Deep learning and automatic control chemical reactor and resource recovery method

Abstract

The invention discloses a deep learning and automatic control chemical reactor and a resource recovery method, and relates to the technical field of hydrometallurgy, and the deep learning and automatic control chemical reactor comprises a chemical reactor, a motor, a heating device, a monitoring device, a data center and a PLC control system. Based on a PLC control system, a Transform-LSTM time sequence prediction model is integrated, so that real-time monitoring of parameters such as mixed liquid temperature, reaction chamber pressure, mixed liquid viscosity, conductivity and the like in a wet leaching reaction process can be realized; and the heating power of the heating device, the rotating speed of the motor, the raw material adding amount of the feeding port, the mixed material discharging time of the discharging port and the air displacement of the air exhaust port can be automatically and accurately controlled. The problems that stirring efficiency is low, parameter regulation and control are lagged, and manual experience is relied on are solved, through a predictive control strategy, the real-time monitoring and dynamic regulation level is remarkably improved, the wet leaching process is optimized, energy consumption is reduced, the high-grade metal leaching efficiency is improved, and the method is suitable for various wet metallurgy processes.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrometallurgy, and particularly to a chemical reactor with deep learning and automatic control and a resource recovery method. Background Art

[0002] Through the selective reaction of the leaching agent with the metal in the ore, the hydrometallurgical leaching process can efficiently extract valuable metals from low-grade ores, and is currently widely used in the comprehensive utilization of mineral resources and environmental protection fields. For example, the secondary zinc oxide powder volatilized by the pyro-revolving kiln is subjected to hydrometallurgical extraction, which can remove more than 40 kinds of impurities and extract valuable metals such as zinc, indium, lead, copper, and bismuth.

[0003] The existing hydrometallurgical leaching process is affected by various parameters such as temperature, raw material ratio, and stirring speed, and the hydrometallurgical leaching process has a highly complex and non-linearly changing working condition, resulting in difficulty in accurately controlling the leaching process. In addition, it is difficult to realize real-time monitoring and dynamic adjustment of various parameters and the state change of raw materials in the hydrometallurgical leaching process. The prior art usually adopts the method of timed sampling for monitoring and observation, highly relying on the experience of operators and being unable to quickly respond to process fluctuations.

[0004] In view of this, how to provide a hydrometallurgical leaching reaction device that can achieve adaptive optimization and precise control of multiple parameters during the hydrometallurgical leaching process and real-time monitor and control the reaction process is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a chemical reactor with deep learning and automatic control and a resource recovery method, which are used to realize the predictive control and adaptive optimization of multiple parameters in the hydrometallurgical leaching process to solve the problems existing in the prior art.

[0006] To achieve the above purpose, the present invention provides a chemical reactor with deep learning and automatic control, including:

[0007] A chemical reactor that defines a reaction chamber for hydrometallurgical leaching and is provided with a feed port, a discharge port, and an exhaust port communicating with the reaction chamber;

[0008] A motor, whose output end extends into the reaction chamber and is connected to a stirring paddle, for stirring the mixed liquid in the reaction chamber;

[0009] A heating device, which is arranged on the outer wall of the chemical reactor, for adjusting the temperature of the reaction chamber;

[0010] The monitoring device is used to monitor the raw material addition amount at the feed inlet, the discharge time of the mixture at the discharge outlet, the temperature of the reaction chamber, the pressure of the reaction chamber, and the viscosity of the mixed liquid. The monitoring device is communicatively connected to the data center and uploads the monitored data to the data center. The data center integrates a deep learning module, which performs time series prediction on the temperature of the mixed liquid, the viscosity of the mixed liquid, and the conductivity in the reaction chamber based on the historical reaction data uploaded by the monitoring device, and generates the best raw material ratio-temperature-stirring speed-reaction time combination (the best parameter combination) by analyzing the data characteristics of multiple batches.

[0011] The PLC control system is communicatively connected to the data center and dynamically adjusts the heating power of the heating device, the rotation speed of the motor, the raw material addition amount at the feed inlet, the discharge time of the mixture at the discharge outlet, and the exhaust volume of the exhaust port according to the best parameter combination generated by the data center (2).

[0012] Furthermore, it also includes:

[0013] The raw material cylinder is communicated with the feed inlet through a feed pipeline, and a first electromagnetic valve is arranged on the feed pipeline. A second electromagnetic valve is arranged at the discharge outlet, and a third electromagnetic valve is arranged at the exhaust port. The first electromagnetic valve is used to control the opening and closing of the feed inlet, the second electromagnetic valve is used to control the opening and closing of the discharge outlet, and the third electromagnetic valve is used to control the opening and closing of the exhaust port.

[0014] Flow sensors are arranged at the feed inlet and the discharge outlet. The monitoring device monitors the raw material addition amount at the feed inlet and the discharge time of the mixture at the discharge outlet through the flow sensors. The PLC control system controls the raw material addition amount at the feed inlet and the raw material discharge amount at the discharge outlet by controlling the opening and closing time of the first electromagnetic valve and the second electromagnetic valve. A pressure sensor is arranged in the reaction chamber. The monitoring device monitors the pressure of the reaction chamber through the pressure sensor. The PLC control system controls the exhaust volume of the exhaust port by controlling the opening and closing time of the third electromagnetic valve.

