A dust control method in a calcium carbonate manufacturing process

Abstract

The present application provides a dust control method in a calcium carbonate manufacturing process, including: preprocessing a dataset of dust monitoring data and process parameters, fusing data from different sources, in different formats, and with different timestamps, and normalizing the data into a form for time series analysis; under preset constraints of production efficiency, cost control, and quality stability, performing an optimization search on equipment decision variable parameters, product formulas, and transportation methods to obtain a balanced solution for dust control and production management, and forming an improved process plan; sending the optimized process plan to the control systems of each production device, adjusting the temperature, pressure, and flow parameters of each process in real time, and dynamically monitoring the change in dust concentration to form a closed-loop optimization control of dust.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular, to a method for controlling dust in the calcium carbonate manufacturing process. Background Art

[0002] In the calcium carbonate manufacturing process, the generation of dust is closely related to the technological process. To effectively control dust, it is necessary to monitor and collect data on key parameters such as the chemical composition, particle size, and pH value of the dust in real time. Among them, how to collect data on multiple attributes of dust is a major problem. However, the massive dust monitoring data poses a huge challenge to data analysis and process optimization. How to extract the internal relationship between dust and the manufacturing process from the huge amount of data and formulate a precise process optimization plan accordingly has become a technical problem to be solved urgently. The correlation analysis between dust data and the manufacturing process involves multiple complex links, including data preprocessing, feature extraction, association rule mining, etc. Among them, data preprocessing requires cleaning, integrating, and transforming the original dust monitoring data to improve data quality; feature extraction requires selecting the most representative and discriminative feature subset from high-dimensional data to reduce the analysis complexity; association rule mining requires using algorithms such as frequent pattern mining and sequential pattern mining to discover the implicit relationship between dust attributes and process parameters. These links are interdependent and iteratively optimized, posing extremely high requirements on the accuracy, efficiency, and robustness of the algorithms. In addition, the calcium carbonate manufacturing process involves multiple processes such as raw material processing, calcination, digestion, and drying, and there are many types of process parameters, such as temperature, pressure, flow rate, rotation speed, etc. The correlation between dust data and process parameters is diverse, dynamic, and uncertain, further increasing the difficulty of correlation analysis. At the same time, process optimization requires comprehensive decision-making in multiple dimensions such as equipment parameters, product formula, transportation method, and ventilation management. The optimization process is interrelated, and how to balance production efficiency, cost control, quality stability, etc. while meeting the dust control target is a multi-objective optimization problem with great solution difficulty. Summary of the Invention

[0003] The present invention provides a method for controlling dust in the calcium carbonate manufacturing process, which mainly includes:

[0004] During the calcium carbonate manufacturing process, an infrared spectrometer is installed to monitor the chemical composition data of the dust, a laser particle size analyzer is used to monitor the particle size data of the dust, and a pH meter is used to monitor the pH value data of the dust to obtain the dust monitoring data during the calcium carbonate manufacturing process. At the same time, the temperature, pressure, flow rate, rotation speed process parameters of the raw material processing, calcination, digestion, and drying processes are obtained, and a data set of dust monitoring data and process parameters is constructed;

[0005] Preprocess the dataset of dust monitoring data and process parameters, fuse data from different sources, in different formats, and with different timestamps, and normalize the data into a form for time series analysis;

[0006] Extract subsets of dust monitoring data and process parameter features from the preprocessed dataset of dust monitoring data and process parameters. Combine dust attributes including dust chemical composition, particle size, and pH value attributes with process parameters including temperature, pressure, flow rate, and rotation speed to screen out feature combinations;

[0007] Obtain frequent association patterns between dust attributes and process parameters based on the feature combinations. Through a preset association rule mining algorithm and preset algorithm parameter settings, obtain frequent association rules, and based on the analysis results of dust attributes and process parameters in the time series, obtain a dust and process association rule library;

[0008] According to the mined dust and process association library, combined with a preset calcium carbonate production process flow chart, judge the causes of dust generation in each process, determine the dust control points for key processes such as raw material treatment, calcination, digestion, and drying, and formulate improvement measures for the process of equipment parameters, product formula, and transportation method;

[0009] Under the preset constraints of production efficiency, cost control, and quality stability, perform an optimization search on the equipment decision variable parameters, product formula, and transportation method to obtain a balanced solution for dust control and production management, and form an improved process plan;

[0010] Send the optimized process plan to the control systems of each production equipment, adjust the temperature, pressure, and flow rate parameters of each process in real time, and dynamically monitor the change of dust concentration to form a closed-loop optimization control of dust.

[0011] The technical solution provided by the embodiments of the present invention may include the following beneficial effects:

[0012] The present invention discloses a method for controlling dust in the calcium carbonate manufacturing process. This method installs sensors on the production line to monitor dust properties and process parameters in real time, and constructs an associated dataset of dust monitoring and process parameters. After preprocessing and feature extraction of the data, association rule mining and time series analysis are used to establish an association rule base between dust and process. Based on this rule base, the present invention determines the causes of dust generation in each process, determines the dust control points in the key processes, and formulates process improvement measures. Under the constraints of production efficiency, cost, and quality, the process parameters are optimized to form an improvement plan. The optimized plan is sent to the production equipment to adjust the process parameters in real time, and the change of dust concentration is dynamically monitored to form a closed-loop control. By analyzing the relationship between the dust concentration change curve and the process parameters, the present invention can adjust the parameters according to the feedback to achieve precise control of dust. This method realizes the intelligent management of dust emissions in the calcium carbonate production process, and improves production efficiency and product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a flowchart of a method for controlling dust in the calcium carbonate manufacturing process of the present invention.

[0014] Figure 2 It is a schematic diagram of a method for controlling dust in the calcium carbonate manufacturing process of the present invention.

[0015] Figure 3 It is another schematic diagram of a method for controlling dust in the calcium carbonate manufacturing process of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and detailedly described in conjunction with the drawings in the embodiments of the present invention. The described embodiments are only a part of the embodiments of the present invention.

[0017] Such as Figures 1-3 , a method for controlling dust in the calcium carbonate manufacturing process of this embodiment may specifically include:

[0018] S101. During the calcium carbonate manufacturing process, monitor the chemical composition data of dust by installing an infrared spectrometer, monitor the particle size data of dust by a laser particle size analyzer, and monitor the pH value data of dust by a pH meter to obtain the dust monitoring data during the calcium carbonate manufacturing process. At the same time, obtain the temperature, pressure, flow rate, and rotation speed process parameters of the raw material treatment, calcination, digestion, and drying processes, and construct a dataset of dust monitoring data and process parameters.

[0019] Obtain the dust chemical composition data monitored by an infrared spectrometer and the dust particle size data monitored by a laser particle size analyzer; construct a dust characteristic matrix based on the dust chemical composition data and the dust particle size data, where the dust characteristic matrix includes a multi-dimensional feature vector of the chemical composition percentage and particle size distribution of each dust sample; perform principal component analysis on the dust characteristic matrix, and select the principal components with the cumulative contribution rate reaching a preset threshold as the dust key feature vectors; establish a non-linear mapping model between the process parameters and the dust key feature vectors, where the non-linear mapping model is constructed using a random forest regression tool, with the process parameters as input variables and the dust key feature vectors as target variables; calculate the importance scores of each process parameter on the dust characteristics according to the non-linear mapping model, and determine the key process parameters that most significantly affect the dust characteristics; construct a gradient boosting decision tree tool, with the key process parameters and the dust key feature vectors as input features and the pH value as the prediction target, where the pH value is monitored and obtained by a pH meter; use the decision tree tool to analyze the multi-dimensional associated data set including process parameters, dust characteristics and pH value, and establish a relationship model between process parameters and dust characteristics, with the process parameters as feature nodes and the dust characteristics and pH value as leaf nodes.

