CN116822380B - Collaborative optimization method for tail gas recycling in copper smelting process based on digital twin - Google Patents

Abstract

The invention discloses a collaborative optimization method for tail gas recycling in the copper smelting process based on digital twins, and belongs to the field of intelligent control of metal smelting tail gas recycling. Collect the state parameters and process parameters of the sulfuric acid reaction tower to obtain complete data information of the entire production process. Secondly, build a digital twin model of the desulfurization and acid production process based on the above data, and use the constructed digital twin model of the production process to pass The BPNLP neural network continuously iteratively obtains optimal operating parameters, analyzes and compares them with real-time data of physical equipment, and uploads the obtained data to the human-computer interaction interface through the communication protocol to achieve "virtual-real linkage" between the entire industrial production and the client; finally, Based on the optimization parameter results, the operating parameters of the actual production process can be optimized and controlled in real time through the process parameter control system to achieve "virtual control of reality" and ensure long-term maximization of desulfurization efficiency and sulfuric acid production speed.

Description

A collaborative optimization of exhaust gas recovery and utilization in the copper smelting process based on digital twins method

Technical field

The invention relates to the field of intelligent control of metal smelting tail gas recycling, and specifically relates to a collaborative optimization method for tail gas recycling in the copper smelting process based on digital twins.

Background technique

At present, Qiu Zujun, Zhang Feng and others have proposed MGG (Mitsubishi Gas-gas Heater) flue gas heat exchange technology. SO ₂ in the flue gas is absorbed and oxidized into ammonium sulfate, and then evaporated, concentrated and crystallized to obtain ammonium sulfate products. Cai Bing et al. People have proposed a hydrogen peroxide flue gas desulfurization technology, which has become the main means for many factories to solve sulfur dioxide air pollution. However, the entire process involves the distillation operation of liquid phase separation of sulfur dioxide and hydrogen peroxide. This operation is typical For many years, the modeling, optimization, and control issues of complex processes involving multiple input and multiple output (MIMO), large delays, severe nonlinearity, and strong parameter coupling have been hot topics studied by many scholars. And currently a large amount of research is focused on the replacement of traditional desulfurization with desulfurization and acid production for the purpose of pollution prevention and control. In the acid making process, the input of relevant parameters such as liquid level, temperature, and material ratio will have a great impact on the output parameters sulfur dioxide conversion rate and absorption rate. At the same time, most of the existing research on desulfurization and sulfuric acid production focuses on its environmental pollution assessment, or only focuses on the recovery efficiency of sulfuric acid. There are few studies on collaborative optimization methods for pollution prevention and resource recovery in the desulfurization and acid production process.

BP neural network is a model widely used in industrial process optimization; the gradient method used by BP neural network often results in Falls local minima; if the fitting function is too complex, multiple fitting local minima may appear, This will bring great obstacles to practical application.

Contents of the invention

In view of the optimization problem in the acid-making distillation operation, the purpose of the present invention is to provide a collaborative optimization method for exhaust gas recovery and utilization in the copper smelting process based on digital twins, which can realize real-time collection, analysis, and decision-making of state parameters in the acid-making process. According to the order, by analyzing the sulfur dioxide method, coupling the multi-parameter layout and desulfurization process, an interactive, visual and optimizable digital twin model of the flue gas sulfuric acid production system was established to achieve the purpose of "combination of virtual and real, and virtual control of real". To maximize the production rate and greater economic value of sulfur dioxide acid production, the following steps are specifically included:

Step1: Retrieve the state parameters, process parameters, and evaluation indicators of desulfurization and acid production in the smelting process from the previous database, and perform data preprocessing, data prediction, and data mining to build a previous database.

Step2: Collect real-time data through on-site industrial contact and electromagnetic sensors or rely on manual test records, including status parameters and process parameters. These parameters together constitute a real-time database; the status parameters include furnace body temperature, pressure, liquid, etc. position, gas-liquid ratio; the process parameters include the concentration and flow rate of input sulfur dioxide and hydrogen peroxide, and the concentration and flow rate of output sulfuric acid.

