A method, system, and storage medium for controlling the level of a sinter mix tank

Abstract

This invention provides a method, system, and storage medium for controlling the material level in a sintering mixing tank. The method includes: generating a set of candidate parameters under constraints; predicting the material level in the mixing tank, the proportion of qualified particle size (3mm-8mm) in the mixture, and the total steam consumption based on each candidate parameter set; calculating the comprehensive performance index corresponding to each candidate parameter set based on the prediction results; finding the optimal candidate parameter set among the candidate parameter sets with the goal of minimizing the comprehensive performance index; and distributing the optimal candidate parameter set to the basic control system. This invention solves the problem in existing technologies where it is difficult to comprehensively consider multiple objectives such as material level stability, steam energy consumption, and mixture particle size, leading to the system being in a locally optimal state and unable to achieve global optimization.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for iron ore sintering production processes, specifically to a method, system, and storage medium for controlling the material level in a sintering mixing tank. Background Technology

[0002] In the sintering process, the mixing tank serves as a buffer link between the mixing and pelletizing process and the sintering and feeding process. Stable control of the material level in the mixing tank is crucial for ensuring the continuous and stable operation of the sintering machine. Excessive fluctuations in the material level in the mixing tank directly affect the uniformity of the feeding process in the sintering machine, thereby impacting the yield and quality of the sintered ore.

[0003] Currently, the material level control in sintering mixing tanks mainly employs a single-loop control method based on a PID controller. This method monitors the material level deviation in real time and adjusts operational variables such as the batching conveying rate or the rotational speed of the roller feeder according to a proportional-integral-derivative control law to maintain the material level near the set value. This control method can achieve basic material level stability and is widely used in sintering production.

[0004] However, single-loop PID control has the following technical limitations: This method uses only material level stability as the sole control objective, maintaining the material level by adjusting operational variables such as batching feed rate, mixer moisture setpoint, steam flow rate, or roller speed. However, adjusting these operational variables simultaneously affects the particle size distribution of the mixture and steam energy consumption. For example, increasing the batching feed rate to maintain material level stability will simultaneously change the operating conditions of the secondary mixer, thus affecting the moisture and particle size distribution of the mixture, and steam consumption will also change accordingly. Because single-loop control only focuses on the material level indicator, it cannot comprehensively consider the interrelationships and constraints between multiple production objectives such as material level stability, steam energy consumption, and mixture particle size, leading to the system being in a state of local optimization and making it difficult to achieve global optimization.

[0005] Furthermore, existing control methods exhibit a lag in response to changes in operating conditions. During sintering production, factors such as raw material composition, ambient temperature and humidity change slowly, causing the system's optimal operating point to gradually drift. Traditional control methods rely on operators manually adjusting control parameters periodically based on experience, making it difficult to detect and track these slow changes in a timely manner, causing the system to gradually deviate from its optimal operating state. Simultaneously, different operators differ in their judgment and adjustment strategies regarding changes in operating conditions, making optimization methods difficult to replicate and pass on, and production stability is significantly affected by personnel skill levels.

[0006] Therefore, a new method for controlling the material level in the sintering mixing tank is needed. This method can take into account multiple objectives such as steam energy consumption and particle size quality of the mixture while ensuring stable material level, so as to achieve global collaborative optimization of the system and have the ability to automatically adapt to changes in operating conditions. Summary of the Invention

[0007] The purpose of this invention is to provide a method for controlling the material level in a sintering mixing tank, aiming to solve the problem that existing technologies struggle to comprehensively consider multiple objectives such as material level stability, steam energy consumption, and particle size of the mixture, leading to the system being in a locally optimal state and unable to achieve global optimization. The specific technical solution is as follows: A method for controlling the material level in a sintering mixing tank, comprising the following steps: A1. Generate within constraints candidate parameter groups Where X1 represents the target comprehensive conveying capacity setting value of the batching area, X2 represents the target moisture setting value of the secondary mixer, X3 represents the steam flow rate setting value of the secondary mixer, X4 represents the preheating steam flow rate setting value of the mixing tank, and X5 represents the rotational speed setting value of the roller feeder. It is a natural number greater than or equal to 2; A2, based on each candidate parameter group The material level in the mixing tank, the proportion of qualified particle size (3mm-8mm) in the mixture, and the total steam consumption were predicted separately. Based on the prediction results, candidate parameter groups were calculated. The corresponding comprehensive performance indicators, among which, Represents any candidate parameter group. ; A3. With the goal of minimizing the overall performance index, Find the optimal candidate parameter set from the candidate parameter sets. ; A4. Set the optimal candidate parameters It is then distributed to the basic control system.

[0008] Preferably, based on the first candidate parameter groups To predict the material level in the mixing tank, specifically: Real-time data collection of material conveying parameters in the batching area, operating parameters of the primary mixer, operating parameters of the secondary mixer, operating parameters of the mixing tank, and operating parameters of the sintering machine; After aligning the real-time data collected from the material conveying parameters in the batching area, the operating parameters of the primary mixer, and the operating parameters of the secondary mixer, the data is combined with the real-time data collected from the mixing tank and the sintering machine to construct the current time frame. eigenvectors ; eigenvectors In and candidate parameter groups The corresponding eigenvalues ​​are replaced with , , , as well as The optimized feature vector is obtained. ; The optimized feature vector Input the data into the material level prediction model to predict the material level value at future times. .

[0009] Preferably, the material conveying parameters of the batching area, the operating parameters of the primary mixer, and the operating parameters of the secondary mixer, which are collected in real time, are time-series aligned, specifically as follows: In parameters Searching in the collection sequence The collected values ​​at each time point are used as parameters. At the present moment The values ​​input into the material level prediction model, where, For parameters Response delay time to the material level in the mixing tank. This represents any one of the material conveying parameters in the batching area, the operating parameters of the primary mixer, and the operating parameters of the secondary mixer.

[0010] Preferably, based on the first candidate parameter groups The prediction of the proportion of qualified particle size in the mixture is as follows: Based on the Target moisture setpoint for the secondary mixer in the candidate parameter group and the steam flow rate setpoint of the secondary mixer Calculate the moisture content of the mixture fed into the drum. ; The current mix proportions of each material, pellet mill feed flow rate, pellet mill speed, pellet mill tilt angle, and target moisture setting of the secondary mixer are set. Steam flow rate setpoint for secondary mixer and the moisture content of the mixture fed into the drum The data is input into the particle size prediction model, which predicts the percentage of particles with a qualified particle size of 3mm-8mm. ; The moisture content of the mixture fed into the cylinder Represented as: in, The steam condensation efficiency coefficient, ranging from 0.6 to 0.9, represents the proportion of steam converted into process water. This represents the feed flow rate of the pellet mill, expressed in t / h.

[0011] Preferably, based on the first candidate parameter groups The total steam consumption forecast is as follows: in, Indicates the length of the prediction period. Indicates the first The secondary mixer steam flow rate setpoint in the candidate parameter group. Indicates the first The preheating steam flow rate setting for the mixing tank in the candidate parameter group.