[0020] Specifically, a dust characteristic matrix is constructed based on the dust chemical composition data obtained by monitoring with an infrared spectrometer and the dust particle size data obtained by monitoring with a laser particle size analyzer. The percentage of chemical components and the particle size distribution of each dust sample are numerically processed to form a multi-dimensional feature vector. Principal component analysis is used to reduce the dimension of the feature matrix, and the principal components with a cumulative contribution rate reaching 95% are selected as the key dust feature vectors. For the relationships between process parameters such as raw material treatment temperature, calcination pressure, digestion flow rate, and drying rotation speed and the key dust feature vectors, a non-linear mapping model is established using a random forest regression tool. During the model training process, the process parameters are used as input variables, and the key dust feature vectors are used as target variables, and the best fitting result is obtained through multiple iterations of optimization. Using the obtained non-linear mapping model, the importance scores of each process parameter on the dust characteristics are calculated to determine the key process parameters that most significantly affect the dust characteristics. Combining the dust pH value data obtained by monitoring with a pH meter and the non-linear mapping model of the process parameters and dust characteristics established above, a gradient boosting decision tree tool is constructed. In the gradient boosting decision tree, the process parameters and the key dust feature vectors are used as input features, and the pH value is used as the prediction target. The model is trained and optimized through a cross-validation method, and finally a prediction model that can accurately predict the dust pH value under different process parameter combinations is obtained. Based on the non-linear mapping model of the process parameters and dust characteristics and the pH value prediction model, a multi-dimensional correlation dataset including process parameters, dust characteristics, and pH values is constructed. The decision tree tool is used to analyze this dataset to establish a relationship model between the process parameters and the dust characteristics. In the decision tree, the process parameters are used as feature nodes, and the dust characteristics and pH values are used as leaf nodes, and feature selection and tree structure construction are carried out through the information gain criterion. By analyzing the decision tree model, the key process parameter combinations that affect the dust characteristics are identified, providing a basis for judging the optimization of process parameters during the calcium carbonate manufacturing process. During the calcium carbonate manufacturing process, the dust chemical composition data obtained by the infrared spectrometer includes the percentages of components such as CaCO3, MgCO3, SiO2, etc., and the dust particle size data obtained by the laser particle size analyzer includes particle size distribution parameters such as D10, D50, D90, etc. These data are combined into a multi-dimensional feature vector to construct a dust characteristic matrix. Principal component analysis is applied to the feature matrix, and the principal components with a cumulative contribution rate of 95%, such as the first 3 principal components PC1, PC2, PC3, are selected as the key dust feature vectors. For process parameters such as raw material treatment temperature, such as 800°C - 1000°C, calcination pressure, such as 0.1 - 0.5 MPa, digestion flow rate, such as 50 - 100 L / min, and drying rotation speed, such as 10 - 30 rpm, a non-linear mapping model is established using random forest regression. In the model training, 500 decision trees are used, the maximum depth of each tree is set to 10, and the minimum number of samples in the leaf nodes is set to 5. The model parameters are optimized through cross-validation to obtain the best fitting result.Using the trained random forest model, calculate the feature importance scores of each process parameter, such as temperature 0.4, pressure 0.3, flow rate 0.2, and rotation speed 0.1, and determine that temperature and pressure are the key process parameters that most significantly affect the dust characteristics. Combining the dust pH value data obtained by the pH meter, such as pH 7.5 - 9.5, construct the gradient boosting decision tree model XGBoost. In XGBoost, use the process parameters and the key feature vectors PC1, PC2, and PC3 of the dust as input features, and the pH value as the prediction target. Use 100 base learners, set the learning rate to 0.1, and the maximum tree depth to 6. Train and optimize the model through 5-fold cross-validation, and finally obtain a pH value prediction model with a root mean square error RMSE of 0.15. Based on the above model, construct a multi-dimensional association dataset including process parameters, dust characteristics, and pH values. Use the decision tree algorithm C4.5 to analyze this dataset and establish a relationship model between process parameters and dust characteristics. In the decision tree, use the process parameters as feature nodes, and the dust characteristics and pH values as leaf nodes, and perform feature selection and tree structure construction through the information gain ratio criterion. Set the minimum number of samples in a leaf node to 10 and the maximum tree depth to 8. Through the analysis of the decision tree model, identify the key process parameter combinations that affect the dust characteristics. For example, when the temperature is in the range of 950°C - 980°C and the pressure is in the range of 0.3 - 0.4 MPa, ideal dust characteristics and pH values can be obtained.

[0021] S102. Preprocess the dataset of dust monitoring data and process parameters, fuse data from different sources, different formats, and different timestamps, and normalize the data into the form of time series analysis.

[0022] Obtain dust monitoring data and process parameter data, and use an adaptive moving median filter to perform preliminary noise removal on the data to obtain a preliminarily denoised dataset. According to the preliminarily denoised dataset, combine the Z-score method to identify outliers, and judge whether the Z-score value of the data point exceeds the preset threshold range. If it exceeds, mark it as an outlier. For outliers, use the local linear regression method for correction to obtain a corrected dataset. Use a data conversion tool to unify the format of the corrected dataset, convert the data of the infrared spectrometer, laser particle size analyzer, and pH meter into a standardized JSON format to obtain a standardized dataset. Use the Kalman filtering algorithm to perform multi-source data fusion on the standardized dataset, and generate a unified data stream based on the measurement errors and time delays of different data sources. Use a timestamp alignment tool to synchronize the unified data stream, interpolate all data at a unified time interval, and use the cubic spline interpolation method to fill in the missing time point data to obtain a time-aligned and normalized time series dataset.

[0023] Specifically, an adaptive moving median filter is used to preliminarily remove noise from dust monitoring data and process parameter data. The sliding window size is dynamically adjusted by calculating the local data variance, and the median of the data within each window is calculated to replace the original data points, obtaining a preliminarily denoised data set. For edge data points, the mirror filling method is used to maintain data integrity. According to the statistical characteristics of dust monitoring data and process parameter data, combined with the Z-score method and the local outlier factor algorithm, outliers are identified. The Z-score value of the data is calculated, and the threshold is set to ±3. At the same time, the local outlier factor of each data point is calculated, and the data that exceeds the threshold range or whose local outlier factor is significantly higher than the surrounding points is marked as an outlier. The local linear regression method is used to correct the outliers. For data from different sources and in different formats, the data format is unified through a data conversion tool. The data from infrared spectrometers, laser particle size analyzers, and pH meters is converted into a standardized JSON format, and field mapping, unit conversion, and data type unification are performed. At the same time, the format of process parameter data is converted to ensure that all data has a consistent structure and field name, and the original timestamp information is retained during this process. The Kalman filter algorithm is used to fuse multi-source data, considering the measurement errors and time delays of different data sources to generate a unified data stream. Then, a timestamp alignment tool is used to synchronize the fused data with different timestamps. All data is interpolated at a unified time interval, and the cubic spline interpolation method is used to fill in the missing time point data. For intervals with drastic changes, piecewise linear interpolation is used. Finally, a time-aligned normalized time series data set is obtained, laying a foundation for subsequent analysis. During the calcium carbonate manufacturing process, an adaptive moving median filter is used to preliminarily remove noise from dust monitoring data and process parameter data. Specifically, when implementing, the local data variance within 10 minutes is calculated. If the variance is greater than 0.5, the sliding window is set to 3 data points; if the variance is less than 0.5, it is set to 7 data points. The median of the data within each window is calculated to replace the original data points, obtaining a preliminarily denoised data set. For edge data points, the mirror filling method is used to mirror-copy the first 3 or the last 3 data points to the edge to maintain data integrity. Subsequently, the Z-score method and the local outlier factor algorithm are combined to identify outliers. The Z-score value of the data is calculated, and the threshold is set to ±3. At the same time, the local outlier factor of each data point is calculated, and 10 nearest neighbor points are selected. If the local outlier factor is greater than 1.5, the point is marked as an outlier. The local linear regression method is used to correct the outliers, and the abnormal value is replaced with the fitted value by fitting a straight line using 5 normal data points before and after it. For data from different sources and in different formats, the data format is unified through a data conversion tool. The wavenumber range in the infrared spectrometer data is 4000 - 400 cm -1Convert to JSON format. Convert the D10, D50, and D90 values of the laser particle size analyzer to micrometer units, and retain two decimal places for the data of the pH meter. At the same time, perform format conversion on the process parameter data, unify the temperature to Celsius and the pressure to MPa to ensure that all data has a consistent structure and field names. During this process, retain the original timestamp information accurate to the millisecond level. Subsequently, use the Kalman filtering algorithm to fuse multi-source data. Set the measurement noise covariance matrix R as a diagonal matrix, and the diagonal elements are 0.1, 0.2, and 0.05 respectively, representing the measurement errors of different data sources. Set the process noise covariance matrix Q as 0.01×I, where I is the identity matrix. Generate a unified data stream through iterative updates. Finally, use the timestamp alignment tool to synchronize the fused data, and interpolate all data at 1-second time intervals. For most data, use the cubic spline interpolation method with a smoothing parameter of 0.5; for intervals with drastic changes, such as when the temperature changes by more than 10°C within 5 seconds, use piecewise linear interpolation. Finally, obtain a time-aligned and normalized time series dataset, laying a foundation for subsequent analysis.

