A dynamic optimization control method for tin smelting process

Abstract

The present invention relates to a dynamic optimization control method for a tin smelting process, and belongs to the interdisciplinary technical field of metallurgical engineering and artificial intelligence. First, data is collected and preprocessed, outliers are eliminated, and missing values ​​are generated using cGAN to fill data gaps; then, the key state variables in the tin smelting process are modeled using an LSTM model to capture the complex time series change patterns and embed metallurgical knowledge as constraints. Next, a data-driven state update model is added to the reinforcement learning environment. Finally, a deep reinforcement learning algorithm is introduced, and the spray gun operating parameters are used as the action space, and the state of the tin smelting process is used as the state space. In the reward function design, the prediction results of key parameters are incorporated, and a dynamic optimization control strategy is implemented through intelligent agent learning. This method has significant advantages in dynamically optimizing tin purity and reducing CO emissions, and provides an innovative solution for intelligent and efficient production in the metallurgical industry.

Description

Technical Field

[0001] The invention relates to a dynamic optimization control method for a tin smelting process, and belongs to the intersecting technical field of metallurgical engineering and artificial intelligence. Background Art

[0002] Tin smelting, a crucial process in nonferrous metal smelting, involves complex physical and chemical reactions and multivariable coupling. Traditional tin smelting control methods typically rely on empirical rules and static optimization strategies, which struggle to cope with the dynamic changes and complex nonlinear relationships within the furnace. This approach faces numerous challenges in practical application, including the difficulty in accurately predicting the dynamic changes in key process indicators (such as tin purity and CO emission concentration) and the difficulty in optimizing control strategies in real time. These challenges lead to high energy consumption, high emissions, and fluctuating product quality.

[0003] In recent years, with the rapid development of intelligent manufacturing technologies, deep learning and reinforcement learning have provided new solutions for the intelligent optimization of tin smelting processes. The superior performance of recurrent neural networks and their improved models (such as long short-term memory networks) in time series modeling has enabled the accurate prediction of key parameters in the tin smelting process. Deep reinforcement learning technology, on the other hand, provides a theoretical foundation for the development of dynamic optimization control strategies. However, relying solely on data-driven machine learning models can easily overlook physical constraints such as stoichiometry and energy balance in metallurgical processes, potentially leading to unreliable prediction and optimization results.

[0004] To this end, a dynamic optimization control method that integrates data-driven and metallurgical knowledge is being developed. This method fully exploits the data potential of the tin smelting process while embedding domain knowledge to enhance model interpretability and accuracy. By constructing a reinforcement learning environment, using the spray gun operating parameters as the action space and the key states of the tin smelting process as the state space, and designing a reward function that can perceive changes in process conditions in real time, the intelligent agent can autonomously learn and dynamically optimize the control strategy, thereby achieving improved tin purity, reduced CO emissions, and minimized energy consumption. This technical background lays the theoretical and practical foundation for intelligent optimization of the tin smelting process. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a dynamic optimization control method for a tin smelting process, which can optimize the operating parameters in the smelting process in real time, thereby solving the above-mentioned problem.

[0006] The technical solution of the present invention is a dynamic optimization control method for a tin smelting process. First, relevant data is collected from a tin smelting plant database, including state variables and spray gun operating parameters at each moment during the smelting process. Second, the data is preprocessed to remove outliers recorded by sensors. A conditional generative adversarial network (cGAN) is used to generate key variables measured at low frequencies to fill data gaps. Subsequently, a long short-term memory (LSTM) model is used to model key state variables in the tin smelting process (such as tin purity and CO concentration), capturing time series variations. Metallurgical knowledge (such as the oxygen-coal stoichiometric ratio and oxygen flow rate changes) is embedded as constraints to improve the accuracy and reliability of the model's predictions. Within a reinforcement learning environment, a data-driven state update model is employed, taking key state variables and spray gun operating parameters as input. A deep learning model is used to predict the next state at the next moment, incorporating metallurgical knowledge to enhance the credibility of state transitions. Finally, a deep reinforcement learning algorithm was introduced, using the spray gun operating parameters as the action space and the tin smelting state as the state space. A reward function was designed to reflect the optimization goals of increasing tin purity and reducing CO emissions. This enabled the intelligent agent to learn the optimal policy, which guided it to generate optimal spray gun operation recommendations under different environmental conditions.

