A sintering mixture tank level prediction method and system based on a convolutional neural network and a memory

Abstract

The application discloses a sintering mixture tank level prediction method and system based on a convolutional neural network and a memory. The method screens characteristic variables related to the mixture tank level as model input parameters through a random forest algorithm, performs autocorrelation analysis on time series data of the input variables, and constructs a two-dimensional characteristic variable matrix. The two-dimensional matrix is input into a pre-trained mixture tank level prediction model based on a convolutional neural network to obtain the mixture tank level output. Based on the above method, the application develops a sintering mixture tank level prediction system. The system can predict the mixture tank level in advance, help operators discover changes in the mixture tank level in time, and provide an important basis for accurate regulation and control of operating parameters.

Description

Technical Field

[0001] This invention relates to a method for predicting the level of sintering mixture tanks, and more particularly to a method, system, and memory for predicting the level of sintering mixture tanks based on convolutional neural networks, belonging to the field of iron ore sintering production. Background Technology

[0002] Sintering is a crucial step in the steel production process, and its stable, efficient, and low-consumption operation is a prerequisite for cost reduction and efficiency improvement in steel production. After batching, mixing, and granulation, the iron ore mixture is conveyed to the mixing trough, and then unloaded onto the sintering machine trolley via a circular roller and multi-roller distributor. The mixing trough serves as a storage and buffer. Maintaining a stable material level in the mixing trough is critical for the smooth and stable operation of sintering production. Too low a material level can lead to uneven distribution, resulting in a higher material level in the middle and lower levels on the sides of the sintering machine trolley, increasing the load on the blower and potentially requiring machine shutdown due to insufficient material. Conversely, too high a material level can cause compression and damage to the granulated pellets, affecting the permeability of the mixture.

[0003] In actual production, operators typically need to constantly monitor changes in the mixing tank level and take appropriate operational controls. However, the mixing tank level fluctuates frequently and rapidly, and its response time to various operations is relatively long. Therefore, it is necessary to establish a predictive model for the mixing tank level to anticipate level changes and make timely and accurate adjustments to operating parameters, thereby reducing fluctuations in the mixing tank level. Mixing tank level prediction plays a crucial supporting role in improving the stability of the sintering process. Summary of the Invention

[0004] To address the problems existing in the prior art, the first objective of this invention is to provide a method for predicting the level of sintering mixing tanks based on convolutional neural networks. This method filters feature parameters related to the level of the mixing tank, selects data obtained from each feature parameter over time to construct a two-dimensional feature variable matrix, and builds a mixing tank level prediction model using a convolutional neural network. Based on the correlation between sintering feature parameters and the mixing tank level, the resulting prediction method, while maintaining the prediction rate, significantly reduces the impact of input variable fluctuations on the prediction results, thereby greatly improving the prediction accuracy.

[0005] A second objective of the present invention is to provide a readable storage device containing a computer program for implementing the above-described method for predicting the level of sintering mixture tank.

[0006] The third objective of this invention is to provide a sintering mixing tank level prediction system based on a convolutional neural network. This prediction system, based on the synergistic effect of a programmable logic controller, a processor, and a readable storage device, achieves accurate prediction of the mixing tank level, thereby enabling real-time monitoring and control of the mixing tank level without the need for manual adjustment of the mixing tank level, thus realizing intelligent control of the sintering process.

[0007] To address the aforementioned technical problems, this invention provides a method for predicting the material level in a sintering mixing tank based on a convolutional neural network, comprising:

[0008] Step S1: Obtain sintering process data, divide the data into training set and test set, and then filter i sets of feature parameters related to the material level in the mixing tank through the training set;

[0009] Step S2: Extract the data of each feature parameter group 1 to T from each group of feature parameters according to the time interval, and stack them into an i*T two-dimensional feature parameter matrix;

[0010] Step S3: Input the obtained two-dimensional feature variable matrix into the convolutional neural network model to construct the material level prediction model of the mixing tank, and test it through the test set to obtain the result.

[0011] As a preferred embodiment, the process of filtering the feature parameters includes:

[0012] Step S1-1: After preprocessing and standardization, the sintering process data is randomly divided into a training set and a test set with a ratio of 3 to 4:1.

[0013] Step S1-2: Using the stochastic Senli algorithm, establish more than 18 characteristic parameters, including mixture moisture and material layer thickness, and calculate the importance index n of each characteristic variable for the material level in the mixing tank. k ;

[0014] Step S1-3: After normalizing the importance indices of all feature parameters, select features with indices greater than the threshold. As the number of feature parameters increases, the mean squared error of the model built using random forest decreases significantly and tends to stabilize, at which point the appropriate number of feature parameters can be determined. Sort the feature parameters by their importance indices; the minimum value of the feature parameter importance index can then be determined as the threshold.

