs. This significantly improves the model's accuracy and generalization ability.

[0149] The model combination method of parallel learners is as follows: the learners are arranged in parallel as plastic focusing learners, surface bonding focusing learners and mesophase focusing learners.

[0150] Define a weight function π i (x; α) measures each learner h i The contribution of each learner h. Here, α is the parameter of the weight function. The final output of the model can be represented as the contribution of each learner h. i The weighting of the output is shown in equation (5).

[0151] H(y|x)=π1(x;α)×h1(y|x;θ1)+π2(x;α)×h2(y|x;θ2)+……+π i (x;α)×h i (y|x;θ i Equation (5)

[0152] The input indicators for the learner are the coal quality characteristics of a single type of coal, the coal quality characteristics of a blended coal, and the thermal data of a coke oven. The output indicator is the quality index of coke.

[0153] The coal quality characteristics of a single type of coal include, but are not limited to, ash content, sulfur content, volatile matter, caking index, maximum thickness of the plastic layer (Y value), maximum shrinkage of the plastic layer (X value), maximum Gibbs freeness, plastic range, initial flow temperature, maximum flow temperature, final flow temperature, solidification temperature, maximum expansion (b value), maximum shrinkage (a value), solid-soft range, ash composition (SiO2, Al2O3, Fe2O3, TiO2, CaO, MgO, K2O, Na2O, SO3, MnO2, P2O5), and vitrinite reflectance. The standard deviation of reflectance (s), the vitrinite reflectance range (V3-V21), the content of active components, the content of inert components, and other relevant indicators.

[0154] The coal quality characteristics of blended coal include, but are not limited to, ash content, sulfur content, volatile matter, caking index, plastic layer index, and other relevant indicators.

[0155] The thermal data of a coke oven includes, but is not limited to, indicators such as coking temperature and coking time.

[0156] The quality index (output index) of coke is the ash content (A) of coke. d ), total sulfur (S) t,d ), shatter resistance (M) 40 ), abrasion resistance (M) 10 ), thermal reactivity (CRI) and thermal post-reaction strength (CSR).

[0157] The data mining algorithms used in the training of the learner include, but are not limited to, one or more of the following models: least squares linear regression, ridge regression, LASSO regression, support vector machine (SVR), random forest (RF), decision tree (ID3, C4.5, C5.0 and CART), logistic regression, neural network, etc.

[0158] The data mining algorithms in each learner can be exactly the same or completely different. The learning rate (η) of the machine learning algorithm is between 0.01 and 0.5, the maximum depth (μ) is between 3 and 10, the penalty function parameter (γ) is between 0 and 0.4, the number of training iterations (r) is between 100 and 5000, and the regularization penalty coefficient (λ) is between 0 and 5. The training process automatically terminates and the model parameters of each learner are fixed after the maximum number of iterations is reached.

[0159] 8 Model Evaluation

[0160] The technical flowchart for model evaluation is as follows: Figure 4 As shown, a stratified sampling method is used to divide the validation set, maintaining a similar class ratio between the validation set and the training set.

[0161] The model evaluation parameter uses the correlation coefficient (R²). 2 The parameters such as mean square error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) are shown in equations (6) to (9), respectively.

[0162]

[0163]

[0164]

[0165]

[0166] Where y is the actual value, These are predicted values.

[0167] 9. Coke Quality Prediction System Based on Simulated Coking Mechanism

[0168] A coke quality prediction system simulating coking mechanisms includes a data acquisition layer, a data processing layer, a model building layer, a model solving layer, and a model output layer, such as... Figure 5 As shown. Deploy the trained model to the production system.

[0169] The specific implementation process is as follows:

[0170] 1 Data Collection

[0171] During the coking production process in coking enterprises, the laboratory collects samples of single coal types, blended coals, and coke according to the date and shift for testing and analysis, and presents the results in the form of quality reports.

