A pellet production system based on a grate and a belt induration furnace

Abstract

The present application relates to a kind of pellet production system based on chain grate and belt type baking machine, belong to pellet production technical field.It includes: raw material processing module, for generating in accordance with preset ore blending scheme the pelletizing raw material meeting preset particle size distribution;Pellet process parameter generation module, for determining the performance of pelletizing raw material and generating pellet process parameter;Pelletizing optimization module, for obtaining disc pelletizer real-time pelletizing parameter and closed-loop control disc pelletizer equipment parameter generation green ball;Chain grate preheating module, for adjusting the preheating parameter of chain grate according to pellet process parameter, and preheating green ball to obtain preheated pellet;Belt type baking machine baking module, for adjusting the baking parameter of belt type baking machine according to integrated multivariate regression model and real-time baking parameter, and obtaining finished pellet after cooling after completing baking process;Pellet measurement module, for reverse optimization preset ore blending scheme according to pellet index.The present application improves pellet index, meets industrial demand.

Description

Technical Field

[0001] This invention relates to the field of pellet production technology, and in particular to a pellet production system based on a chain grate machine and a belt roaster. Background Technology

[0002] Pellets are an important raw material in the metallurgical process. They are solid particles formed by mixing ore powder, fuel and binder, and then pressing and roasting them. They are usually used in blast furnace smelting or direct reduced iron production.

[0003] Common pellet production equipment includes three types: vertical shaft furnace, belt roaster, and chain grate machine. Vertical shaft furnace roasting has a small single-unit capacity, uneven heating, and poor adaptability to raw materials; although belt roaster is simple to operate, its roasting control precision is still insufficient; chain grate machine has the advantages of uniform roasting and easy control, but it has high energy consumption at high temperatures and is difficult to control the operating parameters at high temperatures.

[0004] Therefore, the present invention provides a pellet production system based on a chain grate machine and a belt roaster. Summary of the Invention

[0005] To solve the above-mentioned technical problems, this invention provides a pellet production system based on a chain grate machine and a belt calciner. The technical solution of this invention is as follows:

[0006] This invention provides a pellet production system based on a chain grate machine and a belt roaster, comprising:

[0007] The raw material processing module is used to generate pelletizing raw materials that conform to the preset particle size distribution according to the preset ore blending scheme;

[0008] The pelletizing process parameter generation module is used to measure the performance of the pelletizing raw materials after pre-activation treatment, and generate pelletizing process parameters based on the performance of the pelletizing raw materials.

[0009] The pelletizing optimization module is used to obtain real-time pelletizing parameters in the pelletizing process of the disc pelletizer, control the equipment parameters of the disc pelletizer in a closed loop based on the pelletizing process parameters and real-time pelletizing parameters, and control the disc pelletizer to generate green pellets from the pelletizing raw materials through the equipment parameters of the disc pelletizer.

[0010] The preheating module of the chain grate machine is used to adjust the preheating parameters of the chain grate machine according to the pelletizing process parameters through a gradient boosting decision tree, and to control the chain grate machine to preheat the green pellets in stages with the preheating parameters to obtain preheated pellets.

[0011] The belt calciner calcination module is used to preheat the pellets by calcining them with a belt calciner and obtain real-time calcination parameters during the calcination process. The real-time calcination parameters are input into an integrated multivariate regression model. Based on the integrated multivariate regression model and the pellet process parameters, calcination adjustment parameters are output. The calcination parameters of the belt calciner are dynamically adjusted according to the calcination adjustment parameters. After the belt calciner completes the calcination process, the pellets are cooled to obtain the finished product.

[0012] The pellet measurement module is used to measure the pellet indicators of the finished pellets and to optimize the preset ore blending scheme based on the pellet indicators.

[0013] Preferably, the pelletizing process parameter generation module includes:

[0014] The pre-activation treatment unit is used to collect quantitative pelletizing raw materials as test samples. After quality inspection of the test samples, the test samples are pre-activated using chemical methods to obtain standard samples.

[0015] The performance testing unit is used to measure the particle size and bulk density of the standard sample using a laser particle size analyzer to obtain the physical properties of the standard sample; and to measure the chemical properties of the standard sample using a chemical analyzer. The physical and chemical properties of the standard sample constitute the performance of the pelletizing raw material.

[0016] The pelletizing process parameter determination unit inputs the performance of the pelletizing raw materials into a pre-trained pelletizing process feature extraction model, which outputs the pelletizing process features of the raw materials and generates pelletizing process parameters based on these features.

[0017] Preferably, the ball-forming optimization module includes:

[0018] The pelletizing data acquisition unit is used to input pelletizing raw materials into the disc pelletizer and obtain real-time pelletizing parameters in the pelletizing process of the disc pelletizer by sensors pre-installed in the disc pelletizer in combination with the pelletizing parameter type.

[0019] The state space construction unit is used to construct the current pelletizing state space based on real-time pelletizing parameters, and to determine the current pelletizing state based on the pelletizing process parameters to obtain the determination result.

[0020] The closed-loop control unit is used to generate a pelletizing adjustment signal based on the judgment result, perform closed-loop control on the equipment parameters of the disc pelletizer based on the pelletizing adjustment signal, and control the disc pelletizer to generate green pellets from the pelletizing raw materials through the equipment parameters of the disc pelletizer.

[0021] The grading and screening unit is used to screen the green pellets generated by the disc pelletizer through a screening device, and to return green pellets whose particle size does not meet the standard particle size in the pelletizing process parameters to the disc pelletizer for reprocessing.