[0015] Furthermore, the data center includes:

[0016] The data storage module is used to store historical reaction data, including the time series data of the mixed liquid temperature, the reaction chamber pressure, the mixed liquid viscosity, the conductivity, and the raw material flow rate.

[0017] Deep learning module, integrating the Transformer-LSTM time series prediction model. The Transformer-LSTM model first uses a long short-term memory neural network to capture the temporal features of the given mixed liquid temperature, mixed liquid viscosity, and mixed liquid conductivity data from the time dimension, strengthening the model's ability to model long-distance dependencies in time series data. Subsequently, the output of the LSTM network is used as the input of the Transformer network, and the multi-head self-attention mechanism is used to capture global dependencies, accurately capturing the global correlation of the mixed liquid temperature, mixed liquid viscosity, and mixed liquid conductivity data in the temporal dimension. Finally, the features strengthened by integrating the LSTM global time series and the Transformer position encoding are used as the input of the fully neural network to predict the mixed liquid temperature, mixed liquid viscosity, and conductivity at the next moment, and instructions are sent to the PLC control system according to the prediction results for the heating power of the heating device, the rotation speed of the motor, the raw material addition amount at the feed inlet, the discharge time of the mixed material at the discharge outlet, and the exhaust volume at the exhaust outlet.

[0018] Adaptive optimization control module, outputting the prediction results and instructions to the PLC control system (3), and the PLC control system (3) dynamically adjusts the heating power of the heating device, the rotation speed of the motor, the raw material addition amount at the feed inlet, the discharge time of the mixed material at the discharge outlet, and the exhaust volume at the exhaust outlet according to the instructions.

[0019] Further, the raw material cylinder includes a solid raw material cylinder for storing solid raw materials and a leaching agent cylinder for storing leaching agents. The output ends of the solid raw material cylinder and the leaching agent cylinder are both connected to the feed inlet, and the flow sensor can respectively monitor the actual flows of the solid raw materials and the leaching agent. The PLC control system controls the opening degree of the first solenoid valve according to the difference between the actual flow and the target flow, and the control logic is:

[0020]

[0021] Where:

[0022] u(t) is the opening degree of the first solenoid valve;

[0023] e1(t) is the difference between the target flow and the actual flow;

[0024] K p , K i , K d are the proportional, integral, and differential gains of the PID controller respectively.

[0025] Further, a conductivity tester is arranged in the reaction chamber. The conductivity tester can monitor the conductivity of the mixed liquid and upload the monitored data to the data center. When it is monitored that the conductivity of the mixed liquid is stable at the set value, the PLC control system opens the discharge outlet through the second solenoid valve.

[0026] Further, the execution logic of the PLC control system includes: when the Transformer-LSTM model predicts that the temperature of the mixed solution will exceed the set range within the next Δt time, dynamically adjust the power of the heating device; when the Transformer-LSTM model predicts that the viscosity of the mixed solution will exceed the set range within the next Δt time, dynamically adjust the power of the motor.

[0027] Further, the PLC control system controls the opening and closing of the third solenoid valve according to the pressure of the reaction chamber monitored by the pressure sensor, and the control logic is: when the pressure of the reaction chamber exceeds the set safety threshold, open the third solenoid valve until the pressure of the reaction chamber is equal to or less than the safety threshold.

[0028] Further, the monitoring device will collect reaction data for 3 to 6 months to construct a training set. The training set includes multi-dimensional time series data such as the temperature of the mixed solution, the viscosity of the mixed solution, and the conductivity, which are used to train the deep learning module Transformer-LSTM to ensure the accuracy of its prediction results.

[0029] Further, the long short-term memory neural network LSTM is a recurrent neural network commonly used to process time series data, which can better adapt to the time series data of the state parameters in the wet leaching process and mine the hidden global time series features in the data. Its main components include:

[0030] Forgot gate F t , which determines the retention of the cell state at the previous moment, and its mathematical expression is:

[0031] F t =θ(M f ·[g t-1 ,y t +d f ),

[0032] where θ is the sigmoid activation function, M f and d f are trainable weights and biases, g t-1 is the information at the previous moment, and y t is the input at the current moment;

[0033] Input gate I t , which is used to control whether to update the cell state, and its mathematical expression is:

[0034] I t =θ(M i ·[g t-1 ,y t +d i ),

[0035] Among them, M i and d i are the trainable weights and biases of the input gate;

[0036] The candidate cell state is used to determine the new information to be stored in the cell state at the current moment, and its mathematical expression is as follows:

[0037]

[0038] Among them, tang is the activation function;

[0039] The new cell state H t , which is used to be updated according to the input at the current moment and the state at the previous moment, and its mathematical expression is as follows:

[0040]

[0041] Among them, ⊙ represents element-wise multiplication;

[0042] The output gate O t , which determines the output amount of the hidden state, and its mathematical expression is:

[0043] O t = θ(M o · [g t-1 , y t + d o )

