A soft measurement method for general powder particle size in the production process of high-temperature electrical grade magnesium oxide powder

Abstract

The present invention discloses a soft measurement method for the particle size of general powder in the production process of high-temperature electrical grade magnesium oxide powder, comprising the following steps: establishing a sample set; standardizing the sample data, performing dimensionality reduction processing using the PCA method, and eliminating redundant relationships between auxiliary variables; using the XGBoost modeling method to establish an initialization objective function that considers the loss function and model complexity for the k-th output variable, and transforming the objective function using a second-order Taylor expansion; defining a decision tree, and introducing the tree structure into the objective function, and constructing the optimal tree using a greedy algorithm, and repeating the above process t k times, and finally generates t k A decision tree is constructed to obtain an ideal soft measurement model; the ideal model uses the outputs of these decision trees combined as the predicted output of the soft measurement results for the general powder particle size during the production process of high-temperature electrical-grade magnesium oxide powder; and the ideal soft measurement model for general powder particle size is corrected using a method that combines deviation compensation correction with real-time learning. The soft measurement model correction method of the present invention enables online and real-time correction of the soft measurement model and is adaptive to changes in working conditions.

Description

Technical Field

[0001] The invention relates to the field of advanced control technology for a high-temperature electrical grade magnesium oxide powder production process, in particular to a soft measurement method for general powder particle size in a high-temperature electrical grade magnesium oxide powder production process. Background Art

[0002] Fused magnesium oxide crushing and screening is an important process in the production of high-temperature electrical grade magnesium oxide powder. It breaks larger lumps of fused magnesium oxide into finer powders, and then screens out fused magnesium oxide powders with a suitable particle size range for subsequent high-temperature calcination and modification processes.

[0003] The crushing and screening process of fused magnesium oxide is as follows: Figure 1 As shown. A vibrating feeder feeds the lumpy fused magnesia into a hammer crusher for crushing. The crushed fused magnesia powder is then transported via a bucket elevator to a higher location and then to a vibrating screen. The vibrating screen then classifies the fused magnesia powder according to particle size. The coarser particles return to the hammer crusher for further crushing, while the finer particles are collected by bags. The particles of appropriate size then enter the air classifier for further screening. The air classifier further separates the fused magnesia powder of appropriate particle size. The coarser particles, considered the fused magnesia powder of appropriate particle size, are transported via a bucket elevator to a higher location for magnetic separation to remove iron. The collected particles are then sent to a high-temperature furnace for calcination. The finer particles are screened and collected by a collector, and the remaining, very small, finer particles are collected by an electrostatic precipitator.

[0004] The quality indicator of the fused magnesia crushing and screening process is the particle size distribution (mass fraction of the range of +35 mesh, -35 mesh to +325 mesh, -325 mesh) of the fused magnesia powder collected after magnetic separation and iron removal. This is also known as the "general powder particle size." Currently, high-temperature electrical-grade magnesia powder manufacturers test general powder particle size by manually sampling the fused magnesia powder every half hour and testing the sample offline using a standard vibrating screen. This method suffers from long sampling cycles and significant lag, making it difficult to provide timely guidance for production and hindering the implementation of advanced control.

[0005] With the development of science and technology, soft sensing technology has provided a feasible solution for online measurement of general powder particle size. Establishing a soft measurement method with strong generalization ability for online measurement of general powder particle size is of great significance for stabilizing the production of crushing and screening processes and implementing advanced control of crushing and screening processes. Summary of the Invention

[0006] Aiming at the problem that the general powder particle size cannot be detected online during the production process of high-temperature electrical grade magnesium oxide powder, the technical problem to be solved by the present invention is to provide a soft measurement method for the general powder particle size in the production process of high-temperature electrical grade magnesium oxide powder, which can continuously provide general powder particle size measurement values ​​online, has strong generalization ability, and is adaptable.

