Method for predicting the optimum moisture content of a sintering mixture

Abstract

This invention discloses a method for predicting the suitable moisture content of sintering mixtures, belonging to the field of sintering production technology. It obtains the proportions of adhering powder and core particles of each individual iron ore powder in the sintering mixture, as well as the moisture capacity of the adhering powder and core particles, and calculates the comprehensive moisture capacity of each individual iron ore powder and the overall moisture capacity of the sintering mixture. After standardized granulation based on actual sintering production parameters, the permeability index is calculated using the Voisin formula. The model input variables are then constructed, with the suitable moisture content corresponding to each set of conditions used as the model output. A PSO-optimized T-S fuzzy neural network MISO prediction model is built to predict the suitable moisture content of the sintering mixture. This invention significantly improves the prediction accuracy and generalization ability of the suitable moisture content by quantifying the material's moisture capacity characteristics, constructing a multi-dimensional input parameter system, and using a PSO-optimized T-S fuzzy neural network model, making it adaptable to various sintering production conditions.

Description

Technical Field

[0001] This invention relates to the field of sintering production technology, specifically to a method for predicting the suitable moisture content of sintering mixtures. Background Technology

[0002] In the sintering production process, the sintering material needs to undergo wetting, mixing, and granulation to obtain a sintering mixture with a good particle size distribution. Moisture mainly plays a role in binding and wetting the material surface during mixing and granulation. During granulation, moisture affects the granulation effect, thus affecting the permeability of the mixture. If the moisture content of the sintering mixture is too low, the material is difficult to form granules without sufficient moisture binding, resulting in more fine powder material, poor granulation effect, and poor sintering permeability. If the moisture content of the sintering mixture is too high, the granules quickly adhere to a large amount of powder material during rolling, resulting in granules with typically large particle sizes. During further rolling granulation, these green granules are prone to agglomeration under the wetting effect of excessive water and mechanical force, and excessive gravitational water on the material surface reduces the porosity of the mixture, also decreasing sintering permeability. Decreased permeability increases the resistance and energy consumption of the sintering machine's ventilation system, affecting the vertical sintering speed and consequently the yield and quality of the sintered ore. Therefore, adjusting the moisture content of the mixture to achieve optimal air permeability is of great significance for improving sintering productivity and economic benefits.

[0003] Currently, in industrial production, the moisture control of sintering mixtures largely relies on operator experience or comprehensive judgment based on subsequent sintering process parameters, resulting in significant lag, subjectivity, and randomness. A search reveals that patent number CN121761647 A discloses a method, apparatus, equipment, and medium for controlling the moisture of sintering mixtures. This method includes acquiring the material type, feed rate, and moisture coefficient of each material in the mixture. The material types include flux type, dust collector ash type, and other types, where "other types" refers to types other than flux and dust collector ash. Based on the material type, feed rate, and moisture coefficient, the method determines the amount of water added per material of different types. The sum of the water added per material is used as the target water addition for the mixture. The actual water addition is adjusted according to the target water addition. This method ensures that the target water addition changes with changes in the material composition of the mixture, allowing the control of the actual water addition to match the current mixture composition and improving the sintering moisture control effect. Patent CN106842919A discloses a closed-loop control method for the moisture content of sintering mixtures. The control process mainly includes: 1) automatically calculating the amount of water added to the mixture using fuzzy control theory based on data from the target moisture content, a microwave moisture meter, and a water flow meter; 2) establishing a sintering moisture quality evaluation index and using an automatic optimization algorithm to periodically and automatically optimize the target moisture value, thereby precisely controlling the amount of water added. This method changes the traditional manual adjustment mode for water addition in sintering mixtures, realizing closed-loop control of the moisture content, improving the accuracy and automation level of water addition control, and ensuring good mixing, pelletizing effects, and ideal air permeability. Patent No. CN 112130454 A discloses a method and system for intelligent moisture adjustment of sintering mixtures. The method involves using a LiDAR scanner to scan the surface of the mixture on a conveyor belt in real time to obtain three-dimensional data of the surface. The shape of the mixture surface is then generated based on the belt speed and the scanning frequency of the LiDAR scanner to calculate its volume. The weight of the mixture is then obtained from the belt scale, and finally the bulk density of the mixture is obtained. The minimum bulk density and the corresponding optimal moisture value of the mixture are then determined using an extreme value search method, so that the sintering mixture has the best air permeability.