[0024] S103. Extract the dust monitoring data and process parameter feature subsets from the preprocessed datasets of dust monitoring data and process parameters. Combine the dust attributes including dust chemical composition, particle size, and pH value with the process parameters including temperature, pressure, flow rate, and rotational speed to screen out the feature combinations.

[0025] According to the preprocessed datasets of dust monitoring data and process parameters, standardize the data to obtain standardized data; use the kernel principal component analysis method to extract the feature subset of the standardized data. Obtain the principal components with a cumulative contribution rate reaching the preset percentage as the feature subset by calculating the eigenvalues and eigenvectors of the kernel matrix of the standardized data; for the process parameter data, calculate the non-linear correlation between temperature, pressure, flow rate, and rotational speed using the maximum information coefficient. If the maximum information coefficient is greater than the threshold, select the parameter with the largest amount of information as the representative to obtain the process parameter feature subset; combine the dust chemical composition, particle size, and pH value attributes, evaluate the relationship between the attributes and the process parameter feature subset, select the attribute with the highest mutual information value, and combine it with the process parameter feature subset to form a preliminary feature combination; use the random forest feature importance evaluation tool to screen the preliminary feature combination, and obtain the feature importance ranking by calculating the contribution of each feature to the model prediction accuracy; combine the recursive feature elimination method, start removing from the least important feature step by step, and evaluate the model performance after each removal until the performance starts to decline significantly or reaches the preset feature quantity threshold to obtain the optimized feature combination.

[0026] Specifically, based on the preprocessed dust monitoring data and process parameter dataset, first, the data is standardized to scale all features to a range with a mean of 0 and a variance of 1. Subsequently, the kernel principal component analysis method is used to extract a feature subset. By calculating the eigenvalues and eigenvectors of the kernel matrix, the principal components with a cumulative contribution rate reaching 95% are selected as the feature subset. For the process parameter data, the maximum information coefficient is used to calculate the non-linear correlation between temperature, pressure, flow rate, and rotational speed. If the maximum information coefficient is greater than 0.8, then the parameter with the largest amount of information is selected as a representative, thereby obtaining a process parameter feature subset. Combining the dust chemical composition, particle size, and pH value attributes, the minimum redundancy maximum correlation algorithm is used to evaluate the relationship between these attributes and the process parameter feature subset. This algorithm selects the top five attributes with the highest mutual information values by maximizing the correlation between features and the target variable while minimizing the redundant information between features, and combines them with the process parameter feature subset to form a preliminary feature combination. The random forest feature importance evaluation tool is used to screen the preliminary feature combination. By calculating the contribution of each feature to the model prediction accuracy, the feature importance ranking is obtained. Combining the recursive feature elimination algorithm, starting from the least important feature, it is gradually removed. After each removal, the model performance is evaluated until the performance starts to decline significantly or reaches the preset feature quantity threshold, and finally, an optimized feature combination is obtained as the input data for subsequent analysis. During the calcium carbonate manufacturing process, feature extraction and selection are performed on the preprocessed dust monitoring data and process parameter dataset. First, all data is standardized to scale the feature values to a range with a mean of 0 and a variance of 1 to ensure the comparability of data at different scales. For example, the temperature data is standardized from the range of 800 - 1000 °C to the range of -1.5 to 1.5. Subsequently, the kernel principal component analysis method is used to extract a feature subset. The Gaussian kernel function is selected, and the kernel parameter σ is set to 1.0. By calculating the eigenvalues and eigenvectors of the kernel matrix, the principal components with a cumulative contribution rate reaching 95% are selected as the feature subset, usually obtaining 10 - 15 principal components. For the process parameter data, the maximum information coefficient is used to calculate the non-linear correlation between temperature, pressure, flow rate, and rotational speed. The α parameter is set to 0.6, and the number of sampling times is 500. If the maximum information coefficient is greater than 0.8, such as the maximum information coefficient between temperature and pressure being 0.85, then the temperature with the largest amount of information is selected as a representative. Combining the dust chemical composition, such as the CaCO3 content, particle size, such as the D50 value, and pH value attributes, the minimum redundancy maximum correlation algorithm is used to evaluate the relationship between these attributes and the process parameter feature subset. The redundancy parameter β is set to 0.5. By maximizing the correlation between features and the target variable while minimizing the redundant information between features, the top five attributes with the highest mutual information values are selected. For example, the CaCO3 content, D50 value, pH value, temperature, and pressure are selected as key attributes.Use a random forest feature importance assessment tool, set the number of trees to 500 and the maximum depth to 10, and calculate the contribution of each feature to the model's prediction accuracy. Combine with the recursive feature elimination algorithm, remove the 10% least important features each time, and evaluate the change in model performance. Stop the iteration when the performance drops by more than 1% or the number of features is reduced to 50% of the original, and finally obtain an optimized feature combination, usually containing 15 - 20 features, as the input data for subsequent analysis.

[0027] S104. Based on the feature combination, obtain the frequent association patterns between the dust properties and process parameters. By presetting the association rule mining algorithm and setting parameters for the preset algorithm, obtain the frequent association rules, and based on the analysis results of the dust properties and process parameters in the time series, obtain the dust - process association rule library.

[0028] Perform equal - frequency discretization on continuous data according to the preset number of intervals to obtain the discretized feature data; use the FP - growth algorithm to mine frequent item sets from the discretized feature data, recursively generate frequent item sets by constructing the FP - tree and conditional FP - tree; construct association rules according to the frequent item sets using the association rule generator; if the association rule meets the preset threshold, it is determined as a strong association rule; obtain the time - series data set, use wavelet transform to extract the frequency - domain features of the time - series data in the time - series data set to obtain the approximation coefficients and detail coefficients at different scales; cluster the strong association rules, construct a rule similarity matrix by calculating the semantic similarity between rule items, and combine with the frequency - domain features, and use the hierarchical clustering algorithm to merge the rules with high similarity to form the dust - process association rule library.

[0029] Specifically, according to the dust properties and process parameters in the feature combination, first, equal-frequency discretization is performed on continuous data, and numerical features are divided into 10 intervals. Subsequently, the FP-growth algorithm is used for frequent itemset mining. The minimum support threshold is set to 0.05. By constructing the FP tree and conditional FP trees, frequent itemsets are recursively generated, the support of the itemsets is calculated, and the frequent itemsets that meet the minimum support requirement are screened out. Using the obtained frequent itemsets, an association rule generator is applied to construct association rules. The minimum confidence threshold is set to 0.7, and the confidence, lift, Kulczynski index, and all-confidence index of the rules are calculated. Strong association rules that meet the minimum confidence requirement, have a lift greater than 1, a Kulczynski index greater than 0.6, and an all-confidence index greater than 0.7 are screened out. For time series data, wavelet transform is used to extract the frequency domain features of the time series. The Daubechies wavelet is selected, and the decomposition level is 3 to obtain the approximation coefficients and detail coefficients at different scales. Then, the adaptive window technique is used to convert continuous time series data into a discrete transaction set. The initial window size is set to 30 minutes, and the window size is dynamically adjusted by calculating the variance of the data within the window. The sliding step is set to 1 / 6 of the window size. The data within each window is processed using the equal-frequency discretization process and frequent itemset mining method, combined with the features extracted by wavelet transform, to obtain association rules with timestamps and frequency domain features. The obtained association rules are integrated through a rule clustering method based on semantic similarity. The Word2Vec tool is used to calculate the semantic similarity between rule items, and a rule similarity matrix is constructed. Combining the time series features, the hierarchical clustering algorithm is used, and the clustering distance threshold is set to 0.2 to merge rules with high similarity. Finally, a dust and process association rule library is formed, which contains information such as rule content, support, confidence, lift, Kulczynski index, all-confidence index, time range, and frequency domain features. During the calcium carbonate manufacturing process, association rule mining is performed on dust properties and process parameters. First, continuous data such as temperature of 800 - 1000 °C and pressure of 0.1 - 0.5 MPa are subjected to equal-frequency discretization and divided into 10 intervals. The FP-growth algorithm is used for frequent itemset mining, and the minimum support threshold is set to 0.05. For example, by constructing the FP tree, it is found that the support of {high temperature, high pressure, large particles} is 0.06, meeting the threshold requirement. Subsequently, an association rule generator is applied to construct rules such as "high temperature, high pressure → large particles", and the confidence is calculated to be 0.75, the lift is 1.2, the Kulczynski index is 0.7, and the all-confidence index is 0.8. For time series data, the Daubechies4 wavelet is used for 3-layer decomposition to obtain the approximation coefficients and detail coefficients. For example, wavelet transform is performed on dust concentration data to obtain the energy distribution of different frequency components.Using the adaptive window technique, the initial window size is set to 30 minutes. By calculating the variance of the data within the window, during the high volatility period, if the variance > 0.5, the window is reduced to 15 minutes; during the low volatility period, if the variance < 0.1, the window is expanded to 45 minutes. The sliding step is set to 1 / 6 of the window size. Apply the FP-growth algorithm and the association rule generator to the data within each window, combine with wavelet features to obtain time-related rules. Finally, use the Word2Vec tool to calculate the semantic similarity of rule items. For example, the similarity between "high temperature" and "temperature increase" is 0.85. Construct a rule similarity matrix and apply the hierarchical clustering algorithm, setting the clustering distance threshold to 0.2. Merge similar rules. For example, "high temperature → large particles" and "temperature increase → particle size increase" are merged into one rule. Finally, form an association rule library containing rule content, various indicators, time range, and frequency domain features to provide a basis for process optimization.

