Intelligent prediction method and system for smelting flue gas acid making process based on double models

Abstract

The invention belongs to the technical field of flue gas acid making, and particularly discloses an intelligent prediction method and system for a smelting flue gas acid making process based on double models. The method comprises the following steps: collecting flue gas acid making data, preprocessing and extracting characteristic parameters; screening features having significant influence on the prediction target; establishing a production process time sequence prediction model by using LSTM; training an LSTM model through historical data to predict a key parameter change trend; establishing a mechanism driving model based on a key chemical reaction basic principle in a flue gas acid making process; the mechanism driving model provides physical and chemical laws of the reaction process, and the data driving model performs dynamic prediction through real-time data; and integrating the dual-mode combined model with smelting equipment and a process, and dynamically adjusting production parameters by adopting a predictive control algorithm based on real-time prediction of a data-driven model and process optimization of a mechanism-driven model. The device has the characteristics of high production efficiency, stable production, low energy consumption and less manual intervention.

Description

Technical Field

[0001] The present invention relates to the technical field of flue gas to sulfuric acid production, and particularly relates to an intelligent prediction method and system for the flue gas to sulfuric acid production process based on a dual model, which has high production efficiency, stable production, low energy consumption, and less manual intervention. Background Art

[0002] The production of sulfuric acid from flue gas is a key process for treating sulfur-containing flue gas in the metallurgical industry. The core steps include flue gas purification, catalytic conversion (SO 2 →SO 3 ), and acid formation by absorption (H 2 SO 4 ), etc. As a pillar of traditional manufacturing, the metallurgical industry has long faced problems such as low production efficiency, serious resource waste, high energy consumption, and heavy pollution. Especially in the process of producing sulfuric acid from flue gas, traditional control methods rely on manual experience and conventional automation control technologies, making it difficult to achieve precise regulation. The instability of sulfur dioxide (SO 2 ) content in high-concentration flue gas, the complexity of oxidation reactions, and the multi-factor coupling in the reaction process make it difficult to ensure the stability of the smelting process and the quality of products.

[0003] In recent years, with the rapid development of information technology, data analysis, and artificial intelligence (AI) technology, intelligent control has become an important direction for the metallurgical industry to improve production efficiency and optimize processes. Data-driven models and mechanism-driven models have gradually been applied to intelligent process control. Among them, the mechanism-driven model is based on physical, chemical, and thermodynamic principles, and describes the system behavior through differential equations or algebraic equations; it has clear physical meaning, strong interpretability, does not require historical data, is suitable for new process design, and has high prediction accuracy under steady-state conditions; however, it also has problems such as difficulty in dealing with dynamic disturbances (such as flue gas flow fluctuations and catalyst deactivation). The data-driven model is based on historical data or real-time sensor data, and mines the input-output relationship through machine learning algorithms; it is good at dealing with non-linear and high-dimensional data (such as multi-sensor fusion), has strong dynamic adaptability, is suitable for complex working conditions, can update the model in real time, and can cope with equipment aging or process changes; however, it has problems such as relying on a large amount of high-quality data, poor effect in small-sample scenarios, low model interpretability, difficulty in meeting the requirements of process safety review, and the risk of overfitting. Therefore, using only data-driven models and mechanism-driven models alone is difficult to meet the non-linear and strongly fluctuating flue gas to sulfuric acid production process.

[0004] In the prior art, although there have been technical solutions that attempt to combine data-driven models and mechanism-driven models, due to the lack of a systematic solution, the integration of data and the application of models are disjointed from each other, and it is difficult to effectively combine data-driven models and mechanism-driven models. Moreover, data-driven models usually rely on a large amount of historical data for training, while mechanism-driven models are built based on physical and chemical principles. Therefore, when the two are combined, there are often problems such as poor model compatibility and difficulty in parameter matching. And for the intelligent transformation of the traditional sulfuric acid production process from smelting flue gas, it mostly focuses on the equipment layer and the data acquisition layer, while the intelligent technologies in the model layer and the application layer have not been widely used.

[0005] Therefore, in order to achieve precise control of the smelting process, improve production efficiency and stability, an intelligent control system urgently needs to be more closely integrated in a multi-level and full-process system. Summary of the Invention

[0006] Aiming at the deficiencies in the prior art, the present invention provides an intelligent prediction method for the sulfuric acid production process from smelting flue gas with high production efficiency, stable production, low energy consumption and less manual intervention, and also provides an intelligent prediction system for the sulfuric acid production process from smelting flue gas based on a dual model.

[0007] The intelligent prediction method for the sulfuric acid production process from smelting flue gas based on a dual model of the present invention is implemented as follows: It includes steps of data-driven model construction, mechanism-driven model construction, dual-model combination, and system application. The specific content is as follows: A. Data-driven model construction: Collect data from the sulfuric acid production process from smelting flue gas and perform preprocessing; then extract characteristic parameters from the preprocessed data; subsequently, through feature importance evaluation and correlation analysis, screen out the features that have a significant impact on the prediction target; then use LSTM to establish a time series prediction model for the production process; finally, train the LSTM model with historical data to predict the change trend of key parameters in the production process, providing timely early warning and feedback for production regulation; B. Mechanism-driven model construction: Based on the basic principles of key chemical reactions in the sulfuric acid production process from smelting flue gas, establish a mechanism-driven model; C. Dual-model combination: The mechanism-driven model provides an in-depth understanding of the physical and chemical laws of the reaction process, while the data-driven model dynamically predicts the reaction process through real-time data; D. System application: Integrate the model after the aforementioned dual-model combination with existing smelting equipment and process flows to form a closed-loop control. Based on the real-time prediction of the data-driven model and the process optimization of the mechanism-driven model, use a predictive control algorithm to dynamically adjust production parameters.