[0004] The existing moisture prediction technologies mentioned above only consider the characteristics of a single iron ore powder or simple proportioning parameters, without comprehensively considering the combined effects of differences in moisture content of iron ore powder particle size, multiple types of auxiliary materials, fuel, and return ore proportions. This results in poor model adaptability, making it unable to adapt to varying ore blending structures and production conditions. Furthermore, traditional prediction models have weak nonlinear fitting capabilities, making it difficult to accurately capture the complex nonlinear relationships between multiple sintering parameters and suitable moisture content, leading to low moisture prediction accuracy and failing to meet the production requirements of modern intelligent sintering. Summary of the Invention

[0005] The purpose of this invention is to provide a method for predicting the appropriate moisture content of sintering mixtures. By quantifying the wet capacity characteristics of iron ore powder, integrating production ratio parameters across all dimensions, and combining an optimized fuzzy neural network model, the method achieves accurate and real-time prediction of the appropriate moisture content. This solves the problems of existing methods for controlling the moisture content of sintering mixtures, which rely on manual experience, have single prediction model parameters, poor nonlinear fitting ability, and weak adaptability to working conditions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting the suitable moisture content of sintering mixtures, comprising the following steps:

[0007] S1: The proportions of adhering powder and core particles of each type of iron ore powder in the sintered blended ore, as well as the wet capacity S of the adhering powder, were obtained through experimental testing. 黏 (%), wet capacity of nuclear particles S 核 (%); whereby, the adhesing powder is iron ore powder with a particle size of -0.5mm, and the core particles are iron ore powder with a particle size of ≥5mm;

[0008] S2: Calculate the comprehensive moisture content S of each type of iron ore powder based on the ratio of adhering powder and core particles and their corresponding moisture content. 综 (%)

[0009] S3: Calculate the overall moisture content S of the sintered blended ore based on the blending ratio of various single iron ore powders and their corresponding comprehensive moisture content. 混匀矿 (%)

[0010] S4: Collect N sets of actual sintering production condition parameters, N≥120. These parameters include different blending ratios of ore, limestone powder, dolomite powder, quicklime, solid fuel, returned ore, dust collector ash, and gas ash. Prepare materials by weight according to the parameters for each set, add granulation water, mix thoroughly, and then granulate using a cylindrical mixer. Measure the pressure drop ΔP of the granulated mixture and calculate the permeability index P using the Voisier formula. e ;

[0011] S5: Based on the variation law of air permeability of the mixture with granulation moisture, determine the suitable moisture value of the sintering mixture corresponding to each working condition; when the air permeability of the mixture tends to level off or decreases for the first time with the increase of moisture, the corresponding moisture value is the suitable granulation moisture.

[0012] S6: Construct the model input variables and introduce the intermediate independent variable, the correction value for the wet capacity of the homogenized ore. Combined with solid fuel ratio P 固体燃料 Limestone powder ratio P 石灰石粉 Dolomite powder ratio P 白云石粉 Quicklime ratio P 生石灰 Gas ash ratio P 瓦斯灰 Dust removal ash ratio P 除尘灰and the return ore ratio P 返矿 Together, they form multidimensional input variables. All input variables are normalized and mapped to the [0,1] interval. The appropriate moisture value corresponding to each working condition is used as the model output.

[0013] S7: Build a PSO-optimized TS fuzzy neural network MISO prediction model, divide the dataset into training and testing sets at a ratio of 9:1, complete model training and verification, and realize the prediction of suitable moisture content of sintering mixture.

[0014] Furthermore, the comprehensive moisture content S of each individual iron ore powder in S2 综 The formula for calculating (%) is as follows:

[0015] .

[0016] Furthermore, the overall moisture content S of the sintered homogenized ore in S3 混匀矿 The formula for calculating (%) is as follows:

[0017]

[0018] In the formula: m represents the type of single iron ore powder; S 综i Let be the moisture content of the i-th type of iron ore powder, %; P i Let be the proportion of the i-th type of iron ore powder participating in the sintering blending.

[0019] Furthermore, the air permeability index P in S4 e The calculation formula is as follows:

[0020]

[0021] In the formula: A is the area of ​​the material layer, in meters. 2 H is the height, in meters; Q is the gas flow rate, in meters. 3 / min.

[0022] Furthermore, the intermediate independent variables introduced in S6 The calculation formula is as follows:

[0023]

[0024] This formula is the product of the moisture content of the sintered blended ore and the proportion of the corresponding blended ore participating in sintering.