[0044] Among them, M o and d o are the trainable weights and biases of the output gate;

[0045] The output hidden state g t , which is jointly determined by the activation value of the output gate and the non-linear transformation of the cell state, and its mathematical expression is:

[0046] g t = O t ⊙ tang(H t )

[0047] Furthermore, the Transformer network is a sequence-to-sequence model based on the self-attention mechanism, which can accelerate the process of model training and inference, and more accurately capture the global dependency relationship of the time series data of the state parameters in the hydrometallurgical process. The input of the Transformer is the output of the LSTM. Assume that the dimension of this output is b k , and its calculation formula is:

[0048]

[0049] Among them, Q, K, and V are the query, key, and value matrices respectively, and b k is the dimensional scaling factor;

[0050] Furthermore, the heating device includes:

[0051] A jacket layer that defines a heating chamber and is provided with a heating medium inlet and a heating medium outlet that communicate with the heating chamber;

[0052] A heating device disposed in the heating chamber for heat exchange with the heating medium, and the PLC control system can control the heating power of the heating device.

[0053] Furthermore, a temperature sensor is disposed in the reaction chamber, the temperature sensor can monitor the actual temperature of the reaction chamber, and the PLC control system controls the power of the heating device through a first frequency converter according to the difference between the actual temperature and the target temperature. The control logic is:

[0054]

[0055] Wherein:

[0056] T(t) is the power of the heating device;

[0057] e2(t) is the difference between the target temperature and the actual temperature;

[0058] K p , K i , K d are the proportional, integral, and derivative gains of the PID controller respectively.

[0059] Furthermore, a viscometer is disposed in the reaction chamber, the monitoring device monitors the actual viscosity of the mixed liquid through the viscometer, and the PLC control system controls the rotation speed of the motor through a second frequency converter according to the difference between the actual viscosity and the target viscosity. The control logic is:

[0060]

[0061] Wherein:

[0062] ω(t) is the power of the motor;

[0063] e3(t) is the difference between the target viscosity and the actual viscosity;

[0064] K p , K i , K d are the proportional, integral, and derivative gains of the PID controller respectively.

[0065] The present invention also provides a resource recovery method, which applies the chemical reactor with deep learning and automatic control and includes the following steps:

[0066] S1: The PLC control system controls the raw materials to enter the reaction chamber from the feed port at the target flow rate;

[0067] S2: The monitoring device monitors in real time the raw material addition amount at the feed port, the raw material discharge amount at the discharge port, the temperature of the reaction chamber, the pressure of the reaction chamber, and the viscosity of the mixed solution. The monitoring device uploads the monitored data to the data center;

[0068] S3: The Transformer-LSTM model in the data center receives the monitoring data in real time, predicts the future change trend, and generates adjustment suggestions;

[0069] S4: The PLC control system dynamically adjusts the heating power of the heating device, the rotation speed of the motor, the raw material addition amount at the feed port, the discharge time of the mixed material at the discharge port, and the exhaust volume of the exhaust port according to the adjustment instructions from the data center.

[0070] The present invention discloses the following technical effects:

[0071] Based on the PLC control system and by integrating the Transformer-LSTM time series prediction model, the present invention can realize the real-time monitoring of parameters such as the temperature of the mixed solution, the pressure of the reaction chamber, the viscosity of the mixed solution, and the conductivity during the wet leaching reaction process, and can automatically and accurately control the heating power of the heating device, the rotation speed of the motor, the raw material addition amount at the feed port, the raw material discharge amount at the discharge port, and the exhaust volume of the exhaust port, so as to realize the predictive control and adaptive optimization of the reaction parameters. Compared with the prior art, the present invention solves the problems of low stirring efficiency, lagging parameter regulation, and dependence on manual experience in the traditional hydrometallurgy process. Through the predictive control strategy, the real-time monitoring and dynamic regulation level are significantly improved, the wet leaching reaction process is optimized, the energy consumption is reduced, the leaching efficiency of high-grade metals is improved, and it is applicable to various hydrometallurgy processes.

[0072] Economic benefits

[0073] Improvement in noble metal recovery rate: By optimizing the raw material ratio and reaction conditions in real time through the deep learning module, noble metals can be efficiently leached from low-grade solid waste, and the recovery rate is significantly improved compared with the traditional method.

[0074] Reduction in energy consumption and cost: Based on the dynamic regulation function of the predictive control strategy, the ineffective stirring time and the fluctuation of the heating power are reduced, the energy consumption is greatly reduced, and at the same time, the amount of chemical reagents used is reduced, thus reducing the cost.

[0075] Environmental friendliness: Precise control of the exhaust volume and the reaction end point reduces the emission of harmful gases and secondary pollution. Brief Description of the Drawings

[0076] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0077] Figure 1 It is a schematic structural diagram of the present invention;

[0078] Figure 2 It is a flowchart of the method of the present invention;

[0079] Figure 3 It is a framework diagram of the deep learning module;

[0080] Figure 4 It is a fitting effect diagram of the Transformer-LSTM time series prediction model for the viscosity of the mixed liquid:

[0081] Figure 5 It is a fitting effect diagram of the LSTM single model for the viscosity of the mixed liquid.

