Intelligent fusion reactor optimization system and method integrating interpretable deep learning and multi-source perception control

Abstract

The invention discloses an intelligent fusion reactor optimization system and method integrating interpretable deep learning and multi-source perception control. The system is composed of a fusion reactor, a rotary kiln, an information perception assembly, a data processing unit, an intelligent prediction engine, an adjustment execution system and an industrial-grade visual interaction platform. Technological parameters in the smelting process are collected in real time through a multi-source sensing assembly and input into an intelligent prediction engine after data processing, and the engine dynamically predicts an optimal technological parameter interval based on a convolutional neural network, a bidirectional long-short-term memory network and an attention mechanism and provides interpretable analysis through an SHAP value. And the adjustment execution system carries out closed-loop feedback adjustment according to the prediction result, process parameters are optimized, the metal extraction rate is increased, and energy consumption is reduced. The industrial-grade visual interaction platform supports real-time monitoring, trend prediction display, abnormity early warning and model interpretation graphical output, and is convenient for operators to intervene and optimize.

Description

Technical Field

[0001] The present invention belongs to the technical field of metallurgical industrial waste slag treatment, and in particular relates to an intelligent optimization system and method for a melting reactor that integrates explainable deep learning and multi-source perception control. Background Art

[0002] In the field of metallurgy, especially in the high-temperature smelting and resource extraction of nonferrous metals, key process variables such as melt pool load, gas flow parameters, smelting temperature distribution, furnace pressure changes, and injection disturbance intensity have a significant impact on metal conversion efficiency and system energy consumption. These factors generally have highly nonlinear coupling and dynamic feedback relationships. Without precise control, they can easily lead to imbalanced reaction processes, significantly increased energy consumption, and fluctuating and unstable smelting conditions, which in turn affect nonferrous metal extraction efficiency and overall system performance.

[0003] Currently, the operational control of traditional melting reactors relies primarily on manual experience and fixed settings, lacking the ability to adapt to complex dynamic conditions. When handling multi-component charges, variable load conditions, or non-steady-state inputs, these reactors often suffer from issues such as delayed response, insufficient regulation accuracy, and fluctuating energy efficiency. These issues severely hinder the promotion and application of efficient smelting and high-value resource utilization.

[0004] In view of this, how to provide an intelligent optimization system that integrates real-time perception, deep learning modeling and closed-loop control, realize dynamic perception, precise modeling and adaptive adjustment of core process parameters such as molten pool load and airflow disturbance, establish a stable, efficient and energy-saving smelting control mechanism, meet the technical requirements of modern metallurgical processes for intelligent, green and flexible operation, and promote the transformation and upgrading of non-ferrous metal resource recovery processes to intelligent manufacturing, is an urgent problem that technical personnel in this field need to solve. Summary of the Invention

[0005] The present invention proposes a melting reactor intelligent optimization system and method that integrates explainable deep learning and multi-source perception control to solve the problems existing in the above-mentioned prior art.

[0006] To achieve the above objectives, the present invention provides a melting reactor intelligent optimization system that integrates interpretable deep learning and multi-source perception control, which is characterized by comprising:

[0007] Melting reactor, used for high-temperature smelting of metals and providing a molten pool reaction environment;

[0008] The rotary kiln is connected to the tail gas outlet of the melting reactor and is used to receive the high-temperature tail gas and preheat and partially reduce the slag raw materials, thereby achieving synergy between the raw material cascade pretreatment and energy recovery;

[0009] An information sensing component is used to collect multi-source process parameters during the smelting process in real time, including molten pool load, gas flow, temperature, pressure, and high-temperature image data, and transmit the collected data to the back-end data processing module;

[0010] The data processing module is used to perform outlier elimination, denoising, normalization, and time window slicing on multi-source process parameters to obtain standardized input data;

[0011] The intelligent prediction module integrates a convolutional neural network layer, a bidirectional long short-term memory network layer, and an attention mechanism layer. It is used to train the nonlinear mapping relationship between metal extraction rate and process parameters using historical multivariate smelting data. It also dynamically infers the current operating conditions during runtime and outputs the optimal control range and metal extraction rate prediction results.

[0012] The regulation execution module is in communication with the intelligent prediction module and is used to perform closed-loop feedback regulation based on the deviation between the output of the intelligent prediction engine and the preset target value, and dynamically coordinate and control the process parameters, such as the gas injection rate, the molten pool feeding rhythm and the operating status of the rotary kiln, to achieve adaptive optimization of the metal extraction rate;

[0013] The industrial-grade visual interactive platform communicates with the intelligent prediction module and the adjustment execution module to set parameters and display system operating conditions, prediction results, process trends, and model interpretation information in real time.

