AI artificial intelligence wisdom tin smelting control system and method

Abstract

The application relates to the technical field of tin smelting, and discloses an AI artificial intelligence intelligent tin smelting control system and method, the system comprising a data acquisition and preprocessing module, an intelligent decision and control module and a real-time monitoring and early warning module; the method comprises the following steps: collecting process data in a tin smelting production process in real time, and cleaning the collected process data; extracting and selecting features that have an influence on smelting quality and efficiency exceeding a preset threshold value, and performing standardization processing on the data; receiving the process data after standardization processing, training a model, processing the real-time monitored data and outputting adjusted process parameters; real-time display of various key parameters and data in the production process, analysis of the real-time data, detection of abnormal conditions, early warning of faults and automatic triggering of response measures. The application realizes the intelligence and automation of the tin smelting process, and can effectively improve the efficiency, quality and safety of the smelting process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tin smelting, and particularly relates to an AI artificial intelligence intelligent tin smelting control system and method. BACKGROUND

[0002] Tin smelting is a process of converting tin ore and other raw materials into high-purity tin through a series of chemical and physical reactions and mechanical processes. The traditional tin smelting process mainly includes roasting, reduction, refining and electrolysis steps. Although these steps basically solve the problem of tin extraction, there are still some challenges in the production process. First, the grade of tin ore is unstable, and the composition of different batches of raw materials differs greatly, making the smelting process and effect difficult to predict and control. Second, the traditional tin smelting method has high energy consumption and causes serious pollution. A large amount of waste gas, waste residue and waste liquid generated during the smelting process has caused serious pollution to the environment. In addition, the traditional tin smelting relies on the experience and manual control of operators, and the process is unstable, the product quality is uneven, and it is easy to cause operation errors and safety accidents. With the rapid development of artificial intelligence technology, especially deep learning and reinforcement learning, it is possible to introduce AI into the intelligent control of the tin smelting process. AI technology can solve many shortcomings of traditional control systems. By introducing reinforcement learning algorithms and using AI to analyze and mine a large amount of historical data and real-time data, the control strategy can be continuously optimized through machine learning models, the state changes in the smelting process can be sensed in real time, and the control parameters can be dynamically adjusted, thereby improving the smelting efficiency and stability and achieving the best production level. Therefore, the intelligent tin smelting control system based on AI can break through the limitations of traditional control systems and significantly improve the intelligent level of the smelting process, which has broad application prospects and great significance. At present, the adjustment of parameters in the production process mainly depends on experienced operators, and the subjective judgment of people inevitably brings volatility, which makes it difficult to guarantee the consistency and stability of the product. The traditional smelting process has high energy consumption and large emissions, which has caused serious pollution to the environment, and energy-saving and pollution control technologies have not been effectively used.

[0003] Prior art one, Chinese patent, application number: 20221049039.3 discloses an intelligent crude tin smelting system and method, by monitoring and data collection statistics in different aspects of the initial, middle and late stages of crude tin smelting, and extracting and labeling the collected data, so as to simultaneously calculate the data in different periods to obtain the evaluation value of the corresponding period. Based on the evaluation value, the state of crude tin smelting in the corresponding period can be analyzed and evaluated, and different ways of early warning and prompting can be realized. Comprehensive monitoring and early warning analysis of crude tin smelting can be realized, so that crude tin smelting can be intelligently and efficiently operated, reducing manual intervention for inspection and adjustment, thereby achieving the purpose of improving the effect of crude tin smelting. Although it solves the technical problem that the existing scheme cannot intelligently and efficiently monitor and prompt crude tin smelting, while taking into account the reasonable use of data processing resources to avoid high load operation. However, it fails to fully utilize the collected data for optimization, and the value of the data is underestimated, resulting in low utilization rate of the data.

[0004] Prior art two, Chinese patent, application number 202111130720.6 discloses an intelligent crude tin smelting system and method. The system includes: an operating management layer for receiving an operating plan and sending operating plan data to an execution manufacturing layer; the execution manufacturing layer includes: an MES management system, a smelting expert system and a spray gun expert system; a control layer for receiving real-time data to control the boiler equipment in the smelting process in real time; a device layer collects real-time data of the boiler equipment in the smelting process and sends it to the control layer and the execution manufacturing layer. The method includes: real-time collection of real-time data of the boiler equipment in the smelting process; calculation and analysis of operating plan data, production management data and real-time data, thereby realizing intelligent analysis, intelligent adjustment, automatic optimization of operating parameters and automatic correction of main operating parameters. Although it solves the problem of manual intervention, independent control and lack of unified control in the original production process. However, it lacks real-time monitoring and fault warning function, and is prone to safety accidents caused by improper parameter setting or equipment failure.

[0005] The prior art three, Chinese patent, application number 202111130677.3 discloses an intelligent crude tin smelting system, comprising: a plan management layer, an execution management layer, a data analysis and evaluation layer, a DCS system management layer, a DCS system control layer and a field device layer; the plan management layer adopts an ERP system, and data interaction with the execution management layer is established through establishment of an enterprise private network; the execution management layer integrates various types of data of the entire crude tin production for use by the data analysis and evaluation layer; the data analysis and evaluation layer analyzes and evaluates the data of each layer; the DCS system management layer writes a production process control algorithm into a DCS controller through an engineer station and a DCS configuration server; data of the DCS system control layer is transmitted to the DCS system management layer for data display through an industrial Ethernet and a switch; and the field device layer transmits collected data to the DCS system control layer. Although manual participation in each link of tin smelting can be reduced, the traditional semi-automatic smelting control condition can be changed, and unified coordinated control of each production management link can be realized. However, it is difficult to dynamically adjust parameters, and it is impossible to quickly respond to the influence of changes in raw material composition and environment, resulting in reduced quality of finished tin products.

[0006] At present, the prior art one, the prior art two and the prior art three have the problems that the collected data cannot be fully utilized for optimization, the value of the data is underestimated, real-time monitoring and fault warning functions are lacking, safety accidents are easily caused by improper parameter setting or equipment failure, and it is difficult to dynamically adjust parameters and quickly respond to the influence of changes in raw material composition and environment. Therefore, the present application provides an AI artificial intelligence intelligent tin smelting control system. SUMMARY

[0007] The main purpose of the present application is to provide an AI artificial intelligence intelligent tin smelting control system and method to solve the problems that the collected data cannot be fully utilized for optimization, real-time monitoring and fault warning functions are lacking, and it is difficult to dynamically adjust parameters in the prior art.

[0008] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0009] An AI artificial intelligence intelligent tin smelting control system, comprising:

[0010] A data acquisition and preprocessing module for real-time acquisition of process data in a tin smelting production process, and cleaning of the collected process data; extracting and selecting features that affect smelting quality and efficiency more than a preset threshold, and standardizing the data; wherein the process data includes monitoring smelting temperature, oxygen concentration in the furnace and tin flow; and detecting the reduction degree of the finished product in the terminal, and real-time monitoring of the temperature change of the slag and flue gas;

[0011] An intelligent decision-making and control module is configured to receive the standardized process data, train the model, and output the adjusted process parameters, including the gun position, compressed gas flow, oxygen concentration, feeding amount of various materials, material ratio, furnace pressure, and temperature, after processing the real-time monitored data.

[0012] A real-time monitoring and early warning module is configured to display the key parameters and data in the production process in real time, analyze the real-time data, detect abnormal conditions, give early warning of faults, and automatically trigger response measures.

[0013] As a further improvement of the present application, the data acquisition and preprocessing module comprises:

[0014] A sensor node submodule is configured to obtain target data of the process data acquisition equipment, match the target data with each node of the smelting equipment and production line, arrange the corresponding process data acquisition equipment according to the matching result, and obtain the process data in the tin smelting process through the process data acquisition equipment;

[0015] A data processing submodule is configured to obtain the original process data collected by the sensor node submodule, clean the collected original process data, perform data filling if there is an intermittent phenomenon, extract and select target features, and perform standardized processing on the collected process data;

[0016] A data caching submodule is configured to set a data cache area between the sensor node submodule and the data processing submodule, and is responsible for temporarily storing and buffering the process data when the data processing submodule network is congested.

[0017] As a further improvement of the present application, the data processing submodule comprises:

[0018] An original process data processing unit is configured to perform standardized processing on the collected original process data, ensure that the mean value of the data is 0 and the variance is 1, and calculate the covariance matrix based on the standardized data;

[0019] A characteristic vector solving unit is configured to solve the process eigenvalues and process eigenvectors of the covariance matrix, sort the obtained process eigenvalues in descending order, select the process eigenvectors corresponding to part of the process eigenvalues as principal components, and form a new feature space;

[0020] A feature extraction unit is configured to form a conversion matrix with the selected process eigenvectors, multiply the conversion matrix with the original data matrix, project into the new feature space, and convert the original process data into low-dimensional representation while retaining the main information of the data.

[0021] As a further improvement of the present application, the intelligent decision-making and control module comprises:

[0022] The reinforcement learning algorithm submodule is used to simulate an agent interacting with an environment and obtaining observation results and rewards from the environment; the agent is trained and makes decisions; the agent takes actions such as adjusting the gun position and changing the feed amount at each state;

[0023] The training and strategy submodule is used to use a reward function to obtain positive rewards and negative rewards for correct actions and incorrect actions, respectively; the value of the current state of the agent is evaluated, the relationship between the new action and the sorted optimal action in the current strategy is obtained, and the agent is trained using a deep Q network;

[0024] The process parameter adjustment submodule is used to dynamically adjust the tin smelting process parameters, including the gun position, compressed gas flow, oxygen concentration, feed amount of various materials, material ratio, furnace pressure and temperature, according to the sorted optimal action in the current strategy output by the reinforcement learning model based on the real-time detected production state.

