An intelligent detection and setting system for alumina dissolution process

Abstract

The present invention relates to the field of aluminum oxide technology, and specifically to an intelligent detection and setting system for aluminum oxide dissolution process. First, real-time data is obtained from the production line, including a first data set, a second data set, and a third data set; then, three module historical data sets in the historical data are obtained and preprocessed to obtain three preprocessing sets; according to the three preprocessing sets, model training is performed respectively to obtain three parameter prediction models; further, the three parameter prediction models analyze the first data set, the second data set, and the third data set respectively to obtain a first prediction result, a second prediction result, and a third prediction result; finally, according to the three prediction results, the first parameter group, the second parameter group, and the third parameter group in the dissolution process are set respectively.

Description

Technical Field

[0001] The present invention relates to the technical field of aluminum oxide, and in particular to an intelligent detection and setting system for an aluminum oxide dissolution process. Background Art

[0002] The alumina dissolution process is a key preliminary stage in the production of aluminum products. The main purpose is to extract alumina from bauxite. This process usually involves reacting the ore with a caustic soda solution under high temperature and pressure conditions, so that the alumina dissolves from the ore to form a solution. The dissolved alumina is then separated from the solution through a series of steps such as cooling, precipitation and filtration to obtain a high-purity alumina product. The entire dissolution process is very sensitive to parameters such as temperature, pressure, chemical concentration and reaction time. The optimization of these parameters directly affects the recovery rate of alumina and product quality.

[0003] The production efficiency and quality of alumina depend largely on the accurate control of the dissolution process. Traditional methods rely on empirical operations and periodic laboratory tests, which are not only time-consuming and labor-intensive, but also difficult to achieve real-time optimization of process parameters. Intelligent detection equipment can monitor key parameters in the dissolution process, such as temperature, pressure, and chemical component concentration in real time, and predict and adjust these parameters through data analysis and machine learning models to ensure process stability and product consistency. In addition, the intelligent system can also identify potential problems and improvement points by analyzing historical data, significantly improving production efficiency and reducing energy and raw material consumption.

[0004] Currently, the monitoring of alumina dissolution process mostly relies on traditional sensors. These systems can provide basic process monitoring, but lack high automation and optimization capabilities. The main drawbacks of this traditional method are slow response time, inability to dynamically optimize based on real-time data, and poor adaptability to environmental changes and non-standard operating conditions.

[0005] The prior art still has certain defects in the intelligent detection setting of the alumina dissolution process. For example, the Chinese patent with application number CN202110679077.6 proposes an automatic control method for the alumina dissolution process. The method is based on the set value of the pre-established knowledge base of the alumina dissolution process, collects process and test data, and dynamically adjusts the set value through data backtracking and feedforward compensation. This method mainly uses historical and real-time data to optimize the setting of the alkali addition flow rate, thereby improving the dissolution efficiency and the qualified rate of the caustic ratio; however, this method adopts a rule-based control strategy, which may lack good flexibility and adaptability when dealing with complex or nonlinear changes. For example, the Chinese patent with application number 202011633801.3 proposes a heat balance control system and method for alumina dissolution. First, the temperature data of the last-stage casing heater is collected through a data acquisition unit, and then the multiple sets of production parameters stored in the knowledge base are used to analyze and fuzzy reason these data through the intelligent processing module to determine the new steam adjustment input set value of each stage casing heater. The operating system automatically adjusts the regulating valve on the steam branch according to these set values, and accurately controls the temperature of the heater in real time to optimize the heat balance and improve the dissolution rate and product quality of alumina. However, this method mainly relies on temperature control and steam regulation, and does not make full use of other relevant data, including chemical composition and pressure, which limits the depth and scope of optimization. At present, the existing technology needs to improve the integrity of data monitoring and the accuracy of parameter setting in the alumina dissolution process.

[0006] Therefore, an intelligent detection and setting system for alumina dissolution process is proposed. Summary of the invention

[0007] The purpose of the present invention is to provide an intelligent detection and setting system for aluminum oxide dissolution process. First, real-time data is obtained from the production line, including a first data set, a second data set and a third data set; then, three module historical data sets in the historical data are obtained and preprocessed to obtain three preprocessing sets; according to the three preprocessing sets, model training is performed respectively to obtain three parameter prediction models; further, the three parameter prediction models analyze the first data set, the second data set and the third data set respectively to obtain the first prediction result, the second prediction result and the third prediction result; finally, according to the three prediction results, the first parameter group, the second parameter group and the third parameter group in the dissolution process are set respectively.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] An intelligent detection and setting system for aluminum oxide dissolution process, comprising:

[0010] A data acquisition unit, used to acquire real-time data from the production line, including a first data set of the crushing working module, a second data set of the fusion working module, and a third data set of the temperature control working module;

[0011] A data storage unit is used to store all data collected from the production line, including real-time data and historical data; the historical data includes multi-source historical data of three working modules, forming three module historical data sets; the three working modules include the crushing working module, the fusion working module and the temperature control working module;

[0012] A model acquisition unit is used to acquire three parameter prediction models, including a first parameter prediction model, a second parameter prediction model and a third parameter prediction model; the model acquisition unit obtains three preprocessing sets by preprocessing the three module historical data sets respectively; and performs model training according to the three preprocessing sets respectively to obtain three parameter prediction models;

[0013] The data processing unit includes three processing modules, namely a first processing module, a second processing module and a third processing module; each of the processing modules corresponds to one of the parameter prediction models, and is used to analyze a data set of one of the working modules to obtain a prediction result;

[0014] Data communication unit, including industrial Ethernet, for transmitting data;

[0015] The parameter setting unit includes a PID control algorithm for setting a first parameter group, a second parameter group and a third parameter group in the dissolution process based on the prediction result of the data processing unit.

[0016] The first data set includes target particle size, raw material hardness, raw material weight, ball mill size, ball mill number, and raw material initial particle size grade.

[0017] The process of obtaining the initial particle size grade of the raw materials includes:

[0018] The particle size of the raw materials is measured using a laser diffraction particle size analyzer to obtain a particle size set;

[0019] Set a particle size threshold; when the particle size exceeds or is equal to the particle size threshold, it is marked as a large particle size;

[0020] When the particle size is smaller than the particle size threshold, it is marked as a small particle size;

[0021] The Malvern Panalytical software is used to calculate the proportion of the large particle size and the small particle size in the particle size set, respectively, to obtain the initial particle size grade of the raw material; the determination formula of the initial particle size grade of the raw material is:

[0022] ;

[0023] in, is the initial particle size grade of the raw material, is the large particle size, is the particle size collection;

[0024] The target particle size is preset; the process of obtaining the target particle size includes:

[0025] Collecting production data for each batch during production operation; the production data includes the particle size of alumina particles entering the nickel-lined tank, the hardness of the raw materials, the concentration of the alkaline solution, the amount of the alkaline solution, and the alumina content of the dissolved material in the nickel-lined tank;

[0026] Processing the production data: merging the production data whose values ​​of the raw material hardness, the alkali solution concentration and the alkali solution amount differ by no more than 10% to form a production data group;

[0027] Using regression analysis method to analyze the relationship between the particle size of the alumina particles in the production data set and the alumina content of the dissolved material in the nickel-lined tank, and determine the target particle size;

[0028] Establish a mechanism for continuously obtaining the target particle size and regularly review the setting of the target particle size;

[0029] The weight of the raw material is obtained by a weighing device; the hardness of the raw material is obtained by an ultrasonic hardness tester; the ball mill size and the ball mill quantity are preset values.

[0030] The second data set includes the aluminum oxide content, the first impurity content, the second impurity content, the third impurity content and the fourth impurity content in the raw material particles, the particle size of the raw material particles and the target purity of aluminum oxide.