Step3: The previous database and the real-time database constitute the digital twin dynamic model database. The data interoperability connection between the digital twin dynamic model database and the server, the server and the front-end interface composed of unity3D is realized through the TCP communication protocol, and the relevant data in the digital twin dynamic database are The visual graphics are displayed on the front-end interface.

Step 4: Introduce the digital twin dynamic model database obtained in Step 3 as the digital twin model simulation signal into the digital twin model of desulfurization and acid production. The model is simulated through data. The simulation model uses nonlinearity based on the two-stage BPNLP neural network. The optimization is combined with an approximate global optimization algorithm to continuously optimize the process parameters corresponding to the optimal parameters obtained through the digital twin model. The most suitable production optimal engineering parameters are obtained based on the maximizing parameters of the sulfur dioxide desulfurization rate and sulfuric acid conversion rate. .

Step 5: Perform data mining similar aggregation, noise reduction, and missing value processing on the process parameters predicted to maximize the sulfur dioxide desulfurization rate and sulfuric acid conversion rate with previous databases, and then compare parameters with the real-time database to determine the current real-time data Which parameters still need to be improved? Finally, the optimal parameter improvement plan is obtained by optimizing the global algorithm under real-time data.

Step6: Visualize the optimal parameter data improvement plan through the TCP protocol, and display the visual graphics of the data on the human-machine interface of the unity3D front-end.

Step7: In the optimal parameter improvement plan, the operator issues control instructions based on the obtained parameter improvement plan and manual experience.

Step8: The control instructions are passed through the TCP protocol and simulink's desulfurization and sulfuric acid production simulation control model (including the liquid level-flow cascade control system for gas and liquid inlet and outlet, the liquid level-pressure cascade control system in the tower, and the dual input and The three control systems (double output hierarchical decoupling control system) are connected to form a control system, as shown in Figure 3. The control system is remotely connected to the actual industrial server, compares the optimal parameter values with the real-time data values, and compares them based on their differences. The actual furnace body is controlled by parameters, and the valves and regulators in production are controlled to guide actual production.

Step9: Continue to repeat the steps of Step2~Step5, collect real-time data, obtain the optimal control parameters through simulating the neural network digital twin model, and then compare the two data, and then feed back the control furnace process parameters, and repeat continuously for a long time. Ensure the desulfurization efficiency and sulfuric acid production speed are maximized.

Since the final sulfuric acid output and desulfurization conversion rate of the entire reaction process are coupled together by multiple factors such as temperature, pressure, catalyst and hydrogen peroxide level in the tower, the simulation model uses the non-linear method based on the two-stage BPNLP neural network described in S4. Linear optimization is combined with an approximate global optimization algorithm to build the two-stage BPNLP network model. After the neural network preprocessing in the first stage forms the fitting of the smelting process, the smelting parameter optimization process is performed based on the fitted neural network model; according to Based on the characteristics of the entire two-stage process, as well as the corresponding conditional constraints and the smoothing process of the BP neural network output function, an approximate global optimization algorithm is used to form a two-stage BPNLP network model with unified form constraints. The specific composition, involved routes, and operating procedures are shown in the figure. 3, specifically including the following:

S1: Process the data in the data model library and specify the input samples transmitted from the database; according to the production process of desulfurization and acid production in copper smelting, when the input samples obey the normal distribution, select the mean and sum of the samples Standard deviation σ.

S2: The BP neural network model belongs to the error back propagation neural network and needs to be constructed in the following three aspects: input layer, hidden layer and output layer; the model includes neurons, weights, thresholds, layers and activation functions; the present invention is based on the desulfurization process A BP neural network model was established for the real-time state parameters of the acid process and the recovery and utilization conversion rate of pollutant sulfur.