[0007] The specific steps are:

[0008] Step 1: Collect relevant parameter data during the tin smelting process, including the state variable values ​​and spray gun operating parameter values ​​at each moment in the smelting furnace, to provide a comprehensive data basis for subsequent analysis and modeling;

[0009] Step 2: Preprocess the relevant parameter data, delete outliers recorded by the sensors, and use cGAN to generate variables with measurement frequencies below a preset threshold, thereby filling data gaps and improving data integrity.

[0010] Step 3: Use the improved LSTM model to model the tin purity and CO concentration during the tin smelting process, capture the time series variation patterns of state variable values ​​and spray gun operating parameter values, and improve the accuracy and reliability of the prediction;

[0011] Step 4: Build a reinforcement learning environment and adopt a data-driven state update model. The state variable values ​​and spray gun operating parameter values ​​during the tin smelting process are used as input. A deep learning model is used to predict the state value at the next moment. By integrating metallurgical knowledge constraints, the reliability of state transitions and the actual performance of the environment model are enhanced.

[0012] Step 5: Design a reward function to guide the reinforcement learning agent toward improving tin purity and reducing CO emissions based on the captured time series variation patterns. This reward function provides real-time feedback on the agent's decision-making effectiveness, providing effective guidance for the optimization process.

[0013] Step 6: Introduce the Deep Reinforcement Learning (DRL) algorithm, design the spray gun operation parameters as an action space, and the state of the tin smelting process as a state space. The intelligent agent and the state update model interact and learn, and continuously optimize the control strategy to achieve dynamic optimization goals. This strategy can guide the intelligent agent to generate optimal spray gun operation suggestions under different environmental conditions.

[0014] The Step 2 is specifically as follows:

[0015] Step 2.1: Based on the sensor measurement range and the experience of process experts, threshold values ​​for each variable are set, and outliers exceeding the threshold are directly eliminated to ensure the accuracy and reliability of subsequent modeling data;

[0016] Step 2.2: Perform data partitioning, use the complete high-frequency measurement variables as the conditional input of the model, and the low-frequency measurement variables as the target generation variables, and use the cGAN model to generate missing values. The high-frequency measurement variables are used as conditional inputs to the generator and discriminator to provide additional constraint information, thereby improving data integrity and consistency.

[0017] The Step 3 is specifically as follows:

[0018] Step 3.1: Based on existing industrial data from the tin smelting process, core variables reflecting the dynamic changes within the furnace are extracted. New feature columns related to the smelting process are generated through mathematical transformations. These features not only enrich the data dimensions but also provide more meaningful input information for subsequent modeling.

[0019] Step 3.2: Embed metallurgical physical constraints into the original LSTM model's loss function. This constraint penalty term guides the LSTM model to learn predictions that conform to actual process rules, improving the scientific nature and interpretability of the predictions.

[0020] Step 3.3: Use the improved LSTM model to predict the indicator variables in the tin smelting process. The gating mechanism captures the complex time dependencies and accurately models the dynamic changes between variables.

[0021] Step 3.4: After training is complete, use an independent test dataset to verify the model's predictive performance, evaluate the prediction accuracy of tin purity and CO concentration, and save the LSTM model parameters to ensure its applicability in industrial scenarios.

[0022] The Step 4 is specifically as follows:

[0023] Step 4.1: Obtain the state changes of consecutive time steps through time series data, including state variables and action variables, and organize the data into triples:

[0024]

[0025] Where, Represents the state variable at the current moment, Represents the action variable at the current moment, Represents the state variables at the next moment. This data format lays the foundation for the training of the subsequent state update model;

[0026] Step 4.2: Build a state update model of the fully connected neural network, and update the current state and the current action As input, predict the next state ,Use the constructed triple data to train the state update model;

[0027] Step 4.3: Evaluate the performance of the state update model on the validation set and save the state update model parameters that meet the preset indicator requirements to provide a reliable state update mechanism for subsequent embedding in the reinforcement learning environment.