[0082] Among them, 1. Chemical reactor; 2. Data center; 3. PLC control system; 4. First frequency converter; 5. Second frequency converter; 6. Solid raw material cylinder; 7. Leaching agent cylinder; 8. First solenoid valve; 9. Second solenoid valve; 10. Third solenoid valve; 11. Motor; 12. Stirring paddle; 13. Feed inlet; 14. Discharge outlet; 15. Monitoring device; 151. Flow sensor; 152. Temperature sensor; 153. Pressure sensor; 154. Viscometer; 155. Conductivity tester; 16. Jacket layer; 161. Heating medium inlet; 162. Heating medium outlet; 17. Heating device; 18. Exhaust port. Detailed Embodiments

[0083] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0084] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0085] The embodiments of the present invention provide a chemical reactor for deep learning and automatic control, including:

[0086] A chemical reactor 1 defines a reaction chamber for wet leaching and is provided with a feed inlet 13, a discharge outlet 14 and an exhaust outlet 18 communicating with the reaction chamber; the feed inlet 13 and the exhaust outlet 18 are generally located at the top of the chemical reactor 1, and the exhaust outlet 18 is located at the bottom of the chemical reactor 1;

[0087] A motor 11, whose output end extends into the reaction chamber and is connected to a stirring paddle 12, is used for stirring the mixed liquid in the reaction chamber;

[0088] A heating device 17 is arranged on the outer wall of the chemical reactor 1 and is used for adjusting the temperature of the reaction chamber;

[0089] A monitoring device 15 is used for monitoring the raw material addition amount at the feed inlet 13, the discharge time of the mixed material at the discharge outlet 14, the temperature of the reaction chamber, the pressure of the reaction chamber and the viscosity of the mixed liquid. The monitoring device 15 is communicatively connected to a data center 2 and uploads the monitored data to the data center 2;

[0090] The data center 2 integrates a deep learning module, performs time series prediction on the temperature of the mixed liquid, the viscosity of the mixed liquid and the conductivity in the reaction chamber based on the historical reaction data uploaded by the monitoring device 15, and generates an optimal raw material ratio-temperature-stirring speed-reaction time combination, i.e., the optimal parameter combination, by analyzing the characteristics of multi-batch data;

[0091] A PLC control system 3 is communicatively connected to the data center 2 and dynamically adjusts the heating power of the heating device 17, the rotation speed of the motor 11, the raw material addition amount at the feed inlet 13, the discharge time of the mixed material at the discharge outlet 14 and the exhaust volume of the exhaust outlet 18 according to the optimal parameter combination generated by the data center 2;

[0092] A PLC control system 3 is communicatively connected to the data center 2 and controls the heating power of the heating device, the rotation speed of the motor 11, the raw material addition amount at the feed inlet 13, the raw material discharge amount at the discharge outlet 14 and the exhaust volume of the exhaust outlet 18 respectively according to the data uploaded by the monitoring device 15.

[0093] In this embodiment, it further includes:

[0094] A raw material cylinder is communicated with the feed inlet 13 through a feed pipeline, and a first solenoid valve 8 is arranged on the feed pipeline; a second solenoid valve 9 is arranged at the discharge outlet 14, and a third solenoid valve 10 is arranged at the exhaust outlet 18; the first solenoid valve 8 is used to control the opening and closing of the feed inlet 13, the second solenoid valve 9 is used to control the opening and closing of the discharge outlet 14, and the third solenoid valve 10 is used to control the opening and closing of the exhaust outlet 18;

[0095] Flow sensors 151 are provided at the feed inlet 13 and the discharge outlet 14. The monitoring device 15 monitors the raw material addition amount at the feed inlet 13 and the discharge time of the mixed material at the discharge outlet 14 through the flow sensors 151. The PLC control system 3 controls the raw material addition amount at the feed inlet 13 and the discharge time of the mixed material at the discharge outlet 14 by controlling the opening and closing times of the first solenoid valve 8 and the second solenoid valve 9; A pressure sensor 153 is provided near the bottom of the reaction chamber. The monitoring device 15 monitors the pressure of the reaction chamber through the pressure sensor 153. The PLC control system 3 controls the exhaust gas volume of the exhaust port 18 by controlling the opening and closing time of the third solenoid valve 10.

[0096] In this embodiment, the data center 2 includes:

[0097] A data storage module for storing historical reaction data, including time series data of the mixed liquid temperature, reaction chamber pressure, mixed liquid viscosity, conductivity, and raw material flow rate;

[0098] A deep learning module integrating a Transformer-LSTM time series prediction model. For the Transformer-LSTM model, first, the long short-term memory neural network is used to capture the time series features of the given mixed liquid temperature, mixed liquid viscosity, and mixed liquid conductivity data from the time dimension, strengthening the model's ability to model long-distance dependencies of time series data; Subsequently, the output of the LSTM network is used as the input of the Transformer network, and the global dependency relationship is captured through the multi-head self-attention mechanism, accurately capturing the global correlation of the mixed liquid temperature, mixed liquid viscosity, and mixed liquid conductivity data in the time series dimension; Finally, the features strengthened by integrating the LSTM global time series and the Transformer position encoding are used as the input of the fully-connected neural network to predict the mixed liquid temperature, mixed liquid viscosity, and conductivity at the next moment, and send instructions to the PLC control system 3 according to the prediction results to dynamically adjust the heating power, stirring speed, and discharge timing of the reactor, thereby reducing the control lag during the reaction process and improving the reaction efficiency.