S3: First, the parameters of the input layer, hidden layer and output layer are established. The five state parameters in the database are used as input variables of the BP natural network model, namely pressure, temperature in the tower, liquid level, gas-liquid ratio and sulfur gas. Flow input, the output layer is the conversion rate and desulfurization efficiency of sulfuric acid. Based on the hidden layer calculation of the BP neural network model, the hidden layer between the smelting parameters of the secondary desulfurization process and the collaborative optimization of sulfuric acid is set to two layers, and based on Actual process setting node.

S4: The conversion function for the above two hidden layers can be defined as

where p and q are the dimensions of the input and output variables, N is the number of samples; i is the hidden layer of the neuron, i=1 represents the input layer, That is the input variable, such as/> is the reactor pressure during the smelting process/> is the temperature inside the tower,/> is the liquid level in the tower,/> is the gas-liquid ratio from the input pipe to the tower and/> The flow rate of sulfur gas is introduced, etc./> It means output variable parameters, and for the above/> The number of samples for each type of input variable in is defined as/> , to calculate each sample, as follows:

Because i is the hidden layer of the neuron, its output value needs to be calculated for each sample details as follows:

It is assumed here that for the k-th layer, there are s neurons; l (l=1,2, _... H) is the number of hidden layers; where represents the weight corresponding to each of the above input variables, θ is the offset value,/> That is the bias value of the first hidden layer of the kth layer, p is a fixed value indicating the number of samples;/> is the sample node of the input variable in the kth layer.

Similarly, with As input to the model, the output can be obtained/> :

In order to make the result close to the target, the output The error with the actual output is transmitted backward from the output layer to the next layer of the network;/> Represents the bias value of the i-th hidden layer in each pass. Each pass, the threshold and weight of the neuron are adjusted; this recursive process is repeated until the k-th hidden layer, so the function of the output layer can be obtained value/> :

S5: However, due to the insufficient generalization ability of the BP neural network model, outliers may appear in the model simulation results. Therefore, the model output needs to be smoothed twice; the generalization accuracy of the model before non-smoothing is defined as a singular value (negative number) The proportion of the predicted values of all samples; so that the generalization accuracy can be set to an acceptable range;

According to the core of the present invention: under what conditions, under what state and combination of process parameters, the maximum conversion rate and maximum desulfurization efficiency can be obtained; then the generalization accuracy of the above model is optimized, and the parameter optimized model is used The BP neural network performs two-stage prediction simulation learning of BPNLP.

S6: Set the output function of the BP neural network model to the maximum conversion rate of sulfuric acid and the maximum desulfurization efficiency in the smelting flue gas. In order to conduct better simulation predictions of the BP neural network model and solve the problem of minimizing the output function of the smelting flue gas desulfurization ;The BPNLP neural network model used for pollution reduction optimization is expressed as the following mathematical function:

J represents the index function, which is used to represent the function index of the optimal desulfurization conversion rate and rate to be obtained. The function Min represents the mapping of the output from the two-stage BPNLP estimation results, where the input variable x is the desired index; where c is the linear transformation The vector is a 2×1 vector of sulfuric acid conversion rate and desulfurization efficiency; represents the estimation result predicted by the BP model, Represents the linear transformation from the output estimate of a BPNLP model with input variables to the desired index.

To minimize the emission problem, the nonlinear programming is as follows:

Among them, b, L, U are constant column vectors, f is a nonlinear function, x is a decision variable vector, which includes pressure, tower temperature, liquid level, gas-liquid ratio and sulfur dioxide flow rate. The incoming 5× 1 vector; A is a constant matrix, and the linear truncation function Ax represents the data after passing through the normalized domain and eliminating anomalies.

S7: Through the above fitting function N(x), x can generate multiple local minima, generate multiple solutions and further affect the global optimization; use the random scatter method to select the initial value to adapt to the equal step changes of the initial value optimization ; Confirm that the minimum value of the model simulation is selected from the optimization results in the optimal value; for the entire neural network model, grid the area of the variable to be optimized using the "grid and sample" method, combined with local search skills, The solution process within the evolutionary framework is used to find the global optimal solution.