[0028] Step 4.4: In the reinforcement learning environment, the trained state is used to update the model parameters. The state of the reinforcement learning environment at the next moment is predicted through the current environment and the currently executed action, guiding the agent to make action selections. This embedding process enhances the environment's ability to describe complex dynamic changes, provides the agent with more realistic state feedback, guides it to optimize action selection, and achieves efficient control of the tin smelting process.

[0029] The Step 5 is specifically as follows:

[0030] Step 5.1: Establish optimization goals: improve tin purity and reduce CO emissions. Also set auxiliary goals, including reducing energy consumption and maintaining the stability of key process parameters such as furnace pressure, to achieve multi-objective collaborative optimization and ensure efficient operation of the tin smelting process.

[0031] Step 5.2: Use the saved LSTM model parameters to predict the impact of adjusting the current process parameters on tin purity and CO emissions in a preset time period in the future. Use the predicted results as the basis for calculating the reward function to accurately reflect the contribution of the optimization strategy to the long-term goal.

[0032] Step 5.3: Design the reward function, the formula is:

[0033] ( )

[0034] Where R is the reward function, 、 、 is the weight coefficient, which is used to balance the importance of multiple objectives. is the tin purity content at the current moment, is the carbon monoxide content at the current moment, is the tin purity content at the next moment, is the carbon monoxide content at the next moment, is the energy consumption value at the next moment;

[0035] To enhance real-time performance and robustness, the reward function was designed by incorporating the prediction results of key parameters based on the LSTM model, decomposing the long-term optimization goal of the tin smelting process into several short-term optimization sub-goals. This allows the system to more accurately reflect the impact of current process adjustments on future states, thereby effectively guiding the intelligent agent to learn and continuously optimize the control strategy.

[0036] The Step 6 is specifically as follows:

[0037] Step 6.1: Regulate the spray gun operating parameters based on the DDPG algorithm. Define the operations performed by the spray gun in the furnace as the action space, and define the ten core variables in the furnace as the state space: total concentration, melt pool accumulation, oxygen content percentage, furnace bottom middle temperature, furnace bottom external temperature, furnace bottom internal temperature, furnace rise temperature, furnace pressure, exhaust CO analysis, and total energy consumption.

[0038] Step 6.2: Use the Actor network in the DDPG algorithm to predict the optimal action based on the current state to adjust the spray gun operating parameters. Use the Critic network to evaluate the Q value of the current strategy to guide Actor optimization.

[0039] Step 6.3: The interaction data between the reinforcement learning agent and the furnace environment is stored in an experience pool. Randomly sample several furnace status and spray gun operating parameter data from this pool for training. This breaks the temporal correlation and enables the reinforcement learning agent to learn operational experience that was not previously available in the continuous furnace period data, thereby improving the training stability and generalization ability of the model.

[0040] Step 6.4: By adding noise, the agent's ability to explore the action space is enhanced to avoid falling into local optimality. At the same time, the factory failure situation is simulated to complete the model training.

[0041] Step 6.5: Utilize the trained model to implement intelligent dynamic optimization control during the tin smelting process. By real-time monitoring of state variables and combining the model optimization results, optimization suggestions are provided to the spray gun operator to dynamically adjust operating parameters, thereby improving tin purity, reducing CO emissions, and maximizing process energy efficiency.

[0042] The Step 3.2 is specifically as follows:

[0043] Add metallurgical knowledge constraints to the loss function of the LSTM model. The constraint formula is:

[0044]

[0045] in, is the total loss value, is the prediction loss value of the LSTM model, is a constant, The function needs to be combined with metallurgical knowledge. The specific formula is as follows:

[0046]

[0047] in, is a constant, is the difference between the predicted value and the theoretical value. The theoretical value of CO is calculated as follows:

[0048]

[0049] Where, represents the combustion coal flow rate, is a constant related to the calorific value and combustion efficiency of fuel coal, is the oxygen flow rate, is the theoretical stoichiometric ratio for burning coal.

