An advanced process control system and method for alumina slurry grinding

Abstract

The present invention relates to the technical field of advanced process control for alumina ore slurry milling, and specifically to an advanced process control system and method for alumina ore slurry milling. The present invention first classifies alumina ores that need to be slurried through an alumina ore classification unit to obtain a set of alumina ores with qualified sizes and a set of alumina ores with unqualified sizes; secondly, the present invention pulverizes the alumina ores in the set of unqualified alumina ores with qualified sizes through a crushing unit; thirdly, the present invention obtains the proportion of the mixing liquid according to the alumina ore parameters corresponding to the set of qualified alumina ores with qualified sizes through a mixing liquid control unit; then, the present invention obtains the slurry milling parameters through a slurry milling parameter acquisition unit, and slurries the set of qualified alumina ores with qualified sizes through the slurry milling unit; after the slurry milling time is reached, the present invention detects the generated alumina ore slurry through a slurry milling result detection unit. The present invention can improve the slurry milling effect of alumina ore through this system.

Description

Technical Field

[0001] The present invention relates to the technical field of advanced process control for alumina slurry mills, and in particular to an advanced process control system and method for alumina slurry mills. Background Art

[0002] Alumina ore is a key mineral used in alumina production. Its chemical composition primarily consists of alumina, silicates, and other impurities. Alumina slurry grinding is the process of mechanically grinding coarsely ground alumina ore to the desired particle size range, producing an alumina slurry suitable for dissolution and dilution.

[0003] Numerous control methods exist for alumina slurry milling. For example, the document ("Improvements to a Center-Drive Double-Bin Overflow Ball Mill," Nonferrous Metal Design, Li Xing, 2015-03-15) addresses production-related faults and improves ball mills, eliminating key production issues encountered by ball mills in alumina plants, mineral processing plants, and other industries. However, this document only improves the ball mill's hardware and fails to optimize the mill's grinding parameters based on the alumina ore, making it insufficiently intelligent. Another example is Chinese patent application number CN117138933A, which proposes an advanced process control system and method for alumina slurry milling. This method utilizes soft instrumentation technology to calculate the slurry specific gravity using a soft instrumentation algorithm with auxiliary variables. Combined with a model predictive control algorithm, the method adjusts the mill's input parameters in real time based on the liquid-to-solid ratio parameter and optimization target set by the quality controller. Furthermore, Chinese patent publication number CN111196715B proposes a method for preparing inert alumina chemical fillers from activated alumina balls, a type of industrial solid waste or hazardous waste. This method involves placing the raw materials separately in a ball mill and subjecting them to wet ultrafine grinding to obtain an aluminum sol. Zirconium sol is then added to the aluminum sol, followed by drying and other processes to produce the inert alumina chemical filler. This method fails to optimize the amount of conditioning agent added and the milling parameters during the milling process based on the characteristics of the alumina ore, resulting in poor milling results.

[0004] Therefore, an advanced process control system and method for alumina slurry grinding is proposed. Summary of the Invention

[0005] The object of the present invention is to provide an advanced process control system and method for alumina pulp grinding. First, the present invention classifies the alumina raw ore to be pulp-grinded through an alumina raw ore classification unit to obtain a set of alumina raw ore with qualified size and a set of alumina raw ore with unqualified size. Secondly, the present invention pulverizes the alumina raw ore in the set of alumina raw ore with unqualified size through a pulverizing unit. Thirdly, the present invention obtains the ratio of the blending liquid according to the alumina raw ore parameters corresponding to the set of alumina raw ore with qualified size through a blending liquid control unit. Then, the present invention obtains the pulp grinding parameters through a pulp grinding parameter obtaining unit and performs pulp grinding on the set of alumina raw ore with qualified size through a pulp grinding unit. After reaching the pulp grinding time, the present invention detects the generated alumina raw ore pulp through a pulp grinding result detection unit.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An advanced process control system for an alumina pulp mill, comprising:

[0008] An alumina raw ore classification unit: used to classify the alumina raw ore to be pulp-grinded to obtain a set of alumina raw ore with qualified size and a set of alumina raw ore with unqualified size.

[0009] Further, the classification of the alumina raw ore to be pulp-grinded includes:

[0010] Set the sieve holes of the vibrating screen according to the maximum feeding particle size of the ball mill, and the size of the sieve holes is not greater than the maximum feeding particle size of the ball mill; screen the alumina raw ore to be pulp-grinded through the vibrating screen; divide the alumina raw ore remaining on the vibrating screen into a set of alumina raw ore with unqualified size; divide the alumina raw ore sieved out through the vibrating screen into a set of alumina raw ore with qualified size.

[0011] A pulverizing unit: used to pulverize the alumina raw ore in the set of alumina raw ore with unqualified size.

[0012] Further, the pulverizing process includes:

[0013] Step 1: When the number of alumina raw ore contained in the set of alumina raw ore with unqualified size is less than the set value W, jump to step 5; otherwise, execute step 2;

[0014] Step 2: Select S alumina raw ore from the set of alumina raw ore with unqualified size for hardness detection, S < W, and obtain the surface hardness values of the S alumina raw ore;

[0015] Step 3: Obtain the pulverizing parameters according to the maximum value of the S surface hardness values and the set maximum feeding particle size of the ball mill;

[0016] Step 4: The crushing unit crushes according to the crushing parameters. When the crushing time is reached, the alumina ore that fails to be discharged through the discharge port is divided into a set of unqualified alumina ore, and the process goes to step 1.

[0017] Step 5: The crushing unit does not perform crushing.

[0018] Furthermore, the calculation process of the surface hardness value includes:

[0019] A plurality of detection points are evenly selected on each of the alumina ore; a hardness tester is used to detect the hardness value of each of the detection points; and the hardness values corresponding to the plurality of detection points are averaged to obtain the surface hardness value of each of the alumina ore.

[0020] A mixing liquid control unit is used to obtain the mixing liquid ratio according to the alumina ore parameters corresponding to the set of qualified alumina ores.

[0021] Furthermore, the step of obtaining the proportion of the blending solution according to the alumina ore parameters corresponding to the set of qualified alumina ores includes:

[0022] The alumina ore parameters include alumina ore size, alumina ore hardness, alumina ore density, alumina ore pH value, and alumina ore moisture. The process of obtaining the alumina ore size includes: randomly selecting N alumina ore samples from the set of alumina ores with qualified sizes; calculating the minimum circumscribed sphere diameters of the N alumina ores, averaging the N minimum circumscribed sphere diameters, and using the calculated result as the alumina ore size.

[0023] The process of obtaining the hardness of the alumina ore is as follows: calculating the hardness of N alumina ores, averaging the hardness of the N alumina ores, and using the calculated result as the hardness of the alumina ore;

[0024] The process of obtaining the density, pH value and humidity of the alumina ore can refer to the process of obtaining the hardness of the alumina ore;

[0025] The alumina ore parameters are input into the mixing liquid ratio model to obtain the ratio of the mixing liquid.

[0026] Furthermore, the mixing liquid ratio model is constructed based on a deep belief network, and the mixing liquid ratio model adaptively adjusts the learning rate according to the model loss during the training process; the training process of the mixing liquid ratio model includes:

[0027] Collecting a historical alumina ore blending solution ratio data set; the historical alumina ore blending solution ratio data set includes multiple historical alumina blending solution ratio data; the historical alumina blending solution ratio data includes alumina ore parameters and their corresponding standard blending solution ratios;

[0028] The historical alumina ore blending solution ratio dataset is divided into a training set and a test set; the learning rate in the blending solution ratio model is initialized; the alumina ore parameters in the training set are used as input, and the blending solution ratio prediction result is used as output; the blending loss function value of the blending solution ratio model is calculated, and the expression of the blending loss function is:

[0029]

[0030] Among them, ss t is the value of the blending loss function at the tth iteration; M is the number of historical alumina blending solution ratio data in the training set; N is the number of additive types contained in the blending solution; pb (i,j,t) It is expressed as the proportion of the jth additive in the predicted result of the mixing solution ratio corresponding to the i-th historical alumina mixing solution ratio data output by the model after the t-th iteration; and bzpb (i,j) It is expressed as the ratio of the jth additive in the standard mixing solution ratio corresponding to the i-th historical alumina mixing solution ratio data; and exp() represents an exponential function with a natural constant as the base;

[0031] When the value of the mixing loss function converges, the training ends and the learning rate in the mixing liquid ratio model is output; otherwise, the learning rate is adjusted according to the mixing loss function and the training continues until the value of the mixing loss function converges; the formula for adjusting the learning rate according to the mixing loss function is:

[0032]

[0033] Among them, xxl t It represents the learning rate after the tth iteration; β1 represents the set increment value, β1>0; β2 represents the set decrement value, β2>0; ss t Expressed as the value of the deployment loss function at the t-th iteration; It represents the set loss reduction threshold; the test set is used to test the mixing liquid ratio model.