S8: In addition, the output function is smoothed to make the predicted value meaningful; assuming that the output sample takes a certain dimension value of the output value, then the comparison of the model needs to be further expanded to simulate the output value, so that the output function smoothly takes the domain value, and filter out unreasonable values (for example, under basically the same conditions, the difference between the predicted output value and its adjacent output value is obviously too large or too small).

S9: Finally complete the solution of the optimal parameters of the data twin model, and the results obtained, such as a collaborative optimization method based on digital twins for exhaust gas recycling in the copper smelting process. In specific step S6, the data is visualized and sent to unity3D to complete human-computer interaction. follow-up content.

Another object of the present invention is to provide a collaborative optimization system based on digital twins for tail gas recycling in the copper smelting process. As shown in Figure 2, it includes a database module, a data analysis module, a visualization module, an equipment management control module, and a communication module. Service module.

The database module is used to collect real-time data during the operation of desulfurization and sulfuric acid production in the smelting process, and obtain the data information of the tower in the entire process flow; including a geometric parameter data unit, an operating status parameter unit, a real-time process parameter unit, and a neural network. Network model unit.

The data analysis module obtains status data from the above-mentioned database module, preprocesses and analyzes the data to obtain the optimal parameter prediction values through the digital twin model, and achieves the purpose of collaborative optimization of resource recovery in the desulfurization and sulfuric acid production process; including algorithm library units , data preprocessing unit, feature extraction unit, data mining unit.

The visualization module is mainly used to graphically display the data of the above two modules; it includes a fusion model unit, a result transmission unit, a data graphical unit, and a graphic dynamic display unit.

The equipment intelligent management module is used to combine the above-mentioned module results, including the above-mentioned database module, to achieve equipment life cycle information acquisition, complete equipment real-time intelligent optimization operation and maintenance, and according to the optimal operating parameters and the output of the control system, Real-time and efficient control of actual production status parameters; including evaluation index unit, control module unit, intelligent decision-making unit and control instruction unit.

The communication service module mainly realizes the correlation between the above four modules, transfers the data in the database module to the data analysis module in real time, and sends the results obtained by the data analysis module to the visualization module, and finally obtains the results based on the data visualization. Applied to the equipment intelligent management module, it completes the information retrieval and association between various modules; including production data text, on-site images and other types of messages; and provides real-time communication functions to complete the real-time update function of images in the visualization module, and ensure The correctness, stability and security of communication ensure that the intelligent operation and maintenance module of the equipment issues correct instructions and completes the control; and records the operation logs of each module to record the platform for subsequent analysis and troubleshooting.

Preferably, the geometric parameter data unit of the present invention contains actual production field equipment data: including the relevant dimensions of the production furnace body, inlet and outlet pipes, gas and liquid flow rates, production parameters, etc., and stores and retrieves long-term operation data;

The operating status parameter unit of the present invention contains operating data during production and smelting, including temperature, pressure, liquid level and other relevant information in the production furnace body, describing the status and performance of the system or equipment during operation; and providing information about the entire system or equipment. Health status, work efficiency, resource utilization and other aspects of information, and store and retrieve long-term operating data.

The real-time process parameter unit of the present invention contains the pipeline input amount and production output amount that update the reaction in real time, and is used to describe and evaluate the current status and performance of the system, equipment or process, and provide information on the health status and work efficiency of the entire system or equipment. , resource utilization and other aspects of information, and store and retrieve long-term running data.

The neural network model unit of the present invention records each operation log record of the neural network model under long-term operation, and records each relevant parameter weight, number of iterations, and output-to-input ratio.

The algorithm library unit of the present invention is used to perform secondary smoothing processing on the output of the BP model, and utilizes its strong nonlinear mapping ability and flexible network structure to improve prediction reliability, and is suitable for large-scale projects such as the copper smelting desulfurization and acid production process. The sample data tends to fit better, and combined with the global optimal search method, the optimal parameters of the data twin model are solved.