[0172] The coal quality characteristics of a single type of coal include moisture, ash, volatile matter, sulfur, G value, Y value, Gibbs freeness, Oya expansibility, vitrinite reflectance and distribution, active-inert ratio, mineral catalytic index, and coke index (A) of experimental coke ovens for a single type of coal. d S t,d M 40 M 10 (CRI, CSR). Moisture, ash, volatile matter, sulfur, G value, and Y value of a single type of coal were tested daily during night, day, and afternoon shifts. Vitrin reflectance and distribution were also measured in the experimental coke oven (A). d S t,d M 40 M 10 CRI and CSR are only tested upon arrival at the factory. Gibbs flowability, Oya expansion, active-inert ratio, and mineral catalytic index are only tested when necessary.

[0173] The coal quality characteristics of blended coal include moisture, ash content, volatile matter yield, sulfur content, fineness, and particle size distribution of the blending materials. Moisture, ash content, volatile matter yield, sulfur content, and fineness of the blended coal are tested daily during night shifts, day shifts, and afternoon shifts. The caking index (G) of the blended coal is also measured. R.I The maximum thickness of the plastic layer (Y value) is only measured when the coal blending scheme is changed. Vitrin reflectance and distribution, active-inert ratio, mineral catalytic index, etc., are measured irregularly.

[0174] The quality indicators of coke include moisture, ash content, volatile matter yield, sulfur content, and crush strength (M). 40 ), abrasion resistance (M) 10The following parameters were measured: coke thermal reactivity (CRI), coke post-thermal strength (CSR), and coke particle size distribution. Moisture, ash content, volatile matter yield, sulfur content, and crush strength (M) of the coke were measured daily during night, day, and afternoon shifts. 40 ), abrasion resistance (M) 10 The coke particle size distribution was measured, and the thermal reactivity (CRI) and post-thermal strength (CSR) of the coke were measured only during the daytime.

[0175] The coal blending plan includes information such as the coal blending date, coal type, and blending ratio.

[0176] The operating conditions of coking production are relatively stable, and the thermal data of the coke oven include coking temperature, coking time, etc.

[0177] To ensure data uniformity, the corresponding production data for the day shift, afternoon shift, and night shift are processed by weighted average.

[0178] Table 1. Detection Frequency of Industrial Coke Production Data in Coking Enterprises

[0179]

[0180]

[0181] 2 Data Matching

[0182] The date of the coal blending plan formulation does not coincide with the date of quality testing of the blended coal and coke. The production sequence and data information for the coal preparation, blending, and coking processes in coking production are as follows: Figure 6 As shown.

[0183] Based on the time sequence of coking production, data such as the quality of single coal, the quality of blended coal, the coal blending scheme (ratio), and the quality of coke are matched and organized into a complete set of production data.

[0184] 3 Data Conversion

[0185] The test data of single coal, blended coal, and coke were converted into a relatively stable analytical benchmark, and the total moisture content was converted into the received basis (M). t ), ash and sulfur are converted to dry basis (A) respectively. d S t,d ), the volatile matter yield was converted to dry ash-free basis (V daf The transformed features can effectively describe the distribution and interrelationships of the data, thus enabling the construction of a model more suitable for the data.

[0186] 4 Data Cleaning

[0187] A comprehensive inspection of the raw data was conducted. Based on the data quality requirements for coke quality prediction, the categories and characteristics of industrial production data were analyzed, and data cleaning rules and schemes were established. Operations such as filling in missing data, merging duplicate data, and removing redundant data were performed to obtain high-quality data for model building, ensuring that the data prediction results meet the actual requirements of industrial production.

[0188] Outliers in coking coal blending production data can impact model construction. It's crucial to distinguish whether outliers are due to data anomalies or coal quality anomalies. Outlier analysis methods in coking coal blending production data fall into two categories: physical and statistical criteria. Physical criteria remove outliers from the original data that do not conform to the criteria (data anomalies). Statistical criteria remove data that is statistically abnormal.