[0008] Further, the specific process of step A is as follows: A10. Data collection and preprocessing: Through intelligent sensors and data acquisition systems, real-time production data is collected from each key equipment in the sulfuric acid production process from smelting flue gas, and then the collected real-time production data is denoised, missing values are filled, outliers are detected, and data standardization processing is performed. A20. Feature selection and engineering: Based on the preprocessed data, feature engineering techniques are used to extract feature parameters closely related to the sulfuric acid production process from smelting flue gas; then through feature importance evaluation and correlation analysis, the values of the feature parameters are adjusted and the change degree of the prediction target is observed, and the features that have a significant impact on the prediction target are screened out, and redundant or feature parameters with a low impact on the prediction target are removed. A30. LSTM model construction: An LSTM network is used to establish a time series prediction model for the production process, and the long-term dependence characteristics in the time series are captured through the gating mechanism. A40. LSTM model prediction: The LSTM model is trained and optimized through historical data, and the optimized LSTM model is used to predict the change trend of key parameters in the production process, providing timely early warning and feedback for production regulation.

[0009] Further, the key equipment includes reactors, absorption towers, fans, compressors, and heat exchangers, the production data includes temperature, gas concentration, fan speed, reactor pressure, current, and voltage, the feature parameters include the relationship between temperature and gas concentration, fan speed and gas flow, and reactor pressure and SO 2 concentration, and the key parameters include temperature, pressure, and SO 2 concentration.

[0010] Further, the basic principles of the key chemical reactions in step B include SO 2 catalytic oxidation reaction and SO 3 absorption reaction; the chemical kinetic model in the mechanism-driven model is used to describe the reaction process of SO 2 converted to SO 3 ; and the physical process is used to simulate the gas flow, temperature distribution, and pressure change in the reactor, and the production process is optimized according to the chemical reaction process and physical process simulation.

[0011] Further, in step C, the theoretical output of the mechanism-driven model is used as the input feature of the data-driven model, or the data-driven model is used to correct the error of the mechanism-driven model, and at the same time, the model parameters are dynamically adjusted through physical constraints and real-time data feedback.

[0012] Further, abnormal detection and handling are also included in step D. When the predicted value exceeds the standard or is abnormal during the production process, the early warning mechanism is automatically identified and triggered.

[0013] The intelligent prediction system for the sulfuric acid production process based on dual models of the present invention is implemented as follows: It includes a data-driven module, a mechanism-driven module, a dual-model combination module, and a system application module. The data-driven module is used to collect data from the sulfuric acid production process of smelting flue gas and perform preprocessing; then extract characteristic parameters from the preprocessed data; subsequently, through feature importance evaluation and correlation analysis, select features that have a significant impact on the prediction target; then use LSTM to establish a time series prediction model for the production process; finally, train the LSTM model with historical data to predict the change trend of key parameters in the production process, providing timely early warning and feedback for production control. The mechanism-driven module establishes a mechanism-driven model based on the basic principles of key chemical reactions in the sulfuric acid production process of smelting flue gas. The dual-model combination module is used to enable the mechanism-driven model in the mechanism-driven module to provide an in-depth understanding of the physical and chemical laws of the reaction process, while the data-driven module dynamically predicts the reaction process through real-time data. The system application module is used to integrate the aforementioned dual-model combination module with existing smelting equipment and process flows to form a closed-loop control. Based on the real-time prediction of the data-driven module and the process optimization of the mechanism-driven module, a predictive control algorithm is used to dynamically adjust production parameters.

[0014] Furthermore, the data-driven module includes a data collection and preprocessing unit, a feature selection and engineering unit, an LSTM model construction unit, and an LSTM model prediction unit. The data collection and preprocessing unit is used to collect real-time production data from each key equipment in the sulfuric acid production process of smelting flue gas through intelligent sensors and a data collection system, and then perform denoising, filling missing values, outlier detection, and data standardization processing on the collected real-time production data. The feature selection and engineering unit is used to, based on the preprocessed data, adopt feature engineering techniques to extract characteristic parameters closely related to the sulfuric acid production process of smelting flue gas; then, through feature importance evaluation and correlation analysis, adjust the values of the characteristic parameters and observe the degree of change of the prediction target, select features that have a significant impact on the prediction target, and remove redundant or less influential characteristic parameters on the prediction target. The LSTM model construction unit is used to establish a time series prediction model for the production process using an LSTM network, and capture the long-term dependence characteristics in the time series through a gating mechanism. The LSTM model prediction unit is used to train and optimize the LSTM model with historical data, and use the trained and optimized LSTM model to predict the change trend of key parameters in the production process, providing timely early warning and feedback for production control.