[0031] The acquisition of the particle size of the raw material particles includes: using a laser diffraction particle size analyzer to measure the particle size of the raw material particles to obtain a particle size set; calculating the average value of the particle size in the particle size set as the particle size of the raw material particles; the target purity of alumina is a value preset according to production requirements;

[0032] The aluminum oxide content, the first impurity content, the second impurity content, the third impurity content and the fourth impurity content in the raw material particles are obtained by atomic absorption spectrometer; the first impurity is iron, the second impurity is silicate, the third impurity is calcium, and the fourth impurity is magnesium.

[0033] The third data set includes raw material particle size, raw material hardness, raw material weight, aluminum oxide mass concentration in the dissolution solution, sodium carbonate mass concentration in the dissolution solution, target mass concentration of the dissolution solution, and carbon-alkali ratio of the dissolution solution.

[0034] The mass concentration of aluminum oxide in the dissolution solution is obtained by monitoring the absorbance of aluminum oxide in the solution by a near-infrared spectrometer; the mass concentration of sodium carbonate in the dissolution solution is obtained by a near-infrared spectrometer; the target mass concentration of the dissolution solution is a value preset according to production requirements; the carbon-alkali ratio of the dissolution solution is the ratio of sodium carbonate to sodium hydroxide in the dissolution solution, which is obtained by an online ion chromatograph.

[0035] The first parameter group includes a crushing speed of a crushing device and a crushing time of a crushing device; the first processing module includes a first parameter prediction model for analyzing the first data set to obtain a first prediction result; and the first parameter group is set according to the first prediction result;

[0036] The first parameter prediction model is trained by the first module historical data set; each data in the first module historical data set includes multiple feature columns and multiple target columns; the multiple feature columns include target particle size, raw material hardness, raw material weight, ball mill size, ball mill quantity and raw material initial particle size; the multiple target columns include crushing speed of crushing equipment and crushing time of crushing equipment;

[0037] The step of performing the model training on the first parameter prediction model using the first module historical data set includes:

[0038] Data preprocessing: normalize all data to obtain the first preprocessing set;

[0039] Data set division: the first preprocessed set is divided into a training set and a test set in a ratio of 7:3;

[0040] Training model: inputting the training set into the first parameter prediction model until the model training stop condition is met to obtain an initial first parameter prediction model; the model training stop condition is that the loss of the first parameter prediction model reaches convergence;

[0041] Testing model: inputting the test set into the first parameter prediction model for testing, optimizing the first parameter prediction model according to the test result, and obtaining the optimal first parameter prediction model;

[0042] The first parameter prediction model includes a first preprocessing layer, a first feature analysis layer and a first parameter prediction layer; the first preprocessing layer is used to perform preprocessing operations on the first data set to obtain a first standard data set; the preprocessing operations include data cleaning, standardization, normalization and missing value processing; the first feature analysis layer includes an LSTM network structure, which is used to analyze the first standard data set to obtain a first feature vector; the first parameter prediction layer includes two output neurons, each of which corresponds to a prediction target; the prediction targets include the crushing speed of the crushing equipment and the crushing time of the crushing equipment; the first parameter prediction layer is used to predict the parameters in the first parameter group according to the first feature vector to obtain a first prediction result.

[0043] The second parameter group includes the amount of alkaline solution and the concentration of alkaline solution; the second processing module is used to analyze the second data set to obtain a second processing result; the second processing module includes a second parameter prediction model, which is used to analyze the second data set to obtain a second prediction result; the second parameter group is set according to the second prediction result;

[0044] The second parameter prediction model includes a second preprocessing layer, a second feature analysis layer, a second feature fusion layer and a second parameter prediction layer;

[0045] The second preprocessing layer is used to perform a preprocessing operation on the second data set to obtain a second standard data set; the preprocessing operation includes data cleaning, standardization, normalization and missing value processing;

[0046] The second feature analysis layer includes two LSTM network structures, which are used to analyze the second standard data set and the impurity information respectively to obtain a second feature vector and an impurity feature vector; the impurity information includes a first impurity content, a second impurity content, a third impurity content and a fourth impurity content;

[0047] The second feature fusion layer includes an impurity attention mechanism; the second feature fusion layer is used to concatenate the second feature vector and the impurity feature vector to obtain a comprehensive feature vector; the impurity features in the comprehensive feature vector are weighted by the impurity attention mechanism to obtain a weighted comprehensive feature vector;

[0048] The calculation formula of the impurity attention mechanism is:

[0049] ;

[0050] ;

[0051] in, is at the time step Attention score; is the parameter of the attention mechanism; is the hyperbolic tangent activation function; is the weight matrix; is the LSTM network at time step Output: is the weighted sum of all impurity impact scores; is the bias term; are network parameters; It is The content of impurities; For the The impact score of each impurity; is the activation function, which is used to ensure that the output is between 0 and 1;

[0052] The second parameter prediction layer includes two output neurons, each of which corresponds to a prediction target; the prediction targets include the amount of alkaline solution and the concentration of alkaline solution; the second parameter prediction layer is used to predict the parameters in the second parameter group according to the second eigenvector to obtain a second prediction result.

[0053] The third parameter group includes a heating rate, a target temperature and a holding time during the dissolution process; the third processing module is used to analyze the third data set to obtain a third processing result; the third processing module includes a third parameter prediction model, which is used to analyze the third data set to obtain a third prediction result;

[0054] Setting the third parameter group according to the third prediction result;

[0055] The third parameter prediction model is trained by the historical data set of the third module; each data in the historical data set of the third module includes multiple feature columns and multiple target columns; the multiple feature columns include raw material particle size, raw material hardness, raw material weight, aluminum oxide mass concentration in the dissolution solution, sodium carbonate mass concentration in the dissolution solution, target mass concentration of the dissolution solution and carbon-alkali ratio of the dissolution solution; the multiple target columns include heating rate, target temperature, holding time and energy consumption; the energy consumption is obtained by an electric energy meter;

[0056] The step of performing the model training on the third parameter prediction model through the third module historical data set comprises:

[0057] Data preprocessing: normalize all data to obtain the third preprocessing set;

[0058] Data set division: the third preprocessing set is divided into a training set and a test set in a ratio of 7:3;

[0059] Training model: input the training set into the third parameter prediction model until the model training stop condition is met to obtain an initial third parameter prediction model; the model training stop condition is that the loss of the third parameter prediction model reaches convergence; the training model includes a reward mechanism function for evaluating the benefit of the current operating state; the formula of the reward mechanism function is:

[0060] ;

[0061] ;

[0062] ;

[0063] ;

[0064] in, is the reward value at time point t; is the weight coefficient; is the attenuation coefficient; Indicates energy consumption; They represent the effects of heating rate, target temperature and holding time on total energy consumption; It indicates the extraction efficiency of aluminum oxide; is the current mass concentration of alumina in the dissolution solution; is the target alumina mass concentration; Composite functions representing raw material properties; It is a coefficient used to adjust the effect of heating rate, target temperature and holding time on energy consumption; are the coefficients for target particle size, raw material hardness and raw material weight, respectively, is the target particle size; is the hardness of the raw material; is the weight of raw materials;

[0065] Testing model: inputting the test set into the third parameter prediction model for testing, optimizing the third parameter prediction model according to the test result, and obtaining the optimal third parameter prediction model;

[0066] The third parameter prediction model includes a third preprocessing layer, a third feature analysis layer and a third parameter prediction layer;

[0067] The third preprocessing layer is used to perform a preprocessing operation on the third data set to obtain a third standard data set; the preprocessing operation includes data cleaning, standardization, normalization and missing value processing;

[0068] The third feature analysis layer includes an LSTM network structure, which is used to analyze the third standard data set to obtain a third feature vector;

[0069] The third parameter prediction layer includes three output neurons, each of which corresponds to a prediction target; the prediction targets include heating rate, target temperature and holding time; the third parameter prediction layer is used to predict the parameters in the third parameter group according to the third eigenvector to obtain a third prediction result.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] 1. The present invention realizes precise control of the relevant parameters of the crushing equipment by real-time monitoring of the target particle size, hardness, weight and particle size grade of the raw materials, as well as the ball mill size and number of the crushing equipment, and proposes a first parameter prediction model trained by the historical data set of the first module. The model ensures that the raw materials accurately reach the preset target particle size and optimizes the raw material preparation process. By classifying the initial particle size of the raw materials and combining historical data analysis to optimize the target particle size, the system not only improves the dissolution efficiency of alumina, but also ensures the consistency and quality control of the raw material processing, thereby improving the efficiency and economic benefits of the entire production process.