[0025] Furthermore, the method for constructing the TS fuzzy neural network MISO prediction model in S7 is as follows:

[0026] S701: Initialize the position and velocity of the particles in the D-dimensional search space, and then randomly generate the position and velocity of the particles in the D-dimensional search space;

[0027] S702: Establish the TS model and calculate the consequent network weights of the learning algorithm based on the output error. The training error of the TS model is used as the fitness value of the particle swarm.

[0028] S703: Determine the individual extreme value P based on the initial fitness value. j and the global optimal solution P g Then update the particle's velocity and position;

[0029] S704: Judgment: Determine whether the number of iterations of the particle swarm has reached the maximum value. If "yes", calculate the optimal parameters of the TS model; if "no", return to S703.

[0030] S705: Second judgment: Determine whether the iteration of the TS model has reached the maximum value. If "yes", the model iteration ends; otherwise, return to S702.

[0031] Furthermore, in S701, the particle represents the network weight of the consequent in the TS fuzzy neural network. Assuming there are n particles in a D-dimensional search space, the calculation formula is as follows:

[0032]

[0033]

[0034] In the formula: Represents the current velocity of the j-th particle in the D-dimensional search space (j = 1, 2, ..., n; i = 1, 2, ..., D); Let be the position of the j-th particle in the population in D-dimensional space, i.e., the weight population of the consequent network of the TS fuzzy neural network; Let pbest be the individual optimal value of the j-th particle, which is the optimal solution of the consequent weights of the TS fuzzy neural network. The population optimum gbest represents the population; k is the current iteration number; ω is the inertia weight; c1 and c2 are acceleration constants; r1 and r2 are random numbers distributed in the interval [0, 1]; the particle position and velocity are restricted to a certain interval [-Xmax, Xmax] and [-Vmax, Vmax] to prevent it from blindly searching.

[0035] Furthermore, the fuzzy rule establishment method for the TS model in S702 is as follows: Let a MISO system input vector be... Each component x i All are fuzzy linguistic variables, and are set as follows: ;in, It is x iThe value of the j-th linguistic variable is defined in the domain x. i If a fuzzy set is given, then the TS fuzzy rule can be expressed by the following formula:

[0036]

[0037] then

[0038] In the formula: Let k be the k-th rule, where k = 1, 2, ..., z; .

[0039] Furthermore, the TS model consists of two parts: an antecedent network and a consequent network. The antecedent network is used to match fuzzy rules, and the consequent network is used to generate fuzzy rules, as detailed below:

[0040] (1) In the preamble network: the first layer is the input layer with N1 = D nodes, where D is the number of input parameters; the second layer is the fuzzification layer, whose main function is to fuzzify the input data, and its membership function is a Gaussian function:

[0041]

[0042] In the formula: i = 1, 2, ..., D; j = 1, 2, ..., m; m is the number of fuzzy segments for each input parameter; and They are respectively The center and width;

[0043] The third layer is the fuzzy rule layer, used to match the antecedents of fuzzy rules and calculate the fitness of each rule. The node function is:

[0044]

[0045] Number of nodes N3 = m;

[0046] The fourth layer is the normalization layer, and its formula is as follows:

[0047] ;

[0048] (2) In the consequent network: the first layer is the input layer, with the 0th node inputting the value x0 = 1, providing the constant term in the consequent of the fuzzy rule; the second layer has m nodes, which calculate the consequent of each rule, i.e.

[0049]

[0050] The third layer is the system output, namely:

[0051] .

[0052] Furthermore, it also includes verifying the prediction accuracy of the prediction model, with the specific formula as follows:

[0053] Based on the actual conditions during sintering production, a prediction error of ±0.20% of the moisture content of the sintering mixture was selected to calculate the prediction accuracy of the model. The hit rate was calculated as follows:

[0054]

[0055] when hour, ;otherwise ;

[0056] In the formula: l is the length of the predicted sample; Indicates whether the hit or not; These are the model's predicted values; This is the actual value;

[0057] The mean relative error (MRE) is calculated as follows:

[0058]

[0059] In the formula: n is the number of samples; These are the model's predicted values; This is the actual value.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] 1. The method for predicting the suitable moisture content of sintering mixtures of the present invention quantifies the difference in moisture capacity of different particle sizes of iron ore powder for the first time. By weighting the proportion of adhering powder and core particles, the comprehensive moisture capacity is calculated, which accurately characterizes the water absorption and granulation characteristics of the mixed ore. This solves the defect of traditional technology that ignores the influence of iron ore powder particle size structure on moisture demand.

