Clean coal yield prediction method based on support vector machine

Abstract

The invention relates to the technical field of coal processing and utilization, and discloses a clean coal yield prediction method based on a support vector machine, which comprises a data acquisition module used for analyzing factors influencing the clean coal yield, acquiring related data and integrating the data into a data set, and a data preprocessing module connected with the data acquisition module and used for preprocessing the data. The data preprocessing module is used for randomly dividing a data set according to a 70% training set and a 30% test set and carrying out standardization and normalization preprocessing, the parameter optimization module is connected with the data preprocessing module and optimizes hyper-parameters of a support vector machine through an improved grey wolf algorithm, and the model building module is connected with the parameter optimization module and is used for building a model. The model building module is used for building and training a support vector regression model based on the optimized hyper-parameters, and the model verification module is connected with the model building module and uses a test set to verify the performance of the model. According to the method, the hyper-parameters of the support vector machine are optimized through the improved grey wolf algorithm, the problem that a traditional optimization method is prone to falling into local optimum is effectively avoided, and the precision of clean coal yield prediction is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of coal processing and utilization technology, specifically to a method for predicting clean coal yield based on support vector machines. Background Technology

[0002] With rapid economic development, coal, as one of the main energy sources, remains an important part of my country's energy consumption, despite a decline in its share of total consumption. To meet coal demand, the country promotes the clean and efficient use of coal. Currently, common algorithms in the prediction field include multiple linear regression (MLR), backpropagation neural networks, and support vector regression (SVR), each with its own advantages and disadvantages. In order to improve the utilization rate of coal resources and further promote the green and low-carbon transformation of the country's energy sector, it is necessary to seek better technologies to improve coal combustion efficiency.

[0003] The shortcomings of existing common algorithms in the prediction field are:

[0004] 1. In patent document CN116415741B, a dual-database prediction method combining CNN and Transformer is adopted. The reliability is improved by cross-checking the two types of data, and the prediction accuracy is improved by capturing sequence dependencies using Transformer. Although the reliability is improved, the impact of feature engineering on model performance is not fully considered, and the problem of hyperparameter optimization is not involved.

[0005] 2. In patent document CN116307075B, PSO-SVM is used to predict coal consumption for power supply and optimize coal blending scheme. Although SVR is used, it relies on traditional parameter optimization methods (such as PSO and grid search), resulting in low optimization efficiency.

[0006] 3. In patent document CN112362842B, an SVR was used to establish a predictive model for the release characteristics of harmful elements. However, no special feature extraction strategy was designed for the multi-source heterogeneous data (such as coal source ratio and production process parameters) unique to the coal industry. As a result, there is still room for improvement in the prediction accuracy and generalization ability of the model in complex industrial scenarios.

[0007] 4. In patent document CN115577881A, the application scenarios of SVR were further expanded, and it was used for deep coal gas content prediction and coal blending coking hot state prediction. The generalization ability of the model was improved through data screening and normalization. However, it lacked targeted processing of data noise and the application of dynamic weight strategy in the optimization process, making it difficult to fully adapt to the data volatility and nonlinear characteristics in the coal production process. Summary of the Invention

[0008] This invention provides a method for predicting clean coal yield based on support vector machines, in order to address the problems mentioned in the background that existing technologies lack targeted processing of data noise and that the application of dynamic weighting strategies in the optimization process is difficult to fully adapt to the data volatility and nonlinear characteristics in the coal production process.

[0009] This invention provides a method for predicting clean coal yield based on support vector machines, comprising:

[0010] The system includes a data acquisition module for analyzing factors affecting clean coal yield and acquiring relevant data, which is then integrated into a dataset. A data preprocessing module, connected to the data acquisition module, randomly divides the dataset into a 70% training set and a 30% test set, and performs standardization and normalization preprocessing. A parameter optimization module, connected to the data preprocessing module, optimizes the hyperparameters of the support vector machine using an improved gray wolf algorithm. A model building module, connected to the parameter optimization module, builds and trains a support vector regression model based on the optimized hyperparameters. A model validation module, connected to the model building module, validates the model performance using a test set and evaluates the prediction effect using assessment metrics.