[0025] As a further improvement of the present application, the reinforcement learning algorithm submodule comprises:

[0026] The data establishment unit is used to establish a process parameter table and initialize random numbers, and the parameters in the table include multiple process parameter states such as current temperature, pressure and gas concentration, actions such as adjusting the gun position and changing the feed amount, and gains; an experience pool is established, and the agent collects experience through interaction with the environment during the training process and stores the collected experience in the experience pool;

[0027] The state updating unit is used to update the process parameter state and the gain, and the agent selects the adjustment gun position and the change feed amount action with the largest value in the process parameter table value based on the current process parameter table, executes and observes the new process parameter state and the gain; the process parameter table is updated according to the new process parameter state and the gain;

[0028] The stable training unit is used to stabilize the training process, and a fixed target network is used to calculate the target process parameter table value; the process of selecting actions and calculating the target process parameter table value is separated, the process parameters are optimized using the backpropagation algorithm until the target network converges, and the performance of the agent is evaluated regularly during the training process.

[0029] As a further improvement of the present application, the training and strategy submodule comprises:

[0030] The function construction unit is used to design a composite reward function based on a multi-objective optimization reward function, and the agent considers simultaneous optimization of multiple objectives when optimizing the smelting process;

[0031] The dynamic adjustment unit is used to generate action selection probabilities through the strategy model, and uses an adaptive learning rate to dynamically adjust the learning rate according to the learning performance of the agent.

[0032] A decision feedback unit is configured to introduce a feedback mechanism at each time step to evaluate whether the process parameters under the current strategy are optimal based on the state monitoring data and environmental changes, and to timely find situations deviating from the reward function target; based on the feedback of historical decision performance, self-supervised learning is introduced to enable the agent to continuously extract lessons from past experiences.

[0033] As a further improvement of the present application, the process parameter adjustment submodule comprises:

[0034] A strategy network unit is configured to select an adjustment gun position and a change in the amount of feed action, and output an adjustment gun position and a change in the amount of feed action based on the current temperature, pressure, gas concentration and other process parameter states;

[0035] A value network unit is configured to evaluate the value of the current strategy, and output an estimated value or advantage function based on the current temperature, pressure, gas concentration and other process parameter states, adjustment gun position and change in the amount of feed action, to guide the update direction of the strategy network subunit;

[0036] An optimal strategy unit is configured to set a process probability value between 0 and 1 during the training process, and at each decision moment, to select an action with the highest estimated value by using the complement of the process probability value, and to balance the relationship between exploring new actions and using the optimal action in the current strategy after sorting in the current strategy by randomly selecting an action with the process probability value; through continuous updating, the optimal strategy after sorting in the current strategy is gradually reached.

[0037] As a further improvement of the present application, the real-time monitoring and early warning module comprises:

[0038] A real-time data analysis unit is configured to receive the current output action and the various process parameter states, extract target features, extract key parameters from the target features, and display the key parameters;

[0039] An anomaly detection unit is configured to detect the extracted current output target features and key parameters, and compare them with the optimal process parameters after sorting in the current strategy in the optimal process parameter table after sorting in the current strategy; if the current output target features and key parameters are less than the optimal process parameters after sorting in the current strategy, an early warning signal is sent out;

[0040] An early warning and response unit is configured to automatically trigger a preset emergency response program after receiving the early warning signal, and remind relevant personnel through a visual tool and an alarm sound.

[0041] As a further improvement of the present application, the real-time data analysis unit comprises:

[0042] The target feature selection subunit is configured to perform preprocessing operations such as data cleaning and correction, select the most relevant target features from the original data based on the current strategy in the sorted optimal strategy process parameter table, and form an output feature set;

[0043] The association rule calculation subunit is configured to scan the output feature set, calculate the support of each item, derive a candidate item set, mark the candidate item set as a frequent item set if the support of the candidate item set is greater than a preset threshold, find a frequent item set 1, generate a candidate item set 2 using the frequent item set 1, scan the output feature set again, calculate the support, find a frequent item set 2, repeat the above process until no new frequent item set can be generated, and finally output the association rules between the key parameters and the target features that satisfy the preset threshold.

[0044] The support calculation formula is as follows:

[0045] Support = number of features of the item set / total number of features

[0046] The model training subunit is configured to train the model using the key parameters and the target features that satisfy the preset threshold as the training set of the model, evaluate the performance of the model by cross-validation, and visualize the key parameters after inputting the current output target features.

[0047] To achieve the above object, the present application further provides the following technical solutions.

[0048] An AI artificial intelligence wisdom tin smelting control method, comprising the following steps:

[0049] Real-time acquisition of process data in the tin smelting production process, and cleaning of the acquired process data; extraction and selection of features that affect smelting quality and efficiency more than a preset threshold, and standardization of the data;

[0050] Receiving the standardized process data, training the model, and outputting the adjusted process parameters including gun position, compressed gas flow, oxygen concentration, various material feeding amounts, material ratio, furnace pressure, and temperature after processing the real-time monitored data;

[0051] Real-time display of various key parameters and data in the production process, analysis of real-time data, detection of abnormal conditions, early warning of faults, and automatic triggering of response measures.

[0052] The data acquisition and preprocessing module of the present application collects key process data in real time during smelting (such as smelting temperature, oxygen concentration in the furnace, tin flow, slag and flue gas temperature, etc.), ensuring the timeliness and accuracy of the data; the collected data is cleaned to remove noise and outliers, and important features related to smelting quality and efficiency are extracted, which are converted into standardized form for subsequent processing. Significance: Through cleaning and standardization, it ensures that the data input to the intelligent decision module is accurate and available, thereby improving the reliability of the decision; extracting key features enables the decision system to more accurately understand and analyze changes in the smelting process, helping to better predict and optimize smelting results. The intelligent decision and control module analyzes and trains the standardized process data, uses machine learning or deep learning algorithms to form a prediction model; based on the data obtained through real-time monitoring, it outputs adjusted process parameters (such as gun position, compressed gas flow, oxygen concentration, etc.), achieving automatic control of the smelting process. Significance: Accurate adjustment of process parameters can improve smelting efficiency and tin yield, while reducing the production of substandard products; automated intelligent decision-making reduces human intervention and reduces the risk of human error, ensuring the stability and consistency of production. The real-time monitoring and early warning module monitors key parameters in real time and analyzes them, and when abnormal conditions (such as excessive temperature, abnormal oxygen concentration, etc.) occur, it automatically identifies and issues an early warning; in the event of a fault or abnormal condition, it can automatically trigger appropriate response measures (such as adjusting parameters or stopping production) to prevent accidents. Significance: Real-time monitoring and early warning systems can detect potential hazards in the production process in a timely manner, ensuring the safety of operators and equipment, and preventing accidents; by adjusting and responding to abnormal conditions in a timely manner, the stability of the production line is improved, reducing downtime, thereby improving overall production efficiency and economic benefits. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 The functional module schematic diagram of an embodiment of the AI artificial intelligence intelligent tin smelting control system of the present application;

[0054] Figure 2 The functional module schematic diagram of an embodiment of the AI artificial intelligence intelligent tin smelting control system of the present application;

[0055] Figure 3 The functional module schematic diagram of the data acquisition and preprocessing module of an embodiment of the AI artificial intelligence intelligent tin smelting control system of the present application;

[0056] Figure 4 The functional module schematic diagram of the intelligent decision and control module of an embodiment of the AI artificial intelligence intelligent tin smelting control system of the present application;

[0057] Figure 5A functional module schematic diagram of a real-time monitoring and early warning module of an embodiment of the AI artificial intelligence smart tin smelting control system;

[0058] Figure 6 A step flow schematic diagram of an embodiment of the AI artificial intelligence smart tin smelting control method;

[0059] Figure 7 A structural schematic diagram of an embodiment of the electronic device;

[0060] Figure 8 A structural schematic diagram of an embodiment of the storage medium. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0062] As shown in the drawings, Figure 1 The present embodiment provides an embodiment of the AI artificial intelligence smart tin smelting control system, which specifically comprises, in sequence, a data acquisition and preprocessing module 1, an intelligent decision and control module 2, and a real-time monitoring and early warning module 3;

[0063] The data acquisition and preprocessing module 1 is used to collect process data in real time during the tin smelting production process, and to clean the collected process data; to extract and select features that affect smelting quality and efficiency beyond a preset threshold, and to standardize the data; wherein the process data includes monitoring smelting temperature, oxygen concentration in the furnace, and tin flow; and detecting the reduction degree of the finished product in the terminal, and real-time monitoring the temperature changes of the slag and flue gas, etc.; the intelligent decision and control module 2 is used to receive the standardized process data, train the model, and output the adjusted process parameters after processing the real-time monitored data, including gun position, compressed gas flow, oxygen concentration, various material feeding amount, material ratio, furnace pressure and temperature, etc.; the real-time monitoring and early warning module 3 is used to real-time display various key parameters and data in the production process, analyze the real-time data, detect abnormal conditions, give early warning of faults, and automatically trigger response measures.

[0064] Preferably, the data acquisition and preprocessing module of the present embodiment collects key process data in real time during the smelting process (such as smelting temperature, oxygen concentration in the furnace, tin flow, slag and flue gas temperature, etc.), ensuring the timeliness and accuracy of the data; the collected data is cleaned to remove noise and outliers, and important features related to smelting quality and efficiency are extracted, which are converted into standardized form for subsequent processing. Significance: Through cleaning and standardization, the data input to the intelligent decision module is accurate and available, thereby improving the reliability of the decision; extracting key features enables the decision system to more accurately understand and analyze changes in the smelting process, which helps better predict and optimize smelting results. The intelligent decision and control module analyzes and trains the standardized process data, uses machine learning or deep learning algorithms to form a prediction model; according to the data obtained by real-time monitoring, the adjusted process parameters (such as gun position, compressed gas flow, oxygen concentration, etc.) are output to realize automatic control of the smelting process. Significance: Accurate adjustment of process parameters can improve smelting efficiency and tin yield, while reducing the production of substandard products; automated intelligent decision reduces human intervention and reduces the risk of human error, ensuring the stability and consistency of production. The real-time monitoring and early warning module monitors various key parameters in real time and analyzes them, and when abnormal conditions (such as high temperature, abnormal oxygen concentration, etc.) occur, it automatically identifies and issues an early warning; in the event of a fault or abnormal condition, it can automatically trigger the appropriate response measures (such as adjusting parameters or stopping production) to prevent accidents. Significance: Real-time monitoring and early warning systems can detect potential hazards in the production process in a timely manner, ensuring the safety of operators and equipment, and preventing accidents; by adjusting and responding to abnormal conditions in a timely manner, the stability of the production line is improved, reducing downtime, thereby improving overall production efficiency and economic benefits (for specific principles, refer to the attached Figure 2 ).

