DE-MK-MSVR-based tin smelting process multi-class energy consumption prediction method

Through the DE-MK-MSVR multi-category energy consumption prediction method, combined with multi-output support vector regression and differential evolution algorithm, the problem of low prediction accuracy of multiple types of energy consumption during tin smelting is solved, and the correlation consideration of energy consumption is achieved and improve prediction accuracy.

CN118917463APending Publication Date: 2024-11-08KUNMING UNIV OF SCI & TECH
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410950723.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-16
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The prior art is difficult to accurately predict multi-type energy consumption in multiple processes during tin smelting, and the traditional single model ignores the cross-correlation between different energy types, resulting in low prediction accuracy.

Method used

Using a multi-category energy consumption prediction method based on DE-MK-MSVR, through data acquisition, preprocessing, multi-output support vector regression model and differential evolution algorithm, combined with multi-core learning and distance correlation coefficient algorithm, we can build a multi-type energy prediction at the same time. consumed model.

Benefits of technology

The accuracy of energy consumption prediction in the tin smelting process is improved, the correlation between different energy consumption is taken into account, and the nonlinear fitting ability and prediction accuracy of the model are enhanced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917463A_ABST
    Figure CN118917463A_ABST
Patent Text Reader

Abstract

The invention relates to a DE-MK-MSVR-based tin smelting process multi-class energy consumption prediction method, and belongs to the technical field of energy consumption prediction. The method specifically comprises the steps that missing value, abnormal value and data dimension processing is conducted on data collected in the tin smelting process, a distance correlation coefficient matrix is adopted to analyze energy consumption influence factors, a multi-output energy consumption prediction model is built, multi-kernel learning is introduced to improve the model, and a differential evolution algorithm is adopted to optimize model parameters. And finally, a classic multi-output prediction model is adopted for comparison. According to the method, the problem of comprehensive energy consumption prediction caused by multi-type energy consumption and strong correlation in the tin smelting process can be solved; the problems that the overall prediction effect is poor and the calculation cost is too high due to the fact that a plurality of single energy consumption prediction models are constructed and correlation between energy sources is ignored are solved. The defect that a traditional model can only predict single energy consumption is overcome, the consumption conditions of different types of energy under the same energy consumption influence factor can be predicted at the same time, and the model prediction precision is high.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of energy consumption prediction, and in particular to a multi-category energy consumption prediction method for a tin smelting process based on DE-MK-MSVR. Background Art

[0002] As a high-emission and high-energy-consuming industry, non-ferrous metal smelting is the key to its sustainable development. Non-ferrous metals consume a large amount of natural mineral resources such as coal and ore during the smelting process. Energy consumption is one of the important costs of enterprises. Energy consumption prediction plays a very important role in tapping energy-saving potential. The current energy consumption prediction methods can be roughly divided into three categories: mechanism modeling, data-driven modeling and hybrid modeling. Mechanism-based modeling methods are mostly methods in which domain experts, with rich domain knowledge, establish process models based on the reaction operation mechanism within the object process and the laws involved in the process, such as chemical reaction laws, thermodynamics, and the law of conservation of energy and mass. This method is generally more complex and costly; data-driven modeling methods do not rely too much on the mechanism knowledge of the reaction process, and establish energy consumption prediction models by analyzing process characteristics and data features. They have high prediction accuracy and fast response. Representative methods include BP neural network prediction, long short-term memory neural network (LSTM), support vector machine (SVM), random forest (RF) and other models; hybrid modeling is a method that combines mechanism analysis with data-driven. By introducing prior knowledge into the modeling and analysis process, it can not only greatly improve the efficiency of modeling optimization, but also solve the problem of poor model generalization ability.