[0015] Furthermore, in the mechanism-driven module, the basic principles of key chemical reactions include SO 2 catalytic oxidation reaction and SO 3 absorption reaction; the chemical kinetic model in the mechanism-driven model is used to describe the reaction process of SO 2 converted to SO 3 ; and the physical process is used to simulate the gas flow, temperature distribution and pressure change in the reactor, and the production process is optimized according to the chemical reaction process and physical process simulation.

[0016] Furthermore, the system application module also includes an anomaly detection and handling unit, which is used to automatically identify and trigger an early warning mechanism when the predicted value exceeds the standard or is abnormal during the production process.

[0017] The beneficial effects of the present invention are as follows: 1. The present invention deeply combines the mechanism-driven model and the data-driven model. By providing the physical and chemical laws and theoretical basis of the reaction process through the mechanism-driven model (chemical kinetics, physical process simulation), and combining the real-time data capture of the dynamic changes and uncertainties of the data-driven model (LSTM time series prediction), it solves the limitations of a single model in non-linear and dynamic disturbance (such as flue gas fluctuation, catalyst deactivation) scenarios, and also solves the problem of model fragmentation existing in the existing dual-model combination, so as to ensure that it can more accurately reflect the changes in the production process when facing a complex production environment, and can also realize the dynamic optimization and precise control of the production process; moreover, by real-time prediction and dynamic adjustment to optimize the production process, it can not only improve the production efficiency, stability and product quality of the sulfuric acid production process from smelting flue gas, but also reduce energy consumption and reduce manual intervention.

[0018] 2. The data-driven model of the present invention establishes a time series prediction model of the production process through LSTM, so as to be able to adapt to complex non-linear characteristics and dynamic changes. Combined with the scientific simulation of the mechanism-driven model, it can accurately grasp the change trend of key parameters in the production process, so as to provide reliable support for process adjustment in the production process.

[0019] 3. The present invention integrates closed-loop control and adopts anomaly early warning. By comparing the prediction results of the dual models with real-time process parameters and using the predictive control algorithm to dynamically optimize production parameters (such as converter temperature, fan speed, etc.), it ensures that key parameters (such as temperature, pressure, etc.) are stable within a predetermined range, thus reducing the risk of shutdown caused by production fluctuations, ensuring the stability and efficiency of the sulfuric acid production process, significantly improving production capacity and product quality, and also reducing energy consumption; moreover, based on the prediction trend of the LSTM model, it automatically identifies abnormal conditions exceeding the standard (such as catalyst layer blockage, SO 2Concentration mutations, etc.) and trigger an early warning mechanism, which can further reduce the need for manual intervention, improve the safety and reliability of the sulfuric acid production process from smelting flue gas, and promote the intelligent transformation of the smelting industry.

[0020] 4. By evaluating the importance of features, the present invention screens key parameters (such as temperature - gas concentration, fan speed - gas flow rate), which can eliminate the interference of redundant data, thereby improving the generalization ability of the model; moreover, it breaks through the limitations of the traditional equipment layer / collection layer, realizes the collaborative optimization of the model layer (dual - model fusion) and the application layer (closed - loop control), and achieves the deep integration of data and mechanism.

[0021] In summary, the present invention has the characteristics of high production efficiency, stable production, low energy consumption, and less manual intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a schematic diagram of the principle of the sulfuric acid production process from high - concentration flue gas according to an embodiment of the present invention; Figure 2 is a schematic diagram of the architecture principle of the intelligent prediction system of the present invention; Wherein: 1 - Heat exchanger Ⅰ, 2 - Heat exchanger Ⅱ, 3 - Pre - converter, 4 - SO 3 Cooler, 5 - Pre - absorption tower, 6 - Relay fan, 7 - SO 2 Main fan, 8 - First - layer converter, 9 - Second - layer converter, 10 - Tube - side inlet, 11 - Tube - side outlet, 12 - Shell - side inlet, 13 - Shell - side outlet. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0024] As Figure 1 and 2 shown, the intelligent prediction method for the sulfuric acid production process from smelting flue gas based on a dual - model of the present invention includes steps of constructing a data - driven model, constructing a mechanism - driven model, combining the two models, and system application. The specific content is as follows: A. Construction of a data - driven model: Collect data from the sulfuric acid production process from smelting flue gas and perform pre - processing; then extract characteristic parameters from the pre - processed data; subsequently, through feature importance evaluation and correlation analysis, screen out features that have a significant impact on the prediction target; then use LSTM to establish a time - series prediction model for the production process; finally, train the LSTM model with historical data to predict the change trend of key parameters in the production process, providing timely early warning and feedback for production regulation; B. Mechanism-driven model construction: Based on the basic principles of key chemical reactions in the sulfuric acid production process from smelting flue gas, a mechanism-driven model is established; C. Dual-mode combination: The mechanism-driven model provides an in-depth understanding of the physical and chemical laws of the reaction process, while the data-driven model dynamically predicts the reaction process through real-time data; D. System application: Integrate the model after the aforementioned dual-mode combination with existing smelting equipment and process flows to form a closed-loop control. Based on the real-time prediction of the data-driven model and the process optimization of the mechanism-driven model, use the model predictive control algorithm (MPC) to dynamically adjust production parameters.