[0189] (1) Statistical criteria for exclusion

[0190] Based on the crushing strength (M) of coke 40 For example, the detection fluctuation range of this indicator is ±3%. According to the statistical criteria for exclusion, the coke crush resistance (M...) in the production data... 40 ) Create a scatter plot of the indicator and find points outside the fluctuation range, such as... Figure 7 As shown. A thorough analysis of the reasons for the abnormal data deviation is needed, tracing back to the coal blending scheme, and analyzing whether the anomaly was caused by the coal blending scheme or a data testing error. If it is a data testing error, the affected data set needs to be removed.

[0191] Other indicators are handled in the same way.

[0192] (2) Physical criteria elimination

[0193] According to formula (2), the coking rate (K) of a certain enterprise under normal production conditions (without coking) is about 75%, which is unlikely to exceed 80%. Therefore, K≤80% is used as the criterion for removing ash data, and the data is removed.

[0194] Other indicators are handled in the same way.

[0195] 5 Feature Analysis

[0196] The correlation coefficients between coke quality indicators and blended coal quality indicators were obtained using the Pearson correlation coefficient, as shown in Table 2. The ash content, volatile matter, sulfur content, caking index, and plastic layer index of the blended coal showed strong independence and were therefore chosen as inputs to the model.

[0197] Table 2. Pearson correlation coefficients between coke quality indicators and blended coal quality indicators.

[0198]

[0199] The crush resistance of coke (M) 40), abrasion resistance (M) 10 ), thermal reactivity (CRI), thermal post-reaction strength (CSR), and ash content (A) of blended coal. d ), sulfur (S) t,d ), volatile matter yield (V) daf ), cohesion index (G) R.I Correlation analysis was performed on the raw data of the maximum thickness (Y) of the gelatinous layer. The results of the correlation heatmap are shown below. Figure 8 As shown.

[0200] 6. Data Normalization

[0201] According to formula (4), the data is normalized to the range [0,1] to eliminate the influence of different dimensions between different features and ensure the comparability between data.

[0202] 7. Construction of Data Mining Models

[0203] The initial coal char dataset D was randomly divided into three datasets of equal size, D1, D2, and D3, using a resampling method (sampling with replacement). Three independent base learners, denoted as learner A, learner B, and learner C, were constructed to simulate three coking mechanisms: plastic coking, surface-bonded coking, and mesophase coking, respectively. Each learner was trained and used to predict on the training set samples.

[0204] Each subset trains its own learner (learning a model on each dataset). By changing the weights of similar samples, the learner learns the optimal model on each dataset.

[0205] Each of the three learners has an independent output, and they are combined in parallel. Figure 3 This is a weighted combination method for the learners. The three learners are trained using different algorithms and objective functions, and each outputs a completely different model independently, thus preventing them from producing identical errors on the test set.

[0206] Based on the confidence (accuracy) features output by the three learners, the weight coefficients of each learner are obtained through iterative calculation. An optimization model for minimizing prediction bias is then established, and a coke quality prediction method based on data mining simulation of coking mechanism is constructed.

[0207] The output of the ensemble model is a linear weighted sum of the outputs of the three learners. The outputs of the three learners are linearly weighted using multiple linear regression, as shown in Equation (10).

[0208] H(x)=α1×h1(x)+α2×h2(x)+α3×h3(x)……Equation (10)

[0209] Where α1, α2, and α3 are the weights of learner 1, learner 2, and learner 3, respectively, and all are non-negative; and α1 + α2 + α3 = 1.