[0062] 2. The method for predicting the suitable moisture content of sintering mixtures in this invention integrates core operating parameters such as the characteristics of the mixed ore, flux, fuel, return ore, and industrial auxiliary materials, covering all batching variables in sintering production. The model has comprehensive input dimensions and can be adapted to any ore blending and batching conditions.

[0063] 3. The method for predicting the suitable moisture content of sintered mixtures in this invention uses the trend of air permeability change as the basis for determining the suitable moisture content, which is in line with the process essence of sintering and granulation. The calibrated sample data is real and accurate, effectively improving the quality of model training.

[0064] 4. The method for predicting the suitable moisture content of sintering mixtures in this invention optimizes the consequent weights of the TS fuzzy neural network through the PSO algorithm, which solves the problems of traditional fuzzy neural network parameters easily getting trapped in local optima and slow convergence speed. The model has high prediction accuracy and good stability.

[0065] 5. The method for predicting the suitable moisture content of sintering mixtures of the present invention constructs a multi-input single-output MISO system to achieve nonlinear accurate prediction under multi-parameter coupling. It can completely replace manual experience adjustment, realize intelligent and precise control of sintering moisture, and effectively improve sintering production efficiency and finished product quality. Attached Figure Description

[0066] Figure 1 This is a flowchart of a method according to an embodiment of the present invention;

[0067] Figure 2 This is a schematic diagram illustrating the process of optimizing the TS fuzzy neural network using the PSO algorithm in an embodiment of the present invention.

[0068] Figure 3 This is a schematic diagram of the TS fuzzy neural network structure in an embodiment of the present invention;

[0069] Figure 4 This is a trend chart comparing the model's predicted values ​​with the actual values ​​in the experimental examples of this invention. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] To address the problems of existing methods for controlling the moisture content of sintering mixtures, which rely on manual experience, have limited predictive model parameters, poor nonlinear fitting ability, and weak adaptability to operating conditions, this invention provides a method for predicting the suitable moisture content of sintering mixtures, such as... Figure 1 As shown, it includes the following steps:

[0072] Step 1: Obtain the percentage (%) of adhering powder (-0.5mm particle size fraction) and core particles (≥5mm particle size fraction) of each type of iron ore powder in the sintered blended ore, and the moisture content S of the adhering powder through experimental testing. 黏 (%), and the wet capacity S of the nuclear particles. 核 (%).

[0073] Step 2: The proportion (%) of the adhering powder (-0.5mm particle size fraction) and core particles (≥5mm particle size fraction) of each type of iron ore powder in the above-mentioned sintered and blended ore can be automatically processed and obtained by the particle size image detection device system installed on the feed belt of the batching and feeding belt before the sintered and blended ore is piled up.

[0074] Step 3: Based on the blending ratios of various single iron ore powders and their corresponding comprehensive moisture content, calculate the overall moisture content S of the sintered blended ore. 混匀矿 (%); The specific formula is as follows:

[0075]

[0076] If a sintered blended ore is composed of m types of single iron ore powders, the corresponding wet capacity S of the sintered blended ore can be calculated using the following formula. 混匀矿 :

[0077]

[0078] In the formula: S 混匀矿 The moisture content of the sintered blended ore, %; S 综i Let be the moisture content of the i-th type of iron ore powder, %; P i The proportion of the i-th type of iron ore powder participating in the sintering blending is %; b 黏 The percentage is the proportion of adhesive powder.

[0079] Step 4: Obtain the actual parameters of different sintering blended ore blending structures and corresponding blended ore ratios, different flux (limestone powder, dolomite powder and quicklime) ratios, different solid fuel ratios, different return ore ratios, and different auxiliary material (dust removal ash and gas ash) ratios during actual sintering production for N groups (≥120 groups).

[0080] Step 5: Produce the materials by weight according to the specified proportions for each sintering material. Add the water required for granulation and mix thoroughly. Then, load the mixture into a 600 mm * 300 mm cylindrical mixer and granulate for 4 minutes at 15 r / min. The air permeability of the granulated mixture is tested by measuring the pressure drop ΔP of the gas passing through the mixture layer. The air permeability index Pe is calculated using the Voisel formula.

[0081]

[0082] In the formula: the material layer area A is 7.85 × 10 -2 m 2 The height H is 0.2m, and the gas flow rate Q is 0.167 m³ / s. 3 / min.