[0030] By segmenting the dust and process parameter data according to the time series to form time windows, obtaining the time series analysis results of the dust attributes and process parameters according to the preset time series model, combining with the mined frequent association rules, screening out the association rules that meet the preset practical application value, and organizing the screened rules into a database format to establish a dust and process association rule library.

[0031] Use the adaptive sliding window method to perform time series segmentation on the dust and process parameter data, and dynamically adjust the window size according to the variance of the data within the window; if the variance exceeds the preset threshold, the window is reduced, otherwise it is expanded. Analyze the dust attributes and process parameter data within the time window, eliminate the non-stationarity of the data through the seasonal autoregressive integrated moving average model, and use the discrete wavelet transform to extract the multi-scale features of the time series. According to the time series analysis results and the previously mined frequent association rules, set the support threshold, confidence threshold, Kulczynski index threshold, and full confidence index threshold to judge whether the rules continuously meet all threshold conditions. Weight the rules that meet the conditions to obtain the energy distribution to determine the weight coefficients. Convert the weighted rules into a structured data format to establish a relational database table structure including a rule table and a time window table; the rule table contains fields such as rule ID, antecedent, consequent, support, confidence, lift, Kulczynski index, full confidence index, and weight coefficient; the time window table contains fields such as window ID, rule ID, start time, end time, and window features.

[0032] Specifically, the adaptive sliding window method is used to segment the time series of dust and process parameter data. The initial window size is set to 60 minutes, and the window size is dynamically adjusted by calculating the variance of the data within the window. When the variance exceeds the preset threshold, the window is shrunk; otherwise, it is expanded. The sliding step is set to 1 / 6 of the window size, generating a continuous time window sequence. Each window contains the time series data of dust attributes and process parameters. The seasonal autoregressive integrated moving average model and wavelet transform are used to analyze the dust attributes and process parameter data within each time window. The differencing operation is used to eliminate the non-stationarity of the data, and the autocorrelation function and partial autocorrelation function are used to determine the order of the model and the seasonal period. At the same time, the discrete wavelet transform is applied to extract the multi-scale features of the time series. The Daubechies wavelet is selected, and the decomposition level is 3 to obtain the energy distribution of different frequency components. Combining the time series analysis results and the previously mined frequent association rules, the rules are screened by setting the support threshold to 0.1, the confidence threshold to 0.7, the Kulczynski index threshold to 0.6, and the all-confidence index threshold to 0.7. It is judged whether the rules continuously meet all threshold conditions within 3 consecutive time windows, and the lift of the rules is calculated. If the lift is greater than 1.5, it is determined that the rule has practical application value. The time series features are used to weight the rules, and the weight coefficient is determined according to the energy distribution obtained by the wavelet transform to highlight the rules with significant time series features. The selected association rules with practical application value are converted into a structured data format, and the relational database table structure is designed, including the main table "Rule Table" and the sub-table "Time Window Table". The Rule Table contains fields such as Rule ID, which is the primary key, antecedent, consequent, support, confidence, lift, Kulczynski index, all-confidence index, and weight coefficient. The Time Window Table contains fields such as Window ID, which is the primary key, Rule ID, which is the foreign key, start time, end time, and window features. An index is established on the Rule ID in the Rule Table and the Rule ID in the Time Window Table to optimize the query performance, forming a dust and process association rule base. During the calcium carbonate manufacturing process, time series analysis and association rule mining are carried out on the dust and process parameter data. First, the adaptive sliding window method is used for time series segmentation, and the initial window size is set to 60 minutes. By calculating the variance of the data within the window, when the variance exceeds 0.5, the window is shrunk to 30 minutes; when the variance is less than 0.1, it is expanded to 90 minutes. The sliding step is set to 1 / 6 of the window size. The seasonal autoregressive integrated moving average model and wavelet transform are applied to the data within each window. The differencing operation is used to eliminate non-stationarity, and the model order is determined by the autocorrelation function and partial autocorrelation function, such as SARIMA(1,1,1)(0,1,1)12. At the same time, the Daubechies4 wavelet is used for 3-layer discrete wavelet transform to obtain the approximation coefficients and detail coefficients, and the energy distribution of different frequency components is calculated.For example, the energy proportion of the low-frequency component is 60%, the medium-frequency is 30%, and the high-frequency is 10%. Combining the time series analysis results and the pre-mined frequent association rules, the screening conditions are set as the support threshold 0.1, the confidence threshold 0.7, the Kulczynski index threshold 0.6, and the all-confidence index threshold 0.7. Judge whether the rule continuously meets all threshold conditions within 3 consecutive time windows, and calculate the lift at the same time. For example, for the rule "temperature increase → dust particle size increase", the indicators within 3 consecutive windows are support, including 0.12, 0.13, 0.11, confidence, including 0.75, 0.78, 0.72, Kulczynski index, including 0.65, 0.68, 0.63, all-confidence index, including 0.73, 0.75, 0.71, and lift 1.8, and it is determined as a rule with practical application value. Determine the weight coefficient using the energy distribution obtained by wavelet transform. For example, the weight of this rule is 0.6×0.7 + 0.3×0.2 + 0.1×0.1 = 0.49. Finally, store the selected rules in a relational database, and design a "rule table" and a "time window table". The rule table contains fields such as rule ID, which is the primary key, such as R001, antecedent, which includes temperature increase, consequent, which includes dust particle size increase, support 0.12, confidence 0.75, lift 1.8, Kulczynski index 0.65, all-confidence index 0.73, and weight coefficient 0.49. The fields of the time window table include window ID, which is the primary key, such as W001, rule ID, which is the foreign key, R001, start time 2024-03-15 10:00:00, end time 2024-03-15 11:00:00, and window characteristics, which can store wavelet coefficients in JSON format. Create an index on the rule ID to optimize the query performance and form a complete dust and process association rule library.

[0033] S105. According to the mined dust and process association library, combined with the preset calcium carbonate production process flow chart, judge the dust generation reasons for each process, determine the dust control points for key processes such as raw material handling, calcination, digestion, and drying, and formulate process improvement measures for equipment parameters, product formula, and transportation method.

[0034] Convert the preset calcium carbonate production process flow chart into a directed graph, where nodes represent processes, edges represent the associations between processes, and edge weights are set according to the association rules; calculate the shortest paths between nodes in the directed graph to obtain the dust generation causes and influencing factors of each process; according to the dust generation causes and influencing factors, use the fuzzy comprehensive evaluation method combined with principal component analysis to evaluate the four key processes of raw material treatment, calcination, digestion, and drying, and set the evaluation indicators of dust concentration, particle size distribution, and chemical composition; extract the main characteristics of the evaluation indicators through principal component analysis, determine the weights of each indicator, calculate the comprehensive scores of each process, and the node with the highest score is determined as the dust control point; for the dust control point, use the multi-objective optimization algorithm to optimize the equipment parameters, product formula, and transportation method; adopt a method combining the particle swarm optimization algorithm and the simulated annealing algorithm to solve the Pareto optimal solution set of the objective function and obtain the optimal parameter combination of each control point; according to the optimal parameter combination, use a knowledge graph to construct a process improvement measure library; extract key concepts from the optimal parameter combination through entity recognition methods, and use relationship extraction to identify the relationships between entities; store the key concepts and the relationships between entities in a graph database, and use the graph database query language for relationship reasoning to obtain specific process improvement measures.

