Coal powder fineness prediction method based on data driving

Abstract

The invention discloses a pulverized coal fineness prediction method based on data driving, and relates to the technical field of pulverized coal fineness prediction. Comprising the following steps: S1, sampling and analyzing pulverized coal: taking a pulverized coal sample, carrying out hardness test on the pulverized coal, and recording initial fineness and a hardness value; carrying out stress analysis in the transportation process, and testing the stress conditions, including loading and unloading, vibration, impact and the like, of the pulverized coal in each link in the transportation process; data recording: recording the stress data according to a time sequence to form a data set; s2: data processing and modeling: data preprocessing, data cleaning, and extracting key features from the collected data set; and model selection and training: adopting a support vector machine as a basic algorithm, and performing parameter optimization in combination with a particle swarm algorithm. According to the method, the situation that in the conveying process, the finenesses of pulverized coal can be changed based on external force, and consequently the final finenesses and the preset finenesses are different is considered, analysis is conducted on the finenesses, and the obtained final result is more accurate and high in reliability.

Description

Technical Field

[0001] The invention relates to the technical field of coal powder fineness prediction, and in particular to a coal powder fineness prediction method based on data driving. Background Art

[0002] In modern thermal power generation and industrial boilers, coal is the main fuel, and its combustion efficiency directly affects energy utilization efficiency and environmental pollution. The fineness of coal powder is one of the important factors affecting its combustion performance. In traditional methods, the fineness of coal powder is controlled by adjusting the parameters of the coal mill, but this method has many limitations. It is difficult to adjust the fineness of coal powder in real time and accurately to adapt to different combustion requirements; during the transportation process, the fineness of coal powder may change due to external forces, resulting in a difference between the final fineness and the preset fineness.

[0003] After searching, the application scheme of Chinese patent application number CN202311204101.6 discloses a method for predicting the fineness of coal powder in a coal mill, which belongs to the field of coal mill control. It includes the following steps: Step 1: Collect historical data from the operation of the coal mill, directly pre-process the data, take the parameters affecting the fineness of coal powder as input variables, and the fineness of coal powder as output variables, and divide the training set and the test set; Step 2: Construct a coal powder fineness prediction model of the LSTM neural network based on the data of the training set; Step 3: Initialize the LSTM neural network prediction model, and use the particle swarm optimization algorithm to optimize the LSTM neural network model; Step 4: Use the prediction model to predict the test sample, obtain the predicted value of coal powder fineness, and judge the accuracy of the prediction model according to the predicted value. The prediction method in the above patent has the following shortcomings: Although it has the prediction ability, it cannot take into account the changes in the fineness of pulverized coal during transportation, resulting in differences between the fineness at the transportation terminal and the predicted fineness, which needs to be improved. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a coal powder fineness prediction method based on data drive.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A data-driven coal powder fineness prediction method includes:

[0007] S1: Coal pulverized gas sampling and analysis:

[0008] Take a sample of coal powder, perform a hardness test on the coal powder, and record the initial fineness and hardness values;

[0009] Stress analysis during transportation: testing the stress of coal powder in each link during transportation, including loading and unloading, vibration, impact, etc.

[0010] Data recording, recording the force data in chronological order to form a data set.

[0011] S2: Data processing and modeling:

[0012] Data preprocessing, data cleaning, and extracting key features from the collected data sets;

[0013] For model selection and training, support vector machine is used as the basic algorithm, combined with particle swarm algorithm for parameter optimization; radial basis function is selected as kernel function, and the objective function is set to minimize the prediction error; the details are as follows:

[0014]

[0015] in, is the weight coefficient vector, is the mapping function, is the intercept term; introduce relaxation factor and , allowable fitting error , then the model minimizes the objective function through constraints, as follows:

[0016]

[0017]

[0018]

[0019]

[0020] in, is the penalty coefficient, Exceeding the allowable fitting error The number of

[0021] Particle swarm optimization;

[0022] S3: Model evaluation and revision;

[0023] The mean square error and mean absolute error were used to evaluate the model performance. The model was tested using an independent validation set to ensure the generalization ability of the model.