[0050] The beneficial effects of the present invention are:

[0051] (1) Dynamic optimization control capability: Compared with existing static optimization methods, this invention achieves dynamic optimization control of the tin smelting process by combining deep reinforcement learning algorithms and data-driven models. The intelligent agent can monitor the status in real time and autonomously optimize the spray gun operation strategy to cope with complex process changes;

[0052] (2) Accurate modeling integrating metallurgical knowledge: Different from traditional data-driven models, this invention introduces metallurgical physical constraints such as the oxygen-coal stoichiometric ratio and combines it with an improved recurrent neural network (RNN) to accurately capture the dynamic changes of key parameters and improve prediction accuracy and reliability;

[0053] (3) Efficient environment construction: The present invention adopts a data-driven state update model to accurately reflect the dynamic characteristics of tin smelting, provide a reliable state transfer mechanism for reinforcement learning, avoid the failure of optimization strategy due to accumulation of environmental errors, and thus improve learning efficiency and optimization performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is the overall framework diagram of the present invention;

[0055] Figure 2 This is a comparison chart of the predicted value and the actual value of the total amount of concentration in the embodiment of the present invention;

[0056] Figure 3 1 is a comparison chart of the predicted value and the actual value of the melt pool accumulation in an embodiment of the present invention;

[0057] Figure 4 1 is a comparison chart of the predicted value and the actual value of the oxygen content percentage in an embodiment of the present invention;

[0058] Figure 5 This is a comparison chart of the predicted value and the actual value of the temperature in the middle of the furnace bottom according to an embodiment of the present invention;

[0059] Figure 6 1 is a comparison chart of the predicted value and the actual value of the external temperature of the furnace bottom according to an embodiment of the present invention;

[0060] Figure 7 This is a comparison chart of the predicted value and the actual value of the furnace bottom temperature in an embodiment of the present invention;

[0061] Figure 8 is a comparison chart of the predicted value and the actual value of the temperature rise of the furnace according to an embodiment of the present invention;

[0062] Figure 9 1 is a comparison chart of the predicted value and the actual value of the furnace pressure in an embodiment of the present invention;

[0063] Figure 10 This is a comparison chart of the predicted value and the actual value of the CO content in the exhaust gas according to the embodiment of the present invention;

[0064] Figure 11 This is a comparison chart of the predicted value and the actual value of the total energy consumption of the embodiment of the present invention. DETAILED DESCRIPTION

[0065] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0066] Example 1: Figure 1 As shown, a dynamic optimization control method for tin smelting process, the specific steps are:

[0067] Step 1: Collect relevant parameter data during the tin smelting process. The relevant parameter data includes the state variable values ​​and spray gun operation parameter values ​​at each moment in the smelting furnace, providing a comprehensive data basis for subsequent analysis and modeling.

[0068] Specifically, to validate the real-time dynamic optimization effectiveness of the proposed model in a tin smelting process, this example evaluated the effectiveness of improving tin purity, reducing CO emissions, and optimizing energy consumption by simulating the interaction between the reinforcement learning model and the industrial production environment, based on actual operational data from a tin smelting plant. Thirty consecutive days of smelting data were extracted from the data system of the Ausmel furnace at the tin smelting plant, with a time resolution of 1 minute. Data size: 43,200 records, including 10 state variables and 8 operating parameters. Details are shown in Tables 1 and 2.

[0069] Table 1 Action variables performed by the spray gun in the tin factory data

[0070] Action variable name unit Total material flow tons / hour Fuel coal flow kg / hour Coal-carrying air flow Standard cubic meter / h Coal loading air pressure kPa Gun back pressure kPa Gun Position mm Oxygen flow Standard cubic meters per hour Air flow Standard cubic meter / h

[0071] Table 2 State variables of the Ausmel furnace in the tin factory data

[0072] State variable name unit Total amount of concentrate ton Melt pool accumulation ton oxygen concentration percentage Temperature of the middle part of furnace bottom temperature Furnace bottom external temperature temperature Furnace bottom internal temperature temperature Furnace heating temperature temperature Furnace pressure Pa Exhaust gas CO content Content per million Energy consumption kilowatt

[0073] Step 2: Preprocess the relevant parameter data, delete the outliers recorded by the sensors, and use cGAN to generate variables with a measurement frequency lower than a preset threshold, thereby filling the data gaps and improving data integrity.