[0099] An adaptive optimization control module outputs the prediction results and instructions to the PLC control system 3. The PLC control system 3 dynamically adjusts the heating power of the heating device 17, the rotation speed of the motor 11, the raw material addition amount at the feed inlet 13, the discharge time of the mixed material at the discharge outlet 14, and the exhaust gas volume of the exhaust port 18 according to the instructions.

[0100] In this embodiment, the raw material cylinder includes a solid raw material cylinder 6 for storing solid raw materials and a leaching agent cylinder 7 for storing leaching agents. The output ends of the solid raw material cylinder 6 and the leaching agent cylinder 7 are both connected to the feed inlet 13. A total of three first solenoid valves 8 are provided, which are respectively arranged at the output end of the solid raw material cylinder 6, the output end of the leaching agent cylinder 7, and the feed inlet 13. A total of two flow sensors 151 are provided, which are respectively arranged at the output end of the solid raw material cylinder 6 and the output end of the leaching agent cylinder 7, and can respectively monitor the actual flow rates of the solid raw materials and the leaching agents; the PLC control system 3 controls the opening degree of the first solenoid valve 8 according to the difference between the actual flow rate and the target flow rate, and the control logic is as follows:

[0101]

[0102] Wherein:

[0103] u(t) is the opening degree of the first solenoid valve 8;

[0104] e1(t) is the difference between the target flow rate and the actual flow rate;

[0105] K p ,K i ,K d are respectively the proportional, integral, and differential gains of the PID controller included in the PLC control system 3 (the same hereinafter).

[0106] In this embodiment, a conductivity tester 155 is arranged in the reaction chamber. The conductivity tester 155 is fixed at a position on the inner side wall of the reactor close to the liquid level of the mixed solution. The probe of the conductivity tester 155 is inserted into the mixed solution, and can monitor the conductivity of the mixed solution and upload the monitored data to the data center 2. The probe position is away from the stirring paddle 12 to avoid the interference of stirring eddies or solid particles on the probe monitoring results. When it is monitored that the conductivity of the mixed solution is stable at the set value, the PLC control system 3 opens the discharge port 14 through the second solenoid valve 9.

[0107] In this embodiment, the execution logic of the PLC control system includes:

[0108] When the Transformer-LSTM model predicts that the temperature of the mixed solution will exceed the set range within the next Δt time, dynamically adjust the power of the heating device 17;

[0109] When the Transformer-LSTM model predicts that the viscosity of the mixed solution will exceed the set range within the next Δt time, dynamically adjust the power of the motor 11;

[0110] In this embodiment, the PLC control system 3 controls the opening and closing of the third solenoid valve 10 according to the pressure of the reaction chamber monitored by the pressure sensor 153. The control logic is as follows: when the pressure of the reaction chamber exceeds the set safety threshold, the third solenoid valve 10 is opened until the pressure of the reaction chamber is equal to or less than the safety threshold.

[0111] In this embodiment, through the monitoring device 15 in the reaction device, the system will collect reaction data for 3 to 6 months to construct a training set. The training set includes multi-dimensional time series data such as the temperature of the mixed solution, the viscosity of the mixed solution, and the conductivity, which are used to train the deep learning module Transformer-LSTM to ensure the accuracy of its prediction results.

[0112] In this embodiment, the long short-term memory neural network LSTM is a recurrent neural network commonly used to process time series data, which can better adapt to the time series data of the state parameters in the hydrometallurgical leaching process and mine the hidden global time series features in the data. Its main components include:

[0113] Forgot gate F t , which determines the retention of the previous cell state. Its mathematical expression is:

[0114] F t =θ(M f ·[g t-1 ,y t +d f ),

[0115] where θ is the sigmoid activation function, M f and d f are trainable weights and biases, g t-1 is the information of the previous moment, and y t is the input of the current moment;

[0116] Input gate I t , which is used to control whether to update the cell state. Its mathematical expression is:

[0117] I t =θ(M i ·[g t-1 ,y t +d i ),

[0118] where M i and d i are the trainable weights and biases of the input gate;

[0119] Candidate cell state is used to determine the new information that needs to be stored in the cell state at the current moment. Its mathematical expression is as follows:

[0120]

[0121] Among them, tang is the activation function;

[0122] The new cell state H t , which is used to update according to the input at the current moment and the state at the previous moment, and its mathematical expression is as follows:

[0123]

[0124] Among them, ⊙ represents element-wise multiplication;

[0125] The output gate O t , which determines the output amount of the hidden state, and its mathematical expression is:

[0126] O t = θ(M o ·[g t-1 , y t +d o )

[0127] Among them, M o and d o are the trainable weights and biases of the output gate;

[0128] The output hidden state g t , which is jointly determined by the activation value of the output gate and the non-linear transformation of the cell state, and its mathematical expression is:

[0129] g t = O t ⊙tang(H t )

[0130] In this embodiment, the Transformer network is a sequence-to-sequence model based on the self-attention mechanism, which can accelerate the process of model training and inference, and more accurately capture the global dependence relationship of the time series data of the state parameters in the hydrometallurgical process. The input of the Transformer is the output of the LSTM. Assume that the dimension of this output is b k , and its calculation formula is:

[0131]

[0132] Among them, Q, K, and V are the query, key, and value matrices respectively, and b k is the dimension scaling factor.