[0024] Analyze the source of model error based on the verification results; adjust the relevant parameters of SVM and PSO for the parts with large errors and retrain the model;

[0025] S4: Final prediction;

[0026] The latest data on coal powder hardness and force during transportation are input into the model; the model outputs the final predicted value of coal powder fineness.

[0027] Preferably: in S1, sampling is specifically as follows:

[0028] S111: Sampling preparation: Select appropriate sampling tools and methods based on the characteristics of the coal powder;

[0029] S112: Sampling: Randomly select coal powder samples from the coal mill outlet;

[0030] S113: Sample quantity control: The weight of each comprehensive sample should be controlled within 100g to facilitate subsequent processing and testing;

[0031] S114: Record information: record sampling time, location, and environmental conditions;

[0032] S115: Preliminary inspection: Inspect the coal powder after sampling to eliminate obvious foreign matter or impurities.

[0033] Preferably: in S1, the force analysis during transportation includes:

[0034] S121: Data collection: Based on sensors installed at the detection position (such as conveyor belts, etc.), various force data of coal powder during transportation are recorded in real time, including impact force, pressure, vibration, etc.;

[0035] S122: Data classification: Classify and organize the collected data by type (such as impact force, pressure, etc.) and source (such as loading and unloading, vibration, etc.);

[0036] S123: Feature extraction: Use signal processing technology to extract key features of various data, such as peak value, mean value, standard deviation, etc., to provide a basis for subsequent analysis;

[0037] S124: Analysis of influencing factors: Study the influence of different factors (such as loading and unloading methods, transportation speed, road conditions, etc.) on the stress of coal powder and identify the main influencing factors;

[0038] S125: Modeling and prediction: Establish a force prediction model based on historical data, input new transportation conditions, and predict the forces that the coal powder may be subjected to;

[0039] S126: Verification and optimization: Verify the accuracy of the model through actual transportation tests, and continuously optimize the model parameters based on the feedback results.

[0040] Preferably: in S3, the model evaluation includes:

[0041] S311: Performance indicator calculation: Use mean square error and mean absolute error statistical indicators to quantify the model prediction accuracy.

[0042] S312: Sensitivity analysis: Change the model input parameters (such as learning rate, number of iterations, etc.) and observe the changes in the prediction results to evaluate the model's sensitivity to parameter changes;

[0043] S313: Stability test: Repeatedly test the model performance on different time periods and different data sets to ensure the predictive ability of the model.

[0044] Preferably: in S3, the model modification includes:

[0045] S321: Accuracy analysis and preliminary corrections;

[0046] S322: In-depth error analysis and feature engineering;

[0047] S323: Correction of model overfitting;

[0048] S324: Parameter tuning and algorithm optimization.

[0049] Preferably, in S321, the accuracy analysis and preliminary correction are as follows:

[0050] Evaluate model performance;

[0051] Comparative analysis: Compare the current model with existing benchmark models or industry standards to assess their relative strengths and weaknesses;

[0052] Sensitivity analysis: Change the model input parameters (such as learning rate, number of iterations, etc.) and observe the changes in the prediction results to evaluate the model's sensitivity to parameter changes.

[0053] Preferably, in S322, the accuracy analysis and preliminary correction are as follows:

[0054] Error analysis: Analyze the source and nature of forecast errors to determine whether they are systematic deviations or random fluctuations, and whether they are related to specific input variables;

[0055] Feature engineering: Optimize the feature selection process, remove redundant or irrelevant features, and introduce new features to improve the generalization ability of the model.

[0056] Preferably, in S323, the accuracy analysis and preliminary correction are as follows:

[0057] Model adjustment: reduce the number of layers or nodes in the neural network, or use fewer split points in the decision tree;

[0058] Increase the regularization parameter: Increase the weight of the regularization term to prevent the model from overfitting the training data;

[0059] Cross-validation: The performance of the modified model was evaluated using the cross-validation method.