[0014] Optionally, the melt reactor comprises:

[0015] The furnace shell is made of high-temperature resistant steel plate and covered with a composite insulation layer. The insulation layer includes a multi-layer structure composed of alumina ceramic fiber felt, mullite insulation bricks and nano-aerogel composite blanket, which is used to withstand the high-temperature smelting reaction environment and effectively isolate heat conduction;

[0016] Auxiliary feeding hopper, located in the upper center area of ​​the furnace body, is used to add auxiliary raw materials, including reducing agent and flux;

[0017] The molten pool area is located at the bottom of the furnace and is used to contain and maintain the high-temperature molten material to complete the reduction, sublimation and phase separation reactions of the metal oxides. The area is connected to the molten pool load sensor and participates in the multi-source sensing feedback loop;

[0018] An intelligent control spray gun, located at the top of the furnace, includes several adjustable nozzles with adjustable spray angles and airflow rates. This nozzle injects gas into the area above the molten pool to create turbulent flow and enhance the gas-liquid reaction process. The spray gun is in communication with the control execution module, and its air inlet channel is equipped with a solenoid valve for closed-loop control of the spray flow rate.

[0019] The discharge port is provided on the lower side wall of the furnace body and is used to discharge the slag or unreacted residue produced after the reaction;

[0020] The tail gas discharge port is located in the top area of ​​the rear of the furnace body and is used to guide the high-temperature tail gas to be discharged. The tail gas discharge port is connected to the air inlet end of the rotary kiln through a high-temperature heat-resistant pipe for energy cascade recovery and coordinated treatment of tail gas;

[0021] The waste heat recovery module is installed between the exhaust gas outlet and the air inlet of the rotary kiln. It integrates a heat exchanger and a heat exchange control device to transfer the heat carried by the exhaust gas to the feed area of ​​the rotary kiln, thereby realizing cascade utilization of energy and improving the overall thermal efficiency of the system.

[0022] Optionally, the rotary kiln comprises:

[0023] Piping system for introducing high-temperature tail gas generated during the smelting process into the rotary kiln;

[0024] The preheating heat exchange zone is used to preheat the slag to be treated by high-temperature tail gas so that the slag reaches a predetermined temperature level before entering the melting reactor;

[0025] The collaborative processing area is used to receive the target parameters and feedback signals output by the intelligent prediction module, and dynamically adjust the operating status of the rotary kiln according to the real-time smelting conditions.

[0026] Optionally, the information perception component includes:

[0027] The molten pool load sensor is located at the bottom of the melting reactor and is used to monitor the changes in the molten pool load in real time;

[0028] The gas flow meter is installed in the air inlet pipe of the intelligent control spray gun to measure the gas injected into the furnace and transmit the monitoring results to the regulation execution module for dynamic control;

[0029] Temperature sensors are placed above the molten pool and in the middle of the furnace wall to continuously collect temperature data in the molten pool area, as well as thermal state changes and reaction intensity in the reaction furnace;

[0030] Pressure sensors are installed in the furnace cover and tail gas outlet of the melting reactor to monitor the pressure inside the furnace or the tail gas back pressure fluctuations, and assist in judging the system operation stability and ventilation safety status;

[0031] The image acquisition device is installed at the observation port on the top of the furnace body and is used to obtain infrared images of the high-temperature area or dynamic pictures of the temperature field for the intelligent prediction module to perform working condition identification, model enhancement training and visualization display.

[0032] Optionally, the intelligent prediction module includes:

[0033] Convolutional neural network unit, used to extract local spatial combination features of multi-source input variables;

[0034] Bidirectional long short-term memory network units for learning temporal dependencies between variables;

[0035] Attention mechanism unit, used to dynamically enhance the model's responsiveness to changes in key parameters;

[0036] The interpretability analysis unit outputs feature contribution explanation results based on the SHAP algorithm.

[0037] Optionally, the adjustment execution module includes:

[0038] A prediction receiving unit, configured to receive the output result of the intelligent prediction module;

[0039] PID controller, used to generate dynamic control signals based on the deviation between the output of the prediction module and the process feedback parameters collected in real time;

[0040] A control instruction generation module is used to convert control signals into process execution instructions;

[0041] A process execution device, used to adjust process parameters according to process execution instructions;

[0042] The extraction rate feedback tuning unit is used to optimize PID parameters or control strategies based on the error between the actual metal extraction rate and the predicted target.

[0043] Optionally, the industrial-grade visual interaction platform includes:

[0044] Multi-source data fusion interface, used to display time-series synchronization diagrams and trend overlay diagrams of process variables;

[0045] Prediction analysis visualization layer, used to dynamically present prediction results and control suggestions;

[0046] Causal interpretation rendering unit, used to generate feature influence ranking maps and sensitivity heat maps;

[0047] An abnormal operating condition warning mechanism is used to trigger an early warning when the system operates abnormally;

[0048] Graphical control interactive panel to support users in parameter setting and strategy management.