[0015] As a preferred option, the importance index n k The calculation process is as follows:

[0016] Equation 1: n k =w k *G k -w left *G left -w right*G right ;

[0017] Formula 2:

[0018] In Equations 1 and 2, w k w left and w right G represents the ratio of the number of training samples in node k and its left and right child nodes to the total number of training samples. k G left and G right Let c be the Gini index of node k and its left and right child nodes, respectively. i For a certain feature variable, v ij N is a split value of the variable. s For the total number of training samples, y left and y right These represent the number of training samples for the left and right child nodes after the segmentation, respectively.

[0019] As a preferred embodiment, the process of constructing the two-dimensional feature parameter matrix is ​​as follows:

[0020] Step S2-1: Arrange the selected feature parameters in chronological order, and denote them as X. i,1 ,X i,2 ,…,X i,T , where i is the number of the feature parameter and T is the time series;

[0021] Step S2-2: Calculate the autocorrelation coefficient of each variable according to the time intervals from 1 to K. Select the time interval when all autocorrelation coefficients are greater than the threshold (the threshold is usually 0.6), which is the time interval of each feature parameter.

[0022] Step S2-3: Obtain the i*T two-dimensional feature parameter matrix by using the types of feature parameters as columns and the time intervals as rows.

[0023] As a preferred option, the autocorrelation coefficient r i,k The calculation process is as follows:

[0024] Formula 3:

[0025] In Equation 3, Let be the time series mean of variable i.

[0026] As a preferred embodiment, the structure of the convolutional neural network model consists of a convolutional module and a fully connected regression module.

[0027] As a preferred embodiment, the convolution module includes three convolution-batch normalization-activation function modules. The first two convolution modules use a 3x3 kernel with a stride of 2 to perform convolution operations, and the last module uses a 3x3 kernel with a stride of 1 to perform convolution operations.

[0028] As a preferred embodiment, the fully connected regression module includes a global average pooling layer and a fully connected ReLU activation function-Dropout layer.

[0029] As a preferred embodiment, Dropout refers to randomly selecting 10-20% of neurons to deactivate during each training session.

[0030] As a preferred embodiment, the number of neurons in the fully connected regression module is 64-32-1.

[0031] The present invention also provides a computer-readable storage device containing a computer program that can implement the sintering mixture tank level prediction method based on convolutional neural networks as described in any of the above claims.

[0032] The present invention also provides a sintering mixture tank level prediction system based on convolutional neural network, comprising: a programmable logic controller (100), a memory (200) as described in claim 9, and a processor (300).

[0033] Compared with the prior art, the beneficial technical effects of the technical solution of the present invention are as follows:

[0034] 1) The prediction method provided by this invention selects feature parameters related to the level of the mixing tank, selects the data obtained by each feature parameter with time as the scale, constructs a two-dimensional feature variable matrix, and constructs a mixing tank level prediction model through a convolutional neural network. Based on the correlation between sintering feature parameters and mixing tank level, the obtained prediction method ensures the prediction rate of mixing tank level while significantly reducing the impact of input variable fluctuations on the prediction results, thereby greatly improving the prediction accuracy.

[0035] 2) The technical solution provided by this invention uses convolutional neural networks for feature extraction based on time series reconstruction of variables, deeply mines the correlation between process variables and their temporal relationships and the material level in the mixing tank, and makes full use of the temporal characteristics of process detection parameters to improve the accuracy of mixing tank material level prediction.

[0036] 3) The prediction system provided by this invention is based on the synergistic effect between the programmable logic controller, the processor and the readable memory, which realizes the accurate prediction of the material level in the mixing tank. In this way, the material level in the mixing tank can be monitored and controlled in real time without the need for manual control of the material level change in the mixing tank, thus realizing intelligent control of the sintering process. Attached Figure Description

[0037] Figure 1 A flowchart illustrating the method for predicting the material level in a sintering mixing tank;

[0038] Figure 2 This is a graph showing the accuracy of prediction results in the example.

[0039] Figure 3 Diagram showing the configuration of the sintering mixing tank level prediction system;

[0040] Figure 4 This is the computer program interface for the sintering mixing tank level prediction system. Detailed Implementation

[0041] To facilitate understanding of the present invention, the present invention will be described more fully and in detail below with reference to the accompanying drawings and preferred embodiments, but the scope of protection of the present invention is not limited to the following specific embodiments.