[0133] To verify the accuracy of model prediction, the present invention constructs an integrated Transformer-LSTM time series prediction model and an LSTM single model, and conducts a comparative experiment verification using simulation data. Taking the prediction result of the mixed liquid viscosity as an example, the prediction results of the Transformer-LSTM time series prediction model and the LSTM single model are as Figure 4 and Figure 5 shown. The R 2 of the Transformer-LSTM model reaches 0.9633, and the MAPE and MAE only reach 0.0372 and 3.1445 respectively. Compared with the single model LSTM, the R 2 increases by 5.9%, and the MAPE and MAE decrease by 11.42% and 15.1% respectively.

[0134] In this embodiment, the heating device includes:

[0135] A jacket layer 16 that defines a heating chamber and is provided with a heating medium inlet 161 and a heating medium outlet 162 communicating with the heating chamber;

[0136] Two heating devices 17 are symmetrically arranged in the heating chamber for heat exchange with the heating medium, and the PLC control system 3 can control the heating power of the heating device 17.

[0137] In this embodiment, a temperature sensor 152 is arranged in the reaction chamber. The temperature sensor 152 is close to the bottom of the reaction chamber and can monitor the actual temperature of the reaction chamber, the temperature of the mixed liquid. The PLC control system 3 controls the power of the heating device 17 through the first frequency converter 4 according to the difference between the actual temperature and the target temperature. The control logic is:

[0138]

[0139] Wherein:

[0140] T(t) is the power of the heating device 17;

[0141] e2(t) is the difference between the target temperature and the actual temperature;

[0142] K p , K i , K d are the proportional, integral, and differential gains of the PID controller respectively.

[0143] In this embodiment, a viscometer 154 is arranged in the reaction chamber. The monitoring device 15 monitors the actual viscosity of the mixed liquid through the viscometer 154. The PLC control system 3 controls the rotation speed of the motor 11 through the second frequency converter 5 according to the difference between the actual viscosity and the target viscosity. The control logic is:

[0144]

[0145] Wherein:

[0146] ω(t) is the power of the motor 11;

[0147] e3(t) is the difference between the target viscosity and the actual viscosity;

[0148] K p , K i , K d are the proportional, integral, and derivative gains of the PID controller, respectively.

[0149] An embodiment of the present invention further provides a wet leaching method, which applies the above chemical reactor for deep learning and automatic control, and includes the following steps:

[0150] S1: The PLC control system 3 controls the raw materials to enter the reaction chamber from the feed port 13 at the target flow rate. In order to prevent the raw materials from reacting in advance and causing the blockage of the feed port 13, the PLC control system 3 first opens the first solenoid valve 8 corresponding to the solid raw material cylinder 6, and then opens the first solenoid valve 8 corresponding to the leaching agent cylinder 7, and feeds the materials in sequence.

[0151] The motor 11 starts, drives the stirring paddle 12 to stir the raw materials at a preset speed, so that the solid raw materials and the leaching solution are fully mixed. At the same time, the heating device 17 is started to heat the mixed solution in the reaction chamber to the target temperature.

[0152] S2: The monitoring device 15 monitors the raw material addition amount at the feed port 13, the discharge time of the mixed material at the discharge port 14, the temperature of the reaction chamber, the pressure of the reaction chamber, and the viscosity of the mixed solution in real time. The monitoring device 15 uploads the monitored data to the data center 2;

[0153] S3: The Transformer-LSTM model in the data center 2 receives the monitoring data in real time, predicts the future change trend, screens the key time series features through the attention mechanism, so as to more accurately predict the parameter fluctuations, and dynamically adjusts the heating power, stirring speed, and discharge timing of the reactor according to the prediction results, thereby reducing the control lag in the reaction process and improving the reaction efficiency;

[0154] S4: The PLC control system 3 dynamically adjusts the heating power of the heating device, the rotation speed of the motor 11, the raw material addition amount at the feed port 13, the discharge time of the mixed material at the discharge port 14, and the exhaust volume of the exhaust port 18 according to the adjustment instructions of the data center 2.

[0155] Specifically, the PID control algorithm is used to adjust the heating power and cooling power in real time to regulate the temperature of the mixed liquid in the reaction chamber. When the temperature is too low, the power of the heating device 17 is increased through the first frequency converter 4 to raise the medium temperature; when the temperature is too high, the power of the heating device 17 is decreased through the first frequency converter 4 to lower the medium temperature. Ensure that the temperature of the mixed liquid always remains within the target range.

