An intelligent control method for concentrating and dosing coal sludge water

Abstract

The present invention relates to an intelligent control method for coal slurry water concentration and dosing, which belongs to the field of intelligent control and includes the following steps: (a) collecting various parameters of the sedimentation process of the concentration tank; (b) preprocessing and feature engineering the collected data; (c) analyzing the correlation between the process parameters of the concentration tank and the sedimentation index, and calculating the maximum information coefficient between the variables through the MINE algorithm; (d) constructing a multi-algorithm prediction meta-model, and obtaining the optimal model weight combination through model fusion; (e) according to the real-time process parameters of the concentration tank and the height of the clear water layer and the overflow turbidity at the current moment, on the basis of the prediction meta-model, establishing an optimization model based on a genetic algorithm, and automatically adjusting the dosing frequency of the concentration tank under the premise of meeting the process parameter control index. The advantages of the present invention are: comprehensively considering the control conditions of the history and current period of the concentration tank, according to the model trained by historical data, the automatic control of the dosing system of the concentration tank is realized, and the dosing control operation can be timely and accurate.

Description

Technical Field

[0001] The invention relates to an intelligent control method for coal sludge water concentration and dosing, and belongs to the field of intelligent control. Background Art

[0002] The turbidity of coal slurry water is an important indicator in the water resource circulation process of coal preparation plants. Turbid circulating water will affect the production of the main production lines of coal preparation plants. The turbidity of coal slurry water is affected by many factors, such as the quality of raw coal, the amount of washing, the amount of flocculants and coagulants added, etc., which will affect the sedimentation effect of coal slurry water; in addition, during the sedimentation of coal slurry water, the height of the clear water layer in the concentration tank determines the current operating state of the concentration tank. If the height of the clear water layer is too low, that is, the current coal slurry feed exceeds the upper limit of the current operating state of the concentration tank, it is necessary to adjust the dosing frequency or reduce the feed amount; in addition, there are also large differences in the coal slurry water recovery process of coal preparation plants. Therefore, it is necessary to comprehensively consider different processes and adjust the dosing frequency under different production conditions. For example, in CN1958116A, the industrial computer determines whether to add drugs based on the coal slurry water concentration and flow data obtained by the concentration meter and flow meter and the overflow water concentration obtained by the sensor. CN105174402A first calculates the optimal dosage of flocculants and coagulants according to the total dry coal slime amount, and adjusts the real-time dosage of the reagents according to the actual coal slime water concentration and flow rate and the data information fed back by the concentration monitoring system in the coal slime water deep clarification system.

[0003] The above scheme only considers the concentration and flow rate of coal slurry water and the clear water layer of the thickening tank as the main control parameters. It does not establish an analysis model based on historical trends, and cannot comprehensively calculate the optimal dosage of coal slurry water overflow concentration and the height of the clear water layer of the thickening tank. Since it takes a certain amount of time for the control system to respond to the thickening tank from execution, if the adjustment frequency is too frequent, it may cause system indicators to fluctuate and fail to achieve steady state. Therefore, for the automatic control scenario of thickening and dosing,

[0004] It is necessary to comprehensively consider not only the overflow concentration and clear water layer height of the thickening tank, but also the feed coal slurry concentration and flow rate, rake pressure state, underflow concentration, control response speed and changes in process parameters under different production conditions.