[0072] 2. The present invention proposes a second parameter prediction model obtained through training. The model is trained using the historical data set of the second module. It can analyze the alumina content in the raw materials and the content of various impurities in the raw materials, the particle size of the raw material particles and the target purity of alumina in real time, and accurately set the amount and concentration of the alkaline solution. In this way, the optimal ratio of chemical components in the dissolution solution can be ensured, and the dissolution efficiency of alumina can be optimized. In addition, the model integrates an impurity attention mechanism, which predicts the possible reaction of impurities with alkaline solution by weighted analysis of the effects of various impurities. This not only improves the extraction efficiency of alumina, but also helps to reduce chemical waste and control production costs, while maintaining environmental sustainability and the stability of the dissolution reaction.

[0073] 3. The present invention proposes a third parameter prediction model trained with the historical data set of the third module, and incorporates an innovative reward mechanism function in the training process to achieve a balance between efficiency and energy consumption in the process of optimizing alumina dissolution; the third parameter prediction model is used to analyze the particle size of raw materials, the mass concentration of alumina in the dissolution solution, the mass concentration of sodium carbonate in the dissolution solution, the target mass concentration of the dissolution solution, and the carbon-alkali ratio of the dissolution solution to obtain the pressure, heating rate, target temperature, and insulation time in the optimal dissolution process. The setting of the fourth parameter group helps to accurately control the dynamics of the chemical reaction and ensure that the chemical reaction is complete and efficient. This fine temperature control optimizes energy use, reduces energy consumption, and ensures the quality and consistency of the final product, thereby improving overall production efficiency and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 A flowchart of an implementation method of an intelligent detection and setting system for aluminum oxide dissolution process provided by an embodiment of the present invention;

[0075] Figure 2 A structural diagram of an intelligent detection and setting system for an aluminum oxide dissolution process provided by an embodiment of the present invention;

[0076] Figure 3 A process flow chart of an intelligent detection and setting system for an alumina dissolution process provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0077] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0078] The dissolution process of alumina is as follows: bauxite is crushed in a ball mill, and then the alumina in the ore is dissolved in an alkaline solution of a certain concentration in a hydrothermal reactor; specifically, the crushed ore and the alkaline solution are added to a nickel-lined tank in a certain proportion, the reactor is heated to a preset temperature, and then dissolved after a period of insulation. When the dissolution time is reached, the reactor is opened and the nickel-lined tank is taken out when the temperature is quickly cooled to a certain temperature with water.

[0079] In the process of alumina dissolution, the following parameters should be considered: particle size of raw material particles, alumina content and impurity content (including iron, silicate, calcium and magnesium), concentration and addition amount of alkaline solution, reaction temperature and pressure, dissolution time and mass concentration of alumina and sodium carbonate in the dissolution solution. These parameters directly affect the efficiency of the dissolution reaction and the quality of the final product, and must be accurately monitored and dynamically adjusted to optimize the entire production process.

[0080] In the present invention, the dissolution process of aluminum oxide is divided into three modules, namely, a crushing working module, a fusion working module and a temperature control working module; further, the present invention proposes three parameter groups, which respectively include important parameters in these three working modules.

[0081] There are many studies on intelligent detection and setting of aluminum oxide dissolution process in the prior art, but there are some defects. In order to solve the defects in the prior art, the purpose of the present invention is to provide an intelligent detection and setting system for aluminum oxide dissolution process. The specific implementation method flow chart and system structure diagram of the system are shown in FIG. Figure 1 and Figure 2 .

[0082] Embodiment 1

[0083] This example analyzes the dissolution process of the aluminum oxide sample.

[0084] The process flow chart of the intelligent detection and setting system for aluminum oxide dissolution process proposed in this embodiment is as follows: Figure 3 .

[0085] like Figure 3 As shown, this embodiment stores acquired real-time data, historical data and prediction models through a data storage unit; wherein the real-time data includes a first data set, a second data set and a third data set; the first data set includes target particle size, raw material hardness, raw material weight, ball milling size, ball milling quantity and raw material initial particle size grade; the second data set includes alumina content, a first impurity content, a second impurity content, a third impurity content and a fourth impurity content in raw material particles, as well as raw material particle size and alumina target purity; the third data set includes raw material particle size, raw material hardness, raw material weight, alumina mass concentration in the dissolution solution, sodium carbonate mass concentration in the dissolution solution, target mass concentration of the dissolution solution and carbon-alkali ratio of the dissolution solution; the parameter prediction model includes a first parameter prediction model, a second parameter prediction model and a third parameter prediction model; the parameter prediction model is trained by historical data, and the real-time data is analyzed by the trained parameter prediction model to obtain the corresponding prediction result; wherein the first parameter prediction model is used to analyze the first data set to obtain a first prediction result; the second parameter prediction model is used to analyze the second data set to obtain a second prediction result; the third parameter prediction model is used to analyze the third data set to obtain a third prediction result.

[0086] The parameter setting unit sets relevant parameters in a specific working module through the first prediction result, the second prediction result and the third prediction result obtained by the three parameter prediction models;

[0087] Specifically, first, the raw material bauxite is put into the crushing equipment, and the first data set obtained is analyzed by the first parameter prediction model to obtain a first prediction result. The parameter setting unit sets the crushing equipment speed and time in the crushing working module according to the first prediction result to obtain the crushed raw material; further, the crushed raw material is put into the reactor, and the second data set obtained is analyzed by the second parameter prediction model to obtain a second prediction result. The parameter setting unit sets the amount and concentration of the alkaline solution in the fusion working module according to the second prediction result; finally, the crushed raw material fused with the alkaline solution enters the heating equipment, and the third data set obtained is analyzed by the third parameter prediction model to obtain a third prediction result. The parameter setting unit sets the heating rate, target temperature and insulation time in the temperature control working module according to the third prediction result to obtain the dissolved material.

[0088] The system proposed in this embodiment will be described in detail below in conjunction with a specific implementation method flow chart and a system structure diagram.

[0089] Reference Figure 1 S10 in which S10 is applied to a data acquisition unit of an intelligent detection setting system for an alumina dissolution process, and is used to acquire real-time data from a production line, including a first data set of a crushing working module, a second data set of a fusion working module, and a third data set of a temperature control working module;

[0090] Furthermore, the first data set includes target particle size, raw material hardness, raw material weight, ball milling size, ball milling number and raw material initial particle size grade.

[0091] In this embodiment, the raw material is mined bauxite.

[0092] When setting the parameters of the crushing equipment for crushing bauxite, the target particle size determines the particle size standard that the crushing equipment needs to achieve, which directly affects the rate and completeness of the subsequent chemical reaction. The appropriate particle size can increase the reaction surface area, speed up the reaction speed, and improve the extraction efficiency of alumina. Raw materials with high hardness require more energy to crush to the target particle size, which affects the wear and energy consumption of the crushing equipment. Reasonable adjustment of crushing parameters to adapt to raw materials of different hardness can optimize energy consumption and extend equipment life. The weight of raw materials is directly related to the amount of material processed in each batch, which affects the load and operating efficiency of the crusher. Accurately control the input amount of raw materials to ensure that the equipment operates at the optimal load and avoid overload or inefficiency. Understanding the initial particle size grade distribution of raw materials helps to adjust the settings of the crushing equipment to adapt to raw materials of different particle sizes. This can ensure a more uniform crushing effect, reduce over-crushing or insufficient crushing, and further optimize energy consumption and processing efficiency;

[0093] The size of the ball mill and the number of balls are key factors in determining the parameters of the crushing equipment. Specifically, a larger ball mill or a larger number of balls can increase the contact area of ​​the grinding media, thereby improving the crushing efficiency and speeding up the crushing of materials, but the crushing equipment may need to start at a higher speed.