[0012] Preferably, the comprehensive performance index is expressed as follows: in, Indicates overall performance indicators; It is the forecast period Variance of internal material level Indicates to The parameters after dimensionless processing; It is the forecast period Total steam consumption within the facility. Indicates to The parameters after dimensionless processing; It is the forecast period The percentage of internally qualified particles with a size of 3mm-8mm. Indicates to The parameters after dimensionless processing; , and All are weighting coefficients. , , All are greater than or equal to 0 and less than or equal to 1. , , The sum of the three equals 1.

[0013] Preferably, the constraint condition is expressed as follows: in, This indicates the target total conveying capacity setpoint for the batching area. This indicates the target moisture setting value for the secondary mixer. This indicates the setpoint for the steam flow rate of the secondary mixer. This indicates the set value of the preheating steam flow rate in the mixing tank. This indicates the setpoint for the rotational speed of the roller feeder; This indicates the lower limit of the target total conveying capacity in the ingredient preparation area. This indicates the upper limit of the target total conveying capacity in the ingredient preparation area. This indicates the lower limit of the target moisture content for the secondary mixer. This indicates the upper limit of the target moisture content for the secondary mixer. This indicates the lower limit of the steam flow rate of the secondary mixer. This indicates the upper limit of the steam flow rate of the secondary mixer. This indicates the lower limit of the preheating steam flow rate in the mixing tank. This indicates the upper limit of the preheating steam flow rate in the mixing tank. This indicates the lower limit of the rotational speed of the roller feeder. This indicates the upper limit of the rotational speed of the roller feeder.

[0014] Preferably, the material level control method includes a periodic trigger mode, an event trigger mode, and a manual trigger mode; wherein, the event trigger mode refers to performing steps A1-A4 for optimization when the variance of the predicted material level or the total steam consumption continuously deviates from a preset threshold.

[0015] The present invention also provides a material level control system for a sintering mixing tank, including a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor to perform the material level control method.

[0016] The present invention also provides a storage medium storing a computer program, which, when run, executes the material level control method.

[0017] The application of the technical solution of the present invention has the following beneficial effects: First, zero-risk optimization. By conducting massive simulations and optimizations in a high-precision digital twin environment, the risks of direct trial and error in actual production are completely avoided, fundamentally ensuring production safety and the reliability of the optimization process. The virtual evaluation system, based on multiple high-precision prediction models, can accurately predict the impact of candidate parameter combinations on material level, energy consumption, and quality, ensuring that the optimization process takes place entirely in virtual space without causing any disturbance to actual production.

[0018] Second, it achieves coordinated optimization of operating conditions. The focus has shifted from simply controlling material level stability to comprehensively optimizing multiple key operating conditions such as stability, energy consumption, and quality. By constructing a multi-objective comprehensive performance index that includes material level variance, steam consumption, and particle size distribution, and employing a weighted approach to achieve coordinated trade-offs among these objectives, the technical challenges of strong coupling and conflicting objectives among the three indicators have been resolved. Industrial testing shows that, while maintaining material level stability and particle size distribution quality, steam consumption is reduced by approximately 2.9%, material level variance remains below 2.5 t², and particle size distribution is increased by 0.3%.

[0019] Third, it possesses self-optimization capabilities. Through multi-trigger mechanisms and online rolling optimization algorithms, it can automatically adapt to slow changes in process characteristics, promptly detect and track operational drift, and ensure that the mixing tank system continuously maintains optimal operating conditions. The system can automatically complete optimization calculations and parameter adjustments without manual intervention; the optimization methods do not rely on personal experience and are easily replicated and passed on.

[0020] Fourth, the decision-making basis is transparent and credible. The optimization objective function is clearly defined, the weight coefficients can be flexibly adjusted according to production strategies, and the optimization decision-making process is clear, traceable, and easily understood and accepted by on-site operators. The design of the parameter conversion module reflects a deep understanding of the sintering process mechanism, ensuring the accuracy of multi-model collaborative evaluation.

[0021] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0022] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the iron ore sintering process; Figure 2 This is a flowchart of the material level prediction method in the sintering mixing tank in Example 1; Figure 3 This is a diagram of the deep neural network architecture based on gated recurrent units (GRU) in Example 1; Figure 4 This is a flowchart of the sintering mixture tank level control method in Example 2. Detailed Implementation

[0023] To facilitate understanding of the present invention, a more complete description is provided below, along with preferred embodiments. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the present invention.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0025] Example 1: Figure 1This is a schematic diagram of the iron ore sintering process. The batching area is equipped with a return ore trough, a mixing ore trough, a solvent trough, a fuel trough, a dust trough, a dolomite trough, and a quicklime trough. The materials are metered and fed to the mixing-1A belt conveyor by the corresponding batching small belt scales to form a mixture. The mixture is then passed through a primary mixer and a secondary mixer for mixing and granulation before entering a mixing trough for temporary storage. Subsequently, the mixture is transported from the mixing trough to the sintering machine to complete the sintering operation. The sinter discharged from the sintering machine enters an annular cooler for cooling. After cooling, the sinter is classified by a screening system and divided into three paths: one path is transported to the return ore trough for recycling, one path is transported to the bottom material trough as the bottom material for the sintering machine, and the other path is transported to the blast furnace for smelting or sent to the finished ore bin for storage.

[0026] To address the inherent limitations of existing technologies in sensing dimensions, temporal alignment, and model selection, which lead to problems such as lag, inaccuracy, and poor robustness in predicting material levels in mixing troughs, such as... Figure 2 As shown, this embodiment provides a method for predicting the material level in a sintering mixing tank, aiming to accurately depict the causal chain of material flow and achieve early, high-precision material level prediction. The prediction method includes the following steps: S1. Obtain historical material conveying parameters of the batching area, operating parameters of the primary mixer, operating parameters of the secondary mixer, operating parameters of the mixing tank, and operating parameters of the sintering machine; S2. Based on the response delay time of different parameters to the material level in the mixing tank, perform time-series alignment of the historical parameters. S3. Divide the time-series aligned parameters into training, validation, and test sets to complete the training of the prediction model; S4. Deploy the trained prediction model online and collect the material conveying parameters in the batching area, the working parameters of the primary mixer, the working parameters of the secondary mixer, the working parameters of the mixing tank, and the working parameters of the sintering machine. S5. After aligning the real-time collected material conveying parameters of the batching area, the operating parameters of the primary mixer, and the operating parameters of the secondary mixer with the time sequence, input them together with the real-time collected operating parameters of the mixing trough and the sintering machine into the prediction model to predict the material level value of the mixing trough at future times. .

[0027] Steps S1-S5 will be explained in detail below: Preferably, the material conveying parameters of the batching area include the target comprehensive conveying capacity of the batching area X101 and the actual comprehensive conveying capacity of the batching area X102.

[0028] Specifically, the target comprehensive conveying capacity X101 and the actual comprehensive conveying capacity X102 of the batching area can be the actual material conveying capacity of the batching area and the batching error (i.e. the difference between the target comprehensive conveying capacity X101 and the actual comprehensive conveying capacity X102 of the batching area) can be used as inputs to the prediction model. The batching error, as a powerful feedforward signal, can provide early warning of future material level fluctuations caused by inaccurate batching.