[0025] The specific process of step A is as follows: A10. Data collection and preprocessing: Through intelligent sensors and data collection systems, collect real-time production data from each key equipment in the sulfuric acid production process from smelting flue gas, and then perform denoising, filling missing values, outlier detection, and data standardization on the collected real-time production data; Ensure the quality and consistency of the data through preprocessing, and provide a reliable data basis for subsequent analysis and modeling; A20. Feature selection and engineering: On the basis of the preprocessed data, use feature engineering techniques to extract feature parameters closely related to the sulfuric acid production process from smelting flue gas; Then, through feature importance evaluation and correlation analysis, adjust the values of the feature parameters and observe the degree of change of the prediction target, and screen out the features that have a significant impact on the prediction target (such as temperature change, SO 2 concentration, etc.), and remove redundant or less influential feature parameters on the prediction target to simplify the model and improve the prediction accuracy; A30. LSTM model construction: Use the LSTM (Long Short-Term Memory) network to establish a time series prediction model for the production process, and capture the long-term dependence characteristics in the time series through the gating mechanism to adapt to complex non-linear characteristics and dynamic changes; A40. LSTM model prediction: Train and optimize the LSTM model through historical data, and use the trained and optimized LSTM model to predict the change trend of key parameters in the production process to provide timely early warning and feedback for production control.

[0026] The key equipment includes reactors, absorption towers, fans, compressors, and heat exchangers. The production data includes temperature, gas concentration, fan speed, reactor pressure, current, and voltage. The feature parameters include the relationship between temperature and gas concentration, fan speed and gas flow, and reactor pressure and SO 2 concentration. The key parameters include temperature, pressure, and SO 2 concentration.

[0027] The basic principles of the key chemical reactions in Step B include SO 2 catalytic oxidation reaction and SO 3 absorption reaction; the chemical kinetic model in the mechanism-driven model is used to describe the reaction process of SO 2 being converted to SO 3 ; and the physical process is used to simulate the gas flow, temperature distribution and pressure change in the reactor, and the production process is optimized according to the simulation of the chemical reaction process and the physical process.

[0028] The input variables of the mechanism-driven model include the initial concentrations of SO 2 , O 2 , the instantaneous flue gas temperature during the reaction, the instantaneous conversion rate during the reaction, the reaction rate constant, the virtual contact time between the gas and the catalyst, etc., and the output variable is the reaction conversion rate completed per second. The mechanism-driven model simulates the interaction of multiple factors such as temperature change, reaction rate, and fan load during the conversion process of SO 2 through the mathematical expressions of physical and chemical reactions. The mechanism-driven model not only considers the basic chemical kinetics of the reaction process, but also integrates the influence of equipment performance and environmental conditions on the reaction process, thus providing a scientific production process model based on physical and chemical principles.

[0029] The mechanism-driven model constructs a scientific production process model based on physical and chemical principles by integrating the basic chemical kinetics of the reaction process, equipment performance, and the influence of environmental conditions. This model not only considers the kinetic characteristics of key reactions such as SO 2 catalytic oxidation and SO 3 absorption, but also combines the operating parameters of equipment such as converters and absorption towers (such as temperature, flow rate, catalyst performance, etc.), and dynamically adapts to changes in environmental conditions (such as temperature, humidity, pressure). Through experimental data verification and the application of optimization algorithms, the model can achieve accurate prediction and optimal control of the entire process, thereby improving production efficiency, stabilizing product quality, and reducing energy consumption.

[0030] In Step C, the theoretical output of the mechanism-driven model is used as the input feature of the data-driven model, or the data-driven model is used to correct the error of the mechanism-driven model. At the same time, the model parameters are dynamically adjusted through physical constraints and real-time data feedback, so as to give full play to the advantages of both, improve the accuracy and adaptability of the model, and achieve intelligent optimization control of the entire process.

[0031] The combination of the data-driven model and the mechanism-driven model can be achieved through various methods such as feature enhancement, model integration, residual modeling, parameter calibration, hybrid modeling, physics-informed neural network (PINN), and dynamic linkage and real-time optimization.

[0032] In step D, the integration to form a closed-loop control is to monitor various key process parameters of the system in real time during the production process, compare the true values of various key process parameters with the model prediction results, and dynamically adjust the model parameters according to the error to ensure that the production process is always in the best working state and ensure that the process parameters are stable within a predetermined range.

[0033] In step D, the model predictive control algorithm (MPC) performs rolling optimization based on the fused model, dynamically adjusts key process parameters (such as temperature, gas concentration, fan speed, etc.) by predicting the future system output and minimizing the error, and ensures that the production process operates in an optimal state.

[0034] The dynamic adjustment of production parameters in step D is based on real-time data and prediction results, and intelligently adjusts process parameters (such as temperature, gas concentration, fan speed, etc.) to ensure that the production process proceeds in an optimal state.

[0035] Step D also includes anomaly detection and handling. When the predicted value exceeds the standard or an abnormal situation occurs during the production process, it automatically identifies and triggers an early warning mechanism.