[0074] Step 2.1: Based on the sensor measurement range and the experience of process experts, threshold values ​​for each variable are set, and outliers exceeding the threshold are directly eliminated to ensure the accuracy and reliability of subsequent modeling data;

[0075] Step 2.2: Perform data partitioning, use complete high-frequency measurement variables (such as oxygen flow rate, furnace pressure, etc.) as conditional inputs of the model, and low-frequency measurement variables (such as tin purity, etc.) as target generation variables, and use the cGAN model to generate missing values. The high-frequency measurement variables are input as conditions into the generator and discriminator to provide additional constraint information, thereby improving data integrity and consistency.

[0076] Step 3: Use the improved LSTM model to model the tin purity and CO concentration in the tin smelting process, capture the time series change patterns of state variable values ​​and spray gun operating parameter values, and improve the accuracy and reliability of the prediction.

[0077] Step 3.1: Based on existing industrial data from the tin smelting process, core variables reflecting dynamic changes within the furnace (such as oxygen flow rate, fuel coal flow rate, furnace pressure, etc.) are extracted. Through mathematical transformations, new feature columns related to the smelting process are generated. These features not only enrich the data dimension but also provide more meaningful input information for subsequent modeling.

[0078] Specifically, the new feature columns are: oxygen flow rate change, combustion coal flow rate change, and oxygen-coal stoichiometric ratio. The oxygen flow rate change and the combustion coal flow rate change help the model capture the temporal dynamics and nonlinear relationship of the combustion reaction in the furnace. The oxygen-coal stoichiometric ratio reflects the supply and demand relationship between oxygen and fuel. The specific calculation formula is:

[0079]

[0080] Where, Indicates the change in oxygen flow rate, represents the oxygen flow rate at the current time step, represents the oxygen flow rate in the previous time step;

[0081]

[0082] Where, Indicates the change in combustion coal flow rate, represents the burning coal flow rate at the current time step, represents the combustion coal flow rate in the previous time step;

[0083]

[0084] Where, represents the stoichiometric ratio of oxygen to coal, Indicates the oxygen flow rate, Indicates oxygen purity, represents the density of oxygen, represents the combustion coal flow rate, Indicates the purity of the burning coal.

[0085] Step 3.2: Embed metallurgical physical constraints into the original LSTM model's loss function. Using constraint penalties (such as deviations from the oxygen-coal stoichiometric ratio), the LSTM model is guided to learn predictions that align with actual process rules, improving the scientific nature and interpretability of the predictions.

[0086] Specifically, metallurgical knowledge constraints are added to the loss function of the LSTM model, and the constraint formula is:

[0087]

[0088] in, is the total loss value, is the prediction loss value of the LSTM model, is a constant, The function needs to be combined with metallurgical knowledge. The specific formula is as follows:

[0089]

[0090] in, is a constant, is the difference between the predicted value and the theoretical value. The theoretical value of CO is calculated as follows:

[0091]

[0092] Where, represents the combustion coal flow rate, is a constant related to the calorific value and combustion efficiency of fuel coal, is the oxygen flow rate, is the theoretical stoichiometric ratio for burning coal.

[0093] Step 3.3: Use the improved LSTM model to predict indicator variables in the tin smelting process (such as tin purity and CO concentration). The gating mechanism captures complex time dependencies and accurately models the dynamic changes between variables.

[0094] Step 3.4: After training is complete, use an independent test dataset to verify the model's predictive performance, evaluate the prediction accuracy of tin purity and CO concentration, and save the LSTM model parameters to ensure its applicability in industrial scenarios.

