Low-energy-consumption furnace charge steady-state control method for discharging magnesium from bottom of vertical tank by Pidgeon method

Abstract

The invention provides a Pidgeon method vertical tank bottom magnesium discharging low-energy-consumption furnace charge steady-state control method which comprises the following steps: acquiring real-time distribution data of particles from a furnace charge feeding system, extracting position and density information of the particles by utilizing an image recognition algorithm, calculating a distribution uniformity index in combination with a preset uniform distribution model, and calculating the distribution uniformity of the particles according to the distribution uniformity index. The discrete degree of particle distribution is quantified; evaluating the parameter correlation matrix, judging whether parameter conflicts exist or not, and if the conflicts are detected, optimizing parameter combinations through a genetic algorithm, obtaining process parameter data again and updating the parameter correlation matrix until the parameters are coordinated and consistent; and constructing a comprehensive prediction model based on a machine learning algorithm by utilizing the updated distribution uniformity index, the heat transfer cumulative change trend and the coordinated and consistent process parameter combination, and predicting the reduction reaction efficiency and the product quality.

Description

Technical Field

[0001] This invention relates to the field of magnesium bottom tapping technology in the Pidgeon process, and particularly to a low-energy-consumption steady-state control method for magnesium bottom tapping in the Pidgeon process. Background Technology

[0002] In the bottom-out magnesium reduction process of the Pidgeon process, a complex technical problem exists involving the steady-state feeding and distribution of low-energy-consumption charge under multi-source excitation. Precise control of charge particle size is required during feeding to meet the demands of high efficiency and large charge capacity in the vertical regenerative reduction furnace. However, the distribution of charge particles directly affects the gas flow distribution within the furnace, thus impacting the uniformity and efficiency of the reduction reaction. To achieve graded charging and central coking, it is necessary to address how to maintain a stable charge distribution during feeding, avoiding gas flow segregation or localized overheating due to uneven particle distribution, which would negatively affect the overall reduction effect. Another technical challenge lies in the coupling and cumulative effects of heat transfer during bottom-out magnesium reduction. Multiple physical fields interact within the furnace, including temperature, stress, and flow fields. The coupling between these fields is not only reflected in the heat transfer process between the fluid and solid but also involves complex phenomena such as flow, radiation, and ablation at the fluid-solid interface. Particularly in the reduction reaction, the existence of conjugate heat transfer causes the heat transfer effect to accumulate continuously within the furnace, thus affecting the magnesium reduction mechanism. Accurately describing and predicting this coupling effect and its cumulative phenomenon is key to improving reduction efficiency. Furthermore, optimizing the process parameters of the bottom-outlet magnesium reduction furnace also faces challenges. Key process parameters such as temperature, pressure, and flow rate directly impact energy consumption and magnesium product quality. However, these parameters exhibit complex interrelationships; adjusting a single parameter can trigger a chain reaction on others. Achieving precise control and optimization within the context of multi-parameter coupling is the core issue for improving product quality and reducing energy consumption. Summary of the Invention

[0003] This invention provides a method for steady-state control of low-energy-consumption furnace charge at the bottom of the Pidgeon process vertical treasury, mainly comprising: S1. Obtain real-time particle distribution data from the furnace charge feeding system, extract particle position and density information using image recognition algorithms, and calculate the distribution uniformity index by combining a uniform distribution model; S2. Analyze the distribution uniformity index. If it is lower than the preset threshold, adjust the feeding speed and feeding angle, reacquire particle distribution data and update the index until it meets the standard. S3. When the uniformity index of distribution meets the standard, acquire the temperature field and flow field data in the furnace, use the particle distribution data as the boundary condition, and calculate the heat transfer and flow field coupling effect through the heat conduction model and the fluid dynamics model to obtain the heat distribution characteristics. S4. Based on the heat distribution characteristics, the cumulative heat transfer trend is predicted by combining the heat balance equation Q=mcΔT. If there are local overheated areas, the gas flow rate and temperature distribution are adjusted, the coupling effect is recalculated and the trend is updated until the overheating is eliminated. S5. When the cumulative heat transfer trend is stable, real-time data of process parameters including temperature, pressure and flow rate are obtained. A multi-parameter correlation model is constructed based on statistical regression methods to analyze the mutual influence of parameters and obtain the parameter correlation matrix. S6. Evaluate the parameter correlation matrix. If parameter conflicts are detected, optimize the parameter combination using a genetic algorithm, reacquire process data, and update the matrix until the parameters are consistent. S7. Using the updated distribution uniformity index, heat transfer accumulation change trend and coordinated process parameter combination, a comprehensive prediction model is constructed based on machine learning algorithm to predict reduction reaction efficiency and product quality; S8. Analyze the prediction results of the comprehensive prediction model. If the process conditions are not optimal, generate a parameter adjustment plan, reacquire the furnace charge distribution data, heat transfer data and process parameter data, and update the prediction results until the process conditions reach the optimal state.