[0034] Slurry grinding parameter acquisition unit: used to obtain slurry grinding parameters according to alumina characteristic parameters.

[0035] Furthermore, obtaining the slurry grinding parameters according to the alumina characteristic parameters includes:

[0036] The alumina characteristic parameters include the alumina ore parameters, the target particle diameter of the alumina ore, and the alumina ore addition rate; the slurry milling parameters include the ball mill parameters and the mixing liquid addition rate; the ball mill parameters include the ball mill speed, motor power, and slurry milling time; the mixing liquid addition rate is obtained based on the alumina ore addition rate; the process of obtaining the ball mill parameters includes:

[0037] Collecting a historical alumina slurry grinding data set, wherein the historical alumina slurry grinding data set includes Q historical alumina characteristic parameters and corresponding historical ball mill parameter standard values;

[0038] The historical alumina characteristic parameters include historical alumina ore parameters, historical alumina ore target particle diameter and historical alumina ore addition rate;

[0039] Clustering the Q historical alumina characteristic parameters to obtain P clusters; using the data in the P clusters to train a ball mill slurry grinding parameter prediction model to obtain P trained ball mill slurry grinding parameter prediction models;

[0040] Calculating the distance between the alumina characteristic parameters and the centers of P clusters, and selecting the cluster with the closest distance as the predicted cluster; inputting the alumina characteristic parameters into the trained ball mill slurry grinding parameter prediction model corresponding to the predicted cluster to obtain the ball mill parameters corresponding to the alumina characteristic parameters;

[0041] The training process of the ball mill grinding parameter prediction model includes: randomly initializing the weights and biases of the ball mill grinding parameter prediction model;

[0042] The historical alumina characteristic parameters in the cluster are input into the ball mill slurry grinding parameter prediction model to obtain the historical ball mill parameter prediction value; the ball mill loss function value in the ball mill slurry grinding parameter prediction model is calculated, and the calculation formula is:

[0043]

[0044] Among them, csyc t It represents the ball mill loss function value after the tth iteration in the ball mill grinding parameter prediction model training process; m represents the number of data items contained in the cluster; qmcs (e,d,t) bzcs represents the dth historical ball mill parameter prediction value corresponding to the eth historical alumina characteristic parameter output by the ball mill slurry grinding parameter prediction model at the tth iteration; (e,d)It represents the dth historical ball mill parameter standard value corresponding to the eth historical alumina characteristic parameter; d=1 corresponds to the ball mill speed; d=2 corresponds to the ball mill motor power; d=3 corresponds to the ball mill slurry grinding time;

[0045] When the ball mill loss function value converges, the training ends and the weight and bias of the ball mill slurry grinding parameter prediction model are output; otherwise, the training continues until the ball mill loss function value converges.

[0046] The pulp grinding unit is used to pulp grind the alumina ore of qualified size according to the pulp grinding parameters.

[0047] Pulp grinding result detection unit: used to detect the generated alumina ore pulp after the pulp grinding time is reached.

[0048] Furthermore, the pulping result detection unit includes:

[0049] Randomly select M liters of the alumina slurry as a slurry test sample;

[0050] Obtaining a first weight of solid particles in the raw ore slurry having a particle diameter greater than a set target particle diameter of the alumina ore; and simultaneously obtaining a second weight of all solid particles in the raw ore slurry;

[0051] Calculate the ratio of the first weight to the second weight; when the ratio is greater than the set ratio, output a pulp grinding failure warning; otherwise, obtain the alumina ore slurry and corresponding parameters, and retrain the mixing liquid ratio model and the ball mill pulp grinding parameter prediction model.

[0052] Furthermore, the retraining of the mixing liquid ratio model and the ball mill mill parameter prediction model includes: obtaining corresponding parameters in the milling process; the corresponding parameters include alumina ore parameters, alumina characteristic parameters, mixing liquid ratio and milling parameters;

[0053] The ratio of the prepared liquid is used as the standard prepared liquid ratio corresponding to the parameters of the alumina ore;

[0054] Adding the alumina ore parameters and the corresponding standard blending solution ratio to a historical alumina ore blending solution ratio dataset, and retraining the blending solution ratio model;

[0055] The alumina characteristic parameters and the ball mill parameters in the slurry grinding parameters are added to a historical alumina slurry grinding data set, and the ball mill slurry grinding parameter prediction model is trained again.

[0056] An advanced process control method for alumina slurry grinding, characterized by comprising:

[0057] Classify the alumina raw ore to be pulp-milled to obtain a set of alumina raw ore with qualified size and a set of alumina raw ore with unqualified size; crush the alumina raw ore in the set of alumina raw ore with unqualified size. The specific process includes:

[0058] Step 1: When the number of alumina raw ore in the set of alumina raw ore with unqualified size is less than the set value W, jump to Step 5; otherwise, execute Step 2;

[0059] Step 2: Select S alumina raw ore from the set of alumina raw ore with unqualified size for hardness detection, S < W, and obtain the surface hardness values of the S alumina raw ore;

[0060] Step 3: Obtain the crushing parameters according to the maximum value of the S surface hardness values and the maximum feed particle size of the set ball mill;

[0061] Step 4: The crushing unit crushes according to the crushing parameters. When the crushing time is reached, the alumina raw ore that fails to be discharged through the discharge port is classified into the set of alumina raw ore with unqualified size, and jump to Step 1;

[0062] Step 5: The crushing unit does not perform crushing;

[0063] Obtain the ratio of the blending liquid according to the alumina raw ore parameters corresponding to the set of alumina raw ore with qualified size;

[0064] Obtain the pulp-milling parameters according to the alumina characteristic parameters; the alumina characteristic parameters include the alumina raw ore parameters, the target particle diameter of the alumina raw ore, and the addition rate of the alumina raw ore; pulp-mill the set of alumina raw ore with qualified size according to the pulp-milling parameters; detect the generated alumina raw ore pulp after the pulp-milling time is reached.

[0065] Compared with the prior art, the beneficial effects of the present invention are:

[0066] 1. During the crushing process of the alumina raw ore in the present invention, first, the hardness values of multiple points on the surface of the alumina raw ore are averaged to obtain the surface hardness value of the alumina raw ore. The present invention sets different crushing parameters according to the surface hardness value to crush the alumina raw ore. This method can better represent the hardness level of the overall sample by measuring the hardness at multiple points and taking the average value, reducing the error caused by local non-uniformity; at the same time, obtaining the crushing parameters of the crusher according to the maximum value of the surface hardness value and the maximum feed particle size of the set ball mill can enable the crusher to accurately crush the alumina raw ore with large size, thereby improving the pulp-milling effect in the subsequent pulp-milling process.

[0067] 2. The present invention constructs a mixing liquid ratio model, which uses historical alumina ore mixing liquid ratio data for training, takes alumina ore parameters as input, and mixing liquid ratio as output, and uses the error between the mixing liquid ratio output in each iteration and the standard mixing liquid ratio as the loss function value of the mixing liquid ratio model, while dynamically adjusting the learning rate according to the change of the loss function. This model can avoid rapid overfitting caused by an excessively large learning rate or slow training speed caused by an excessively small learning rate. The present invention uses a trained mixing liquid ratio model to obtain the mixing liquid ratio. This method can combine the various characteristics of the alumina ore to accurately obtain the mixing liquid ratio, thereby improving the slurry grinding effect of the alumina ore.