[0156] Based on the data of the viscosity change of the mixed liquid collected in real time, the PLC control system 3 adjusts the power of the motor 11 in real time through the second frequency converter 5 to change the stirring speed. When the viscosity is too small, the stirring power is reduced and the stirring speed is lowered; when the viscosity is too large, the stirring power is increased and the stirring speed is raised.

[0157] The PLC control system 3 controls the opening and closing of the exhaust port 18 according to the pressure of the reaction chamber collected in real time; when the pressure of the reaction chamber exceeds the set safety threshold, the third solenoid valve 10 is automatically opened to discharge the gas from the exhaust port 18 to prevent damage to the reaction system.

[0158] During the stirring process, the PLC control system 3 collects the conductivity data of the mixed liquid in real time. When the conductivity reaches the set stable value and lasts for a certain period of time, the second solenoid valve 9 is opened to discharge the mixed liquid.

[0159] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention.

[0160] The embodiments described above are only for describing the preferred mode of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solution of the present invention should fall within the protection scope determined by the claims of the present invention.

Claims

1. A chemical reactor integrating deep learning and automatic control, characterized in that, Comprising: A chemical reactor (1) defining a reaction chamber for wet leaching and having a feed inlet (13), a discharge outlet (14) and an exhaust outlet (18) communicating with the reaction chamber; A motor (11) whose output end extends into the reaction chamber and is connected to a stirring paddle (12) for stirring the mixed liquid in the reaction chamber; A heating device (17) disposed on the outer wall of the chemical reactor (1) for adjusting the temperature of the reaction chamber; A monitoring device (15) for monitoring the raw material addition amount at the feed inlet (13), the discharge time of the mixed material at the discharge outlet (14), the temperature of the reaction chamber, the pressure of the reaction chamber and the viscosity of the mixed liquid. The monitoring device (15) is communicatively connected to a data center (2) and uploads the monitored data to the data center (2); the data center (2) integrates a deep learning module to perform time series prediction on the temperature of the mixed liquid, the viscosity of the mixed liquid and the conductivity in the reaction chamber based on the historical reaction data uploaded by the monitoring device (15), and generates an optimal parameter combination by analyzing the data characteristics of multiple batches; A PLC control system (3) communicatively connected to the data center (2) to dynamically adjust the heating power of the heating device (17), the rotation speed of the motor (11), the raw material addition amount at the feed inlet (13), the discharge time of the mixed material at the discharge outlet (14) and the exhaust volume of the exhaust outlet (18) according to the optimal parameter combination generated by the data center (2).

2. The chemical reactor for deep learning and automatic control according to claim 1, characterized in that, Further comprising: A raw material cylinder communicated with the feed inlet (13) through a feed pipeline, and a first electromagnetic valve (8) is provided on the feed pipeline; a second electromagnetic valve (9) is provided at the discharge outlet (14), and a third electromagnetic valve (10) is provided at the exhaust outlet (18); the first electromagnetic valve (8) is used to control the opening and closing of the feed inlet (13), the second electromagnetic valve (9) is used to control the opening and closing of the discharge outlet (14), and the third electromagnetic valve (10) is used to control the opening and closing of the exhaust outlet (18); Flow sensors (151) are provided at the feed inlet (13) and the discharge outlet (14). The monitoring device (15) monitors the raw material addition amount at the feed inlet (13) and the raw material discharge amount at the discharge outlet (14) through the flow sensors (151). The PLC control system (3) controls the raw material addition amount at the feed inlet (13) and the discharge time of the mixed material at the discharge outlet (14) by controlling the opening and closing time of the first electromagnetic valve (8) and the second electromagnetic valve (9); a pressure sensor (153) is provided in the reaction chamber. The monitoring device (15) monitors the pressure of the reaction chamber through the pressure sensor (153). The PLC control system (3) controls the exhaust volume of the exhaust outlet (18) by controlling the opening and closing time of the third electromagnetic valve (10).

3. The chemical reactor for deep learning and automatic control according to claim 2, characterized in that, The data center (2) includes: A data storage module for storing historical reaction data, including time series data of the mixed liquid temperature, the reaction chamber pressure, the mixed liquid viscosity, the conductivity and the raw material flow rate; Deep learning module, integrating the Transformer-LSTM time series prediction model; the Transformer-LSTM model first uses the long short-term memory neural network to capture the temporal features of the given mixed liquid temperature, mixed liquid viscosity, and mixed liquid conductivity data from the time dimension, strengthening the model's ability to model long-distance dependencies of time series data; subsequently, the output of the LSTM network is used as the input of the Transformer network, and the multi-head self-attention mechanism is used to capture global dependencies, capturing the global correlation of the mixed liquid temperature, mixed liquid viscosity, and mixed liquid conductivity data in the temporal dimension; finally, the features strengthened by integrating the LSTM global time series and the Transofmer position encoding are used as the input of the fully-connected neural network to predict the mixed liquid temperature, mixed liquid viscosity, and conductivity at the next moment, and send instructions to the PLC control system (3) according to the prediction results to dynamically adjust the heating power of the heating device (17), the rotation speed of the motor (11), the raw material addition amount at the feed inlet (13), the discharge time of the mixed material at the discharge outlet (14), and the exhaust volume of the exhaust port (18); Adaptive optimization control module, outputting the prediction results and instructions to the PLC control system (3), and the PLC control system (3) dynamically adjusts the heating power of the heating device (17), the rotation speed of the motor (11), the raw material addition amount at the feed inlet (13), the discharge time of the mixed material at the discharge outlet (14), and the exhaust volume of the exhaust port (18) according to the instructions.