[0035] Specifically, according to the rules in the dust and process association library, the shortest path algorithm in graph theory is used to analyze the preset calcium carbonate production process flow chart. The process flow chart is transformed into a directed graph, where nodes represent processes and edges represent the associations between processes. The edge weights are set according to the association rules. The shortest paths between each pair of nodes are calculated through the Floyd-Warshall algorithm to identify key nodes and paths, and the dust generation causes and influencing factors of each process are obtained. The fuzzy comprehensive evaluation method combined with principal component analysis is used to evaluate the four key processes of raw material treatment, calcination, digestion, and drying, and evaluation indicators such as dust concentration, particle size distribution, and chemical composition are set. The main characteristics of the data are extracted through principal component analysis, and the weights of each indicator are determined in combination with expert experience. The comprehensive scores of each process are calculated, and the node with the highest score is determined as the dust control point. For the identified dust control points, a multi-objective optimization algorithm is used to optimize the equipment parameters, product formula, and transportation method, and objective functions such as dust control, product quality, and energy consumption are set. At the same time, constraint conditions such as the operating range of the equipment and raw material supply are considered. The particle swarm optimization algorithm combined with the simulated annealing algorithm is used to solve the Pareto optimal solution set to obtain the optimal parameter combination of each control point. The Monte Carlo simulation method is used to evaluate the stability and reliability of the optimization scheme. By randomly perturbing the input parameters, a large number of simulation data are generated, and the expected value and variance of the objective function are calculated to screen out the optimization scheme with high stability. Based on the optimization results, a process improvement measure library is constructed using a knowledge graph. Key concepts such as "equipment parameters" and "product formula" are extracted from the optimization scheme through entity recognition, and the relationships between entities such as "influence" and "control" are identified using relation extraction. The extracted information is stored in a graph database, and complex relation reasoning is performed using the graph database query language to obtain specific process improvement measures and form an executable process improvement plan. In the process of optimizing the calcium carbonate production process, first, the process flow chart is transformed into a directed graph. Nodes represent processes such as raw material treatment, calcination, digestion, and drying, edges represent the associations between processes, and the edge weights are set according to the association rules. For example, the edge weight from raw material treatment to calcination is 0.8. The Floyd-Warshall algorithm is used to calculate the shortest paths between each pair of nodes, and the calcination process is identified as a key node. The main reasons for dust generation in this process are mainly related to temperature and pressure. Subsequently, the four key processes are evaluated, and the evaluation indicators are set to include dust concentration (mg / m³), particle size distribution (μm), and chemical composition (%). The eigenvalues 2.5, 1.2, and 0.3 are obtained through principal component analysis, and the weights 0.5, 0.3, and 0.2 are determined in combination with expert experience. The comprehensive score is calculated, and the calcination process has the highest score of 0.85 and is determined as the main dust control point. For the calcination process, the optimization objectives are set as dust concentration ≤ 10 mg / m³, product purity ≥ 98%, and energy consumption ≤ 500 kWh / t. Constraint conditions such as the equipment temperature range of 800 - 1000 °C and the pressure range of 0.1 - 0.5 MPa are considered.The particle swarm optimization algorithm with a swarm size of 50 and 100 iteration times is combined with the simulated annealing algorithm with an initial temperature of 100 and a cooling coefficient of 0.95 to solve and obtain the optimal parameter combination as a temperature of 950 °C and a pressure of 0.3 MPa. The Monte Carlo method is used for 1000 simulations. The expected value of the objective function is calculated as a dust concentration of 8.5 mg / m³, a product purity of 98.5%, and an energy consumption of 480 kWh / t, and the variances are all less than 5%, confirming that the solution has high stability. Finally, a knowledge graph is constructed to extract entities such as "calcination temperature" and "pressure", and to identify relationships such as "affecting dust concentration" and "controlling product purity". The information is stored in the Neo4j graph database, and the Cypher query language is used for reasoning to obtain the improvement measures of controlling the calcination temperature within 950 ± 10 °C and maintaining the pressure at 0.3 ± 0.02 MPa, forming an executable process improvement plan.

[0036] S106. Under the preset constraints of production efficiency, cost control, and quality stability, perform an optimization search on the equipment decision variable parameters, product formula, and transportation mode to obtain a balanced solution for dust control and production management, and form an improved process plan.

[0037] According to the preset constraints of production efficiency, cost control, and quality stability, obtain the equipment decision variable parameters, product formula, and transportation mode as optimization variables, and the dust control level, production efficiency, cost, and product quality as objective functions. Use the ε-constraint method to construct a multi-objective optimization model, select the dust control level as the main objective function, and transform the other objective functions into constraints. Use the particle swarm optimization algorithm combined with the simulated annealing algorithm to solve the optimization model, and determine the particle swarm size, maximum number of iterations, initial temperature, and cooling coefficient therefrom. By continuously updating the particle positions, velocities, and temperature parameters during the annealing process, obtain the optimal solution within the global scope. Adopt an adaptive weight adjustment strategy to dynamically adjust the weights of each sub-objective in the objective function according to the results of each iteration. Calculate the moving window standard deviation of each sub-objective and judge the sub-objective with a large standard deviation. If the standard deviation is large, assign it a high weight. Achieve the dynamic balance between objectives through the weight update formula, and evaluate the Pareto optimality of the current optimal solution after each iteration. Conduct a sensitivity analysis and Monte Carlo simulation on the solution obtained by the optimization algorithm to obtain the sensitivity coefficients of each parameter. Use Latin hypercube sampling to generate input parameter combinations and conduct Monte Carlo simulations to evaluate the stability and reliability of the optimization results.

[0038] Specifically, according to the preset production efficiency, cost control, and quality stability constraints, a multi-objective optimization model is established. The equipment decision variable parameters, product formula, and transportation method are taken as optimization variables, and the dust control degree, production efficiency, cost, and product quality are taken as objective functions. The ε-constraint method is used to construct the multi-objective optimization model. The dust control degree is selected as the main objective function, and other objective functions are transformed into constraint conditions. A series of single-objective optimization problems are generated by systematically changing the constraint bound ε value. The particle swarm optimization algorithm combined with the simulated annealing algorithm is used to solve the established optimization model. The particle swarm size is set to 100, the maximum number of iterations is 1000, the initial temperature is 100, and the cooling coefficient is 0.95. By continuously updating the particle positions, velocities, and temperature parameters during the annealing process, the optimal solution is searched globally. An adaptive weight adjustment strategy is adopted to dynamically adjust the weights of each sub-objective in the objective function according to the results of each iteration. Calculate the moving window standard deviation of each sub-objective, with the window size set to 10, and assign a higher weight to the sub-objective with a larger standard deviation. The weight update formula is wi = σi / Σσj, where σi is the standard deviation of the i-th sub-objective, wi is the weight of the i-th sub-objective, and Σσj is the sum of the standard deviations of all j sub-objectives. In this way, the dynamic balance between objectives is achieved, and the Pareto optimality of the current optimal solution is evaluated after each iteration. Sensitivity analysis and Monte Carlo simulation are performed on the solution obtained by the optimization algorithm. By changing the values of the input parameters, observe the changes in the output results and calculate the sensitivity coefficients of each parameter. Use Latin hypercube sampling to generate 10,000 groups of input parameter combinations for Monte Carlo simulation to evaluate the stability and reliability of the optimization results. According to the sensitivity analysis and Monte Carlo simulation results, screen out the key parameters that have a greater impact on the results. Use digital twin technology to construct a virtual production environment, import the optimization plan, simulate the actual production process, and evaluate the feasibility and effectiveness of the plan. Based on the simulation results, fine-tune the optimization plan to form the final improved process plan. In the process of optimizing the calcium carbonate production process, first establish a multi-objective optimization model, and set the dust control degree, production efficiency, cost, and product quality as objective functions. Adopt the ε-constraint method, select the dust control degree as the main objective, and transform other objectives into constraint conditions. For example, set the production efficiency ≥ 90%, the cost ≤ 5000 yuan / ton, and the product quality ≥ 98% purity. The optimization variables include the calcination temperature of 800 - 1000 °C, the pressure of 0.1 - 0.5 MPa, the CaCO3 ratio of 95 - 99%, and the conveyor belt speed of 0.5 - 2 m / s. Use the particle swarm optimization algorithm combined with the simulated annealing algorithm to solve, with the particle swarm size set to 100, the maximum number of iterations of 1000, the initial temperature of 100, and the cooling coefficient of 0.95. During the iteration process, dynamically adjust the weights, calculate the standard deviations of each objective within 10 iterations, such as the standard deviation of the dust control degree is 0.05, and the production efficiency is 0.02, and update the weights accordingly. Evaluate the Pareto optimality every 100 iterations and retain the non-dominated solutions.After the optimization is completed, sensitivity analysis and Monte Carlo simulation are carried out. 10,000 groups of parameter combinations are generated using Latin hypercube sampling, such as the calcination temperature of 950 ± 10 °C and the pressure of 0.3 ± 0.02 MPa. The sensitivity coefficients are calculated, and it is found that the temperature sensitivity is 0.8 and the pressure is 0.5, identifying the temperature as the key parameter. A virtual production line is constructed using digital twin technology, and the optimized scheme is imported as the calcination temperature of 955 °C, the pressure of 0.32 MPa, the CaCO3 ratio of 97.5%, and the conveyor belt speed of 1.2 m / s. The 100-hour production process is simulated to evaluate the feasibility of the scheme. According to the simulation results, the parameters are fine-tuned to form the final improvement scheme, including the calcination temperature of 953 °C, the pressure of 0.31 MPa, the CaCO3 ratio of 97.8%, and the conveyor belt speed of 1.1 m / s, which is expected to reduce dust emissions by 15% while maintaining the stability of production efficiency and product quality.