[0011] The steps for predicting clean coal yield are as follows:

[0012] S1. Analyze the factors affecting clean coal yield and obtain relevant information about these factors;

[0013] S2. The acquired data information is divided and preprocessed into a dataset, and the resulting dataset is randomly divided into a training set and a test set.

[0014] S3. Obtain the optimal parameter combination of support vector machine hyperparameters through the improved Grey Wolf algorithm;

[0015] S4. Establish a support vector regression model with the optimal hyperparameter combination, import the test set from S2 into the model for training, and obtain a prediction model for clean coal yield.

[0016] S5. The test set is predicted using a prediction model to obtain the clean coal yield under different input feature vector conditions, and the prediction results are compared and analyzed with the actual values.

[0017] Preferably, the data preprocessing module implements a support vector regression model using Python software and the scikit-learn library.

[0018] Preferably, the data acquisition module is also used to collect coal source ratio information and production process parameters. The coal source ratio information comes from the company's daily production technology inspection and testing report. The production process parameters include total moisture, ash, volatile matter, total sulfur, caking index, and plastic layer index. The data comes from the feedback of raw coal test data in the south area of ​​the company's quality inspection workshop.

[0019] Preferably, in S2, the training set also needs to be standardized, and the standardization function is: ,in and Let be the sample mean and variance, respectively. After standardization, the training set is normalized using the following normalization function: This maps the data to the [0,1] interval.

[0020] Preferably, the improved gray wolf algorithm in S3 specifically includes setting the number of individuals in the population, the maximum number of iterations, the range of initialization penalty parameters, and the range of kernel function parameters, and using a Bernoulli chaotic mapping sequence to initialize the wolf pack positions.

[0021] Preferably, a dynamic weighting strategy is introduced in S3, the mathematical expression of which is:

[0022]

[0023]

[0024] D1, D2, and D3 represent the Euclidean distances between individual wolves and the top three winners.

[0025] Preferably, in S5, the prediction effect is evaluated using evaluation indicators, including mean deviation, mean absolute error, coefficient of determination, and root mean square error, the formulas of which are as follows:

[0026]

[0027]

[0028]

[0029] Where n represents the sample size. Let be the predicted value of the i-th sample. Represented as the actual value of the i-th sample. This represents the average of the actual values.

[0030] Preferably, the model building module specifies the kernel function type, SVM type, and insensitive loss coefficient in the parameter string of the svm_train function, and passes the optimized hyperparameter combination to the function.

[0031] Preferably, the data preprocessing module organizes the raw data into CSV format, where the input feature vectors are coal source ratio information, total moisture, ash content, volatile matter, total sulfur content, caking index, and plastic layer index, and the output space vector is the clean coal yield.

[0032] Preferably, slack variables should also be introduced into the vector regression model of S4. and This is used to define the width of the isolation band and set constraints based on the sample point locations:

[0033] If the sample point is located within or at the edge of the isolation zone, then ,and ;

[0034] If the sample point is located above the isolation zone, then ,and ,at this time This indicates the distance the sample point extends beyond the isolation zone.

[0035] If the sample point is located below the isolation zone, then ,and ,at this time This indicates the distance the sample point extends beyond the isolation zone.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] Improving prediction accuracy: By improving the Grey Wolf algorithm to optimize the hyperparameters of the support vector machine, the problem of traditional optimization methods easily getting trapped in local optima is effectively avoided, and the accuracy of clean coal yield prediction is significantly improved.

[0038] Optimize data processing workflow: Ensure reasonable data distribution through standardization and normalization preprocessing, accelerate model convergence, and improve training stability;

[0039] Enhanced model generalization ability: The introduction of dynamic weight strategy and slack variable mechanism enables the support vector regression model to better adapt to complex nonlinear relationships, thereby improving the model's generalization ability and robustness.

[0040] Promoting the clean and efficient use of coal: By accurately predicting the yield of clean coal, we can provide a scientific basis for optimizing the coal processing process, which will help improve the utilization rate of coal resources, reduce energy waste, and meet the needs of green and low-carbon development. Attached Figure Description

[0041] Figure 1 This is an overall flowchart of the present invention;

[0042] Figure 2 This is a schematic diagram illustrating the process of optimizing hyperparameters using the improved Grey Wolf algorithm of this invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] This invention discloses a method for predicting clean coal yield based on support vector machines. Example 1: The overall process is as follows... Figure 1 As shown, it includes the following modules:

[0045] The data acquisition module is used to analyze the key factors affecting the yield of clean coal and collect relevant data. Specifically, the factors affecting the yield of clean coal include coal source ratio information (such as the proportion of different coal sources) and production process parameters (such as total moisture, ash, volatile matter, total sulfur, caking index, and plastic layer index).