[0049] Optionally, the collaborative processing area includes:

[0050] An intelligent screw feeder, installed at the front end of the rotary kiln, is used to adjust the charge feeding rate based on the predicted output results;

[0051] The variable frequency drive discharging motor is installed at the tail of the rotary kiln to dynamically adjust the slag discharging speed and material residence time;

[0052] The electric hydraulic inclination adjustment unit is installed under the rotary kiln support structure and is used to adjust the cylinder inclination in real time.

[0053] The present invention also provides a method for intelligent optimization of a melting reactor that integrates interpretable deep learning and multi-source perception control, comprising the following steps:

[0054] The slag raw materials are preheated by a rotary kiln and transported to a melting reactor;

[0055] The information sensing component collects multi-source process parameters in real time and processes the data to generate standardized input data;

[0056] The processed data is input into the intelligent prediction module, which outputs the metal extraction rate prediction results and the optimal control range;

[0057] Based on the deviation between the predicted results and the preset targets, the process parameters are dynamically adjusted to perform closed-loop optimization control of the smelting conditions.

[0058] Optionally, when the deviation between the predicted result and the set threshold is less than or equal to 5%, the injection airflow rate is increased for adjustment to enhance the gas-liquid phase disturbance inside the molten pool and promote the decomposition of metal oxides and the gas phase migration reaction; when the predicted deviation is greater than 5% and not more than 10%, a combined adjustment strategy of increasing the gas flow rate and reducing the feed rate of the melting reactor is simultaneously implemented to reduce the load intensity of the molten pool, reduce the material diffusion resistance, and enhance the reduction reaction environment; the adjustment amplitude is positively correlated with the predicted deviation, and the control parameters are dynamically adjusted by the PID controller in combination with real-time feedback data to achieve a rapid recovery of the metal extraction rate and closed-loop optimization operation of the smelting process.

[0059] Compared with the prior art, the present invention has the following advantages and technical effects:

[0060] The present invention constructs a smelting intelligent optimization system that integrates an interpretable deep learning model with multi-source working condition perception, thereby achieving dynamic prediction and closed-loop control of key process parameters such as molten pool load, gas flow, temperature, and pressure, significantly improving metal extraction rate and reducing unit energy consumption. The system enhances the transparency of model decision-making based on SHAP value analysis, and improves the accuracy and reliability of control. Combining waste heat recovery with the coordinated operation mechanism of the rotary kiln, it further optimizes energy utilization efficiency and reduces carbon emissions. It has multi-working condition adaptability and high robustness, and is suitable for a variety of non-ferrous metal smelting and extraction scenarios. It has significant economic value, environmental benefits and industrial promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0062] Figure 1 Schematic diagram of the system structure of an embodiment of the present invention;

[0063] Figure 2 is a flow chart of a method according to an embodiment of the present invention;

[0064] Figure 3 This is a diagram of the deep learning module architecture of an embodiment of the present invention;

[0065] In the figure: 1. Melting reactor; 2. Rotary kiln; 3. Information perception component; 4. Data processing module; 5. Intelligent prediction module; 6. Adjustment execution module; 7. Industrial-grade visual interaction platform; 11. Furnace shell; 12. Auxiliary feeding hopper; 13. Molten pool area; 14. Intelligent control spray gun; 15. Discharge port; 16. Exhaust gas outlet; 17. Waste heat recovery module; 21. Pipeline system; 22. Preheating and heat exchange area; 23. Co-processing area; 31. Molten pool load sensor; 32. Gas flow meter; 33. Temperature sensor; 34. Pressure sensor; 35. Image acquisition equipment. DETAILED DESCRIPTION

[0066] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0067] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0068] Example 1

[0069] like Figure 1 As shown, this embodiment provides a melting reactor intelligent optimization system that integrates interpretable deep learning and multi-source perception control, including:

[0070] The melting reactor 1 is used for high-temperature smelting of metals and providing a molten pool reaction environment, wherein the load and gas disturbance in the molten pool need to be precisely controlled to achieve efficient reduction and sublimation reactions and improve the extraction efficiency of metals such as metals;

[0071] The rotary kiln 2 is connected to the tail gas outlet 16 of the melting reactor 1 and is used to receive the high-temperature tail gas and preheat and partially reduce the slag raw materials, thereby achieving cascade pretreatment of the raw materials and energy recovery, improving the thermal efficiency of the system and reducing fuel consumption;

[0072] Information sensing component 3, used to collect multi-source process parameters in the smelting process in real time, including but not limited to molten pool load, gas flow, temperature, pressure and high-temperature image data, and transmit the collected data to the back-end data processing module 4;