[0039] S107. Send the optimized process plan to the control systems of each production device, adjust the temperature, pressure, and flow parameters of each process in real time, and dynamically monitor the change of dust concentration to form a closed-loop optimization control of dust.

[0040] Receive an optimization instruction carrying process plan parameters, and the optimization instruction is sent by the central controller; according to the optimization instruction, send the process plan parameters to the programmable logic controller of each production device through the industrial communication network, and the programmable logic controller adjusts the operating state of the production device according to the process plan parameters; set the main and standby channels and use cyclic redundancy check, and the cyclic redundancy check is used to ensure the accuracy of data transmission; if a communication interruption or data transmission error is detected, automatically switch to the standby channel; obtain the temperature, pressure, and flow parameters of each process, and adjust the temperature, pressure, and flow parameters in real time; obtain the dust concentration detection data of each key point of the production line, and the dust concentration detection data is collected by various types of dust detection devices; collect the dust concentration detection data at a preset frequency through the data acquisition module and perform data preprocessing; construct a correlation model between the dust concentration and process parameters; dynamically optimize the process parameters according to the correlation model, and the dynamic optimization is used to adjust the relevant process parameters.

[0041] Specifically, the optimized process plan parameters are sent to the programmable logic controllers of each production device through the industrial communication network, establishing a communication link between the central controller and the control systems of each process device to achieve real-time transmission and update of the parameters. Redundant communication design is adopted, with main and standby channels set up, and the accuracy of data transmission is ensured through cyclic redundancy check. When a communication interruption or data transmission error is detected, it automatically switches to the standby channel and records the abnormal situation. The distributed control system is used to adjust the temperature, pressure, and flow parameters of each process in real time. The adaptive model predictive control algorithm is introduced. By online identifying the system model parameters, the future output is predicted and the control sequence is optimized. The upper and lower limits of parameter adjustment and the change rate limit are set to prevent over-adjustment. Multiple types of dust detection devices are used to continuously monitor the dust concentration at key points on the production line. Combining light scattering and beta-ray attenuation methods improves the comprehensiveness and accuracy of detection. The detection data is collected at a frequency of 1Hz through the data acquisition module and undergoes data preprocessing, including outlier detection and moving average filtering. The processed data is transmitted to the central control system in real time. An association model between dust concentration and process parameters is constructed based on the decision tree algorithm. The sliding window technique is adopted, with the window size set to 1 hour, and the model parameters are updated every 5 minutes to achieve online learning. The process parameters are dynamically optimized in combination with the real-time monitoring data. When the predicted dust concentration exceeds the set threshold, the relevant process parameters are automatically adjusted. At the same time, a safety protection mechanism is set up. When the parameter adjustment range exceeds the preset range, an alarm is triggered and the adjustment range is restricted to ensure the safe and stable operation of the production process. In the dust control optimization of the calcium carbonate production line, first, the optimized process parameters are sent to each device PLC through the OPC UA protocol. For example, parameters such as the calcination temperature of 953°C and pressure of 0.31 MPa are transmitted to the kiln control system. A dual-channel redundancy design is adopted. The main channel uses industrial Ethernet, and the standby channel uses 4G wireless network. The data integrity is ensured through 32-bit CRC check. When a communication interruption exceeding 100 ms is detected, it automatically switches to the standby channel. The parameters of each process are adjusted in real time. For example, in the kiln temperature control, the adaptive model predictive control algorithm is used, with a sampling period of 1 s, a prediction time domain of 60 s, and a control time domain of 10 s. The transfer function parameters are identified online through the least squares method to predict the future temperature change and optimize the control sequence. The upper limit of temperature adjustment is set to ±5°C / min to prevent over-adjustment. In terms of dust monitoring, light scattering and beta-ray dust detectors are installed at key points such as raw material handling, calcination, and screening, with a measurement range of 0 - 1000 mg / m³ and an accuracy of ±2%. The data is collected at a frequency of 1Hz, and outliers are removed through the 3σ criterion, and 5-point moving average filtering is applied. The processed data is transmitted to the central control system, and an association model between dust concentration and process parameters is constructed using the C4.5 decision tree algorithm. A 1-hour sliding window is used, and the model is updated every 5 minutes. The maximum depth of the tree is set to 5 layers.When the predicted dust concentration exceeds 50 mg / m³, automatically adjust the relevant process parameters. For example, if the predicted dust concentration in the calcination process reaches 55 mg / m³, then reduce the furnace temperature by 2°C. At the same time, set up a safety protection mechanism to limit the temperature adjustment range to no more than ±10°C / h to ensure the safe and stable operation of the production process.

[0042] Dynamically monitor the dust concentration, analyze it, obtain the change curve of the dust concentration after optimizing the process, compare the dust concentration in the change curve with the parameters of each process adjusted at the corresponding time to form feedback, analyze the error of the feedback, compare the error of the feedback with the preset error threshold for judgment, trigger the corresponding preset dust control strategy, and adjust each process parameter according to the feedback to reduce or increase the dust concentration to achieve dust control.

[0043] Obtain the dust concentration data collected by the online dust monitor. The dust concentration data is transmitted from the online dust monitor set at the key points of the production line to the central control system at a preset frequency; perform preprocessing on the dust concentration data, and the preprocessing includes removing outliers using the median filtering algorithm and applying wavelet transform for noise reduction; process the preprocessed dust concentration data using the autoregressive integrated moving average model to obtain the curve of the dust concentration changing with time; use the Granger causality test tool to compare and analyze the curve of the dust concentration changing with time and the data of each process parameter to determine the strength of the Granger causal relationship between each process parameter and the dust concentration; according to the strength of the Granger causal relationship, calculate the adjustment amount of the process parameter using the adaptive fuzzy control algorithm, and the adaptive fuzzy control algorithm is calculated based on the dust concentration error value and the error change rate; send the adjustment amount of the process parameter to each process device through the distributed control system to achieve automatic adjustment of the process parameter.