[0046] Furthermore, the data sources are the company's daily production technology inspection and testing reports and the raw coal test data feedback from the south area of ​​the quality inspection workshop. For example, in a certain batch of data, coal source A accounts for 30%, coal source B accounts for 70%, total moisture is 8.5%, ash content is 22.3%, volatile matter is 28.1%, total sulfur content is 1.2%, caking index is 75, and plastic layer index is 12mm, corresponding to a clean coal yield of 65.2%.

[0047] The data preprocessing module, which is connected to the data acquisition module, divides and preprocesses the raw data;

[0048] First, the dataset is randomly divided into a 70% training set and a 30% test set. For example, out of 1000 data points, 700 are used for training and 300 for testing. Then, standardization and normalization are performed using Python software and the scikit-learn library.

[0049] Standardization: Using formulas ,in The mean of the training set. The standard deviation of the training set;

[0050] Normalization: using the formula The data is mapped to the [0,1] interval. For example, the minimum and maximum values ​​of the gray feature are 18.0 and 25.0 respectively, and the normalized value is... .

[0051] The parameter optimization module, connected to the data preprocessing module, optimizes the hyperparameters (penalty parameter C and kernel function parameter g) of the support vector machine by improving the Grey Wolf algorithm. The specific steps are as follows:

[0052] Population initialization: Set the population size N=30, the maximum number of iterations T_max=100, the penalty parameter range C_range=[0.1,100], the kernel function parameter range g_range=[0.001,10], and use the Bernoulli chaotic mapping sequence to initialize the wolf pack position to avoid getting trapped in local optima;

[0053] Dynamic weighting strategy: Calculate the Euclidean distances D1, D2, and D3 between each wolf in the pack and the top three winners using the formula... , and Dynamically adjust the wolf pack position update weights to enhance algorithm convergence;

[0054] Introducing slack variables: Defining slack variables in a support vector regression model and This is used for controlling the width of the isolation zone. For example, if the deviation between the predicted and actual values ​​of a sample is within the isolation zone, then... ,and If the deviation exceeds the area above the isolation zone, then ,and .

[0055] The model building module, connected to the parameter optimization module, establishes a support vector regression model based on the optimized hyperparameters. Specifically, the kernel function type is specified as radial basis function (RBF), the SVM type is ε-SVR, and the insensitive loss coefficient is specified in the svm_train function. The optimized (C) and (g) are passed to this function. For example, the optimized hyperparameters are (C=50) and (g=0.5). The model learns the nonlinear mapping relationship between input features and clean coal yield through the training set.

[0056] The model validation module, connected to the model building module, uses a test set to validate model performance. Predictive performance is evaluated using the following metrics:

[0057] Mean deviation: This reflects the systematic bias in the predicted values;

[0058] Mean absolute error: It measures the absolute magnitude of the prediction error;

[0059] Coefficient of determination: This indicates the degree to which the model explains the variation in the data;

[0060] Root mean square error: Highlighting the impact of larger errors;

[0061] Where n represents the sample size. Let be the predicted value of the i-th sample. Represented as the actual value of the i-th sample. The average of actual values;

[0062] For example, the evaluation results for a certain test set are MBE=0.2, MAE=1.5, and R... 2 =0.92 and RMSE=2.0, indicating that the model has high prediction accuracy.

[0063] By improving the Grey Wolf algorithm to optimize the hyperparameters of the support vector machine, the problem of traditional optimization methods easily getting trapped in local optima is effectively avoided, significantly improving the accuracy of clean coal yield prediction. Through standardization and normalization preprocessing, reasonable data distribution is ensured, accelerating model convergence and improving training stability. The introduction of dynamic weight strategy and slack variable mechanism enables the support vector regression model to better adapt to complex nonlinear relationships, improving the model's generalization ability and robustness. By accurately predicting clean coal yield, a scientific basis is provided for the optimization of coal processing, which helps to improve coal resource utilization, reduce energy waste, and meet the needs of green and low-carbon development.