[0042] The workflow of a sintering mixing tank level prediction method is as follows: Figure 1 As shown, including

[0043] Step 1: Use nine parameters as input variables: moisture content of the sintering mixture, overall conveying capacity, material level in the mixing trough, opening of the material distribution gate, rotational speed of the roller feeder, material layer thickness, trolley speed, negative pressure of the ignition furnace, and negative pressure of the No. 1 wind box.

[0044] Step 2: Extract 10-dimensional data for each input variable with a time interval of 1 minute, and stack them into a 9-row, 10-column two-dimensional feature variable matrix;

[0045] Step 3: Input the two-dimensional feature variables into the pre-trained convolutional neural network model to calculate the material level in the mixing tank at the next moment, thereby achieving advance prediction of the material level.

[0046] Step one specifically includes:

[0047] After preprocessing and standardizing the data collected from the sintering production site, the samples were randomly divided into a training set and a test set in a 4:1 ratio. A random forest algorithm was used to establish a model relating 18-dimensional feature variables, such as mixture moisture and material layer thickness, to the material level in the mixing tank. The importance index of each feature variable was calculated. The importance index of each feature was then normalized, and values ​​greater than a threshold T were selected. hNine characteristics (mixture moisture content, overall conveying capacity, mixture trough level, distribution gate opening, roller feeder speed, material layer thickness, trolley speed, ignition furnace negative pressure, and No. 1 wind box negative pressure) are used as input variables for mixture trough level prediction.

[0048] Step two specifically includes:

[0049] Extract the time series data of the nine input variables, denoted as X. i,1 ,X i,2 ,…,X i,T Where i is the input variable number; T is the time series length, with a value of 20 min; calculate the autocorrelation coefficient of each variable for time intervals from 1 to k; select the autocorrelation coefficient r of all input variables. i,k The time interval k = 10 when all values ​​are greater than the threshold; combine the sequence data of the 9 input variables with a time length of 10 min to obtain a two-dimensional feature vector of 9 rows and 10 columns.

[0050] Step three specifically includes:

[0051] The model's network structure consists of convolutional modules, global average pooling layers, and fully connected regression modules. The convolutional modules include three convolution-batch normalization-activation function modules. The first two convolutional modules use 3x3 kernels with a stride of 2, while the last module uses a 3x3 kernel with a stride of 1. The fully connected regression module includes global average pooling layers and fully connected ReLU activation function-Dropout layers. Dropout randomly selects 20% of neurons to deactivate during each training session. The number of neurons in the fully connected layers is 64-32-1.

[0052] Example 1

[0053] like Figure 1 As shown, this invention provides a method for predicting the tank position of sintering mixtures, comprising the following steps:

[0054] 1) Following the method in step one above, the random forest algorithm was used to select highly correlated variables from the initially chosen 18-dimensional process variables as input variables for predicting the level of the mixing tank. The feature contribution rates of the process variables are shown in Table 1. A threshold T was set. h =0.03, and nine variables were determined as inputs: moisture content of the mixture, total conveying capacity, material level in the mixing trough, opening of the material distribution gate, rotational speed of the roller feeder, material layer thickness, trolley speed, negative pressure of the ignition furnace, and negative pressure of the No. 1 wind box;

[0055] Table 1

[0056]

[0057]

[0058] 2) Following the method in step two above, a time-series correlation analysis was performed on the sintering process variables selected in Table 1. The autocorrelation coefficients of each variable at different time intervals are shown in Table 2. It can be seen that when k = 10, the autocorrelation coefficients of all characteristic variables are greater than 0.6, indicating that the variables have strong autocorrelation over 10 sampling periods. Therefore, a time scale of 10 was chosen. The data were concatenated into a 9*10 sintering process time-series feature matrix, which, together with the corresponding mixing tank level at the time, forms the sample dataset, which is the input-output structure of the mixing tank level prediction model.

[0059] Table 2

[0060]

[0061]

[0062] 3) Based on the method in step three above, the network structure of the sintering mixture tank level prediction model is as follows: Figure 2 As shown, using production process data, after obtaining model input through steps one and two, the network structure is trained to obtain the final mixing tank level prediction model. The prediction results of the model are shown in Table 3, and the model prediction effect over a period of time is shown in Table 4. The accuracy rate is judged by ±8t of the measured value.