[0065] The data collection and preprocessing module of the embodiment ensures that the system can obtain the latest data in the production process in a timely manner, provides a basis for intelligent decision-making, removes redundant and erroneous or inconsistent data, improves the accuracy of subsequent processing and analysis, filters out features that affect smelting quality and efficiency beyond a preset threshold from massive data, reduces computational load and improves model accuracy, converts data of different dimensions and ranges into a unified format for subsequent processing and analysis; the intelligent decision-making and control module trains the model using historical data and real-time data, optimizes smelting process parameters, adjusts process parameters in real time based on standardized processed data to optimize the smelting process, realizes automatic control through AI algorithms, reduces human error, and improves control accuracy; the real-time monitoring and early warning module displays key parameters and data in the production process in real time through a visual interface for easy monitoring by management personnel. In-depth analysis of real-time data can help identify potential problems and abnormalities, and the system can provide early warnings and automatically trigger appropriate response measures; the system provides high-quality and accurate data support for intelligent decision-making, improving data processing efficiency and accuracy, reducing human intervention, and reducing costs, which helps quickly identify and correct potential problems in the production process, improving production stability; the system realizes intelligent control of the smelting process, improves production efficiency and product quality, adjusts process parameters in real time, quickly responds to changes in the production process, improves production stability, reduces dependence on skilled workers, reduces labor costs, and improves production flexibility; the system improves the transparency and controllability of the production process, reduces production risks, and timely alerts and handles abnormal situations to avoid the expansion of faults, reduce losses, improve production safety and stability, and ensure smooth production.

[0066] In summary, the embodiment realizes the intelligentization and automation of the tin smelting process, which can effectively improve the efficiency, quality and safety of the smelting process. Through real-time data monitoring, intelligent decision-making and rapid response, the system greatly reduces manual intervention, improves the production environment, optimizes resource utilization, reduces costs, and promotes the transformation of the tin smelting industry to intelligent manufacturing. This integrated control scheme not only applies to tin smelting, but also provides an example for intelligent production management in other metallurgical and related industries, and has high promotional value.

[0067] Further, as shown in Figure 3 , the data collection and preprocessing module 1 specifically includes:

[0068] A sensor node sub-module for obtaining target data of process data collection equipment, matching the target data with each node of the smelting equipment and production line, arranging corresponding process data collection equipment according to the matching result, and obtaining process data in the tin smelting process through the process data collection equipment;

[0069] The process data acquisition device includes an oxygen content sensor, a temperature sensor, a flow meter, a component analyzer, an infrared temperature measuring instrument and the like; the oxygen content sensor is installed in the flue of the smelting furnace and is responsible for monitoring the oxygen concentration in the furnace; the temperature sensor is arranged in the smelting furnace and is responsible for real-time monitoring of the smelting temperature in the furnace; the flow meter is installed at the conveying inlet and is responsible for monitoring the tin flow; the component analyzer is arranged at the smelting terminal and is responsible for detecting the reduction degree of the finished product; the infrared temperature measuring instrument is installed outside the smelting furnace and is responsible for real-time monitoring and collecting the temperature change of the slag and flue gas;

[0070] The data processing submodule is used for acquiring the original process data collected by the sensor node submodule, cleaning the collected original process data, and if there is an intermittent phenomenon, data filling is required; target features are extracted and selected, and the collected process data is standardized.

[0071] The data cache submodule is used for setting a data cache area between the sensor node submodule and the data processing submodule, and is responsible for temporarily storing and buffering the process data when the data processing submodule network is congested.

[0072] Preferably, the sensor node sub-module of the present embodiment acquires key process data in the tin smelting process in real time by deploying oxygen content sensors, temperature sensors, flow meters, component analyzers, infrared thermometers and other equipment; ensures that the collection equipment matches each node of the smelting equipment and production line, so as to correctly collect data related to the corresponding position and time. Significance: Realize real-time monitoring of key parameters in the smelting process, provide important operation and control information; through appropriate sensor selection and reasonable layout, ensure that the collected data has high accuracy and reliability, thereby providing a basis for subsequent data processing and decision-making. The data processing sub-module cleans the original process data collected by the sensor, removes outliers and noise, and ensures data quality; for missing or intermittent process data, interpolation or estimation methods are used for filling, ensuring data continuity; extract and select key features that affect smelting efficiency and quality from the cleaned process data for subsequent analysis and decision-making; standardize the collected process data and convert it to a unified scale data format to facilitate subsequent analysis and model training. Significance: Data cleaning and filling ensure the integrity and reliability of the data, thereby supporting more accurate decision-making; through feature extraction and selection, the data dimension is reduced, the computational complexity is reduced, and the subsequent intelligent decision-making and control are more efficient; standardized processing makes various data comparable, making it easier to analyze and model. The data cache sub-module temporarily stores and buffers process data when the data processing sub-module experiences network congestion or processing delays, reducing data loss or omission caused by instantaneous traffic surges or transmission problems; by setting a data cache area, the stability of the data stream is improved, ensuring that the data processing module can continuously and efficiently acquire data. Significance: The setting of the data cache area can increase the fault tolerance of the system, so that it can run normally under high load conditions and will not cause the entire system to fail due to data loss; in complex smelting monitoring, the cache mechanism can balance the data input and processing process, improve the overall system performance, and ensure real-time performance.

[0073] The sensor node submodule of the embodiment accurately acquires key process parameters in the smelting process through different types of sensors, intelligently matches and lays out corresponding data acquisition equipment according to the specific nodes of the smelting equipment and production line, and ensures the comprehensiveness and pertinence of data acquisition; the data processing submodule removes noise and outliers in the original data, reasonably fills in missing data, ensures the integrity and consistency of the data, extracts useful features for process analysis from a large amount of data, reduces the data dimension, improves the subsequent processing efficiency, converts data of different dimensions and different ranges into a unified format, facilitates subsequent data analysis and modeling, temporarily stores and buffers process data when the network is congested or the data processing submodule is overloaded, prevents data loss, realizes smooth transmission of data through the cache mechanism, reduces the impact of network fluctuations or processing delays on data acquisition and processing, improves the accuracy and real-time performance of data acquisition, provides a reliable data foundation for subsequent process analysis and optimization, reduces the complexity and cost of data acquisition through intelligent matching and layout, improves the efficiency and flexibility of the overall system, improves data quality, provides high-quality input for subsequent data analysis and mining, simplifies the complexity of the problem through feature extraction and selection, improves the efficiency and accuracy of data analysis, and data standardization enables data from different sources and different time points to be directly compared and analyzed, enhancing the comparability of the data; ensures the continuity and integrity of data acquisition, even if there is a temporary fault in the network or processing system, the data can be saved completely, improving the stability and reliability of the system, and reducing the risk of production interruption caused by data loss or processing delay.

[0074] In summary, the data acquisition and preprocessing module of the embodiment can realize comprehensive and accurate process data monitoring and processing of the tin smelting process. Not only does it improve the reliability and usability of the data, but it also provides a solid foundation for subsequent intelligent decision-making and control, ensuring the quality, efficiency and safety of the smelting process. The configuration and function design of the embodiment greatly promotes the development of the tin smelting industry towards intelligence and automation, contributing to improving production efficiency, saving resources and reducing environmental impact.

[0075] Further, the data processing submodule specifically includes:

[0076] The original process data processing unit performs standardization processing on the collected original process data to ensure that the mean value of the data is 0 and the variance is 1, and on the basis of the standardized data, calculates the covariance matrix thereof;

[0077] The formula for standardization processing is as follows:

[0078]

[0079] Wherein: X is the original process data, μ is the mean of the original process data, and σ is the standard deviation of the original process data; the processed original process data can better perform subsequent covariance calculation;

[0080] The covariance matrix C is used to measure the linear relationship between multiple variables, and its expression is:

[0081]

[0082] Wherein: C is the covariance matrix, the size is m x m, m is the dimension of the feature, X' is the standardized input data matrix, the size is n x m, n is the sample quantity; n-1 is the degree of freedom adjustment, which ensures that the calculated covariance estimate is unbiased;

[0083] The characteristic vector unit is used to solve the process characteristic value and the process characteristic vector of the covariance matrix, and the solved process characteristic value is sorted in descending order, and the process characteristic vector corresponding to part of the process characteristic value is selected as the principal component to form a new feature space:

[0084] Wherein, the eigenvalue and the eigenvector of the covariance matrix C are solved. The characteristic equation is:

[0085] Cv = λv

[0086] Wherein: v is the eigenvector, and λ is the corresponding eigenvalue.

[0087] The eigenvalue represents the variation degree of the corresponding eigenvector in the covariance matrix; the eigenvalues are sorted in descending order, so that a part of the eigenvectors corresponding to the high eigenvalues can be selected as the principal component to construct a new feature space;

[0088] According to the selected process eigenvalue and the corresponding process eigenvector, the mapping matrix P is calculated using the following formula:

[0089]

[0090]

[0091] Wherein, Indicates the mapping measurement in the new feature space, which calculates the distance between samples f i and f j , f i and f j are two samples in the original process data, P is the mapping matrix, which contains the selected eigenvectors, and T represents the transposition operation; the mapping matrix update rule used according to the given condition is:

[0092]

[0093] Wherein: Pt is the mapping matrix obtained at the tth iteration ij is a certain measure (e.g. distance) between samples f i and f j , a learning rate or adjustment parameter, used to control the magnitude of the update, e ij is a certain parameter involved in the update, representing an error or a certain condition;

[0094] a feature extraction unit, which forms a transformation matrix with the selected process feature vectors, and uses this transformation matrix to multiply with the original data matrix to project into a new feature space, converting the original process data into a low dimensional representation while preserving the main information of the data;

[0095] where the process is represented as:

[0096] Y = X'P

[0097] where Y is the reduced dimension data representation, X' is the standardized original data matrix, and P is the transformation matrix formed by the selected feature vectors.