[0068] 3. The present invention obtains a historical alumina slurry grinding data set, clusters the historical alumina characteristic parameters in the historical alumina slurry grinding data set, and obtains multiple cluster clusters. The present invention constructs a ball mill slurry grinding parameter prediction model, and uses the historical alumina slurry grinding data in different clusters to obtain multiple trained ball mill slurry grinding parameter prediction models. During the training process, the error between the historical ball mill parameter prediction value outputted in each iteration and the corresponding historical slurry grinding parameter standard value is used as the loss function value to optimize the weight and bias of the model. The present invention selects the corresponding trained ball mill slurry grinding parameter prediction model according to the distance between the alumina characteristic parameters and different clusters to obtain the ball mill parameters corresponding to the alumina characteristic parameters. This method can train specific models for different clusters, and can more accurately capture the alumina slurry grinding data patterns and changes in these specific clusters. Therefore, targeted prediction of the ball mill slurry grinding parameters of each cluster can significantly improve the prediction accuracy of the ball mill parameters, thereby improving the slurry grinding effect of the alumina ore. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 A schematic structural diagram of an advanced process control system for an alumina slurry mill provided by an embodiment of the present invention;

[0070] Figure 2 A schematic flow chart of an advanced process control method for alumina slurry grinding provided by an embodiment of the present invention;

[0071] Figure 3 A schematic diagram of a crushing process of a crushing unit provided in an embodiment of the present invention;

[0072] Figure 4 A schematic diagram of a pulping parameter acquisition process according to an embodiment of the present invention;

[0073] Figure 5 A schematic flow chart of an advanced pulp mill process control system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0074] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0075] Alumina ore is a key ore used in alumina production. Its chemical composition primarily consists of alumina, silicates, and other impurities. Alumina ore undergoes a series of processing steps to obtain alumina. Alumina slurry grinding is the process of mechanically grinding the coarsely ground alumina ore to the desired particle size range. This process is commonly used in the aluminum smelting industry to produce alumina slurry suitable for dissolution and dilution.

[0076] The present invention provides an advanced process control system for alumina slurry mill. This system is applied to an advanced process control method for alumina slurry mill. For specific system structure diagram and method flow chart, please refer to Figure 1 and Figure 2 .

[0077] Example 1

[0078] Reference Figure 1 The alumina ore classification unit is applied to S10 of an alumina slurry grinding advanced process control method.

[0079] Furthermore, the classification of the alumina ore to be pulp-milled in the alumina processing plant A includes:

[0080] The mesh holes of the vibrating screen are set according to the maximum feed particle size of the ball mill, and the size of the mesh holes is not larger than the maximum feed particle size of the ball mill; in this embodiment, the maximum feed particle size of the ball mill is 40 mm, and the size of the mesh holes is set to 40 mm; the alumina ore that needs to be slurry-milled is screened through the vibrating screen; the alumina ore remaining on the vibrating screen is divided into a collection of alumina ores with unqualified sizes; and the alumina ore screened out by the vibrating screen is divided into a collection of alumina ores with qualified sizes.

[0081] Because ball mills cannot slurry grind oversized alumina ore, the alumina ore needs to be screened based on size. In this embodiment, this step screens the alumina ore based on size, classifying large alumina ore into a collection of unqualified alumina. This method facilitates subsequent processing of alumina ores of different sizes.

[0082] Reference Figure 1The crushing unit is applied to S20 of an advanced process control method for alumina ore pulp grinding.

[0083] Further, referring to Figure 3 , the crushing process includes:

[0084] Step 1: When the number of alumina raw ores contained in the set of alumina raw ores with unqualified size is less than the set value W, jump to Step 5; otherwise, execute Step 2;

[0085] Step 2: Select S alumina raw ores from the set of alumina raw ores with unqualified size for hardness detection, S < W, and obtain the surface hardness values of the S alumina raw ores;

[0086] The calculation process of the surface hardness value includes:

[0087] Uniformly select multiple detection points on each alumina raw ore; use a hardness tester to detect the hardness values of each detection point; average the hardness values corresponding to the multiple detection points to obtain the surface hardness value of each alumina raw ore.

[0088] In this embodiment, during the crushing process of the alumina raw ore, first average the hardness values of multiple points on the surface of the alumina raw ore to obtain the surface hardness value of the alumina raw ore. This method can better represent the hardness level of the overall sample by measuring the hardness at multiple points and taking the average, reducing errors caused by local non-uniformity.

[0089] Step 3: Obtain the crushing parameters of the crusher according to the maximum value of the S surface hardness values and the maximum feed particle size of the ball mill set; the crushing parameters include the discharge port size, motor power, and crushing time of the crusher; the discharge port size is equal to the maximum feed particle size of the ball mill;

[0090] Step 4: The crushing unit crushes according to the crushing parameters. When the crushing time is reached, classify the alumina raw ores that fail to be discharged through the discharge port into the set of alumina raw ores with unqualified size, and jump to Step 1;

[0091] Step 5: The crushing unit does not perform crushing.

[0092] Further, Step 4 also includes: Classify the alumina raw ores discharged through the discharge port into the set of alumina raw ores with qualified size.

[0093] In this embodiment, different pulverization parameters are set based on the surface hardness value to pulverize the alumina ore. Simultaneously, the pulverization parameters of the pulverizer are determined based on the maximum surface hardness value and the set maximum feed particle size of the ball mill. Alumina ore that still fails to meet the pulverization standards is then repeatedly pulverized. This method enables the pulverizer to accurately pulverize large alumina ore, thereby improving the slurry grinding effect during the subsequent slurry grinding process.

[0094] Reference Figure 1 The mixing liquid control unit is applied to S30 of an advanced process control method for alumina slurry grinding.

[0095] Furthermore, after the crushing unit stops crushing, the mixing liquid control unit obtains the mixing liquid ratio according to the alumina ore parameters corresponding to the set of qualified alumina ores, and the specific steps include:

[0096] The alumina ore parameters include alumina ore size, alumina ore hardness, alumina ore density, alumina ore pH value and alumina ore humidity;

[0097] The process of obtaining the size of the alumina ore comprises: randomly selecting N alumina ore samples from the set of alumina ore of qualified size; calculating the minimum circumscribed sphere diameters of the N alumina ores, averaging the N minimum circumscribed sphere diameters, and using the calculated result as the size of the alumina ore;

[0098] The process of obtaining the hardness of the alumina ore is as follows: calculating the hardness of N alumina ores, averaging the hardness of the N alumina ores, and using the calculated result as the hardness of the alumina ore;

[0099] The process of obtaining the density, pH value and humidity of the alumina ore can refer to the process of obtaining the hardness of the alumina ore; the alumina ore parameters are input into the mixing liquid ratio model to obtain the ratio of the mixing liquid.

[0100] In this example, random sampling was used to obtain various parameters of the alumina ore, including size, hardness, density, pH, and moisture. This method ensures that the statistical characteristics of the sampled alumina ore are consistent with those of the population. Furthermore, the various parameters of the alumina ore obtained by this method provide a data foundation for subsequent prediction of the mixing solution ratio and pulp mill parameters.