[0044] Specifically, an online dust monitor is used to monitor the dust concentration at key points of the production line in real time, and the monitoring data is transmitted to the central control system through the data acquisition module at a frequency of 10Hz. The collected data is preprocessed, including removing outliers using median filtering and applying wavelet transform for noise reduction. An autoregressive integrated moving average model is used to process the preprocessed data to generate a curve graph of the dust concentration changing over time. According to the dust concentration change curve, the dust concentration data is compared and analyzed with the process parameter data in the same time period through the Granger causality test tool. The Granger causality strength between each process parameter and the dust concentration is calculated to form a parameter influence quantification index. A parameter influence threshold is set to screen out the key process parameters that significantly affect the dust concentration. A dust concentration error threshold is set, and an exponentially weighted moving average control chart is used to monitor the dust concentration in real time. The upper control limit UCL and the lower control limit LCL are calculated. When it is detected that the dust concentration exceeds the control limit, a preset dust control strategy is triggered. According to the parameter influence index, the process parameters that need to be adjusted are determined, and the adjustment priority is set. Based on the adaptive fuzzy control algorithm, combined with the dust concentration error value and the error change rate, the adjustment amount of the process parameters is calculated. Through an online learning mechanism, the fuzzy rule base is dynamically updated according to the historical adjustment effect. Considering the adjustment constraint conditions of the process parameters, such as the temperature change rate limit, the pressure adjustment range, etc., to ensure that the parameter adjustment is within a safe range. The adjustment instructions are sent to each process equipment through a distributed control system to realize the automatic adjustment of the process parameters and form a closed-loop feedback mechanism for dust control. During the optimization process of dust control in the calcium carbonate production line, first, laser scattering dust monitors are installed at key processes such as raw material processing, calcination, and screening, with a measurement range of 0 - 1000mg / m³ and an accuracy of ±1%. Data is collected at a frequency of 10Hz and transmitted to the central control system through the OPCUA protocol. The 5-point median filter is applied to the original data to remove outliers, and then the db4 wavelet is used for 3-layer decomposition for noise reduction. The ARIMA(2,1,1) model is used to process the data to generate a dust concentration change curve. Subsequently, the relationship between the dust concentration and the process parameters is analyzed through the Granger causality test, with the lag order set to 10 and the significance level α = 0.05. The calculated Granger causality strengths of the calcination temperature, pressure, and raw material particle size are 0.85, 0.72, and 0.63 respectively, exceeding the preset threshold of 0.6, and are determined as key parameters. An EWMA control chart for dust concentration is set, with a smoothing coefficient λ = 0.3 and a control limit coefficient L = 3. The calculated UCL = 55mg / m³ and LCL = 35mg / m³. When it is detected that the dust concentration exceeds the control limit for 3 consecutive points, the control strategy is triggered. Based on the adaptive fuzzy control algorithm, a 7×7 fuzzy rule matrix is designed, with the input variables being the dust concentration error e and the error change rate ec, and the output being the parameter adjustment amount.For example, when e = +10 mg / m³ and ec = +2 mg / m³ / min, the adjustment amount of the calcination temperature is -5°C. By sliding the time window with a window size of 30 minutes, the historical adjustment effect is evaluated, and the fuzzy rules are updated dynamically. Considering process constraints, such as the temperature change rate being limited within ±10°C / h. Finally, the optimized parameters are sent to the on-site PLC through the ModbusTCP protocol to achieve automatic adjustment of process parameters and form a closed-loop feedback mechanism for dust control.

[0045] As described above, this is only the specific implementation manner of this specification. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, modules, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this specification is not limited thereto. Any person skilled in the art within the technical scope disclosed in this specification can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of this specification.

Claims

1. A dust control method in a calcium carbonate manufacturing process, characterized in that, The method includes: during the calcium carbonate manufacturing process, installing an infrared spectrometer to monitor the chemical composition data of dust, a laser particle size analyzer to monitor the particle size data of dust, and using a pH meter to monitor the pH value data of dust, so as to obtain the dust monitoring data during the calcium carbonate manufacturing process. At the same time, obtain the temperature, pressure, flow rate, and rotation speed process parameters of the raw material treatment, calcination, digestion, and drying processes, and construct a dataset of dust monitoring data and process parameters; preprocess the dataset of dust monitoring data and process parameters, fuse data from different sources, different formats, and different timestamps, and normalize the data into a form suitable for time series analysis; extract the dust monitoring data and process parameter feature subsets from the preprocessed dataset of dust monitoring data and process parameters, combine the dust attributes including chemical composition, particle size, and pH value of dust with the process parameters including temperature, pressure, flow rate, and rotation speed, and screen out the feature combinations; obtain the frequent association patterns between dust attributes and process parameters based on the feature combinations, set parameters through a preset association rule mining algorithm and a preset algorithm, obtain the frequent association rules, and obtain a dust and process association rule library according to the analysis results of dust attributes and process parameters in the time series; according to the mined dust and process association rule library, combined with a preset calcium carbonate production process flow chart, judge the dust generation reasons for each process, determine the dust control points for the key processes of raw material treatment, calcination, digestion, and drying, and formulate improvement measures for the process of equipment parameters, product formula, and transportation method; under the preset constraints of production efficiency, cost control, and quality stability, perform an optimization search on the equipment decision variable parameters, product formula, and transportation method to obtain a balanced solution for dust control and production management, and form an improved process plan; issue the optimized process plan to the control systems of each production equipment, adjust the temperature, pressure, and flow rate parameters of each process in real time, and dynamically monitor the change of dust concentration to form a closed-loop optimization control of dust.

2. The method according to claim 1, characterized in that During the calcium carbonate manufacturing process, installing an infrared spectrometer to monitor the chemical composition data of dust, a laser particle size analyzer to monitor the particle size data of dust, and using a pH meter to monitor the pH value data of dust, so as to obtain the dust monitoring data during the calcium carbonate manufacturing process. At the same time, obtain the temperature, pressure, flow rate, and rotation speed process parameters of the raw material treatment, calcination, digestion, and drying processes, and construct a dataset of dust monitoring data and process parameters, including: obtaining the chemical composition data of dust monitored by the infrared spectrometer and the particle size data of dust monitored by the laser particle size analyzer; constructing a dust feature matrix according to the chemical composition data of dust and the particle size data of dust, where the dust feature matrix includes a multi-dimensional feature vector of the chemical composition percentage and particle size distribution of each dust sample; performing principal component analysis on the dust feature matrix, and selecting the principal components with the cumulative contribution rate reaching the preset threshold as the dust key feature vectors. Establish A non - linear mapping model between process parameters and the key feature vector of dust. The non - linear mapping model is constructed using a random forest regression tool, where the process parameters are used as input variables and the key feature vector of dust is used as the target variable; Calculate the importance scores of each process parameter on the dust characteristics according to the non - linear mapping model, and determine the key process parameters that most significantly affect the dust characteristics; construct a gradient - boosting decision tree tool, taking the key process parameters and the key feature vector of dust as input features, and the pH value as the prediction target, where the pH value is monitored and obtained by a pH meter; use the decision tree tool to analyze the multi - dimensional associated data set containing process parameters, dust characteristics and pH values, and establish a relationship model between process parameters and dust characteristics, where the process parameters are used as feature nodes and the dust characteristics and pH values are used as leaf nodes.

3. The method according to claim 1, wherein, Pre - process the data set of dust monitoring data and process parameters, fuse data from different sources, different formats and different time stamps, and normalize the data into a form suitable for time - series analysis, including: obtain dust monitoring data and process parameter data, and use an adaptive moving median filter to preliminarily remove noise from the data to obtain a preliminarily denoised data set; According to the preliminarily denoised data set, combine the Z - score method to identify outliers, and judge whether the Z - score value of the data point exceeds the preset threshold range. If it exceeds, it is marked as an outlier; For outliers, use the local linear regression method for correction to obtain a corrected data set; Unify the format of the corrected data set through a data conversion tool, convert the data of the infrared spectrometer, laser particle size analyzer and pH meter into a standardized JSON format to obtain a standardized data set; use the Kalman filtering algorithm to perform multi - source data fusion on the standardized data set, and generate a unified data stream based on the measurement errors and time delays of different data sources; Use a time - stamp alignment tool to synchronize the unified data stream, interpolate all data at a unified time interval, and use the cubic spline interpolation method to fill in the missing time - point data to obtain a time - aligned and normalized time - series data set.

4. The method according to claim 1, characterized in that, Extract the dust monitoring data and process parameter feature subsets from the pre - processed data set of dust monitoring data and process parameters. Combine the dust attributes including dust chemical composition, particle size, pH value attributes and the process parameters including temperature, pressure, flow rate, rotation speed, and screen out the feature combinations, including: according to the pre - processed data set of dust monitoring data and process parameters, standardize the data to obtain standardized data; use the kernel principal component analysis method to extract the feature subset of the standardized data, and obtain the principal components with a cumulative contribution rate reaching the preset percentage as the feature subset by calculating the kernel matrix eigenvalues and eigenvectors of the standardized data; For process parameter data, calculate the non - linear correlation between temperature, pressure, flow rate and rotation speed using the maximum information coefficient. If the maximum information coefficient is greater than the threshold, select the parameter with the largest amount of information as the representative to obtain the process parameter feature subset; Combined with the chemical composition, particle size, and pH value attributes of dust, evaluate the relationship between the attributes and the subset of process parameter features, select the attribute with the highest mutual information value, and combine it with the subset of process parameter features to form a preliminary feature combination; use the random forest feature importance evaluation tool to screen the preliminary feature combination, calculate the contribution of each feature to the model prediction accuracy to obtain the feature importance ranking; combined with the recursive feature elimination method, gradually remove the least important features, evaluate the model performance after each removal until the performance starts to decline significantly or reaches the preset feature quantity threshold to obtain the optimized feature combination.