4. A chemical reactor for deep learning and automatic control according to claim 3, characterized in that, The raw material cylinder includes a solid raw material cylinder (6) for storing solid raw materials and a leaching agent cylinder (7) for storing leaching agents. The output ends of the solid raw material cylinder (6) and the leaching agent cylinder (7) are both connected to the feed inlet (13). The flow sensor (151) can respectively monitor the actual flow rates of the solid raw materials and the leaching agent; the PLC control system (3) controls the opening degree of the first solenoid valve (8) according to the difference between the actual flow rate and the target flow rate, and the control logic is: Where: u(t) is the opening degree of the first solenoid valve (8); e1(t) is the difference between the target flow rate and the actual flow rate; K p , K i , K d are the proportional, integral, and derivative gains of the PID controller, respectively.

5. The chemical reactor for deep learning and automatic control according to claim 3, characterized in that, A conductivity tester (155) is arranged in the reaction chamber. The conductivity tester (155) can monitor the conductivity of the mixed liquid and upload the monitored data to the data center (2). When it is monitored that the conductivity of the mixed liquid is stable at the set value, the PLC control system (3) opens the discharge outlet (14) through the second solenoid valve (9).

6. The chemical reactor for deep learning and automatic control according to claim 3, characterized in that, The execution logic of the PLC control system (3) includes: When the Transformer-LSTM model predicts that the mixed liquid temperature will exceed the set range within the next Δt time, dynamically adjust the power of the heating device (17); When the Transformer-LSTM model predicts that the mixed liquid viscosity will exceed the set range within the next Δt time, dynamically adjust the power of the motor (11).

7. The chemical reactor for deep learning and automatic control according to claim 3, characterized in that, The PLC control system (3) controls the opening and closing of the third solenoid valve (10) according to the pressure of the reaction chamber monitored by the pressure sensor (153). The control logic is as follows: when the pressure of the reaction chamber exceeds the set safety threshold, the third solenoid valve (10) is opened until the pressure of the reaction chamber is equal to or less than the safety threshold.

8. A chemical reactor for deep learning and automatic control according to claim 1, characterized in that, The heating device includes: A jacket layer (16) that defines a heating chamber and is provided with a heating medium inlet (161) and a heating medium outlet (162) communicating with the heating chamber; A heating device (17) disposed in the heating chamber for heat exchange with the heating medium, and the PLC control system (3) can control the heating power of the heating device (17); A temperature sensor (152) is disposed in the reaction chamber. The temperature sensor (152) can monitor the actual temperature of the reaction chamber. The PLC control system (3) controls the power of the heating device (17) through the first frequency converter (4) according to the difference between the actual temperature and the target temperature. The control logic is as follows: Where: T(t) is the power of the heating device (17); e2(t) is the difference between the target temperature and the actual temperature; K p , K i , K d are the proportional, integral, and derivative gains of the PID controller, respectively.

9. The chemical reactor for deep learning and automatic control according to claim 1, wherein, A viscometer (154) is disposed in the reaction chamber. The monitoring device (15) monitors the actual viscosity of the mixed liquid through the viscometer (154). The PLC control system (3) controls the rotation speed of the motor (11) through the second frequency converter (5) according to the difference between the actual viscosity and the target viscosity. The control logic is as follows: Where: ω(t) is the power of the motor (11); e3(t) is the difference between the target viscosity and the actual viscosity; K p , K i , K d are the proportional, integral, and derivative gains of the PID controller, respectively.

10. A resource recovery method, characterized in that, Applying the chemical reactor with deep learning and automatic control according to any one of claims 3-9, comprising the following steps: S1: The PLC control system (3) controls the raw materials to enter the reaction chamber from the feed port (13) at the target flow rate; S2: The monitoring device (15) monitors in real time the raw material addition amount at the feed port (13), the discharge time of the mixture at the discharge port (14), the temperature of the reaction chamber, the pressure of the reaction chamber, and the viscosity of the mixed liquid. The monitoring device (15) uploads the monitored data to the data center (2); S3: The Transformer-LSTM model of the data center (2) receives the monitoring data in real time, predicts the future change trend, and generates adjustment suggestions; S4: The PLC control system (3) dynamically adjusts the heating power of the heating device (17), the rotation speed of the motor (11), the raw material addition amount at the feed port (13), the discharge time of the mixture at the discharge port (14), and the exhaust volume at the exhaust port (18) according to the adjustment instructions of the data center (2).

Images