[0098] Preferably, the original process data processing unit of the present embodiment standardizes the original process data by formula, so that the mean of the processed data is 0 and the variance is 1; it helps to eliminate the dimensional influence between features and ensures that different features are compared on the same scale; on the basis of standardized data, the covariance matrix is calculated to measure the linear relationship between different variables. The covariance matrix can reveal the correlation, importance and variability between input features. Meaning: data standardization eliminates the instability of model training caused by different dimensions, making subsequent analysis more reliable and accurate; the covariance matrix lays the foundation for subsequent feature selection and data dimensionality reduction. Through in-depth analysis of the covariance matrix, the system can identify important features, thereby optimizing decision support. The characteristic vector unit obtains the characteristic value and characteristic vector by solving the characteristic equation of the covariance matrix. In this process, the characteristic value represents the variability of the corresponding feature, and the characteristic vector indicates the most important direction in the data; the characteristic values are arranged in descending order, so that the characteristic vector corresponding to the high characteristic value can be selected as the principal component to identify the main source of variation in the data. Meaning: by selecting the characteristic vector with the largest variance, subsequent analysis can focus on the most important features, reducing the dimensionality of their influence on the results, thereby improving analysis efficiency; using the selected characteristic vector to form a new feature space can effectively transform the original data into a low-dimensional representation while retaining key information and improving subsequent model performance. The feature extraction unit forms a conversion matrix with the selected process feature vectors and multiplies it with the standardized original data matrix to realize data dimensionality reduction and map high-dimensional data to a low-dimensional space; through this projection operation, the main information is retained as much as possible, while noise and irrelevant information are compressed. Through data dimensionality reduction, the computational complexity of the model is reduced, and the training and inference process becomes more efficient; while maintaining the main information, reducing the number of features can improve the generalization ability of the model and reduce the risk of overfitting, thereby improving the prediction effect and decision quality.

[0099] In summary, in the smelting and related industrial process monitoring, the efficient and accurate data processing of the present embodiment can provide support for real-time monitoring and decision-making, and improve production efficiency; through feature selection and dimensionality reduction, enterprises can more effectively allocate resources and make precise business strategies, thereby reducing production costs and improving economic efficiency; the data processing module provides a high-quality data basis for subsequent machine learning or intelligent algorithms, promoting the intelligentization and automation development of the smelting process. The data processing sub-module plays a crucial role in tin smelting data analysis and intelligent decision-making, enhancing the operability and practicality of the system.

[0100] Further, as shown in Figure 4 , the intelligent decision and control module 2 specifically includes:

[0101] Reinforcement learning algorithm submodule, used for simulating an agent interacting with the environment and obtaining observations and rewards from it; the main body of training and decision-making of the agent; the agent takes actions such as adjusting the gun position and changing the feed amount at each state;

[0102] Among them, in the tin smelting control system, the agent (Agent) is the main body responsible for decision-making in the smelting process, used to control and optimize the smelting parameters;

[0103] Environment (Environment), refers to the whole smelting process and its control system, which is perceived and influenced by the agent;

[0104] State (State), describes the current running situation of the smelting process, including temperature, pressure, gas concentration and other process parameters;

[0105] Action (Action), the action the agent can take at each state, such as adjusting the gun position and changing the feed amount, etc.;

[0106] Reward (Reward), the reward signal fed back by the environment to the agent after the agent takes an action, used to measure the goodness of the action;

[0107] Training and strategy submodule, used to use the reward function to obtain positive and negative rewards for correct and incorrect actions respectively; evaluate the value of the current state of the agent, obtain the relationship between the new action and the optimal action sorted in the current strategy, and use the deep Q network to train the agent;

[0108] Process parameter adjustment submodule, used to dynamically adjust the tin smelting process parameters, including gun position, compressed gas flow, oxygen concentration, various material feed amount, material ratio, furnace pressure and temperature, etc., according to the real-time detected production state and the optimal action sorted in the current strategy output by the reinforcement learning model.

[0109] Preferably, the reinforcement learning algorithm submodule of the present embodiment constructs an interaction model of the agent with the smelting environment. The agent explores the state space by trying different actions (such as adjusting the gun position, changing the feed amount), and obtains observation values and rewards according to the results of the attempts, combining learning process and decision evaluation. The agent perceives the current smelting process parameters (such as temperature, pressure, gas concentration) at each state, and selects the best action on this basis, aiming to improve the efficiency and output of the smelting process. Through reinforcement learning, the system can make quick and effective decisions in a changing smelting environment to cope with different production needs and scenarios, improving automation level; the agent has the ability to learn and self-adjust according to feedback in actual production process, thereby optimizing parameter settings and improving overall production efficiency. The training and strategy submodule evaluates the action results of the agent by setting a reward function, positive rewards are used to encourage correct actions, and negative rewards are used to punish incorrect actions, forming an effective feedback mechanism; the value of the current state is evaluated, and decisions are made according to the relationship between new actions and the optimal actions sorted in the current strategy, and the agent is trained using a deep Q network (DQN) to improve its decision-making ability. Through effective reward feedback, the agent can quickly identify actions that are beneficial or detrimental to the smelting process, combining "exploration and utilization" to make decisions more scientific and accurate; training the agent using a deep Q network can continuously improve the quality of decision-making strategies, ensuring optimal performance in the long run and adapting to complex and changing production needs. The process parameter adjustment submodule adjusts the tin smelting process parameters (such as gun position, oxygen concentration, furnace pressure, etc.) in a timely manner based on real-time monitoring of production status; applies the strategies derived from the reinforcement learning model to practice; combines the reinforcement learning decision output with real-time data to achieve intelligent control of the smelting process, optimizing resource use efficiency while ensuring product quality and safety. Through real-time adjustment of process parameters, the production changes can be quickly responded to, ensuring that various parameters in the smelting process remain within a safe and optimal range, improving overall production stability; optimizing process parameters not only improves production efficiency, but also reduces raw material consumption and emissions of waste gas and waste, in line with the goals of green manufacturing and sustainable development.

[0110] The reinforcement learning algorithm submodule of the embodiment simulates the real-time interaction between the agent and the smelting environment, enabling the system to experiment and learn without interrupting actual production. By receiving observations of the environment and corresponding reward signals, the agent is provided with basic data for learning, serving as the core of decision-making. The agent can select and execute appropriate actions based on the current state to achieve long-term optimization goals. The training and policy submodule uses a clear reward function to distinguish between correct and incorrect actions, motivating the agent to value the current state and helping to select the optimal action after sorting in the current policy. A deep neural network is used to approximate the Q value, solving the dimensionality disaster problem of traditional Q learning in high-dimensional state space and enabling more complex and refined decision-making. The process parameter adjustment submodule obtains real-time production state information through sensors and monitoring devices, providing data support for process parameter adjustment. Based on the optimal action after sorting in the current policy output by the reinforcement learning model, the key process parameters in the tin smelting process are automatically adjusted, optimizing raw material utilization and product quality by adjusting material proportioning and feed quantity. The system achieves intelligent and automated production, reduces human intervention, improves production efficiency and stability, and gradually optimizes production strategies through continuous trial and error and learning to adapt to different working conditions and market demands. The efficiency and accuracy of strategy learning are improved, enabling the agent to quickly find the optimal or near-optimal production strategy in the current policy, enhancing the system's adaptability and robustness and maintaining stable performance under different production conditions and external disturbances. The system achieves fine control and optimization of process parameters, improves production efficiency and product quality, reduces resource consumption and waste generation, reduces production costs and environmental impact, improves flexibility and response speed in the production process, and quickly adapts to market changes and customer demands.

[0111] In summary, the functions of the intelligent decision-making and control module of the embodiment achieve comprehensive intelligent control and optimization of the tin smelting process. The comprehensive significance is that the use of reinforcement learning technology and deep learning algorithms significantly improves the automation and intelligence of the tin smelting process, reducing the need for human intervention and labor intensity. The intelligent decision-making module can achieve precise control through data-driven methods, reducing energy consumption and costs and improving production efficiency, thereby enhancing the competitiveness of enterprises. The system can flexibly respond to changes in market demand, adjust production strategies in real time, improve the ability of enterprises to adapt to market changes, and enhance the resilience and sustainable development capacity of the industry. The intelligent decision-making and control system enhances the intelligence and automation of the entire tin smelting process, achieving high efficiency, high quality, and low loss production goals, and promoting the upgrading and development of industry technology.

[0112] Further, the reinforcement learning algorithm submodule specifically includes:

[0113] The data establishing unit is configured to establish a process parameter table and initialize random numbers, and the parameters in the table include a plurality of process parameter states such as a current temperature, a pressure and a gas concentration in a smelting process, an action of adjusting a gun position, an action of changing a feed amount and a gain; and an experience pool is established, in a training process, an agent collects experiences through interaction with an environment, and stores the collected experiences in the experience pool;

[0114] The state updating unit is configured to update the process parameter states and the gain, the agent selects an action of adjusting a gun position and an action of changing a feed amount with the largest value in the process parameter table based on the current process parameter table, executes and observes new process parameter states and a gain, and updates the process parameter table according to the new process parameter states and the gain;

[0115] In the state updating, at each time step t, the agent updates the state S t and the selected action A t , and the state updating expression is as follows:

[0116] S t+1 = S t + ΔS + ∈(S t , A t )

[0117] In the expression, S t represents the current process parameter state, including important variables such as a temperature, a pressure and a gas concentration, A t represents an action selected by the agent in the state S t , such as an action of adjusting a gun position or an action of changing a feed amount, ΔS represents a state change caused by the action of the agent and the feedback of the environment, reflecting an immediate reaction of the process, ∈(S t , A t ) represents a random disturbance introduced, simulating noise and uncertainty in the environment, and enhancing the adaptability of the model to the real situation; and the dynamic updating mechanism ensures that the agent can make adjustments for real-time feedback and continuously improve the decision-making process.