[0005] At present, coal preparation plants mainly rely on manual experience to carry out water treatment and water circulation operations during the slime water concentration process. Water treatment usually achieves the effect of controlling the separation of mud and water by adjusting the dosage in the concentration tank to control the sedimentation rate, which requires high experience for the operator. In actual operation, the water treatment effect fluctuates greatly, resulting in limited water circulation efficiency. Existing technologies have begun to apply some models based on production data to predict water treatment effects in order to better cope with changes in operating parameters, but the applicability of these models is still limited. They are often only designed for specific types of slime water sedimentation processes and are difficult to fully cover various complex production conditions. In addition, there are also large differences in the slime water treatment process in different coal preparation plants. For example, some coal preparation plants directly reuse the concentrated clean water in the production process, while other coal preparation plants need to undergo further treatment processes such as filtration or chemical treatment before the clean water is reused. This process difference leads to uneven effects and efficiency of the water circulation process, and more accurate data analysis and control methods are needed to ensure the stability of water quality and the improvement of recovery rate during the slime water concentration process. In summary, the main challenges faced by coal preparation plants in slime water concentration and water circulation are that operations rely on manual experience, the level of automation and intelligence is insufficient, and different processes have different requirements for slime water treatment. These problems lead to the instability of slime water treatment effects and fluctuations in production efficiency. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a coal slurry clear water layer and overflow turbidity prediction model and a dosing control optimization method based on historical data analysis and modeling, and reversely guides the optimal setting value of each agent dosage according to the key indicators of predicting coal slurry water concentration. The coal slurry water parameter prediction model is based on the on-site production data and process conditions, constructs the corresponding relationship between the production data and the coal slurry clear water layer height and overflow turbidity, and weights the prediction results according to the onset time to obtain the changes in the main technical indicators of the concentration tank in the future (clear water layer height, overflow turbidity). Combined with the changes in this part, the parameter optimization method is used to calculate the optimal and minimum dosing frequency of the concentration tank under the condition of meeting the parameter limitation conditions.

[0007] In order to overcome the defects of the prior art, the present invention provides an intelligent control method for coal slurry water concentration and dosing. The technical solution of the present invention is:

[0008] An intelligent control method for coal slime water concentration and dosing, comprising the following steps:

[0009] (a) Collect various parameters of the thickener settling process, including coal slime feed flow, feed concentration, underflow concentration, underflow flow, thickener rake pressure, current thickener dosing frequency, reagent concentration, current thickener clear water layer height and overflow turbidity;

[0010] (b) Preprocessing and feature engineering of the collected data;

[0011] (c) Analyze the correlation between the process parameters of the concentration tank and the sedimentation index, and calculate the maximum information coefficient between the variables using the MINE algorithm;

[0012] (d) Construct a multi-algorithm prediction meta-model and obtain the optimal model weight combination through model fusion;

[0013] (e) According to the real-time process parameters of the concentrator and the height of the clear water layer and the overflow turbidity at the current moment, an optimization model based on genetic algorithm is established on the basis of the prediction meta-model to automatically adjust the dosing frequency of the concentrator under the premise of meeting the process parameter control indicators.

[0014] The step (a) is specifically:

[0015] 1.1 Use measuring equipment to monitor the coal slime feed flow in real time to ensure the real-time and accuracy of data;

[0016] 1.2 Measure the feed concentration through an online concentration meter to obtain the solid content information of the coal slurry water;

[0017] 1.3 Install a concentration sensor at the bottom flow outlet of the concentration tank to monitor the bottom flow concentration in real time;

[0018] 1.4 Measure the underflow flow rate through the flow meter to obtain the volume flow data of the underflow of the concentration tank;

[0019] 1.5 Install pressure sensors to monitor the rake pressure of the thickener to assess the solid accumulation in the bottom flow of the thickener;

[0020] 1.6 Obtain the current dosing frequency through the control unit of the dosing system to understand the dosing status of the agent;

[0021] 1.7 Measure the concentration of the reagent to ensure that the concentration of the reagent during dilution and addition meets the process requirements;

[0022] 1.8 Use water level sensor to monitor the height of the clear water layer in the current concentration tank;

[0023] 1.9 Use a turbidity meter to monitor the overflow turbidity of the current thickener to evaluate the sedimentation effect of the thickener.

[0024] The step (b) is specifically:

[0025] 2.1 Data cleaning, removing outliers and missing values;

[0026] 2.2 Data conversion: converting raw data into a format suitable for model training;

[0027] 2.3 Feature selection: select the features most relevant to the prediction target;

[0028] 2.4 Feature construction, creating new features to improve the predictive ability of the model;

[0029] 2.5 Data segmentation: Divide the dataset into training set, validation set, and test set for model training and evaluation.

