An intelligent control method for the crushing and screening process of high-temperature electrical-grade magnesium oxide

Abstract

The present invention discloses an intelligent control method for a high-temperature electrical-grade magnesium oxide crushing and screening process, comprising the following steps: analyzing the process flow, equipment and operation of the high-temperature electrical-grade magnesium oxide crushing and screening process, determining crushing and separation as links affecting particle size, and selecting a crusher speed, a classifier speed and a feed flow rate as operating variables; in an offline production process, taking the particle size distribution range of fused magnesium oxide powder as a target, and using a hybrid intelligent modeling method to calculate preset values ​​of the operating variables based on the magnesium oxide content, hardness, particle size and type of the fused magnesium oxide raw material; in an online production process, making real-time adjustments based on operation feedback and machine load conditions; and finally obtaining set values ​​of the operating variables for control.

Description

Technical Field

[0001] The present invention relates to the technical field of control of a high-temperature electrical-grade magnesium oxide production process, in particular to an intelligent control method for a high-temperature electrical-grade magnesium oxide crushing and screening process. Background Art

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

[0003] The crushing and screening process of high temperature electrical grade magnesium oxide is as follows: Figure 1 As shown. A feeding device feeds the lumpy fused magnesium oxide into a crusher for crushing. The crushed fused magnesium oxide powder is then lifted to a higher level by a bucket elevator and sent to a linear vibrating screen. The linear vibrating screen then classifies the fused magnesium oxide powder according to particle size. Coarser particles are returned to the crusher for further crushing, while finer particles are collected by bags. Particles of suitable size then enter an air classifier for further screening. The air classifier further separates the fused magnesium oxide powder of suitable particle size. The coarser particles, considered as fused magnesium oxide powder of suitable particle size, are lifted to a higher level by a bucket elevator, where they undergo magnetic separation for iron removal. 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, even finer particles, are collected by an electrostatic precipitator.

[0004] The indicator that reflects the quality of the crushing and screening process of high-temperature electrical grade magnesium oxide is the particle size distribution of the fused magnesium oxide powder collected after magnetic separation and iron removal (the mass fraction distributed in the range of +35 mesh, -35 mesh to +325 mesh, -325 mesh). It is tested offline by the inspectors using a standard vibrating screen machine.

[0005] The crushing and screening process for high-temperature electrical-grade magnesium oxide (MgO) involves complex mechanisms, and the raw material types and properties vary widely. In actual production, the particle size distribution of fused MgO powder is controlled by the operator based on subjective experience. This behavior is closely related to factors such as operator experience and responsibility, and is subject to a certain degree of arbitrariness. This leads to large fluctuations in product quality and poor consistency. Therefore, how to timely and accurately determine the appropriate setpoints for operating variables based on actual production conditions to reduce product quality fluctuations has become a pressing issue. Summary of the Invention

[0006] In view of the complex mechanism of the crushing and screening process of high-temperature electrical-grade magnesium oxide, the variability of raw material types and properties, the certain arbitrariness of operation, and the large fluctuations in product quality, the technical problem to be solved by the present invention is to provide an intelligent control method for the crushing and screening process of high-temperature electrical-grade magnesium oxide, which can overcome the changes in raw materials and automatically determine the set values ​​of operating variables according to actual production conditions.

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

[0008] An intelligent control method for the crushing and screening process of high-temperature electrical grade magnesium oxide comprises the following steps:

[0009] Analyze the process flow, equipment, and operation of high-temperature electrical-grade magnesium oxide crushing and screening, identify crushing and separation as the links that affect particle size, and select crusher speed, classifier speed, and feed flow rate as operating variables;

[0010] In the offline stage of the production process, the particle size distribution range of the fused magnesia powder is used as the target. Based on the magnesia content, hardness, particle size and type of the fused magnesia raw material, a hybrid intelligent modeling method is used to calculate the preset values ​​of the operating variables. In the online stage of the production process, real-time adjustments are made based on operational feedback and machine load conditions. Finally, the set values ​​of the operating variables are obtained for control.

[0011] The offline steps of the production process include:

[0012] Construct input and output sample sets based on historical data of production processes;

[0013] Consider the degree of similarity between the new situation and existing operational situations;

[0014] Offline construction of hybrid intelligent models to predict pre-set values ​​of operating variables in new situations;

[0015] Set the preset value Y of the manipulated variable new The data is sent to the basic control system for execution. If the particle size distribution cannot be satisfied, the preset value is corrected empirically.