[0060] Preferably, in S324, the accuracy analysis and preliminary correction are as follows:

[0061] Parameter tuning: Choose one of grid search, random search, or Bayesian optimization as the tuning method; use techniques such as cross-validation to evaluate model performance under different parameter combinations; systematically adjust model parameters based on the selected method and evaluation criteria, record the results of each experiment, and select the best performing parameter combination; optimize hyperparameters such as learning rate, regularization term, and batch size; use visualization tools such as learning curves and feature importance graphs to assist the tuning process;

[0062] Based on the existing algorithms, introduce ensemble learning algorithms or transfer learning algorithms;

[0063] The details are as follows:

[0064] Algorithm optimization: Based on the existing algorithm, an integrated learning algorithm is introduced to select basic models and use the training data set to train each basic model separately; the test data is input into each basic model to generate their own prediction results; the prediction results of each basic model are combined through an integrated strategy to obtain the final prediction result.

[0065] Algorithm optimization: Based on the existing algorithm, a transfer learning algorithm is introduced. A deep learning model pre-trained on a large-scale dataset is selected as the starting point. The last fully connected layer of the pre-trained model is removed, and the outputs of the first few layers are used as feature extractors. A new fully connected layer is added on the top to classify new image categories. Fine-tune the training on the new dataset, adjusting only the parameters of the newly added layer and freezing the rest of the pre-trained model.

[0066] Preferably: in S4, the final prediction includes:

[0067] S41: Model input: input the preprocessed data into the optimized model for prediction operation;

[0068] S42: Result interpretation: Based on the model output results, determine whether the coal powder fineness will be significantly affected, and give the specific degree and scope of impact;

[0069] S43: Report generation: Organize the forecast results and related analysis processes to form a detailed report.

[0070] The beneficial effects of the present invention are:

[0071] 1. The present invention takes into account that during the transportation process, the fineness of the coal powder may change due to external force, resulting in a difference between the final fineness and the preset fineness, and uses this for analysis to obtain a final result that is more accurate and reliable.

[0072] 2. The present invention can monitor and predict the change trend of coal powder fineness in real time, providing a basis for timely adjustment of coal mill operating parameters; ensure the accuracy of measurement data through high-precision sensors and advanced data processing algorithms; and further improve the prediction accuracy by combining the PSO-SVM algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 The present invention provides a flow chart of a method for predicting coal powder fineness based on data drive. DETAILED DESCRIPTION

[0074] The technical solution of the present invention is further described in detail below in conjunction with specific implementation methods.

[0075] Embodiment 1:

[0076] A data-driven coal powder fineness prediction method includes:

[0077] S1: Coal pulverized gas sampling and analysis:

[0078] Take a sample of coal powder, use special equipment to test the hardness of the coal powder, and record the initial fineness and hardness values;

[0079] Stress analysis during transportation: testing the stress of coal powder in each link during transportation, including loading and unloading, vibration, impact, etc.

[0080] Data recording, recording the force data in chronological order to form a data set.

[0081] S2: Data processing and modeling:

[0082] Data preprocessing, data cleaning, extracting key features from the collected data sets, such as coal powder hardness, maximum force during transportation, cumulative force, etc.;

[0083] For model selection and training, support vector machine (SVM) is used as the basic algorithm, combined with particle swarm optimization (PSO) for parameter optimization; radial basis function (RBF) is selected as the kernel function, and the objective function is set to minimize the prediction error; the details are as follows:

[0084]

[0085] in, is the weight coefficient vector, is the mapping function, is the intercept term; introduce relaxation factor and , allowable fitting error , then the model minimizes the objective function through constraints, as follows:

[0086]

[0087]

[0088]

[0089]

[0090] in, is the penalty coefficient, Exceeding the allowable fitting error The number of

[0091] Particle swarm optimization;

[0092] S3: Model evaluation and revision;

[0093] Use indicators such as mean square error (MSE) and mean absolute error (MAE) to evaluate model performance; use an independent validation set to test the model to ensure the generalization ability of the model;

[0094] Analyze the source of model error based on the verification results; adjust the relevant parameters of SVM and PSO for the parts with large errors and retrain the model;

[0095] S4: Final prediction;

[0096] The latest data on coal powder hardness and force during transportation are input into the model; the model outputs the final predicted value of coal powder fineness.