[0083] Step 6: As the moisture content during granulation increases, the permeability of the mixture improves significantly. However, when the moisture content reaches a certain level, the increase in permeability slows down or even decreases. The purpose of granulation is to allow fine particles to adhere to relatively coarse particles, thereby reducing the content of fine particles in the mixture and ensuring good permeability of the sintered material layer. Therefore, based on the purpose of granulation, the criterion for determining the suitable moisture content for granulation can be obtained: the suitable moisture content is defined as the point at which the permeability of the mixture begins to decrease or its increase slows down. The suitable moisture content values ​​for the sintered mixtures corresponding to the above N groups of different ore blends and under different blending conditions are obtained using this method.

[0084] Step 7: Introduce intermediate independent variables Its formula The value is given as the product of the moisture content of the sintered blended ore and the proportion of the corresponding blended ore participating in sintering. The different groups mentioned above... Solid fuel ratio P 固体燃料 Limestone powder ratio P 石灰石粉 Dolomite powder ratio P 白云石粉 Quicklime ratio P 生石灰 Gas ash ratio P 瓦斯灰 Dust removal ash ratio P 除尘灰 and the return ore ratio P 返矿 The eight variables mentioned above are normalized and mapped to [0, 1], serving as input parameters for the suitable moisture rate prediction model. The suitable moisture values ​​are then used as the model's output. To facilitate observation of the model's output, the normalized data needs to be restored through inverse normalization.

[0085] Step 8: Build a PSO-optimized TS fuzzy neural network MISO prediction model. Divide the dataset into training and testing sets at a 9:1 ratio to complete model training and validation, achieving the prediction of suitable moisture content for sintering mixtures. Details are as follows:

[0086] First, the positions and velocities of particles are initialized in a D-dimensional search space. All particles are evaluated based on a fitness value, where the particle's position is determined by its weights in the TS fuzzy neural network. The fitness value in the PSO algorithm is the difference between the predicted and actual values ​​from the TS fuzzy model. The particle's direction and distance are determined by its velocity, which is dynamically adjusted based on its own operational experience and that of other particles in the swarm. The particle position is updated by tracking its individual optimal value (pbest) and the global optimal value (gbest). Suppose there are n particles in a D-dimensional search space, and their mathematical expressions are as follows:

[0087]

[0088]

[0089]

[0090]

[0091]

[0092]

[0093] In the formula: Represents the current velocity of the j-th particle in the D-dimensional search space (j=1, 2, ..., n; i=1, 2, ..., D); Let be the position of the j-th particle in the population in D-dimensional space, i.e., the weight population of the consequent network of the TS fuzzy neural network; This represents the individual optimal value (pbest) of the j-th particle, which is the optimal solution for the consequent weights of the TS fuzzy neural network. represents the population's optimal value (gbest); k is the current iteration number; ω is the inertia weight; c1 and c2 are acceleration constants; r1 and r2 are random numbers distributed in the interval [0, 1]. The particle position and velocity are restricted to certain intervals [-Xmax, Xmax] and [-Vmax, Vmax] to prevent blind searching.

[0094] like Figure 2 As shown, in this step, the weights of the consequent network of the TS fuzzy neural network are optimized using PSO. The steps are as follows:

[0095] (1) Initialization: Randomly generate the position and velocity of the particles in the D-dimensional search space.

[0096] (2) Establish the TS model, calculate the consequent weights of the learning algorithm based on the output error, and use the training error of the TS model as the fitness value of the particle swarm.

[0097] (3) Determine the individual extreme value P based on the initial fitness value. j and the global optimal solution P g Then update the particle's velocity and position.

[0098] (4) Judgment: Determine whether the number of iterations of the particle swarm has reached the maximum value. If it is "yes", calculate the optimal parameters of the TS model; if it is "no", return to (3).

[0099] (5) Second judgment: Determine whether the iteration of the TS model has reached the maximum value. If it is "yes", the model iteration ends; otherwise, return to (2).

[0100] In the above TS model, let there be an MISO system input vector. Each component xi is a fuzzy linguistic variable, and is set as follows: .in, Let be the j-th linguistic variable value of xi, which is a fuzzy set defined on the universe of discourse xi. Then the TS fuzzy rule can be expressed by the following formula:

[0101]

[0102] then

[0103] In the formula: Let k be the k-th rule, where k = 1, 2, ..., z; .

[0104] like Figure 3 As shown, the TS fuzzy neural network structure model built in the above steps consists of two parts: an antecedent network and a consequent network. The antecedent network is used to match fuzzy rules, and the consequent network is used to generate fuzzy rules, as detailed below:

[0105] (1) In the preamble network: Layer 1 is the input layer with N1 = D nodes, where D is the number of input parameters (8); Layer 2 is the fuzzification layer, whose main function is to fuzzify the input data. Its membership function is a Gaussian function:

[0106]

[0107] In the formula: i = 1, 2, ..., D; j = 1, 2, ..., m, where m is the number of fuzzy segments for each input parameter. and They are respectively The center and width.