[0007] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:

[0008] A soft measurement method for the general powder particle size in the production process of high-temperature electrical-grade magnesium oxide powder is provided. The method performs the following steps to obtain an ideal soft measurement model and calibrate it in real time. The method is used to perform soft measurement prediction of the unknown general powder particle size in the production process of unknown high-temperature electrical-grade magnesium oxide powder. The method includes the following steps:

[0009] Establish a sample set: select auxiliary variables in the fused magnesium oxide crushing and screening process, measure the general powder particle size as a label, and establish a sample set with {auxiliary variables as input and fitted labels as output};

[0010] Data preprocessing: standardize the sample data, use the PCA method to reduce the dimension, and eliminate the redundant relationship between auxiliary variables;

[0011] Establish a soft sensor model based on the XGBoost method: Use the XGBoost modeling method to establish an initialization objective function for the k-th output variable that takes into account the loss function and model complexity, and use the second-order Taylor expansion to transform the objective function; define a decision tree, and introduce the tree structure into the objective function, use the greedy algorithm to build the optimal tree, and repeat the above process t k times, and finally generates t k A decision tree is generated to obtain an ideal soft measurement model; the ideal model uses the outputs of these decision trees added together as the prediction output of the soft measurement results of the general powder particle size in the production process of high-temperature electrical grade magnesium oxide powder;

[0012] Soft measurement model correction: The ideal soft measurement model of general powder particle size is corrected by combining deviation compensation correction with real-time learning.

[0013] The auxiliary variables include original particle size, hardness, feed amount, crusher current, return amount, and air classifier speed; the fitting labels are manual detection values ​​under two particle size distribution ranges.

[0014] The standardization adopts the normalization method to perform the following steps:

[0015]

[0016] Where u ij ′ represents the jth sample value of the i-th input variable, is the sample mean of the i-th input variable, σi ′ is the sample standard deviation of the i-th input variable, u ij for u ij ′ is the standardized value, i=1,2,…,6; j=1,2,…,N; N is the number of samples.

[0017] The dimensionality reduction process using the PCA method is as follows:

[0018] Calculate the new input variable matrix X=UP after data preprocessing; where U∈R N×6 The matrix formed by the input data after normalization, the matrix P is the first s columns of the feature matrix P0;

[0019] Let the tth value of the kth output variable be k The formula for initializing the objective function of a decision tree is as follows:

[0020]

[0021] Among them, the loss function of the i-th (i=1,2,...,N) sample is y ki represents the i-th sample value of the k-th output variable; represents the first t of the k-th output k The predicted value of the i-th sample by all decision trees together; Ω ka (f ka ) represents the model complexity of the a-th tree of the k-th output.

[0022] The initial objective function is transformed and simplified by Taylor expansion to remove the constant term to obtain the objective function:

[0023]

[0024] Define a decision tree

[0025] Bundle Defined as in Represents the result of a tree, assuming that the tree has leaf nodes, used to convert the input x i ∈R s Mapped to a leaf node, Is the length of A one-dimensional vector representing the weight of the leaf node, that is, the predicted value of the decision tree;

[0026] The following formula is used to measure the complexity of a tree:

[0027]

[0028] in, and Both are hyperparameters that control the degree of penalty, and their values ​​range from [0, 1]. The objective function shown in formula (8) is transformed into:

[0029]

[0030] All samples x belonging to the jth leaf node i , is classified into a leaf node sample set, mathematically expressed as: Calculate the loss of all leaf node samples as make The objective function becomes To minimize the objective function, let right The partial derivative of is zero, so

[0031] Will Substitute into the objective function and get the objective function

[0032] Using the greedy algorithm to construct the optimal tree includes:

[0033] Determination of candidate split points: For each new input variable, the range of values ​​is divided into three or four quartiles according to the number of samples. Each quartile is a candidate split point.

[0034] A greedy algorithm is used to split the tree nodes: considering all new input variables, split at each candidate split point of each new input variable; after each split, the original leaf node is further split into two left and right child leaf nodes, and the sample set in the original leaf node will be dispersed to the left and right leaf nodes based on the comparison between the node and the value at the candidate split point; after a new node is split, it is tested whether this split will bring gain to the loss function. The gain is defined as follows:

[0035]

[0036] The L subscript is the optimal value of the objective function when the value is divided into the left subtree, and the R subscript is the optimal value of the objective function when the value is divided into the right subtree;

[0037] If Gain < 0, stop the split; if the number of samples of any leaf node is lower than the threshold Nu min , stop this split, in which Nu min The value range is [1, 10]; if the depth of the tree reaches the maximum value dep max , stop this split, where dep maxThe value range is a positive integer in the interval [3, 9]. Otherwise, after all candidate split points have completed the split, the candidate split point with Gain>0 and the largest value is selected as the best split point for this round of splitting. Then, at the next leaf node, all candidate split points of all new input variables are considered, and the above process is repeated to perform the next round of splitting on the leaf node until the condition for stopping splitting is met. The final number of leaf nodes is obtained as Substitute into the objective function.