[0004] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a low-energy-consumption steady-state control method for magnesium production in the Pidgeon process vertical furnace, addressing the issues of gas flow segregation and localized overheating caused by uneven particle distribution in the furnace charge. The invention acquires particle distribution data in real time, extracts position and density information using image recognition algorithms, calculates distribution uniformity indices using a uniform distribution model, and adjusts the feed rate and angle accordingly to achieve steady-state charging. Simultaneously, the invention acquires furnace temperature and flow field data, uses updated particle distribution as boundary conditions, calculates the coupling effect of heat transfer and flow field using heat conduction and fluid dynamics models, predicts the cumulative heat transfer trend, and adjusts the gas flow rate and temperature distribution accordingly to eliminate localized overheating. Furthermore, the invention analyzes the interrelationships between process parameters based on a multi-parameter correlation model, optimizes parameter combinations using a genetic algorithm, and achieves parameter coordination. Finally, the invention utilizes machine learning algorithms to construct a comprehensive prediction model, predicting reduction reaction efficiency and product quality, and generating parameter adjustment schemes to achieve continuous optimization of process conditions. The technical advantages of this invention are that it improves the uniformity of furnace charge distribution, optimizes heat transfer, and enables precise control of process parameters, thereby significantly improving reduction efficiency and product quality while reducing energy consumption. Attached Figure Description

[0005] Figure 1 This is a flowchart of a low-energy-consumption steady-state control method for magnesium bottom discharge in the Pidgeon process according to the present invention.

[0006] Figure 2 This is a schematic diagram of a low-energy-consumption steady-state control method for magnesium bottom discharge in the Pidgeon process according to the present invention.

[0007] Figure 3 This is another schematic diagram of a low-energy-consumption steady-state control method for magnesium bottom discharge in the Pidgeon process according to the present invention. Detailed Implementation

[0008] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0009] like Figure 1-3 This embodiment of a low-energy-consumption furnace charge steady-state control method for bottom-discharge magnesium production in the Pidgeon process may specifically include: S101. Obtain real-time particle distribution data from the furnace charge feeding system, extract particle position and density information using image recognition algorithms, and calculate distribution uniformity index using a preset uniform distribution model to quantify the dispersion of particle distribution.

[0010] Particle distribution data is collected during the feeding process of the furnace charge system. Real-time image data is acquired using image capture equipment to obtain an initial image set. Based on the initial image set, image recognition algorithms are applied to extract features from the particles, identifying their location and density information to determine preliminary particle distribution characteristics. For this preliminary characteristic data, a pre-set uniformity model is used for comparative analysis to calculate distribution indices and determine whether the dispersion of the particle distribution meets a preset threshold. If the distribution indices exceed the preset threshold, the preliminary characteristic data undergoes secondary correction processing, using image denoising techniques to optimize data quality and obtain corrected characteristic data. Based on the corrected characteristic data, the uniformity model is applied again to calculate new distribution indices, obtaining more accurate dispersion assessment results. If the new distribution indices still exceed the preset threshold, trend analysis is performed using historical data records to identify the specific points of distribution anomalies. Based on the trend analysis results, adjustment parameter suggestions for the furnace charge system's feeding process are generated, completing the dynamic monitoring and optimization of particle distribution quality.

[0011] For example, when acquiring real-time particle distribution data from the furnace charge feeding system, a high-definition industrial camera installed above the feed inlet can capture image data of the falling particles at a frequency of 30 frames per second, with the image resolution set to 1920x1080 pixels to ensure that particle details are clearly visible. Next, an image recognition algorithm, such as the YOLOv5 model based on deep learning, is used to process the image. After pre-training, the model is fine-tuned for particle features, achieving an accuracy of over 95% in identifying particle positions. Specifically, each frame of the image is divided into a 100x100 pixel grid, and the number of particles in each grid is calculated to obtain the particle center coordinates (x, y) and density value. For example, if a grid contains 5 particles, the density value is 0.05 particles / pixel². Subsequently, based on a pre-set uniform distribution model (assuming an ideal particle density of 0.04 particles / pixel²), the deviation between the actual and ideal distribution is calculated. The distribution uniformity index is then calculated using the standard deviation formula: σ = √[(Σ(ρi - ρ_mean)²) / N], where ρi is the density of each grid cell, ρ_mean is 0.04, and N is the total number of grid cells (10,000). Assuming a calculated σ of 0.015, this indicates a low degree of distribution dispersion. Finally, by comparing the uniformity index with historical data, if the σ value is higher than 0.02 for 10 consecutive minutes, the system automatically triggers a feed rate adjustment command, reducing the feed rate from 10 tons / hour to 8 tons / hour to optimize distribution uniformity. After adjustment, data is collected again to verify the effect, forming a closed-loop control logic to ensure particle distribution stability.