[0016] Delete the historical data with the greatest similarity to the new question and update the stored sample set.

[0017] Constructing the input and output sample set based on the historical data of the production process includes: defining the input data matrix The output data is stored in the output data matrix Among them, X i =[x i1 ,x i2 ,...,x i6 ,x i7 ](i=1,2,...,U) represents the input variable of the i-th sample data, xi1 ~x i7 Y represents the magnesium oxide content, hardness and particle size of the fused magnesium oxide raw material, the mass fraction of the desired range of +35 mesh, -35 mesh to +325 mesh, and -325 mesh, and the type of the fused magnesium oxide raw material; i =[y i1 ,y i2 ,y i3 ](i=1,2,...,U) represents the output variable of the i-th sample data, y i1 ~y i3 They represent the preset values ​​of crusher speed, classifier speed and feed flow respectively.

[0018] During production operations, the similarity between the new situation and the existing operation situation is considered as the similarity between the new problem input and the sample input;

[0019]

[0020] Among them, X new =[x new1 ,x new2 ,...,x new6 ,x new7 ] indicates new question input, x newj Represents the jth (j=1, 2, ..., 7) component of the new problem input.

[0021] Consider the similarity between the new situation and the existing operation situation; build a hybrid intelligent model offline, specifically:

[0022] If the similarity is high and the number of similar samples is large, the extreme learning machine (ELM) method is used to establish a pre-set model of the operating variables;

[0023] If the similarity is high and the number of similar samples is small, the SMOTE algorithm is used to expand the similar samples, and then the Bagging LSSVM method is used to build a pre-set model of the operating variables;

[0024] If the similarity is low, ask domain experts or excellent operators to directly provide the preset values ​​of the operating variables in the current situation.

[0025] If the similarity D(X new ,X i ) is greater than or equal to the similarity threshold δ s The number of samples N is greater than or equal to the sample number threshold δ sn , the extreme learning machine (ELM) method is used to establish a prediction model for the preset values ​​of the operating variables, as follows:

[0026]

[0027] Among them, Y new Enter X for new questions new The preset values ​​of the operating variables when a i 、b i The input weight vector and bias value of the hidden layer node of the ELM method are randomly generated.

[0028] If the similarity D(X new ,X i ) is greater than or equal to δ s The number of samples N is less than δ sn , the SMOTE algorithm is used to expand similar samples, and then the Bagging LSSVM method is used to establish a pre-set model of the operating variables. The steps are as follows:

[0029] ① Set the upsampling ratio to AA, and use the SMOTE method to perform linear interpolation to obtain new samples (NSI k ,NSO k );

[0030] (NSI k ,NSO k )=(X i ,Y i )+rand i,z (0,1)×((X z -X i ),(Y z -Y i )) (5)

[0031] Where k = 1, 2, ..., AA, rand i,z (0,1) means using existing samples (X i ,Y i ) and (X z ,Y z )When generating a new sample (NSI k ,NSO k ) Random numbers generated in the interval [0,1];

[0032] The final expanded B = A × AA new samples are put together with the selected samples to form an extended sample set (EX j ,EY j ), where j = 1, 2, ..., (A + B);

[0033] ②Use the Bagging LSSVM method to establish a prediction model for the preset values ​​of the operating variables:

[0034] a. In the interval [1, (A+B)], the Bagging method is used according to the sampling probability Randomly extract s samples from the input and output sample sets, and repeat Z times to obtain Z pre-set sample subsets: {C1, V1}, {C2, V2}, ..., {C Z ,V Z};

[0035] Among them, the new sample subset consisting of s samples extracted from the input and output sample sets for the i-th (i=1,2,...,Z) time is set as is the input data of the zth sample in the i-th subset, is the output data of the zth sample in the i-th subset;

[0036] b. Establish the LSSVM model f of the jth output based on the i-th subset ij (X new );

[0037] c. Use the average method to synthesize the results of the Z LSSVM sub-models to obtain the preset value of the j-th operating variable:

[0038]

[0039] Therefore, the new preset value of the manipulated variable is Y new =[y new1 ,y new2 ,y new3 ].