[0022] Preferably, the state space construction unit includes:

[0023] The state variable selection subunit is used to construct a covariance matrix based on the real-time ball-making parameters corresponding to all ball-making parameter types, perform eigenvalue decomposition on the covariance matrix to obtain principal component eigenvectors, which are composed of eigenvalues ​​of each ball-making parameter type. The eigenvalues ​​of the ball-making parameter type in the principal component eigenvectors that are less than the spatial threshold are deleted to obtain the state variables and their eigenvalues.

[0024] The state health calculation subunit constructs a ball-making state space based on state variables and their characteristic values, calculates the state health of the ball-making state space at the current moment, and determines that the ball-making state at the current moment is abnormal if the state health of the ball-making state space at the current moment is greater than the health threshold, otherwise it determines that the ball-making state at the current moment is normal.

[0025] Preferably, when the closed-loop control unit performs closed-loop control of the equipment parameters of the disc pelletizer based on the pelletizing adjustment signal, it includes:

[0026] The pelletizing adjustment signal is projected onto the preset feasible domain of the equipment to obtain the control parameters of the disc pelletizer. The control parameters are distributed to the corresponding equipment of the disc pelletizer. The equipment parameters of each equipment of the disc pelletizer are adjusted according to the control parameters. Real-time pelletizing parameters are obtained during the operation of each equipment of the disc pelletizer according to its equipment parameters. The real-time pelletizing parameters are fed back to the state space construction unit. The equipment parameters of the disc pelletizer are controlled in a closed loop according to the judgment result of the state space construction unit.

[0027] Preferably, the chain grate machine's segmented preheating of green pellets includes a forced-air drying section, an exhaust drying section, and a preheating section, and the chain grate machine's preheating module includes:

[0028] The objective function construction unit is used to construct a preheating prediction model based on the preheating parameter type, generate constraints for each segment based on the pelletizing process parameters and the preheating conditions of each segment, and construct the objective function for each segment based on the constraints and the preheating prediction model.

[0029] The decision tree unit is used to calculate the preheating parameters for each segment based on the gradient boosting decision tree and the objective function of each segment.

[0030] The preheating unit is used to control the chain grate machine to preheat the green pellets in segments according to the preheating parameters of each segment, so as to obtain preheated pellets.

[0031] Preferably, the decision tree unit includes:

[0032] The error subunit is used to calculate the error between the predicted result of the objective function of each segment and its target preheating parameter, and to use the error as the input for the next iteration of the gradient boosting decision tree.

[0033] The iterative subunit is used to calculate the error gradient based on the error of the previous iteration of the decision tree, update the function parameters of the objective function of each segment based on the error gradient, obtain the new objective function of each segment, and calculate the error between the prediction result of the new objective function of each segment and its target preheating parameters. The iteration stops when the number of iterations meets the preset number of decision tree nodes, and all error gradients are obtained.

[0034] The output sub-unit is used to fit all error gradients obtained from the gradient boosting decision tree to obtain the updated gradient. The updated gradient and the prediction results of each iteration are weighted and summed to obtain the warm-up parameters for each segment.

[0035] Preferably, the belt roaster roasting module includes:

[0036] The multivariate regression model building unit is used to generate parameter tensors based on preset pellet indicators and preset roasting parameter types in the pelletizing process parameters, and to build a regression model for each preset roasting parameter type based on the parameter tensors.

[0037] An integration unit is used to calculate the weight coefficient of each preset roasting parameter type based on the real-time roasting parameters during the roasting process, and to generate an integrated multivariate regression model based on the weight coefficient of each preset roasting parameter type and the regression model of each preset roasting parameter type.

[0038] The parameter adjustment unit inputs real-time roasting parameters into an integrated multivariate regression model, which outputs roasting adjustment parameters for each preset roasting parameter type. The roasting parameters of the belt roaster are adjusted in real time according to the roasting adjustment parameters for each preset roasting parameter type. After the belt roaster completes the roasting process with the real-time adjusted roasting parameters, it is cooled to obtain the finished pellets.

[0039] Preferably, when the integration unit calculates the weighting coefficient for each preset roasting parameter type based on the real-time roasting parameters during the roasting process, it includes:

[0040] Calculate the parameter sensitivity of each preset roasting parameter type based on the real-time roasting parameters of each preset roasting parameter type;

[0041] Construct a sensitivity matrix based on the parameter sensitivity of all preset roasting parameter types, and calculate the initial weight of each preset roasting parameter type based on the sensitivity matrix;

[0042] The initial reward function is updated based on the real-time roasting parameters to obtain a new reward function, and the initial weight of each preset roasting parameter type is updated based on the new reward function to obtain the weight coefficient of each preset roasting parameter type.

[0043] All of the above-mentioned optional technical solutions can be combined arbitrarily, and the present invention will not provide a detailed description of the structure after each combination.

[0044] By means of the above solution, the beneficial effects of the present invention are as follows:

[0045] By controlling the equipment parameters of the disc pelletizer in a closed loop based on pelletizing process parameters and real-time pelletizing parameters, high-standard green pellets are provided for subsequent processing, improving the performance of the finished pellets obtained from preheating and roasting. A method for intelligently adjusting the preheating parameters of the chain grate machine is provided by adjusting the preheating parameters of the chain grate machine using a gradient boosting decision tree based on the pelletizing process parameters, and controlling the chain grate machine to preheat the green pellets in stages with the preheating parameters, ensuring the stability of preheated pellet production. By roasting the preheated pellets with a belt roaster and acquiring real-time roasting parameters during the roasting process, these real-time roasting parameters are input into an integrated multivariate regression model. Based on the integrated multivariate regression model and the pelletizing process parameters, roasting adjustment parameters are output, realizing real-time adjustment of the belt roaster's roasting parameters. This allows for real-time adjustment of roasting parameters according to the roasting status of the pellets, improving the stability of the finished pellet production process and the performance of the finished pellets. By using an integrated multivariate regression model to determine the output roasting adjustment parameters, the subjective nature of traditional experience-based decision-making methods is avoided, making the process more scientific and intelligent.