[0097] Among them, in said S1, sampling is specifically as follows:

[0098] S111: Sampling preparation: Select appropriate sampling tools and methods based on the characteristics of the coal powder;

[0099] S112: Sampling: Randomly select coal powder samples from the coal mill outlet;

[0100] S113: Sample quantity control: The weight of each comprehensive sample should be controlled at about 100g to facilitate subsequent processing and testing;

[0101] S114: Record information: record sampling time, location, and environmental conditions;

[0102] S115: Preliminary inspection: Inspect the coal powder after sampling to eliminate obvious foreign matter or impurities.

[0103] Among them, in said S1, the force analysis during transportation includes:

[0104] S121: Data collection: Based on sensors installed at the detection position (such as conveyor belts, etc.), various force data of coal powder during transportation are recorded in real time, including impact force, pressure, vibration, etc.;

[0105] S122: Data classification: Classify and organize the collected data by type (such as impact force, pressure, etc.) and source (such as loading and unloading, vibration, etc.);

[0106] S123: Feature extraction: Use signal processing technology to extract key features of various data, such as peak value, mean value, standard deviation, etc., to provide a basis for subsequent analysis;

[0107] S124: Analysis of influencing factors: Study the influence of different factors (such as loading and unloading methods, transportation speed, road conditions, etc.) on the stress of coal powder and identify the main influencing factors;

[0108] S125: Modeling and prediction: Establish a force prediction model based on historical data, input new transportation conditions, and predict the forces that the coal powder may be subjected to;

[0109] S126: Verification and optimization: Verify the accuracy of the model through actual transportation tests, and continuously optimize the model parameters based on the feedback results.

[0110] Among them, in S3, the model evaluation includes:

[0111] S311: Performance indicator calculation: Use mean square error (MSE) and mean absolute error (MAE) statistical indicators to quantify the model prediction accuracy.

[0112] S312: Sensitivity analysis: Change the model input parameters (such as learning rate, number of iterations, etc.) and observe the changes in the prediction results to evaluate the model's sensitivity to parameter changes;

[0113] S313: Stability test: Repeatedly test the model performance on different time periods and different data sets to ensure that the model has stable and reliable predictive capabilities.

[0114] Wherein, in said S3, the model modification includes:

[0115] S321: Accuracy analysis and preliminary corrections;

[0116] S322: In-depth error analysis and feature engineering;

[0117] S323: Correction of model overfitting;

[0118] S324: Parameter tuning and algorithm optimization.

[0119] Among them, in S321, the accuracy analysis and preliminary correction are as follows:

[0120] Evaluate model performance;

[0121] Comparative analysis: Compare the current model with existing benchmark models or industry standards to assess their relative strengths and weaknesses;

[0122] Sensitivity analysis: Change the model input parameters (such as learning rate, number of iterations, etc.) and observe the changes in the prediction results to evaluate the model's sensitivity to parameter changes.

[0123] Among them, in S322, the accuracy analysis and preliminary correction are as follows:

[0124] Error analysis: Analyze the source and nature of forecast errors to determine whether they are systematic deviations or random fluctuations, and whether they are related to specific input variables;

[0125] Feature engineering: Optimize the feature selection process, remove redundant or irrelevant features, and introduce new features to improve the model generalization ability.

[0126] Among them, in S323, the accuracy analysis and preliminary correction are as follows:

[0127] Model tuning: reducing the number of layers or nodes in a neural network, or using fewer split points in a decision tree;

[0128] Increase the regularization parameter: Increase the weight of the regularization term to prevent the model from overfitting the training data;

[0129] Cross-validation: The performance of the modified model was evaluated using the cross-validation method.