[0029] like Figure 1 As shown, the batching area is fed by several small batching belt scales, all of which fall onto the mixing-1A conveyor belt below for conveying. When the mixing-1A conveyor belt starts, feeding begins from the first batching belt. After a delay of a few seconds (usually about 5 seconds), the next batching belt starts, and so on, until the last batching belt starts. This ensures that all materials are fed according to the proportions at startup, without material waste during the transition period. Similarly, if the mixing-1A conveyor belt stops for any reason during normal operation, all batching belts will stop with material on them. Upon startup, they must start simultaneously to ensure that the proportions are not affected. In this embodiment, the actual comprehensive conveying volume X102 of the batching area is the material volume at the material drop point of the last batching belt (i.e., the material volume where all materials converge simultaneously).

[0030] Preferably, the operating parameters of the primary mixer include the primary mixer inlet belt conveyor capacity x103, the primary mixer outlet belt conveyor capacity x104, and the primary mixer outlet material moisture content x105.

[0031] Specifically, the moisture content of the primary mixer outlet material multiplied by 10⁵ is one of the key inputs to the prediction model. This parameter characterizes the material's viscosity and flowability. Abnormal material moisture content (too high / too low) is one of the root causes of abnormal operating conditions such as overflowing or artificially high material levels. Using the primary mixer's operating parameters as input parameters allows the prediction model to learn the impact of changes in these parameters on the material level in the mixing trough, thus ensuring the accuracy of the material level prediction.

[0032] Preferably, the operating parameters of the secondary mixer include: secondary mixer inlet belt conveyor flow rate X106, secondary mixer outlet belt conveyor flow rate X107, secondary mixer outlet material moisture content X108, secondary mixer added steam flow rate X109, secondary mixer added steam pressure X110, secondary mixer added steam temperature X111, the proportion of secondary mixer (granulator) outlet material with a particle size distribution less than 3mm X112, the proportion of secondary mixer (granulator) outlet material with a particle size distribution of 3-8mm X113, and the proportion of secondary mixer (granulator) outlet material with a particle size distribution greater than 8mm X114.

[0033] Specifically, the moisture content of the material at the outlet of the secondary mixer, X108, is one of the key inputs to the prediction model. This parameter is used to characterize the viscosity and flowability of the material. Abnormal material moisture content (too high / too low) is one of the root causes of abnormal operating conditions such as material overflow or artificially high material levels.

[0034] Meanwhile, the steam flow rate X109, steam pressure X110, and steam temperature X111 added by the secondary mixer reflect the final conditioning of the material, affecting its temperature and pre-humidity, and are the final fine-tuning of the material's state before it enters the mixing tank.

[0035] Finally, the proportion of particles smaller than 3mm at the outlet of the secondary mixer (granulator) X112, the proportion of particles between 3-8mm at the outlet of the secondary mixer (granulator) X113, and the proportion of particles larger than 8mm at the outlet of the secondary mixer (granulator) X114 are the key physical properties of the material's bulk density and flowability. Specifically, a high proportion of fine particles (<3mm) leads to a high bulk density, occupying less space in the trough, resulting in a lower material level; at the same time, poor flowability easily causes poor discharge. A high proportion of qualified particles (3-8mm) indicates good material flowability and stable packing, which is an ideal state. A high proportion of coarse particles (>8mm) indicates excessively good material flowability and low bulk density, resulting in a lighter weight for the same volume, which easily leads to artificially high and fluctuating material levels. Parameters X112, X113, and X114 are key to explaining the unexplained fluctuations in the level of the mixing tank. Using them as inputs to the prediction model can directly explain and predict many unexplained level fluctuations, and they are key variables to make up for the blind spots in the existing prediction input parameter system.

[0036] Preferably, the operating parameters of the mixing tank include the inlet conveying rate of the mixing tank X115, the outlet material temperature of the mixing tank X116, the real-time material level of the mixing tank X117, the material level change rate of the mixing tank per unit time X118, the opening degree of each fine-tuning gate at the outlet of the mixing tank X119, the preheating steam flow rate of the mixing tank X120, the preheating steam pressure of the mixing tank X121, and the preheating steam temperature of the mixing tank X122; furthermore, depending on the size of the mixing tank, there are generally 5-8 fine-tuning gates. In this embodiment, the opening degree of each fine-tuning gate is recorded as X119_1-X119_N and input into the prediction model, where N represents the total number of fine-tuning gates.

[0037] Specifically, the operating parameters of the mixing tank can reflect the consumption of the mixture, the trend of material level changes, and momentum information, helping the prediction model learn the correlation between material level changes and various input parameters. Among them, the material temperature X116 at the outlet of the mixing tank can comprehensively reflect the thermal state of the material, indirectly verifying the regulation effect of steam.

[0038] Preferably, the sintering machine operating parameters include the rotary feeder speed x123, the sintering machine trolley speed x124, and the sintering machine trolley material layer thickness x125. Specifically, the sintering machine operating parameters reflect the consumption of the mixture in the mixing tank, helping the prediction model learn the correlation between sintering consumption and material level changes.

[0039] Specifically, in this embodiment, the units of the target comprehensive conveying capacity X101, the actual comprehensive conveying capacity X102, the inlet belt conveying capacity of the primary mixer X103, the outlet belt conveying capacity of the primary mixer X104, the inlet belt conveying capacity of the secondary mixer X106, the outlet belt conveying capacity of the secondary mixer X107, and the inlet conveying capacity of the mixing trough X115 are all t / h; the units of the moisture content of the material at the outlet of the primary mixer X105, the moisture content of the material at the outlet of the secondary mixer X108, the proportion of the particle size distribution of the material at the outlet of the secondary mixer (granulator) less than 3mm X112, the proportion of the particle size distribution of the material at the outlet of the secondary mixer (granulator) with a particle size distribution of 3-8mm X113, and the proportion of the particle size distribution of the material at the outlet of the secondary mixer (granulator) greater than 8mm X114, and the unit hourly rate are all t / h. The units for the material level change rate X118 in the mixing tank and the opening degree of each fine-tuning gate at the outlet of the mixing tank X119 are all %; the units for the steam flow rate X109 added to the secondary mixer and the preheating steam flow rate X120 in the mixing tank are t / h; the units for the steam pressure X110 added to the secondary mixer and the preheating steam pressure X121 in the mixing tank are MPa; the unit for the real-time material level X117 in the mixing tank is t; the unit for the rotational speed X123 of the roller feeder is rpm; the unit for the sintering machine trolley speed X124 is m / min; the unit for the material layer thickness X125 on the sintering machine trolley is mm; and the units for the steam temperature X111 added to the secondary mixer, the material temperature at the outlet of the mixing tank X116, and the preheating steam temperature X122 in the mixing tank are all °C.