[0036] The intelligent prediction system for the sulfuric acid production process based on a dual model of the present invention includes a data-driven module, a mechanism-driven module, a dual-model combination module, and a system application module. The data-driven module is used to collect data on the sulfuric acid production process from the smelting flue gas and perform preprocessing; then extract characteristic parameters from the preprocessed data; subsequently, through feature importance evaluation and correlation analysis, screen out the features that have a significant impact on the prediction target; then use LSTM to establish a time series prediction model for the production process; finally, train the LSTM model with historical data to predict the change trend of key parameters in the production process, providing timely early warning and feedback for production control. The mechanism-driven module establishes a mechanism-driven model based on the basic principles of key chemical reactions in the sulfuric acid production process from the smelting flue gas. The dual-model combination module is used to enable the mechanism-driven model in the mechanism-driven module to provide an in-depth understanding of the physical and chemical laws of the reaction process, while the data-driven module dynamically predicts the reaction process through real-time data. The system application module is used to integrate the aforementioned dual-model combination module with existing smelting equipment and process flows to form a closed-loop control, and based on the real-time prediction of the data-driven module and the process optimization of the mechanism-driven module, adopt a model predictive control algorithm to dynamically adjust production parameters.

[0037] The data-driven module includes a data collection and preprocessing unit, a feature selection and engineering unit, an LSTM model construction unit, and an LSTM model prediction unit. The data acquisition and preprocessing unit is used to collect real-time production data from each key equipment in the sulfuric acid production process from smelting fumes through intelligent sensors and a data acquisition system, and then perform denoising, missing value filling, outlier detection and data standardization on the collected real-time production data; The feature selection and engineering unit is used to extract feature parameters closely related to the sulfuric acid production process from smelting fumes based on the preprocessed data by using feature engineering techniques; then, through feature importance evaluation and correlation analysis, adjust the values of the feature parameters and observe the degree of change of the prediction target, screen out the features that have a significant impact on the prediction target, and remove redundant or feature parameters that have a low impact on the prediction target; The LSTM model construction unit is used to establish a time series prediction model of the production process by using the LSTM network, and capture the long-term dependence characteristics in the time series through the gating mechanism; The LSTM model prediction unit is used to train and optimize the LSTM model through historical data, and use the trained and optimized LSTM model to predict the change trend of key parameters in the production process, providing timely early warning and feedback for production control.

[0038] In the mechanism-driven module, the basic principles of key chemical reactions include SO 2 catalytic oxidation reaction, SO 3 absorption reaction; use the chemical kinetic model in the mechanism-driven model to describe the reaction process of SO 2 converted to SO 3 ; and use physical process simulation to simulate the gas flow, temperature distribution and pressure change in the reactor, and optimize the production process according to the chemical reaction process and physical process simulation.

[0039] The system application module also includes an anomaly detection and processing unit, which is used to automatically identify and trigger an early warning mechanism when the predicted value exceeds the standard or is abnormal during the production process.

[0040] As Figure 2 shown, the intelligent prediction system for the sulfuric acid production process from smelting fumes based on a dual model divides the system into four parts: equipment layer, data layer, model layer and application layer on an industrial basis, where the model layer is the core, specifically involving the dual-mode drive of a data-driven model and a mechanism-driven model.

[0041] Part 1: Equipment layer The equipment layer includes key equipment in the process of smelting flue gas to sulfuric acid (such as reactors, absorption towers, fans, compressors, heat exchangers, etc.). In the present invention, the equipment layer not only undertakes the core tasks of actual production (chemical reactions and physical processes), but also is seamlessly connected to the data layer through sensors and intelligent devices, collecting the operating status of the equipment and process parameters (such as temperature, pressure, gas concentration, fan speed, current and voltage, etc.) in real time and uploading them to the data platform, providing raw data support for data processing and model training.

[0042] The second part: Data layer The data layer collects and stores data in the process of smelting flue gas to sulfuric acid through various sensors, intelligent metering devices and Internet of Things technologies. These data include but are not limited to: temperature (such as reactor temperature, absorption tower temperature, etc.), gas concentration (such as SO 2 , O 2 concentration), pressure, fan speed, current and voltage, etc.

[0043] In the data layer, the collected data is transmitted to the data processing center in the cloud or locally through the data acquisition system (DCS) and industrial Internet of Things (IioT). The data is processed through standardization, converted into a format that can be used for model training, and a real-time data stream is provided for subsequent analysis and prediction.

[0044] The third part: Model layer (core layer) The model layer mainly includes two types of models: data-driven models (such as LSTM neural networks); mechanism-driven models (reaction models based on chemical reaction kinetics and physical process models).

[0045] The dual-mode driving method of the present invention combines data-driven models and mechanism-driven models to ensure that the optimization control of the process of smelting flue gas to sulfuric acid can be carried out by simultaneously using historical data and chemical and physical theories.

[0046] 3.1 Data-driven model: LSTM neural network The data-driven model mainly relies on historical data for modeling and prediction. The LSTM neural network is used for time series analysis of production data to accurately predict key parameters in the production process (such as temperature, SO 2 concentration, pressure, etc.). The specific steps are as follows: Data processing and preprocessing: As mentioned above, the collected raw data is preprocessed and then input into the LSTM model for training.

[0047] Model training: Using historical production data as input, the LSTM model learns the long-term dependencies in the production process and predicts the change trends of key parameters.

[0048] Prediction and Optimization: By predicting future process parameter changes, the LSTM model provides a basis for real-time adjustment of production control.

[0049] 3.2 Mechanism-Driven Model: Simulation of Chemical Reactions and Physical Processes The mechanism-driven model optimizes the production process by simulating the chemical reactions and physical processes in the processes such as SO 2 catalytic oxidation and SO 3 absorption. This model is based on the principles of chemical reaction kinetics and thermodynamics, considering the relationships among factors such as reaction rate, temperature, and gas concentration.