[0130] Among them, in S324, the accuracy analysis and preliminary correction are as follows:

[0131] Parameter tuning: Choose one of Grid Search, Random Search, or Bayesian optimization as the tuning method; use cross-validation and other techniques to evaluate model performance under different parameter combinations; systematically adjust model parameters based on the selected method and evaluation criteria, record the results of each experiment, and select the best performing parameter combination; optimize hyperparameters such as learning rate, regularization term, and batch size; use visualization tools such as learning curves and feature importance graphs to assist the tuning process;

[0132] Based on the existing algorithms, introduce ensemble learning algorithms or transfer learning algorithms;

[0133] The details are as follows:

[0134] Algorithm optimization: Based on the existing algorithm, an integrated learning algorithm is introduced to select basic models and use the training data set to train each basic model separately; the test data is input into each basic model to generate their own prediction results; the prediction results of each basic model are combined through an integration strategy (such as weighted average) to obtain the final prediction result.

[0135] Algorithm optimization: Based on the existing algorithm, a transfer learning algorithm is introduced. A deep learning model pre-trained on a large-scale dataset is selected as the starting point. The last fully connected layer of the pre-trained model is removed, and the outputs of the first few layers are used as feature extractors. A new fully connected layer is added on the top to classify new image categories. Fine-tune the training on the new dataset, adjusting only the parameters of the newly added layer and freezing the rest of the pre-trained model.

[0136] Among them, in S4, the final prediction includes:

[0137] S41: Model input: input the preprocessed data into the optimized model for prediction operation;

[0138] S42: Result interpretation: Based on the model output results, determine whether the coal powder fineness will be significantly affected, and give the specific degree and scope of impact;

[0139] S43: Report generation: Organize the forecast results and related analysis processes to form a detailed report.

[0140] 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. A data-driven method for predicting coal powder fineness, characterized in that: include: S1: Coal pulverized gas sampling and analysis: Take a sample of coal powder, perform a hardness test on the coal powder, and record the initial fineness and hardness values; Stress analysis during transportation: testing the stress of coal powder in each link during transportation, including loading and unloading, vibration, and impact; Data recording, recording the force data in chronological order to form a data set; S2: Data processing and modeling: Data preprocessing, data cleaning, and extracting key features from the collected data sets; For model selection and training, support vector machine is used as the basic algorithm, combined with particle swarm algorithm for parameter optimization; radial basis function is selected as kernel function, and the objective function is set to minimize the prediction error; the details are as follows: in, is the weight coefficient vector, is the mapping function, is the intercept term; introduce relaxation factor and , allowable fitting error , then the model minimizes the objective function through constraints, as follows: in, is the penalty coefficient, Exceeding the allowable fitting error The number of Particle swarm algorithm optimization; S3: Model evaluation and revision; The mean square error and mean absolute error (MBE) were used to evaluate the model performance; the model was tested using an independent validation set to ensure the generalization ability of the model; Analyze the source of model error based on the verification results; adjust the relevant parameters of SVM and PSO for the parts with large errors and retrain the model; S4: Final prediction; The latest data on coal powder hardness and force during transportation are input into the model; the model outputs the final predicted value of coal powder fineness.

2. A data-driven coal powder fineness prediction method according to claim 1, characterized in that: In S1, sampling is specifically as follows: S111: Sampling preparation: Select appropriate sampling tools and methods based on the characteristics of the coal powder; S112: Sampling: randomly select coal powder samples from the coal mill outlet; S113: Sample quantity control: The weight of each comprehensive sample should be controlled within 100g to facilitate subsequent processing and testing; S114: Record information: record sampling time, location, and environmental conditions; S115: Preliminary inspection: Inspect the coal powder after sampling to eliminate obvious foreign matter or impurities.

3. The method for predicting coal powder fineness based on data drive according to claim 1, characterized in that: In S1, the force analysis during transportation includes: S121: Data collection: Based on sensors installed at the detection position, various force data of coal powder during transportation are recorded in real time, including impact force, pressure, and vibration; S122: Data classification: Classify and organize the collected data by type and source; S123: Feature extraction: Use signal processing techniques to extract key features of various types of data; S124: Analysis of influencing factors: Study the influence of different factors on the stress of coal powder and identify the main influencing factors; S125: Modeling and prediction: Establish a force prediction model based on historical data, input new transportation conditions, and predict the forces that the coal powder may be subjected to; S126: Verification and optimization: Verify the accuracy of the model through actual transportation tests, and continuously optimize the model parameters based on the feedback results.