[0095] Step 4: Build a reinforcement learning environment and adopt a data-driven state update model. Take the state variable values ​​and spray gun operation parameter values ​​in the tin smelting process as input, use the deep learning model to predict the state value at the next moment, and enhance the reliability of state transfer and the actual performance of the environmental model by integrating metallurgical knowledge constraints.

[0096] Step 4.1: Obtain the state changes of consecutive time steps through time series data, including state variables and action variables, and organize the data into triples:

[0097]

[0098] Where, Represents the state variable at the current moment, Represents the action variable at the current moment, Represents the state variables at the next moment. This data format lays the foundation for the training of the subsequent state update model;

[0099] Step 4.2: Build a state update model of the fully connected neural network, and update the current state and the current action As input, predict the next state ,Use the constructed triple data to train the state update model;

[0100] Step 4.3: Evaluate the performance of the state update model on the validation set, save the state update model parameters that meet the preset indicator requirements, and provide a reliable state update mechanism for subsequent embedded reinforcement learning environment. The state update effect is as follows: Figure 2-Figure 11 The figure shows the comparison between the predicted values ​​and true values ​​of the reinforcement learning state update model for ten different states. The ten states are the total amount of concentration, the accumulated molten pool, the percentage of oxygen content, the middle temperature of the furnace bottom, the external temperature of the furnace bottom, the internal temperature of the furnace bottom, the temperature rise of the furnace, the furnace pressure, the CO content of the exhaust gas, and the total energy consumption. The predicted value represents the state at the next moment predicted by the current state and action, and the true value represents the actual state at the next moment.

[0101] Step 4.4: In the reinforcement learning environment, the trained state is used to update the model parameters. The state of the reinforcement learning environment at the next moment is predicted through the current environment and the currently executed action, guiding the agent to make action selections. This embedding process enhances the environment's ability to describe complex dynamic changes, provides the agent with more realistic state feedback, guides it to optimize action selection, and achieves efficient control of the tin smelting process.

[0102] Step 5: Design a reward function to guide the reinforcement learning agent to improve tin purity and reduce CO emissions based on the captured time series changes. The reward function provides real-time feedback on the agent's decision-making effect, providing effective guidance for the optimization process.

[0103] Step 5.1: Establish optimization goals: improve tin purity and reduce CO emissions. Also set auxiliary goals, including reducing energy consumption and maintaining the stability of key process parameters such as furnace pressure, to achieve multi-objective collaborative optimization and ensure efficient operation of the tin smelting process.

[0104] Step 5.2: Use the saved LSTM model parameters to predict the impact of adjusting the current process parameters on tin purity and CO emissions over the next three minutes. Use the predicted results as the basis for calculating the reward function to accurately reflect the contribution of the optimization strategy to the long-term goal.

[0105] Step 5.3: Design the reward function, the formula is:

[0106] ( )

[0107] Where R is the reward function, 、 、 is the weight coefficient, which is used to balance the importance of multiple objectives. is the tin purity content at the current moment, is the carbon monoxide content at the current moment, is the tin purity content at the next moment, is the carbon monoxide content at the next moment, is the energy consumption value at the next moment.

[0108] Furthermore, when the LSTM model prediction results show that the adjustment of the current operating parameters can effectively increase the tin purity and reduce the CO concentration, the reward function will provide positive incentives; on the contrary, if the operation causes the CO concentration to increase or the tin purity to decrease, the reward function will provide negative feedback.

[0109] Step 6: Introduce the Deep Reinforcement Learning (DRL) algorithm, design the spray gun operation parameters as an action space, and the state of the tin smelting process as a state space. The intelligent agent and the state update model interact and learn, and continuously optimize the control strategy to achieve dynamic optimization goals. This strategy can guide the intelligent agent to generate optimal spray gun operation suggestions under different environmental conditions.

[0110] Step 6.1: Regulate the spray gun operating parameters based on the DDPG algorithm. The operations performed by the spray gun in the furnace (such as oxygen flow rate and coal-carrying air pressure) are defined as the action space. The ten core variables in the furnace (total concentration, melt pool accumulation, oxygen content percentage, furnace bottom middle temperature, furnace bottom external temperature, furnace bottom internal temperature, furnace rise temperature, furnace pressure, exhaust CO analysis, and total energy consumption) are defined as the state space.