The data preprocessing unit of the present invention preprocesses the missing data and large deviation noise that often occur in real-time data collection in advance by changing the optimal parameters in historical data to avoid repeated large amounts of the same optimization of equipment data.

The feature extraction unit of the present invention uses the BP neural network model and uses the five state parameters in the database as the input variables of the BP natural network model. The output layer is the conversion rate and desulfurization efficiency of sulfuric acid, as well as the settings of the hidden layer. Predict training and validate sample classification.

The data mining unit of the present invention discovers hidden patterns, association rules and trends in long-term production data. Before feature extraction, it performs advance data analysis and mines the overall characteristics of the obtained production input information, thereby identifying Trends, unusual patterns, and correlations in data.

The fusion model unit model fusion of the present invention combines the prediction results of multiple data analysis modules to obtain more accurate and reliable predictions; by combining the prediction results obtained in the model with different data and different dates, the entire system performance to achieve higher accuracy and robustness.

The result transmission unit of the present invention regularly retrieves the optimal parameters obtained by the data analysis module, updates the obtained optimal parameters in real time, and transmits these output results to the back-end data analysis system in a suitable form to support Make decisions, provide insights, drive subsequent visualization processes and conduct further analysis.

The data graphical unit of the present invention performs a visual graphical display of the result optimal parameter data obtained by the result transmission unit, and uses bar charts, line charts, scatter charts, pie charts, etc. to convey output parameters, process parameters, etc. Relationships, trends, and patterns between data.

The graphic dynamic display unit of the present invention mainly drives the entire module to continuously retrieve real-time data from the data analysis module and perform graphic updates for data visualization, so that the graphics can be updated with real-time or frequent changes in data, so as to present the latest information and information in a timely manner. trend.

The evaluation index unit of the present invention mainly compares the obtained optimal operating parameters with the real-time operating parameters, and determines which parameters need to be weighted according to the evaluation index. Under different evaluation indexes, high weights are required for regulation and measurement models. accuracy, performance, and reliability to help workers understand model performance and make decisions.

The control module unit of the present invention is a liquid level-flow cascade control system for gas and liquid in and out, a liquid level-pressure cascade control system in the tower, and dual input and dual input of the temperature in the tower based on the changed values obtained from the comparison. The decoupled control of output classification, under the regulation of feedback, completes the control from digital twin simulation to actual production.

The intelligent decision-making unit of the present invention uses the existing data and information of the data analysis module as a basis, and identifies patterns, trends and laws through analysis, mining and integration of optimal parameters and actual on-site production data, thereby providing targeted Advice on production decisions regarding performance and reliability.

The control instruction unit of the present invention refers to the key component responsible for generating, verifying and sending control instructions after intelligent decision-making, monitoring the system status and receiving feedback after the intelligent decision-making unit; completing the operation of the operator to control the actual production by sending commands to the system. It issues control instructions regarding valve opening and controller opening adjustment to the actual production control parameters of the valves, regulators, and controllers to guide the collaborative optimization of the ultimate goal of copper smelting tail gas desulfurization and sulfuric acid resource recovery.

The communication service module of the present invention provides the ability to connect, transmit and interact with the above four modules; the communication module allows communication between different systems, devices or nodes through physical or virtual channels, and supports data collection from the database module Send, receive and process message transmission, as well as the transmission and reception of data analysis module messages, including production data text, on-site images and other types of messages; and provide real-time communication functions to complete the real-time update function of images in the visualization module, and ensure The correctness, stability and security of communication ensure that the intelligent operation and maintenance module of the equipment issues correct instructions and completes the control; and records the operation logs of each module to record the platform for subsequent analysis and troubleshooting.