4. The method for predicting coal powder fineness based on data drive according to claim 1, characterized in that: In S3, the model evaluation includes: S311: Performance indicator calculation: Use mean square error and mean absolute error statistical indicators to quantify the model prediction accuracy. S312: Sensitivity analysis: Change the model input parameters and observe the changes in the prediction results to evaluate the model's sensitivity to parameter changes; S313: Stability test: Repeatedly test the model performance on different time periods and different data sets to ensure the predictive ability of the model.

5. A data-driven coal powder fineness prediction method according to claim 4, characterized in that: In S3, the model modification includes: S321: Accuracy analysis and preliminary corrections; S322: In-depth error analysis and feature engineering; S323: Correction of model overfitting; S324: Parameter tuning and algorithm optimization.

6. A data-driven coal powder fineness prediction method according to claim 5, characterized in that: In S321, the accuracy analysis and preliminary correction are as follows: Evaluate model performance; Comparative analysis: Compare the current model with existing benchmark models or industry standards to assess their relative strengths and weaknesses; Sensitivity analysis: Change the model input parameters and observe the changes in the prediction results to evaluate the model's sensitivity to parameter changes.

7. The method for predicting coal powder fineness based on data drive according to claim 5, characterized in that: In S322, the accuracy analysis and preliminary correction are as follows: Error analysis: Analyze the source and nature of forecast errors to determine whether they are systematic deviations or random fluctuations, and whether they are related to specific input variables; Feature engineering: Optimize the feature selection process, remove redundant or irrelevant features, and introduce new features to improve the generalization ability of the model.

8. The method for predicting coal powder fineness based on data drive according to claim 5, characterized in that: In S323, the accuracy analysis and preliminary correction are as follows: Model adjustment: reduce the number of layers or nodes in the neural network, or use fewer split points in the decision tree; Increase the regularization parameter: Increase the weight of the regularization term to prevent the model from overfitting the training data; Cross-validation: The performance of the modified model was evaluated using the cross-validation method.

9. The method for predicting coal powder fineness based on data drive according to claim 5, characterized in that: In S324, the accuracy analysis and preliminary correction are as follows: Parameter tuning: Choose one of grid search, random search, or Bayesian optimization as the tuning method; use cross-validation techniques to evaluate model performance under different parameter combinations; systematically adjust model parameters based on the selected method and evaluation criteria, record the results of each experiment, and select the best performing parameter combination; optimize parameters such as learning rate, regularization term, and batch size; use learning curves and feature importance graph visualization tools to assist the tuning process; Based on the existing algorithms, introduce ensemble learning algorithms or transfer learning algorithms; The details are as follows: Algorithm optimization: Based on the existing algorithm, an integrated learning algorithm is introduced to select basic models and use the training data set to train each basic model separately; the test data is input into each basic model to generate their own prediction results; the prediction results of each basic model are combined through the integration strategy to obtain the final prediction result; Algorithm optimization: Based on the existing algorithm, a transfer learning algorithm is introduced, and a deep learning model pre-trained on a large-scale data set is selected as the starting point; Remove the last fully connected layer of the pre-trained model and use the outputs of the first few layers as feature extractors; Add a new fully connected layer on top to classify new image categories; perform fine-tuning training on the new dataset, adjusting only the parameters of the newly added layer and freezing the rest of the pre-trained model.

10. The method for predicting coal powder fineness based on data drive according to claim 5, characterized in that: In S4, the final prediction includes: S41: Model input: input the preprocessed data into the optimized model for prediction operation; S42: Result interpretation: Based on the model output results, determine whether the coal powder fineness will be significantly affected, and give the specific degree and scope of impact; S43: Report generation: Organize the forecast results and related analysis processes to form a detailed report.

Images