[0210] Five-fold cross-validation is used, where the dataset is randomly divided into five equal-sized subsets. One subset is retained as the validation set at each iteration, and the model is built through five rounds of cross-validation. Figure 9 As shown. Specific Implementation Example 1:

[0212] Nine hundred sets of industrial production data from a 7-meter top-loading coke oven at a coking plant, collected from January 2018 to June 2022, were used to predict coke quality. After data matching, transformation, cleaning, feature analysis, correlation analysis, and normalization, a parallel model was built using a regression tree (CART) model as the learner on the Python-based scikit-learn machine learning framework. The model learned from each training set individually, and the optimal value of the learning result was used as the model for the learner. Multiple linear regression was used to linearly weight the outputs of the three learners, resulting in the output of the ensemble model.

[0213] The traditional SVR algorithm and the Adaboost algorithm were used to predict coke quality (CSR), and the results were compared with the algorithm of this invention. The trend of model accuracy with the number of iterations is as follows: Figure 10 As shown, the trend of model accuracy with sample size is as follows: Figure 11 As shown.

[0214] Depend on Figure 10 and Figure 11 It is known that the traditional machine learning algorithm SVR performs poorly in predicting coke quality, while the Adaboost algorithm can improve the accuracy of coke quality prediction. As the number of samples participating in model training increases, the classification accuracy of traditional machine learning algorithms does not significantly improve, while the Adaboost algorithm shows some improvement. The algorithm of this invention can adaptively adjust the weights of the samples participating in model training, resulting in a significant improvement in the prediction performance of coke quality.

[0215] The present invention also provides an apparatus for implementing the parallel prediction method for coke quality based on a simulated coking mechanism, comprising a processor and a memory connected thereto; wherein the processor is configured to execute the parallel prediction method for coke quality based on a simulated coking mechanism; and the memory is used to store executable instructions of the processor.

[0216] The present invention also provides a computer storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the aforementioned parallel prediction method for coke quality based on simulating coking mechanism.

[0217] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0218] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0219] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0220] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0221] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0222] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A parallel prediction method for coke quality based on simulating coking mechanisms, characterized in that, Includes the following steps: Step 1: Data Collection Based on the characteristics of coking coal blending and coking production data, actual coking production data of coking enterprises are collected. From the quality reports and production process reports of coking enterprises, data on the coking production process that affect coke quality are analyzed and extracted, including data on single coal, blended coal, coke and coke oven thermal performance. Step 2: Data matching, constructing the initial coal and coke dataset Based on the time-series relationship of coking production, data matching was performed on the coal quality characteristics of single coal, coal quality characteristics of blended coal, coal blending ratio, coke quality indicators and coke oven thermal data to construct an initial coal and coke dataset. Step 3: Data Conversion The analytical standards for various test data of coal and coke are different. The coal and coke test data are converted into relatively stable analytical standards, including: total moisture is expressed on an as-received basis, ash and sulfur are expressed on a dry basis, and volatile matter yield is expressed on a dry ash-free basis. Step 4: Data Cleaning Based on the data quality requirements for coke quality prediction, this paper analyzes the categories and characteristics of industrial production data, establishes data cleaning rules and plans, conducts comprehensive inspection of raw data to detect missing, irrelevant / redundant, and abnormal / erroneous data, confirms the nature and location of problematic data, and obtains comprehensive data quality information; and corrects the data by filling in missing data, merging duplicate data, removing redundant data, and detecting abnormal / erroneous data. Step 5: Feature Analysis Feature analysis removes unnecessary or redundant features, selects the features that best fit the modeling problem, reduces model size, and prevents overfitting. Feature analysis methods include stepwise regression, sequence feature selection, regularization, and nearest neighbor analysis. Step 6: Data Normalization Some bias-based machine learning algorithms are sensitive to outliers. Ash content, sulfur content, volatile matter, G value, Y value, and coke quality indicators have high dimensionality and large range. In order to eliminate the influence of different dimensions between features and ensure the comparability of data, normalization is used for feature transformation. Step 7: Building the Data Mining Model A high-quality coal and coke dataset meeting data mining requirements was established. A coke quality prediction model simulating the coking mechanism was built using a rule-based sampling-learner construction-model combination approach; this included the following: 7.1 Rule Sampling The initial coal and coke dataset D is randomly divided into k datasets of equal size, D1, D2, ..., D3, using a resampling method. k D j and D (-j) The training set and test set are for the j-th execution; 7.2 Building the Learner Construct i independent base learners, named h1, h2, ..., h1 respectively. i This indicates that each learner h simulates one or more coking mechanisms; for a given input x, each learner h... i Both attempts aim to approximate the distribution of the original data y and obtain a local output h. i (y|x; θ i ); where θ i It is the i-th learner h i Parameters; Each subset trains its own learner, that is, learns a model on each dataset; by changing the weights of similar samples, the learner learns the optimal model on each dataset respectively. 7.3 Model Combination The learners simulate the focusing process from different angles and are combined in parallel. Each learner has an independent output, so that the final result can complement each other and learn a stable model that performs well in all aspects. By combining the outputs, a more powerful model is generated, and the model accuracy and generalization ability are greatly improved.