[0101] Furthermore, in the alumina ore slurry grinding process, the mixing liquid (also known as the grinding agent) plays a vital role. It can affect the effect of ore grinding and the efficiency of the process. The mixing liquid is a mixture of water, lime, caustic soda, etc. The mixing liquid ratio model is constructed based on a deep belief network. The mixing liquid ratio model adaptively adjusts the learning rate according to the model loss during the training process; the training process of the mixing liquid ratio model includes: collecting the historical alumina ore mixing liquid ratio data set of alumina processing plant A, the historical alumina ore mixing liquid ratio data set includes 450 historical alumina mixing liquid ratio data; the historical alumina mixing liquid ratio data includes alumina ore parameters and its corresponding standard mixing liquid ratio;

[0102] The historical alumina ore blending solution ratio dataset is divided into a training set and a test set; the number of data items contained in the training set is 315, and the number of data items contained in the test set is 135;

[0103] Initializing parameters in the formulation ratio model; the parameters include learning rate, batch size, number of hidden layers, and number of neurons in each hidden layer;

[0104] The alumina ore parameters in the training set are used as input, and the blending liquid ratio prediction result is used as output; the blending loss function value of the blending liquid ratio model is calculated, and the expression of the blending loss function is:

[0105]

[0106] Among them, ss t is the value of the blending loss function at the tth iteration; M is the number of historical alumina blending solution ratio data in the training set; N is the number of additive types contained in the blending solution; pb (i,j,t) It is expressed as the proportion of the jth additive in the predicted result of the mixing solution ratio corresponding to the i-th historical alumina mixing solution ratio data output by the model after the t-th iteration; and bzpb (i,j) It is expressed as the ratio of the jth additive in the standard mixing solution ratio corresponding to the i-th historical alumina mixing solution ratio data; and exp() is an exponential function with a natural constant as the base. When the ratio of the prepared liquid output by the model is exactly the same as the ratio of the standard prepared liquid, ss t =1;ss t The larger it is, the worse the prediction effect of the model.

[0107] In this embodiment, this step obtains the blending loss function value for each iteration based on the difference between the blending ratio output by the blending ratio model and the standard blending ratio. The blending loss function value can measure the prediction effect of the blending ratio model and provide a data basis for subsequent adjustments to the learning rate of the blending ratio model.

[0108] Furthermore, when the value of the mixing loss function converges, the training ends and the learning rate in the mixing liquid ratio model is output; otherwise, the learning rate is adjusted according to the mixing loss function, and the training continues until the value of the mixing loss function converges; the formula for adjusting the learning rate according to the mixing loss function is:

[0109]

[0110] Among them, xxl t It represents the learning rate after the tth iteration; β1 represents the set increment value, β1>0; β2 represents the set decrement value, β2>0; ss t Expressed as the value of the deployment loss function at the t-th iteration; It represents the set loss reduction threshold; the test set is used to test the mixing liquid ratio model.

[0111] Furthermore, in order to verify the effectiveness of the mixed liquid ratio model in this embodiment, this embodiment also adopted four other models for comparison. The comparison results are shown in Table 1.

[0112] Table 1 Comparison of the predicted effects of the formulation ratio model of Example 1

[0113] Model Model 1 Model 2 Model 3 Model 4 Mean square error 0.021 0.043 0.064 0.069 Mean absolute error 0.046 0.096 0.106 0.132 Root mean square error 0.032 0.084 0.095 0.121 Mean absolute percentage error 0.072％ 0.167％ 0.194％ 0.249％

[0114] Furthermore, Model 1 in Table 1 is the mixed liquid ratio model proposed in the present invention, Model 2 is trained based on Model 1 using a fixed learning rate, Model 3 is trained based on Model 1 using an absolute value loss function, and Model 4 is trained using a traditional deep belief network. Table 1 shows that the mixed liquid ratio model proposed in this embodiment outperforms the other three models in all indicators.

[0115] This step of the present embodiment constructs a mixing liquid ratio model, which uses historical alumina ore mixing liquid ratio data for training, uses alumina ore parameters as input, and mixing liquid ratio as output, and uses the error between the mixing liquid ratio output in each iteration and the standard mixing liquid ratio as the loss function value of the mixing liquid ratio model, while dynamically adjusting the learning rate according to the change of the mixing loss function. This model can avoid rapid overfitting caused by an excessively large learning rate or slow training speed caused by an excessively small learning rate. The present invention uses a trained mixing liquid ratio model to obtain the mixing liquid ratio. This method can combine the various characteristics of the alumina ore to accurately obtain the mixing liquid ratio, thereby improving the slurry grinding effect of the alumina ore.

[0116] Reference Figure 1 A pulp mill parameter acquisition unit is provided, wherein the pulp mill parameter acquisition unit is applied to S40 of an advanced process control method for alumina pulp milling.

[0117] Further, refer to Figure 4 , the obtaining of slurry grinding parameters according to the characteristic parameters of alumina includes:

[0118] The alumina characteristic parameters include the alumina ore parameters, the target particle diameter of the alumina ore and the alumina ore addition rate; the slurry milling parameters include the ball mill parameters and the mixing liquid addition rate; the ball mill parameters include the ball mill speed, motor power and slurry milling time; the mixing liquid addition rate is obtained based on the alumina ore addition rate; the ball mill model and size data selected by the alumina processing factory A in this embodiment refer to Table 2.

[0119] Table 2 Ball mill models and dimensions selected by Alumina Processing Plant A

[0120] model Cylinder diameter Cylinder length Motor model Effective volume Processing power Maximum feed size MQYg2436 2400mm 3600mm JR137-6 <![CDATA[13.8m 3 ]]> 8~110t / h 40mm

[0121] Furthermore, the process of obtaining the ball mill parameters includes:

[0122] Collecting a historical alumina slurry grinding data set, wherein the historical alumina slurry grinding data set includes 450 historical alumina characteristic parameters and corresponding historical ball mill parameter standard values;

[0123] Cluster the 450 historical alumina characteristic parameters to obtain P clusters, each cluster containing at least k historical alumina characteristic parameters, where k is a set value to ensure that the model training data is sufficient; in this embodiment, k is set to 50; a ball mill slurry grinding parameter prediction model is constructed based on a multi-layer perceptron; the ball mill slurry grinding parameter prediction model is trained using the data in the P clusters to obtain P trained ball mill slurry grinding parameter prediction models; the distance between the alumina characteristic parameters and the centers of the P clusters is calculated, and the cluster with the closest distance is selected as the prediction cluster; the alumina characteristic parameters are input into the trained ball mill slurry grinding parameter prediction model corresponding to the prediction cluster to obtain the ball mill parameters corresponding to the alumina characteristic parameters.

[0124] The training process of the ball mill grinding parameter prediction model includes:

[0125] Randomly initializing the weights and biases of the ball mill slurry grinding parameter prediction model;

[0126] The historical alumina characteristic parameters in the cluster are input into the ball mill slurry grinding parameter prediction model to obtain the historical ball mill parameter prediction value; the ball mill loss function value in the ball mill slurry grinding parameter prediction model is calculated, and the calculation formula is:

[0127]

[0128] Among them, csyc t It represents the ball mill loss function value after the tth iteration in the ball mill grinding parameter prediction model training process; m represents the number of data items contained in the cluster; qmcs (e,d,t) bzcs represents the dth historical ball mill parameter prediction value corresponding to the eth historical alumina characteristic parameter output by the ball mill slurry grinding parameter prediction model at the tth iteration; (e,d) It represents the dth historical ball mill parameter standard value corresponding to the eth historical alumina characteristic parameter; d=1 corresponds to the speed of the ball mill; d=2 corresponds to the motor power of the ball mill; d=3 corresponds to the slurry grinding time of the ball mill.

[0129] In this embodiment, this step obtains the ball mill loss function value for each iteration based on the difference between the various ball mill parameters output by the model and the historical standard values of the ball mill parameters. The ball mill loss function value can measure the prediction performance of the ball mill slurry grinding parameter prediction model and provide a data basis for subsequent parameter adjustments of the ball mill slurry grinding parameter prediction model.

[0130] Furthermore, when the ball mill loss function value converges, the training ends and the weight and bias of the ball mill slurry grinding parameter prediction model are output; otherwise, the training continues until the ball mill loss function value converges.

[0131] Furthermore, this example selected 10 different alumina characteristic parameters, obtained a prediction cluster corresponding to each alumina characteristic parameter, and obtained ball mill parameters based on the corresponding trained ball mill slurry grinding parameter prediction model. To verify the effectiveness of this method, this example also selected three other methods to predict ball mill parameters, and selected the ball mill loss function value in this example as the error indicator to verify the effectiveness of the ball mill parameter prediction method proposed in this example. Specific results are shown in Table 3.