[0094] By accurately monitoring the target particle size, raw material hardness, raw material weight, ball mill size, ball mill number and raw material initial particle size grade, the present invention can adjust the operating parameters of the crushing equipment, including crushing speed and time. These parameters directly affect the processing quality and efficiency of the raw materials, so selecting these data for analysis can ensure that the raw materials are finely processed according to the predetermined standards, maximize the crushing efficiency and particle size uniformity, thereby optimizing the downstream chemical reaction conditions and improving the extraction rate of alumina, achieving maximum cost-effectiveness and environmental sustainability of the production process.

[0095] Furthermore, the process of obtaining the initial particle size grade of the raw material includes:

[0096] The particle size of the raw materials is measured using a laser diffraction particle size analyzer to obtain a particle size set;

[0097] Set a particle size threshold; when the particle size exceeds or is equal to the particle size threshold, it is marked as a large particle size;

[0098] When the particle size is smaller than the particle size threshold, it is marked as a small particle size;

[0099] The Malvern Panalytical software is used to calculate the proportion of the large particle size and the small particle size in the particle size set, respectively, to obtain the initial particle size grade of the raw material; the determination formula of the initial particle size grade of the raw material is:

[0100] ;

[0101] in, is the initial particle size grade of the raw material, is the large particle size, is the particle size collection;

[0102] According to the national standard particle size requirements for bauxite ore, in this embodiment, the particle size threshold for distinguishing large particle size from small particle size is set to 100 mm;

[0103] In this embodiment, the final calculation result of the Malvern Panalytical software is: the large particle size accounts for 62.74% of the particle size set, and the small particle size accounts for 37.26% of the particle size set; so in this embodiment, the initial particle size grade of the raw material is 3;

[0104] Furthermore, the target particle size is preset; and the process of obtaining the target particle size includes:

[0105] Collect production data for each batch during production operation;

[0106] The production data include the particle size of the alumina particles entering the nickel-lined tank, the hardness of the raw materials, the concentration of the alkaline solution, the amount of the alkaline solution and the alumina content of the dissolved matter in the nickel-lined tank;

[0107] Processing the production data: merging the production data whose values ​​of the raw material hardness, the alkali solution concentration and the alkali solution amount differ by no more than 10% to form a production data group;

[0108] Using regression analysis method to analyze the relationship between the particle size of the alumina particles in the production data set and the alumina content of the dissolved material in the nickel-lined tank, and determine the target particle size;

[0109] Establish a mechanism for continuously obtaining the target particle size and regularly review the setting of the target particle size;

[0110] Taking the concentration of alkaline solution as an example, in this embodiment, the determination method for determining whether the concentration of alkaline solution in two pieces of production data does not exceed 10% is:

[0111] beg ,in is the concentration of the alkaline solution in a piece of production data, is the concentration of the alkaline solution in another production data, and , is the ratio of the alkaline solution concentration of the two production data. When the concentration of the alkaline solution in the two production data does not exceed 10%;

[0112] In this embodiment, some production data of alumina sample 1 is shown in Table 1:

[0113] Table 1 Partial production data of alumina sample 1

[0114] Serial number Particle size of alumina particles entering the nickel-lined tank (mm) Raw material hardness (Mohs) Alkali solution concentration (%) Alkaline solution volume (liters) Alumina content of dissolved substances in nickel-lined tanks (%) 1 224.52 4.85 33.65 1320 51.42 2 457.77 9.43 49.2 1085 43.17 3 122.87 5.76 24.81 945 33.38 4 114.35 2.12 30.12 547 44.66 5 396.45 7.41 37.24 1418 39.62 6 370.09 8.55 47.36 1023 56.45 7 285.65 5.43 36.95 1362 52.62 ... ... ... ... ... ...

[0115] According to the partial production data shown in Table 1, the production group data obtained is shown in Table 2:

[0116] Table 2 Production group data table

[0117]

[0118] As shown in Table 1, in the data with serial numbers 2 and 6, the hardness of the raw materials differs by 9.33%, the amount of the alkaline solution differs by 3.74%, and the alumina content of the dissolved substances in the nickel-lined tank differs by 5.71%, which is consistent with the fact that the three values ​​of the hardness of the raw materials, the concentration of the alkaline solution and the amount of the alkaline solution do not differ by more than 10%. Therefore, the production data with serial numbers 2 and 6 are combined into one data group, namely, the data group with serial number 1 in Table 2;

[0119] In this embodiment, the particle size of the alumina particles in the production data set is analyzed using a regression analysis method. Alumina content of dissolved substances in nickel-lined tanks The relationship between them is as follows:

[0120] ;

[0121] in, is the intercept term, is the regression coefficient, is the error term;

[0122] Calculate the target particle size by regression equation The formula is:

[0123] ;

[0124] in, Obtained by analyzing the production data set;

[0125] Furthermore, the weight of the raw material is obtained by a weighing device; the hardness of the raw material is obtained by an ultrasonic hardness tester; and the ball mill size and the ball mill quantity are preset values.

[0126] In the ball mill used in this embodiment, the number of balls is 93 and the size is 25 mm.

[0127] Through fine raw material particle size classification, this system divides the comprehensive particle size of raw materials into three levels, so as to achieve precise adjustment of crushing equipment parameters. For raw materials with particle size grade 3, it indicates that it contains more large-size particles, and the operating intensity of the crushing equipment may need to be increased; while for raw materials with particle size grade 1, since it contains more small-size particles, the operating parameters of the crushing equipment may be relatively reduced. In addition, the target particle size determined by historical data analysis helps to optimize the dissolution process of raw materials, while the weight and hardness of raw materials and the size and number of ball mills in the crushing equipment directly affect the parameter setting of the crushing equipment, making the entire processing process more efficient and precise. This method not only improves processing efficiency, but also ensures processing quality, which meets the production needs of energy conservation and emission reduction.

[0128] Furthermore, the second data set includes the aluminum oxide content, the first impurity content, the second impurity content, the third impurity content and the fourth impurity content in the raw material particles, as well as the particle size of the raw material particles and the target purity of aluminum oxide.

[0129] In the fusion work module, the alumina content in the raw materials is a key factor affecting the dissolution efficiency because it determines how much alumina can be extracted from the raw materials. A higher alumina content indicates a greater potential yield, which requires adjusting the dissolution conditions to optimize the dissolution process. At the same time, impurities in the raw materials may react with the added alkaline solution to form complexes or produce scaling, so understanding the various impurity contents is crucial to adjusting the chemical reaction conditions to maximize the efficiency of alumina extraction, reduce energy consumption, and reduce the consumption of chemical reagents. In addition, the particle size of the raw materials directly affects the dissolution rate and efficiency. Although smaller particle sizes help accelerate chemical reactions due to their larger surface area to volume ratio, they may also cause incomplete dissolution or subsequent filtration difficulties. Therefore, particle size control is a factor that must be considered in the dissolution process. Finally, setting the target purity of alumina is key to ensuring product quality and meeting the needs of specific downstream applications. By real-time monitoring and adjusting production parameters, the final product can be ensured to meet quality standards, thereby avoiding additional reprocessing costs. This comprehensive consideration strategy not only improves production efficiency, but also optimizes resource utilization, ensuring the economy and environmental friendliness of the process.