[0073] Data processing module 4 is used to perform outlier elimination, denoising, normalization and time window slicing processing on multi-source process parameters to obtain standardized input data;

[0074] Intelligent prediction module 5 is used to integrate the convolutional neural network layer, the bidirectional long short-term memory network layer and the attention mechanism layer, and is used to train the nonlinear mapping relationship between metal extraction rate and process parameters based on historical multivariate smelting data, and dynamically infer the current working conditions during runtime to output the optimal control range and metal extraction rate prediction results;

[0075] The adjustment execution module 6 is in communication with the intelligent prediction module 5 and is used to perform closed-loop feedback adjustment based on the deviation between the output of the intelligent prediction engine and the preset target value, dynamically coordinate and control process parameters, including gas injection rate, molten pool feeding rhythm and rotary kiln operation status, to achieve adaptive optimization of metal extraction rate;

[0076] The industrial-grade visual interactive platform 7 is connected to the intelligent prediction module 5 and the adjustment execution module 6 for parameter setting and real-time display of system operating conditions, prediction results, process trends, and model interpretation information.

[0077] In this embodiment, the melt reactor 1 comprises:

[0078] Furnace shell 11: Made of high-temperature resistant steel plate, covered with a composite insulation layer. The insulation layer includes a multi-layer structure consisting of alumina ceramic fiber felt, mullite insulation bricks, and nano-aerogel composite blanket. It has high temperature tolerance, low thermal conductivity, and excellent thermal shock resistance. It is used to support the high-temperature smelting reaction environment and effectively isolate heat conduction.

[0079] Auxiliary feeding hopper 12: located in the upper center of the furnace body, used to add auxiliary raw materials, including but not limited to reducing agents, fluxes or other regulating materials, to be smelted together with the pre-treated waste slag from the rotary kiln 2 to adjust the reaction atmosphere and chemical composition in the molten pool;

[0080] Molten pool area 13: Located at the bottom of the furnace, it is used to contain and maintain high-temperature molten material and complete the reduction, sublimation, and phase separation reactions of the metal oxides. This area is connected to the molten pool load sensor 31 and participates in the multi-source sensing feedback loop;

[0081] Intelligent control spray gun 14: Located at the top of the furnace, it includes multiple adjustable nozzles with adjustable spray angle and airflow rate. It is used to inject gas into the area above the molten pool to create turbulent disturbances and enhance the gas-liquid phase reaction process. The spray gun is in communication with the control execution module 6, and its air inlet channel is equipped with a solenoid valve to achieve closed-loop precise control of the spray flow rate.

[0082] Discharge port 15: located on the lower side wall of the furnace body, used to discharge the slag or unreacted residue produced after the reaction;

[0083] Tail gas discharge port 16: located in the top area of ​​the rear of the furnace body, used to guide the high-temperature tail gas to be discharged. The tail gas discharge port 16 is connected to the air inlet end of the rotary kiln 2 through a high-temperature heat-resistant pipe to achieve energy cascade recovery and coordinated treatment of tail gas;

[0084] Waste heat recovery module 17: It is arranged between the tail gas outlet 16 and the air inlet end of the rotary kiln 2, and integrates a heat exchanger and a heat exchange control device to transfer the heat carried by the tail gas to the feed area of ​​the rotary kiln, thereby realizing cascade utilization of energy and improving the overall thermal efficiency of the system.

[0085] In this embodiment, the rotary kiln 2 comprises:

[0086] The piping system 21 is connected to the tail gas outlet 16 of the melting reactor 1 and the air inlet of the rotary kiln 2, and is used to introduce the high-temperature tail gas generated during the smelting process into the rotary kiln 2 to achieve cascade recovery and effective reuse of energy;

[0087] Preheating heat exchange zone 22: Located in the front section of the rotary kiln 2, it is used to utilize the introduced high-temperature tail gas to perform heat exchange and preheating on the slag to be processed, so that the slag reaches a predetermined temperature level before entering the melting reactor 1, thereby significantly reducing the afterburning load of the rotary kiln and improving the overall thermal efficiency and energy saving level of the system;

[0088] Collaborative processing area 23: communicates with the adjustment execution module 6, is used to receive the target parameters and feedback signals output by the intelligent prediction module 5, and dynamically adjust the operating state of the rotary kiln 2 according to the real-time smelting conditions, including but not limited to the feed rate, discharge rhythm and cylinder inclination, so as to realize the multi-stage process collaborative control and responsive intelligent adjustment of the rotary kiln 2 and the melting reactor 1.