[0046] In summary, this invention combines the advantages of both a chain grate machine and a belt roaster. It improves preheating efficiency by controlling the preheating parameters of the chain grate machine and enhances roasting efficiency by controlling the roasting parameters of the belt roaster. The chain grate machine, responsible for preheating, operates at a preheating temperature much lower than the roasting temperature, thus avoiding the drawbacks of high energy consumption and difficulty in controlling high-temperature operating parameters associated with chain grate machines. The belt roaster, through a multivariate regression model, adjusts roasting parameters to avoid poor pellet performance due to insufficient roasting control precision. Finally, after the finished pellets are generated, the pre-defined ore blending scheme is optimized by using the pellet indicators, thereby improving the pellet blending scheme, reducing energy consumption in pellet production, improving the pellet indicators, and meeting industrial requirements.

[0047] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0048] Figure 1This is a schematic diagram of a pellet production system based on a chain grate machine and a belt roaster, provided by an embodiment of the present invention. Detailed Implementation

[0049] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0050] like Figure 1 As shown, this embodiment of the invention provides a pellet production system based on a chain grate machine and a belt roaster, which includes:

[0051] The raw material processing module is used to generate pelletizing raw materials that conform to the preset particle size distribution according to the preset ore blending scheme;

[0052] The pelletizing process parameter generation module is used to measure the performance of the pelletizing raw materials after pre-activation treatment, and generate pelletizing process parameters based on the performance of the pelletizing raw materials.

[0053] The pelletizing optimization module is used to obtain real-time pelletizing parameters in the pelletizing process of the disc pelletizer, control the equipment parameters of the disc pelletizer in a closed loop based on the pelletizing process parameters and real-time pelletizing parameters, and control the disc pelletizer to generate green pellets from the pelletizing raw materials through the equipment parameters of the disc pelletizer.

[0054] The preheating module of the chain grate machine is used to adjust the preheating parameters of the chain grate machine according to the pelletizing process parameters through a gradient boosting decision tree, and to control the chain grate machine to preheat the green pellets in stages with the preheating parameters to obtain preheated pellets.

[0055] The belt calciner calcination module is used to preheat the pellets by calcining them with a belt calciner and obtain real-time calcination parameters during the calcination process. The real-time calcination parameters are input into an integrated multivariate regression model. Based on the integrated multivariate regression model and the pellet process parameters, calcination adjustment parameters are output. The calcination parameters of the belt calciner are dynamically adjusted according to the calcination adjustment parameters. After the belt calciner completes the calcination process, the pellets are cooled to obtain the finished product.

[0056] The pellet measurement module is used to measure the pellet indicators of the finished pellets and to optimize the preset ore blending scheme based on the pellet indicators.

[0057] Specifically, the pre-determined ore blending scheme is a predetermined raw material ratio scheme for pelletizing. In practice, high-sulfur ore, magnetite concentrate, bentonite, limestone, and other raw materials are precisely batched from their respective silos using electronic belt scales and then transported to a mixer to generate pelletizing raw materials that conform to the pre-determined particle size distribution. The pre-determined particle size distribution is obtained by scientifically planning and designing the particle size of the pelletizing raw materials based on their characteristics and final requirements, resulting in optimal strength, permeability, and sintering properties. Pelletizing process parameters include pre-determined pelletizing indicators, green pellet quality requirements, standard pelletizing parameters, standard particle size, standard preheating parameters, and pelletizing process parameters.

[0058] Real-time pelletizing parameters include acceleration, humidity, and rotational speed. Equipment parameters for disc pelletizers include water spray volume and operating parameters.

[0059] Gradient boosting decision trees are an ensemble learning method that combines multiple decision tree nodes. The ensemble learning method performs fitting calculations by each decision tree node correcting the errors of the previous decision tree node.

[0060] Real-time roasting parameters include temperature, roasting time, and atmosphere. An integrated multivariate regression model is a comprehensive model framework obtained by integrating regression models of multiple variables. Roasting parameters for a belt roaster include roasting temperature, roasting time, and roasting gas concentration.

[0061] The specifications of finished pellets include compressive strength, FeO content, and sulfur content. The compressive strength of finished pellets is generally required to be greater than 2250 N / p or even higher to meet the requirements of subsequent processes such as transportation, storage, and blast furnace smelting, ensuring that the finished pellets do not break during blast furnace charging and descent, and maintaining good permeability. The sulfur content of finished pellets should be controlled at a low level to reduce the sulfur load during blast furnace smelting, lower desulfurization costs, improve the quality of molten iron, and avoid the deterioration of steel properties due to excessive sulfur content. Preferably, the FeO content of finished pellets should be in the range of 1.5%-2%; a lower FeO content indicates that the finished pellets are fully roasted and have a high degree of oxidation, which is beneficial to improving the reducibility and softening properties of the finished pellets, promoting indirect reduction reactions in the blast furnace, reducing the coke ratio, and improving blast furnace production efficiency.

[0062] The first pellet after roasting is cooled in an annular cooler. The second pellet after cooling is screened to remove powder and then the finished pellets are stored in a finished product warehouse or directly transported to the blast furnace for use.