[0130] Furthermore, the acquisition of the particle size of the raw material particles includes: using a laser diffraction particle size analyzer to measure the particle size of the raw material particles to obtain a particle size set; calculating the average value of the particle size in the particle size set as the particle size of the raw material particles; the target purity of alumina is a value preset according to production requirements;

[0131] The aluminum oxide content, the first impurity content, the second impurity content, the third impurity content and the fourth impurity content in the raw material particles are obtained by atomic absorption spectrometer; the first impurity is iron, the second impurity is silicate, the third impurity is calcium, and the fourth impurity is magnesium.

[0132] In this embodiment, the average value of the particle size of the raw material is obtained by using a laser diffraction particle size analyzer, which helps to adjust the dissolution process and ensure a more uniform chemical reaction rate, thereby improving the extraction efficiency of aluminum oxide. At the same time, the impurity content of iron, silicate, calcium, magnesium, etc. accurately measured by an atomic absorption spectrometer allows the chemical dosage in the dissolution process to be adjusted more accurately, reducing unnecessary consumption and possible environmental pollution in the chemical reaction; in the dissolution process of aluminum oxide, the impurities present include iron, silicate, calcium and magnesium, which will react differently with the alkaline solution, thereby affecting the efficiency and cost of the entire process; iron and silicate may form precipitation, resulting in equipment blockage and reduced extraction efficiency; precipitation generated by calcium and magnesium consumes too much alkali, reduces the alkali efficiency of the system, and increases additional processing costs; correct management of the content of these impurities is the key to optimizing the production process, reducing costs and improving the extraction efficiency of aluminum oxide. In addition, the set target purity of aluminum oxide is preset according to production requirements, ensuring that the product quality meets the strict requirements of downstream processing, and effectively avoiding quality problems and cost increases in subsequent processing. These measures not only optimize the production process as a whole, but also bring the dual advantages of economic benefits and environmental sustainability to enterprises.

[0133] Furthermore, the third data set includes raw material particle size, raw material hardness, raw material weight, aluminum oxide mass concentration in the dissolution solution, sodium carbonate mass concentration in the dissolution solution, target mass concentration of the dissolution solution, and carbon-alkali ratio of the dissolution solution.

[0134] The particle size, hardness and weight of the raw materials directly affect the reaction rate and heat conduction efficiency of the raw materials during the heating process, thereby affecting the required heating rate and the target temperature achieved. At the same time, the chemical composition of the dissolution solution, especially the mass concentration of alumina and the concentration of sodium carbonate, determines the length of time to maintain the target temperature to ensure the completeness of the reaction. The target mass concentration of the dissolution solution sets the final state that the dissolution reaction should reach, affecting the setting of the heating rate and the holding time to ensure the maximum dissolution efficiency. This fine parameter control not only ensures the efficiency of the process, but also effectively saves energy, optimizes the reaction conditions, and ensures the efficiency and environmental friendliness of the alumina extraction process.

[0135] Furthermore, the mass concentration of aluminum oxide in the dissolution solution is obtained by monitoring the absorbance of aluminum oxide in the solution by a near-infrared spectrometer; the mass concentration of sodium carbonate in the dissolution solution is obtained by a near-infrared spectrometer; the target mass concentration of the dissolution solution is a value preset according to production requirements; the carbon-alkali ratio of the dissolution solution is the ratio of sodium carbonate to sodium hydroxide in the dissolution solution, which is obtained by an online ion chromatograph.

[0136] By accurately monitoring the mass concentration of alumina and sodium carbonate in the dissolution solution, the system can better control the heating rate, target temperature and holding time. These parameters directly affect the thermodynamic and kinetic conditions of the dissolution process, thereby affecting the dissolution efficiency and energy consumption of alumina. Setting the target mass concentration of the dissolution solution and optimizing the carbon-alkali ratio help to develop a more effective heating strategy, ensuring that while achieving the ideal dissolution rate and final product quality, energy consumption can be minimized and the economy of the process can be improved. In this way, energy consumption can be finely controlled while maintaining reaction efficiency, optimizing energy management in the production process and ensuring environmental sustainability.

[0137] Reference Figure 1 S20 in the embodiment, S20 is applied to a model acquisition unit of an intelligent detection setting system for alumina dissolution process, and is used to acquire three parameter prediction models, including a first parameter prediction model, a second parameter prediction model, and a third parameter prediction model; the model acquisition unit pre-processes the three module historical data sets respectively to obtain three pre-processing sets; and performs model training respectively according to the three pre-processing sets to obtain three parameter prediction models;

[0138] Taking the first parameter prediction model as an example, the first parameter prediction model is trained by the first module historical data set; each data in the first module historical data set includes multiple feature columns and multiple target columns; the multiple feature columns include target particle size, raw material hardness, raw material weight, ball mill size, ball mill quantity and raw material initial particle size; the multiple target columns include crushing speed of crushing equipment and crushing time of crushing equipment;

[0139] The step of performing the model training on the first parameter prediction model using the first module historical data set includes:

[0140] Data preprocessing: normalize all data to obtain the first preprocessing set;

[0141] Data set division: the first preprocessed set is divided into a training set and a test set in a ratio of 7:3;

[0142] Training model: inputting the training set into the first parameter prediction model until the model training stop condition is met to obtain an initial first parameter prediction model; the model training stop condition is that the loss of the first parameter prediction model reaches convergence;

[0143] Testing model: inputting the test set into the first parameter prediction model for testing, optimizing the first parameter prediction model according to the test result, and obtaining the optimal first parameter prediction model;

[0144] Furthermore, the intelligent detection and setting system for alumina dissolution process also includes a data storage unit; the data storage unit is used to store all data collected from the production line, including real-time data and historical data; the historical data includes multi-source historical data of three working modules, forming three module historical data sets; the three working modules include the crushing working module, the fusion working module and the temperature control working module;

[0145] Reference Figure 1 S30 in the embodiment, S30 is applied to a data processing unit of an intelligent detection setting system for alumina dissolution process, comprising three processing modules, namely a first processing module, a second processing module and a third processing module; each of the processing modules corresponds to one of the parameter prediction models, and is used to analyze a data set of one of the working modules to obtain a prediction result;

[0146] Reference Figure 1 S40 in the data processing unit is applied to a parameter setting unit of an intelligent detection and setting system for an aluminum oxide dissolution process, and includes a PID control algorithm for setting a first parameter group, a second parameter group, and a third parameter group in the dissolution process based on the prediction result of the data processing unit.

[0147] Furthermore, the first parameter group includes a crushing speed of a crushing device and a crushing time of a crushing device; the first processing module includes a first parameter prediction model, which is used to analyze the first data set to obtain a first prediction result; and the first parameter group is set according to the first prediction result;

[0148] In this embodiment, the crushing equipment is a ball mill; the grinding balls are steel balls;

[0149] The basic working principle of ball mill is as follows:

[0150] Filling materials and grinding balls: The ball mill is filled with a certain amount of grinding balls, which are usually made of steel balls, ceramic balls or wear-resistant alloy balls.

[0151] Rotation: The ball mill is driven by a motor and a transmission mechanism to rotate the mill cylinder on a horizontal axis. This rotation causes the grinding balls and materials in the cylinder to be lifted to a certain height along with the cylinder, and then fall due to gravity.

[0152] Impact and grinding: When the grinding balls fall freely, they will have a strong impact and grinding effect on the materials in the cylinder. The materials are gradually crushed in this continuous impact, extrusion and grinding.

[0153] Material refinement: After a period of grinding, the material is ground to the required particle size and then sent to the next process.