[0050] Reaction Rate Equation: Use a chemical kinetics model to describe the reaction process of SO 2 converted to SO 3 ; Physical Process Simulation: Simulate the gas flow, temperature distribution, and pressure changes in the reactor.

[0051] 3.3 Dual-Mode Driving: Combination of Data-Driven and Mechanism-Driven The dual-mode driving method of the present invention combines a data-driven model with a mechanism-driven model. The data-driven model is used to make real-time predictions about the production process, and the mechanism-driven model provides theoretical support for the reaction process.

[0052] The combination of the two can: Optimize the control strategy: Based on the prediction results of the data-driven model, the mechanism-driven model conducts theoretical verification on the production process to ensure the scientificity and accuracy of the control strategy; Enhance the model accuracy: Through continuous iteration and optimization, improve the prediction ability of the model and make up for the deficiencies of a single model.

[0053] Part Four: Application Layer (System Integration and Application) The application layer is responsible for applying the output of the model layer to the actual production of the sulfuric acid production process from smelting flue gas. Through an intelligent control system, the prediction results of the data-driven model and the mechanism-driven model are combined to dynamically adjust the production process parameters to ensure optimal control of the production process.

[0054] System integration and application include the following steps: Real-time data collection and analysis: By real-time monitoring of production data, the intelligent system predicts the change trends of key parameters (such as temperature, gas concentration, etc.) in the production process through the data-driven model; Process adjustment: Based on the prediction results of the LSTM model, combined with the optimization of the reaction process by the mechanism-driven model, the operation data of the reactor is obtained in real time through the data acquisition system, and the LSTM model is used to train the historical data to predict the change trend of future process parameters; at the same time, a mechanism-driven model is established based on the physical and chemical principles of the reaction process to provide theoretical guidance; then, the prediction results of the LSTM model are combined with the theoretical output of the mechanism-driven model, and the control instructions for the same parameter are coordinated by methods such as weighted average or residual correction to adjust process parameters such as the reactor temperature, fan speed, and gas flow rate.

[0055] Feedback control: Through the feedback mechanism, the actual data is compared with the prediction results in real time, and the control strategy is dynamically adjusted. The MPC algorithm uses the feedback correction mechanism to continuously correct the control instructions by optimizing the constraints of the problem to adapt to the changes in the production process. The combined use of reinforcement learning and MPC further improves the adaptability and robustness of the control system to dynamic changes, corrects the deviations in the production process, and ensures stable and efficient production.

[0056] Specific steps for application integration: Data input and control instruction generation: In the process of data input and control instruction generation, the real-time data is input into the model layer through the data acquisition system, and the LSTM model and the mechanism-driven model generate control instructions respectively. In order to coordinate the control instructions for the same parameter, methods such as weighted average method, residual correction method, multi-model voting mechanism or dynamic adjustment strategy can be adopted; Execute system adjustment: The control instructions adjust the parameters of the production equipment through the execution system, such as the reactor temperature, absorption tower pressure, fan speed, etc.; System optimization and adaptive adjustment: Through the feedback control mechanism, the system parameters are adjusted in real time, and the production process is continuously optimized to ensure the optimal state.

[0057] Example 1 As Figure 1 shown, soft measurement is carried out on the outlet pressure of the high-concentration flue gas sulfuric acid blower of a copper smelting enterprise. The purified smelting flue gas mainly contains SO 2 and O 2 , and after passing through the shell side of heat exchanger I1 and heat exchanger II2 through the SO 2 main blower 7, it exchanges heat with the tube-side gas and reaches the inlet of the first-layer converter 8 of the pre-converter 3. No chemical reaction occurs during this process; SO 2 and O 2 react in the first-layer converter 8 of the pre-converter 3 under the action of the catalyst to generate SO 3 ; the mixed gas leaves the first-layer converter 8 of the pre-converter 3 and is cooled through the tube side of heat exchanger II2, and then enters the second-layer converter 9 of the pre-converter 3 for further conversion. During the conversion process, the gas is SO2 , O 2 and SO 3 mixed gas; the mixed gas after secondary conversion enters the SO 3 cooler 4 after being cooled in the tube side of heat exchanger I1, and the separated SO 3 enters the pre-absorption tower 5 for absorption, and the remaining unabsorbed mixed gas enters the SO 3 cooler 4 again through the relay fan 6.

[0058] The intelligent prediction in the process of smelting flue gas to sulfuric acid is as follows: S100: Collect and preprocess the data in the process of smelting flue gas to sulfuric acid; then extract the characteristic parameters from the preprocessed data; subsequently, through the evaluation of feature importance and correlation analysis, screen out the features that have a significant impact on the prediction target; then use LSTM to establish a time series prediction model of the production process; finally, train the LSTM model with historical data to predict the change trend of key parameters in the production process, providing timely early warning and feedback for production regulation. The specific process is as follows: S110: Through intelligent sensors and data acquisition systems, collect real-time production data (including temperature, gas concentration, fan speed, reactor pressure, current and voltage) from each key equipment (including reactors, absorption towers, fans, compressors and heat exchangers) in the process of smelting flue gas to sulfuric acid, and then perform denoising, filling missing values, outlier detection and data standardization processing on the collected real-time production data; ensure the quality and consistency of the data through preprocessing, providing a reliable data basis for subsequent analysis and modeling.