[0040] Preferably, the material conveying parameters of the batching area, the operating parameters of the primary mixer, the operating parameters of the secondary mixer, the operating parameters of the mixing tank, and the operating parameters of the sintering machine are obtained by measuring sensors distributed in the sintering equipment.

[0041] Preferably, due to material flow delays during sintering, the response delay times of various parameters to the material level in the mixing tank differ. To align the parameters in time and ensure the accuracy of subsequent predictions, the method for aligning the collected parameters in step S2 is as follows: If parameter The response delay to the material level in the mixing tank is a material flow delay, therefore, in the parameters... Searching in the collection sequence The collected values ​​at each time point are used as parameters. At the present moment The values ​​input into the prediction model, where, For parameters Response delay time to the material level in the mixing tank; If parameter It can affect the material level changes in the mixing tank in real time (i.e., parameters). If there is no response delay for the material level in the mixing tank, then the parameters are directly set. The current time in the acquisition sequence The collected values ​​are input into the prediction model.

[0042] Specifically, parameters If the sampling point is located in front of the mixing tank, the delay is due to the material flow transportation (such as the material transportation parameters in the batching area, the operating parameters of the primary mixer, and the operating parameters of the secondary mixer, i.e., parameters X101-X114). Its response delay time depends on the material flow parameters. The time required for the sample to travel from the collection point to the center of the mixing trough can be calculated based on the material's conveying distance and speed, or determined by analyzing the response curves of historical adjustment records and material level change rates. This type of parameter should be sought in the preceding data collection sequence. The collected value at time t is used as the current time. Input the values ​​into the prediction model.

[0043] Furthermore, the operating parameters of the mixing tank and the sintering machine can be considered as influencing the material level in real time or near real time. Therefore, the response delay time of these parameters to the material level in the mixing tank is... These parameters directly use the current time. The collected values ​​are input into the prediction model. These parameters specifically include: mixing trough inlet conveying rate X115, mixing trough outlet material temperature X116, mixing trough real-time material level X117, mixing trough material level change rate per unit time X118, mixing trough outlet fine-tuning gate opening X119, mixing trough preheating steam flow rate X120, mixing trough preheating steam pressure X121, mixing trough preheating steam temperature X122, roller feeder speed X123, sintering machine trolley speed X124, and sintering machine trolley material layer thickness X125.

[0044] Furthermore, by normalizing the parameters after time alignment, the current time step can be constructed. eigenvectors eigenvectors These are the input parameters of the prediction model, which can be used to predict the material level at future times. .

[0045] Preferably, the prediction model in this embodiment is a Long Short-Term Memory (LSTM) network or a deep neural network architecture based on Gated Recurrent Units (GRUs). In this embodiment, a deep neural network architecture based on GRUs is selected. Compared to LSTM, the deep neural network based on GRUs merges the input gate and forget gate of LSTM into a single update gate, reducing the number of parameters by approximately 25%. While ensuring the ability to effectively capture time-series dependencies of equal length (minutes to tens of minutes) during sintering, the training speed is improved by 30%–40%, better meeting the real-time requirements of industry. In addition, the fewer parameters of the GRU-based deep neural network enable the model to generalize more effectively in industrial scenarios with limited data, significantly reducing the risk of overfitting.

[0046] like Figure 3 As shown, the specific structure of the GRU-based deep neural network is as follows: Input layer: Receives feature vectors after batch normalization. ; Sequence coding layer: A three-layer bidirectional GRU structure is adopted, with the number of units being 128, 64 and 32 respectively. The temporal features are extracted progressively from short-term to long-term. A Dropout layer (dropout rate of 0.2) is connected after the first and second bidirectional GRU layers to suppress overfitting.

[0047] Attention Mechanism Layer: An attention mechanism is applied to the output of the third-layer bidirectional GRU, enabling the model to adaptively focus on the most critical historical moments for the current prediction, thereby enhancing the interpretability of the model.

[0048] Fully connected regression layer: The output of the attention mechanism is sequentially mapped to a 64-dimensional fully connected layer and a 32-dimensional fully connected hidden layer, and finally the predicted material level value is obtained from the linear output layer.

[0049] Furthermore, the training scheme for the prediction model in this embodiment is as follows: Data preparation: Collect no less than 6 months of sintering production data, construct more than 200,000 sets of time-aligned sample data, and after cleaning and normalizing the sample data, divide them into training set, validation set and test set according to proportion.

[0050] Specifically, sample data is cleaned using process rules (upper and lower limits) or the sliding window Z-score method (window=30, threshold=3) to remove outliers generated during the production process, thus ensuring data quality. Scale unification is achieved through a dimensionless method to eliminate dimensional differences between different features, ensuring that all input features are within a comparable numerical range, thereby improving the stability and convergence speed of model training.

[0051] Input sequence construction: except for the current time step eigenvectors In addition, feature vectors from the first 5 time steps are constructed to form a 3D input tensor of [batch size, 6, feature dimension] to help the model capture dynamic trends. Here, batch size represents the number of samples fed into the model for training or prediction in a single run; 6 is the time step size, representing the number of consecutive historical time points contained in each sample input to the model; and feature dimension represents the number of all variables used to describe the system state at each time point (i.e., X101-X125).

[0052] Loss Function: In this embodiment, adaptive Huber Loss is used as the loss function for training the prediction model. Its core design involves gradually decreasing the Huber Loss threshold parameter from an initial value (e.g., 1.0) to a final value (e.g., 0.5) as training epochs increase; or dynamically adjusting it based on the prediction model's performance on the independent validation set—when the prediction model training stabilizes, the threshold is automatically reduced in steps of 0.05 or 0.1 to enhance the loss function's sensitivity to prediction errors, thereby improving the model's fitting accuracy for small deviations.

[0053] Meanwhile, to prevent the GRU prediction model from overfitting to noisy industrial data during training and to improve its generalization ability, an L2 regularization term (with a weight decay coefficient set to 0.001) is introduced on top of the adaptive Huber Loss. This measure effectively constrains the size of the model parameters, making the mapping relationship between the learned input features and the material level smoother and more robust, thereby ensuring the reliability of the prediction system under actual complex working conditions.

[0054] Optimization and Preset Targets: The AdamW optimizer is used, with the learning rate decaying from 0.001 to 0.0001 over 100 epochs using cosine annealing. An early stopping strategy is employed, terminating training if the validation set loss does not decrease for 10 consecutive epochs. On the test set, if the root mean square error (RMSE) < 1.5t, the mean absolute error (MAE) < 1.2t, and 95% of the predicted point absolute errors are ≤ 2.0t, then the preset performance target (≤ 400) is met. The process requirements for sintering machine production capacity are as follows: the larger the sintering machine, the greater the production flow rate and the wider the allowable error range.

[0055] Preferably, after the prediction model is trained, TensorRT is used to optimize its inference, achieving a single prediction time of <10ms on an NVIDIA T4 GPU. Simultaneously, a model performance monitoring system is deployed to continuously track prediction errors. When performance degrades, a retraining process can be automatically triggered, forming a self-evolving capability.