[0111] Step 6.2: Use the Actor network in the DDPG algorithm to predict the optimal action based on the current state to adjust the spray gun operating parameters. Use the Critic network to evaluate the Q value of the current strategy to guide Actor optimization.

[0112] Step 6.3: The interaction data between the reinforcement learning agent and the furnace environment is stored in an experience pool. Randomly sample several furnace status and spray gun operating parameter data from this pool for training. This breaks the temporal correlation and enables the reinforcement learning agent to learn operational experience that was not previously available in the continuous furnace period data, thereby improving the training stability and generalization ability of the model.

[0113] Step 6.4: By adding noise, the agent's ability to explore the action space is enhanced to avoid falling into local optimality. At the same time, the factory failure situation is simulated to complete the model training.

[0114] Step 6.5: Utilize the trained model to implement intelligent dynamic optimization control during the tin smelting process. By real-time monitoring of state variables and combining the model optimization results, optimization suggestions are provided to the spray gun operator to dynamically adjust operating parameters, thereby improving tin purity, reducing CO emissions, and maximizing process energy efficiency.

[0115] Step 7: Validate the performance of the reinforcement learning model on the test dataset to evaluate its effectiveness in improving tin purity, reducing CO emissions, and optimizing energy consumption. The experimental results are as follows: Current furnace status: Total concentration: 5.777605 tons, Total molten pool: 5.4335 tons, Oxygen concentration: 1.2%, Furnace bottom center temperature: 594.0 temperature units, Furnace bottom outer temperature: 518.0 temperature units, Furnace bottom inner temperature: 518.0 temperature units, Furnace ramp temperature: 740.0 temperature units, Furnace pressure: 3.490 Pa, Exhaust gas CO content: 2.000 ppm (parts per million), Energy consumption: 1.780 kW. The algorithm recommended the following actions: total material flow rate: 86.5 tons / hour, fuel coal flow rate: 9,960 kg / hour, coal-carrying airflow: 0.106 Nm³ / hour, coal-carrying air pressure: 32.5 kPa, spray gun back pressure: 410 kPa, gun position: 7,031 mm, oxygen flow rate: 17,832 Nm³ / hour, and air flow rate: 17,832 Nm³ / hour. When the agent executed the spray gun operating parameters according to the optimal strategy, tin purity increased by an average of 1.5% and CO emissions decreased by 5%.

[0116] The above describes the specific embodiments of the present invention in detail with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary technicians in this field without departing from the scope of the present invention.