[0030] 4. The intelligent control method for concentrating and dosing coal slurry according to claim 1 or 2, characterized in that the step (c) is specifically:

[0031] 3.1 Select variables: determine the relevant process parameters and sedimentation indicators for analysis, including but not limited to coal slime feed flow, feed concentration, underflow concentration, underflow flow, thickener rake pressure, thickener current dosing frequency, reagent concentration, current thickener clear water layer height and overflow turbidity; 3.2 Data preparation: input the data set of the selected variables into the MINE algorithm to calculate the mutual information between the variables;

[0032] 3.3 Calculation of MIC: The MINE algorithm was used to calculate the maximum information coefficient between each pair of variables to assess the strength of the linear and nonlinear relationships between them;

[0033] 3.4 Variable screening: select the process parameters related to sedimentation index according to the MIC value. The higher the MIC value, the stronger the correlation between the variables.

[0034] 3.5 Correlation analysis: further analysis of the selected variables was conducted to determine their specific relationships with the settlement indicators;

[0035] 3.6 Feature construction, based on the results of correlation analysis, construct or select feature sets for subsequent model training to improve the accuracy of model prediction of clear water layer height and overflow turbidity.

[0036] The step (d) is specifically:

[0037] 4.1 According to the results of data correlation analysis, the parameters related to each prediction target are obtained, and a multi-algorithm model is established based on it, and then combined into a meta-model in a weighted manner;

[0038] 4.2 Each meta-model predicts various parameters at different times, namely, feed concentration, instantaneous feed flow, concentration at different liquid levels, height of clear water layer in the thickener, overflow turbidity in the thickener, rake pressure in the thickener, dosing frequency of dosing pump in the thickener, height of clear water layer and overflow turbidity at the current moment, and predicts the change value of overflow turbidity of clear water layer height after a fixed time:

[0039] H n =F n (X);

[0040] D n =G n (X);

[0041]

[0042] Where n is the number of base models, X is the process parameter of the concentration tank in the current period, Fn and Gn are the base models (XGBoost, RandomForest, LightGBM, GBDT, etc.) trained by historical data and applying Bayesian optimization hyperparameters to obtain the trained base algorithms; Wn is the weight of each model, Hn and Dn are the predicted clear water layer height and overflow turbidity of the nth model, respectively, and H and D are the final predicted values ​​after weighting;

[0043] 4.3 Construct a calculation function that combines the clean water layer height optimization objective and the overflow turbidity optimization objective, and assign different weights to each objective for importance adjustment:

[0044] P=min m1·|f1(x)-T1|+m2·|f2(x)-T2|;

[0045] Among them, m1 and m2 are the weight coefficients of the objective function, and the sum of the weights should be 1, that is, m1+m2=1; T1 is the ideal target value of the clear water layer height, T2 is the ideal target value of the overflow turbidity, x is the dosing frequency, and P is the deviation of the comprehensive optimization target. The overall goal is to minimize the deviation. The weight coefficient is determined by the correlation coefficient between the dosing frequency and the optimization target obtained in the correlation analysis.

[0046] The advantages of the present invention are:

[0047] 1. Taking into account the historical and current control conditions of the concentration tank, the model trained based on historical data is used to realize the automatic control of the concentration tank dosing system. By comprehensively considering the changing trends of the main parameter indicators, the dosing control operation can be timely and accurate.

[0048] 2. Model the main influencing parameters, analyze the impact of individual parameters on overflow turbidity and clear water layer height, use the maximum information coefficient (MIC) calculation to obtain the impact of each parameter on the two main indicators, and combine actual operating experience and process principles to conduct preliminary data screening to improve the prediction effect.