[0056] Furthermore, in this embodiment, at a steel plant with a capacity of 400... Special tests were conducted on the sintering machine. A continuous production period (8 hours) was selected. During this period, the particle size distribution of the material at the outlet of the secondary mixer changed significantly due to adjustments in the upstream raw material ratio, while operating parameters such as the batching rate and roller speed remained constant. The prediction method of this embodiment and the traditional prediction method (a BP neural network model based on flow parameters and a fixed time window) were used to predict the material level, and the results were compared with the actual material level.

[0057] The test period includes three stages: T1, T2, and T3. Stage T1 is 0-120 min, stage T2 is 120-360 min, and stage T3 is 360-480 min. The particle size distribution of each stage is shown in Table 1.

[0058] Table 1. Particle size distribution in different time periods In this test, the prediction model uses the current time. The input parameters are used to predict the material level value 15 minutes later. (i.e., prediction) The material level at time T2. In the T2 stage, the actual material level gradually decreases due to the increase in fine particles (slowly decreasing from the baseline of 35t to 31t). However, the traditional prediction model, which does not introduce particle size parameters, fails to detect the change in physical properties and still predicts that the material level will remain around 35t based on the flow rate data, resulting in a continuous increase in prediction error (the maximum deviation reaches 3.4t).

[0059] The prediction method in this embodiment collects particle size data X112, X113, and X114 in real time and accurately correlates the causal relationship between the current particle size and the subsequent material level through time-series alignment. After capturing the increasing trend of fine particle proportion, the GRU model in this embodiment predicts in advance that the discharge level will gradually decrease, and the predicted value is highly consistent with the actual value. Table 2 shows a comparison of prediction data from typical moments within stage T2. Table 2 Comparison of Predicted Data for Typical Moments in Stage T2 Comparing the results of the prediction method in this embodiment with those of the traditional prediction method, it can be seen that in stage T2, the prediction method of this embodiment can accurately capture the downward trend of material level caused by the increase of fine particles 15 minutes in advance. The predicted value is highly consistent with the actual material level 15 minutes later, with an average absolute error (MAE) of 0.3t, a maximum absolute error of 0.6t, and an error of ≤0.8t at 95% of the prediction points. The traditional prediction method has an average absolute error (MAE) of 2.1t, a maximum absolute error of 3.4t, and an error of ≤3.2t at 95% of the prediction points. This fully demonstrates that particle size is a key physical variable affecting material level, and introducing a prediction model can significantly improve prediction accuracy (especially when material properties fluctuate). The prediction method of this embodiment solves the problems of lag, inaccuracy, and poor robustness of the traditional prediction method.

[0060] The prediction method in this embodiment systematically incorporates material physical property parameters (particle size, moisture) and process parameters into the prediction model, and performs time-series alignment on each parameter, so that the input data has physical completeness and time-series correctness, realizing the synergistic gain effect of the multi-source parameter system.

[0061] Among these parameters, particle size directly affects the bulk density and angle of repose of the material in the mixing tank during sintering. Finer particle size increases bulk density, leading to a higher material level (weight) for the same volume; conversely, finer particle size decreases. The predictive model uses this parameter to detect the invisible, slow drift in material level caused by changes in raw material source or fluctuations in granulation effect. Moisture content significantly affects the flowability and viscosity of the material during sintering. Excessive moisture reduces flowability, making the material prone to adhering in the mixing tank, resulting in poor discharge and a false increase in material level. Insufficient moisture may exacerbate segregation. The predictive model uses this parameter to identify material level distortion or discharge fluctuations caused by abnormal moisture levels. Steam parameters alter the material temperature and surface properties during sintering, indirectly affecting flowability and granulation effect, serving as a crucial bridge variable connecting process operation and material state. Material flow parameters (such as the secondary mixer outlet belt conveyor speed X107), roller feeder speed X123, and sintering machine trolley speed X124 directly reflect the balance between input and output in the mixing tank. The prediction method in this embodiment utilizes the synergistic effects of material conveying parameters in the batching area, operating parameters of the primary mixer, operating parameters of the secondary mixer, operating parameters of the mixing trough, and operating parameters of the sintering machine to construct a virtual material state-dynamic equilibrium digital image in the prediction model. Therefore, the system can provide high-precision early warnings of material level changes caused by gradual changes in physical properties or operational adjustments 5-15 minutes in advance. The measured absolute error of over 95% of prediction points is ≤2.0 tons, far superior to models relying solely on historical material level data.

[0062] The prediction method in this embodiment addresses the problem of traditional methods that simply stack parameters collected at different locations and times, thus disrupting the strict causal relationship between the model input features and the prediction target (material level). By aligning the parameters over time, this method ensures that the prediction model learns genuine physical causal relationships, rather than spurious correlations. Therefore, when faced with unfamiliar operating conditions or disturbances, the model can make more reasonable inferences based on physical logic, exhibiting stronger generalization ability and robustness, and reducing the risk of overfitting.

[0063] This embodiment employs a GRU-based deep neural network, which, while ensuring the ability to capture long-term temporal dependencies, has fewer parameters, higher training efficiency, and is easier to deploy in industrial real-time systems. Furthermore, the simpler structure helps reduce the risk of overfitting and improves model stability in industrial scenarios where the data volume is not infinite. The adaptive Huber loss function and L2 regularization used during predictive model training further ensure the robustness and generalization of the predictive model when dealing with industrial data noise. This embodiment's method, through mechanism-based multi-source parameter selection and temporal alignment, completes information purification at the input level, using the most relevant and timely data to drive the model, avoiding the computational burden and unreliability caused by brute-force fitting with large datasets.

[0064] The predictive method in this embodiment transforms material level control from a post-event reaction to a pre-event intervention, achieving an upgrade in the control paradigm. Simultaneously, its clear input-output relationship makes the decision-making process of the entire intelligent system interpretable and traceable, providing a high-quality, high-reliability perception core for constructing a digital twin of the sintering process and realizing full-process automation.

[0065] Example 2: See Figure 4 This embodiment provides a material level control method for a sintering mixing tank, aiming to solve the problem that existing technologies struggle to comprehensively consider multiple objectives such as material level stability, steam energy consumption, and particle size of the mixture, leading to the system being in a local optimum state and unable to achieve global optimization. The specific details are as follows: A1. Generate within constraints candidate parameter groups ; Where X1 represents the target total conveying capacity of the batching area. The unit is t / h, and X2 represents the target moisture setpoint for the secondary mixer. The unit is %, and X3 represents the setpoint for the steam flow rate of the secondary mixer. The unit is t / h, and X4 represents the set value of the preheating steam flow rate of the mixing tank. The unit is t / h, and X5 represents the set value of the rotational speed of the roller feeder. The unit is rpm. It is a natural number greater than or equal to 2; A2, based on each candidate parameter group The material level in the mixing tank, the proportion of qualified particle size (3mm-8mm) in the mixture, and the total steam consumption were predicted separately. Based on the prediction results, candidate parameter groups were calculated. The corresponding comprehensive performance indicators, among which, Represents any candidate parameter group. ; A3. With the goal of minimizing the overall performance index, Find the optimal candidate parameter set from the candidate parameter sets. ; A4. Set the optimal candidate parameters It is then distributed to the basic control system.