[0108] The third layer is the fuzzy rule layer, used to match the antecedents of fuzzy rules and calculate the fitness of each rule. Node functions:

[0109]

[0110] The number of nodes N3 = m. Let be the membership degree of the i-th feature of the k-th sample to the j-th fuzzy rule. Let be the activation strength of the j-th fuzzy rule for the k-th sample.

[0111] The fourth layer is the normalization layer, and its formula is as follows:

[0112]

[0113] (2) In the consequent network: Layer 1 is the input layer, with the 0th node inputting the value x0 = 1, providing the constant term in the consequent of the fuzzy rule. Layer 2 has m nodes, which calculate the consequent of each rule, i.e.

[0114]

[0115] For the k-th sample, the i-th dimension is the original input independent variable. For the j-th rule, input the corresponding linear weight coefficients for the i-th dimension. This is the local linear output corresponding to the j-th fuzzy rule. All D-dimensional inputs are weighted and accumulated.

[0116] The third layer is the system output, namely:

[0117]

[0118] Step 9: Based on the prediction model built in Step 8, set a threshold (0.15%) between the model's predicted moisture content and the current target moisture content of the sintering mixture. If the difference is within the threshold range, the control flag for the target moisture content will not be adjusted; otherwise, the current model prediction value will be used to replace the target moisture content control value for water control of the sintering mixture.

[0119] Specifically, based on the actual conditions during sintering production, a prediction error of ±0.20% of the moisture content of the sintering mixture can be selected to calculate the model's prediction accuracy. The hit rate is calculated as follows:

[0120]

[0121] when hour, ;otherwise .

[0122] In the formula: l is the length of the predicted sample; Indicates whether the hit or not; These are the model's predicted values; This is the actual value.

[0123] The mean relative error (MRE) is calculated as follows:

[0124]

[0125] In the formula: n is the number of samples; These are the model's predicted values; This is the actual value.

[0126] To further verify the effectiveness and feasibility of the method of the present invention, the following specific experimental data are provided for illustration:

[0127] First, the proportions of adhering powder and core particles of each individual ore in the blended ore participating in the sintering batch and their corresponding moisture content values ​​were obtained through experiments, as shown in Table 1:

[0128] Table 1. Particle size distribution and moisture content of various iron ore powders

[0129]

[0130] Subsequently, the proportion of the sintered blended ore during sintering production and the corresponding moisture content S of the sintered blended ore were obtained. 混匀矿 See Table 2 for details:

[0131] Table 2. Structure of sintered homogenized ore and calculated moisture content S 混匀矿

[0132]

[0133] Then obtain the batching ratio and calculations during sintering production. The appropriate moisture content for the laboratory is shown in Table 3.

[0134] Table 3 Suitable Moisture Content and Batching Parameters for Sintered Blended Ore

[0135]

[0136] Using the data in Table 3 above, a PSO-optimized TS fuzzy neural network was constructed to predict the suitable moisture content of sintered ore. The ratio of the training set to the test set was 9:1. In actual production, new actual production parameters were predicted using the model, and the model's prediction accuracy reached over 95%, with an average relative error of 0.0056. The specific trends of the actual and predicted values ​​are as follows: Figure 4 As shown.

[0137] The model predicts the moisture content of the sinter mixture and uses it to control the water addition in the sintering mixture. After using this method, the sintering yield increased by 1.36%, the strength of the sinter drum increased by 1.12%, and the power consumption of the main pumping unit decreased by 1.51 kWh / t of ore, as shown in the table below:

[0138]

[0139] Calculations based on the above data show that the predicted moisture content of the model can be used as the target moisture value for controlling the water addition of the sintering mixture during actual sintering production. A certain threshold can be set: if the difference between the model's current predicted moisture content and the current target moisture content of the mixture is less than 0.15%, the moisture content control value of the mixture will not be adjusted; otherwise, the target control value will be replaced by the current model-predicted moisture content.