Claims

1. A dynamic optimization control method for a tin smelting process, characterized in that: The method specifically includes: Step 1: Collect relevant parameter data during the tin smelting process, including the state variable values ​​and spray gun operation parameter values ​​at each moment in the smelting furnace; Step 2: Preprocess the relevant parameter data, delete the abnormal values ​​recorded by the sensor, and use cGAN to generate variables with measurement frequency below the preset threshold; Step 3: Use the improved LSTM model to model the tin purity and CO concentration during the tin smelting process, capturing the time series variation patterns of state variable values ​​and spray gun operating parameter values. Step 4: Build a reinforcement learning environment and adopt a data-driven state update model. Take the state variable values ​​and spray gun operation parameter values ​​of the tin smelting process as input, and use the deep learning model to predict the state value at the next moment. Step 5: Design a reward function to guide the reinforcement learning agent based on the captured time series variation patterns. Step 6: Introducing a deep reinforcement learning algorithm, the spray gun operating parameters are designed as an action space, and the state of the tin smelting process is designed as a state space. The intelligent agent and the state update model interact and learn, continuously optimizing the control strategy to achieve the dynamic optimization goal; The Step 3 is specifically as follows: Step 3.1: Based on the existing industrial data of the tin smelting process, the core variables reflecting the dynamic changes in the furnace are extracted, and new feature columns related to the smelting process are generated through mathematical transformation; Step 3.2: Embed the physical constraints of the metallurgical field into the loss function of the original LSTM model. The constraint penalty term guides the LSTM model to learn prediction results that conform to the actual process laws. Step 3.3: Use the improved LSTM model to perform prediction training on the indicator variables in the tin smelting process; Step 3.4: After training is complete, use an independent test dataset to verify the model's prediction performance, evaluate the prediction accuracy of tin purity and CO concentration, and save the LSTM model parameters. The Step 5 is specifically as follows: Step 5.1: Set optimization goals; Step 5.2: Use the saved LSTM model parameters to predict the impact of adjusting the current process parameters on tin purity and CO emissions in a preset time period in the future, and use the prediction results as the basis for calculating the reward function; Step 5.3: Design the reward function, the formula is: ( ) ； Where R is the reward function, 、 、 is the weight coefficient, which is used to balance the importance of multiple objectives. is the tin purity content at the current moment, is the carbon monoxide content at the current moment, is the tin purity content at the next moment, is the carbon monoxide content at the next moment, is the energy consumption value at the next moment; The Step 6 is specifically as follows: Step 6.1: Regulate the spray gun operating parameters based on the DDPG algorithm. Define the operations performed by the spray gun in the furnace as the action space, and define the ten core variables in the furnace as the state space: total concentration, melt pool accumulation, oxygen content percentage, furnace bottom middle temperature, furnace bottom external temperature, furnace bottom internal temperature, furnace rise temperature, furnace pressure, exhaust CO analysis, and total energy consumption. Step 6.2: Use the Actor network in the DDPG algorithm to predict the optimal action based on the current state to adjust the spray gun operating parameters. Use the Critic network to evaluate the Q value of the current strategy to guide Actor optimization. Step 6.3: The interaction data between the reinforcement learning agent and the furnace environment is stored in the experience pool. A number of furnace status and spray gun operating parameter data are randomly sampled from the experience pool for training. This allows the reinforcement learning agent to learn operating experience that was not previously available in the continuous furnace period data. Step 6.4: Add noise to simulate a factory failure and complete model training. Step 6.5: Using the trained model, intelligent dynamic optimization control is implemented during the tin smelting process. The Step 3.2 is specifically as follows: Add metallurgical knowledge constraints to the loss function of the LSTM model. The constraint formula is: ； in, is the total loss value, is the prediction loss value of the LSTM model, is a constant, The function needs to be combined with metallurgical knowledge. The specific formula is as follows: ； in, is a constant, is the difference between the predicted value and the theoretical value. The theoretical value of CO is calculated as follows: ； Where, represents the combustion coal flow rate, is a constant related to the calorific value and combustion efficiency of fuel coal, is the oxygen flow rate, is the theoretical stoichiometric ratio for burning coal.

2. The method for dynamic optimization control of a tin smelting process according to claim 1, characterized in that: The Step 2 is specifically as follows: Step 2.1: Set the threshold values ​​for each variable based on the sensor measurement range and the experience of process experts, and directly eliminate abnormal values ​​that exceed the threshold; Step 2.2: Perform data partitioning, use the complete high-frequency measurement variables as the conditional input of the model, and the low-frequency measurement variables as the target generation variables, and use the cGAN model to generate missing values, wherein the high-frequency measurement variables are input as conditions to the generator and discriminator to provide additional constraint information.

3. The method for dynamic optimization control of a tin smelting process according to claim 1, characterized in that: The Step 4 is specifically as follows: Step 4.1: Obtain the state changes of consecutive time steps through time series data, including state variables and action variables, and organize the data into triples: ； Where, Represents the state variable at the current moment, Represents the action variable at the current moment, Represents the state variable at the next moment; Step 4.2: Build a state update model of the fully connected neural network, and update the current state and the current action As input, predict the next state ,Use the constructed triple data to train the state update model; Step 4.3: Evaluate the performance of the state update model on the validation set and save the state update model parameters that meet the preset indicator requirements; Step 4.4: In the reinforcement learning environment, use the trained state to update the model parameters, predict the environment state of reinforcement learning at the next moment through the current environment and the currently executed action, and guide the intelligent agent to make action selections.

Images