[0003] Faced with the high temperature and strong dust in the tin smelting process, which involves complex physical and chemical reactions and energy-to-mass conversion of multiphase flow, it is difficult to use mechanism analysis modeling when the key state parameters of the smelting process cannot be accurately perceived. The data-driven modeling method is more suitable for energy consumption analysis of the entire tin smelting process. Traditional energy consumption prediction models only perform prediction and analysis on a certain production process or a single energy source. These single prediction models cannot meet the needs of multi-category output prediction for complex production processes with multiple processes. The tin smelting process is accompanied by cross-consumption of multiple types of energy and there is a strong correlation between energy consumption; if a single model is used for prediction, the potential cross-correlation between different energy types will be ignored. Summary of the invention

[0004] In order to solve the problem of multi-type energy consumption prediction in a production process involving multiple processes under the condition of a small sample data set in the tin smelting process in the prior art, the present invention proposes a multi-category comprehensive energy consumption prediction method for the tin smelting process based on DE-MK-MSVR. In order to fully consider the correlation between different energy consumptions, the problem of low prediction accuracy of multi-category energy consumption by a single model is solved by adopting the multi-output support vector regression (MSVR) method, thereby improving the prediction effect of energy consumption data of tin smelting enterprises.

[0005] To achieve the above object, the present invention provides a method for predicting multi-energy consumption in tin smelting process based on DE-MK-MSVR, which specifically comprises the following steps:

[0006] S1. Collecting energy consumption data, production process data and real-time status data of the tin smelting process through a data acquisition device;

[0007] The data acquisition device includes: an electronic multifunctional smart meter, a smart thermal meter, a thermal integrator, a smart gas flow meter, a smart liquid flow meter, a weighing meter, a pressure, temperature, and speed sensor, and uses the Internet and 5G network communication methods for data transmission;

[0008] S2, taking the energy consumption data obtained in S1 as output variables, and the process data and real-time status data as input variables and preprocessing them;

[0009] Since the collected data have missing values, outliers, and differences in collection period and dimension, the data collected on site cannot be used directly for modeling, so the data needs to be preprocessed. The preprocessing includes: data format conversion, missing value filling, outlier detection, and data dimension unification. The specific steps are as follows:

[0010] S2.1. Convert the categorical data in the original data into numerical data using one-hot encoding;

[0011] S2.2. Lagrange interpolation is used to make up for missing data. The specific expression is as follows:

[0012]

[0013]

[0014] Where: y j represents the position of the jth output variable, l j (x) represents the interpolation function, x represents the input variable, and x i represents the position of the i-th input variable, x j represents the position of the jth input variable, and k represents the number of independent variables;

[0015] S2.3. Use box plots to detect and remove outliers;

[0016] The box plot consists of five values: minimum value (min), lower quartile (Q1), median (median), upper quartile (Q3) and maximum value (max); in the box plot, IQR represents the interquartile range, IQR = Q3-Q1, the minimum observation value is set to Q1-1.5*IQR, the maximum observation value is Q3+1.5*IQR, and the value beyond the maximum and minimum range is considered an "outlier";

[0017] The box plot does not impose any restrictive requirements on the data. It only truly and intuitively shows the true appearance of the data distribution. The standard for judging outliers in the box plot is based on quartiles and interquartile ranges. Quartiles have a certain degree of robustness. The results of the box plot in identifying outliers are relatively objective, and it has certain advantages in identifying outliers.

[0018] S2.4. Use a normalization algorithm to eliminate the dimensional differences between variables. The specific expression is as follows:

[0019]

[0020] In the formula: μ represents the mean, σ represents the standard deviation, x represents the input variable, and x * represents the normalized data;

[0021] S3, using the distance correlation coefficient algorithm to analyze the input and output variables preprocessed by S2, and removing redundant input variables according to the correlation coefficient ratio, and screening out input variables with high correlation with the output variables;

[0022] The correlation coefficient ratio between the input and output variables is calculated by the distance correlation coefficient algorithm. The coefficient ratio range is [0,1]. If the ratio is 0, it means that the variables are independent of each other. The closer the ratio is to 1, the stronger the correlation between the variables. The correlation coefficient ratio of 0.9 is used as the threshold. If the correlation coefficient is higher than 0.9, it indicates that the variables are strongly correlated. The correlation between the two strongly correlated redundant variables and the output variables is compared respectively, and a group of variables with a higher correlation with the output variable is retained.