[0038] The form of the soft measurement model based on the XGBoost method is:

[0039]

[0040] Among them, u I Represents the input data u I The results after standardization, t represents the kth output variable k The sum of the outputs of the decision trees is the soft measurement result of the kth output variable of the general powder particle size in the high-temperature electrical grade magnesium oxide production process obtained based on the input data.

[0041] The soft sensor model correction includes:

[0042] a) comparing the manual detection value of the general powder particle size with the soft measurement result; if the absolute value of the error is less than a predefined first threshold, the original soft measurement model is kept unchanged and continues to be used;

[0043] b) If the absolute value of the error is greater than the first threshold and less than the second threshold, a deviation compensation method is used for correction, and the actual error is added to the soft measurement result as the final soft measurement result;

[0044] c) If the absolute values ​​of the errors are all greater than the second threshold, Z sample data with the highest similarity to the new input data are selected, the new input data are preprocessed, the soft measurement model is rebuilt using the XGBoost method, and the new input and output data are added to the training samples.

[0045] The present invention has the following advantages and beneficial effects:

[0046] 1. It can measure the "general powder particle size" indicator which is difficult to detect online by traditional methods but is very important, providing the prerequisite for the implementation of advanced control in the production process of high-temperature electrical grade magnesium oxide powder;

[0047] 2. The XGBoost method is used to establish a soft sensing model, which enhances the generalization ability of the model and overcomes the risk of overfitting;

[0048] 3. Soft measurement model correction can perform online real-time correction on the soft measurement model and has the ability to adapt to changes in working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a schematic diagram of the crushing and screening process in the production of high-temperature electrical grade magnesium oxide;

[0050] Figure 2 This is the flow chart of the soft measurement process of general powder particle size. DETAILED DESCRIPTION

[0051] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the specific implementation methods of the present invention are further described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art can make similar improvements without violating the scope of the invention. Therefore, the present invention is not limited to the specific implementation disclosed below.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art of the art to which the present invention pertains. The terms used in the specification of the invention herein are for the purpose of describing specific embodiments only and are not intended to limit the present invention.

[0053] 1. Selection of auxiliary variables:

[0054] Through a comprehensive analysis of the fused magnesia crushing and screening process used in the production of high-temperature electrical-grade magnesia powder, we determined that six key parameters, y1 (the mass fraction of the +35 mesh range) and y2 (the mass fraction of the -35 mesh to +325 mesh range), influence the particle size of the general powder (mass fractions distributed within the ranges of +35 mesh, -35 mesh, +325 mesh, and -325 mesh) and affect the particle size of the general powder. These parameters include the original particle size of the fused magnesia, hardness, feed rate, crusher current, return rate, and air classifier speed. Therefore, these six variables are used as auxiliary variables in the soft sensor model and are represented by u1′ to u6′.

[0055] 2. Data preprocessing:

[0056] The training samples are standardized and the input variables are PCA-dimensionality reduced.

[0057] (1) Standardization:

[0058] Regularization method is used to standardize the training samples.

[0059]

[0060] Where u ij′ represents the i-th input variable u i ′’s jth sample value, is the i-th input variable u i The sample mean of ′, σ i ′ is the i-th input variable u i ′’s sample standard deviation, u ij for u ij ′ is the standardized value, i=1,2,…,6; j=1,2,…,N; N is the number of samples.

[0061] (2) PCA dimensionality reduction of input variables:

[0062] Suppose the matrix formed by the data after standardization of U' is U∈R N×6 Follow these steps:

[0063] Step 1: Calculate the covariance matrix

[0064] Step 2: Find the eigenvalues ​​and corresponding eigenvectors of the covariance matrix C, and arrange the eigenvalues ​​from large to small. Let the arranged eigenvalues ​​and their corresponding eigenvectors be θ i and p i , i=1,2,…,6.

[0065] Step 3: Calculation s is the number of new input variables obtained after the input variables are reduced in dimension. The specific value needs to be solved according to the inequality on the left.

[0066] Step 4: Arrange the eigenvectors by column from left to right according to the corresponding eigenvalues ​​to form the characteristic matrix P0, and take the first s columns of P0 to form the matrix P.