[0012] S102. Analyze the distribution uniformity index to determine whether the furnace charge meets the steady-state feeding requirements. If the index is lower than the preset threshold, adjust the feeding speed and feeding angle, reacquire particle distribution data and update the distribution uniformity index until the requirements are met.

[0013] Particle distribution data of the furnace charge is collected in real time using sensor devices. The raw data is preprocessed to remove noise interference and obtain preliminary particle distribution characteristic values. Based on the preliminary particle distribution characteristic values, a distribution uniformity index is calculated and analyzed using a pre-established evaluation model to determine whether a preset threshold has been reached. If the distribution uniformity index is lower than the preset threshold, the feed rate adjustment is calculated based on historical data and the current particle distribution characteristic values, and the feed rate parameters are updated. If the distribution uniformity index still does not reach the preset threshold, the feed angle adjustment is calculated based on the particle distribution characteristic values ​​and the furnace charge analysis results, and the feed angle parameters are updated. Using the adjusted feed rate and feed angle parameters, particle distribution data of the furnace charge is collected again to obtain new particle distribution characteristic values. For the new particle distribution characteristic values, the distribution uniformity index is recalculated and compared using the same evaluation model to determine whether the steady-state feed standard is met. If the distribution uniformity index still does not meet the requirements, the feed rate adjustment and feed angle adjustment steps are repeated, continuously updating the particle distribution characteristic values ​​and uniformity index until the preset threshold is reached.

[0014] For example, regarding the analysis and adjustment of the uniformity index of furnace charge distribution, firstly, particle distribution data inside the furnace is collected in real time using sensors. Assume the collected distribution data is a two-dimensional matrix, where each element represents the particle density of a certain area. The initial data are [[2.5, 3.1, 2.8], [3.0, 2.7, 3.2], [2.9, 3.3, 2.6]], in grams per cubic centimeter. The uniformity index is calculated using the standard deviation as the metric. The algorithm is as follows: first, the average of all elements in the matrix is ​​calculated, resulting in a mean of approximately 2.9. Then, the sum of squared deviations of each element from the mean is calculated and divided by the total number of elements (9), yielding a variance of approximately 0.06 and a standard deviation of 0.24. A preset uniformity threshold of 0.2 is set. If the standard deviation is greater than the threshold, it indicates uneven distribution, and the feed parameters need to be adjusted. The system automatically triggers an adjustment mechanism, reducing the feeding speed from the initial 5 tons per hour to 4.5 tons per hour, while simultaneously increasing the feeding angle from 30 degrees to 35 degrees to alter the particle drop distribution. After adjustment, the distribution data is re-collected via sensors. Assuming the new data is [[2.8, 2.9, 2.8], [2.9, 2.8, 3.0], [2.9, 2.9, 2.8]], the standard deviation is recalculated, with a mean of 2.86, a variance of approximately 0.004, and a standard deviation of 0.06, all less than the threshold of 0.2, meeting the steady-state feeding requirements. If this is still not met, the system will continue to iteratively adjust the speed and angle, decreasing the speed by 0.5 tons per hour and increasing the angle by 5 degrees each time, until the targets are met. The entire process is automatically completed by the control system. Data is transmitted to the central processor for analysis via the Industrial Internet of Things (IIoT), and adjustment commands are issued to the feeding equipment via the PLC controller, ensuring the accuracy of closed-loop control. To enhance logical consistency, the system can also incorporate furnace temperature distribution data (such as an average temperature of 800 degrees Celsius and a standard deviation of 10 degrees Celsius) as an auxiliary judgment. If the temperature distribution also tends to be uniform, the effectiveness of the feed adjustment can be further confirmed, forming a multi-dimensional verification chain of thought.

[0015] S103. When the uniformity of distribution meets the requirements, acquire the temperature field and flow field data inside the furnace, use the updated particle distribution data as boundary conditions, and calculate the coupling effect of heat transfer and flow field through the heat conduction model and the fluid dynamics model to obtain the heat distribution characteristics.

[0016] Temperature and flow field data are acquired from the furnace environment via a data acquisition module. The raw data is preprocessed to obtain preliminary cleaned temperature and flow field datasets. Based on these datasets and particle distribution data, boundary conditions are constructed, and a pre-established heat conduction model is used to simulate the heat transfer effect, determining intermediate results for the heat transfer distribution. A fluid dynamics model is then used to analyze the flow field data, and the influence of flow field coupling effects is calculated based on the intermediate results of the heat transfer distribution, obtaining the characteristics of the coupled flow field distribution. Using these coupled flow field distribution characteristics and the output data from the heat conduction model, detailed heat distribution characteristics are further iteratively calculated to obtain the final heat distribution dataset. If the final heat distribution dataset deviates from the uniformity index, the particle distribution data is adjusted, boundary conditions are reconstructed, and the effects of heat transfer and flow field coupling are iteratively calculated to determine if the preset uniformity conditions are met. Based on the adjusted particle distribution data and the final heat distribution dataset, combined with furnace environment parameters, an optimized mapping relationship between heat distribution and flow field coupling is determined through comparative analysis. If the optimized mapping relationship meets the preset threshold, the heat distribution features are finally calibrated based on the mapping relationship to obtain the calibrated heat distribution feature dataset.