[0089] In this embodiment, the information perception component 3 includes:

[0090] Molten pool load sensor 31: located at the bottom of the melting reactor 1, used to monitor the changes in the molten pool load in real time;

[0091] Gas flow meter 32: installed in the air inlet pipeline of the intelligent control spray gun 14, used to measure the gas injected into the furnace and transmit the monitoring results to the regulation execution module 6 for dynamic control;

[0092] Temperature sensor 33: arranged above the molten pool and in the middle of the furnace wall, used to continuously collect temperature data of the molten pool area 13, thermal state changes and reaction intensity in the reaction furnace;

[0093] Pressure sensor 34: installed in the furnace cover and exhaust outlet of the melting reactor 1, used to monitor the pressure in the furnace or the exhaust back pressure fluctuation, and assist in judging the system operation stability and ventilation safety status;

[0094] Image acquisition device 35: installed at the observation port on the top of the furnace body, used to obtain infrared images of the high-temperature area or dynamic pictures of the temperature field, which are used by the intelligent prediction module 5 for working condition identification, model enhancement training and visualization display support.

[0095] In this embodiment, data processing first uses the Z-Score statistical method to detect and eliminate outliers in the original multi-source process parameters, and identifies and removes data points that deviate from the normal distribution; secondly, the sliding average filter algorithm is applied to perform time series denoising on the cleaned data to suppress high-frequency disturbances and enhance data stability; then Min-Max is used for normalization to standardize all types of data to a unified numerical range, thereby improving the efficiency and numerical stability of model training; finally, the normalized data is structured based on a fixed-length time window slicing mechanism to generate a multi-dimensional input feature data set as the input format of the deep learning model.

[0096] In this embodiment, the intelligent prediction module 5 is as follows Figure 3 As shown, including:

[0097] Convolutional neural network unit, consisting of two layers of one-dimensional convolutional structure, each layer has a convolution kernel size of 3, a stride of 1, a channel number of 16, and an activation function of ReLU, which is used to extract the local spatial combination features of multiple source input variables such as melt pool load, gas flow, temperature, and pressure;

[0098] A bidirectional long short-term memory network unit with 25 hidden units and a "last" output mode, which learns the temporal dependencies and fluctuation trends between variables through synchronous forward and backward state learning;

[0099] The attention mechanism unit is a single-head attention structure with a key channel and a query channel dimension of 2. It outputs the importance distribution of input features at each moment, which is used to dynamically enhance the model's responsiveness to changes in key parameters.

[0100] The explainability analysis unit, based on the SHAP algorithm, calculates the marginal contribution of input features in a single prediction output through the perturbation method. The output includes indicators such as variable contribution value, influence direction, sensitivity score and cumulative explanation rate. The explanation results are graphically embedded in the industrial-grade visualization interactive platform. The presentation forms include variable importance ranking bar charts, heat distribution maps, sample-level local explanation curves and time window sensitivity trend charts, which are used to assist operators in understanding the basis of model decisions, tracing the causal path of predictions, and formulating process control or intervention recommendations, thereby realizing a closed-loop transformation from deep model prediction to visual decision support.

[0101] In this embodiment, the adjustment execution module 6 includes:

[0102] A prediction receiving unit, configured to receive the target parameter value output by the intelligent prediction module 5, including the metal extraction rate prediction result and its corresponding optimal control range;

[0103] PID controller, used to generate dynamic control signals based on the deviation between the predicted target value and the process feedback parameters collected in real time;

[0104] The control instruction generation module converts the control signal output by the PID controller into parameter adjustment instructions that can be recognized by the process execution device and supports multi-channel synchronous output;

[0105] A process execution device, which is used to adjust multiple process variables according to control instructions, including but not limited to regulating the melt pool feed rate, injection air flow rate, rotary kiln speed, inclination angle and heat energy input intensity;

[0106] The extraction rate feedback tuning unit dynamically optimizes PID parameters or control strategies based on the error between the system's actual metal extraction rate and the predicted target, improving the system's adaptive control capabilities and metal recovery efficiency under variable working conditions.

[0107] In this embodiment, the platform is built on an edge computing terminal or industrial control host and is linked to the intelligent prediction module, the regulation execution module, and the information perception component in a two-way communication manner. It is used to achieve digital perception of the entire metallurgical smelting process, predictive-driven visual display, and operator interactive intervention. The prediction results specifically include:

[0108] A multi-source data fusion interface is used to present process variables such as melt pool load, gas flow, furnace temperature fluctuation, and extraction rate evolution from the sensor network in the form of time series synchronization diagrams, trend overlay diagrams, and abnormal threshold bands, realizing integrated visualization of high-frequency data;

[0109] The prediction and analysis visualization layer dynamically presents the extraction rate prediction value and the optimal control recommendation range output by the deep learning engine in real time through dynamic graphics, trajectory animation, or multi-dimensional coordinate projection, and supports response comparison with historical feedback results;

[0110] The causal interpretation rendering unit generates feature influence ranking diagrams, sensitivity heat maps, and causal pathway diagrams based on SHAP value analysis to assist operators in understanding prediction logic and intervention paths.