[0118] The action selection process of the agent is based on the Q value of the current state, and the influence of the state change is also considered, and the following formula is used to realize the action selection process:

[0119] A t = argmax a∈A (Q(S t , a) + αΔS(a, S t ))

[0120] In the formula, Q(S t , a) represents an expected return value of the selected action a in the current state, A represents a set of all possible actions, and α is a regulation parameter, used to balance the influence of the Q value estimation and the state change on the action selection, and ΔS(a, St ) represents the expected change in state after performing action a, obtained through historical data fitting or modeling; the selection mechanism makes the agent not only focus on short-term immediate rewards, but also consider the potential impact of actions on future states, making more intelligent decisions;

[0121] An incremental learning mechanism is applied to update the Q values of states:

[0122] Q(S t , A t )←Q(S t , A t )+β(R t +γmax a Q(S t+1 , a)-Q(S t , A t ))

[0123] where R t represents the reward received at time t, γ represents the discount factor, which determines the weight of future rewards in current decision-making, and β represents the incremental learning rate, which can be gradually decayed as the learning process proceeds, so that early experiences have a greater impact and later experiences have a smaller impact, thus smoothing the learning process; the incremental update mechanism provides flexibility for the agent to adjust learning in real time, promoting learning efficiency in complex situations;

[0124] The state information is analyzed in the frequency domain through Fourier transform, focusing on periodic changes:

[0125]

[0126] where F(S t ) represents the frequency domain representation of the state vector, used to capture and analyze periodic characteristics or patterns in the state; the formula for optimizing action selection in the decision-making process is:

[0127] A t =argmax a∈A (Q(F(S t ), a)+αΔS(a, S t ))

[0128] helps the agent identify trends and periodic fluctuations in process parameters, making more targeted adjustments when making decisions;

[0129] Stable training unit, used to stabilize the training process, uses a fixed target network to calculate the target process parameter table value; separate the process of selecting actions and calculating the target process parameter table value, use the backpropagation algorithm to optimize the process parameters until the target network converges, and evaluate the agent's performance regularly during the training process.

[0130] Preferably, the data establishment unit of the present embodiment initializes and maintains a table containing key process parameters (such as temperature, pressure, gas concentration, etc.), and performs random number initialization to ensure the diversification of the initial state; collect the experience obtained by the agent in the process of interacting with the environment, and store these experiences in the experience pool for subsequent learning. The significance achieved: through reasonable random initialization, the system can explore more possible states, thereby avoiding premature convergence into local optimum; the establishment of the experience pool allows the agent to reuse experience in multiple learning tasks, improves learning efficiency, and reduces dependence on real-time data. The state updating unit updates the process parameter state by executing the selected action, and observes the new state and gain; through the state updating formula, the immediate response and external influence in the process are reflected. The significance achieved: the agent can continuously correct its behavior according to environmental feedback, improving the flexibility and adaptability of decision-making; the introduction of random disturbance enables the system to better cope with uncertainties and noise in reality, avoiding overfitting problems. The action selection mechanism considers not only the Q value of the current state, but also the potential impact of state changes when making decisions, in order to optimize action selection; the implementation of this selection. The significance achieved: makes the agent not only pursue short-term rewards when making decisions, but also focuses on potential future state benefits, thereby making better decisions; the optimized action selection process improves learning efficiency, enabling the agent to quickly adapt to complex environments. The incremental learning mechanism continuously optimizes the agent's Q value evaluation through the incremental update formula, gradually improving learning effectiveness using real-time feedback. The significance achieved: the incremental update strategy reduces the instability of learning, enabling the agent to learn smoothly and gradually converge to the optimal strategy; over time, the agent can better adjust its learning rate and focus on important historical experiences. Fourier transform and frequency domain analysis Through Fourier transform, the frequency domain characteristics of state parameters are obtained to help identify periodic changes. The significance achieved: the agent can make more targeted adjustments based on trends and periodic fluctuations in process parameters, thereby improving decision-making accuracy; frequency domain analysis provides the agent with more rich information, promoting more complex decision-making processes. The stable training unit uses a fixed target network to calculate the target process parameter value, making the training process more stable; the steps of selecting actions and calculating target process parameter values are separated, and the process parameters are optimized through the backpropagation algorithm. The significance achieved: by fixing the target network, the shock caused by target value updating is reduced, improving the convergence speed and stability of training; regularly evaluate the performance of the agent to enable it to adapt to environmental changes in a timely manner and continuously optimize its strategy.

[0131] The data establishment subunit of the embodiment provides the system with initial smelting environment data, including key process parameters such as temperature, pressure and gas concentration, and operable actions such as adjusting gun position and changing feed amount and their corresponding gains. By randomly initializing process parameters and action gains, the system diversity and explorability are increased, which helps to cover a wider range of smelting conditions in the early training stage. The experience pool stores the operation and its result of each step as experience in the process of interaction between the agent and the environment. The experience pool can efficiently manage and reuse these experiences for subsequent learning and optimization of the agent. The state updating subunit updates the process parameter table based on the current process parameter table. The agent selects the action such as adjusting gun position and changing feed amount that can bring the largest gain after sorting. After executing the selected action, the new process parameter state and gain are observed and recorded. According to the new state and gain, the process parameter table is updated to reflect the latest state of the current smelting environment. The stable training subunit uses a fixed target network when calculating the target process parameter table value, reduces the fluctuations in the training process, and separates the two processes so that the selection action is not affected by the current network parameter update. The process parameters are optimized using the back propagation algorithm. By continuously adjusting the network parameters, the performance of the agent gradually approaches the optimal solution after sorting in the current strategy. The performance of the agent is evaluated regularly during the training process to ensure the correctness of the training direction. The embodiment provides a rich and explorable initial environment for the agent, which is the basis for intelligent control optimization. Random number initialization promotes the exploration of unknown areas by the agent and avoids falling into the locally optimal solution after sorting in the current strategy too early. The data utilization rate is improved, enabling the agent to learn from past experiences and speed up learning. By storing and reusing experiences, the data correlation and non-stationary distribution problems are alleviated, and the learning stability and efficiency are improved. The embodiment realizes real-time control and optimization of the smelting process by the agent. Through continuous trial and error and learning, the agent can gradually find the optimal smelting strategy after sorting in the current strategy, improving smelting efficiency and product quality. The stability and convergence of the training are improved, enabling the agent to learn effective smelting strategies more reliably. The processes of selecting actions and calculating target values are separated, avoiding overfitting bias in the training process. The performance of the agent is evaluated regularly, providing immediate feedback and adjustment opportunities for the training process and ensuring that the training effect meets expectations.

[0132] In summary, the entire reinforcement learning system of the embodiment can more efficiently construct models, adapt to environments, and optimize decisions, thereby effectively performing intelligent control in complex smelting processes. Not only does it improve the flexibility and robustness of the system, but it also provides strong support and theoretical foundation for practical applications.

[0133] Further, the training and strategy sub-module specifically includes:

[0134] A function construction unit is configured to design a composite reward function based on a multi-objective optimization reward function, so that the agent considers simultaneous optimization of multiple objectives when optimizing the smelting process.

[0135] Components of the reward function:

[0136] Economic reward W economic： Related to smelting efficiency and raw material cost, using the formula:

[0137] W economic = α·(Y output -C input )-β·C fuel

[0138] In the formula, Y output is the product yield, C input is the raw material cost, C fuel is the energy consumption cost, and α and β are weight coefficients.

[0139] Safety reward W safety : Based on the safety performance index in the smelting process, using negative scoring of risk factors such as fire and overheating:

[0140] W safety =-γ·R risk

[0141] In the formula, R risk is the risk assessment index in the current state, and γ is the weight coefficient.

[0142] Environment-friendly reward W environment : To encourage the reduction of waste and emissions, using the formula:

[0143] W environment =-δ·E emissions

[0144] In the formula, E emissions is the emission amount, and δ is the weight coefficient.

[0145] By weighting and summing the rewards of all parts, the final reward function is obtained:

[0146] R(S t , A t , S t+1 ) = W economic + W safety + W environment

[0147] A dynamic adjustment unit is configured to generate action selection probabilities through a policy model, and uses an adaptive learning rate to dynamically adjust the learning rate according to the learning performance of the agent.

[0148] A decision feedback unit is configured to introduce a feedback mechanism at each time step to evaluate whether the process parameters under the current strategy are optimal based on state monitoring data and environmental changes, and to timely discover situations deviating from the reward function target; based on historical decision performance feedback, a self-supervised learning is introduced to enable the agent to continuously extract lessons from past experiences.

[0149] Preferably, the function construction unit of the embodiment combines economic, safety and environmental friendliness factors to form a comprehensive reward function, which can capture multiple objectives in the smelting process and avoid negative effects caused by single objective optimization; the weight coefficients in the reward function can be flexibly adjusted at different operation stages or environmental conditions to adapt to specific production requirements and external environmental changes. The significance achieved is that through multi-objective optimization, the smelting process can improve economic benefits while effectively controlling safety risks and environmental impacts, striving to achieve the optimal overall production effect; the agent no longer focuses on a single indicator in the decision-making process, but makes a more scientific and reasonable decision based on comprehensive consideration, thereby improving the overall system intelligence. The dynamic adjustment unit generates action selection probabilities for specific states through the strategy model, so that the agent can take optimal or suboptimal actions when facing corresponding environmental changes, thereby achieving more accurate decision-making; the learning rate is adaptively adjusted according to the learning performance of the agent to improve the convergence speed of the model and control fluctuations in the training process. The significance achieved is that the adaptive learning rate helps the agent to adapt to new production environments more quickly, avoiding the impact caused by too large or too small learning rate in the initial learning stage; the action selection probabilities generated by the strategy model enable the agent to make flexible adjustments in complex and variable environments, improving the real-time control capability of the production process. The decision feedback unit evaluates the effectiveness of the current strategy execution through real-time state monitoring data at each time step, and timely discovers possible deviations and deficiencies; based on historical decision performance, the agent can learn from past experiences to provide a basis for future decision-making, thereby continuously optimizing its strategy. The significance achieved is that the real-time feedback mechanism enables the agent to adjust decisions in a timely manner according to feedback from environmental changes and production states, ensuring the optimality of process parameters; the self-supervised learning mechanism enables the agent to continuously learn and improve, supporting long-term automated decision-making and continuous optimization, making the smelting process more efficient and precise in automation and intelligence.

[0150] In summary, the training and strategy sub-module of the present embodiment plays a crucial role in the tin smelting control system. Considering the economy, safety and environment, the indicators of the production process are comprehensively improved. The dynamic adjustment capability and feedback mechanism enable the process parameters to be adjusted in real time according to the actual situation, thereby ensuring the continuity and controllability of production; through feedback and self-supervision, the agent has the ability to learn and adapt continuously, continuously improving the accuracy and timeliness of decision-making. Promote the intelligent level and efficiency of intelligent decision and control module in the tin smelting process, help to achieve the goal of safe, environmentally friendly and efficient smelting production.