[0059] S120. On the basis of the preprocessed data, adopt feature engineering techniques to extract the characteristic parameters closely related to the process of smelting flue gas to sulfuric acid (including the relationship between temperature and gas concentration, fan speed and gas flow, reactor pressure and SO 2 concentration); then through the evaluation of feature importance and correlation analysis, adjust the values of the characteristic parameters and observe the degree of change of the prediction target, screen out the features that have a significant impact on the prediction target (such as temperature change, SO 2 concentration, etc.), and remove the redundant or less influential characteristic parameters on the prediction target to simplify the model and improve the prediction accuracy.

[0060] S130: Use the LSTM (Long Short-Term Memory) network to establish a time series prediction model of the production process, and capture the long-term dependence characteristics in the time series through the gating mechanism to adapt to complex non-linear characteristics and dynamic changes.

[0061] S140: Train and optimize the LSTM model with historical data, and use the trained and optimized LSTM model to predict the key parameters in the production process (including temperature, pressure and SO2 The change trend of (concentration) provides timely warning and feedback for production regulation.

[0062] S200: Based on the basic principles of key chemical reactions in the sulfuric acid production process from smelting flue gas (including the catalytic oxidation reaction of SO 2 and the absorption reaction of SO3), a mechanism-driven model is established (the mechanism-driven model simulates the interaction of multiple factors such as temperature change, reaction rate, and fan load during the SO 2 conversion process through mathematical expressions of physical and chemical reactions. The mechanism-driven model not only considers the basic chemical kinetics of the reaction process but also integrates the influence of equipment performance and environmental conditions on the reaction process, thus providing a scientific production process model based on physical and chemical principles).

[0063] Among them, the chemical kinetics model in the mechanism-driven model is used to describe the reaction process of SO 2 converted to SO 3 ; and the physical process is used to simulate the gas flow, temperature distribution, and pressure change in the reactor, and the production process is optimized according to the simulation of the chemical reaction process and the physical process.

[0064] S300: The mechanism-driven model provides an in-depth understanding of the physical and chemical laws of the reaction process, while the data-driven model dynamically predicts the reaction process through real-time data; among them, the theoretical output of the mechanism-driven model is used as the input feature of the data-driven model, or the data-driven model is used to correct the error of the mechanism-driven model, and at the same time, the model parameters are dynamically adjusted through physical constraints and real-time data feedback, so as to give full play to the advantages of both and improve the accuracy and adaptability of the model, realizing intelligent optimization control of the whole process.

[0065] S400: Integrate the combined model of the two modes with the existing smelting equipment and process flow to form a closed-loop control (in the production process, key process parameters of the real-time monitoring system are monitored, and the key process parameters are compared with the model prediction results, and the control strategy is adjusted according to the error between the true value of the parameters and the prediction results to ensure that the production process is always in the best working state and ensure that the process parameters are stable within the predetermined range). Based on the real-time prediction of the data-driven model and the process optimization of the mechanism-driven model, the model predictive control algorithm (MPC) is used to intelligently adjust the process parameters (such as temperature, gas concentration, fan speed, etc.) based on real-time data and prediction results to ensure that the production process is carried out in the optimal state.

[0066] Moreover, the combined model of the two modes will also continuously learn and optimize the historical production data, continuously improve the prediction accuracy and control strategy, and improve the intelligent level of the production process.

[0067] It also includes anomaly detection and handling, which automatically identifies and triggers an early warning mechanism when the predicted value exceeds the standard or shows anomalies during the production process.

[0068] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. An intelligent prediction method for smelting flue gas acid production process based on dual models, characterized by: It includes data-driven model construction, mechanism-driven model construction, dual-mode combination, and system application steps. The specific contents are as follows: A. Data-driven model construction: Collect data from the smelting flue gas acid production process and preprocess it; then extract feature parameters from the preprocessed data; then screen out features that have a significant impact on the prediction target through feature importance evaluation and correlation analysis; then use LSTM to establish a time series prediction model for the production process; finally, train the LSTM model through historical data to predict the change trend of key parameters in the production process, and provide timely warning and feedback for production control; B. Construction of mechanism-driven model: Based on the basic principles of key chemical reactions in the process of smelting flue gas acid production, a mechanism-driven model is established; C. Dual-mode combination: The mechanism-driven model provides an in-depth understanding of the physical and chemical laws of the reaction process, while the data-driven model dynamically predicts the reaction process through real-time data; D. System application: Integrate the aforementioned dual-mode model with existing smelting equipment and process flow to form a closed-loop control, based on real-time prediction of the data-driven model and process optimization of the mechanism-driven model, and use predictive control algorithms to dynamically adjust production parameters.