[0063] In one specific embodiment, the pellet process parameter generation module includes:

[0064] The pre-activation treatment unit is used to collect quantitative pelletizing raw materials as test samples. After quality inspection of the test samples, the test samples are pre-activated using chemical methods to obtain standard samples.

[0065] The performance testing unit is used to measure the particle size and bulk density of the standard sample using a laser particle size analyzer to obtain the physical properties of the standard sample; and to measure the chemical properties of the standard sample using a chemical analyzer. The physical and chemical properties of the standard sample constitute the performance of the pelletizing raw material.

[0066] The pelletizing process parameter determination unit inputs the performance of the pelletizing raw materials into a pre-trained pelletizing process feature extraction model, which outputs the pelletizing process features of the raw materials and generates pelletizing process parameters based on these features.

[0067] In the pre-activation treatment unit, the chemical method used in this embodiment of the invention is to adjust the surface properties of the test sample by reacting it with acids, alkalis, or other chemical reagents. For example, certain unwanted components in the ore of the test sample can be removed by acid washing, or the reactivity of the test sample can be improved by adding some chemical substances.

[0068] In the performance testing unit, when measuring the chemical properties of a standard sample using a chemical analyzer, the main elements of the standard sample are measured using the chemical analyzer, and the main elements are used as the chemical properties of the standard sample.

[0069] In the pelletizing process parameter determination unit, the pelletizing process feature extraction model generally uses a random forest model. This model is trained using a large number of historical finished pellets to obtain a well-trained pelletizing process feature extraction model. The input to the pelletizing process feature extraction model is the performance of the pelletizing raw material, and the output is the pelletizing process features. These features include particle size, cohesiveness, and moisture content. The pelletizing process features are combined with existing technical standards, and pelletizing process parameters are generated through rules and logical reasoning. For example, if the particle size feature in the pelletizing process features is 80% of the particles are smaller than 5mm and 20% are larger than 5mm, then according to technical standards, the recommended particle size range is that the proportion of particles smaller than 5mm should reach more than 80%. Since 80% of the particles in this batch of raw material are smaller than 5mm, it meets this standard, and therefore the particle size is considered to meet the requirements. The generated pelletizing process parameters would then be: the rotation speed of the disc pelletizer in the standard pelletizing parameters is 30 rpm.

[0070] In one specific embodiment, the ball-forming optimization module includes:

[0071] The pelletizing data acquisition unit is used to input pelletizing raw materials into the disc pelletizer and obtain real-time pelletizing parameters in the pelletizing process of the disc pelletizer by sensors pre-installed in the disc pelletizer in combination with the pelletizing parameter type.

[0072] The state space construction unit is used to construct the current pelletizing state space based on real-time pelletizing parameters, and to determine the current pelletizing state based on the pelletizing process parameters to obtain the determination result.

[0073] The closed-loop control unit is used to generate a pelletizing adjustment signal based on the judgment result, perform closed-loop control on the equipment parameters of the disc pelletizer based on the pelletizing adjustment signal, and control the disc pelletizer to generate green pellets from the pelletizing raw materials through the equipment parameters of the disc pelletizer.

[0074] The grading and screening unit is used to screen the green pellets generated by the disc pelletizer through a screening device, and to return green pellets whose particle size does not meet the standard particle size in the pelletizing process parameters to the disc pelletizer for reprocessing.

[0075] In the state space construction unit, the pelletizing state space is a multi-dimensional mathematical space, where each dimension corresponds to a type of pelletizing parameter. Each point in the pelletizing state space represents the state of the entire disk pelletizing machine system at a certain moment under the influence of all real-time pelletizing parameters. The pelletizing state at the current moment in the pelletizing state space refers to a health assessment result reflecting the current pelletizing process, obtained based on real-time pelletizing parameters (such as acceleration, humidity, rotational speed, etc.), including both normal and abnormal pelletizing states.

[0076] In a closed-loop control unit, a PID controller, a state feedback controller, or an adaptive controller can be used. The pelletizing adjustment signal is a signal used to indicate the adjustment value of the equipment parameters.

[0077] In the grading and screening unit, green pellets are graded by a screening device. Green pellets that meet the standard particle size are fed into a chain grate machine for drying and preheating, while those that do not meet the standard size are returned to the disc pelletizer for re-pelletizing. The screening device can be a vibrating screen, drum screen, or air classifier, etc.

[0078] In one specific embodiment, the state space construction unit includes:

[0079] The state variable selection subunit is used to construct a covariance matrix based on the real-time ball-making parameters corresponding to all ball-making parameter types, perform eigenvalue decomposition on the covariance matrix to obtain principal component eigenvectors, which are composed of eigenvalues ​​of each ball-making parameter type. The eigenvalues ​​of the ball-making parameter type in the principal component eigenvectors that are less than the spatial threshold are deleted to obtain the state variables and their eigenvalues.

[0080] The state health calculation subunit constructs a ball-making state space based on state variables and their characteristic values, calculates the state health of the ball-making state space at the current moment, and determines that the ball-making state at the current moment is abnormal if the state health of the ball-making state space at the current moment is greater than the health threshold, otherwise it determines that the ball-making state at the current moment is normal.