[0040] In the online production process, based on the difference between the target range and the detected value of the particle size distribution of the fused magnesium oxide powder, a feedback compensation strategy is used to calculate the feedback compensation value ΔY(t) of the operating variable. The crusher load condition is judged based on the crusher return flow rate and bucket elevator current, and the adjustment value of the operating variable is calculated using an intelligent adjustment strategy.

[0041] ΔY(t)=[Δy1(t),Δy2(t),Δy3(t)]

[0042]

[0043] Finally, the manipulated variable is set to

[0044] The present invention has the following advantages and beneficial effects:

[0045] The method of the present invention does not require analysis of the complex mechanism of the crushing and screening process of high-temperature electrical-grade magnesium oxide, fully utilizes historical data with good operating effects and the operating experience and knowledge of on-site experts, and can automatically provide set values ​​of operating variables in a timely manner according to changes in raw materials and products, thereby achieving stable control of the particle size distribution of the fused magnesium oxide powder, improving the consistency of the particle size distribution, making the crushing and screening process operation more regular, and eliminating the arbitrariness and blindness of operators as much as possible. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is the principle diagram of the crushing and screening process of high-temperature electrical grade magnesium oxide;

[0047] Figure 2 Flowchart for calculation of operating variables. DETAILED DESCRIPTION

[0048] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, the specific implementation methods of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth many specific details 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 may make similar modifications without departing from the scope of the invention. Therefore, the present invention is not limited to the specific implementation methods disclosed below.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the art to which this invention belongs. The terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The present invention will be further described in detail below with reference to the accompanying drawings and examples.

[0050] The technical problem to be solved by the present invention is to provide an intelligent control method for the crushing and screening process of high-temperature electrical grade magnesium oxide, which can overcome the changes in raw materials and automatically determine the set values ​​of operating variables according to actual production conditions. Figure 2 The figure shows the calculation flow chart of the operating variables, and the specific steps are as follows.

[0051] 1. Selection of operating variables:

[0052] An analysis of the process flow, equipment and operation of the crushing and screening process of high-temperature electrical grade magnesium oxide revealed that the links that have a greater impact on particle size are crushing and separation. The variables that affect the quality and efficiency of the crushing link are feed flow and crusher speed, and the main factor affecting the quality and efficiency of the separation link is the classifier speed. Therefore, crusher speed, classifier speed and feed flow were selected as operating variables.

[0053] 2. Calculation of the set value of the operating variable:

[0054] (1) Presetting of operating variables

[0055] During production operations, the preset values ​​of the operational variables are predicted based on the similarity between the new situation (i.e., new problem) and the existing operational situation. If the similarity is high and the number of similar samples is large, the ELM (Extreme Learning Machine) method is used to establish the preset model of the operational variables. If the similarity is high and the number of similar samples is small, the SMOTE (Synthetic Minority Oversampling) algorithm is used to expand the similar samples, and then the Bagging LSSVM (Bagging Least Squares Support Vector Machine) method is used to establish the preset model of the operational variables. If the similarity is low, domain experts or excellent operators are asked to directly provide the preset values ​​of the operational variables for the current situation.

[0056] (1) Construct input and output sample sets

[0057] An analysis of the crushing and screening process of high-temperature electrical-grade magnesium oxide and the operator's performance revealed that operators determine the preset values ​​for crusher speed, classifier speed, and feed rate based on the target particle size, the magnesium oxide content, hardness, and particle size of the fused magnesium oxide raw material, and the target particle size distribution (mass fraction within the ranges of +35 mesh, -35 mesh to +325 mesh, and -325 mesh). Therefore, the pre-set algorithm selects the following input variables: magnesium oxide content, hardness, and particle size of the fused magnesium oxide raw material; the target mass fraction within the ranges of +35 mesh, -35 mesh to +325 mesh, and -325 mesh; and the type of fused magnesium oxide raw material; and the output variables are the preset values ​​for crusher speed, classifier speed, and feed rate.