2. The parallel prediction method for coke quality based on simulating coking mechanism according to claim 1, characterized in that, In step 4, the data correction includes detecting abnormal / erroneous data, as detailed below: Statistical methods, distance-based methods, and machine learning models are used to find outlier data that is outside the range of the vast majority of data, and to determine whether outlier data is erroneous or an anomaly. Outliers in coking coal blending production data can affect model construction. It is necessary to distinguish whether outliers are data anomalies or coal quality anomalies. Outlier analysis methods in coking coal blending production data are divided into two categories of criteria: physical and statistical. Physical criteria are used to remove outlier data from the original data that does not conform to the criteria; Statistical criteria exclude statistically abnormal data; ①Statistical criteria for exclusion When the coking process operating conditions are fixed, the production data follows a normal distribution; abnormal data are processed according to the Schauville criterion, that is, data with a deviation greater than σ are removed, where σ is the standard deviation of the production data; the formula for calculating σ is shown in equation (1). ...Equation (1) ② Physical criteria elimination Volatile matter V in coke from coking plants d ≤1.5%, volatile matter V of blended coal daf If the coking rate is controlled between 23% and 29%, then the formula for calculating the coking rate K is shown in equation (2): ...Equation (2) In the formula: a is the correction coefficient for the coking rate K, which is related to the volatile matter content of coal, furnace structure, and operating procedures, and is 1.5%; Under normal production conditions, no coking occurs, and the coking rate K is between 73.2% and 79.3%. K≤α is used as the rejection criterion for coal and coke data, and α is between 74% and 80%. The data is then rejected.

3. The parallel prediction method for coke quality based on simulating coking mechanism according to claim 1, characterized in that, In step 5, the feature analysis specifically includes the following: Pearson correlation coefficient was used to select features with high correlation to coke quality indicators and good data independence; according to the Pearson correlation formula, the correlation P between two continuous variables j and X was calculated. j,X for: ... Equation (3) Among them, P j,X Let σ be the Pearson correlation coefficient. j , σ X Let j be the characteristic variable of the blended coal and μ be the standard deviation of the coke quality X. j μ K Let j and X be the expected values, respectively, and cov(j,X) be the covariance of j and X. The Pearson correlation coefficient P ranges from -1 to 1. P>0 indicates a positive correlation, P<0 indicates a negative correlation, and P=0 indicates zero correlation. The larger the absolute value of P, the stronger the correlation. |P|=0.8-1.0 indicates a very strong correlation, 0.6-0.8 indicates a strong correlation, 0.4-0.6 indicates a moderate correlation, 0.2-0.4 indicates a weak correlation, and 0.0-0.2 indicates a very weak correlation or no correlation.

4. The parallel prediction method for coke quality based on simulating coking mechanism according to claim 1, characterized in that, Step 6 specifically involves: normalizing the data to the range [0,1], and normalizing the variables using equation (4): ... Equation (4) In the formula: V is the characteristic value of the data, V max and V min These represent the maximum and minimum values ​​of each feature in each sample group. This represents the normalized eigenvalues.