Compared with the prior art, the present invention at least has the following beneficial effects:

(1) The digital twin platform of the present invention based on the two-stage BPNLP network model for collaborative optimization of resource recovery in copper smelting tail gas desulfurization and sulfuric acid production is mainly aimed at typical multi-input multi-output, large delay, severe nonlinearity, and strong parameter coupling. For complex processes and complex problems in modeling, optimization, and control, we use digital twin technology and BPNLP neural network algorithm to establish real-time data upload interaction, establish a digital twin model to conduct real-time data analysis, and obtain desulfurization efficiency and sulfuric acid. The optimal operating parameters for conversion rate prediction, real-time data comparison and instruction issuance achieve collaborative optimization of resource recovery through the digital twin platform, which can effectively improve environmental protection capabilities and corporate benefits, and has broad application prospects.

(2) According to the characteristics of the desulfurization and acid production process of copper smelting in actual production, the present invention uses the BPNLP neural network based on the secondary smoothing process of the BP neural network and nonlinear optimization to obtain the optimal desulfurization using an approximate global optimization method. parameter. The present invention builds a simulation platform, which can obtain key parameter information in the flue gas acid-making distillation process in real time through TCP communication, such as pressure, temperature, and liquid level in the tower, and can realize data sharing of each unit for virtual and real purposes. Linkage, digital twin simulation is performed based on the conversion reaction of sulfur dioxide, and then the optimal sulfur dioxide desulfurization rate and the optimal equilibrium conversion rate of sulfuric acid in the flue gas sulfuric acid production process are predicted, as well as the corresponding temperature, pressure and other processes at this optimal equilibrium rate. parameters, and finally achieve the best desulfurization efficiency through feedback control of the actual production equipment controller.

Description of the drawings

In order to explain the technical solutions of the embodiments of the present invention more clearly, the drawings needed to describe the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without exerting creative efforts.

Figure 1 is a flow chart of the present invention's method for collaboratively optimizing digital twins for resource recovery in copper smelting tail gas desulfurization and sulfur production based on the two-stage BPNLP network model;

Figure 2 is the desulfurization and acid production process flow and digital twin model constructed by the present invention;

Figure 3 is a flow chart of the operation of the two-stage BPNLP neural network model used in the present invention;

Figure 4 is a graph showing the optimal parameter control parameters of sulfuric acid in the embodiment; (a) the iterative optimization and control process of the maximum sulfuric acid conversion rate under the BPNLP algorithm in Example 1; (b) the maximum desulfurization rate under the BPNLP algorithm in Example 1 Iterative optimization control process; (c) In the embodiment, it is assumed that the liquid level of 40 meters is the optimal output value, and the control process of the actual liquid level.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

Example 1

A collaborative optimization method for tail gas recycling in the copper smelting process based on digital twins, which specifically includes the following steps:

S1: Retrieve the state parameters, process parameters, and evaluation indicators of desulfurization and sulfuric acid production in the smelting process from the previous database, and perform data preprocessing, data prediction, and data mining to build an interactive, computable, and optimizable past database.

The data content used in the previous database specifically includes: tower equipment geometric characteristics such as tower height, tower diameter, tower head size, air supply volume, diesel volume, oxygen supply volume introduced through the pipeline, etc., as well as desulfurization systems. In the acid tower, the status parameters of the tower body temperature, liquid level, pressure, ratio, etc. of the desulfurization and acid-making tower; as well as the input end (the flow rate, flow rate, concentration and other related parameters of sulfur dioxide and hydrogen peroxide), the output end (characterization Process parameters such as sulfur dioxide conversion rate, sulfuric acid content, and water output that characterizes desulfurization rate); and related evaluation indicators such as tail gas treatment conversion rate, desulfurization efficiency, sulfuric acid conversion rate, and output tail gas sulfur content for the result output. .

S2: Collect real-time data through on-site industrial contact and electromagnetic sensors or rely on manual test records. The state parameters include the state parameters of temperature, pressure, liquid level, and gas-liquid ratio in the furnace. The process parameters include Input the concentration and flow rate of sulfur dioxide and hydrogen peroxide, and output the process parameters of the concentration and flow rate of sulfuric acid. These parameters together form a real-time database.