[0067] Step 5: Calculate X = UP. X is the new input variable matrix after data preprocessing.

[0068] 3. Soft sensor modeling based on XGBoost method:

[0069] Follow these steps:

[0070] Step 1: Set k = 1 and go to Step 2;

[0071] Step 2: Let t k =1, go to Step 3:

[0072] Step 3: Initialize and construct the objective function.

[0073] Let the tth value of the kth output variable be k The objective function formula of a decision tree is as follows:

[0074]

[0075] Among them, the loss function of the i-th (i=1,2,...,N) sample is y ki represents the i-th sample value of the k-th output variable; represents the first t of the k-th output k The predicted value of the i-th sample by all decision trees together; Ω ka (f ka ) represents the model complexity of the a-th tree of the k-th output. Go to Step 4;

[0076] Step 4: Objective function transformation.

[0077] Substituting formula (5) into formula (4), we can get

[0078]

[0079] Among them, x i represents the i-th row of X, And is a constant. Taylor expansion of formula (6) yields

[0080]

[0081] in They are all constants. After simplification and removing the constant terms, the objective function is Equation (8), and we go to Step 5.

[0082]

[0083] Step 5: Define a decision tree and introduce the tree into the objective function.

[0084] The expression of a decision tree is defined as follows: Bundle Defined as in Represents the result of a tree, which is to input x i ∈R s Mapped to a leaf node (assuming the tree has leaf nodes), Is the length of A one-dimensional vector representing the weight of the leaf node (i.e., the predicted value of the decision tree).

[0085] The following formula is used to measure the complexity of a tree:

[0086]

[0087] in, and Both are hyperparameters that control the degree of penalty, and their value range is [0,1]. for No. components. Therefore, the objective function is transformed into:

[0088]

[0089] All samples x belonging to the jth leaf node i , is classified into a leaf node sample set, mathematically expressed as: Calculate the loss of all leaf node samples as follows make

[0090]

[0091] Then the objective function becomes To minimize the objective function, let right The partial derivative of is zero, so we have

[0092] Will Substituting into the objective function, we get

[0093] Go to Step 6;

[0094] Step 6: Build the optimal tree.

[0095] Determination of candidate split points: For each new input variable, the range of values ​​is divided into three or four quartiles according to the number of samples. Each quartile is a candidate split point.

[0096] Use greedy algorithm to split tree nodes:

[0097] Starting from the depth of the tree at 0, a round of splitting is attempted for each leaf node in the tree, that is, considering all new input variables, splitting is performed at each candidate splitting point of each new input variable; after each split, the original leaf node is further split into two left and right child leaf nodes, and the sample set in the original leaf node will be distributed to the left and right leaf nodes according to the judgment rule of the node (that is, compared with the value at the candidate splitting point); after a new node is split, it is tested whether this split will bring gain to the loss function. The gain is defined as follows:

[0098]

[0099] The L subscript is the optimal value of the objective function when the value is divided into the left subtree, and the R subscript is the optimal value of the objective function when the value is divided into the right subtree. If Gain < 0, stop the split; if the number of samples of any leaf node is lower than the threshold Nu min , stop this split, in which Nu min The value range is [1, 10]; if the depth of the tree reaches the maximum value dep max , stop this split, where dep max The value range is a positive integer in the interval [3, 9]. Otherwise, after all candidate split points have completed the split, the candidate split point with Gain>0 and the largest value is selected as the best split point for this round of split. Then, at the next leaf node, all candidate split points of all new input variables are considered, and the above process is repeated to perform the next round of splitting on the leaf node until the conditions for stopping splitting are met. The final number of leaf nodes is Go to Step 7;

[0100] Step 7: Determine t k Whether the maximum value t is reached kmax , t kmax The value range is a positive integer in the interval [1, 10]. If it is reached, go to Step 8, otherwise let t k =t k +1, go to Step 2;

[0101] Step 8: Determine whether k reaches 2. If so, the soft measurement model is established and go to Step 9. Otherwise, set k = k + 1 and go to Step 1.