[0049] 3. Multi-algorithm fusion prediction, by integrating multiple machine learning model algorithms and performing training optimization. According to the hysteresis of the dosing reaction time of the concentration tank, that is, multiple modeling is performed for multiple time periods in the future and weighted to obtain the final prediction model. This method can better reflect the changes in the sedimentation of the concentration tank and improve the prediction accuracy.

[0050] 4. Optimize the target fusion, by combining the prediction indicators and adjusting the weights, simplify the multi-objective optimization problem into a single-objective optimization problem, so that the overflow turbidity and the height of the clear water layer reach an ideal balance.

[0051] 5. Genetic algorithm optimization: Through genetic algorithm, the dosing frequency of the concentration tank is optimized to obtain the minimum dosing frequency that meets the overall optimization goal. This method can realize real-time dosing frequency control based on future changes in the concentration tank and comprehensive consideration of multiple indicator restrictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0053] The present invention will be further described below in conjunction with specific embodiments, and the advantages and features of the present invention will become clearer as the description proceeds. However, these embodiments are exemplary only and do not constitute any limitation to the scope of the present invention. It should be understood by those skilled in the art that the details and forms of the technical solution of the present invention may be modified or replaced without departing from the spirit and scope of the present invention, but these modifications and replacements all fall within the scope of protection of the present invention.

[0054] See also Figure 1 The present invention relates to an intelligent control method for coal slime water concentration and dosing, comprising the following steps:

[0055] (a) Collect various parameters of the thickener settling process, including coal slime feed flow, feed concentration, underflow concentration, underflow flow, thickener rake pressure, current thickener dosing frequency, reagent concentration, current thickener clear water layer height and overflow turbidity;

[0056] (b) Preprocessing and feature engineering of the collected data;

[0057] (c) Analyze the correlation between the process parameters of the concentration tank and the sedimentation index, and calculate the maximum information coefficient between the variables using the MINE algorithm;

[0058] (d) Construct a multi-algorithm prediction meta-model and obtain the optimal model weight combination through model fusion;

[0059] (e) According to the real-time process parameters of the concentration tank and the height of the clear water layer and the overflow turbidity at the current moment, a genetic algorithm-based optimization model is established on the basis of the prediction meta-model to automatically adjust the concentration tank dosing frequency on the premise of meeting the process parameter control indicators. Among them, when establishing a genetic algorithm-based optimization model, the following specific steps are usually involved:

[0060] Initialize the population: Generate an initial population, where each individual is usually represented by a string of codes (such as binary codes). The number of individuals in the population and the length of the code depend on the specific requirements of the problem.

[0061] Evaluate fitness: Calculate the fitness of each individual. Individuals with high fitness are more likely to be selected to generate the next generation. The fitness function is usually directly related to the objective function of the problem and is used to evaluate the quality of individuals.

[0062] Selection process: Select several excellent individuals according to their fitness to prepare for crossover and mutation. Common selection methods include roulette selection, tournament selection, etc.

[0063] Crossover operation: Pair selected individuals in some way (such as single-point crossover, multi-point crossover, uniform crossover, etc.) and exchange some of their genes to produce new offspring. The crossover operation helps explore new solution spaces.

[0064] Mutation operation: modify certain genes of newborn individuals with a small probability to introduce mutations and increase the diversity of the population. The mutation operation can be expressed as: for gene x nk , the mutation operation can be expressed as x′ nk =x nk +δ, where δ is a small random change and the mutation rate is usually represented by μμ.

[0065] Generate a new population: From the excellent individuals of the previous generation and the newly generated individuals, a new population is formed. Various strategies can be used to form a new population, such as elite strategy, generation replacement, etc.

[0066] Termination condition: If the preset termination condition (such as the maximum number of iterations or fitness threshold) is reached, the iteration is stopped. Otherwise, the fitness is evaluated and the next generation of evolution is continued. In practical applications, the parameter settings of the genetic algorithm (such as population size, crossover rate, mutation rate, etc.) have an important impact on the performance of the algorithm. In addition, the genetic algorithm can be combined with other optimization techniques to form a hybrid algorithm to improve the convergence speed and maintain the diversity of the algorithm.