[0081] In the state variable selection subunit, a covariance matrix is ​​constructed based on the real-time ball-making parameters corresponding to all ball-making parameter types. The principal component eigenvectors are obtained by eigenvalue decomposition of the covariance matrix, which is achieved through formulas (1) and (2):

[0082] (1);

[0083] (2);

[0084] In formula (1), X represents the covariance matrix, x 1m Let x represent the covariance between the real-time ball-forming parameters corresponding to the first ball-forming parameter type and the real-time ball-forming parameters corresponding to the m-th ball-forming parameter type. n1 Let v represent the covariance between the real-time ball-making parameters corresponding to the nth ball-making parameter type and the real-time ball-making parameters corresponding to the 1st ball-making parameter type; in formula (2), v represents the feature matrix obtained based on the covariance matrix, λ represents the principal component eigenvector obtained based on the diagonal elements of the feature matrix, and T represents the transpose of the matrix. Solving the linear equation system of formula (2) yields the principal component eigenvector. The spatial threshold is a mathematical critical value used to screen key principal components, set through empirical values.

[0085] In the state health calculation subunit, the state space for ball-making is constructed based on the state variables and their characteristic values. When calculating the state health H(t) of the ball-making state space at the current time t, it is achieved through formula (3):

[0086] (3);

[0087] In formula (3), This indicates that there are a total of k state variables. This represents the eigenvalue of the i-th state variable. This represents the preset weight of the i-th state variable. The preset weight for each state variable is determined empirically. The health threshold is the critical criterion for distinguishing between normal and abnormal ball-forming states; its general empirical value is 0.5.

[0088] In one specific embodiment, when the closed-loop control unit performs closed-loop control of the equipment parameters of the disc pelletizer according to the pelletizing adjustment signal, it includes: projecting the pelletizing adjustment signal onto a preset equipment feasible domain to obtain the control parameters of the disc pelletizer; distributing the control parameters to the corresponding equipment of the disc pelletizer; adjusting the equipment parameters of each equipment of the disc pelletizer according to the control parameters; acquiring real-time pelletizing parameters during the operation of each equipment of the disc pelletizer according to its equipment parameters; feeding back the real-time pelletizing parameters to the state space construction unit; and performing closed-loop control of the equipment parameters of the disc pelletizer according to the determination result of the state space construction unit.

[0089] Specifically, the preset equipment feasible region is a vector space that defines the process safety boundary of the physically realizable range of the control parameters of the disc pelletizer. The preset equipment feasible region is constructed based on the preset safety parameter range for each equipment parameter, and the pelletizing adjustment signal is projected into the preset equipment feasible region using the rate of change projection method. After feasibility constraint verification, the control parameters of the disc pelletizer are obtained. The corresponding equipment of the disc pelletizer includes a pelletizing disc, a feeder, sensors, and a hopper.

[0090] In one specific embodiment, the chain grate preheating module includes:

[0091] The objective function construction unit is used to construct a preheating prediction model based on the preheating parameter type, generate constraints for each segment based on the pelletizing process parameters and the preheating conditions of each segment, and construct the objective function for each segment based on the constraints and the preheating prediction model.

[0092] The decision tree unit is used to calculate the preheating parameters for each segment based on the gradient boosting decision tree and the objective function of each segment.

[0093] The preheating unit is used to control the chain grate machine to preheat the green pellets in segments according to the preheating parameters of each segment, so as to obtain preheated pellets.

[0094] In the objective function construction unit, preheating parameters include temperature, heating time, and thermal efficiency. The preheating prediction model is a mathematical model that predicts the preheating effect of pellets during the preheating process based on given process parameters, preheating conditions for each segment, and historical data. Green pellets sequentially pass through a forced-air drying section, a forced-air drying section, and a preheating section on a chain grate machine. In the forced-air and forced-air drying sections, hot air evaporates and removes moisture from the green pellets; in the preheating section, the green pellets are heated to a certain temperature to prepare for the subsequent calcination process. The preheating temperature in the forced-air drying section is generally 150-250℃, and the preheating time is 20-40 min. The preheating temperature in the forced-air drying section is generally 300-500℃, and the preheating time is 15-30 min. The preheating temperature in the preheating section is generally 550-800℃, and the preheating time is 10 min. The objective function for each segment is based on the preheating prediction model; the only difference lies in the constraints.

[0095] In the decision tree unit, preheating parameters refer to a series of adjustable process variables during the preheating process of the chain grate, including parameters such as preheating temperature, airflow rate, oxygen concentration, and pressure.

[0096] In the preheating unit, the preheating temperature is crucial during the preheating process. For example, under the pre-defined ore blending scheme with a concentrate ratio of 50% for green pellets, a preheating temperature of 750℃, and a preheating time of 2.5 minutes, the compressive strength of the generated preheated pellets is 446 N. With other conditions unchanged, increasing the preheating temperature from 750℃ to 900℃ results in a compressive strength of 578 N for the generated preheated pellets, an increase of 132 N compared to the 750℃ preheating temperature condition. This demonstrates that increasing the preheating temperature is effective in improving the compressive strength of the pellets.

[0097] In one specific embodiment, the decision tree unit includes:

[0098] The error subunit is used to calculate the error between the predicted result of the objective function of each segment and its target preheating parameter, and to use the error as the input for the next iteration of the gradient boosting decision tree.

[0099] The iterative subunit is used to calculate the error gradient based on the error of the previous iteration of the decision tree, update the function parameters of the objective function of each segment based on the error gradient, obtain the new objective function of each segment, and calculate the error between the prediction result of the new objective function of each segment and its target preheating parameters. The iteration stops when the number of iterations meets the preset number of decision tree nodes, and all error gradients are obtained.

[0100] The output sub-unit is used to fit all error gradients obtained from the gradient boosting decision tree to obtain the updated gradient. The updated gradient and the prediction results of each iteration are weighted and summed to obtain the warm-up parameters for each segment.