[0111] An abnormal operating condition warning mechanism automatically triggers interface highlighting, operation lock suggestions, or policy rollback prompts when system operation deviates from the set threshold or extraction rate prediction deviation continues to increase, enabling feedforward intervention capabilities within a closed-loop prediction system.

[0112] The graphical control interactive panel allows operators to set prediction targets, adjust key parameters, manage strategy start and stop, and switch model interpretations through the interface. It has permission control and instruction tracing functions, and supports on-site industrial operations and remote collaborative deployment.

[0113] In this embodiment, the waste heat recovery module is arranged between the tail gas discharge port of the melting reactor and the air inlet section of the rotary kiln, and a heat exchange path is formed through a heat exchanger, which is used to efficiently transfer the heat of the high-temperature tail gas generated during the smelting process to the rotary kiln for raw material preheating or heat energy supply, thereby realizing the cascade recovery and reuse of the tail gas waste heat, significantly reducing the system unit fuel consumption and improving the overall thermal efficiency.

[0114] In this embodiment, the collaborative processing area includes:

[0115] An intelligent screw feeder, installed at the front end of the rotary kiln, is used to adjust the charge feed rate based on the predicted output results. Its electronically controlled drive unit can receive process adjustment instructions and implement continuous or intermittent speed control;

[0116] The variable frequency drive discharging motor is installed at the tail of the rotary kiln and is used to dynamically adjust the slag discharging speed and material residence time. It is equipped with a closed-loop feedback interface for synchronous monitoring of the discharging status and coordination of working conditions with the main control system;

[0117] The electric hydraulic inclination adjustment unit is arranged under the rotary kiln support structure and is used to adjust the cylinder inclination in real time to achieve precise control of the material gravity flow state and heat exchange efficiency. The adjustment system is communicated with the adjustment execution module and supports dynamic response adjustment based on the metal extraction rate optimization target.

[0118] In this embodiment, the device includes the following execution units for real-time control and linkage response of key process variables in the smelting process:

[0119] A raw material feeding unit, including an electrically controlled screw conveyor or a variable frequency vibration feeding mechanism, is used to dynamically convey slag or industrial waste to the melting reactor according to a target feed rate;

[0120] The spray adjustment unit is composed of a multi-channel programmable spray gun cluster with adjustable spray angle, air duct back pressure control and programmable oxygen flow distribution functions, which are used to optimize the gas-liquid interface disturbance and enhance the reduction reaction;

[0121] Temperature control heating unit, including hot air heat exchange device, used to adjust the heat load intensity of the smelting zone and the thermal field distribution of the furnace;

[0122] The rotary kiln operation drive components, including the inclination adjustment actuator, the drum rotation servo system and the tail discharge valve control mechanism, coordinate with the temperature-time curve to control the retention behavior of the raw materials in the kiln and the heat exchange efficiency;

[0123] The signal response execution interface communicates with the PID controller or advanced regulation logic unit, automatically responds to the set action after receiving the predictive control instruction, and ensures the accurate implementation of the control strategy at the physical layer and the state feedback closed loop.

[0124] like Figure 2 As shown, this embodiment also provides a parameter optimization method, which is applied to the melting reactor intelligent optimization system integrating interpretable deep learning and multi-source perception control, comprising the following steps:

[0125] S1: After the slag raw materials are preheated in the rotary kiln, they are transported to the molten pool area of ​​the melting reactor at a set flow rate. Under the instructions of the PLC main control system, a high-temperature smelting environment is established to achieve the coupling of raw material cascade pretreatment and the main reaction process;

[0126] S2: The information perception component collects multi-source process parameters such as melt pool load, gas flow, temperature, pressure, and images in real time. The data processing module then performs anomaly removal, denoising, normalization, and time window slicing to construct a standardized input data set.

[0127] S3: The processed data is input into the intelligent prediction module that has been trained based on historical data. The CNN-BiLSTM-Attention combined model is used to infer the current metal extraction rate and its corresponding optimal parameter control range. The feature contribution interpretation results are output based on SHAP value analysis.

[0128] S4: The adjustment execution module receives the prediction results and compares them with the preset extraction rate target. If there is a significant deviation, it automatically generates control instructions based on the PID control logic to drive the process execution device to adjust multiple process variables such as feed rate, blowing intensity, and kiln operation status, thereby realizing closed-loop dynamic adjustment of the smelting conditions and adaptive optimization control of the metal extraction rate.