[0151] Further, the process parameter adjustment sub-module specifically includes:

[0152] The strategy network unit is configured to select an action of adjusting the gun position and changing the feed amount, and output an action of adjusting the gun position and changing the feed amount based on the current temperature, pressure, gas concentration and other process parameter states.

[0153] The value network unit is configured to evaluate the value of the current strategy, and output an estimated value or advantage function based on the current temperature, pressure, gas concentration and other process parameter states, the action of adjusting the gun position and changing the feed amount, to guide the update direction of the strategy network sub-unit.

[0154] The optimal strategy unit is configured to set a process probability value between 0 and 1 during the training process, and at each decision-making moment, select the action with the highest estimated value using the complement of the process probability value, and randomly select an action using the process probability value, to balance the relationship between exploring new actions and using the optimal action in the current strategy after sorting. Through continuous updating, the optimal strategy after sorting in the current strategy is gradually reached.

[0155] Preferably, the strategy network unit of the present embodiment generates targeted decisions based on input real-time process parameters (such as temperature, pressure, gas concentration, etc.), selects the most suitable actions, including adjusting the gun position and changing the feed amount; can select the appropriate control strategy in a multi-element and dynamically changing state environment, improve the reaction speed and the effectiveness of decision-making. The significance achieved: by monitoring the process parameters in real time, the strategy network can quickly respond, reduce the reaction delay, and ensure the accurate control of the smelting process; constantly adjust the strategy according to the feedback and historical data, ensure that the selected action can achieve better production efficiency and product quality under the current state. The value network unit can evaluate the effect of the strategy based on the current state and action, thereby outputting an estimated value or advantage function reflecting the performance of the current strategy; by providing value evaluation feedback, guide the update direction of the strategy network, so that the strategy gradually adjusts towards a higher value. The significance achieved: provides a method to quantify the pros and cons of the strategy, helps to find the shortcomings of the current strategy, and thus carries out targeted optimization; through the evaluation of the strategy performance, promote the benign interaction between the strategy and the environment, help the agent to continuously improve in learning and find a better solution. The optimal strategy unit will set a probability value between 0 and 1 for the strategy, and when taking action, it can balance between exploration (randomly selecting actions) and exploitation (selecting the best action) according to the probability value; by setting the exploration probability to explore unknown actions, while not neglecting the use of the current best known action. The significance achieved: encourages the agent to try new actions during training, which helps to discover potentially overlooked efficient operation strategies, thereby reducing the risk of overfitting; by randomly selecting actions, ensure that the strategy maintains a balance between diversity and stability to adapt to various changes and challenges that may occur in the smelting process.

[0156] The strategy network subunit of the embodiment selects and adjusts the gun position and changes the feeding amount based on the current process parameter state, realizes real-time control and optimization of the process, and through continuous updating and learning, the strategy network subunit can gradually adapt to different process conditions and changes, improve the accuracy and stability of control; the value network subunit evaluates the current strategy and outputs an estimated value or advantage function to guide the updating direction of the strategy network subunit, and through interaction and feedback with the strategy network subunit, the value network subunit can continuously optimize and update its evaluation ability to more accurately reflect the pros and cons of the strategy; the optimal strategy in the current strategy after sorting subunit sets a process probability value during the training process to balance the relationship between exploring new actions and using the optimal action in the current strategy after sorting in the current strategy, and selects the action with the highest estimated value at each decision-making moment by the complement of the process probability value, and selects an action randomly by the process probability value. The optimal strategy in the current strategy after sorting subunit can gradually converge to the optimal strategy in the current strategy after sorting; intelligent control of the process is realized, production efficiency and product quality are improved, manual intervention and dependence are reduced, and production cost and labor cost are reduced; an effective evaluation and guidance mechanism is provided for the strategy network subunit, which promotes continuous optimization and improvement of the strategy, improves the self-adaptation ability and robustness of the whole system, and makes it better cope with various process conditions and changes; an effective balance between exploration and utilization is achieved, avoiding unstable strategies or performance degradation caused by excessive exploration or excessive utilization, and through gradual convergence to the optimal strategy in the current strategy after sorting, the performance and efficiency of the whole system are improved, providing strong support for intelligent control of the process.

[0157] In summary, the process parameter adjustment submodule of the embodiment significantly improves the intelligent level and effect of the decision-making process through the cooperative operation of the three subunits. The strategy, value, and optimal strategy units work together to ensure that the agent can make flexible and accurate decisions in a dynamic environment, improving smelting efficiency and quality; through real-time data feedback and evaluation, continuously optimize and adjust measures to enhance the system's ability to adapt to different process conditions; while optimizing measures to improve yield and efficiency, it also creates conditions for reducing safety risks and environmental impact. The process parameter adjustment submodule provides strong technical support for intelligent decision-making in the smelting process through flexible strategy selection, value evaluation, and exploration mechanisms, helping to achieve efficient, safe, and sustainable production goals.

[0158] Further, as shown in Figure 5 , the real-time monitoring and early warning module specifically includes:

[0159] The real-time data analysis unit is used to receive the current output action and multiple process parameter states, extract target features, extract key parameters from the target features, and display the key parameters;

[0160] an abnormality detection unit configured to detect the extracted current output target features and key parameters, and compare the current output target features and key parameters with the optimal process parameters in the optimal process parameter table in the current strategy, and if the current output target features and key parameters are less than the optimal process parameters in the current strategy, an early warning signal is sent out;

[0161] an early warning and response unit configured to, after receiving the early warning signal, automatically trigger a preset emergency response program, and remind relevant personnel through a visual tool and an alarm sound.

[0162] Preferably, the real-time data analysis unit of the embodiment receives various process parameter states (such as temperature, pressure, gas concentration, etc.) and current output actions in the smelting process in real time, and performs data analysis and analysis; representative target features are extracted from the raw data, and key parameters are further identified, which are the core information in the monitoring process. The significance achieved: by continuously collecting and processing data, the system can quickly reflect the current state of the production process, providing basic information for emergency response; visualizing and displaying key parameters helps operators identify production dynamics in a timely manner, enhancing data readability and operability, and helping decision-making process. The abnormality detection unit compares with the optimal process parameters in the current strategy, and evaluates the current output target features and key parameters in real time; through comparison, it can quickly identify the deviation from the normal state; once the target features or key parameters are lower than the set optimal process parameter values, an early warning signal is automatically sent out. The significance achieved: through real-time monitoring and abnormality detection, deviations can be detected in the early stages of problems, preventing potential accidents or failures and ensuring production safety; having an abnormality detection function helps predictive maintenance, reduces downtime and maintenance costs, and improves the overall reliability of the system. The early warning and response unit can quickly and automatically execute the preset emergency response program after receiving the early warning signal, ensuring efficient handling of any abnormal situation; the early warning information is displayed through a visual tool, and relevant operators are reminded through sound alarms and other methods to attract enough attention. The significance achieved: the automation function of the early warning and response unit can ensure timely measures when abnormalities occur, significantly improving the reaction ability and emergency response ability of on-site personnel; effective early warning and response mechanisms enhance the safety awareness of operators and improve their alertness to potential risks, helping to build a safer working environment.

[0163] The real-time data analysis unit of the embodiment can receive current output actions and various process parameter states in real time, ensure the timeliness and integrity of the data, extract target-related features from a large amount of data, which can reflect key information in the process or production process; further, the key parameters are displayed in an intuitive manner, which facilitates relevant personnel to quickly understand the current process or production state; the abnormality detection unit can monitor the extracted current output target features and key parameters in real time, ensure that abnormal situations are found in time, compare the current output target features and key parameters with the current strategy in the sorted optimal process parameter table in the current strategy to evaluate whether the current process or production state deviates from the sorted optimal state in the current strategy; when it is found that the current output target feature and key parameter value is less than the current strategy in the sorted optimal process parameter value, an early warning signal is automatically sent to remind the relevant personnel; the warning and response unit can receive the early warning signal from the abnormality detection unit, ensure the timely transmission of information, automatically trigger the preset emergency response program to deal with possible abnormal situations, remind the relevant personnel of the abnormal situation through the visual tool and the alarm sound, and guide them to take corresponding measures; the efficiency and accuracy of data processing are improved, which provides a reliable foundation for subsequent abnormality detection and early warning; through the display of key parameters, the transparency and controllability of the system are enhanced, which helps the relevant personnel to find problems in time and take measures; through real-time monitoring and comparative analysis, abnormal situations in the process or production process can be found in time to avoid potential problems, and the issuance of the early warning signal helps the relevant personnel to take measures in time to reduce losses and ensure production safety; through the triggering of the emergency response program, abnormal situations can be quickly dealt with to reduce losses and ensure production safety, and the visual and alarm reminding functions enhance the usability and reliability of the system, which helps the relevant personnel to quickly understand the abnormal situation and respond.

[0164] In summary, the real-time monitoring and early warning module of the embodiment establishes an efficient monitoring and response system through the synergistic effect of the three units, ensuring the safety and stability of the tin smelting production process. The goals achieved include: through real-time data analysis and abnormality detection, the monitoring capability of the production process is effectively improved, making the system intelligent with self-adjustment and real-time early warning features; with an effective early warning mechanism, potential problems are found and handled in time, reducing the risk of unexpected accidents and ensuring the safety and health of employees; through fast decision-making and response, the stagnation time in the production process is reduced, ensuring the efficiency of the overall operation and helping to achieve lean production. The efficient monitoring and early warning mechanism enables the tin smelting process to continuously move towards safety, intelligence, and green, laying a foundation for achieving the goal of sustainable development.