[0023] The specific expression of the distance correlation coefficient algorithm is as follows:

[0024]

[0025] Where: x represents the input variable, y represents the output variable, dcor(x,y) represents the distance correlation coefficient value, dcov(x,y) represents the distance covariance, dvar(x) and dvar(y) represent the distance covariance of x and y respectively;

[0026] S4, setting the input variables and output variables as training samples and dividing them into training set and test set, building a DE-MK-MSVR energy consumption prediction model, using the redundant input variables removed in S3 to train the model, and using the test set to verify the prediction effect of the model;

[0027] The training set and the test set are divided into 8:2. The specific steps of building the DE-MK-MSVR energy consumption prediction model are as follows:

[0028] S4.1. Establish a multi-output support vector regression model;

[0029] For the function fitting problem, let the training sample be S = {(x i ,y i )}, where x i Indicates the position of the i-th input variable, y i Indicates the position of the i-th output variable, i=1,2,...,N p , N p Represents the population size, and the specific expression of constructing the regression function model is as follows:

[0030]

[0031] In the formula, Φ(x) represents the nonlinear mapping function of the high-dimensional space, W is the weight coefficient of the mapping function Φ(x), B represents the bias vector of the multi-output function F(x), W = [w 1 ,w 2 ,…,w N ],B=[b 1 ,b 2 ,…,b N ], f1(x)…f N (x) represents an N-dimensional output function, where N = 1, 2, 3...n, f1(x) represents a 1-dimensional output function, f N (x) represents the N-th dimension output function; φ 1 (x)…φ N (x) represents an N-dimensional nonlinear mapping function, where N = 1, 2, 3...n, φ 1 (x) represents a 1-dimensional nonlinear mapping function, φ N (x) represents the Nth nonlinear mapping function; b 1 ...b n represents an N-dimensional bias vector, where N = 1, 2, 3...n, b 1 represents a 1-dimensional bias vector, b n represents the N-th dimension bias vector;

[0032] For the case of N-dimensional output, the regression function can be expressed as:

[0033] y=φ T (x)φ(x)β+b

[0034] In the formula, φ(x) represents the nonlinear mapping function, β represents the coefficient of the model, and b represents the bias vector;

[0035] S4.2. Multi-kernel learning is introduced to perform convex combination of the kernel function of the model. Multi-kernel learning is used to perform linear combination of the kernel function of the MSVR model. The calculation process is as follows:

[0036]

[0037]

[0038] In the formula, κ * (x,y) represents the combined multiple kernel function, r i represents the weight of the i-th kernel function, κ i (x, y) represents different kernel functions, and m represents the number of kernel functions;

[0039] In order to reduce the dependence of the multi-core function on each base core function and reduce the base core weight β i The computational complexity of each base kernel function is determined according to the size of the root mean square error obtained by modeling each base kernel function, which is defined as follows:

[0040]

[0041] In the formula, n is the number of original training sample data, y i represents the i-th output variable, represents the i-th predicted value;

[0042] S4.3. Use differential evolution algorithm to optimize the hyperparameters of the model. The differential optimization algorithm mainly includes four parts: population initialization, mutation, crossover and selection. The specific calculation process is as follows:

[0043] S4.3.1 Population initialization, the specific expression is as follows:

[0044]

[0045] Where N p represents the population size, D represents the individual dimension, U represents the upper bound of the variable, L represents the lower bound of the variable, and rand[0,1] represents a random number generated between 0 and 1;

[0046] S4.3.2 Mutation is the operation of generating new individuals from the original individuals. The new variable vector is generated by the following formula. The specific expression is as follows:

[0047] v i,G+1 =x r1,G +F×(x r2,G -x r3,G )

[0048] In the formula, x r1 Indicates the random first individual number, x r2 Represents the random second individual number, x r3 represents the random third individual number, F represents the mutation operator, the value range is [0,2], and G represents the evolutionary generation;

[0049] S4.3.3 Crossover is the operation of generating new individuals from mutant individuals and current individuals according to certain rules. It is mainly to increase the diversity of interference parameter vectors. The specific expression is as follows:

[0050]