[0058] Select the input and output variable data with good control effects in the past to form the input and output modeling sample set, and store the input variable data in the input modeling data matrix, and store the output variable data in the output modeling data matrix. Suppose a total of U groups of input and output data are found, and the input data among them are stored in the input data matrix (symbol T Indicates matrix transposition, the same below), the output data is stored in the output data matrix In, X i =[x i1 ,x i2 ,...,x i6 ,x i7 ](i=1,2,...,U) represents the input variable of the i-th sample data, x i1 ~x i7 Y represents the magnesium oxide content, hardness and particle size of the fused magnesium oxide raw material, the mass fraction of the desired range of +35 mesh, -35 mesh to +325 mesh, and -325 mesh, and the type of the fused magnesium oxide raw material; i =[y i1,y i2 ,y i3 ](i=1,2,...,U) represents the output variable of the i-th sample data, y i1 ~y i3 They represent the preset values ​​of crusher speed, classifier speed and feed flow respectively.

[0059] (2) Determine the similarity between the new question input and the sample input

[0060] Query the input and output sample sets to see if there is a sample with the same seventh component as the new question input. If so, retrieve the sample with the same seventh component as the new question input and use formula (1) to determine the similarity between the new question input and the sample input; if not, the similarity between the new question input and the sample input is 0.

[0061]

[0062] Among them, X new =[x new1 ,x new2 ,...,x new6 ,x new7 ] indicates new question input, x newj Represents the jth (j=1, 2, ..., 7) component of the new problem input.

[0063] (3) Establish a prediction model for preset values ​​of operating variables based on the similarity results

[0064] A. Similarity is greater than or equal to the similarity threshold δ s , δ s The value is 0.65, and the similarity is greater than or equal to δ s The number of samples is greater than or equal to the sample size threshold δ sn , δ sn Take it as 80.

[0065] Assume that the similarity is greater than or equal to δ s The number of samples is N, and the sample input matrix is The sample output matrix is X Ii =[x Ii1 ,x Ii2 ,...,x Ii6 ](i=1,2,...,N),Y Oi =[y Oi1 ,y Oi2 ,y Oi3 ]i(1,2,...,N). The extreme learning machine (ELM) method is used to establish a prediction model for the preset values ​​of the operating variables. The steps are as follows:

[0066] ① Randomly generate ELM input weight vector a i and the bias value b of the hidden layer node i , i=1,2,…,L, L represents the number of hidden layer nodes, L=10;

[0067] ②Calculate the ELM hidden layer output matrix H;

[0068]

[0069] Among them, h i (x j )=G(a i ,b i ,X Ij ) represents the output of the i-th hidden layer node, which is the ELM nonlinear feature map, G(a i ,b i ,X Ij )=exp(-b i ||X Ij -a i ||), sample data (X Ij ,Y Oj )∈R 6 ×R 3 , j=1,2,…,N,a i =[a i1 ,a i2 ,…,a i6 ] T represents the input weight connecting the i-th hidden layer node of ELM, β i =[β i1 ,β i2 ,β i3 ] T represents the output weight of the i-th hidden layer node; b i Represents the bias value of the i-th hidden layer node.

[0070] ③ Use formula (3) to calculate the output weight matrix β:

[0071]

[0072] in, represents the Moore-Penrose generalized inverse of the hidden layer output matrix H, which is obtained using the singular value decomposition method.

[0073] ④Use formula (4) to find the new problem input X new The preset value Y of the manipulated variable at new , which represents the result after the new problem input is standardized.

[0074]

[0075] B. Similarity is greater than or equal to δ s , and the similarity is greater than or equal to δ s The number of samples is less than δ sn The steps are as follows:

[0076] ①Use the SMOTE method to perform linear interpolation to obtain new samples:

[0077] Assume that the number of selected samples is A, the upsampling ratio is AA, and for each sample

[0078] (X i ,Y i ), i=1,2,…,A, select the W samples with the highest similarity (X z ,Y z ), z=1,2,…,W, W is usually no more than 5, randomly select AA samples from the selected W samples, and use formula (5) to expand AA new samples (NSI k ,NSO k ):

[0079] (NSI k ,NSO k )=(X i ,Y i )+rand i,z (0,1)×((X z -X i ),(Y z -Y i )) (5)

[0080] Where k = 1, 2, ..., AA, rand i,z (0,10 means using existing samples (X i ,Y i ) and (X z ,Y z )When generating a new sample (NSI k ,NSO k ) is a random number generated in the interval [0,1].