[0140] In summary, this invention provides a method for predicting the suitable moisture content of sintering mixtures. For the first time, it quantifies the differences in moisture capacity among different particle sizes of iron ore powder. By weighting the proportions of adhering powder and core particles, it calculates the comprehensive moisture capacity, accurately characterizing the water absorption and granulation characteristics of the blended ore. This overcomes the shortcomings of traditional technologies that neglect the influence of iron ore powder particle size structure on moisture requirements. Simultaneously, it integrates core operating parameters such as blended ore characteristics, flux, fuel, return ore, and industrial auxiliary materials, covering all batching variables in sintering production. The model input dimensions are comprehensive and adaptable to any ore blending and batching conditions. The use of permeability change trends as the criterion for determining suitable moisture content aligns with the essence of sintering granulation processes, and the calibrated sample data is authentic and accurate, effectively improving the model training quality. Furthermore, this invention optimizes the consequent weights of the TS fuzzy neural network using the PSO algorithm, solving the problems of traditional fuzzy neural network parameter iteration easily getting trapped in local optima and slow convergence speed. The model exhibits high prediction accuracy and good stability. In addition, the construction of a multi-input single-output MISO system enables precise nonlinear prediction under multi-parameter coupling, which can completely replace manual experience-based adjustment, realize intelligent and precise control of sintering moisture, and effectively improve sintering production efficiency and finished product quality.

[0141] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting the suitable moisture content of sintering mixtures, characterized in that, Includes the following steps: S1: The proportions of adhering powder and core particles of each type of iron ore powder in the sintered blended ore, as well as the wet capacity S of the adhering powder, were obtained through experimental testing. 黏 (%), wet capacity of nuclear particles S 核 (%); whereby, the adhesing powder is iron ore powder with a particle size of -0.5mm, and the core particles are iron ore powder with a particle size of ≥5mm; S2: Calculate the comprehensive moisture content S of each type of iron ore powder based on the ratio of adhering powder and core particles and their corresponding moisture content. 综 (%) S3: Calculate the overall moisture content S of the sintered blended ore based on the blending ratio of various single iron ore powders and their corresponding comprehensive moisture content. 混匀矿 (%) S4: Collect N sets of actual sintering production condition parameters, N≥120. These parameters include different blending ratios of ore, limestone powder, dolomite powder, quicklime, solid fuel, returned ore, dust collector ash, and gas ash. Prepare materials by weight according to the parameters for each set, add granulation water, mix thoroughly, and then granulate using a cylindrical mixer. Measure the pressure drop ΔP of the granulated mixture and calculate the permeability index P using the Voisier formula. e ; S5: Based on the variation law of air permeability of the mixture with granulation moisture, determine the suitable moisture value of the sintering mixture corresponding to each working condition; when the air permeability of the mixture tends to level off or decreases for the first time with the increase of moisture, the corresponding moisture value is the suitable granulation moisture. S6: Construct the model input variables and introduce the intermediate independent variable, the correction value for the wet capacity of the homogenized ore. Combined with solid fuel ratio P 固体燃料 Limestone powder ratio P 石灰石粉 Dolomite powder ratio P 白云石粉 Quicklime ratio P 生石灰 Gas ash ratio P 瓦斯灰 Dust removal ash ratio P 除尘灰 and the return ore ratio P 返矿 Together, they form multidimensional input variables. All input variables are normalized and mapped to the [0,1] interval. The appropriate moisture value corresponding to each working condition is used as the model output. S7: Build a PSO-optimized TS fuzzy neural network MISO prediction model, divide the dataset into training and testing sets at a ratio of 9:1, complete model training and verification, and realize the prediction of suitable moisture content of sintering mixture.

2. The method for predicting the suitable moisture content of sintering mixtures as described in claim 1, characterized in that: The combined moisture content S of each individual iron ore powder in S2 综 The formula for calculating (%) is as follows: In the formula, b 黏 The percentage is the proportion of adhesive powder.

3. The method for predicting the suitable moisture content of sintering mixtures as described in claim 1, characterized in that: The overall moisture content of S3 medium-sintered mixed ore S 混匀矿 The formula for calculating (%) is as follows: In the formula: m represents the type of single iron ore powder; S 综i Let be the moisture content of the i-th type of iron ore powder, %; P i Let be the proportion of the i-th type of iron ore powder participating in the sintering blending.

4. The method for predicting the suitable moisture content of sintering mixtures as described in claim 1, characterized in that: S4 has an air permeability index P e The calculation formula is as follows: In the formula: A is the area of ​​the material layer, in meters. 2 H is the height, in meters; Q is the gas flow rate, in meters. 3 / min.

5. The method for predicting the suitable moisture content of sintering mixtures as described in claim 1, characterized in that: Intermediate independent variables introduced in S6 The calculation formula is as follows: This formula is the product of the moisture content of the sintered blended ore and the proportion of the corresponding blended ore participating in sintering.