[0017] For example, after achieving the required uniformity of particle distribution, temperature and flow field data within the furnace are first collected via a sensor network. It is assumed that the collected temperature field data is distributed across 1000 points on a spatial grid, with a temperature range of 1150°C to 1250°C and a flow velocity range of 0.5 m / s to 2.0 m / s. The data is stored in matrix form and smoothed using Fourier transform to filter out noise, resulting in a stable temperature gradient distribution and flow field vector diagram. The analysis shows that the maximum temperature gradient is 5°C / cm and the peak flow vorticity is 0.8 s^-1. Subsequently, updated particle distribution data is input into the system as boundary conditions. The particle distribution data is assumed to have a uniformity index of 0.95, a particle size range of 0.1 mm to 0.5 mm, and a density of 2500 kg / m^3. These parameters are imported into the heat conduction model and the fluid dynamics model via a data interface. The finite element method is used for mesh generation, with 50,000 mesh elements and a time step of 0.01 s to ensure computational accuracy. Next, heat transfer within the furnace was calculated using a heat conduction model. Based on the Fourier heat conduction equation with a thermal conductivity of 50 W / (m·K), and combined with the Navier-Stokes equations in the fluid dynamics model, heat transfer and flow field effects were coupled and calculated. The iterative convergence condition was set to a residual of less than 10^-5. The calculation results showed that the high-temperature zone was concentrated in the center of the furnace, with a peak temperature of 1150°C and a heat flux density of 3000 W / m^2. The influence of the flow field on the heat distribution was manifested in a boundary layer thickness of approximately 2 cm. Analysis of the heat distribution characteristics revealed a positive correlation between the concentrated heat area and the particle density, with a correlation coefficient of 0.85. This provides data support for subsequent optimization of particle distribution, logically forming a complete chain from data acquisition to model calculation and feature analysis, ensuring the rigor and traceability of the technology implementation.

[0018] S104. Based on the heat distribution characteristics, predict the cumulative heat transfer trend by combining the heat balance equation, where the heat balance equation is Q=mcΔT, Q represents the heat change, m represents the material mass, c represents the specific heat capacity, and ΔT represents the temperature change. If the prediction results show that there are local overheated areas, adjust the gas flow rate and temperature distribution, recalculate the coupling effect and update the cumulative heat transfer trend until the overheated areas are eliminated.

[0019] By collecting heat and temperature distribution data within the system, preliminary calculations are performed using the heat balance equation to obtain the initial trend of heat transfer accumulation. Based on this initial trend, the existence of local overheating areas is analyzed. If local overheating is detected, the corresponding temperature distribution and heat change data are extracted to determine the specific location and extent of the overheated area. For the identified overheated area, the gas flow rate and temperature distribution parameters are adjusted, and a pre-defined coupling effect model is used for recalculation to obtain updated heat distribution data. Using the updated heat distribution data, combined with material mass and specific heat capacity, heat and temperature changes are recalculated to obtain a new trend of heat transfer accumulation. Based on this new trend, the existence of local overheating areas is determined. If they still exist, the latest coupling effect data is extracted, and the gas flow rate and temperature distribution parameters are adjusted, with the calculation repeated until the overheated area disappears. The final heat transfer accumulation trend data is obtained, and combined with the heat balance equation, the consistency of heat and temperature changes is verified to determine the stable state of the system's heat distribution. Using the heat distribution data under the stable state, the final gas flow rate and temperature distribution parameters are recorded and stored in the system database for subsequent heat balance analysis.

[0020] For example, in analyzing and predicting heat distribution characteristics, firstly, temperature distribution data inside the equipment is acquired using thermal imaging technology. Assuming the temperature distribution range inside an industrial furnace is 500 to 800 degrees Celsius, with the central region reaching 1200 degrees Celsius and the edge region at 1150 degrees Celsius, heat change is predicted using the heat balance equation Q=mcΔT. With a material mass m of 1000 kg, a specific heat capacity c of 0.5 kJ / kg·°C, and an initial temperature change ΔT of 50 degrees Celsius, the initial heat change Q is calculated as Q=1000×0.5×50=25000 kJ. Based on the temperature distribution data, the cumulative heat transfer trend is predicted. Using a finite element analysis algorithm, the furnace space is divided into 1000 grid cells. The heat transfer rate is calculated for each cell. Assuming the heat accumulation rate in the central region reaches 300 kJ / s and at the edge 100 kJ / s, the prediction results show that the central region accumulates 540,000 kJ of heat within 30 minutes, indicating a risk of localized overheating. Subsequently, the system automatically identifies overheated areas and adjusts the gas flow rate and temperature distribution parameters. The gas flow rate in the central area is reduced from 5 m / s to 3 m / s, and the target temperature is lowered from 780 degrees Celsius to 700 degrees Celsius. Using a thermo-hydrodynamic coupling model, the system recalculates, assuming the adjusted heat accumulation rate in the central area drops to 200 kJ / s while the edge remains unchanged. The heat transfer accumulation trend is updated, predicting that the central heat accumulation will reach 360,000 kJ / s within 30 minutes, thus reducing the overheating risk. If overheating still exists, the system iteratively adjusts the flow rate to 2.5 m / s and the temperature to 680 degrees Celsius, repeating the calculation until the heat distribution is uniform and the difference in heat accumulation between the center and edge is less than 10%. Finally, the optimized heat distribution map and parameter settings are output to ensure stable equipment operation.