[0132] Table 3 Ball mill parameter prediction results of Example 1

[0133] Alumina characteristic parameter serial number Method 1 Method 2 Method 3 Method 4 1 0.046 0.086 0.116 0.119 2 0.057 0.089 0.114 0.121 3 0.048 0.082 0.107 0.116 4 0.051 0.094 0.108 0.117 5 0.049 0.090 0.110 0.118 6 0.055 0.086 0.112 0.116 7 0.037 0.090 0.114 0.118 8 0.042 0.085 0.113 0.120 9 0.038 0.093 0.109 0.123 10 0.047 0.091 0.112 0.119

[0134] Furthermore, the first method in Table 3 is the method proposed in this embodiment, which includes first clustering the characteristic parameters of aluminum oxide and then selecting the corresponding trained ball mill slurry grinding parameter prediction model for prediction; the second method is based on the first method and uses the mean absolute error as the loss function to train the ball mill slurry grinding parameter prediction model; the prediction process of the third method includes not clustering the characteristic parameters of aluminum oxide, but directly using the historical aluminum oxide slurry grinding data set to train a ball mill slurry grinding parameter prediction model, and inputting the ball mill slurry grinding parameters into the same ball mill slurry grinding parameter prediction model for prediction; the fourth method is based on the third method and replaces the ball mill slurry grinding parameter prediction model with a traditional multi-layer perceptron for prediction. It can be seen from Table 3 that the method proposed in this embodiment can effectively improve the accuracy of slurry grinding parameter prediction.

[0135] This step of the present embodiment obtains a historical alumina slurry grinding data set, clusters the historical alumina characteristic parameters in the historical alumina slurry grinding data set, and obtains multiple cluster clusters. The present invention constructs a ball mill slurry grinding parameter prediction model based on a multilayer perceptron, and uses the historical alumina slurry grinding data in different clusters to obtain multiple trained ball mill slurry grinding parameter prediction models. During the training process, the error between the historical ball mill parameter prediction value obtained in each iteration and the corresponding historical slurry grinding parameter standard value is used as the ball mill loss function value to optimize the weight and bias of the model. The present invention selects the corresponding trained ball mill slurry grinding parameter prediction model according to the distance between the alumina characteristic parameters and different clusters to obtain the ball mill parameters corresponding to the alumina characteristic parameters. This method can train specific models for different clusters, and can more accurately capture the data patterns and changes of these specific clusters. Therefore, targeted prediction of the ball mill slurry grinding parameters of each cluster can significantly improve the prediction accuracy of the ball mill parameters, thereby improving the slurry grinding effect of the alumina ore.

[0136] Reference Figure 1The pulp grinding unit and the pulp grinding result detection unit are applied to S50 and S60 of an advanced process control method for alumina pulp grinding.

[0137] Furthermore, the pulping result detection unit includes:

[0138] M liters of the alumina ore slurry are randomly selected as the ore slurry test sample; a first weight of solid particles in the ore slurry with a particle diameter greater than the set target particle diameter of the alumina ore is obtained; a second weight of all solid particles in the ore slurry is simultaneously obtained; the ratio of the first weight to the second weight is calculated; when the ratio is greater than the set ratio, a slurry grinding failure warning is output; otherwise, the alumina ore slurry and corresponding parameters are obtained, and the mixing liquid ratio model and the ball mill slurry grinding parameter prediction model are retrained. The alumina ore slurry is then classified to screen out solid particles with a particle diameter greater than the set target particle diameter of the alumina ore, thereby obtaining an alumina ore slurry suitable for dissolution and dilution operations.

[0139] This step in this embodiment tests the alumina slurry generated after pulp grinding, calculates the weight percentage of particles whose diameters do not reach the target diameter, and issues an early warning if the percentage is too high. This method can provide real-time information on abnormalities in the alumina slurry process, helping production managers identify and resolve potential problems promptly.

[0140] Furthermore, the retraining of the mixing liquid ratio model and the ball mill mill parameter prediction model includes: obtaining corresponding parameters in the milling process; the corresponding parameters include alumina ore parameters, alumina characteristic parameters, mixing liquid ratio and milling parameters;

[0141] The ratio of the prepared liquid is used as the standard prepared liquid ratio corresponding to the parameters of the alumina ore;

[0142] Adding the alumina ore parameters and the corresponding standard blending solution ratio to a historical alumina ore blending solution ratio dataset, and retraining the blending solution ratio model;

[0143] The alumina characteristic parameters and the ball mill parameters in the slurry grinding parameters are added to the data corresponding to the prediction cluster in the historical alumina slurry grinding data set, and the ball mill slurry grinding parameter prediction model corresponding to the prediction cluster is trained again.

[0144] After outputting the qualified pulp milling result, this step in this embodiment obtains the corresponding parameters of the pulp milling process and retrains the mixing liquid ratio model and the ball mill pulp mill parameter prediction model. This method can expand the training data of the two models, thereby helping the models better learn features and thus improve the accuracy of model prediction.

[0145] This embodiment obtains a flow chart of an advanced process control system for alumina slurry mill according to the above steps. Figure 5 . In this embodiment, the alumina ore is input into the alumina ore control unit for classification, and a set of alumina ore with qualified size and a set of alumina ore with unqualified size are obtained; the set of alumina ore with unqualified size is input into the crushing unit for crushing, and the alumina ore with qualified size is classified into the set of alumina ore with qualified size; then, the alumina ore parameters are obtained according to the set of alumina ore with qualified size, and the mixing liquid control unit is controlled according to the alumina ore parameters to obtain the ratio of the mixing liquid and generate the mixing liquid; at the same time, the alumina characteristic parameters are obtained, and the slurry milling parameters of the ball mill are obtained through the slurry milling parameter acquisition unit; then, the mixing liquid and the alumina ore in the set of alumina ore with qualified size are input into the slurry milling unit, and the ball mill is controlled by the slurry milling parameters for slurry milling; and the alumina ore slurry output from the ball mill is input into the slurry milling result detection unit for detection.

[0146] Example 2

[0147] Reference Figure 1 The alumina ore classification unit is applied to S10 of an alumina slurry grinding advanced process control method.

[0148] Furthermore, the classification of the alumina ore to be pulp-milled in the alumina processing plant B includes:

[0149] The mesh holes of the vibrating screen are set according to the maximum feed particle size of the ball mill, and the size of the mesh holes is not larger than the maximum feed particle size of the ball mill; in this embodiment, the maximum feed particle size of the ball mill is 35 mm, and the size of the mesh holes is set to 35 mm; the alumina ore that needs to be slurry-milled is screened through the vibrating screen; the alumina ore remaining on the vibrating screen is divided into a collection of alumina ores with unqualified sizes; and the alumina ore screened out by the vibrating screen is divided into a collection of alumina ores with qualified sizes.

[0150] Reference Figure 1 The pulverizing unit is applied to an S20 of an advanced process control method for alumina slurry grinding.

[0151] Furthermore, the crushing process includes:

[0152] Step 1: When the number of alumina ores in the set of unqualified alumina ores is less than the set value W, jump to step 5; otherwise, execute step 2.

[0153] Step 2: Select S alumina raw ores from the set of alumina raw ores with unqualified size, where S < W, and perform hardness testing on these S alumina raw ores to obtain the surface hardness values of the S alumina raw ores;

[0154] The calculation process of the surface hardness value includes:

[0155] Uniformly select multiple detection points on each of the alumina raw ores; use a hardness tester to detect the hardness values of each of the detection points; average the hardness values corresponding to the multiple detection points to obtain the surface hardness value of each alumina raw ore.