[0081] After the expansion operation is performed on all A selected samples, B = A × AA new samples are expanded. The expanded new samples are put together with the selected samples to form an expanded sample set (EX j ,EY j ), where j = 1, 2, ..., (A + B).

[0082] ②Use the Bagging LSSVM method to establish a prediction model for the preset values ​​of the operating variables:

[0083] a. Use the Bagging method to create sample subsets.

[0084] Assign equal sampling probability to each positive integer in the interval [1, (A+B)] With this probability, randomly generate s = (A + B) positive integers k from the interval [1, (A + B)] i ∈]1,(A+B)](i=1,2,...,s), extract the kth i samples form a new sample subset, let in, is the input data of the i-th sample in the first subset, is the output data of the i-th sample in the first subset, then the first preset sample subset generated from the extended sample set is {C1, V1}; repeat the above steps Z times, and Z preset sample subsets are obtained: {C1, V1}, {C2, V2}, ..., {C Z ,V Z}, Z is taken as 50.

[0085] b. Establish an LSSVM sub-model for each sample subset.

[0086] The coefficients of the LSSVM model for the jth output built based on the i-th subset are given by equations (6) and (7):

[0087]

[0088] Solve α by formula (8) ij and d ij ,Right now:

[0089]

[0090] Among them, σ ij is the width of the radial basis function, and its value range is [0.1, 20]. s =[1,1,...,1] T , α ij =[α ij1 ,α ij2 ,...,α ijs ] T , α ij represents the weight parameter vector, V ij Indicates V i The j-th (j=1,2,3) column of (i=1,2,...,Z). Among them, r=1,2,...,s, v=1,2,...,s, i=1,2,...,Z, j=1,2,3. Is is the s-order identity matrix. ij is the regularization parameter, and its value range is in the interval [0.1,10]. ij Represents the bias parameter.

[0091] c. Sub-model weighted synthesis

[0092] The results of the Z LSSVM sub-models are synthesized using the average method, and the preset value of the j-th operating variable is:

[0093]

[0094] Therefore, the new preset value of the manipulated variable is Y new =[y new1 ,y new2 ,y new3 ]

[0095] C. Similarity is less than the threshold δ s .

[0096] Ask domain experts or experienced operators to directly provide solution Y new .

[0097] (4) Modification of preset values ​​of operating variables

[0098] Set the preset value Y of the manipulated variable new The results are sent to the basic control system for execution and evaluated. If the obtained particle size distribution meets production requirements, the result remains unchanged. If the obtained particle size distribution does not meet the requirements, the preset value needs to be corrected. This correction method uses a correction solution provided by field experts (manual correction) until the obtained particle size distribution meets the requirements.

[0099] (5) Storage of preset results of operation variables

[0100] If the maximum similarity between the new question and the samples in the input and output sample sets is greater than or equal to the similarity threshold SIM max , then delete the historical input and output data with the greatest similarity to the new problem from matrices X and Y, and then store the input and output data of the new problem in the input and output sample sets respectively. Otherwise, store the input and output data of the new problem directly in the input and output sample sets. SIM max Take 0.95.

[0101] (2) Feedback compensation

[0102] Feedback compensation can correct the preset values ​​of crusher speed, classifier speed and feed flow rate to compensate for the influence of unknown interference, so that the particle size distribution is within the target range, thereby better adapting to changes in working conditions.

[0103] Feedback compensation is implemented using an expert rule method based on rule reasoning, and its form is shown in the following rules:

[0104] 1) If the difference between the expected value and the measured value of the mass fraction of the particle size distribution in the range of +35 mesh is less than or equal to △a1%, the crusher speed is increased by △y1(t) = △A11r / min, the classifier speed is reduced by △y2(t) = △A21r / min, and the feed flow rate is reduced by △y3(t) = △A31t / h;

[0105] 2) If the difference between the expected value and the measured value of the mass fraction of the particle size distribution in the +35 mesh range is greater than △a1% and less than △a2%, the crusher speed is increased by △y1(t) = △A12r / min, the classifier speed is reduced by △y2(t) = △A22r / min, and the feed flow rate is reduced by △y3(t) = △A32t / h;