[0051] Where rand(j) represents the jth estimated value of a random number generator between [0,1]; rnbr(i) represents a randomly selected sequence; v ji,G+1 represents the value of the individual generated by the mutation operation in the jth dimension, x ji,G+1 represents the value of the current individual in the jth dimension, C R Represents the crossover operator, with a value range of [0,1];

[0052] S4.3.4 Selection

[0053] The selection operation mainly screens the new individuals generated by the crossover operation and selects the individuals that will enter the next generation;

[0054] In this algorithm, the value of the optimization function is called fitness value. The screening rule is to compare the fitness value of the new individual with the current individual. The individual with the smaller fitness value enters the next generation, which can ensure that the fitness value of the individual is continuously iterated and reduced.

[0055] S5. Evaluate the prediction model results; verify the prediction effect of the DE-MK-MSVR energy consumption model by comparing the prediction performance of different models; the evaluation indicators of the model include: determination coefficient (R 2 ), root mean square error (RMSE), mean error (MAE) and percentage error (MAPE), the calculation formulas are as follows:

[0056]

[0057]

[0058]

[0059]

[0060] In the formula, m represents the total number of samples, y i and represent the output variable and the predicted value respectively, Indicates the average of the measured values.

[0061] The beneficial effects of the present invention are:

[0062] 1. The multi-energy consumption prediction method for tin smelting process based on DE-MK-MSVR proposed in the present invention can simultaneously predict multiple types of energy consumption under the same input parameters. The method incorporates the correlation between energy consumption into the calculation, and the prediction accuracy is significantly improved.

[0063] 2. The multi-energy consumption prediction method for tin smelting process based on DE-MK-MSVR proposed in the present invention adopts differential evolution algorithm to find the optimal parameter value for the model, and improves the MSVR model by multi-kernel learning, so that the nonlinear fitting ability of the model is stronger and the prediction accuracy of the model is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 Flow chart of multi-energy consumption prediction method for tin smelting process;

[0065] Figure 2 Tin smelting data collection framework diagram;

[0066] Figure 3 Data cleaning and energy consumption influencing factors analysis process;

[0067] Figure 4 This is a flow chart of energy consumption prediction modeling based on DE-MK-MSVR;

[0068] Figure 5 The results of the proposed method are compared with those of other prediction models. DETAILED DESCRIPTION

[0069] The following is a detailed description of an embodiment of the present invention. This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and a specific operation process are given, but the protection scope of the present invention is not limited to the following embodiment.

[0070] Example 1

[0071] like Figure 1 to Figure 2 As shown in FIG. 1 , a multi-energy consumption prediction method for tin smelting process based on DE-MK-MSVR is shown in FIG. 1 . The specific steps are as follows:

[0072] S1. Collecting energy consumption data, production process data and real-time status data of the tin smelting process through a data acquisition device;

[0073] The data acquisition device includes: an electronic multifunctional smart meter, a smart thermal meter, a thermal integrator, a smart gas flow meter, a smart liquid flow meter, a weighing meter, a pressure, temperature, and a speed sensor, and uses the Internet and 5G network communication methods for data transmission. The collected data is shown in Table 1;

[0074] Table 1 Data collected during tin smelting and statistics on energy consumption types

[0075]

[0076] like Figure 3 As shown in the figure, the specific steps of data cleaning and energy consumption influencing factor analysis process are as follows:

[0077] S2, preprocessing the data obtained in S1, including: data format conversion, missing value filling, outlier detection and data dimension unification;

[0078] Since the collected data have missing values, outliers, and differences in collection cycles and dimensions, the data collected on site cannot be used directly for modeling, so the data needs to be preprocessed. The preprocessing includes: data format conversion, missing value filling, outlier detection, and data dimension unification. The data format conversion is to convert the categorical data in the original data into numerical data using One-Hot Encoding. The specific steps are as follows:

[0079] S2.1. Convert the categorical data in the original data into numerical data using one-hot encoding;

[0080] S2.2. Lagrange interpolation is used to make up for missing data. The specific expression is as follows:

[0081]

[0082]

[0083] Where: y j represents the value of the jth independent variable position, l j (x) represents the interpolation function, x represents the input variable, and x i represents the position of the i-th input variable, x j represents the position of the jth input variable, and k represents the number of independent variables;