[0067] Based on the above steps (a) to (e), the following are achieved:

[0068] Improve sedimentation efficiency: By real-time monitoring of multiple key parameters of the thickening tank, such as coal slime feed flow rate, feed concentration, etc., and using intelligent algorithms for data analysis, the dosing frequency and reagent concentration can be controlled more accurately, thereby improving the sedimentation efficiency of coal slime water.

[0069] Optimize the use of chemicals: The intelligent control system can automatically adjust the dosage of chemicals according to the real-time data of the concentration tank, reduce excessive use of chemicals, reduce production costs, and reduce environmental pollution.

[0070] Reduce manual intervention: The automated intelligent control system reduces dependence on manual operation, reduces workers' labor intensity, and improves production safety.

[0071] Improve system stability and reliability: Through real-time monitoring and intelligent control, the system can quickly respond to changes in the properties of coal slurry water, adjust the dosing strategy in time, and ensure the stable operation of the concentration tank.

[0072] Data-driven decision-making: Using the MINE algorithm to calculate the maximum information coefficient between variables can better understand the relationship between process parameters and sedimentation indicators, thereby making more scientific decisions.

[0073] Model fusion technology: By building a multi-algorithm prediction meta-model and fusing the models to obtain the optimal weight combination, the accuracy and robustness of the prediction can be improved.

[0074] Optimization of genetic algorithm: The optimization model based on genetic algorithm can automatically adjust the dosing frequency of the concentration tank and further optimize the dosing process while meeting the process parameter control indicators.

[0075] The step (a) is specifically:

[0076] 1.1 Use measuring equipment to monitor the coal slime feed flow in real time to ensure the real-time and accuracy of data;

[0077] 1.2 Measure the feed concentration through an online concentration meter to obtain the solid content information of the coal slurry water;

[0078] 1.3 Install a concentration sensor at the bottom flow outlet of the concentration tank to monitor the bottom flow concentration in real time;

[0079] 1.4 Measure the underflow flow rate through the flow meter to obtain the volume flow data of the underflow of the concentration tank;

[0080] 1.5 Install pressure sensors to monitor the rake pressure of the thickener to assess the solid accumulation in the bottom flow of the thickener;

[0081] 1.6 Obtain the current dosing frequency through the control unit of the dosing system to understand the dosing status of the agent;

[0082] 1.7 Measure the concentration of the reagent to ensure that the concentration of the reagent during dilution and addition meets the process requirements;

[0083] 1.8 Use water level sensor to monitor the height of the clear water layer in the current concentration tank;

[0084] 1.9 Use a turbidity meter to monitor the overflow turbidity of the current thickener to evaluate the sedimentation effect of the thickener.

[0085] The step (b) is specifically:

[0086] 2.1 Data cleaning, removal of outliers and missing values; including but not limited to: identifying and processing data points that are beyond the normal operating range; filling or deleting missing data records; identifying and correcting obvious errors or deviations.

[0087] 2.2 Data conversion: converting raw data into a format suitable for model training; including but not limited to: normalizing or standardizing numerical data to eliminate the impact of different dimensions; encoding categorical data, such as using one-hot encoding or label encoding; converting time series data into a format suitable for analysis.

[0088] 2.3 Feature selection: select the features most relevant to the prediction target; including but not limited to: using statistical tests to evaluate the correlation between each feature and the prediction target; using feature selection algorithms such as recursive feature elimination (RFE) or model-based feature selection; manually selecting key features based on domain knowledge and process principles.

[0089] 2.4 Feature construction, creating new features to improve the predictive ability of the model; including but not limited to: deriving new features from existing data, such as calculating derived variables or interaction terms; using data transformation techniques, such as logarithmic transformation or polynomial expansion; combining multiple features to form composite features.