[0154] Further, the first parameter prediction model includes a first preprocessing layer, a first feature analysis layer and a first parameter prediction layer;

[0155] The first preprocessing layer is used to perform a preprocessing operation on the first data set to obtain a first standard data set; the preprocessing operation includes data cleaning, standardization, normalization and missing value processing;

[0156] The first feature analysis layer includes an LSTM network structure, which is used to analyze the first standard data set to obtain a first feature vector;

[0157] The first parameter prediction layer includes two output neurons, each of which corresponds to a prediction target; the prediction targets include the crushing speed of the crushing equipment and the crushing time of the crushing equipment; the first parameter prediction layer is used to predict the parameters in the first parameter group according to the first feature vector to obtain a first prediction result.

[0158] In this embodiment, by applying the LSTM network structure, the operating parameters of the ball mill, including the crushing speed and crushing time, are intelligently predicted and adjusted, which greatly improves the efficiency and accuracy of the crushing process. By real-time monitoring and data analysis of raw material characteristics, including initial particle size grade, target particle size, hardness and weight, the system can dynamically adjust the operating parameters of the ball mill according to real-time data to ensure that the material achieves the optimal particle size distribution, thereby optimizing the effect of downstream processing steps. This fine control not only improves the processing efficiency of raw materials, but also further improves the economy and environmental sustainability of the entire production process by reducing energy waste and improving production quality.

[0159] Further, the second parameter group includes the amount of alkaline solution and the concentration of alkaline solution; the second processing module is used to analyze the second data set to obtain a second processing result; the second processing module includes a second parameter prediction model, which is used to analyze the second data set to obtain a second prediction result; the second parameter group is set according to the second prediction result;

[0160] The second parameter prediction model includes a second preprocessing layer, a second feature analysis layer, a second feature fusion layer and a second parameter prediction layer;

[0161] The second preprocessing layer is used to perform a preprocessing operation on the second data set to obtain a second standard data set; the preprocessing operation includes data cleaning, standardization, normalization and missing value processing;

[0162] The second feature analysis layer includes two LSTM network structures, which are used to analyze the second standard data set and the impurity information respectively to obtain a second feature vector and an impurity feature vector; the impurity information includes a first impurity content, a second impurity content, a third impurity content and a fourth impurity content;

[0163] The second feature fusion layer includes an impurity attention mechanism; the second feature fusion layer is used to concatenate the second feature vector and the impurity feature vector to obtain a comprehensive feature vector; the impurity features in the comprehensive feature vector are weighted by the impurity attention mechanism to obtain a weighted comprehensive feature vector;

[0164] The calculation formula of the impurity attention mechanism is:

[0165] ;

[0166] ;

[0167] in, is at the time step The attention score of , which is a scalar value that calculates the importance of the relative contribution of that time step to the final output; is the parameter of the attention mechanism, which is used to The value transformed by the function is converted into a single attention score; is the hyperbolic tangent activation function, which helps compress the input into the range of [−1,1], helps deal with the problem of activation values ​​being too large or too small, and makes the gradient more stable; is the weight matrix; is the LSTM network at time step The output of , the hidden state, contains information about the sequence observed so far; is the weighted sum of all impurity impact scores, where Is the connection impurity affecting the score To the weight matrix processed by the impurity attention mechanism, It is The impact score of the impurity reflects the effect of the impurity on the reaction of the alkaline solution, that is, how it affects the consumption of the alkaline solution and the precipitation produced; is a bias term used in the calculation of the attention mechanism to ensure that there is a baseline output even when all input data are zero; are network parameters; It is The content of impurities; is the activation function, which is used to ensure that the output is between 0 and 1;

[0168] The second parameter prediction layer includes two output neurons, each of which corresponds to a prediction target; the prediction targets include the amount of alkaline solution and the concentration of alkaline solution; the second parameter prediction layer is used to predict the parameters in the second parameter group according to the second eigenvector to obtain a second prediction result.

[0169] In this embodiment, a second parameter prediction model is proposed, which significantly optimizes the dissolution efficiency of alumina by accurately predicting the parameters in the second parameter group. The model introduces an impurity attention mechanism, which specifically analyzes the impurity components in the dissolution process. The mechanism can distinguish the specific effects of the four impurities of iron, silicate, calcium and magnesium in the raw materials and the reaction with the alkaline solution, and adjust the amount and concentration of the alkaline solution by weighted comprehensive feature vectors. This not only improves the efficiency of extracting alumina from raw materials, but also reduces ineffective chemical consumption and energy waste by optimizing chemical reaction conditions, thereby achieving a dual improvement in environmental sustainability and economic benefits of the production process. This intelligent control mechanism ensures the stability of the dissolution process and the quality of the final product, demonstrating a high degree of innovation and practical value.

[0170] The third parameter group includes a heating rate, a target temperature and a holding time during the dissolution process; the third processing module is used to analyze the third data set to obtain a third processing result; the third processing module includes a third parameter prediction model, which is used to analyze the third data set to obtain a third prediction result; and the third parameter group is set according to the third prediction result;

[0171] The third parameter prediction model is trained by the historical data set of the third module; each data in the historical data set of the third module includes multiple feature columns and multiple target columns; the multiple feature columns include raw material particle size, raw material hardness, raw material weight, aluminum oxide mass concentration in the dissolution solution, sodium carbonate mass concentration in the dissolution solution, target mass concentration of the dissolution solution and carbon-alkali ratio of the dissolution solution; the multiple target columns include heating rate, target temperature, holding time and energy consumption; the energy consumption is obtained by an electric energy meter;

[0172] The step of performing the model training on the third parameter prediction model through the third module historical data set comprises:

[0173] Data preprocessing: normalize all data to obtain the third preprocessing set;

[0174] Data set division: the third preprocessing set is divided into a training set and a test set in a ratio of 7:3;

[0175] Training model: input the training set into the third parameter prediction model until the model training stop condition is met to obtain an initial third parameter prediction model; the model training stop condition is that the loss of the third parameter prediction model reaches convergence; the training model includes a reward mechanism function for evaluating the benefit of the current operating state; the formula of the reward mechanism function is:

[0176] ;

[0177] ;

[0178] ;

[0179] ;

[0180] in, is the reward value at time point t; is the weight coefficient; is the attenuation coefficient; Indicates energy consumption; They represent the effects of heating rate, target temperature and holding time on total energy consumption; It indicates the extraction efficiency of aluminum oxide; is the current mass concentration of alumina in the dissolution solution; is the target alumina mass concentration; Composite functions representing raw material properties; It is a coefficient used to adjust the effect of heating rate, target temperature and holding time on energy consumption; are the coefficients for target particle size, raw material hardness and raw material weight, respectively, is the target particle size; is the hardness of the raw material; is the weight of raw materials;

[0181] Testing model: inputting the test set into the third parameter prediction model for testing, optimizing the third parameter prediction model according to the test result, and obtaining the optimal third parameter prediction model;

[0182] The third parameter prediction model includes a third preprocessing layer, a third feature analysis layer and a third parameter prediction layer;

[0183] The third preprocessing layer is used to perform a preprocessing operation on the third data set to obtain a third standard data set; the preprocessing operation includes data cleaning, standardization, normalization and missing value processing;

[0184] The third feature analysis layer includes an LSTM network structure, which is used to analyze the third standard data set to obtain a third feature vector;

[0185] The third parameter prediction layer includes three output neurons, each of which corresponds to a prediction target; the prediction targets include heating rate, target temperature and holding time; the third parameter prediction layer is used to predict the parameters in the third parameter group according to the third eigenvector to obtain a third prediction result.

[0186] In this embodiment, the effectiveness of model training and the accuracy of operation are significantly improved by introducing an innovative reward mechanism function during the training process of the third parameter prediction model. The reward mechanism combines the relationship between the extraction efficiency and energy consumption of aluminum oxide, and effectively balances production efficiency and energy consumption. During the model training stage, this mechanism can adjust the model weights in real time to ensure that the settings of the heating rate, target temperature and holding time can achieve the relatively optimal aluminum oxide extraction efficiency while reducing energy consumption, thereby improving the economy and environmental sustainability of the entire dissolution process.