[0129] In this embodiment, when the real-time predicted metal extraction rate is lower than the set threshold, the adjustment execution module performs graded dynamic response adjustment according to the prediction deviation amplitude. The specific strategies include: when the deviation amplitude between the predicted value and the set threshold is less than or equal to 5%, the airflow rate of the blowing system is preferentially increased for adjustment to enhance the gas-liquid phase disturbance inside the molten pool and promote the decomposition of metal oxides and the gas phase migration reaction; when the prediction deviation amplitude is greater than 5% and not more than 10%, a combined adjustment strategy of increasing the gas flow rate and reducing the feed rate of the melting reactor is simultaneously executed to reduce the load intensity of the molten pool, reduce the material diffusion resistance, and enhance the reduction reaction environment; the adjustment amplitude is positively correlated with the prediction deviation amount, and the PID controller dynamically adjusts the control parameters in combination with the real-time feedback data to achieve a rapid recovery of the metal extraction rate and closed-loop optimization operation of the smelting process.

[0130] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An intelligent optimization system for a melting reactor that integrates interpretable deep learning and multi-source perception control, characterized in that: include: A melting reactor (1) is used for high-temperature smelting of metals and providing a molten pool reaction environment; The rotary kiln (2) is connected to the tail gas discharge port (16) of the melting reactor (1) and is used to receive the high-temperature tail gas and preheat and partially reduce the slag raw materials, thereby achieving synergy between the raw material cascade pretreatment and energy recovery; An information sensing component (3) is used to collect multi-source process parameters in the smelting process in real time, the multi-source process parameters including molten pool load, gas flow, temperature, pressure and high-temperature image data, and transmit the collected data to a back-end data processing module (4); A data processing module (4) is used to perform outlier elimination, denoising, normalization and time window slicing processing on multi-source process parameters to obtain standardized input data; An intelligent prediction module (5) integrates a convolutional neural network layer, a bidirectional long short-term memory network layer, and an attention mechanism layer, and is used to train the nonlinear mapping relationship between metal extraction rate and process parameters through historical multivariate smelting data, and dynamically infer the current working conditions during operation to output the optimal control range and metal extraction rate prediction results; The adjustment execution module (6) is in communication with the intelligent prediction module (5) and is used to perform closed-loop feedback adjustment according to the deviation between the output result of the intelligent prediction engine and the preset target value, and dynamically coordinate and control the process parameters, wherein the control process parameters include the gas injection rate, the molten pool feeding rhythm and the rotary kiln operation state, so as to achieve adaptive optimization of the metal extraction rate; An industrial-grade visual interactive platform (7) is connected to the intelligent prediction module (5) and the adjustment execution module (6) for parameter setting and real-time display of system operating conditions, prediction results, process trends, and model interpretation information.

2. The system according to claim 1, wherein: The melt reactor (1) comprises: The furnace shell (11) is made of a high-temperature resistant steel plate and is covered with a composite heat-insulating layer. The heat-insulating layer comprises a multi-layer structure composed of alumina ceramic fiber felt, mullite insulation bricks and nano-aerogel composite blanket, and is used to support the high-temperature smelting reaction environment and effectively isolate heat conduction. An auxiliary feeding hopper (12), provided in the central area of ​​the upper portion of the furnace body, for feeding auxiliary raw materials, including reducing agents and fluxes; The molten pool area (13) is arranged at the bottom of the furnace body and is used to contain and maintain high-temperature molten materials to complete the reduction, sublimation and phase separation reactions of metal oxides. The area is connected to the molten pool load sensor (31) and participates in the multi-source sensing feedback loop; An intelligent control spray gun (14) is arranged on the top of the furnace body and includes a plurality of adjustable nozzles with adjustable spray angle and air flow rate, and is used to inject gas into the area above the molten pool to form turbulent disturbance and strengthen the gas-liquid phase reaction process; the spray gun is connected to the control execution module (6) in communication, and its air inlet channel is equipped with a solenoid valve for realizing closed-loop control of the spray flow rate; A discharge port (15) is provided on the lower side wall of the furnace body and is used to discharge the slag or unreacted residue produced after the reaction; An exhaust gas discharge port (16) is located in the rear top area of ​​the furnace body and is used to guide the high-temperature exhaust gas to be discharged. The exhaust gas discharge port (16) is connected to the air inlet end of the rotary kiln (2) through a high-temperature heat-resistant pipe for energy cascade recovery and coordinated exhaust gas treatment; The waste heat recovery module (17) is arranged between the tail gas discharge port (16) and the air inlet of the rotary kiln (2), and integrates a heat exchanger and a heat exchange control device to transfer the heat carried by the tail gas to the feed area of ​​the rotary kiln, thereby realizing energy cascade utilization and improving the overall thermal efficiency of the system.