[0101] In the error sub-unit, each segment corresponds to a target preheating parameter. The error e between the predicted objective function of any segment and its target preheating parameter at the j-th iteration is calculated. j When, this is achieved through formula (4):

[0102] (4);

[0103] In formula (4), e j-1 Let L represent the error of the (j-1)th iteration, and let L() represent the error function. j Let F(x) represent the prediction result of the j-th iteration, x represent the preheating condition of the segment, F() represent the state response function, and F(x) represent the state response value obtained according to the preheating condition of the segment. Formula (4) calculates the error through the gradient.

[0104] In the iterative subunit, each iteration is based on the decision tree nodes on the gradient boosting decision tree. After iterating over any decision tree node, the error gradient is filled into the decision tree node. When every decision tree node on the gradient boosting decision tree is filled, that is, when the number of iterations meets the preset number of decision tree nodes, the iteration stops.

[0105] In the output sub-unit, the updated gradient is obtained by fitting all error gradients to a pre-trained error decision tree, which is trained using various historical error gradient data. When the updated gradient and the prediction results of each iteration are weighted and summed to obtain the warm-up parameters for each segment, the weight coefficients of the prediction results of each iteration are obtained from the empirical values ​​pre-set on the error decision tree.

[0106] In one specific embodiment, the belt roaster roasting module includes:

[0107] The multivariate regression model building unit is used to generate parameter tensors based on preset pellet indicators and preset roasting parameter types in the pelletizing process parameters, and to build a regression model for each preset roasting parameter type based on the parameter tensors.

[0108] An integration unit is used to calculate the weight coefficient of each preset roasting parameter type based on the real-time roasting parameters during the roasting process, and to generate an integrated multivariate regression model based on the weight coefficient of each preset roasting parameter type and the regression model of each preset roasting parameter type.

[0109] The parameter adjustment unit inputs real-time roasting parameters into an integrated multivariate regression model, which outputs roasting adjustment parameters for each preset roasting parameter type. The roasting parameters of the belt roaster are adjusted in real time according to the roasting adjustment parameters for each preset roasting parameter type. After the belt roaster completes the roasting process with the real-time adjusted roasting parameters, it is cooled to obtain the finished pellets.

[0110] In the multivariate regression model construction unit, the parameter tensor is a high-dimensional data structure (usually a tensor) composed of preset pellet indicators and preset roasting parameter types from the pelletizing process parameters according to certain rules. In this embodiment, the parameter tensor is typically an N×M matrix, where N represents the number of samples (dataset size) and M represents the number of features (input dimension of each sample). Using the standard roasting parameters of each preset roasting parameter type in the parameter tensor as the target variable, an initial regression model is trained based on historical data for each preset roasting parameter type, resulting in a regression model for each preset roasting parameter type.

[0111] The specific method for calculating the weight coefficients of each preset roasting parameter type in the integration unit will be explained in the following embodiments. The integrated multivariate regression model is a framework that includes a regression model for each preset roasting parameter type and its weight coefficients. When real-time roasting parameters are input into the integrated multivariate regression model, the roasting adjustment parameters for each preset roasting parameter type are obtained by combining the outputs of the regression models for each preset roasting parameter type through a weighted sum and intersection method.

[0112] In the parameter adjustment unit, the roasting parameters include temperature distribution, material residence time, and atmosphere. The roasting adjustment parameters are obtained by adjusting the roasting parameters.

[0113] In one specific embodiment, when the integrated unit calculates the weight coefficient of each preset roasting parameter type based on the real-time roasting parameters during the roasting process, the process includes: calculating the parameter sensitivity of each preset roasting parameter type based on the real-time roasting parameters of each preset roasting parameter type; constructing a sensitivity matrix based on the parameter sensitivity of all preset roasting parameter types; calculating the initial weight of each preset roasting parameter type based on the sensitivity matrix; updating the initial reward function based on the real-time roasting parameters to obtain a new reward function; and updating the initial weight of each preset roasting parameter type based on the new reward function to obtain the weight coefficient of each preset roasting parameter type.

[0114] Specifically, when calculating the parameter sensitivity Sq of the q-th preset calcination parameter type, it is achieved through formula (5):

[0115] (5);

[0116] In formula (5), exp() represents the logarithmic function, uq R represents the preset sensitivity weight coefficient for the q-th preset calcination parameter type. q r represents the deviation between the real-time roasting parameter and the standard roasting parameter for the q-th preset roasting parameter type. q This represents the real-time roasting parameter of the q-th preset roasting parameter type. Formula (5) uses the exp function to calculate the parameter sensitivity. The preset sensitivity weight coefficient is a parameter obtained based on historical experience values, representing the degree of influence of each preset roasting parameter type on the sensitivity.

[0117] Specifically, in the sensitivity matrix, each element represents the derivative of the parameter sensitivity of each preset roasting parameter type with the target process index of each preset roasting parameter type. A row of the sensitivity matrix represents a preset roasting parameter type, and a column represents a target process index. The preset weight of each preset roasting parameter type is obtained by obtaining the diagonal elements in the sensitivity matrix, where each diagonal element represents the preset weight of the preset roasting parameter type in its row.

[0118] Furthermore, the system energy consumption of the belt calciner at the current moment is calculated based on the real-time calcination parameters. The penalty term of the initial reward function is updated based on the system energy consumption, and a new reward function is obtained by updating the initial reward function based on the penalty term. The gradient of the initial weights for each preset calcination parameter type is then calculated using the new reward function to obtain the deviation value for each preset calcination parameter type. The deviation value for each preset calcination parameter type is then superimposed onto the initial weights for each preset calcination parameter type to obtain the weight coefficient for each preset calcination parameter type. Preferably, the initial reward function can be a reward function from reinforcement learning.