[0187] Furthermore, the intelligent detection and setting system for alumina dissolution process also includes a data communication unit; the data communication unit includes industrial Ethernet for transmitting data.

[0188] An intelligent detection and setting system for an alumina dissolution process proposed in this embodiment integrates comprehensive data acquisition, storage, processing and parameter setting units to improve production efficiency and ensure optimization of the production process.

[0189] The system divides the dissolution process into three modules, namely the crushing working module, the fusion working module and the temperature control working module. By collecting the key data of these three modules from the production line in real time, the system can dynamically monitor and adjust the production conditions to achieve precise process control. The data storage unit is set up to store not only real-time data, but also a large amount of historical data resources, providing a basis for deep learning and model optimization. The data of each working module is analyzed by a special parameter prediction model, which is pre-processed and trained based on historical data sets to ensure accurate prediction of future production status. In addition, the high-speed data communication of industrial Ethernet ensures the real-time transmission of information flow, while the advanced PID control algorithm allows the real-time adjustment of production parameters according to the prediction results, optimizes the chemical reaction conditions, reduces energy consumption, and improves product quality and production efficiency. The comprehensive application of this system not only improves the economy and sustainability of the alumina extraction process, but also significantly enhances the factory's adaptability and management efficiency to complex production environments.

[0190] Embodiment 2

[0191] This embodiment is based on the description of an intelligent detection and setting system for aluminum oxide dissolution process in the first embodiment, and analyzes the dissolution process of aluminum oxide sample 2.

[0192] In this embodiment, the final calculation result of the Malvern Panalytical software is: the large particle size accounts for 43.75% of the particle size set, and the small particle size accounts for 56.25% of the particle size set; so in this embodiment, the initial particle size grade of the raw material is 2;

[0193] In this embodiment, some production data of alumina sample 2 is shown in Table 3:

[0194] Table 3 Partial production data of alumina sample 2

[0195] Serial number Particle size of alumina particles entering the nickel-lined tank (mm) Raw material hardness (Mohs) Alkali solution concentration (%) Alkaline solution volume (liters) Alumina content of dissolved substances in nickel-lined tanks (%) 1 479.36 5.98 30.82 986 37.65 2 314.07 4.06 48.99 746 35.41 3 345.63 6.42 28.35 1040 39.52 4 499.02 1.02 22.44 825 52.74 ... ... ... ... ... ...

[0196] According to the partial production data shown in Table 3, the production group data obtained is shown in Table 4:

[0197] Table 4 Production group data table

[0198]

[0199] As shown in Table 3, in the data with serial numbers 1 and 3, the hardness of the raw materials differs by 6.85%, the amount of the alkaline solution differs by 8.71%, and the alumina content of the dissolved substances in the nickel-lined tank differs by 5.19%, which is consistent with the fact that the three values ​​of the hardness of the raw materials, the concentration of the alkaline solution and the amount of the alkaline solution do not differ by more than 10%. Therefore, the production data with serial numbers 1 and 3 are combined into one data group, namely, the data group with serial number 1 in Table 4;

[0200] In the ball mill used in this embodiment, the number of balls is 73 and the size is 25 mm.

[0201] In this embodiment, taking the second parameter prediction model as an example, the second parameter prediction model is trained by the second module historical data set; each data in the second module historical data set includes multiple feature columns and multiple target columns; the multiple feature columns include the aluminum oxide content, the first impurity content, the second impurity content, the third impurity content and the fourth impurity content in the raw material particles, the particle size of the raw material particles and the target purity of aluminum oxide; the multiple target columns include the amount of alkaline solution and the concentration of the alkaline solution;

[0202] The step of performing the model training on the second parameter prediction model using the second module historical data set comprises:

[0203] Data preprocessing: normalize all data to obtain the first preprocessing set;

[0204] Data set division: the first preprocessed set is divided into a training set and a test set in a ratio of 7:3;

[0205] Training model: inputting the training set into the second parameter prediction model until the model training stop condition is met to obtain an initial second parameter prediction model; the model training stop condition is that the loss of the second parameter prediction model reaches convergence;

[0206] Testing model: inputting the test set into the second parameter prediction model for testing, optimizing the first parameter prediction model according to the test result, and obtaining the optimal second parameter prediction model;

[0207] In order to better evaluate the performance and generalization ability of the first parameter prediction model, a cross-validation experiment was performed in this embodiment. The experimental results of the cross-validation are shown in Table 5.

[0208] Table 5 Cross-validation experiment results

[0209]

[0210] The specific steps of the cross-validation experiment include: dividing the data set into 5 non-overlapping subsets, one of which is used as a validation set and the other four subsets are used as training sets for the experiment; conducting 5 experiments of EM1, EM2, EM3, EM4 and EM5 respectively, and obtaining the final evaluation index by calculating the average value of the results of EM1, EM2, EM3, EM4 and EM5.

[0211] As shown in Table 5, the first parameter prediction model showed high precision (P%), recall (R%), and F1 score (F1%) in each experiment;