[0066] The material level control method of this embodiment will be described in detail below: The material level control method in this embodiment includes a periodic trigger mode, an event trigger mode, and a manual trigger mode. The periodic trigger mode automatically initiates and executes steps A1-A4 for optimization at a fixed cycle, for example, a cycle set to 30 to 60 minutes, preferably 60 minutes. The event trigger mode performs optimization when the predicted material level variance or total steam consumption continuously deviates from a preset threshold, for example, when the predicted material level variance exceeds 3.0 for three consecutive cycles. Optimization is triggered immediately when the total steam consumption exceeds 2.0 t / h for three consecutive cycles. The manual triggering mode allows operators to manually request system optimization based on production needs. This embodiment's multi-trigger mechanism improves the timeliness of optimization, enabling regular optimization to maintain the system's optimal state, rapid response to abnormal operating conditions, and retaining the flexibility of manual intervention.

[0067] Specifically, the constraints are set to limit the optimization search space to a safe process range, preventing optimization results from exceeding equipment capabilities or violating process requirements. In this embodiment, the constraints are expressed as follows: in, This indicates the target total conveying capacity setpoint for the batching area. This indicates the target moisture setting value for the secondary mixer. This indicates the setpoint for the steam flow rate of the secondary mixer. This indicates the set value of the preheating steam flow rate in the mixing tank. This indicates the setpoint for the rotational speed of the roller feeder; This indicates the lower limit of the target total conveying capacity in the ingredient preparation area. This indicates the upper limit of the target total conveying capacity in the ingredient preparation area. This indicates the lower limit of the target moisture content for the secondary mixer. This indicates the upper limit of the target moisture content for the secondary mixer. This indicates the lower limit of the steam flow rate of the secondary mixer. This indicates the upper limit of the steam flow rate of the secondary mixer. This indicates the lower limit of the preheating steam flow rate in the mixing tank. This indicates the upper limit of the preheating steam flow rate in the mixing tank. This indicates the lower limit of the rotational speed of the roller feeder. This indicates the upper limit of the rotational speed of the roller feeder.

[0068] Preferably, based on the first candidate parameter groups To predict the material level in the mixing tank, specifically: Real-time data collection of material conveying parameters in the batching area, operating parameters of the primary mixer, operating parameters of the secondary mixer, operating parameters of the mixing tank, and operating parameters of the sintering machine; After aligning the real-time data collected from the material conveying parameters in the batching area, the operating parameters of the primary mixer, and the operating parameters of the secondary mixer, the data is combined with the real-time data collected from the mixing tank and the sintering machine to construct the current time frame. eigenvectors ; eigenvectors In and candidate parameter groups The corresponding eigenvalues ​​are replaced with , , , as well as The optimized feature vector is obtained. ; The optimized feature vector Input the data into the material level prediction model to predict the material level value at future times. .

[0069] The material level prediction model in this embodiment is the same as that in Embodiment 1, and the number of variables input into the material level prediction model is exactly the same. Therefore, this embodiment only focuses on how to obtain the feature vector. Detailed explanation: eigenvectors The characteristic value of the target comprehensive conveying capacity X101 in the middle batching area is replaced with... Replace the characteristic value of the moisture content of the material at the outlet of the secondary mixer with 108. Replace the characteristic value of adding steam flow rate X109 to the secondary mixer with Replace the characteristic value of the preheating steam flow rate X120 in the mixing tank with... Replace the characteristic value of the roller feeder speed X123 with Other unoptimized parameters remain unchanged.

[0070] Therefore, it can be predicted that according to the candidate parameter set... The adjusted material level value, in this embodiment, forms a complete model input feature vector by capturing the current process state parameters, providing complete initial conditions for virtual simulation, ensuring the accuracy of material level prediction, and ensuring that the subsequent material level, total steam consumption, and qualified particle size ratio meet the actual production needs after adjustment according to the optimal candidate parameters, thus achieving global optimization.

[0071] This embodiment performs time-series alignment on the material conveying parameters of the batching area, the operating parameters of the primary mixer, and the operating parameters of the secondary mixer collected in real time. Specifically: In parameters Searching in the collection sequence The collected values ​​at each time point are used as parameters. At the present moment The values ​​input into the material level prediction model, where, For parameters Response delay time to the material level in the mixing tank. This represents any one of the material conveying parameters in the batching area, the operating parameters of the primary mixer, and the operating parameters of the secondary mixer.

[0072] For example, if the response delay time of the overall conveying capacity of the batching area to the material level is 8 minutes, then search for the material level in the collection sequence of the overall conveying capacity of the batching area. The data collected at -8 minutes is used as the current input to the level prediction model. Time-series alignment takes into account the time delay characteristics of the impact of different parameters on the level, improving the accuracy of the prediction model. For details regarding time-series alignment not explicitly stated, please refer to Example 1.

[0073] Preferably, based on the first candidate parameter groups The prediction of the proportion of qualified particle size in the mixture is as follows: Based on the Target moisture setpoint for the secondary mixer in the candidate parameter group and the steam flow rate setpoint of the secondary mixer Calculate the moisture content of the mixture fed into the drum. ; The moisture content of the mixture fed into the cylinder Represented as: in, The steam condensation efficiency coefficient is taken as 0.6 to 0.9 (t condensate / t steam) depending on the process conditions. It represents the proportion of steam converted into process water, and the specific value is obtained by sampling and testing the inlet and outlet water ratios on site. The feed flow rate of the pellet mill is obtained through material tracking or metering belt scale, and the unit is t / h.

[0074] The current mix proportions of each material, pellet mill feed flow rate, pellet mill speed, pellet mill tilt angle, and target moisture setting of the secondary mixer are set. Steam flow rate setpoint for secondary mixer and the moisture content of the mixture fed into the drum The data is input into the particle size prediction model, which predicts the percentage of particles with a qualified particle size of 3mm-8mm. ; Specifically, the particle size prediction model in this embodiment is detailed in Chinese patent application CN113962150A, which discloses a method and system for predicting the particle size of sintered mixtures. The variables input to the particle size prediction model in this embodiment are consistent with those in the patent application, wherein the target moisture content of the secondary mixer is set according to the target moisture content of the secondary mixer. Moisture content of the mixture fed into the drum The real-time water addition amount, which is the input variable in this patent, can then be calculated. For details regarding the particle size prediction model, please refer to Chinese patent application CN113962150A.

[0075] Of course, in some embodiments, other machine learning-based regression prediction models may be used as granular prediction models. Machine learning-based regression prediction models are common knowledge in the field, so they will not be described in detail in this embodiment.

[0076] Preferably, based on the first candidate parameter groups The total steam consumption forecast is as follows: in, Indicates the length of the prediction period. Indicates the first The secondary mixer steam flow rate setpoint in the candidate parameter group. Indicates the first The preheating steam flow rate setting for the mixing tank in the candidate parameter group.