[0119] Based on all the above embodiments, this invention proposes a pellet production system based on a chain grate machine and a belt roaster. First, the performance of the pelletizing raw materials is obtained, and pelletizing process parameters are generated based on the performance of the pelletizing raw materials. The pelletizing process parameters are optimizable items. In the implementation of this invention, the performance of the finished pellets is improved by dynamically correcting them based on the pellet indicators of the subsequent finished pellets.

[0120] Subsequently, by controlling the equipment parameters of the disc pelletizer in a closed loop, and using these parameters to generate green pellets from the pelletizing raw materials, the precision of green pellet production can be improved, and the quality stability of the green pellets can be enhanced.

[0121] Next, based on the pelletizing process parameters, the preheating parameters of the chain grate machine were adjusted using a gradient boosting decision tree. The chain grate machine was then used to preheat the green pellets in stages to obtain preheated pellets. After that, the preheated pellets were roasted using a belt roaster, and real-time roasting parameters were acquired during the roasting process. Based on the integrated multivariate regression model and the pelletizing process parameters combined with the real-time roasting parameters, the roasting parameters of the belt roaster were dynamically adjusted. After the belt roaster completed the roasting process, the pellets were cooled to obtain the finished pellets. This method combines the convenient control of the chain grate machine with the low energy consumption of the belt roaster during high-temperature roasting, optimizes the roasting regime of the finished pellets, and obtains finished pellets with good indicators, significantly improving the preheating and roasting effects of the finished pellet manufacturing process.

[0122] Finally, the pellet indicators of the finished pellets are measured, and the preset ore blending scheme is optimized in reverse based on the pellet indicators, so that the production process of finished pellets is more precise, flexible and efficient.

[0123] Specifically, when the pellet measurement module optimizes the preset ore blending scheme based on pellet indicators, it considers the possible proportion of concentrate in the finished pellet blending scheme in production practice. This embodiment of the invention conducted experimental research on pelletizing and roasting when the concentrate proportion in the preset ore blending scheme is 30%–70%. The research found that when the concentrate proportion is below 60%, the pelletizing performance of green pellets is poor, requiring an increase in bentonite content or improvement of pelletizing process parameters to improve the drop strength of wet pellets and the compressive strength of green pellets. In specific implementation, two measures were taken: increasing the amount of bentonite and increasing the moisture content of the mixture. Therefore, the optimization direction of the preset ore blending scheme and the corresponding pelletizing results are shown in Table 1. As shown in Table 1, by increasing the proportion of bentonite in the ore blending scheme and appropriately increasing the moisture content of the mixture, the number of green pellet drops and the compressive strength of wet pellets can be significantly improved.

[0124]

[0125] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A pellet production system based on a chain grate machine and a belt roaster, characterized in that, include: The raw material processing module is used to generate pelletizing raw materials that conform to the preset particle size distribution according to the preset ore blending scheme; The pelletizing process parameter generation module is used to measure the performance of the pelletizing raw materials after pre-activation treatment, and generate pelletizing process parameters based on the performance of the pelletizing raw materials. The pelletizing optimization module is used to obtain real-time pelletizing parameters in the pelletizing process of the disc pelletizer, control the equipment parameters of the disc pelletizer in a closed loop based on the pelletizing process parameters and real-time pelletizing parameters, and control the disc pelletizer to generate green pellets from the pelletizing raw materials through the equipment parameters of the disc pelletizer. The ball-forming optimization module includes: The pelletizing data acquisition unit is used to input pelletizing raw materials into the disc pelletizer and obtain real-time pelletizing parameters in the pelletizing process of the disc pelletizer by sensors pre-installed in the disc pelletizer in combination with the pelletizing parameter type. The state space construction unit is used to construct the current pelletizing state space based on real-time pelletizing parameters, and to determine the current pelletizing state based on the pelletizing process parameters to obtain the determination result. The closed-loop control unit is used to generate a pelletizing adjustment signal based on the judgment result, perform closed-loop control on the equipment parameters of the disc pelletizer based on the pelletizing adjustment signal, and control the disc pelletizer to generate green pellets from the pelletizing raw materials through the equipment parameters of the disc pelletizer. The grading and screening unit is used to screen the green pellets generated by the disc pelletizer through a screening device, and to return green pellets whose particle size does not meet the standard particle size in the pelletizing process parameters to the disc pelletizer for reprocessing. The preheating module of the chain grate machine is used to adjust the preheating parameters of the chain grate machine according to the pelletizing process parameters through a gradient boosting decision tree, and to control the chain grate machine to preheat the green pellets in stages with the preheating parameters to obtain preheated pellets. The belt calciner calcination module is used to preheat the pellets by calcining them with a belt calciner and obtain real-time calcination parameters during the calcination process. The real-time calcination parameters are input into an integrated multivariate regression model. Based on the integrated multivariate regression model and the pellet process parameters, calcination adjustment parameters are output. The calcination parameters of the belt calciner are dynamically adjusted according to the calcination adjustment parameters. After the belt calciner completes the calcination process, the pellets are cooled to obtain the finished product. The pellet measurement module is used to measure the pellet indicators of the finished pellets and to optimize the preset ore blending scheme based on the pellet indicators.