3. The system according to claim 1, wherein: The rotary kiln (2) comprises: a piping system (21) for introducing high-temperature tail gas generated during the smelting process into the rotary kiln (2); A preheating heat exchange zone (22) is used to preheat the slag to be treated by using high-temperature tail gas, so that the slag reaches a predetermined temperature level before entering the melting reactor (1); The collaborative processing area (23) is used to receive the target parameters and feedback signals output by the intelligent prediction module (5), and dynamically adjust the operating state of the rotary kiln (2) according to the real-time smelting working conditions.

4. The system according to claim 1, wherein: The information perception component (3) includes: A molten pool load sensor (31) is provided at the bottom of the melting reactor (1) for monitoring the change of the molten pool load in real time; A gas flow meter (32) is provided in the air inlet pipe of the intelligent control spray gun (14) for measuring the gas injected into the furnace and transmitting the monitoring result to the regulation execution module (6) for dynamic control; The temperature sensor (33) is arranged on the upper part of the molten pool and the middle part of the furnace wall, and is used to continuously collect temperature data of the molten pool area (13), thermal state changes in the reaction furnace and reaction intensity; A pressure sensor (34) is provided in the furnace cover and the tail gas outlet of the melting reactor (1) to monitor the pressure in the furnace or the tail gas back pressure fluctuation, and to assist in judging the system operation stability and ventilation safety status; The image acquisition device (35) is installed at the observation port on the top of the furnace body and is used to obtain infrared images of the high-temperature area or dynamic images of the temperature field for the intelligent prediction module (5) to perform working condition identification, model enhancement training and visual display.

5. The system according to claim 1, wherein: The intelligent prediction module (5) comprises: Convolutional neural network unit, used to extract local spatial combination features of multi-source input variables; Bidirectional long short-term memory network units for learning temporal dependencies between variables; Attention mechanism unit, used to dynamically enhance the model's responsiveness to changes in key parameters; The interpretability analysis unit outputs feature contribution explanation results based on the SHAP algorithm.

6. The system according to claim 1, wherein: The adjustment execution module (6) comprises: A prediction receiving unit, configured to receive the output result of the intelligent prediction module; PID controller, used to generate dynamic control signals based on the deviation between the output of the prediction module and the process feedback parameters collected in real time; A control instruction generation module is used to convert control signals into process execution instructions; A process execution device, used to adjust process parameters according to process execution instructions; The extraction rate feedback tuning unit is used to optimize PID parameters or control strategies based on the error between the actual metal extraction rate and the predicted target.

7. The system according to claim 1, wherein: The industrial-grade visual interaction platform (7) includes: Multi-source data fusion interface, used to display time-series synchronization diagrams and trend overlay diagrams of process variables; Prediction analysis visualization layer, used to dynamically present prediction results and control suggestions; Causal interpretation rendering unit, used to generate feature influence ranking maps and sensitivity heat maps; An abnormal operating condition warning mechanism is used to trigger an early warning when the system operates abnormally; Graphical control interactive panel to support users in parameter setting and strategy management.

8. The system according to claim 3, wherein: The collaborative processing area (23) includes: An intelligent screw feeder, installed at the front end of the rotary kiln, is used to adjust the charge feeding rate based on the predicted output results; The variable frequency drive discharging motor is installed at the tail of the rotary kiln to dynamically adjust the slag discharging speed and material residence time; The electric hydraulic inclination adjustment unit is installed under the rotary kiln support structure and is used to adjust the cylinder inclination in real time.

9. An intelligent optimization method for a melting reactor that integrates interpretable deep learning and multi-source perception control, characterized in that: The following steps are involved: The slag raw materials are preheated by a rotary kiln and transported to a melting reactor; The information sensing component collects multi-source process parameters in real time and processes the data to generate standardized input data; The processed data is input into the intelligent prediction module, which outputs the metal extraction rate prediction results and the optimal control range; Based on the deviation between the predicted results and the preset targets, the process parameters are dynamically adjusted to perform closed-loop optimization control of the smelting conditions.

10. The method according to claim 9, characterized in that When the deviation between the predicted result and the set threshold is less than or equal to 5%, the injection airflow rate is increased for adjustment to enhance the gas-liquid phase disturbance inside the molten pool and promote the decomposition of metal oxides and the gas phase migration reaction; when the predicted deviation is greater than 5% and does not exceed 10%, a combined adjustment strategy of increasing the gas flow rate and reducing the feed rate of the melting reactor is implemented simultaneously to reduce the load intensity of the molten pool, reduce the material diffusion resistance, and enhance the reduction reaction environment; the adjustment amplitude is positively correlated with the predicted deviation, and the PID controller dynamically adjusts the control parameters in combination with real-time feedback data to achieve a rapid recovery of the metal extraction rate and closed-loop optimization operation of the smelting process.

Images