[0102] Step 9:

[0103]

[0104] When the new input data u I When ′ arrives, the soft measurement result of the kth (k=1,2) output variable is obtained using formula (13) (i.e., the mass fraction distributed in the range of +35 mesh and -35 mesh to +325 mesh). The mass fraction distributed in the range of -325 mesh is obtained using It can be obtained. Among them, u I Represents the newly arrived input data u I ′The result after normalization, f ka (u I P) represents the output of the a-th decision tree of the k-th output variable, t represents the kth output variable k The sum of the outputs of the decision trees is the soft measurement result of the kth output variable of the general powder particle size in the high-temperature electrical grade magnesium oxide production process obtained based on the input data.

[0105] 4. Calibration of soft sensor model:

[0106] The soft-sensor model for general dust particle size is calibrated using a method that combines bias compensation and real-time learning. The manually detected values ​​of general dust particle size are compared with the soft-sensor results. If the absolute value of the error is less than a predefined threshold of 1, the original soft-sensor model remains unchanged and continues to be used. If the absolute value of the error is greater than threshold 1 and less than threshold 2, a bias compensation method is used for correction, where the actual error is added to the soft-sensor result as the final soft-sensor result. If the absolute value of the error is greater than threshold 2, the Z sample data with the highest similarity to the newly arrived input data are selected, the newly arrived input data is preprocessed, and the soft-sensor model is rebuilt using the XGBoost method. The newly arrived input and output data are then added to the training samples.

[0107] Among them, the threshold 1 range of the mass score distributed in the range of +35 mesh is between [0.5% and 1.5%], the threshold 2 range is between (1.5% and 2.5%], the threshold 1 range of the mass score distributed in the range of -35 mesh to +325 mesh is between [3% and 6%], the threshold 2 range is between (6% and 10%], and the similarity is calculated using formula (14). The value range of Z is between 100 and 200.

[0108]

[0109] in

[0110] It should be noted that the present invention is not limited to the embodiments described above. The above-described specific embodiments are merely illustrative and not restrictive. Without departing from the spirit of the present invention and the scope of protection of the claims, those skilled in the art may make various specific variations based on the teachings of the present invention, all of which fall within the scope of protection of the present invention.

Claims

1. A soft measurement method for general powder particle size in the production process of high-temperature electrical grade magnesium oxide powder, characterized in that: The following steps are performed to obtain an ideal soft-sensing model and perform real-time correction to perform soft-sensing prediction of unknown general powder particle size in the production process of unknown high-temperature electrical-grade magnesium oxide powder. The method includes the following steps: Establish a sample set: select auxiliary variables in the fused magnesium oxide crushing and screening process, measure the general powder particle size as a label, and establish a sample set with {auxiliary variables as input and fitted labels as output}; Data preprocessing: standardize the sample data, use the PCA method to reduce the dimension, and eliminate the redundant relationship between auxiliary variables; Establish a soft sensor model based on the XGBoost method: Use the XGBoost modeling method to establish an initialization objective function for the k-th output variable that takes into account the loss function and model complexity, and use the second-order Taylor expansion to transform the objective function; define a decision tree, and introduce the tree structure into the objective function, use the greedy algorithm to build the optimal tree, and repeat the above process t k times, and finally generates t k A decision tree is generated to obtain an ideal soft measurement model; the ideal model uses the outputs of these decision trees added together as the prediction output of the soft measurement results of the general powder particle size in the production process of high-temperature electrical grade magnesium oxide powder; Soft measurement model correction: The ideal soft measurement model of general powder particle size is corrected by combining deviation compensation correction with real-time learning.

2. The soft measurement method for general powder particle size in a high-temperature electrical grade magnesium oxide powder production process according to claim 1, characterized in that: The auxiliary variables include original particle size, hardness, feed amount, crusher current, return amount, and air classifier speed; the fitting labels are manual detection values ​​under two particle size distribution ranges.

3. The soft measurement method for general powder particle size in a high-temperature electrical grade magnesium oxide powder production process according to claim 1, characterized in that: The standardization is performed using the regularization method: Where u ij ′ represents the jth sample value of the i-th input variable, is the sample mean of the i-th input variable, σ i ′ is the sample standard deviation of the i-th input variable, u ij for u ij ′ is the standardized value, i=1,2,…,6; j=1,2,…,N; N is the number of samples.

4. The soft measurement method for general powder particle size in a high-temperature electrical grade magnesium oxide powder production process according to claim 1, characterized in that: The dimensionality reduction process using the PCA method is as follows: Calculate the new input variable matrix X=UP after data preprocessing; where U∈R N×6 It is the matrix formed by the input data after standardization, and the matrix P is the first s columns of the feature matrix P0.