[0063] Table 3

[0064] Serial Number 1 2 3 4 …… 82 83 84 Measured values 76.41 77.70 82.50 82.90 …… 97.99 100.18 101.38 Predicted value 76.01 80.47 77.83 80.16 …… 99.64 99.79 102.10

[0065] Table 4

[0066] Root mean square error Mean relative error (%) Accuracy (%) 11.24 3.63 96.60

[0067] This invention also provides a sintering mix tank position prediction system based on a convolutional neural network, the structure of which is as follows: Figure 3 As shown: a programmable logic controller (100), a memory (200), a processor (300), and a computer program stored in the memory (200) and capable of running on the processor (300). The computer program is based on the Vue+SpringBoot framework and is primarily written in Java. The program interface is shown below. Figure 4 As shown.

Claims

1. A method for predicting the material level in a sintering mixing tank based on a convolutional neural network, characterized in that, include: Step S1: Obtain sintering process data, divide the data into training set and test set, and then filter i sets of feature parameters related to the material level in the mixing tank through the training set; Step S2: Extract the data of each feature parameter group 1 to T from each group of feature parameters according to the time interval, and stack them into an i*T two-dimensional feature parameter matrix; Step S3: Input the obtained two-dimensional feature variable matrix into the convolutional neural network model to construct the material level prediction model of the mixing tank, and test it through the test set to obtain the result; The process of selecting i sets of feature parameters related to the material level in the mixing tank through the training set includes: Step S1-1: After preprocessing and standardization, the sintering process data is randomly divided into a training set and a test set with a ratio of 3 to 4:

1. Step S1-2: Use the random forest algorithm to establish more than 18 feature parameters, including mixture moisture and material layer thickness, and calculate the importance index of each feature variable for the material level in the mixing tank. ; Step S1-3: After normalizing the importance index of all feature parameters, select features that are greater than the threshold. As the number of feature parameters increases, the mean square error of the model built using random forest decreases significantly and tends to stabilize. Determine the appropriate number of feature parameters. Sort the importance index of the feature parameters. At this time, the minimum value of the feature parameter importance index is determined as the threshold.

2. The method for predicting the material level in a sintering mixing tank based on a convolutional neural network according to claim 1, characterized in that: The importance index The calculation process is as follows: Formula 1: ; Formula 2: ; In Equations 1 and 2, , and These represent the ratios of the number of training samples in node k and its left and right child nodes to the total number of training samples, respectively. , and Let be the Gini indexes of node k and its left and right child nodes, respectively. For a certain feature variable, For a slice value of the variable, For the total number of training samples, and These represent the number of training samples for the left and right child nodes after the segmentation, respectively.

3. The method for predicting the material level in a sintering mixing tank based on a convolutional neural network according to claim 1, characterized in that: The process of constructing the two-dimensional feature parameter matrix is ​​as follows: Step S2-1: Arrange the selected feature parameters in chronological order, and denote them as follows: ,in i The feature parameter number, T It is a time series; Step S2-2: Calculate the autocorrelation coefficient of each variable according to the time intervals from 1 to K. Select the time interval when all autocorrelation coefficients are greater than the threshold, which is the time interval of each feature parameter. Step S2-3: Obtain the i*T two-dimensional feature parameter matrix by using the types of feature parameters as columns and the time intervals as rows.

4. The method for predicting the material level in a sintering mixing tank based on a convolutional neural network according to claim 3, characterized in that: The autocorrelation coefficient The calculation process is as follows: Formula 3: ; In Equation 3, Let be the time series mean of variable i.

5. The method for predicting the material level in a sintering mixing tank based on a convolutional neural network according to claim 1, characterized in that: The structure of the convolutional neural network model consists of a convolutional module and a fully connected regression module.

6. The method for predicting the material level in a sintering mixing tank based on a convolutional neural network according to claim 5, characterized in that: The convolution module includes three convolution-batch normalization-activation function modules. The first two convolution modules use a 3x3 kernel with a stride of 2 to perform convolution operations, while the last module uses a 3x3 kernel with a stride of 1 to perform convolution operations.

7. The method for predicting the material level in a sintering mixing tank based on a convolutional neural network according to claim 5, characterized in that: The fully connected regression module includes a global average pooling layer and a fully connected ReLU activation function-Dropout layer; Dropout refers to randomly selecting 10-20% of neurons to deactivate during each training process; the number of neurons in the fully connected regression module is 64-32-1.

8. A computer-readable storage device comprising a computer program, characterized in that: The computer program can implement the sintering mixture tank level prediction method based on convolutional neural networks as described in any one of claims 1 to 7.

9. A sintering mixing tank level prediction system based on convolutional neural networks, characterized in that, include: Programmable logic controller (100), computer-readable storage (200) as claimed in claim 8, and processor (300).

Images