[0156] Step 3: Obtain the crushing parameters according to the maximum value of the S surface hardness values and the maximum feed particle size of the set ball mill. The crushing parameters include the discharge port size, motor power, and crushing time;

[0157] Step 4: The crushing unit performs crushing according to the crushing parameters. When the crushing time is reached, the alumina raw ores that fail to be discharged through the discharge port are classified into the set of alumina raw ores with unqualified size, and then jump to Step 1;

[0158] Step 5: The crushing unit does not perform crushing.

[0159] Refer to Figure 1 the dispensing liquid control unit in, and the dispensing liquid control unit is applied to S30 of an advanced process control method for an alumina ore mill.

[0160] Further, the obtaining of the ratio of the dispensing liquid according to the alumina raw ore parameters corresponding to the set of alumina raw ores with qualified size includes: the alumina raw ore parameters include the alumina raw ore size, alumina raw ore hardness, alumina raw ore density, alumina raw ore pH value, and alumina raw ore humidity; the obtaining process of the alumina raw ore size includes: randomly select N alumina raw ore samples from the set of alumina raw ores with qualified size;

[0161] Calculate the minimum circumscribed sphere diameter of the N alumina raw ores, average the N minimum circumscribed sphere diameters, and use the calculation result as the alumina raw ore size;

[0162] The obtaining process of the alumina raw ore hardness is:

[0163] Calculate the hardness of the N alumina raw ores, average the hardness of the N alumina raw ores, and use the calculation result as the alumina raw ore hardness; the obtaining processes of the alumina raw ore density, alumina raw ore pH value, and alumina raw ore humidity can refer to the obtaining process of the alumina raw ore hardness; input the alumina raw ore parameters into the dispensing liquid ratio model to obtain the ratio of the dispensing liquid.

[0164] Furthermore, the mixing liquid ratio model is constructed based on a deep belief network, and the mixing liquid ratio model adaptively adjusts the learning rate according to the model loss during the training process; the training process of the mixing liquid ratio model includes:

[0165] Collect a historical alumina ore-mixing solution ratio dataset from alumina processing plant B, the dataset comprising 500 pieces of historical alumina-mixing solution ratio data; the historical alumina-mixing solution ratio data comprises alumina ore parameters and their corresponding standard mixing solution ratios;

[0166] The historical alumina raw ore blending solution ratio dataset is divided into a training set and a test set; the number of data items contained in the training set is 350, and the number of data items contained in the test set is 150;

[0167] Initializing parameters in the formulation ratio model; the parameters include learning rate, batch size, number of hidden layers, and number of neurons in each hidden layer;

[0168] The alumina ore parameters in the training set are used as input, and the blending liquid ratio prediction result is used as output; the blending loss function value of the blending liquid ratio model is calculated, and the expression of the blending loss function is:

[0169]

[0170] Among them, ss t is the value of the blending loss function at the tth iteration; M is the number of historical alumina blending solution ratio data in the training set; N is the number of additive types contained in the blending solution; pb (i,j,t) It is expressed as the proportion of the jth additive in the predicted result of the mixing solution ratio corresponding to the i-th historical alumina mixing solution ratio data output by the model after the t-th iteration; and bzpb (i,j) It is expressed as the ratio of the jth additive in the standard mixing solution ratio corresponding to the i-th historical alumina mixing solution ratio data; and exp() is an exponential function with a natural constant as the base. When the ratio of the prepared liquid output by the model is exactly the same as the ratio of the standard prepared liquid, ss t =1;ss t The larger the value, the worse the prediction effect of the model. When the allocation loss function value converges, the training ends and the learning rate in the allocation liquid ratio model is output. Otherwise, the learning rate is adjusted according to the allocation loss function and the training continues until the allocation loss function value converges. The formula for adjusting the learning rate according to the allocation loss function is:

[0171]

[0172] Among them, xxl t It represents the learning rate after the tth iteration; β1 represents the set increment value, β1>0; β2 represents the set decrement value, β2>0; ss t Expressed as the deployment loss function of the t-th iteration; It represents the set loss reduction threshold; the test set is used to test the mixing liquid ratio model.

[0173] Furthermore, to verify the effectiveness of the formulation ratio model in this embodiment, this embodiment also uses four other models for comparison. The comparison results are shown in Table 4.

[0174] Table 4 Comparison of the predicted effects of the mixing ratio model of Example 2

[0175] Model Model 1 Model 2 Model 3 Model 4 Mean square error 0.018 0.040 0.059 0.064 Mean absolute error 0.037 0.086 0.124 0.142 Root mean square error 0.027 0.073 0.094 0.126 Mean absolute percentage error 0.071％ 0.157％ 0.206％ 0.237％

[0176] Furthermore, in Table 4, Model 1 is the mixed liquid ratio model proposed in the present invention, Model 2 is trained based on Model 1 using a fixed learning rate, Model 3 is trained based on Model 1 using an absolute value loss function, and Model 4 is trained using a traditional deep belief network. Table 4 shows that the mixed liquid ratio model proposed in this embodiment outperforms the other three models in all indicators.

[0177] Reference Figure 1 A pulp mill parameter acquisition unit is provided, wherein the pulp mill parameter acquisition unit is applied to S40 of an advanced process control method for alumina pulp milling.

[0178] Furthermore, obtaining the slurry grinding parameters according to the alumina characteristic parameters includes:

[0179] The alumina characteristic parameters include the alumina ore parameters, the target particle diameter of the alumina ore and the alumina ore addition rate; the slurry milling parameters include the ball mill parameters and the mixing liquid addition rate; the ball mill parameters include the ball mill speed, motor power and slurry milling time; the mixing liquid addition rate is obtained based on the alumina ore addition rate; the ball mill model and size data selected by the alumina processing factory B in this embodiment refer to Table 5.

[0180] Table 5 Ball mill models and dimensions selected by Alumina Processing Plant B

[0181] model Cylinder diameter Cylinder length Motor model Effective volume Processing power Maximum feed size QYgDJ 1800mm 3620mm JR136-8 <![CDATA[8.2m 3 ]]> 4.5~29t / h 35mm

[0182] Furthermore, the process of obtaining the ball mill parameters includes:

[0183] Collecting a historical alumina slurry grinding data set, wherein the historical alumina slurry grinding data set includes 500 historical alumina characteristic parameters and corresponding historical ball mill parameter standard values;

[0184] Clustering the 500 historical alumina characteristic parameters to obtain P clusters, each cluster containing at least k historical alumina characteristic parameters, where k is a set value to ensure sufficient model training data;

[0185] In this embodiment, k is set to 50; a ball mill slurry grinding parameter prediction model is constructed based on a multi-layer perceptron; the ball mill slurry grinding parameter prediction model is trained using data from P clusters to obtain P trained ball mill slurry grinding parameter prediction models; the distance between the alumina characteristic parameters and the centers of the P clusters is calculated, and the cluster with the closest distance is selected as the prediction cluster; the alumina characteristic parameters are input into the trained ball mill slurry grinding parameter prediction model corresponding to the prediction cluster to obtain the ball mill parameters corresponding to the alumina characteristic parameters.

[0186] The training process of the ball mill grinding parameter prediction model includes: randomly initializing the weights and biases of the ball mill grinding parameter prediction model; inputting the historical alumina characteristic parameters in the cluster into the ball mill grinding parameter prediction model to obtain historical ball mill parameter prediction values; and calculating the ball mill loss function value in the ball mill grinding parameter prediction model, using the following formula:

[0187]

[0188] Among them, csyc t It represents the ball mill loss function value after the tth iteration in the ball mill grinding parameter prediction model training process; m represents the number of data items contained in the cluster; qmcs (e,d,t) bzcs represents the dth historical ball mill parameter prediction value corresponding to the eth historical alumina characteristic parameter output by the ball mill slurry grinding parameter prediction model at the tth iteration; (e,d) It represents the dth historical ball mill parameter standard value corresponding to the eth historical alumina characteristic parameter; d=1 corresponds to the ball mill speed; d=2 corresponds to the ball mill motor power; d=3 corresponds to the ball mill slurry grinding time;

[0189] When the ball mill loss function value converges, the training ends and the weight and bias of the ball mill slurry grinding parameter prediction model are output; otherwise, the training continues until the ball mill loss function value converges.