5. The parallel prediction method for coke quality based on simulating coking mechanism according to claim 1, characterized in that, Step 7 further includes the following: Define a weight function π i (x; α) measures each learner h i The contribution of each learner, h, is given by α, where α is the parameter of the weight function; the final output of the model is represented by each learner h. i The weighted output is shown in equation (5): H(y|x)= π1(x; α) ×h1(y|x; θ1)+ π2(x; α) ×h2(y|x; θ2)+ ……+π i (x; α) ×h i (y|x; θ i ) …… Equation (5).

6. The parallel prediction method for coke quality based on simulating coking mechanism according to claim 1, characterized in that, Step 7 further includes the following: The input indicators for the learner are the coal quality characteristics of a single type of coal, the coal quality characteristics of a blended coal, and the thermal data of a coke oven. The output indicator is the quality index of coke. The coal quality characteristics of a single type of coal include ash content, sulfur content, volatile matter, caking index, maximum thickness of the plastic layer (Y value), maximum shrinkage of the plastic layer (X value), maximum Gibbs freeness, plastic range, initial flow temperature, maximum flow temperature, final flow temperature, solidification temperature, maximum expansion (b value), maximum shrinkage (a value), solid-soft range, coal ash composition: SiO2, Al2O3, Fe2O3, TiO2, CaO, MgO, K2O, Na2O, SO3, MnO2, P2O5, and vitrinite reflectance. , The standard deviation of reflectance (s), the vitrinite reflectance range (V3-V21), the content of active components, the content of inert components, and other relevant indicators; The coal quality characteristics of blended coal include ash content, sulfur content, volatile matter, caking index, plastic layer index, and other relevant indicators; The thermal data of a coke oven include coking temperature and coking time. The quality index of coke, i.e., the output index, is the ash content (A) of coke. d Total sulfur S t,d , shatter resistance M 40 Abrasion resistance M 10 Thermal reactivity (CRI) and thermal post-reaction strength (CSR).

7. The parallel prediction method for coke quality based on simulating coking mechanism according to claim 1, characterized in that, Step 7 further includes the following: The data mining algorithms used in the training of the learner include one or more of the following: least squares linear regression, ridge regression, LASSO regression, support vector machine (SVR), random forest (RF), decision tree (ID3, C4.5, C5.0 and CART), logistic regression, and neural network models. The data mining algorithms in each learner may be exactly the same or completely different; the learning rate η of the machine learning algorithm is between 0.01 and 0.5, the maximum depth μ is between 3 and 10, the penalty function parameter γ is between 0 and 0.4, the number of iterations r is between 100 and 5000, and the regularization penalty coefficient λ is between 0 and 5; the training process is automatically terminated and the model parameters of each learner are fixed after the maximum iteration period is reached.

8. The parallel prediction method for coke quality based on simulating coking mechanism according to claim 1, characterized in that, Also includes: Step 8: Model Evaluation A stratified sampling method is used to divide the validation set, maintaining a similar class ratio between the validation set and the training set; The model evaluation parameter uses the correlation coefficient R. 2 The parameters of mean square error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) are shown in equations (6) to (9), respectively. ...Equation (6) ...Equation (7) ...Equation (8) ...Equation (9) Where y is the actual value, These are predicted values.

9. An apparatus for implementing the parallel prediction method for coke quality based on simulating coking mechanisms as described in any one of claims 1 to 7, characterized in that, This includes the processor and the memory connected to it; The processor is configured to execute the parallel prediction method for coke quality based on simulating coking mechanism as described in any one of claims 1 to 7. The memory is used to store the executable instructions of the processor.

10. A computer storage medium, characterized in that, It stores a computer program, which is executed by a processor to implement a parallel prediction method for coke quality based on simulating coking mechanism as described in any one of claims 1 to 7.

Images