[0084] S2.3. Use box plots to detect and remove outliers;

[0085] The box plot consists of five values: minimum value (min), lower quartile (Q1), median (median), upper quartile (Q3) and maximum value (max); in the box plot, IQR represents the interquartile range, IQR = Q3-Q1, the minimum observation value is set to Q1-1.5*IQR, the maximum observation value is Q3+1.5*IQR, and the value beyond the maximum and minimum range is considered an "outlier";

[0086] The box plot does not impose any restrictive requirements on the data. It only truly and intuitively shows the true appearance of the data distribution. The standard for judging outliers in the box plot is based on quartiles and interquartile ranges. Quartiles have a certain degree of robustness. The results of the box plot in identifying outliers are relatively objective, and it has certain advantages in identifying outliers.

[0087] S2.4. Use a normalization algorithm to eliminate the dimensional differences between variables. The specific expression is as follows:

[0088]

[0089] In the formula: μ represents the mean, σ represents the standard deviation, x represents the input variable, and x * represents the normalized data;

[0090] S3, in order to select variables with strong correlation with energy consumption and eliminate redundant variable learning, the distance correlation coefficient algorithm is used to analyze the input and output variables preprocessed by S2, and the redundant input variables are removed according to the correlation coefficient ratio, and the input variables with high correlation with the output variables are screened out;

[0091] The correlation coefficient ratio between the input and output variables is calculated by the distance correlation coefficient algorithm. The coefficient ratio range is [0,1]. If the ratio is 0, it means that the variables are independent of each other. The closer the ratio is to 1, the stronger the correlation between the variables. The correlation coefficient ratio of 0.9 is used as the threshold. If the correlation coefficient is higher than 0.9, it indicates that the variables are strongly correlated. The correlation between the two strongly correlated redundant variables and the output variables is compared respectively, and a group of variables with a higher correlation with the output variable is retained.

[0092] The specific expression of the distance correlation coefficient algorithm is as follows:

[0093]

[0094] Where: x represents the input variable, y represents the output variable, dcor(x,y) represents the distance correlation coefficient value, dcov(x,y) represents the distance covariance, dvar(x) and dvar(y) represent the distance covariance of x and y respectively;

[0095] like Figure 4 As shown, the present invention provides a multi-energy consumption prediction method for tin smelting process based on DE-MK-MSVR, including building a multi-output support vector regression (MSVR) model, improving the MSVR model by multi-kernel learning, optimizing model parameters by using a differential optimization algorithm, building a comparison model and predictive performance evaluation, which includes the following steps:

[0096] S4, setting the input variables and output variables as training samples and dividing them into training set and test set, building a DE-MK-MSVR energy consumption prediction model, using the redundant input variables removed in S3 to train the model, and using the test set to verify the prediction effect of the model;

[0097] The training set and the test set are divided into 8:2. The single-output support vector regression model is generalized according to the principle of structural risk minimization. Considering the strong correlation between multiple outputs, the absolute error is used to describe the overall error of the sample. At the same time, the norm of all regression weight vectors, the fitting error of a single output variable and the overall fitting error of the sample are minimized.

[0098] The specific steps of building the DE-MK-MSVR energy consumption prediction model are:

[0099] S4.1. Establish a multi-output support vector regression model;

[0100] For the function fitting problem, let the training sample be S = {(x i ,y i )}, where x i Indicates the position of the i-th input variable, y i Indicates the position of the i-th output variable, i=1,2,...,N p , N p Represents the population size, and the specific expression of constructing the regression function model is as follows:

[0101]