6. The method for predicting the suitable moisture content of sintering mixtures as described in claim 1, characterized in that: The method for constructing the TS fuzzy neural network MISO prediction model in S7 is as follows: S701: Initialize the position and velocity of the particles in the D-dimensional search space, and then randomly generate the position and velocity of the particles in the D-dimensional search space; S702: Establish the TS model and calculate the consequent network weights of the learning algorithm based on the output error. The training error of the TS model is used as the fitness value of the particle swarm. S703: Determine the individual extreme value P based on the initial fitness value. j and the global optimal solution P g Then update the particle's velocity and position; S704: Judgment: Determine whether the number of iterations of the particle swarm has reached the maximum value. If "yes", calculate the optimal parameters of the TS model; if "no", return to S703. S705: Second judgment: Determine whether the iteration of the TS model has reached the maximum value. If "yes", the model iteration ends; otherwise, return to S702.

7. The method for predicting the suitable moisture content of sintering mixtures as described in claim 6, characterized in that: In S701, a particle represents the network weight of the consequent in the TS fuzzy neural network. Assuming there are n particles in a D-dimensional search space, the calculation formula is as follows: In the formula: Represents the current velocity of the j-th particle in the D-dimensional search space (j = 1, 2, ..., n; i = 1, 2, ..., D); Let be the position of the j-th particle in the population in D-dimensional space, i.e., the weight population of the consequent network of the TS fuzzy neural network; Let pbest be the individual optimal value of the j-th particle, which is the optimal solution of the consequent weights of the TS fuzzy neural network. The population optimum gbest represents the population; k is the current iteration number; ω is the inertia weight; c1 and c2 are acceleration constants; r1 and r2 are random numbers distributed in the interval [0, 1]; the particle position and velocity are restricted to a certain interval [-Xmax, Xmax] and [-Vmax, Vmax] to prevent it from blindly searching.

8. The method for predicting the suitable moisture content of sintering mixtures as described in claim 6, characterized in that: The fuzzy rule establishment method for the TS model in S702 is as follows: Let there be an MISO system input vector. Each component x i All are fuzzy linguistic variables, and are set as follows: ;in, It is x i The value of the j-th linguistic variable is defined in the domain x. i If a fuzzy set is given, then the TS fuzzy rule can be expressed by the following formula: then In the formula: Let k be the k-th rule, where k = 1, 2, ..., z; .

9. The method for predicting the suitable moisture content of sintering mixtures as described in claim 8, characterized in that: The TS model consists of two parts: an antecedent network and a consequent network. The antecedent network is used to match fuzzy rules, and the consequent network is used to generate fuzzy rules, as detailed below: (1) In the preamble network: the first layer is the input layer with N1 = D nodes, where D is the number of input parameters; the second layer is the fuzzification layer, whose main function is to fuzzify the input data, and its membership function is a Gaussian function: In the formula: i = 1, 2, ..., D; j = 1, 2, ..., m; m is the number of fuzzy segments for each input parameter; and They are respectively The center and width; The third layer is the fuzzy rule layer, used to match the antecedents of fuzzy rules and calculate the fitness of each rule. The node function is: Number of nodes N3 = m; Let be the membership degree of the i-th feature of the k-th sample to the j-th fuzzy rule. Let be the activation strength of the j-th fuzzy rule for the k-th sample; The fourth layer is the normalization layer, and its formula is as follows: ； (2) In the consequent network: the first layer is the input layer, with the 0th node inputting the value x0 = 1, providing the constant term in the consequent of the fuzzy rule; the second layer has m nodes, which calculate the consequent of each rule, i.e. In the formula, For the k-th sample, the i-th dimension is the original input independent variable. For the j-th rule, input the corresponding linear weight coefficients for the i-th dimension. This is the local linear output corresponding to the j-th fuzzy rule. Weighted summation of all D-dimensional inputs; The third layer is the system output, namely: 。 10. The method for predicting the suitable moisture content of sintering mixtures as described in claim 9, characterized in that: It also includes verifying the prediction accuracy of the prediction model, with the specific formula as follows: Based on the actual conditions during sintering production, a prediction error of ±0.20% of the moisture content of the sintering mixture was selected to calculate the prediction accuracy of the model. The hit rate was calculated as follows: when hour, ;otherwise ; In the formula: l is the length of the predicted sample; Indicates whether the hit or not; These are the model's predicted values; This is the actual value; The mean relative error (MRE) is calculated as follows: In the formula: n is the number of samples; These are the model's predicted values; This is the actual value.

Images