5. The method according to claim 1, wherein The frequent association patterns between dust attributes and process parameters are obtained based on the feature combination. By presetting the association rule mining algorithm and setting parameters of the preset algorithm, frequent association rules are obtained, and according to the analysis results of dust attributes and process parameters in the time series, a dust and process association rule library is obtained, including: performing equal-frequency discretization processing on continuous data according to the preset number of intervals to obtain discretized feature data; using the FP-growth algorithm to perform frequent item set mining on the discretized feature data, and recursively generating frequent item sets by constructing an FP tree and a conditional FP tree. Construct association rules according to the frequent item sets using an association rule generator; if the association rules meet the preset threshold, they are determined as strong association rules; obtain a time series dataset, use wavelet transform to extract the frequency domain features of the time series data in the time series dataset to obtain approximate coefficients and detail coefficients at different scales; cluster the strong association rules, construct a rule similarity matrix by calculating the semantic similarity between rule items, and combine with the frequency domain features, and use the hierarchical clustering algorithm to merge the rules with high similarity to form a dust and process association rule library; also include: dividing the dust and process parameter data according to the time series to form a time window, obtaining the analysis results of dust attributes and process parameters in the time series according to the preset time series model, combining with the mined frequent association rules, screening out the association rules that meet the preset practical application value, sorting the screened rules into a database format, and establishing a dust and process association rule library.

6. The method according to claim 5, wherein The dust and process parameter data are divided according to the time series to form a time window, the analysis results of dust attributes and process parameters in the time series are obtained according to the preset time series model, combined with the mined frequent association rules, the association rules that meet the preset practical application value are screened out, and the screened rules are sorted into a database format to establish a dust and process association rule library, including: using the adaptive sliding window method to perform time series segmentation on the dust and process parameter data, and dynamically adjusting the window size according to the data variance within the window; if the variance exceeds the preset threshold, the window is reduced, otherwise the window is enlarged; analyze the dust attributes and process parameter data within the time window, eliminate the data non-stationarity through the seasonal autoregressive integrated moving average model, and use discrete wavelet transform to extract the multi-scale features of the time series. According to the time series analysis results and the frequently mined association rules, set the support threshold, confidence threshold, Kulczynski index threshold, and full confidence index threshold, and determine whether the rules continuously meet all the threshold conditions; weight the rules that meet the conditions to obtain the energy distribution and determine the weight coefficients; convert the weighted rules into a structured data format and establish a relational database table structure including a rule table and a time window table; the rule table contains fields such as rule ID, antecedent, consequent, support, confidence, lift, Kulczynski index, full confidence index, and weight coefficient; the time window table contains fields such as window ID, rule ID, start time, end time, and window characteristics.

7. The method according to claim 1, wherein Based on the mined dust and process association library, combined with the preset calcium carbonate production process flow chart, judge the dust generation reasons for each process, determine the dust control points for the key processes of raw material treatment, calcination, digestion, and drying, and formulate process improvement measures for equipment parameters, product formulas, and transportation methods, including: converting the preset calcium carbonate production process flow chart into a directed graph, where nodes represent processes and edges represent the associations between processes, and the edge weights are set according to the association rules; calculating the shortest paths between nodes in the directed graph to obtain the dust generation causes and influencing factors for each process; according to the dust generation causes and influencing factors, use the fuzzy comprehensive evaluation method combined with principal component analysis to evaluate the four key processes of raw material treatment, calcination, digestion, and drying, and set evaluation indicators for dust concentration, particle size distribution, and chemical composition; extract the main characteristics of the evaluation indicators through principal component analysis, determine the weights of each indicator, calculate the comprehensive scores of each process, and the node with the highest score is determined as the dust control point; for the dust control point, use a multi-objective optimization algorithm to optimize the equipment parameters, product formulas, and transportation methods; use a method combining the particle swarm optimization algorithm and the simulated annealing algorithm to solve the Pareto optimal solution set of the objective function to obtain the optimal parameter combinations for each control point; according to the optimal parameter combinations, use a knowledge graph to construct a process improvement measure library; extract key concepts from the optimal parameter combinations through entity recognition methods, and use relationship extraction to identify the relationships between entities; store the key concepts and the relationships between entities in a graph database and use the graph database query language for relationship reasoning to obtain specific process improvement measures.

8. The method according to claim 1, wherein Under the preset constraints of production efficiency, cost control, and quality stability, optimize and search for the equipment decision variable parameters, product formula, and transportation mode to obtain a balance solution for dust control and production management, and form an improved process plan, including: obtaining the equipment decision variable parameters, product formula, and transportation mode as optimization variables, and the dust control level, production efficiency, cost, and product quality as objective functions according to the preset constraints of production efficiency, cost control, and quality stability; constructing a multi-objective optimization model using the ε-constraint method, selecting the dust control level as the main objective function, and transforming the other objective functions into constraints; using the particle swarm optimization algorithm combined with the simulated annealing algorithm to solve the optimization model, and determining the particle swarm size, maximum number of iterations, initial temperature, and cooling coefficient; obtaining the optimal solution globally by continuously updating the particle position, velocity, and temperature parameters during the annealing process; adopting an adaptive weight adjustment strategy to dynamically adjust the weights of each sub-objective in the objective function according to the results of each iteration; calculating the moving window standard deviation of each sub-objective, and judging the sub-objective with a large standard deviation; if the standard deviation is large, assign it a high weight; achieving dynamic balance between objectives through the weight update formula, and evaluating the Pareto optimality of the current optimal solution after each iteration; performing sensitivity analysis and Monte Carlo simulation on the solution obtained by the optimization algorithm to obtain the sensitivity coefficients of each parameter; using Latin hypercube sampling to generate input parameter combinations and performing Monte Carlo simulation to evaluate the stability and reliability of the optimization results.

9. The method according to claim 1, wherein Sending the optimized process plan to the control systems of each production equipment, adjusting the temperature, pressure, and flow parameters of each process in real time, and dynamically monitoring the change in dust concentration to form a closed-loop optimization control of dust, including: receiving an optimization instruction carrying the process plan parameters, and the optimization instruction is issued by the central controller; sending the process plan parameters to the programmable logic controller of each production equipment through the industrial communication network according to the optimization instruction, and the programmable logic controller adjusts the operating state of the production equipment according to the process plan parameters; setting up the main and standby channels and adopting cyclic redundancy check, and the cyclic redundancy check is used to ensure the accuracy of data transmission; automatically switching to the standby channel if a communication interruption or data transmission error is detected; obtaining the temperature, pressure, and flow parameters of each process and adjusting them in real time; obtaining the dust concentration detection data at each key point of the production line, and the dust concentration detection data is collected by various types of dust detection equipment; collecting the dust concentration detection data at a preset frequency through the data acquisition module and performing data preprocessing; constructing an association model between the dust concentration and process parameters; Dynamically optimize process parameters according to the association model, where the dynamic optimization is used to adjust relevant process parameters; it also includes: dynamically monitoring the dust concentration, analyzing it, obtaining the change curve of the dust concentration after the optimized process, comparing the dust concentration in the change curve with the adjusted parameters of each process at the corresponding time to form feedback, analyzing the error of the feedback, comparing the error of the feedback with a preset error threshold for judgment, triggering the corresponding preset dust control strategy, and adjusting each process parameter according to the feedback to reduce or increase the dust concentration to achieve dust control.

10. The method according to claim 9, characterized in that The dynamic monitoring of the dust concentration, analyzing it, obtaining the change curve of the dust concentration after the optimized process, comparing the dust concentration in the change curve with the adjusted parameters of each process at the corresponding time to form feedback, analyzing the error of the feedback, comparing the error of the feedback with a preset error threshold for judgment, triggering the corresponding preset dust control strategy, and adjusting each process parameter according to the feedback to reduce or increase the dust concentration to achieve dust control includes: obtaining the dust concentration data collected by the on-line dust monitor, and the dust concentration data is transmitted to the central control system by the on-line dust monitor set at the key points of the production line at a preset frequency; preprocessing the dust concentration data, and the preprocessing includes removing outliers by using the median filtering algorithm and denoising by applying wavelet transform; Processing the preprocessed dust concentration data by using the autoregressive integrated moving average model to obtain the curve graph of the dust concentration changing with time; using the Granger causality test tool to compare and analyze the curve graph of the dust concentration changing with time and the data of each process parameter to determine the Granger causality strength between each process parameter and the dust concentration; according to the Granger causality strength, calculating the adjustment amount of the process parameters by using the adaptive fuzzy control algorithm, and the adaptive fuzzy control algorithm is calculated based on the dust concentration error value and the error change rate; sending the adjustment amount of the process parameters to each process device through the distributed control system to realize the automatic adjustment of the process parameters.

Images