[0064] Example 2

[0065] Building upon Example 1, the data acquisition module further refines the data sources and types. Coal source ratio information comes from the company's daily production technology inspection and testing reports; for example, a daily report might show coal source A accounting for 40% and coal source B accounting for 60%. Production process parameters are derived from raw coal test data feedback from the south area of ​​the quality inspection workshop, specifically including:

[0066] Total moisture content: determined according to GB / T212-2008 standard, for example, 8.0%;

[0067] Ash content: determined according to GB / T212-2008 standard, for example, 20.5%;

[0068] Volatile matter: determined according to GB / T212-2008 standard, for example, 30.0%;

[0069] Total sulfur content: determined according to GB / T214-2007 standard, for example, 1.0%;

[0070] Adhesion index: determined according to GB / T5447-1997 standard, for example 80;

[0071] Gelatinous layer index: determined according to GB / T479-2000 standard, for example, 15mm;

[0072] The above data, together with the clean coal yield (e.g., 68.5%), constitute the original dataset.

[0073] Example 3

[0074] Based on Example 1, the specific implementation of the data preprocessing module is as follows:

[0075] Data partitioning and preprocessing were implemented using Python 3.8 and the scikit-learn 0.24.2 library.

[0076] The dataset is randomly divided into training and test sets in a 7:3 ratio using the train_test_split function. The random seed random_state=42 is set to ensure that the results are reproducible.

[0077] Standardization: Standardize the training set using the StandardScaler class and apply the same transformation to the test set;

[0078] Normalization: The MinMaxScaler class is used to map the data to the [0,1] interval. For example, if the maximum and minimum values ​​of the gray feature are 25.0 and 18.0 respectively, the normalized value is... .

[0079] Example 4

[0080] Based on Example 1, the specific parameters for the standardization and normalization processing of the data preprocessing module are as follows:

[0081] Standardization processing: Taking gray features as an example, the mean of the training set... =21.2, standard deviation =2.1, the ash content of a certain sample is 22.5, and the standardized value is ;

[0082] Normalization processing: Taking volatile matter characteristics as an example, the minimum value of the training set maximum value A sample has a volatile content of 30.0, and its normalized value is... .

[0083] Example 5

[0084] Based on Example 1, the improved Grey Wolf algorithm initialization parameters of the parameter optimization module are as follows:

[0085] Population size N=50, maximum number of iterations T max =200, penalty parameter range [C] min C max ]=[1, 1000], kernel function parameter range [g min g max ] = [0.01, 100],

[0086] The initial parameter r=0.4 of the Bernoulli chaotic mapping sequence generates the initial wolf pack positions, for example, the position of the i-th wolf is [C i ,g i = [50, 2.5].

[0087] Example 6

[0088] Based on Example 1, here is an example of calculating the dynamic weight strategy for the parameter optimization module:

[0089] A wolf pack has an individual position [C,g]=[30,1.0]. The top three winners, D1, D2, and D3, have positions of [50,2.0], [40,1.5], and [60,3.0], respectively. The calculation process is as follows:

[0090]

[0091]

[0092]

[0093] The dynamic weight calculation process is as follows:

[0094]

[0095]

[0096] .

[0097] Example 7

[0098] Based on Example 1, here is an example of calculating the evaluation metrics for the model validation module:

[0099] The test set contains 10 samples, and the actual values ​​are... The predicted values ​​are [65,67,66,68,70,69,67,66,68,70]. The values ​​are [64, 68, 65, 67, 71, 68, 66, 65, 69, 69]. The performance indicators are calculated as follows:

[0100] Average deviation: ;

[0101] Mean absolute error: ;

[0102] Coefficient of determination: Mean of actual values Calculated R 2 ;

[0103] Root mean square error: .

[0104] Example 8

[0105] Based on Example 1, here is an example of CSV format data from the data preprocessing module:

[0106] Input feature vector: Coal source A accounts for 30%, coal source B accounts for 70%, total moisture 8.5%, ash 22.3%, volatile matter 28.1%, total sulfur 1.2%, caking index 75, and plastic layer index 12mm;

[0107] Output space vector: Clean coal yield 65.2%;

[0108] The CSV file format is as follows: coal source A percentage, coal source B percentage, total moisture, ash, volatile matter, total sulfur, caking index, plastic layer index, and clean coal yield, for example: 30, 70, 8.5, 22.3, 28.1, 1.2, 75, 12, and 65.2.