[0021] S105. When the cumulative heat transfer trend is stable, acquire real-time data of process parameters, including temperature, pressure and flow rate, construct a multi-parameter correlation model based on statistical regression method, analyze the mutual influence relationship between parameters, and obtain the parameter correlation matrix.

[0022] Real-time data, covering temperature, pressure, and flow rates, is acquired from the process system via a sensor network to form an initial dataset. Based on this initial dataset, data cleaning methods are used to remove outliers and missing values, resulting in a processed dataset. If uneven data distribution exists in the processed dataset, data standardization methods are used to adjust the data scale, determining a standardized dataset. For the standardized dataset, a multi-parameter model is constructed using statistical regression methods to analyze the correlation between temperature, pressure, and flow rates, revealing the parameter influence relationships. The mutual influences between key variables are extracted from these relationships, generating a preliminary correlation matrix. Based on this preliminary correlation matrix, correlation coefficient calculation methods are used to further verify the strength of the relationships between parameters, determining the final correlation matrix. If the correlation coefficients between some parameters in the final correlation matrix are lower than a preset threshold, the original data is re-analyzed using data backtracking methods to obtain supplementary data and update the correlation matrix.

[0023] For example, when the cumulative heat transfer trend is stable, the system first collects process parameter data in real time through a sensor network, such as temperature, pressure, and flow rate. Assuming a metallurgical equipment is operating stably, the collected temperature data is 85.5 degrees Celsius, the pressure is 2.8 MPa, and the flow rate is 150.3 cubic meters per hour. This data is automatically uploaded to a cloud database via an industrial IoT platform, forming a time-series dataset. Subsequently, a multi-parameter correlation model is constructed based on statistical regression methods. The system uses a multiple linear regression algorithm, with temperature, pressure, and flow rate as independent variables and heat transfer efficiency as the dependent variable, to calculate the regression coefficients. Assuming the regression equation is efficiency = 0.35 * temperature + 0.22 * pressure - 0.15 * flow rate + constant term, the data is fitted using the least squares method to obtain the significance test results for each parameter. The p-value for temperature is 0.01, for pressure it is 0.03, and for flow rate it is 0.05, indicating that temperature has the most significant impact on heat transfer efficiency. Next, the system analyzes the interrelationships between parameters, using the Pearson correlation coefficient to calculate the correlations between each parameter. The correlation coefficients are found to be 0.75 for temperature and pressure, -0.42 for temperature and flow rate, and -0.38 for pressure and flow rate, indicating a positive correlation between temperature and pressure, while flow rate is negatively correlated with other parameters. Finally, the system generates a parameter correlation matrix, which is stored in a table format in the database for subsequent process optimization. For example, the correlation matrix reveals that an increase in temperature may lead to an increase in pressure, thus adjusting the flow rate to 145.0 cubic meters per hour to balance the system load, forming a closed-loop optimization logic. To ensure the comprehensiveness of the analysis, the system can also perform trend verification using historical data. Assuming the average temperature change over the past week is less than 1.0 degrees Celsius, after confirming the stability of the current data, the analysis results are compared with historical models, with the error controlled within 5%, ensuring model reliability. The entire process is automated, with data transfer and calculations completed by the backend system.

[0024] S106. Evaluate the parameter correlation matrix to determine if there are any parameter conflicts. If a conflict is detected, optimize the parameter combination using a genetic algorithm, reacquire the process parameter data, and update the parameter correlation matrix until the parameters are consistent.

[0025] The parameter matrix undergoes initial processing. Raw process data is retrieved from the repository, and correlation analysis is used to calculate the correlation strength between parameters, resulting in an initial form of the parameter correlation matrix. Based on this initial form, conflict detection is performed. If significant conflicts are detected, the conflict locations and related parameter pairs are recorded, determining the range of parameters requiring optimization. For the recorded conflict locations and parameter pairs, a genetic algorithm is introduced for parameter optimization. Multiple parameter combinations are generated through iterative calculation, yielding a pre-adjusted parameter set. Based on this pre-adjusted parameter set, process data is re-collected, the parameter records in the repository are updated, and a new parameter correlation matrix is ​​constructed. For the newly constructed parameter correlation matrix, conflict detection is performed again. If conflicts still exist, the genetic algorithm is returned for a new round of parameter optimization, resulting in an updated parameter combination. Through multiple iterations of optimization and matrix reconstruction, the process data is continuously updated until the detection results show parameter consistency, determining the final parameter combination scheme. Based on the final parameter combination scheme, the optimized process data and correlation matrix are saved, completing the entire parameter coordination process.