[0190] Furthermore, this example selected five different alumina characteristic parameters, obtained a prediction cluster corresponding to each alumina characteristic parameter, and obtained ball mill parameters based on the corresponding trained ball mill slurry grinding parameter prediction model. To verify the effectiveness of this method, this example also selected three other methods to predict ball mill parameters, and selected the ball mill loss function value in this example as the error indicator to verify the effectiveness of the ball mill parameter prediction method proposed in this example. Specific results are shown in Table 6.

[0191] Table 6 Ball mill parameter prediction results of Example 2

[0192] Alumina characteristic parameter serial number Method 1 Method 2 Method 3 Method 4 1 0.041 0.084 0.096 0.112 2 0.039 0.083 0.102 0.111 3 0.043 0.079 0.098 0.108 4 0.040 0.082 0.101 0.112 5 0.038 0.080 0.094 0.110

[0193] Furthermore, method 1 in Table 6 is the method proposed in this embodiment, which includes first clustering the characteristic parameters of aluminum oxide and then selecting the corresponding trained ball mill slurry grinding parameter prediction model for prediction; method 2 is based on method 1, and uses the mean absolute error as the loss function to train the ball mill slurry grinding parameter prediction model; the prediction process of method 3 includes not clustering the characteristic parameters of aluminum oxide, but directly using the historical aluminum oxide slurry grinding data set to train a ball mill slurry grinding parameter prediction model, and inputting the ball mill slurry grinding parameters into the same ball mill slurry grinding parameter prediction model for prediction; method 4 is based on method 3, and replaces the ball mill slurry grinding parameter prediction model with a traditional multi-layer perceptron for prediction. As can be seen from Table 6, the method proposed in this embodiment can effectively improve the accuracy of slurry grinding parameter prediction.

[0194] Reference Figure 1 The pulp grinding unit and the pulp grinding result detection unit are applied to S50 and S60 of an advanced process control method for alumina pulp grinding.

[0195] Furthermore, the pulping result detection unit includes:

[0196] Randomly select M liters of the alumina slurry as the slurry test sample; obtain the first weight of solid particles in the slurry whose particle diameter is greater than the set alumina target particle diameter; simultaneously obtain the second weight of all solid particles in the slurry; calculate the proportion of the first weight in the second weight; when the proportion is greater than the set ratio, output a slurry grinding failure warning; otherwise, obtain the alumina slurry and corresponding parameters, and re-train the mixing liquid ratio model and the ball mill slurry grinding parameter prediction model.

[0197] Furthermore, the retraining of the mixing liquid ratio model and the ball mill mill parameter prediction model includes: obtaining corresponding parameters in the milling process; the corresponding parameters include alumina ore parameters, alumina characteristic parameters, mixing liquid ratio and milling parameters;

[0198] The ratio of the prepared liquid is used as the standard prepared liquid ratio corresponding to the parameters of the alumina ore;

[0199] Adding the alumina ore parameters and the corresponding standard blending solution ratio to a historical alumina ore blending solution ratio dataset, and retraining the blending solution ratio model;

[0200] The alumina characteristic parameters and the ball mill parameters in the slurry grinding parameters are added to the data corresponding to the prediction cluster in the historical alumina slurry grinding data set, and the ball mill slurry grinding parameter prediction model corresponding to the prediction cluster is trained again.

[0201] In this embodiment, during the pulverization of alumina ore, the hardness values at multiple points on the surface of the alumina ore are first averaged to obtain the surface hardness value of the alumina ore. This embodiment sets different pulverization parameters based on the surface hardness value to pulverize the alumina ore. By measuring the hardness at multiple points and averaging the average, this method can better represent the hardness level of the entire sample and reduce errors caused by local heterogeneity. Furthermore, determining the pulverizer parameters based on the maximum surface hardness value and the set maximum feed size of the ball mill enables the pulverizer to accurately pulverize large alumina ore, thereby improving the slurry milling effect during the subsequent slurry milling process. This embodiment constructs a formulation ratio model. This model is trained using historical alumina ore formulation ratio data, using alumina ore parameters as input and formulation ratio as output. The error between the formulation ratio output at each iteration and the standard formulation ratio is used as the formulation ratio model's loss function value. The learning rate is dynamically adjusted based on changes in the loss function. This model can avoid rapid overfitting caused by an excessively large learning rate or slow training speed caused by an excessively small learning rate. This embodiment uses a trained mixing liquid ratio model to obtain the ratio of the mixing liquid. This method can combine the various characteristics of the alumina ore to accurately obtain the ratio of the mixing liquid, thereby improving the slurry grinding effect of the alumina ore. This embodiment obtains a historical alumina slurry grinding data set, clusters the historical alumina characteristic parameters in the historical alumina slurry grinding data set, and obtains multiple cluster clusters. This embodiment constructs a ball mill slurry grinding parameter prediction model based on a multi-layer perceptron, and uses historical alumina slurry grinding data in different clusters to obtain multiple trained ball mill slurry grinding parameter prediction models. During the training process, the error between the historical ball mill parameter prediction value and the corresponding historical slurry grinding parameter standard value is used as a loss function to optimize the weight and bias of the model. This embodiment selects the corresponding trained ball mill slurry grinding parameter prediction model based on the distance between the alumina characteristic parameters and different clusters to obtain the ball mill parameters corresponding to the alumina characteristic parameters. This method can train specific models for different clusters, and can more accurately capture the data patterns and changes of these specific clusters. Therefore, targeted prediction of the ball mill slurry grinding parameters of each cluster can significantly improve the prediction accuracy of the ball mill parameters, thereby improving the slurry grinding effect of alumina ore.

[0202] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An advanced process control system for alumina slurry grinding, characterized in that: Comprising: Aluminum oxide raw ore classification unit: used to classify aluminum oxide raw ore that needs to be pulp-milled, and obtain a set of aluminum oxide raw ore with qualified size and a set of aluminum oxide raw ore with unqualified size; Crushing unit: used to crush the aluminum oxide raw ore in the set of aluminum oxide raw ore with unqualified size. The specific process includes: Step 1: When the number of aluminum oxide raw ore contained in the set of aluminum oxide raw ore with unqualified size is less than the set value W, jump to Step 5; otherwise, execute Step 2; Step 2: Select S of the aluminum oxide raw ore in the set of aluminum oxide raw ore with unqualified size for hardness detection, S < W, and obtain the surface hardness values of the S aluminum oxide raw ore; [[ID=!5]]Step 3: Obtain crushing parameters based on the maximum value of the S surface hardness values and the maximum feed particle size of the ball mill set; Step 4: The crushing unit crushes according to the crushing parameters. When the crushing time is reached, the aluminum oxide raw ore that fails to be discharged through the discharge port is classified into the set of aluminum oxide raw ore with unqualified size, and jump to Step 1; Step 5: The crushing unit does not perform crushing; Blending liquid control unit: used to obtain the blending ratio of the blending liquid according to the aluminum oxide raw ore parameters corresponding to the set of aluminum oxide raw ore with qualified size; Pulp-milling parameter acquisition unit: used to obtain pulp-milling parameters according to aluminum oxide characteristic parameters; Pulp-milling unit: used to pulp-mill the set of aluminum oxide raw ore with qualified size according to the pulp-milling parameters; Pulp-milling result detection unit: used to detect the generated aluminum oxide raw ore pulp after the pulp-milling time is reached.

2. The alumina slurry mill advanced process control system according to claim 1, characterized in that: The classification of the aluminum oxide raw ore that needs to be pulp-milled includes: Set the sieve holes of the vibrating screen according to the maximum feed particle size of the ball mill. The size of the sieve holes is not greater than the maximum feed particle size of the ball mill; screen the aluminum oxide raw ore that needs to be pulp-milled through the vibrating screen; classify the aluminum oxide raw ore remaining on the vibrating screen into the set of aluminum oxide raw ore with unqualified size; classify the aluminum oxide raw ore sieved out through the vibrating screen into the set of aluminum oxide raw ore with qualified size.