[0090] 2.5 Data segmentation: Divide the dataset into training set, validation set, and test set for model training and evaluation.

[0091] The step (c) is specifically:

[0092] 3.1 Select variables: determine the relevant process parameters and sedimentation indicators for analysis, including but not limited to coal slime feed flow, feed concentration, underflow concentration, underflow flow, thickener rake pressure, thickener current dosing frequency, reagent concentration, current thickener clear water layer height and overflow turbidity; 3.2 Data preparation: input the data set of the selected variables into the MINE algorithm to calculate the mutual information between the variables;

[0093] 3.3 Calculation of MIC: The MINE algorithm was used to calculate the maximum information coefficient between each pair of variables to assess the strength of the linear and nonlinear relationships between them;

[0094] 3.4 Variable screening: process parameters with significant correlation with sedimentation indicators (clear water layer height and overflow turbidity) were screened out according to the MIC value. The higher the MIC value, the stronger the correlation between the variables.

[0095] 3.5 Correlation analysis: further analysis of the selected variables was conducted to determine their specific relationships with the settlement indicators, including the direction of correlation (positive or negative correlation) and the degree of correlation;

[0096] 3.6 Feature construction, based on the results of correlation analysis, construct or select feature sets for subsequent model training to improve the accuracy of model prediction of clear water layer height and overflow turbidity.

[0097] The step (d) is specifically:

[0098] 4.1 Based on the results of data correlation analysis, the parameters related to each prediction target are obtained, and a multi-algorithm model is established based on this, and then combined into a meta-model in a weighted manner; the steps for establishing the multi-algorithm model generally include:

[0099] Data preprocessing: Before performing correlation analysis, the data needs to be cleaned and preprocessed to ensure the quality and consistency of the data. This may include steps such as handling missing values, outliers, and data conversion.

[0100] Correlation analysis: Use statistical methods (such as Pearson correlation coefficient, Spearman rank correlation coefficient, etc.) to analyze the correlation between different parameters. The purpose of this step is to identify the parameters that are most relevant to the prediction target. Correlation analysis can help us understand the relationship between variables and how they affect each other. For example, the Pearson correlation coefficient can measure the linear relationship between two continuous variables, while the Spearman rank correlation coefficient is suitable for non-normally distributed data or ordered categorical data.

[0101] Feature selection: Based on the results of correlation analysis, the parameters most relevant to the prediction target are selected as features for subsequent model training.

[0102] Model selection and training: Choose multiple different algorithms to build the model, including linear regression, decision tree, support vector machine, neural network, etc. Each algorithm has its own unique advantages and applicable scenarios.

[0103] Model optimization: Tune the parameters of each model to improve the performance of the model. This may involve using techniques such as cross-validation and grid search to find the optimal model parameters.

[0104] Model fusion: Combine the prediction results of multiple models in a weighted manner to form a meta-model. Model fusion can improve the accuracy and robustness of predictions. Common fusion methods include simple averaging, weighted averaging, voting, stacking, etc.

[0105] Weight determination: In weighted fusion, the weight of each model needs to be determined. The weight can be assigned based on the performance of the model (such as accuracy, F1 score, etc.), or determined by more complex methods (such as optimization algorithms).

[0106] Meta-model training: Using the selected weights, the meta-model is trained using the predictions of the individual models as input. In stacking, the meta-model can be any machine learning algorithm that learns how to best combine the predictions of the primary models.

[0107] Model validation and evaluation: Evaluate the performance of the meta-model on an independent test set to ensure that the model has good generalization ability. Use appropriate evaluation metrics (such as mean square error, accuracy, etc.) to measure the performance of the model.

[0108] Model deployment: Once the model is validated, it can be deployed in real-world applications for prediction or decision support.