[0106] 3) If the difference between the expected mass fraction and the measured value of the particle size distribution in the range of -325 mesh is less than or equal to △b1%, the crusher speed is reduced by △y1(t) = △B11r / min, the classifier speed is increased by △y2(t) = △B21r / min, and the feed flow rate is increased by △y3(t) = △B31t / h;

[0107] 4) If the difference between the expected mass fraction and the measured value in the particle size distribution range of -325 mesh is greater than △b1% and less than △b2%, the crusher speed is reduced by △y1(t) = △B12r / min, the classifier speed is increased by △y2(t) = △B22r / min, and the feed flow rate is increased by △y3(t) = △B32t / h;

[0108] The parameter values ​​in the rule are as follows: △a1=-3; △A11=60; △A21=10; △A31=0.3; △a2=1; △A12=39; △A22=6; △A32=0.1; △b1=-3; △B11=45; △B21=8; △B31=0.5; △b2=1.5; △B12=27; △B22=5; △B32=0.1.

[0109] (3) Crusher load monitoring and adjustment

[0110] The hammer crusher (crusher for short) is one of the most critical pieces of equipment in general powder processing. It must be operated under the appropriate load during production to prevent failures and significant disruptions. This component monitors the return feed rate and crusher current, and uses expert rules to comprehensively assess the crusher load. If abnormalities are detected, adjustments to the feed rate and crusher speed are provided to restore the crusher to normal operating conditions, avoiding failures.

[0111] 1) If the crusher return material volume is greater than M1 and less than or equal to M2t / h, or the crusher current is greater than N1 and less than or equal to N2A, the feed flow adjustment amount

[0112] 2) If the crusher return material volume is greater than M2 and less than or equal to M3t / h, or the crusher current is greater than N2 and less than or equal to N3A, the feed flow adjustment amount

[0113] 3) If the crusher return material volume is greater than M3t / h, or the crusher current is greater than N3A, the feed flow adjustment amount

[0114] The parameter values ​​in the rule are as follows: M1=1.8; M2=2.3; M3=2.6; N1=54.2; N2=57.3; N3=58.5; △K1=-0.3; △K2=-0.5; △K3=-0.6.

[0115] Therefore, the final manipulated variable setting value is Where ΔY(t)=[Δy1(t),Δy2(t),Δy3(t)],

[0116] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural change made to the above embodiment based on the technical essence of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. An intelligent control method for the crushing and screening process of high-temperature electrical grade magnesium oxide, characterized in that: The following steps are involved: Analyze the process flow, equipment, and operation of high-temperature electrical-grade magnesium oxide crushing and screening, identify crushing and separation as the links that affect particle size, and select crusher speed, classifier speed, and feed flow rate as operating variables; In the offline stage of the production process, the particle size distribution range of the fused magnesia powder is used as the target. Based on the magnesia content, hardness, particle size and type of the fused magnesia raw material, a hybrid intelligent modeling method is used to calculate the preset values ​​of the operating variables; including: 1) Construct input and output sample sets based on historical data of production processes; 2) Consider the degree of similarity between the new situation and existing operational situations; 3) Offline construction of a hybrid intelligent model to predict the preset values ​​of the operating variables in the new situation; specifically: If the similarity is high and the number of similar samples reaches the set value, the extreme learning machine (ELM) method is used to establish a pre-set model of the operating variables; If the similarity is high and the number of similar samples does not reach the set value, the SMOTE algorithm is used to expand the similar samples, and then the Bagging LSSVM method is used to build a pre-set model of the operating variables; If the similarity is low, the expert gives the preset values ​​of the operational variables in the current situation; 4) Set the preset value Y of the manipulated variable new Output to the basic control system for execution. If the particle size distribution cannot be met, the preset value is empirically corrected. 5) Delete the historical data with the greatest similarity to the new question and update the stored sample set; In the online production process, real-time adjustments are made based on operational feedback and machine load conditions; ultimately, the set values ​​of the operating variables are obtained for control.