[0109] Example 9

[0110] Based on Example 1, here is an example of slack variable constraints for the parameter optimization module:

[0111] Actual value of a certain sample , width of the isolation zone ;

[0112] because The sample was located below the isolation zone, therefore, , ;

[0113] This slack variable is used to calculate the penalty term in the objective function, ensuring that the model imposes appropriate constraints on samples that deviate from the isolation zone.

[0114] The contents not described in detail in this description are existing technologies known to those skilled in the art. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for clean coal yield prediction based on support vector machines, characterized in that, The application relates to a coal concentrate yield prediction method and device. The application relates to a coal concentrate yield prediction method and device. S1, factors influencing the coal concentrate yield are analyzed, and related information of the factors is obtained; S2, the obtained data information is divided into data sets and preprocessed, and the obtained data sets are randomly divided into a training set and a test set; S3, an optimal parameter combination of support vector machine superparameters is obtained through an improved grey wolf algorithm; S4, a support vector regression model is established based on the optimal superparameter combination, the test set in S2 is introduced into the model for training, and a prediction model about the coal concentrate yield is obtained; S5, the test set is predicted through the prediction model, the coal concentrate yield under different input feature vector conditions is obtained, and the prediction result is compared with the true value. The data preprocessing module realizes the support vector regression model through Python software and a scikit-learn library.

2. The method of predicting clean coal yield based on support vector machine according to claim 1, characterized in that: The data acquisition module is also used for collecting coal source matching information and production process parameters, wherein the coal source matching information is obtained from enterprise production technology inspection and daily report forms, the production process parameters include total moisture, ash content, volatile matter, total sulfur content, caking index and gel layer index, and the data is obtained from enterprise quality inspection workshop south raw coal test data feedback.

3. The method of predicting clean coal yield based on support vector machine according to claim 1, characterized in that: In S3, the improved grey wolf algorithm specifically comprises the following steps: setting the number of population individuals, the maximum number of iterations, the initialization penalty parameter range and the kernel function parameter range, and initializing the wolf swarm position by using a Bernoulli chaotic mapping sequence.

4. The method of predicting clean coal yield based on support vector machine according to claim 1, characterized in that: In S2, the training set also needs to be standardized, and the standardization function is: where and are the sample mean and variance, respectively. After standardization, the training set is normalized, and the normalization function is: which maps the data to the interval [0, 1].

5. The method of predicting clean coal yield based on support vector machine according to claim 1, characterized in that: In S3, a dynamic weight strategy is introduced, and the mathematical expression is as follows:

6. The method of predicting clean coal yield based on support vector machine according to claim 1, characterized in that: In S5, the prediction effect is evaluated through evaluation indexes, the evaluation indexes include average deviation, average absolute error, determination coefficient and root mean square error, and the formulas are as follows: where D1, D2, D3 are the Euclidean distances of the wolf individual to the first three winners.

7. The method of predicting clean coal yield based on support vector machine according to claim 1, characterized in that: The model construction module specifies the kernel function type, the SVM type and the insensitive loss coefficient in the svm_train function parameter string, and transmits the optimized superparameter combination to the function. where n denotes the number of samples, denotes the predicted value for the i-th sample, denotes the actual value for the i-th sample, denotes the mean of the actual values.

8. The support vector machine based clean coal yield prediction method of claim 1, wherein: The data preprocessing module arranges the original data into a CSV format, wherein the input feature vector is the coal source matching information, total moisture, ash content, volatile matter, total sulfur content, caking index and gel layer index, and the output space vector is the coal concentrate yield.

9. The support vector machine based clean coal yield prediction method according to claim 1, characterized in that: ​ 10. The support vector machine based clean coal yield prediction method according to claim 1, characterized in that: Relaxation variables are introduced in the vector regression model of S4 and to define the isolation bandwidth and set constraints according to sample point locations: If the sample point is located in the isolation band or the edge, then , and ; If the sample point is above the isolation band, then , and , in which case represents the distance by which the sample point exceeds the isolation band. If the sample point is below the isolation band, then , and , in which case represents the distance by which the sample point exceeds the isolation band below.

Images