[0102] In the formula, Φ(x) represents the nonlinear mapping function of the high-dimensional space, W is the weight coefficient of the mapping function Φ(x), B represents the bias vector of the multi-output function F(x), W = [w 1 ,w 2 ,…,w N ],B=[b 1 ,b 2 ,…,b N ], f1(x)…f N (x) represents an N-dimensional output function, where N = 1, 2, 3...n, f1(x) represents a 1-dimensional output function, f N (x) represents the N-th dimension output function; φ 1 (x)…φN (x) represents an N-dimensional nonlinear mapping function, where N = 1, 2, 3...n, φ 1 (x) represents a 1-dimensional nonlinear mapping function, φ N (x) represents the Nth nonlinear mapping function; b 1 ...b n represents an N-dimensional bias vector, where N = 1, 2, 3...n, b 1 represents a 1-dimensional bias vector, b n represents the N-th dimension bias vector;

[0103] For the case of N-dimensional output, the regression function can be expressed as:

[0104] y=φ T (x)φ(x)β+b

[0105] In the formula, φ(x) represents the nonlinear mapping function, β represents the coefficient of the model, and b represents the bias vector;

[0106] S4.2. Multi-kernel learning is introduced to perform convex combination of the kernel function of the model. Multi-kernel learning is used to perform linear combination of the kernel function of the MSVR model. The calculation process is as follows:

[0107]

[0108]

[0109] In the formula, κ * (x,y) represents the combined multiple kernel function, r i represents the weight of the i-th kernel function, κ i (x, y) represents different kernel functions, and m represents the number of kernel functions;

[0110] In order to reduce the dependence of the multi-core function on each base core function and reduce the base core weight β i The computational complexity of each base kernel function is determined according to the root mean square error (RMSE) obtained by modeling each base kernel function, which is defined as follows:

[0111]

[0112] In the formula, n is the number of original training sample data, y i represents the i-th output variable, represents the i-th predicted value;

[0113] In order to further improve the prediction accuracy of the model, the differential evolution algorithm is used to optimize the model hyperparameters and the parameters of the kernel function. Among them, the parameters of the MSVR model include the penalty coefficient C; the kernel parameters include: gamma, degree, coef;

[0114] S4.3. Finally, the differential evolution algorithm is used to optimize the hyperparameters of the model. The differential optimization algorithm mainly includes four parts: population initialization, mutation, crossover and selection. The specific calculation process is as follows:

[0115] S4.3.1 Population initialization, the specific expression is as follows:

[0116]

[0117] In the formula, i represents the number of the i-th individual in the population; j represents the value of the individual in the j-th dimension; N p represents the population size; D represents the individual dimension; U represents the upper bound of the variable; L represents the lower bound of the variable, and rand[0,1] represents a random number generated between 0 and 1;

[0118] S4.3.2 Mutation is the operation of generating new individuals from the original individuals. The new variable vector is generated by the following formula. The specific expression is as follows:

[0119] v i,G+1 =x r1,G +F×(x r2,G -x r3,G )

[0120] In the formula, x r1 Indicates the random first individual number, x r2 Represents the random second individual number, x r3 represents the random third individual number, F represents the mutation operator, and its value range is [0,2]; G represents the evolutionary generation;

[0121] S4.3.3 Crossover is the operation of generating new individuals from mutant individuals and current individuals according to certain rules. It is mainly to increase the diversity of interference parameter vectors. The specific expression is as follows:

[0122]

[0123] Where rand(j) represents the jth estimated value of a random number generator between [0,1]; rnbr(i) represents a randomly selected sequence; v ji,G+1 represents the value of the individual generated by the mutation operation in the jth dimension, x ji,G+1 represents the value of the current individual in the jth dimension, C R Represents the crossover operator, with a value range of [0,1];

[0124] S4.3.4 Selection

[0125] The selection operation mainly screens the new individuals generated by the crossover operation and selects the individuals that will enter the next generation;

[0126] In this algorithm, the value of the optimization function is called the fitness value. The screening rule is to compare the fitness value of the new individual with the current individual. The individual with a smaller fitness value enters the next generation, thus ensuring that the fitness value of the individual is continuously iteratively reduced.

[0127] S5. Evaluate the prediction model results;

[0128] The prediction performance of different models is compared to verify the prediction effect of DE-MK-MSVR energy consumption model. The evaluation indicators of the model include: determination coefficient (R 2 ), root mean square error (RMSE), mean error (MAE) and percentage error (MAPE), the calculation formulas are as follows:

[0129]

[0130]

[0131]

[0132]

[0133] In the formula, m represents the total number of samples, y i and represent the true value and predicted value of the output variable respectively, Indicates the average of the measured values.