[0165] Further, the real-time data analysis unit specifically includes:

[0166] The target feature selection subunit is configured to perform preprocessing operations such as data cleaning and correction, select the most relevant target features from the original data based on the current strategy in the sorted optimal process parameter table in the current strategy, and form an output feature set;

[0167] The association rule calculation subunit is configured to scan the output feature set, calculate the support of each item, obtain a candidate item set, mark the candidate item set as a frequent item set if the support of the candidate item set is greater than a preset threshold, find a frequent item set 1, generate a candidate item set 2 using the frequent item set 1, and again scan the output feature set, calculate the support, and find a frequent item set 2; the above process is repeated until no new frequent item set can be generated, and finally all association rules between the key parameters and the target features that satisfy the preset threshold of support are output;

[0168] The support calculation formula is as follows:

[0169] Support = number of features of the item set / total number of features

[0170] The model training subunit is configured to train the model using the association rules, use the key parameters and the target features that satisfy the preset threshold of support as a training set of the model, evaluate the performance of the model by a cross-validation method, and after the model is constructed and trained, input the current output target feature to visually output the key parameters.

[0171] Preferably, the target feature selection subunit of the present embodiment cleanses and corrects the original data, ensures data quality, eliminates outliers and errors, and improves the reliability of subsequent analysis; based on the sorted optimal process parameters in the current strategy, the most relevant target features are selected from the standardized and cleaned original data to form an output feature set; the feature is the key indicator that has the greatest impact on the output result in the process parameters. The significance achieved: through data cleaning and feature selection, subsequent analysis and modeling are based on high-quality data, improving the accuracy and reliability of the results; extracting the most relevant features can provide more accurate basis to help subsequent decision-making and model optimization, improving production efficiency and safety. The association rule calculation subunit scans the output feature set, calculates the support of each feature item, and finds the frequent item set that meets the preset threshold; this subunit will repeat this process to generate multiple frequent item sets, and finally output the association rules between the key parameters and target features that meet the support requirement; the application of the calculation formula enables the frequency of different item sets to be quantified, providing a basis for subsequent algorithms. The significance achieved: through the mining of frequent item sets and association rules, the potential relationship between target features and key parameters can be discovered to help understand the mutual influence between process parameters; this analysis provides rich features and parameter support for subsequent model building, enabling the model to have better prediction ability, thereby optimizing the control and management of the smelting process. The model training subunit uses the key parameters and target features that meet the support threshold as the training set, and trains and optimizes the selected model through cross-validation and other evaluation methods; in this way, patterns and rules in the data are extracted; after training, the current output target features are input, and the system can visualize the impact of key parameters, which enables operators to more intuitively understand the data. The significance achieved: through model construction and training, the smelting process can be predicted and optimized to provide guidance for subsequent actual production; visual output helps decision-makers better understand the relationship between key parameters and target features, enhances the initiative in process control, and improves the transparency and control ability of production.

[0172] In summary, the various subunits in the real-time data analysis unit of the present embodiment construct a scientific decision support system through refined data processing and feature analysis, closely combining the needs of the tin smelting process. The goals achieved include: generating scalable models using real-time data to effectively refresh and improve the understanding and control of the production process; through feature mining and association rule establishment, the system has the ability of intelligent decision-making, improving production efficiency and product quality; achieving accurate analysis and rapid response, which helps to optimize the smelting process and enhances the system's ability to respond to abnormal situations, ensuring production safety; the real-time data analysis unit can provide deep intelligent support for the tin smelting process, promoting a more efficient, intelligent, and sustainable production process.

[0173] The target feature selection subunit of the embodiment ensures the accuracy and consistency of the data through cleaning and correction operations, etc., provides a reliable basis for subsequent feature selection, filters out the target features most relevant to the optimal process parameters in the current strategy from the original data based on the optimal process table sorted in the current strategy, constructs an output feature set, removes redundant information, and retains features that have a significant impact on model training; the association rule calculation subunit filters out frequent item sets by calculating the support of each item, thereby finding the association rules between the key parameters and the target features, revealing the potential associations in the data, and providing valuable information for model training; candidate item sets are generated using the frequent item sets, and the support is repeatedly calculated until no new frequent item set can be generated, enabling all association rules that meet the preset threshold to be mined, providing comprehensive information support for the model; the model training subunit trains the model using the training set generated by the association rules, ensures that the model can accurately predict the key parameters, improves the prediction performance of the model, enables it to cope with complex process, evaluates the performance of the model through cross-validation and other methods, ensures the stability and reliability of the model, verifies the applicability of the model, provides strong support for the practical application of the model, outputs the key parameters predicted by the model in a visual manner, facilitates relevant personnel to quickly understand the process state, enhances the ease of use and understandability of the system, and improves the overall performance of the system.

[0174] As shown in Figure 6 The embodiment also provides an AI artificial intelligence smart tin smelting control method, which is applied to the AI artificial intelligence smart tin smelting control system in the above embodiment. The AI artificial intelligence smart tin smelting control method specifically includes the following steps:

[0175] Step S1: Real-time acquisition of process data in the tin smelting production process, and cleaning of the acquired process data; extraction and selection of features that affect smelting quality and efficiency more than a preset threshold, and standardization processing of the data;

[0176] Step S2: Receiving the standardized process data, training the model, processing the real-time monitored data, and outputting the adjusted process parameters, including gun position, compressed gas flow, oxygen concentration, various material feeding amount, material ratio, furnace pressure and temperature, etc.

[0177] Step S3: Real-time display of various key parameters and data in the production process, analysis of real-time data, detection of abnormal conditions, early warning of faults, and automatic triggering of response measures.

[0178] Preferably, step S1 of the present embodiment ensures the timeliness and accuracy of the data, provides timely information support for subsequent analysis and decision-making, removes noise, repetition and outliers in the data, improves data quality, lays a foundation for subsequent feature extraction and model training; filters out features that affect smelting quality and efficiency beyond a preset threshold, reduces redundant information, and improves model training efficiency; converts the data into a unified standard form, eliminates the dimensional differences between different features, and improves the stability and prediction performance of the model; step S2 trains the model using the standardized process data, so that the model can learn the rules and patterns in the data; adjusts the process parameters output by the model according to the real-time monitored data, including gun position, compressed gas flow and oxygen concentration, etc., to optimize the smelting process; step S3 displays the key parameters and data in the production process in a visual manner, which is convenient for operators to understand the production status in real time, analyze real-time data, detect abnormal conditions, and timely discover potential problems in the production process; the detected abnormal conditions are warned, and automatic response measures such as shutdown inspection and adjustment of process parameters are triggered to prevent faults; real-time acquisition and preprocessing of process data provide timely and accurate information support for subsequent analysis and decision-making, reduce the complexity and time cost of model training through feature extraction and selection, and improve the prediction accuracy of the model; data standardization helps to eliminate the dimensional differences between data and improve the generalization ability of the model; through model training, intelligent monitoring and prediction of the smelting process are realized, production efficiency and product quality are improved, process parameters are adjusted in real time, and the smelting process can be dynamically optimized according to the production situation to reduce energy consumption and cost; real-time display of key parameters and data improves the production monitoring ability of operators and ensures the smooth progress of the production process; through data analysis and early warning, abnormal conditions in the production process are discovered and handled in time to reduce the failure rate and downtime, and automatic response measures are triggered to quickly respond to emergencies in the production process and ensure production safety.

[0179] As shown in Figure 7 The electronic device 4 includes a processor 41 and a memory 42 coupled to the processor 41.

[0180] The memory 42 stores program instructions for implementing the AI artificial intelligence intelligent tin smelting control method of any of the above embodiments.

[0181] The processor 41 is configured to execute the program instructions stored in the memory 42 to perform AI artificial intelligence intelligent tin smelting control.

[0182] The processor 41 can also be referred to as a CPU (Central Processing Unit). The processor 41 can be an integrated circuit chip having a processing capability for signals. The processor 41 can also be a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application-Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0183] Further, Figure 8 For a structural schematic diagram of the storage medium of an embodiment of the present application, the storage medium 5 of the embodiment of the present application stores program instructions 51 capable of implementing all the methods described above, wherein the program instructions 51 can be stored in the storage medium in the form of a software product, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a ROM (Read-Only Memory), a RAM (Random Access Memory), a magnetic disk or an optical disk, and various media capable of storing program codes, or a terminal device such as a computer, a server, a mobile phone, a tablet, etc.

[0184] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are merely schematic, for example, the division of units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0185] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit. The above is only an embodiment of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent process conversion using the content of the specification and drawings of the present application, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

[0186] The foregoing detailed description of the application has been presented for purposes of illustration and description. It is not intended to be exhaustive or to limit the application to the precise form disclosed, and obviously many modifications and variations are possible in light of the above teaching. It is intended that the scope of the application be governed by the claims and their equivalents.