2. The intelligent prediction method for smelting flue gas acid-making process based on dual models according to claim 1 is characterized by: The specific process of step A is as follows: A10. Data collection and preprocessing: Through intelligent sensors and data collection systems, real-time production data is collected from key equipment in the smelting flue gas acid production process, and then the collected real-time production data is denoised, missing values ​​are filled, outliers are detected, and data is standardized; A20, Feature selection and engineering: Based on the preprocessed data, feature engineering technology is used to extract feature parameters closely related to the smelting flue gas acid production process; then, through feature importance evaluation and correlation analysis, the values ​​of feature parameters are adjusted and the degree of change of the prediction target is observed, and the features with significant impact on the prediction target are screened out, and redundant feature parameters or feature parameters with low impact on the prediction target are removed; A30, LSTM model construction: Use LSTM network to build a time series prediction model for the production process, and capture the long-term dependency characteristics in the time series through the gating mechanism; A40, LSTM model prediction: Train and optimize the LSTM model through historical data, and use the trained and optimized LSTM model to predict the changing trends of key parameters in the production process, providing timely warning and feedback for production control.

3. The intelligent prediction method for smelting flue gas acid-making process based on dual models according to claim 2 is characterized by: The key equipment includes a reactor, an absorption tower, a fan, a compressor and a heat exchanger. The production data includes temperature, gas concentration, fan speed, reactor pressure, current and voltage. The characteristic parameters include the relationship between temperature and gas concentration, fan speed and gas flow, reactor pressure and SO2 concentration. The key parameters include temperature, pressure and SO2 concentration.

4. The intelligent prediction method for smelting flue gas acid-making process based on dual models according to claim 1 is characterized by: The basic principles of the key chemical reactions in step B include SO2 catalytic oxidation reaction and SO3 absorption reaction; the chemical kinetic model in the mechanism-driven model is used to describe the reaction process of SO2 conversion to SO3; and the physical process is used to simulate the gas flow, temperature distribution and pressure change in the reactor, and the production process is optimized based on the chemical reaction process and physical process simulation.

5. The intelligent prediction method for smelting flue gas acid-making process based on dual models according to claim 1 is characterized by: In step C, the theoretical output of the mechanism-driven model is used as the input feature of the data-driven model, or the data-driven model is used to correct the error of the mechanism-driven model, and the model parameters are dynamically adjusted through physical constraints and real-time data feedback.

6. The intelligent prediction method for smelting flue gas acid-making process based on dual models according to claim 1 is characterized by: The D step also includes anomaly detection and processing, which is to automatically identify and trigger an early warning mechanism when the predicted value exceeds the standard or an abnormal situation occurs during the production process.

7. An intelligent prediction system for smelting flue gas acid production process based on dual models, characterized by: Including data-driven module, mechanism-driven module, dual-mode combination module, and system application module. The data-driven module is used to collect and preprocess the data of the smelting flue gas acid-making process; then extract feature parameters from the preprocessed data; then screen out features that have a significant impact on the prediction target through feature importance evaluation and correlation analysis; then use LSTM to establish a time series prediction model for the production process; finally, train the LSTM model through historical data to predict the change trend of key parameters in the production process, and provide timely warning and feedback for production control; The mechanism-driven module establishes a mechanism-driven model based on the basic principles of key chemical reactions in the smelting flue gas acid production process; The dual-mode combination module is used to enable the mechanism-driven model in the mechanism-driven module to provide an in-depth understanding of the physical and chemical laws of the reaction process, while the data-driven module dynamically predicts the reaction process through real-time data; The system application module is used to integrate the aforementioned dual-mode combination module with existing smelting equipment and process flow to form a closed-loop control, based on the real-time prediction of the data-driven module and the process optimization of the mechanism-driven module, and adopts a predictive control algorithm to dynamically adjust production parameters.

8. The intelligent prediction system for smelting flue gas acid production process based on dual models according to claim 7 is characterized by: The data-driven module includes a data acquisition and preprocessing unit, a feature selection and engineering unit, an LSTM model building unit, and an LSTM model prediction unit. The data acquisition and preprocessing unit is used to collect real-time production data from key equipment in the smelting flue gas acid production process through intelligent sensors and data acquisition systems, and then perform denoising, missing value filling, outlier detection and data standardization on the collected real-time production data; The feature selection and engineering unit is used to extract feature parameters closely related to the smelting flue gas acid-making process by using feature engineering technology on the basis of preprocessed data; then, through feature importance evaluation and correlation analysis, the values ​​of the feature parameters are adjusted and the degree of change of the prediction target is observed, the features with significant impact on the prediction target are screened out, and redundant feature parameters or feature parameters with low impact on the prediction target are removed; The LSTM model building unit is used to establish a time series prediction model for the production process using an LSTM network, and to capture long-term dependency characteristics in the time series through a gating mechanism; The LSTM model prediction unit is used to train and optimize the LSTM model through historical data, and use the trained and optimized LSTM model to predict the changing trends of key parameters in the production process, so as to provide timely warning and feedback for production control.

9. The intelligent prediction system for smelting flue gas acid-making process based on dual models according to claim 7 is characterized by: In the mechanism-driven module, the basic principles of key chemical reactions include SO2 catalytic oxidation reaction and SO3 absorption reaction; the chemical kinetic model in the mechanism-driven model is used to describe the reaction process of SO2 conversion to SO3; and the physical process is used to simulate the gas flow, temperature distribution and pressure change in the reactor, and the production process is optimized according to the chemical reaction process and physical process simulation.

10. The intelligent prediction system for smelting flue gas acid-making process based on dual models according to claim 7 is characterized by: The system application module also includes an anomaly detection and processing unit, which is used to automatically identify and trigger an early warning mechanism when the predicted value exceeds the standard or an abnormal situation occurs during the production process.

Images