[0134] The comparison models include: Multi-output Gaussian process regression (MGPR) model and Multi-layer perceptron neural network (MLPNN).

[0135] The DE-MK-MSVR model was built, and the prediction model results were evaluated using the test set. The model prediction performance was compared using the classic multi-output multi-layer perceptron (MLNN) and multi-output Gaussian process regression (MGPR) to verify the prediction effect of the DE-MK-MSVR energy consumption model. The evaluation results are shown in Table 2:

[0136] Table 2 Evaluation indicators of different prediction models

[0137]

[0138]

[0139] The following is a prediction of the energy consumption of a tin smelting enterprise in the southwest region. The prediction results are as follows: Figure 5 shown.

[0140] The multi-output Gaussian process regression (MGPR) model and the multi-layer perceptron neural network (MLPNN) model were experimentally compared with the DE_MK_MSVR model proposed in this paper. The evaluation indicators of each model are shown in Table 2. The energy consumption results predicted by different methods are shown in Figure 5 As shown in Table 2, the DE_MK_MSVR model has the best prediction performance under multiple energy consumption. 2 Reached above 0.98, the R 2 Reaching above 0.96, the R 2 is above 0.93. Among all the models, the DE_MK_MSVR model has the best evaluation index. Figure 5 It can be seen that among all the prediction models, the prediction value of the DE_MK_MSVR model is closer to the true value, the prediction performance is the best, and the model stability is also the best. It can be seen that the model constructed in this paper has a better fitting effect on multi-category energy consumption and better model stability. The model has certain applicability in energy consumption prediction in other similar production enterprises.

Claims

1. A multi-energy consumption prediction method for tin smelting process based on DE-MK-MSVR, characterized in that: The specific steps of the method are as follows: S1. Collecting energy consumption data, production process data and real-time status data of the tin smelting process through a data acquisition device; S2, taking the energy consumption data obtained in S1 as output variables, and the process data and real-time status data as input variables and preprocessing them; S3, using the distance correlation coefficient algorithm to analyze the input and output variables preprocessed by S2, and removing redundant input variables according to the correlation coefficient ratio, and screening out input variables with high correlation with the output variables; S4, setting the input variables and output variables as training samples and dividing them into training set and test set, building a DE-MK-MSVR energy consumption prediction model, using the redundant input variables removed in S3 to train the model, and using the test set to verify the prediction effect of the model; S5. Evaluate the prediction model results.

2. The method for predicting multi-energy consumption in tin smelting process based on DE-MK-MSVR according to claim 1 is characterized in that: The data acquisition device in S1 includes: an electronic multifunctional intelligent electric meter, an intelligent thermal meter, a thermal integrator, an intelligent gas flow meter, an intelligent liquid flow meter, a weighing meter, and pressure, temperature, and speed sensors.

3. The method for predicting multi-energy consumption in tin smelting process based on DE-MK-MSVR according to claim 1, characterized in that: The preprocessing in S2 includes: data format conversion, missing value filling, outlier detection and data dimension unification. The specific process is as follows: S2.

1. Convert the categorical data in the collected data into numerical data using one-hot encoding; S2.

2. Lagrange interpolation is used to supplement the missing data; S2.

3. Use box plots to detect and remove outliers; S2.

4. Use a normalization algorithm to eliminate dimensional differences between variables.

4. The method for predicting multi-energy consumption in tin smelting process based on DE-MK-MSVR according to claim 1, characterized in that: The specific process of S3 is: by using the distance correlation coefficient algorithm, the correlation coefficient ratio between the input and output variables is calculated. The coefficient ratio range is [0,1]. If the ratio is 0, it means that the variables are independent of each other. The closer the ratio is to 1, the stronger the correlation between the variables. The specific expression of the distance correlation coefficient algorithm is as follows: Where: x represents the input variable, y represents the output variable, dcor(x,y) represents the distance correlation coefficient value, dcov(x,y) represents the distance covariance, dvar(x) and dvar(y) represent the distance covariance of x and y respectively.