[0212] Although 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 the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent detection and setting system for aluminum oxide dissolution process, characterized in that: include: A data acquisition unit, used to acquire real-time data from the production line, including a first data set of the crushing working module, a second data set of the fusion working module, and a third data set of the temperature control working module; The first data set includes target particle size, raw material hardness, raw material weight, ball milling size, ball milling number and raw material initial particle size grade; the second data set includes alumina content, first impurity content, second impurity content, third impurity content and fourth impurity content in raw material particles, raw material particle size and alumina target purity; the first impurity is iron, the second impurity is silicate, the third impurity is calcium, and the fourth impurity is magnesium; The third data set includes the particle size of raw material particles, the hardness of raw material, the weight of raw material, the mass concentration of aluminum oxide in the dissolution solution, the mass concentration of sodium carbonate in the dissolution solution, the target mass concentration of the dissolution solution, and the carbon-alkali ratio of the dissolution solution; A data storage unit is used to store all data collected from the production line, including real-time data and historical data; the historical data includes multi-source historical data of three working modules, forming three module historical data sets; the three working modules include the crushing working module, the fusion working module and the temperature control working module; A model acquisition unit is used to acquire three parameter prediction models, including a first parameter prediction model, a second parameter prediction model and a third parameter prediction model; the model acquisition unit obtains three preprocessing sets by preprocessing the three module historical data sets respectively; and performs model training according to the three preprocessing sets respectively to obtain three parameter prediction models; The first parameter prediction model is trained by the first module historical data set; each data in the first module historical data set includes multiple feature columns and multiple target columns; the multiple feature columns include target particle size, raw material hardness, raw material weight, ball mill size, ball mill quantity and raw material initial particle size; the multiple target columns include crushing speed of crushing equipment and crushing time of crushing equipment; The step of performing the model training on the first parameter prediction model using the first module historical data set includes: Data preprocessing: normalize all data to obtain the first preprocessing set; Data set division: the first preprocessed set is divided into a training set and a test set in a ratio of 7:3; Training model: inputting the training set into the first parameter prediction model until the model training stop condition is met to obtain an initial first parameter prediction model; the model training stop condition is that the loss of the first parameter prediction model reaches convergence; Testing model: inputting the test set into the first parameter prediction model for testing, optimizing the first parameter prediction model according to the test result, and obtaining the optimal first parameter prediction model; The second parameter prediction model is trained by the second module historical data set; each data in the second module historical data set includes multiple feature columns and multiple target columns; the multiple feature columns include the aluminum oxide content, the first impurity content, the second impurity content, the third impurity content and the fourth impurity content in the raw material particles, the particle size of the raw material particles and the target purity of aluminum oxide; the multiple target columns include the amount of alkaline solution and the concentration of the alkaline solution; The step of performing the model training on the second parameter prediction model using the second module historical data set comprises: Data preprocessing: normalize all data to obtain the second preprocessing set; Data set division: the second preprocessed set is divided into a training set and a test set in a ratio of 7:3; Training model: inputting the training set into the second parameter prediction model until the model training stop condition is met to obtain an initial second parameter prediction model; the model training stop condition is that the loss of the second parameter prediction model reaches convergence; Testing model: inputting the test set into the second parameter prediction model for testing, optimizing the second parameter prediction model according to the test result, and obtaining the optimal second parameter prediction model; The third parameter prediction model is trained by the historical data set of the third module; each data in the historical data set of the third module includes multiple feature columns and multiple target columns; the multiple feature columns include raw material particle size, raw material hardness, raw material weight, aluminum oxide mass concentration in the dissolution solution, sodium carbonate mass concentration in the dissolution solution, target mass concentration of the dissolution solution and carbon-alkali ratio of the dissolution solution; the multiple target columns include heating rate, target temperature, holding time and energy consumption; the energy consumption is obtained by an electric energy meter; The step of performing the model training on the third parameter prediction model through the third module historical data set comprises: Data preprocessing: normalize all data to obtain the third preprocessing set; Data set division: the third preprocessing set is divided into a training set and a test set in a ratio of 7:3; Training model: input the training set into the third parameter prediction model until the model training stop condition is met to obtain an initial third parameter prediction model; the model training stop condition is that the loss of the third parameter prediction model reaches convergence; the training model includes a reward mechanism function for evaluating the benefit of the current operating state; the formula of the reward mechanism function is: ； ； ； ； in, is the reward value at time point t; is the weight coefficient; is the attenuation coefficient; Indicates energy consumption; They represent the effects of heating rate, target temperature and holding time on total energy consumption; It indicates the extraction efficiency of aluminum oxide; is the current mass concentration of alumina in the dissolution solution; is the target alumina mass concentration; Composite functions representing raw material properties; It is a coefficient used to adjust the effect of heating rate, target temperature and holding time on energy consumption; are the coefficients for target particle size, raw material hardness and raw material weight, respectively, is the target particle size; is the hardness of the raw material; is the weight of raw materials; Testing model: inputting the test set into the third parameter prediction model for testing, optimizing the third parameter prediction model according to the test result, and obtaining the optimal third parameter prediction model; The data processing unit includes three processing modules, namely a first processing module, a second processing module and a third processing module; each of the processing modules corresponds to one of the parameter prediction models, and is used to analyze a data set of one of the working modules to obtain a prediction result; The first parameter prediction model includes a first preprocessing layer, a first feature analysis layer and a first parameter prediction layer; The first preprocessing layer is used to perform a preprocessing operation on the first data set to obtain a first standard data set; the preprocessing operation includes data cleaning, standardization, normalization and missing value processing; The first feature analysis layer includes an LSTM network structure, which is used to analyze the first standard data set to obtain a first feature vector; The first parameter prediction layer includes two output neurons, each of which corresponds to a prediction target; the prediction target includes a crushing speed of a crushing device and a crushing time of a crushing device; the first parameter prediction layer is used to predict the parameters in the first parameter group according to the first feature vector to obtain a first prediction result; The second parameter prediction model includes a second preprocessing layer, a second feature analysis layer, a second feature fusion layer and a second parameter prediction layer; the second preprocessing layer is used to perform preprocessing operations on the second data set to obtain a second standard data set; the preprocessing operations include data cleaning, standardization, normalization and missing value processing; the second feature analysis layer includes two LSTM network structures, which are used to analyze the second standard data set and impurity information respectively to obtain a second feature vector and an impurity feature vector; the impurity information includes a first impurity content, a second impurity content, a third impurity content and a fourth impurity content; the second feature fusion layer includes an impurity attention mechanism; the second feature fusion layer is used to splice the second feature vector and the impurity feature vector to obtain a comprehensive feature vector; the impurity features in the comprehensive feature vector are weighted by the impurity attention mechanism to obtain a weighted comprehensive feature vector; the calculation formula of the impurity attention mechanism is: ； ； in, is at the time step Attention score; is the parameter of the attention mechanism; is the hyperbolic tangent activation function; is the weight matrix; is the LSTM network at time step Output: is the weighted sum of all impurity impact scores; is the bias term; and are network parameters; It is The content of impurities; For the The impact score of each impurity; is the activation function, which is used to ensure that the output is between 0 and 1; The second parameter prediction layer includes two output neurons, each of which corresponds to a prediction target; the prediction targets include the amount of alkaline solution and the concentration of alkaline solution; the second parameter prediction layer is used to predict the parameters in the second parameter group according to the second feature vector to obtain a second prediction result; The third parameter prediction model includes a third preprocessing layer, a third feature analysis layer and a third parameter prediction layer; the third preprocessing layer is used to perform preprocessing operations on the third data set to obtain a third standard data set; the preprocessing operations include data cleaning, standardization, normalization and missing value processing; the third feature analysis layer includes an LSTM network structure, which is used to analyze the third standard data set to obtain a third feature vector; the third parameter prediction layer includes three output neurons, each of which corresponds to a prediction target; the prediction targets include heating rate, target temperature and holding time; the third parameter prediction layer is used to predict the parameters in the third parameter group according to the third feature vector to obtain a third prediction result; Data communication unit, including industrial Ethernet, for transmitting data; a parameter setting unit, comprising a PID control algorithm, for setting a first parameter group, a second parameter group, and a third parameter group in a dissolution process based on the prediction result of the data processing unit; The first parameter group includes the crushing speed and crushing time of the crushing equipment; the second parameter group includes the amount of alkaline solution and the concentration of the alkaline solution; the third parameter group includes the heating rate, target temperature and insulation time during the dissolution process.

2. The intelligent detection and setting system for aluminum oxide dissolution process according to claim 1 is characterized in that: The process of obtaining the initial particle size grade of the raw materials includes: The particle size of the raw materials is measured using a laser diffraction particle size analyzer to obtain a particle size set; Set a particle size threshold; when the particle size exceeds or is equal to the particle size threshold, it is marked as a large particle size; When the particle size is smaller than the particle size threshold, it is marked as a small particle size; The Malvern Panalytical software is used to calculate the proportion of the large particle size and the small particle size in the particle size set, respectively, to obtain the initial particle size grade of the raw material; the determination formula of the initial particle size grade of the raw material is: ； in, is the initial particle size grade of the raw material, is the large particle size, is the particle size set; The target particle size is preset; the process of obtaining the target particle size includes: Collecting production data for each batch during production operation; the production data includes the particle size of alumina particles entering the nickel-lined tank, the hardness of the raw materials, the concentration of the alkaline solution, the amount of the alkaline solution, and the alumina content of the dissolved material in the nickel-lined tank; Processing the production data: merging the production data whose values ​​of the raw material hardness, the alkali solution concentration and the alkali solution amount differ by no more than 10% to form a production data group; Using regression analysis method to analyze the relationship between the particle size of the alumina particles in the production data set and the alumina content of the dissolved material in the nickel-lined tank, and determine the target particle size; Establish a mechanism for continuously obtaining the target particle size and regularly review the setting of the target particle size; The weight of the raw material is obtained by a weighing device; the hardness of the raw material is obtained by an ultrasonic hardness tester; the ball mill size and the ball mill quantity are preset values.

3. The intelligent detection and setting system for aluminum oxide dissolution process according to claim 1 is characterized in that: The mass concentration of aluminum oxide in the dissolution solution is obtained by monitoring the absorbance of aluminum oxide in the solution by a near-infrared spectrometer; the mass concentration of sodium carbonate in the dissolution solution is obtained by a near-infrared spectrometer; the target mass concentration of the dissolution solution is a value preset according to production requirements; the carbon-alkali ratio of the dissolution solution is the ratio of sodium carbonate to sodium hydroxide in the dissolution solution, which is obtained by an online ion chromatograph.

Images