[0026] For example, regarding the evaluation and optimization of the parameter correlation matrix, we take a specific process parameter optimization scenario. Assume we have three process parameters: temperature (T), pressure (P), and time (H), with initial data of T=200°C, P=5.0MPa, and H=10h. We construct a parameter correlation matrix. By calculating the Pearson correlation coefficients between the parameters, the matrix values ​​are: T = 0.85, T = 0.3, and P = 0.75. Analysis shows that the correlation coefficient between T and P exceeds 0.8, indicating a strong positive correlation, which may lead to parameter conflicts and affect process stability. Therefore, the system automatically triggers a genetic algorithm for optimization, setting the population size to 50, the crossover rate to 0.7, the mutation rate to 0.1, and the objective function to minimize the weighted sum of the correlation coefficients between parameters and the process cost (weights of 0.6 and 0.4, respectively). After 10 iterations, the algorithm outputs the optimal parameter combination as T=180°C, P=4.2MPa, and H=11h. Next, the system re-collects process data based on the new parameters, updates the correlation matrix, and calculates that the correlation coefficient between T and P has decreased to 0.65, T and H to 0.28, and P and H to 0.72. All correlation coefficients are below the conflict threshold of 0.8, indicating parameter consistency. If conflicts still exist, the system will repeatedly execute a genetic algorithm for optimization until the conditions are met. To ensure logical rigor, the system is also linked to a process stability assessment module. The optimized parameters are input into simulations to verify their performance under different operating conditions. For example, under high temperature and high pressure, the stability index improves from 0.75 before optimization to 0.88, achieving the expected target. Through this process, the system achieves fully automated closed-loop processing from conflict detection to parameter optimization, ensuring the consistency of process parameters and production efficiency.

[0027] S107. Using the updated distribution uniformity index, heat transfer accumulation change trend and coordinated process parameter combination, a comprehensive prediction model is constructed based on machine learning algorithm to predict reduction reaction efficiency and product quality.

[0028] An initial dataset is constructed by extracting distribution uniformity indicators and heat transfer accumulation data from production data. Preprocessing methods are used to clean and standardize the data, resulting in a structured basic dataset. Based on this basic dataset, the changing trend of heat transfer accumulation is analyzed. Time series analysis is used to extract trend features and determine key time points and fluctuation ranges of trend changes. For these trend features, a feature matrix is ​​constructed by combining process parameters and parameter combination data. A random forest algorithm is used to rank the features by importance, determining the weight of each parameter's impact on reaction efficiency. If the weight of a certain process parameter in the feature matrix is ​​higher than a preset threshold, it is marked as a key parameter. Historical adjustment records of key parameters are obtained to determine their potential impact patterns on product quality. Based on the impact patterns of key parameters, combined with distribution uniformity and reaction efficiency data, a comprehensive prediction model is constructed and trained using a random forest algorithm to obtain predicted outputs for reaction efficiency and product quality. Based on the predicted outputs, the correlation between reaction efficiency and product quality is analyzed. If the efficiency value is lower than a preset standard, the corresponding process parameter combination data is extracted to determine the adjustment direction and magnitude. Obtain the adjusted process parameter combination data, input it into the comprehensive prediction model for secondary verification, determine whether the adjustment meets the preset efficiency and quality standards, and output the final optimized parameter scheme.

[0029] For example, regarding the process of building a comprehensive predictive model to predict reduction reaction efficiency and product quality, we can design and implement it in detail, from data processing to model building. First, for the updated distribution uniformity index, we collect material distribution data within the reactor using sensors. Assuming the collected uniformity index value is 0.85 (range 0 to 1), we calculate its standard deviation as 0.05 using statistical analysis methods. We then perform data standardization using the NumPy library in Python to ensure the data conforms to a normal distribution, providing reliable input for subsequent modeling. Next, regarding the cumulative heat transfer trend, we collect hourly temperature change data during the reaction process. Assuming the initial temperature is 800°C and the cumulative change rate is an increase of 2.5°C per hour, we use time series analysis algorithms such as the ARIMA model to predict the temperature trend for the next 5 hours. The analysis results show that the temperature stabilizes at around 812.5°C, indicating that the heat transfer process is controllable. Subsequently, for a consistent combination of process parameters, we extracted parameters such as reaction time, pressure, and catalyst dosage. Assuming a reaction time of 6 hours, a pressure of 2.5 MPa, and a catalyst dosage of 0.3 kg, we used Principal Component Analysis (PCA) for dimensionality reduction, obtaining a major influencing factor ratio of 85%, ensuring the optimization of the parameter combination. Finally, we constructed a comprehensive prediction model based on machine learning algorithms. We chose the Random Forest algorithm, implemented using the Scikit-learn library, and used the processed uniformity index, heat transfer trend, and process parameters as input features. The training dataset contained 1000 historical reaction records, and the prediction accuracy on the test set was 92%, with a predicted reduction reaction efficiency of 88.5% and a predicted product quality pass rate of 95.2%. By comparing with actual data, the model error was controlled within 3%, verifying the model's reliability. All data processing and analysis in the above stages were implemented through automated scripts, forming a complete logical chain from data acquisition to prediction output, ensuring accurate and traceable prediction results.