3. The alumina slurry mill advanced process control system according to claim 1, characterized in that: The calculation process of the surface hardness value includes: Uniformly select multiple detection points on each of the aluminum oxide raw ore; use a hardness tester to detect the hardness values of each of the detection points; average the hardness values corresponding to the multiple detection points to obtain the surface hardness value of each of the aluminum oxide raw ore.

4. The alumina slurry mill advanced process control system according to claim 1, characterized in that: The steps to obtain the blending ratio of the blending liquid include: The aluminum oxide raw ore parameters include the size of the aluminum oxide raw ore, the hardness of the aluminum oxide raw ore, the density of the aluminum oxide raw ore, the pH value of the aluminum oxide raw ore, and the humidity of the aluminum oxide raw ore; the process of obtaining the size of the aluminum oxide raw ore includes: randomly select N aluminum oxide raw ore samples from the set of aluminum oxide raw ore with qualified size; calculate the minimum circumscribed sphere diameter of the N aluminum oxide raw ore, and average the N minimum circumscribed sphere diameters. The calculation result is used as the size of the aluminum oxide raw ore; The process of obtaining the hardness of the aluminum oxide raw ore is: calculate the hardness of the N aluminum oxide raw ore, and average the hardness of the N aluminum oxide raw ore. The calculation result is used as the hardness of the aluminum oxide raw ore; The process of obtaining the density, pH value and humidity of the alumina ore refers to the process of obtaining the hardness of the alumina ore; the alumina ore parameters are input into the mixing liquid ratio model to obtain the ratio of the mixing liquid.

5. The alumina slurry mill advanced process control system according to claim 4, characterized in that: The formulation ratio model adaptively adjusts the learning rate according to the formulation loss function value of the model during the training process; The training process of the formulation ratio model includes: Collecting a historical alumina ore blending solution ratio data set, wherein the historical alumina ore blending solution ratio data set includes multiple historical alumina blending solution ratio data; the historical alumina blending solution ratio data includes alumina ore parameters of the alumina ore and its corresponding standard blending solution ratio; Dividing the historical alumina ore blending solution ratio dataset into a training set and a test set; initializing the learning rate in the blending solution ratio model; using the alumina ore parameters in the training set as input and the blending solution ratio as output; and calculating the blending loss function value of the blending solution ratio model; When the mixing loss function converges, the training ends and the learning rate in the mixing liquid ratio model is output; otherwise, the learning rate is adjusted according to the mixing loss function and the training continues until the mixing loss function converges; the formula for adjusting the learning rate according to the mixing loss function is: ； in, Expressed as the learning rate after the tth iteration; Indicates the set increment value. ; Indicates the set decrement value, ; It is expressed as the set loss reduction threshold; It is expressed as the value of the mixing loss function at the t-th iteration; a test set is used to test the effect of the mixing liquid ratio model on the prediction of the mixing liquid ratio.

6. The alumina slurry mill advanced process control system according to claim 5, characterized in that: The expression of the deployment loss function is: ； in, It represents the value of the blending loss function at the tth iteration; M represents the number of historical alumina blending solution ratio data in the training set; N represents the number of additive types contained in the blending solution; It is expressed as the ratio of the jth additive in the mixing solution corresponding to the i-th historical alumina mixing solution ratio data output by the model after the t-th iteration; and ; It is expressed as the ratio of the jth additive in the standard mixing solution ratio corresponding to the i-th historical alumina mixing solution ratio data; and ; Expressed as an exponential function with a natural constant as base.

7. The alumina slurry mill advanced process control system according to claim 1, characterized in that: The obtaining of slurry grinding parameters according to the characteristic parameters of alumina includes: The alumina characteristic parameters include the alumina ore parameters, the target particle diameter of the alumina ore, and the alumina ore addition rate; the slurry milling parameters include the ball mill parameters and the mixing liquid addition rate; the ball mill parameters include the ball mill speed, motor power, and slurry milling time; the mixing liquid addition rate is obtained based on the alumina ore addition rate; A historical alumina slurry grinding data set is collected, wherein the historical alumina slurry grinding data set includes Q historical alumina characteristic parameters and corresponding historical ball mill parameter standard values; the Q historical alumina characteristic parameters are clustered to obtain P clusters; a ball mill parameter prediction model is trained using data in the P clusters to obtain P trained ball mill slurry grinding parameter prediction models; the training process of the ball mill parameter prediction model includes: Randomly initializing the weights and biases of the ball mill parameter prediction model; Inputting the historical alumina characteristic parameters in the cluster into the ball mill parameter prediction model to obtain historical ball mill parameter prediction values; calculating the ball mill loss function value in the ball mill parameter prediction model; When the ball mill loss function converges, the training ends and the weight and bias of the ball mill parameter prediction model are output; otherwise, the training continues until the ball mill loss function converges; Calculate the distance between the alumina characteristic parameters and the centers of P clusters, and select the cluster with the closest distance as the prediction cluster; input the alumina characteristic parameters into the trained ball mill slurry grinding parameter prediction model corresponding to the prediction cluster to obtain the ball mill parameters corresponding to the alumina characteristic parameters.

8. The alumina slurry mill advanced process control system according to claim 7, characterized in that: The ball mill loss function is expressed as: ； in, It is represented as the ball mill loss function value after the tth iteration in the ball mill parameter prediction model training process; It represents the number of data items contained in the cluster; d=1 represents the rotation speed of the ball mill; d=2 represents the motor power of the ball mill; d=3 represents the slurry grinding time of the ball mill; It represents the dth historical ball mill parameter prediction value corresponding to the eth historical alumina characteristic parameter output by the ball mill parameter prediction model at the tth iteration; It is expressed as the dth historical ball mill parameter standard value corresponding to the eth historical alumina characteristic parameter.

9. The alumina slurry mill advanced process control system according to claim 1, characterized in that: The specific steps for detecting the generated alumina raw ore pulp after reaching the pulp grinding time include: Randomly select M liters of the alumina raw ore pulp as the raw ore pulp detection sample; obtain the first weight of the solid particles with a particle diameter larger than the set target particle diameter of the alumina raw ore in the raw ore pulp; simultaneously obtain the second weight of all solid particles in the raw ore pulp; calculate the proportion of the first weight in the second weight; when the proportion is greater than the set ratio, output a warning of unqualified pulp grinding; otherwise, obtain the alumina raw ore pulp.

10. An advanced process control method for alumina slurry grinding, characterized in that: Include: Classify the alumina raw ore to be pulp - ground, and obtain a set of alumina raw ore with qualified size and a set of alumina raw ore with unqualified size; Crush the alumina raw ore in the set of alumina raw ore with unqualified size. The specific process includes: Step 1: When the number of alumina raw ore contained in the set of alumina raw ore with unqualified size is less than the set value W, jump to Step 5; otherwise, execute Step 2; Step 2: Select S alumina raw ore from the set of alumina raw ore with unqualified size for hardness detection, S < W, and obtain the surface hardness values of the S alumina raw ore; Step 3: Obtain the crushing parameters according to the maximum value of the S surface hardness values and the maximum feed particle size of the ball mill set; Step 4: The crushing unit crushes according to the crushing parameters. When the crushing time is reached, classify the alumina raw ore that fails to be discharged through the discharge port into the set of alumina raw ore with unqualified size, and jump to Step 1; Step 5: The crushing unit does not perform crushing; Obtain the proportion of the preparation liquid according to the alumina raw ore parameters corresponding to the set of alumina raw ore with qualified size; Obtain the pulp grinding parameters according to the alumina characteristic parameters; the alumina characteristic parameters include the alumina raw ore parameters, the target particle diameter of the alumina raw ore, and the addition rate of the alumina raw ore; pulp - grind the set of alumina raw ore with qualified size according to the pulp grinding parameters; and detect the generated alumina raw ore pulp after reaching the pulp grinding time.

Images