5. The method for predicting multi-energy consumption in tin smelting process based on DE-MK-MSVR according to claim 1, characterized in that: The total sample ratio of the training set to the test set in S4 is 8:

2. The specific steps of building the DE-MK-MSVR energy consumption prediction model are as follows: S4.

1. Establish a multi-output support vector regression function model; For the function fitting problem, let the training sample be S = {(x i ,y i )}, where x i Indicates the position of the i-th input variable, y i Indicates the position of the i-th output variable, i=1,2,...,N p , N p Represents the population size, and the specific expression of constructing the regression function model is as follows: In the formula, Φ(x) represents the nonlinear mapping function of the high-dimensional space, W is the weight coefficient of the mapping function Φ(x), B represents the bias vector of the multi-output function F(x), W = [w 1 ,w 2 ,…,w N ],B=[b 1 ,b 2 ,…,b N ], f1(x)…f N (x) is represented as an N-dimensional output function, φ 1 (x)…φ N (x) represents the N-dimensional nonlinear mapping function, b 1 ...b n represents the N-dimensional bias vector; For the case of N-dimensional output, the regression function model can be expressed as: y=φ T (x)φ(x)β+b In the formula, φ(x) represents the nonlinear mapping function, β represents the coefficient of the model, and b represents the bias vector; S4.

2. Multi-kernel learning is introduced to perform convex combination of the kernel function of the model. Multi-kernel learning is used to perform linear combination of the kernel function of the MSVR model. The calculation process is as follows: In the formula, κ * (x,y) represents the combined multiple kernel function, r i represents the weight of the i-th kernel function, κ i (x, y) represents different kernel functions, and m represents the number of kernel functions; In order to reduce the dependence of the multi-core function on each base core function and reduce the base core weight β i The computational complexity of each base kernel function is determined according to the size of the root mean square error obtained by modeling each base kernel function, which is defined as follows: In the formula, n is the number of original training sample data, y i represents the i-th output variable, represents the i-th predicted value; S4.

3. Use differential evolution algorithm to optimize the hyperparameters of the model. The differential optimization algorithm mainly includes four parts: population initialization, mutation, crossover and selection. The specific calculation process is as follows: S4.3.1 Population initialization, the specific expression is as follows: Where N p represents the population size, D represents the individual dimension, U represents the upper bound of the variable, L represents the lower bound of the variable, and rand[0,1] represents a random number generated between 0 and 1; S4.3.2 variation, the specific expression is as follows: v i,G+1 =x r1,G +F×(x r2,G -x r3,G ) In the formula, x r1 Indicates the random first individual number, x r2 Represents the random second individual number, x r3 represents the random third individual serial number; F represents the mutation operator, and its value range is [0,2]; G represents the evolutionary generation; S4.3.3 Crossover, the specific expression is as follows: Where rand(j) represents the jth estimated value of a random number generator between [0,1]; rnbr(i) represents a randomly selected sequence; v ji,G+1 represents the value of the individual generated by the mutation operation in the jth dimension, x ji,G+1 represents the value of the current individual in the jth dimension, C R Represents the crossover operator, with a value range of [0,1]; S4.3.4 Selection The selection operation mainly screens the new individuals generated by the crossover operation and selects the individuals that will enter the next generation.

6. The method for predicting multi-energy consumption in tin smelting process based on DE-MK-MSVR according to claim 1, characterized in that: The model prediction evaluation indicators in S5 include: determination coefficient, root mean square error, average error and percentage error; the specific expressions of the determination coefficient, root mean square error, average error and percentage error are as follows: In the formula, m represents the total number of samples, y i and Represent the output variable and the predicted value respectively, y represents the average value of the measured value, R 2 represents the coefficient of determination, RMSE represents the root mean square error, MAE represents the average error, and MAPE represents the percentage error.

Citation Information

Cited By

  • Tin smelting anomaly detection method fusing improved optimization algorithm and diffusion model

    CN121185362A

  • A tin smelting anomaly detection method fusing improved optimization algorithm and diffusion model

    CN121185362B