5. The soft measurement method for general powder particle size in the production process of high temperature electrical grade magnesium oxide powder according to claim 1, characterized in that The tth value of the kth output variable k The formula for initializing the objective function of a decision tree is as follows: Among them, the loss function of the i-th (i=1,2,...,N) sample is y ki represents the i-th sample value of the k-th output variable; represents the first t of the k-th output k The predicted value of the i-th sample by all decision trees together; Ω ka (f ka ) represents the model complexity of the a-th tree of the k-th output.

6. The soft measurement method for general powder particle size in a high-temperature electrical grade magnesium oxide powder production process according to claim 1, characterized in that: The initial objective function is transformed and simplified by Taylor expansion to remove the constant term to obtain the objective function:

7. The soft measurement method for general powder particle size in a high-temperature electrical grade magnesium oxide powder production process according to claim 1, characterized in that: Define a decision tree Bundle Defined as in Represents the result of a tree, assuming that the tree has leaf nodes, used to convert the input x i R s Mapped to a leaf node, Is the length of A one-dimensional vector representing the weight of the leaf node, that is, the predicted value of the decision tree; The following formula is used to measure the complexity of a tree: in, and Both are hyperparameters that control the degree of penalty, and their values ​​range from [0, 1]. The objective function shown in formula (8) is transformed into: All samples x belonging to the jth leaf node i , is classified into a leaf node sample set, mathematically expressed as: Calculate the loss of all leaf node samples as make The objective function becomes To minimize the objective function, let right The partial derivative of is zero, so Will Substitute into the objective function and get the objective function 8. The soft measurement method for general powder particle size in a high-temperature electrical grade magnesium oxide powder production process according to claim 1, characterized in that: Using the greedy algorithm to construct the optimal tree includes: Determination of candidate split points: For each new input variable, the range of values ​​is divided into three or four quartiles according to the number of samples. Each quartile is a candidate split point. A greedy algorithm is used to split the tree nodes: considering all new input variables, split at each candidate split point of each new input variable; after each split, the original leaf node is further split into two left and right child leaf nodes, and the sample set in the original leaf node will be dispersed to the left and right leaf nodes based on the comparison between the node and the value at the candidate split point; after a new node is split, it is tested whether this split will bring gain to the loss function. The gain is defined as follows: The L subscript is the optimal value of the objective function when the value is divided into the left subtree, and the R subscript is the optimal value of the objective function when the value is divided into the right subtree; If Gain<0, stop the split; if the number of samples of any leaf node is lower than the threshold Nu min , stop this split, in which Nu min The value range is [1, 10]; if the depth of the tree reaches the maximum value dep max , stop this split, where dep max The value range is a positive integer in the interval [3, 9]. Otherwise, after all candidate split points have completed the split, the candidate split point with Gain>0 and the largest is selected as the best split point for this round of splitting. Then, at the next leaf node, all candidate split points of all new input variables are considered, and the above process is repeated to perform the next round of splitting on the leaf node until the condition for stopping splitting is met. The final number of leaf nodes is obtained as Substitute into the objective function.

9. The soft measurement method for general powder particle size in a high-temperature electrical grade magnesium oxide powder production process according to claim 1, characterized in that: The form of the soft measurement model based on the XGBoost method is: Among them, u I Represents the input data u I The results after standardization, t represents the kth output variable k The sum of the outputs of the decision trees is the soft measurement result of the kth output variable of the general powder particle size in the high-temperature electrical grade magnesium oxide production process obtained based on the input data.

10. The soft measurement method for general powder particle size in the production process of high-temperature electrical grade magnesium oxide powder according to claim 1, characterized in that: The soft measurement model correction includes: a) comparing the manual detection value of the general powder particle size with the soft measurement result, and if the absolute value of the error is less than a predefined first threshold, keeping the original soft measurement model unchanged and continuing to use the original soft measurement model; b) if the absolute value of the error is greater than the first threshold and less than a second threshold, using a deviation compensation method to perform correction, adding the actual error and the soft measurement result as the final soft measurement result; c) if the absolute value of the error is greater than the second threshold, selecting Z sample data with the highest similarity to the new input data, preprocessing the new input data, re-establishing the soft measurement model using the XGBoost method, and adding the new input and output data to the training samples.

Images