Claims

1. An AI artificial intelligence wisdom tin smelting control system, characterized in that, The AI artificial intelligence wisdom tin smelting control system comprises: A data acquisition and preprocessing module is used for collecting process data in a tin smelting production process in real time, and cleaning the collected process data; target features that affect smelting quality and efficiency by more than a preset threshold are extracted and selected, and the data is standardized; wherein the process data includes smelting temperature, oxygen concentration in the furnace and tin flow; and the degree of reduction of the finished product is detected at the terminal, and the temperature change of the slag and flue gas is monitored in real time; An intelligent decision and control module is used for receiving the standardized process data, training the model, processing the real-time monitored data and outputting adjusted process parameters, including gun position, compressed gas flow, oxygen concentration, feeding amount of various materials, material ratio, furnace pressure and temperature; A real-time monitoring and early warning module is used for real-time display of various key parameters and data in the production process, analysis of real-time data, detection of abnormal conditions, early warning of faults and automatic triggering of response measures; The data acquisition and preprocessing module comprises a data processing submodule for obtaining the original process data collected by the sensor node submodule, cleaning the collected original process data, filling in the data if there is an intermittent phenomenon, extracting and selecting target features, and standardizing the collected process data; An original process data processing unit standardizes the collected original process data to ensure that the mean value of the data is 0 and the variance is 1, and calculates the covariance matrix based on the standardized data; A feature vector solving unit solves the process eigenvalues and process eigenvectors of the covariance matrix, and sorts the obtained process eigenvalues in descending order, selects the process eigenvectors corresponding to part of the process eigenvalues as principal components, and constitutes a new feature space; A feature extraction unit uses a conversion matrix composed of selected process eigenvectors to multiply the original data matrix and project it into a new feature space, converting the original process data into low-dimensional representation while retaining the data information; A reinforcement learning algorithm submodule is used to simulate an intelligent agent interacting with the environment and obtaining observation results and rewards from it; the agent is trained and is the main body of decision-making; the agent takes the action of adjusting the gun position and changing the feeding amount at each state; A training and strategy submodule is used to use a reward function to obtain positive rewards and negative rewards for correct actions and incorrect actions respectively; the value of the current state of the agent is evaluated, the relationship between the new action and the sorted optimal action in the current strategy is obtained, and the agent is trained using a deep Q network; A process parameter adjustment submodule is used to dynamically adjust the tin smelting process parameters, including gun position, compressed gas flow, oxygen concentration, feeding amount of various materials, material ratio, furnace pressure and temperature, according to the sorted optimal action in the current strategy output by the reinforcement learning model based on the real-time detected production state; The eigenvalues and eigenvectors of the covariance matrix C are solved, and the characteristic equation is: Cv = lambda v; In the formula, v is the eigenvector, and lambda is the corresponding eigenvalue. According to the selected process characteristic value and the corresponding process characteristic vector, the mapping matrix P is calculated using the following formula: wherein, represents the mapping metric performed in the new feature space, which computes the distance between two samples f i and f j in the original process data, f i and f j are two samples in the original process data, P is the mapping matrix containing the selected feature vectors, and T represents the transpose operation; The mapping matrix update rule used according to the given condition is: where P t is the mapping matrix l ij obtained at the tth iteration i is the distance measure between the sample f j and f ij , a is a learning rate or adjustment parameter that controls the magnitude of the update, ∈ ij is some parameter involved in the update, represents an error or a specific condition; The components of the reward function include: Economic incentive W economic : Related to smelting efficiency and raw material costs, using the formula: W economic = a · (Y output -C input )- β · C fuel In the formula, Y output is the product yield, C input is the raw material cost, C fuel is the energy consumption cost, and α and β are weight coefficients. Safety reward W safety : Based on safety performance indicators in the smelting process, using negative scoring of fire, overheating risk factors: W safety = -γ·R risk In the formula, R risk is the risk assessment index in the current state, and γ is a weight coefficient. Environmentally friendly award W environment for encouraging reduction of waste and emissions, using the formula: W environment = -δ·E emissions In the formula, E emissions is the discharge amount, and δ is a weight coefficient. The final reward function is obtained by weighted summation of the rewards of each part: R(S t ,A t ,S t+1 ) = W economic + W safety + W environment .

2. The AI artificial intelligence tin smelting control system according to claim 1, wherein, The data acquisition and preprocessing module includes: The sensor node submodule is used to obtain the target data of the process data acquisition equipment, match the target data with each node of the smelting equipment and production line, layout the corresponding process data acquisition equipment according to the matching result, and obtain the process data in the tin smelting process through the process data acquisition equipment; The data cache submodule is used to set a data cache area between the sensor node submodule and the data processing submodule, and is responsible for temporarily storing and buffering the process data when the data processing submodule network is congested.

3. The AI artificial intelligence tin smelting control system of claim 1, wherein, The reinforcement learning algorithm submodule includes: The data establishment unit is used to establish a process parameter table and initialize random numbers, and the parameters in the table include current temperature, pressure and gas concentration of the smelting process, adjustment gun position, change in feed amount action and gain; an experience pool is established, and the agent collects experience through interaction with the environment during the training process and stores the collected experience in the experience pool; The state update unit is used to update the process parameter state and gain, and the agent selects the adjustment gun position and change in feed amount action with the largest value in the process parameter table based on the current process parameter table, executes and observes the new process parameter state and gain; and updates the process parameter table according to the new process parameter state and gain; The stable training unit is used to stabilize the training process, and a fixed target network is used to calculate the target process parameter table value; the process of selecting action and calculating the target process parameter table value is separated, and the process parameters are optimized using the back propagation algorithm until the target network converges, and the performance of the agent is evaluated regularly during the training process.

4. The AI artificial intelligence tin smelting control system of claim 3, wherein, The training and strategy submodule includes: The function construction unit is used to design a composite reward function based on the reward function of multi-objective optimization, and the agent considers the simultaneous optimization of multiple objectives when optimizing the smelting process; The dynamic adjustment unit is used to generate action selection probability through the strategy model, and uses an adaptive learning rate to dynamically adjust the learning rate according to the learning performance of the agent; The decision feedback unit is used to introduce a feedback mechanism at each time step, evaluate whether the process parameters under the current strategy are optimal based on state monitoring data and environmental changes, and timely discover situations deviating from the reward function target; based on the feedback of historical decision performance, self-supervised learning is introduced to enable the agent to continuously extract lessons from past experience.

5. The AI artificial intelligence tin smelting control system according to claim 3, wherein, The process parameter adjustment submodule includes: The strategy network unit is used to select the adjustment gun position and change in feed amount action, and outputs an adjustment gun position and change in feed amount action based on the current temperature, pressure and gas concentration of the process parameters; The value network unit is used to evaluate the value of the current strategy, and outputs an estimated value or advantage function based on the current temperature, pressure, gas concentration of the process parameters, adjustment gun position and change in feed amount action, which is used to guide the update direction of the strategy network unit. The optimal strategy unit is configured with a process probability value between 0 and 1 during the training process, and at each decision-making moment, the action with the highest estimated value is selected by the complement of the process probability value, and a random action is selected by the process probability value, to balance the relationship between exploring new actions and using the optimal action in the current strategy after ranking. Through continuous updating, the optimal strategy after ranking in the current strategy is gradually reached.

6. The AI artificial intelligence tin smelting control system of claim 1, wherein, The real-time monitoring and early warning module comprises: The real-time data analysis unit is configured to receive the current output action and various process parameter states, extract target features, and then extract key parameters from the target features and display the key parameters. The abnormality detection unit is configured to detect the extracted current output target features and key parameters, and compare them with the optimal process parameters in the current strategy after ranking in the process parameter table, and if the current output target features and key parameters are less than the optimal process parameters in the current strategy after ranking, an early warning signal is sent. The early warning and response unit is configured to automatically trigger a pre-set emergency response program after receiving the early warning signal, and alert relevant personnel through a visual tool and an alarm sound.

7. The AI artificial intelligence tin smelting control system of claim 6, wherein, The real-time data analysis unit comprises: The target feature selection subunit is configured to perform cleaning and correction preprocessing operations on the data, select the most relevant target features from the original data based on the optimal process parameters in the current strategy after ranking in the process parameter table, and form an output feature set. The association rule calculation subunit is configured to scan the output feature set, calculate the support of each item, obtain a candidate item set, and if the support of the candidate item set is greater than a pre-set threshold, mark it as a frequent item set, and find a frequent item set 1. Candidate item set 2 is generated using the frequent item set 1, and the output feature set is scanned again to calculate the support and find a frequent item set 2. The above process is repeated until no new frequent item set can be generated, and finally the association rules between the key parameters and target features that satisfy the pre-set threshold are output. The support calculation formula is: Support = Number of feature sets containing items / Total number of features The model training subunit is configured to use the association rules, use the key parameters and target features that satisfy the pre-set threshold as the training set of the model to train the model, and evaluate the performance of the model by cross-validation method. After the model is built and trained, the current output target features are input, and the key parameters can be visualized.

8. An AI artificial intelligence wisdom tin smelting control method, characterized in that, The method comprises the following steps: Real-time acquisition of process data in the tin smelting production process, and cleaning of the acquired process data; extraction and selection of features that affect smelting quality and efficiency by more than a pre-set threshold, and standardization of the data; Receiving the standardized process data, training the model, and outputting the adjusted process parameters after processing the real-time monitored data, including gun position, compressed gas flow, oxygen concentration, various material feeding amounts, material ratio, furnace pressure and temperature; Real-time display of key parameters and data in the production process, analysis of real-time data, detection of abnormal conditions, early warning of faults, and automatic triggering of response measures; The subject is trained and makes decisions by simulating an agent interacting with the environment and obtaining observation results and rewards from it; the subject takes actions such as adjusting the gun position and changing the feed amount in each state; A reward function is used to obtain positive and negative rewards for correct and incorrect actions respectively; the value of the current state of the agent is evaluated to obtain the relationship between the new action and the optimal action in the sorted current strategy, and the agent is trained using a deep Q network; According to the real-time detected production state, the optimal action in the current strategy output by the reinforcement learning model is used to dynamically adjust the tin smelting process parameters, including the gun position, the compressed gas flow, the oxygen concentration, the feed amount of various materials, the material ratio, the furnace pressure and the temperature; the eigenvalues and eigenvectors of the covariance matrix C are solved, and the characteristic equation is: Cv=λv; In the formula, v is the eigenvector, and λ is the corresponding eigenvalue; According to the selected process eigenvalue and the corresponding process eigenvector, the mapping matrix P is calculated using the following formula: wherein, represents the mapping metric performed in the new feature space, which computes the distance between two samples f i and f j in the original process data, P is the mapping matrix containing the selected feature vectors, and T denotes the transpose operation. i and f j are two samples in the original process data, P is the mapping matrix containing the selected feature vectors, and T denotes the transpose operation. The mapping matrix update rule used according to the given condition is: In the formula: P t The mapping matrix l is obtained in the t-th iteration. ij It is sample f i and f j The distance metric between them, α is a learning rate or adjustment parameter used to control the magnitude of the updates, ∈ ij It is a parameter involved in the update, representing the error or a specific condition; The components of the reward function include: Economic rewards W economic : related to smelting efficiency and raw material costs, using the formula: W economic = a · (Y output -C input )- β · C fuel In the formula, Y output is the product yield, C input is the raw material cost, C fuel is the energy consumption cost, and α and β are weight coefficients. Safety reward W safety : Based on safety performance indicators in the smelting process, using negative scoring of fire, overheating risk factors: W safety = -γ·R risk In the formula, R risk is the risk assessment index in the current state, and γ is a weight coefficient. Environmentally friendly award W environment for encouraging reduction of waste and emissions, using the formula: W environment = -δ·E emissions In the formula, E emissions is the discharge amount, and δ is a weight coefficient. The rewards of the parts are combined by a weighted sum to give the final reward function: R(S t ,A t ,S t+1 ) = W economic +W safety +W environment .

Images