[0109] 4.2 Each meta-model predicts various parameters at different times, namely, feed concentration, instantaneous feed flow, concentration at different liquid levels, height of clear water layer in the thickener, overflow turbidity in the thickener, rake pressure in the thickener, dosing frequency of dosing pump in the thickener, height of clear water layer and overflow turbidity at the current moment, and predicts the change value of overflow turbidity of clear water layer height after a fixed time:

[0110] H n =F n (X);

[0111] D n =G n (X);

[0112]

[0113] Where n is the number of base models, X is the process parameter of the concentration tank in the current period, Fn and Gn are the base models (XGBoost, RandomForest, LightGBM, GBDT, etc.) trained by historical data and applying Bayesian optimization hyperparameters to obtain the trained base algorithms; Wn is the weight of each model, Hn and Dn are the predicted clear water layer height and overflow turbidity of the nth model, respectively, and H and D are the final predicted values ​​after weighting;

[0114] 4.3 Construct a calculation function that combines the clean water layer height optimization objective and the overflow turbidity optimization objective, and assign different weights to each objective for importance adjustment:

[0115] P=min m1·|f1(x)-T1|+m2·|f2(x)-T2|;

[0116] Among them, m1 and m2 are weight coefficients of the objective function, which reflect the importance of each optimization target. The sum of the weights should be 1, that is, m1+m2=1; T1 is the ideal target value of the clear water layer height, T2 is the ideal target value of the overflow turbidity, x is the dosing frequency, P is the deviation of the comprehensive optimization target, and the overall goal is to minimize the deviation. The weight coefficient is determined by the correlation coefficient between the dosing frequency and the optimization target obtained in the correlation analysis.

[0117] The invention provides an intelligent control method for coal sludge water concentration and dosing.

[0118] Firstly, the historical process parameter data of the thickening tank, the clear water layer height and the overflow turbidity of the thickening tank are collected, and data preprocessing and feature engineering are carried out to analyze the correlation between the process parameter data and the prediction indicators. Then, various machine learning and deep learning algorithms are used as the base algorithm, and the hyper-parameter optimization of the base algorithm is carried out in combination with the correlation analysis results and historical data, and a multi-algorithm prediction meta-model of the clear water layer height and overflow turbidity parameters is constructed. Finally, according to the real-time process parameters of the thickening tank and the clear water layer height and overflow turbidity at the current moment, an optimization model based on genetic algorithm is established on the basis of the prediction meta-model, and it is obtained that the dosing frequency of the thickening tank is automatically adjusted under the premise of meeting the control indicators of the main process parameters, so as to realize the automatic and intelligent control of the thickening tank.

[0119] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An intelligent control method for coal slurry water concentration and dosing, characterized in that: The following steps are involved: (a) Collect various parameters of the thickener settling process, including coal slime feed flow, feed concentration, underflow concentration, underflow flow, thickener rake pressure, current thickener dosing frequency, reagent concentration, current thickener clear water layer height and overflow turbidity; (b) Preprocessing and feature engineering of the collected data; (c) Analyze the correlation between the process parameters of the concentration tank and the sedimentation index, and calculate the maximum information coefficient between the variables using the MINE algorithm; (d) Construct a multi-algorithm prediction meta-model and obtain the optimal model weight combination through model fusion; (e) According to the real-time process parameters of the concentrator and the height of the clear water layer and the overflow turbidity at the current moment, an optimization model based on genetic algorithm is established on the basis of the prediction meta-model to automatically adjust the dosing frequency of the concentrator under the premise of meeting the process parameter control indicators.

2. The intelligent control method for coal slurry concentration and dosing according to claim 1 is characterized in that: The step (a) is specifically: 1.1 Use measuring equipment to monitor the coal slime feed flow in real time to ensure the real-time and accuracy of data; 1.2 Measure the feed concentration through an online concentration meter to obtain the solid content information of the coal slurry water; 1.3 Install a concentration sensor at the bottom flow outlet of the concentration tank to monitor the bottom flow concentration in real time; 1.4 Measure the underflow flow rate through the flow meter to obtain the volume flow data of the underflow of the concentration tank; 1.5 Install pressure sensors to monitor the rake pressure of the thickener to assess the solid accumulation in the bottom flow of the thickener; 1.6 Obtain the current dosing frequency through the control unit of the dosing system to understand the dosing status of the agent; 1.7 Measure the concentration of the reagent to ensure that the concentration of the reagent during dilution and addition meets the process requirements; 1.8 Use water level sensor to monitor the height of the clear water layer in the current concentration tank; 1.9 Use a turbidity meter to monitor the overflow turbidity of the current thickener to evaluate the sedimentation effect of the thickener.