2. The intelligent control method for high-temperature electrical grade magnesium oxide crushing and screening process according to claim 1, characterized in that: Constructing the input and output sample set based on the historical data of the production process includes: defining the input data matrix The output data is stored in the output data matrix Among them, X i =[x i1 ,x i2 ,...,x i6 ,x i7 ](i=1,2,...,U) represents the input variable of the i-th sample data, x i1 ~x i7 Y represents the magnesium oxide content, hardness and particle size of the fused magnesium oxide raw material, the mass fraction of the desired range of +35 mesh, -35 mesh to +325 mesh, and -325 mesh, and the type of the fused magnesium oxide raw material; i =[y i1 ,y i2 ,y i3 ](i=1,2,...,U) represents the output variable of the i-th sample data, y i1 ~y i3 They represent the preset values ​​of crusher speed, classifier speed and feed flow respectively.

3. The intelligent control method for crushing and screening of high-temperature electrical grade magnesium oxide according to claim 1, characterized in that: During production operations, the similarity between the new situation and the existing operation situation is considered, and the similarity between the new problem input and the sample input is calculated; Among them, X new =[x new1 ,x new2 ,...,x new6 ,x new7 ] indicates new question input, x newj Represents the jth (j=1, 2, ..., 7) component of the new problem input.

4. The intelligent control method for the crushing and screening process of high-temperature electrical grade magnesium oxide according to claim 1, characterized in that: If the similarity D(X new ,X i ) is greater than or equal to the similarity threshold δ s The number of samples N is greater than or equal to the sample number threshold δ sn , the extreme learning machine (ELM) method is used to establish a prediction model for the preset values ​​of the operating variables, as follows: Among them, Y new Enter X for new questions new The preset values ​​of the operating variables at time , the weight matrix a i 、b i The input weight vector and bias value of the hidden layer node of the ELM method are randomly generated.

5. The intelligent control method for the crushing and screening process of high-temperature electrical grade magnesium oxide according to claim 1, characterized in that: If the similarity D(X new ,X i ) is greater than or equal to δ s The number of samples N is less than δ sn , the SMOTE algorithm is used to expand similar samples, and then the Bagging LSSVM method is used to establish a pre-set model of the operating variables. The steps are as follows: ① Set the upsampling ratio to AA, and use the SMOTE method to perform linear interpolation to obtain new samples (NSI k ,NSO k ); (NSI k ,NSO k )=(X i ,AND i )+rand i,z (0,1)×((X z -X i ),(AND z -AND i )) (5) Where k = 1, 2, ..., AA, rand i,z (0,1) means using existing samples (X i ,Y i ) and (X z ,Y z )When generating a new sample (NSI k ,NSO k ) Random numbers generated in the interval [0,1]; The final expanded B = A × AA new samples are put together with the selected samples to form an extended sample set (EX j ,EY j ), where j = 1, 2, ..., (A + B); ②Use the Bagging LSSVM method to establish a prediction model for the preset values ​​of the operating variables: a. In the interval [1, (A+B)], the Bagging method is used according to the sampling probability Randomly extract s samples from the input and output sample sets, and repeat Z times to obtain Z pre-set sample subsets: {C1, V1}, {C2, V2}, ..., {C Z ,V Z }; Among them, the new sample subset consisting of s samples extracted from the input and output sample sets for the i-th (i=1,2,...,Z) time is set as is the input data of the zth sample in the i-th subset, is the output data of the zth sample in the i-th subset; b. Establish the LSSVM model f of the jth output based on the i-th subset ij (X new ); c. Use the average method to synthesize the results of the Z LSSVM sub-models to obtain the preset value of the j-th operating variable: Therefore, the new preset value of the manipulated variable is Y new =[y new1 ,y new2 ,y new3 ].

6. The intelligent control method for the crushing and screening process of high-temperature electrical grade magnesium oxide according to claim 1, characterized in that: In the online production process, based on the difference between the target range and the detection value of the particle size distribution of the fused magnesium oxide powder, a feedback compensation strategy is adopted to calculate the feedback compensation value of the operating variable ΔY(t) = [Δy1(t), Δy2(t), Δy3(t)].

7. The intelligent control method for the crushing and screening process of high-temperature electrical grade magnesium oxide according to claim 1, characterized in that: In the online production process, the crusher load condition is judged based on the crusher return flow and bucket elevator current, and the adjustment value of the operating variable is calculated using an intelligent adjustment strategy.

8. The intelligent control method for high-temperature electrical grade magnesium oxide crushing and screening process according to claim 1, 6 or 7, characterized in that: Manipulated variable set value

Images