[0030] S108. Analyze the prediction results of the comprehensive prediction model to determine whether the current process conditions have reached the optimal state. If they have not reached the optimal state, generate a parameter adjustment plan, reacquire the furnace charge distribution data, heat transfer data and process parameter data, update the prediction results of the comprehensive prediction model, until the process conditions reach the optimal state.

[0031] The prediction results generated by the comprehensive prediction model are analyzed using pre-established evaluation rules to determine whether the current process conditions meet the preset threshold for the optimal state, thus obtaining a preliminary condition evaluation conclusion. If the preliminary condition evaluation conclusion indicates that the process conditions have not reached the optimal state, the process parameters are optimized using a support vector machine algorithm based on the deviation between the prediction results and the optimal state to determine a specific parameter adjustment scheme. Based on the determined parameter adjustment scheme, the latest furnace charge distribution data, heat transfer data, and process parameter data are acquired from the production environment through an automated data acquisition system to obtain an updated dataset. For the updated dataset, the pre-trained comprehensive prediction model is recalculated to obtain new prediction results. Based on the new prediction results, the conditions are re-evaluated using the preset evaluation rules. If the process conditions still have not reached the optimal state, a loop iteration mechanism is triggered, and the parameter adjustment scheme generation step is re-executed. If the condition evaluation indicates that the process conditions have reached the optimal state, the loop iteration stops, the current parameter adjustment scheme and prediction results are recorded, and the final process optimization configuration is determined. Using the final process optimization configuration, the process parameter settings in the production environment are updated, and real-time operating data is obtained to verify the optimization effect and determine the stability of the process conditions.

[0032] For example, regarding the analysis of prediction results and optimization of process conditions using the comprehensive prediction model, the system first automatically reads the key indicator data output by the prediction model. For instance, under the current process conditions, the predicted uniformity of the furnace charge distribution is 85.3%, the heat transfer efficiency is 78.6%, and the product qualification rate is 92.1%. This data is then compared with the preset optimal target values ​​(uniformity 90%, heat transfer efficiency 85%, qualification rate 95%). The analysis reveals that the current process conditions have not reached the optimal state. The system then automatically triggers the parameter adjustment module. Based on historical data and the gradient descent algorithm, it calculates the adjustment direction of the current process parameters. For example, the furnace temperature is adjusted from 1250°C to 1270°C, and the air volume is increased from 500 m³ / h to 520 m³ / h. The adjustment range is calculated using the formula ΔP = α * (target value - current value), where α is an adjustment coefficient of 0.1 to ensure smooth adjustment. After adjustment, the system re-collects data on furnace charge distribution (e.g., uniformity improved to 87.5%), heat transfer (efficiency increased to 80.2%), and process parameters (temperature 1270°C, air volume 520 m³ / h) through a sensor network. This data is then input into the comprehensive prediction model in real time, updating the prediction results to uniformity 88.1%, heat transfer efficiency 81.5%, and pass rate 93.2%. The system compares this to the target values ​​again and, finding that the optimal result has not yet been achieved, automatically enters the next iteration, adjusting the temperature to 1280°C and the air volume to 530 m³ / h, cyclically updating the data and predictions until the prediction results reach or approach the target values ​​(uniformity 90.2%, efficiency 85.1%, pass rate 95.3%). To ensure logical rigor, the system is also linked to an energy consumption monitoring module. If the adjusted energy consumption exceeds a threshold (e.g., energy consumption per ton of product exceeds 500 kWh), a genetic algorithm optimizes the parameter combination to balance efficiency and cost, forming a closed-loop control. Finally, the system outputs the optimal combination of process parameters and a prediction result report, which is automatically saved to the database for subsequent analysis.