[0077] In this embodiment, the total steam consumption is obtained by integrating the setpoints for the secondary mixer steam flow rate and the preheating steam flow rate of the mixing tank over the predicted time period. This lightweight estimation method is simple and fast to calculate, and can meet the real-time requirements of online optimization.

[0078] Preferably, the comprehensive performance index is expressed as follows: in, Indicates overall performance indicators; It is the forecast period Variance of internal material level Indicates to Perform dimensionless processing; It is the forecast period Total steam consumption within the facility. Indicates to Perform dimensionless processing; It is the forecast period The percentage of internally qualified particles with a size of 3mm-8mm. Indicates to Perform dimensionless processing; , and All are weighting coefficients. , , All are greater than or equal to 0 and less than or equal to 1. , , The sum of the three equals 1.

[0079] Specifically, weighting coefficients , and Configured according to production strategy. During production phases where maximum stability is prioritized, the configuration is as follows: =0.8、 =0.1、 =0.1, at this point the optimization objective is primarily material level stability. During the cost reduction and efficiency improvement phase, the configuration is... =0.5、 =0.4、 =0.1, at this point the optimization objective balances material level stability and steam energy consumption. During the quality assurance phase, the configuration is as follows: =0.4、 =0.3、 =0.3, at this point the optimization objective comprehensively considers three indicators. A quantitative trade-off between multiple objectives is achieved through weighting coefficients, and the weights can be flexibly adjusted according to the production strategy, making the optimization objective clear and adjustable.

[0080] Furthermore, in this embodiment, a dimensionless processing method is used to standardize the scale of each indicator, eliminating the differences in dimensions between different indicators and making them comparable within the same numerical range. Reference values ​​are preset based on process characteristics, such as material level variance. Divide by reference value (e.g., taking the historical average variance of 2.5) Total steam consumption Divide by reference value (e.g., normal supply flow rate 1.0 t / h), particle size distribution Using itself (already a percentage), the dimensionless indicators are denoted as follows: , , .

[0081] Preferably, this embodiment uses a particle swarm optimization algorithm to generate candidate parameter sets. At the start of the algorithm, a number of particles (e.g., 50) are randomly created, and the position of each particle corresponds to a candidate parameter combination. For particles... The initial value of each variable X is randomly selected within the constraints, for example, X1 in [ , X2 takes a random value within the interval [ ], and X2 is in [ , Random values ​​are selected within the interval; the same applies to X3, X4, and X5. The particle swarm optimization algorithm iteratively updates particle positions. Each particle moves closer to its historically best position based on individual experience, and simultaneously moves closer to the globally best position currently found by the entire particle swarm based on swarm experience. After multiple iterations, all particles gradually move towards the objective function. The smallest combination of parameters converges.

[0082] Specifically, with the goal of minimizing the overall performance index, in Find the optimal candidate parameter set from the candidate parameter sets. The particle swarm optimization algorithm iteratively calculates the comprehensive performance index corresponding to each candidate parameter set. Compare all candidate parameter sets Value, ultimately finding the one that makes The smallest possible set of candidate parameters is selected. The algorithm iterates 50 to 100 times, terminating early when the objective function value changes by less than 0.001 over 10 consecutive iterations. The optimization computation time is controlled within 10 seconds to meet the real-time requirements of online optimization.

[0083] Preferably, in this embodiment, the optimal candidate parameter group is... When the data is sent to the basic control system, a rate of change limit is imposed on each optimal candidate parameter to ensure that it is smoothly transferred to the basic control system at a preset maximum rate of change, thus avoiding any impact on the production process.

[0084] In this embodiment, the target comprehensive conveying capacity variation rate in the batching area is limited to ±5 t / h / min, the target moisture content variation rate in the secondary mixer is limited to ±0.1% / min, the steam flow rate variation rate in the secondary mixer is limited to ±0.05 t / h / min, the preheating steam flow rate variation rate in the mixing tank is limited to ±0.02 t / h / min, and the rotational speed variation rate in the roller feeder is limited to ±0.1 rpm / min. These variation rate limits are adjusted based on the dynamic characteristics of the specific process equipment, reflecting their flexibility and configurability.

[0085] Of course, in some embodiments, it may be feasible to directly send the optimal candidate parameters to the basic control system without imposing a rate of change limit.

[0086] Industrial Case Studies: The following industrial test case illustrates the technical effectiveness of the material level control method in this embodiment. The control method of this embodiment was applied in a 400-ton steel plant in China. The sintering machine underwent a month-long industrial deployment and testing.

[0087] In this case, particle size quality is determined by constraints. ≥58% is guaranteed, therefore the weight of the granularity term in the objective function is 0, and the objective function is set as follows: The system detected that the average steam flow rate remained high for 60 minutes in the previous operating cycle, triggering the event trigger mode optimization. The key operating parameters (before optimization) at the trigger time (T0) are shown in Table 3: Table 3 Key operating parameters at trigger time The system makes a judgment at time T0. If the output exceeds the expected threshold (2.0 t / h), optimization is automatically triggered. The particle swarm optimization algorithm, using the current state as a baseline, completed thousands of virtual simulations within approximately 10 seconds. Through iteration, a better set of setpoints was found: X2 (moisture setpoint) was set to 7.45% (increased by 0.05%), X3 (mixer steam) was set to 1.80 t / h (decreased by 0.05 t / h), and X1, X4, and X5 remained unchanged.

[0088] Based on model evaluation, after adopting this new set of settings, the following prediction variance for the material level within the next hour is expected: It will remain at approximately 2.5. After dimensionless Percentage of qualified particle size It will increase slightly to 58.8%, with average total steam consumption... It is expected to decrease to 2.00 t / h, after dimensionless conversion. Before optimization, after dimensionless transformation , objective function value =0.7×1.0 + 0.3×2.05 = 1.315. After optimization, the objective function value is... =0.7×1.0 + 0.3×2.0 = 1.300, a decrease of approximately 1.14%.

[0089] The optimized parameter values ​​are smoothly transmitted to the basic control system after being limited by the rate of change. After the next operating cycle (T0 to T0+60 minutes) is completed, the actual operating data is shown in Table 4. Table 4 Optimized Key Operating Condition Parameters As shown in Table 4, after optimization, the moisture setpoint was increased to 7.45%, the mixer steam setpoint was decreased to 1.80 t / h, and other optimized variables remained unchanged. The actual variance of the material level was 2.4. Compared to before optimization, the efficiency remained stable with a fluctuation of only -0.1t². Total steam consumption decreased to 1.99t / h, a reduction of 0.06t / h, or approximately 2.9%, compared to before optimization. The average qualified particle size ratio increased to 58.8%, remaining stable with a slight improvement compared to before optimization, representing an increase of 0.3%.