3. The intelligent control method for coal slurry concentration and dosing according to claim 1 or 2, characterized in that: The step (b) is specifically: 2.1 Data cleaning, removing outliers and missing values; 2.2 Data conversion: converting raw data into a format suitable for model training; 2.3 Feature selection: select the features most relevant to the prediction target; 2.4 Feature construction, creating new features to improve the predictive ability of the model; 2.5 Data segmentation: Divide the dataset into training set, validation set, and test set for model training and evaluation.

4. The intelligent control method for coal slurry concentration and dosing according to claim 1 or 2, characterized in that: The step (c) is specifically: 3.1 Select variables: determine the relevant process parameters and sedimentation indicators for analysis, including but not limited to coal slime feed flow rate, feed concentration, underflow concentration, underflow flow rate, thickener rake pressure, thickener current dosing frequency, reagent concentration, current thickener clear water layer height and overflow turbidity; 3.2 Data preparation: The dataset of the selected variables is input into the MINE algorithm to calculate the mutual information between the variables; 3.3 Calculation of MIC: The MINE algorithm was used to calculate the maximum information coefficient between each pair of variables to assess the strength of the linear and nonlinear relationships between them; 3.4 Variable screening: screen out process parameters related to sedimentation index according to MIC value. The higher the MIC value, the stronger the correlation between variables. 3.5 Correlation analysis: further analysis of the selected variables was conducted to determine their specific relationships with the settlement indicators; 3.6 Feature construction, based on the results of correlation analysis, construct or select feature sets for subsequent model training to improve the accuracy of the model prediction of clear water layer height and overflow turbidity.

5. The intelligent control method for coal slurry concentration and dosing according to claim 1 or 2, characterized in that: The step (d) is specifically: 4.1 Based on the results of data correlation analysis, the parameters related to each prediction target are obtained, and a multi-algorithm model is established based on this, and then combined into a meta-model in a weighted manner; 4.2 Each meta-model predicts various parameters at different times, namely, feed concentration, instantaneous feed flow, concentration at different liquid levels, height of clear water layer in the thickener, overflow turbidity in the thickener, rake pressure in the thickener, dosing frequency of dosing pump in the thickener, height of clear water layer and overflow turbidity at the current moment, and predicts the change value of overflow turbidity of clear water layer height after a fixed time: H n =F n (X); D n =G n (X); In the formula, n is the number of base models, X is the process parameter of the concentration tank in the current period, Fn and Gn are the hyperparameters of each base model trained by historical data and applying Bayesian optimization to obtain the trained base algorithm; Wn is the weight of each model, Hn and Dn are the predicted clear water layer height and overflow turbidity of the nth model, respectively, and H and D are the final predicted values ​​after weighting; 4.3 Construct a calculation function that combines the clean water layer height optimization objective and the overflow turbidity optimization objective, and assign different weights to each objective for importance adjustment: P=min m1·|f1(x)-T1|+m2·|f2(x)-T2|; in, m1 and m2 are weight coefficients of the objective function, which reflect the importance of each optimization target; the sum of the weights should be 1, that is, m1+m2=1; T1 is the ideal target value of the clear water layer height, T2 is the ideal target value of the overflow turbidity, x is the dosing frequency, P is the deviation of the comprehensive optimization target, and the overall goal is to minimize the deviation, where the weight coefficient is determined by the correlation coefficient between the dosing frequency and the optimization target obtained in the correlation analysis.

Images