[0033] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for steady-state control of low-energy-consumption furnace charge at the bottom of the vertical tank in the Pidgeon process, characterized in that, include: S1. Obtain real-time particle distribution data from the furnace charge feeding system, extract particle position and density information using image recognition algorithms, and calculate the distribution uniformity index by combining a uniform distribution model; S2. Analyze the distribution uniformity index. If it is lower than the preset threshold, adjust the feeding speed and feeding angle, reacquire particle distribution data and update the index until it meets the standard. S3. When the uniformity index of distribution meets the standard, acquire the temperature field and flow field data in the furnace, use the particle distribution data as the boundary condition, and calculate the heat transfer and flow field coupling effect through the heat conduction model and the fluid dynamics model to obtain the heat distribution characteristics. S4. Based on the heat distribution characteristics, the cumulative heat transfer trend is predicted by combining the heat balance equation Q=mcΔT. If there are local overheated areas, the gas flow rate and temperature distribution are adjusted, the coupling effect is recalculated and the trend is updated until the overheating is eliminated. S5. When the cumulative heat transfer trend is stable, real-time data of process parameters including temperature, pressure and flow rate are obtained. A multi-parameter correlation model is constructed based on statistical regression methods to analyze the mutual influence of parameters and obtain the parameter correlation matrix. S6. Evaluate the parameter correlation matrix. If parameter conflicts are detected, optimize the parameter combination using a genetic algorithm, reacquire process data, and update the matrix until the parameters are consistent. S7. Using the updated distribution uniformity index, heat transfer accumulation change trend and coordinated process parameter combination, a comprehensive prediction model is constructed based on machine learning algorithm to predict reduction reaction efficiency and product quality; S8. Analyze the prediction results of the comprehensive prediction model. If the process conditions are not optimal, generate a parameter adjustment plan, reacquire the furnace charge distribution data, heat transfer data and process parameter data, and update the prediction results until the process conditions reach the optimal state.

2. The method according to claim 1, characterized in that, S1 includes: Real-time image data of particle distribution is acquired through image capture equipment, and image recognition algorithms are applied to extract particle location and density information. A uniform distribution model is used to calculate distribution indicators and determine whether the degree of dispersion meets the threshold. If the indicator exceeds the threshold, the data is corrected through image denoising technology and the indicator is recalculated. If it still exceeds the threshold, historical data trend analysis is combined to generate suggestions for adjusting parameters in the feeding process.

3. The method according to claim 1, characterized in that, S2 includes: real-time acquisition of particle distribution data, preprocessing and calculating the distribution uniformity index; if the index is lower than the threshold, calculating the feed speed adjustment amount based on historical data; if it still does not meet the standard, calculating the feed angle adjustment amount; based on the adjusted speed and angle parameters, re-acquiring data and calculating the index, and repeating the process until the standard is met.

4. The method according to claim 1, characterized in that, S3 includes: acquiring temperature field and flow field data, preprocessing them and constructing boundary conditions by combining them with particle distribution data; simulating heat transfer effects through a heat conduction model and calculating flow field coupling effects by combining them with a fluid dynamics model; iteratively calculating heat distribution characteristics, and if there is a deviation from the uniformity index, adjusting the particle distribution data and recalculating; determining the optimal mapping relationship between heat distribution and flow field coupling, and calibrating the heat distribution characteristics.

5. The method according to claim 1, characterized in that, S4 includes: calculating the initial trend of heat transfer accumulation using the heat balance equation; if a local overheated area is detected, adjusting the gas flow rate and temperature distribution parameters; recalculating the heat distribution data and updating the heat transfer accumulation trend; cyclically adjusting and calculating until the overheated area is eliminated, and verifying the stable state of heat distribution.

6. The method according to claim 1, characterized in that, S5 includes: acquiring real-time data on temperature, pressure, and flow rate, and performing data cleaning and standardization; constructing a multi-parameter model based on statistical regression methods to analyze the correlation between parameters; generating and validating a correlation matrix; and if the correlation coefficient is lower than a threshold, backtracking the data and updating the matrix.

7. The method according to claim 1, characterized in that, S6 includes: calculating the initial shape of the parameter correlation matrix from the process data; if a significant conflict is detected, recording the conflict location and parameter pair; generating a parameter combination scheme through a genetic algorithm, re-collecting data and updating the matrix; and iteratively optimizing until the parameters are consistent.

8. The method according to claim 1, characterized in that, S7 includes: constructing an initial dataset containing distribution uniformity index, heat transfer cumulative data, and process parameter combinations; extracting heat transfer trend features through time series analysis; ranking the importance of features using the random forest algorithm and labeling key parameters; constructing a comprehensive prediction model based on the influence patterns of key parameters, and outputting prediction results for reaction efficiency and product quality after training.

9. The method according to claim 1, characterized in that, S8 includes: determining whether the prediction result has reached the optimal state threshold through evaluation rules; if it has not reached the threshold, generating a parameter adjustment scheme using a support vector machine algorithm; re-acquiring data on furnace charge distribution, heat transfer, and process parameters, and inputting the updated results into the comprehensive prediction model; iterating in a loop until the process conditions reach the optimal, recording the optimized configuration, and verifying its stability.

Images