[0090] The material level control method provided in this embodiment achieves three technical effects. First, zero-risk safety optimization. By conducting massive simulations and optimizations in a high-precision digital twin environment, the risk of direct trial and error in actual production is completely avoided, fundamentally ensuring production safety and the reliability of the optimization process. The virtual evaluation system, based on material level prediction models, particle size prediction models, and steam consumption estimation models, can accurately predict the impact of candidate parameter combinations on material level, energy consumption, and quality, allowing the optimization process to be carried out entirely in virtual space without causing any disturbance to actual production. Second, achieving collaborative optimization of operating conditions. It elevates the optimization from simple material level stability control to comprehensive optimization of multiple key operating condition indicators such as stability, energy consumption, and quality. By constructing a multi-objective comprehensive performance index including material level variance, steam consumption, and particle size distribution, and using a weighted approach to achieve collaborative trade-offs among multiple objectives, the technical challenges of strong coupling and objective conflict among the three indicators are solved. Industrial testing shows that, while maintaining material level stability and particle size quality, steam consumption is reduced by approximately 2.9%, material level variance remains below 2.5t², and particle size distribution is increased by 0.3%. Third, it possesses self-optimization capabilities. Through multi-trigger mechanisms and online rolling optimization algorithms, it can automatically adapt to slow changes in process characteristics, promptly detect and track operational drift, and ensure that the mixing tank system continuously maintains optimal operating conditions. The system can automatically complete optimization calculations and parameter adjustments without manual intervention; the optimization methods do not rely on personal experience and are easily replicated and passed on.

[0091] Example 3: This embodiment provides a material level control system for a sintering mixing tank. The material level control system includes a memory and a processor. The memory stores a computer program, and the computer program is executed by the processor to perform the material level control method in Embodiment 2.

[0092] Example 4: This embodiment provides a storage medium storing a computer program, which, when run, executes the material level control method in Embodiment 2.

[0093] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of controlling the level of a sinter mix bin, characterised by, Includes the following steps: A1, generating within constraints a candidate parameter set ; wherein X1 represents a target total conveying amount setting value of the batching area, X2 represents a target moisture setting value of the secondary mixer, X3 represents a steam flow setting value of the secondary mixer, X4 represents a preheating steam flow setting value of the mixing tank, X5 represents a speed setting value of the circular roller feeder, is a natural number greater than or equal to 2; A2, based on each candidate parameter group The material level in the mixing tank, the proportion of qualified particle size (3mm-8mm) in the mixture, and the total steam consumption were predicted separately. Based on the prediction results, candidate parameter groups were calculated. The corresponding comprehensive performance indicators, among which, Represents any candidate parameter group. ; A3. With the goal of minimizing the overall performance index, Find the optimal candidate parameter set from the candidate parameter sets. ; A4. Set the optimal candidate parameters It is then distributed to the basic control system.

2. The material level control method according to claim 1, characterized in that, Based on the candidate parameter groups To predict the material level in the mixing tank, specifically: Real-time data collection of material conveying parameters in the batching area, operating parameters of the primary mixer, operating parameters of the secondary mixer, operating parameters of the mixing tank, and operating parameters of the sintering machine; After aligning the real-time data collected from the material conveying parameters in the batching area, the operating parameters of the primary mixer, and the operating parameters of the secondary mixer, the data is combined with the real-time data collected from the mixing tank and the sintering machine to construct the current time frame. eigenvectors ; eigenvectors In and candidate parameter groups The corresponding eigenvalues ​​are replaced with , , , as well as The optimized feature vector is obtained. ; The optimized feature vector Input the data into the material level prediction model to predict the material level value at future times. .

3. The material level control method according to claim 2, characterized in that, The real-time collected material conveying parameters of the batching area, the operating parameters of the primary mixer, and the operating parameters of the secondary mixer are time-series aligned, specifically as follows: In parameters Searching in the collection sequence The collected values ​​at each time point are used as parameters. At the present moment The values ​​input into the material level prediction model, where, For parameters Response delay time to the material level in the mixing tank. This represents any one of the material conveying parameters in the batching area, the operating parameters of the primary mixer, and the operating parameters of the secondary mixer.

4. The material level control method according to claim 1, characterized in that, Based on the candidate parameter groups The prediction of the proportion of qualified particle size in the mixture is as follows: Based on the Target moisture setpoint for the secondary mixer in the candidate parameter group and the steam flow rate setpoint of the secondary mixer Calculate the moisture content of the mixture fed into the drum. ; The current mix proportions of each material, pellet mill feed flow rate, pellet mill speed, pellet mill tilt angle, and target moisture setting of the secondary mixer are set. Steam flow rate setpoint for secondary mixer and the moisture content of the mixture fed into the drum The data is input into the particle size prediction model, which predicts the percentage of particles with a qualified particle size of 3mm-8mm. ; The moisture content of the mixture fed into the cylinder Represented as: in, The steam condensation efficiency coefficient represents the proportion of steam converted into process water. This refers to the feed flow rate of the pellet mill.

5. The material level control method according to claim 1, characterized in that, Based on the candidate parameter groups The total steam consumption forecast is as follows: in, Indicates the length of the prediction period. Indicates the first The secondary mixer steam flow rate setpoint in the candidate parameter group. Indicates the first The preheating steam flow rate setting for the mixing tank in the candidate parameter group.

6. The material level control method according to claim 1, characterized in that, The comprehensive performance index is expressed as follows: in, Indicates overall performance indicators; It is the forecast period Variance of internal material level Indicates to The parameters after dimensionless processing; It is the forecast period Total steam consumption within the facility. Indicates to The parameters after dimensionless processing; It is the forecast period The percentage of internally qualified particles with a size of 3mm-8mm. Indicates to The parameters after dimensionless processing; , and All are weighting coefficients. , , All are greater than or equal to 0 and less than or equal to 1. , , The sum of the three equals 1.

7. The material level control method according to claim 1, characterized in that, The constraint condition is expressed as follows: in, This indicates the target total conveying capacity setpoint for the batching area. This indicates the target moisture setting value for the secondary mixer. This indicates the setpoint for the steam flow rate of the secondary mixer. This indicates the set value of the preheating steam flow rate in the mixing tank. This indicates the setpoint for the rotational speed of the roller feeder; This indicates the lower limit of the target total conveying capacity in the ingredient preparation area. This indicates the upper limit of the target total conveying capacity in the ingredient preparation area. This indicates the lower limit of the target moisture content for the secondary mixer. This indicates the upper limit of the target moisture content for the secondary mixer. This indicates the lower limit of the steam flow rate of the secondary mixer. This indicates the upper limit of the steam flow rate of the secondary mixer. This indicates the lower limit of the preheating steam flow rate in the mixing tank. This indicates the upper limit of the preheating steam flow rate in the mixing tank. This indicates the lower limit of the rotational speed of the roller feeder. This indicates the upper limit of the rotational speed of the roller feeder.

8. The material level control method according to claim 1, characterized in that, The material level control method includes a periodic trigger mode, an event trigger mode, and a manual trigger mode; wherein, the event trigger mode refers to executing steps A1-A4 for optimization when the variance of the predicted material level or the total steam consumption continuously deviates from the preset threshold.

9. A material level control system for a sintering mixing tank, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor to perform the material level control method as described in any one of claims 1-8.

10. A storage medium, characterized in that, The storage medium stores a computer program, which, when run, executes the material level control method as described in any one of claims 1-8.

Images