2. The pellet production system based on a chain grate machine and a belt roaster according to claim 1, characterized in that, The pelletizing process parameter generation module includes: The pre-activation treatment unit is used to collect quantitative pelletizing raw materials as test samples. After quality inspection of the test samples, the test samples are pre-activated using chemical methods to obtain standard samples. The performance testing unit is used to measure the particle size and bulk density of the standard sample using a laser particle size analyzer to obtain the physical properties of the standard sample; and to measure the chemical properties of the standard sample using a chemical analyzer. The physical and chemical properties of the standard sample constitute the performance of the pelletizing raw material. The pelletizing process parameter determination unit inputs the performance of the pelletizing raw materials into a pre-trained pelletizing process feature extraction model, which outputs the pelletizing process features of the raw materials and generates pelletizing process parameters based on these features.

3. The pellet production system based on a chain grate machine and a belt roaster according to claim 1, characterized in that, The state space construction unit includes: The state variable selection subunit is used to construct a covariance matrix based on the real-time ball-making parameters corresponding to all ball-making parameter types, perform eigenvalue decomposition on the covariance matrix to obtain principal component eigenvectors, which are composed of eigenvalues ​​of each ball-making parameter type. The eigenvalues ​​of the ball-making parameter type in the principal component eigenvectors that are less than the spatial threshold are deleted to obtain the state variables and their eigenvalues. The state health calculation subunit constructs a ball-making state space based on state variables and their characteristic values, calculates the state health of the ball-making state space at the current moment, and determines that the ball-making state at the current moment is abnormal if the state health of the ball-making state space at the current moment is greater than the health threshold, otherwise it determines that the ball-making state at the current moment is normal.

4. The pellet production system based on a chain grate machine and a belt roaster according to claim 1, characterized in that, When the closed-loop control unit performs closed-loop control of the equipment parameters of the disc pelletizer based on the pelletizing adjustment signal, it includes: The pelletizing adjustment signal is projected onto the preset feasible domain of the equipment to obtain the control parameters of the disc pelletizer. The control parameters are distributed to the corresponding equipment of the disc pelletizer. The equipment parameters of each equipment of the disc pelletizer are adjusted according to the control parameters. Real-time pelletizing parameters are obtained during the operation of each equipment of the disc pelletizer according to its equipment parameters. The real-time pelletizing parameters are fed back to the state space construction unit. The equipment parameters of the disc pelletizer are controlled in a closed loop according to the judgment result of the state space construction unit.

5. A pellet production system based on a chain grate machine and a belt roaster according to claim 1, characterized in that, The chain grate machine preheats the green balls in stages, including a forced-air drying stage, a forced-air drying stage, and a preheating stage. The chain grate machine preheating module includes: The objective function construction unit is used to construct a preheating prediction model based on the preheating parameter type, generate constraints for each segment based on the pelletizing process parameters and the preheating conditions of each segment, and construct the objective function for each segment based on the constraints and the preheating prediction model. The decision tree unit is used to calculate the preheating parameters for each segment based on the gradient boosting decision tree and the objective function of each segment. The preheating unit is used to control the chain grate machine to preheat the green pellets in segments according to the preheating parameters of each segment, so as to obtain preheated pellets.

6. A pellet production system based on a chain grate machine and a belt roaster according to claim 5, characterized in that, The decision tree unit includes: The error subunit is used to calculate the error between the predicted result of the objective function of each segment and its target preheating parameter, and to use the error as the input for the next iteration of the gradient boosting decision tree. The iterative subunit is used to calculate the error gradient based on the error of the previous iteration of the decision tree, update the function parameters of the objective function of each segment based on the error gradient, obtain the new objective function of each segment, and calculate the error between the prediction result of the new objective function of each segment and its target preheating parameters. The iteration stops when the number of iterations meets the preset number of decision tree nodes, and all error gradients are obtained. The output sub-unit is used to fit all error gradients obtained from the gradient boosting decision tree to obtain the updated gradient. The updated gradient and the prediction results of each iteration are weighted and summed to obtain the warm-up parameters for each segment.

7. A pellet production system based on a chain grate machine and a belt roaster according to claim 1, characterized in that, The belt roasting machine roasting module includes: The multivariate regression model building unit is used to generate parameter tensors based on preset pellet indicators and preset roasting parameter types in the pelletizing process parameters, and to build a regression model for each preset roasting parameter type based on the parameter tensors. An integration unit is used to calculate the weight coefficient of each preset roasting parameter type based on the real-time roasting parameters during the roasting process, and to generate an integrated multivariate regression model based on the weight coefficient of each preset roasting parameter type and the regression model of each preset roasting parameter type. The parameter adjustment unit inputs real-time roasting parameters into an integrated multivariate regression model, which outputs roasting adjustment parameters for each preset roasting parameter type. The roasting parameters of the belt roaster are adjusted in real time according to the roasting adjustment parameters for each preset roasting parameter type. After the belt roaster completes the roasting process with the real-time adjusted roasting parameters, it is cooled to obtain the finished pellets.

8. A pellet production system based on a chain grate machine and a belt roaster according to claim 7, characterized in that, When the integrated unit calculates the weighting coefficient for each preset roasting parameter type based on the real-time roasting parameters during the roasting process, it includes: Calculate the parameter sensitivity of each preset roasting parameter type based on the real-time roasting parameters of each preset roasting parameter type; Construct a sensitivity matrix based on the parameter sensitivity of all preset roasting parameter types, and calculate the initial weight of each preset roasting parameter type based on the sensitivity matrix; The initial reward function is updated based on the real-time roasting parameters to obtain a new reward function, and the initial weight of each preset roasting parameter type is updated based on the new reward function to obtain the weight coefficient of each preset roasting parameter type.

Images