S3: The previous database and the real-time database constitute the digital twin dynamic model database, and the data interoperability connection between the digital twin dynamic model database and the server, the server and the front-end interface composed of unity3D is realized through the TCP communication protocol, and the digital twin dynamic database is Visual graphics of relevant data are displayed on this front-end interface.

S4: Use the collected past data and real-time production data to build a digital twin model of the copper smelting tail gas desulfurization and sulfuric acid production process: introduce the digital twin dynamic model database obtained in S3 as a digital twin model simulation signal into the digital twin model of the desulfurization and sulfuric acid production process. , the model conducts simulations through data, and uses nonlinear optimization based on a two-stage BPNLP neural network combined with an approximate global optimization method to continuously optimize the process parameters corresponding to the optimal parameters obtained through the digital twin model. According to the sulfur dioxide desulfurization rate The parameters maximizing the conversion rate of sulfuric acid are the most suitable engineering parameters for production.

S5: Perform data mining similar aggregation, noise reduction, and missing value processing on the predicted process parameters in the state where the sulfur dioxide desulfurization rate and sulfuric acid conversion rate are maximum with the previous database, and then compare the parameters with the real-time database to determine the current real-time data Which parameters still need to be improved? Finally, the optimal parameter improvement plan is obtained by optimizing the global algorithm under real-time data.

S6: Also through the TCP protocol, the optimal parameter data improvement plan is visualized, and the visual graphics of the data are displayed on the human-machine interface of the unity3D front-end.

S7: In the optimal parameter improvement plan, the operator issues control instructions based on the obtained parameter improvement plan and manual experience.

The control instructions pass through the TCP protocol and simulink's desulfurization and sulfuric acid production simulation control model (including the liquid level-flow cascade control system for gas and liquid inlet and outlet, the liquid level-pressure cascade control system in the tower, and the dual input and dual output of the temperature in the tower The hierarchical decoupling control system (these three control systems) are connected to form a control system, as shown in Figure 3. The control system is remotely connected to the actual industrial server, compares the optimal parameter values with the real-time data values, and compares the actual furnace body according to the difference. Carry out parameter control and control valves and regulators in production to guide actual production.

S8: Continue to repeat the steps of S2~S5, collect real-time data, obtain the optimal control parameters through simulating the neural network digital twin model, and then compare the two data, and then feed back the control furnace process parameters, and repeat continuously for a long time. Ensure the desulfurization efficiency and sulfuric acid production speed are maximized.

As a further preferred embodiment of the present invention:

The entire copper smelting flue gas desulfurization and acid production desulfurization process of the present invention is to dry and wash the flue gas discharged from the smelting furnace, and the sulfur dioxide in it is catalytically oxidized into sulfur trioxide in the converter, and then reacts with water in the absorption tower. The reaction produces sulfuric acid. The chemical equation of the reaction process is:

In the formula, Q is the heat released by the chemical reaction.

Since the final sulfuric acid output and desulfurization conversion rate of the entire reaction process are coupled together by multiple factors such as temperature, pressure, catalyst and hydrogen peroxide level in the tower, the simulation model uses the non-linear method based on the two-stage BPNLP neural network described in S4. Linear optimization is combined with an approximate global optimization algorithm to build the two-stage BPNLP network model. After the neural network preprocessing in the first stage forms the fitting of the smelting process, the smelting parameter optimization process is performed based on the fitted neural network model; according to Based on the characteristics of the entire two-stage process, as well as the corresponding conditional constraints and the smoothing process of the BP neural network output function, an approximate global optimization algorithm is used to form a two-stage BPNLP network model with unified form constraints. The specific composition, involved routes, and operating procedures are shown in the figure. 3, specifically including the following:

(1) Process the data in the data model library. Before training the neural network, the sample data needs to be processed; first, the input sample is specified; according to the production process of desulfurization and acid production in copper smelting, it is assumed that the input sample obeys the normal distribution. , using the basic principles of statistical data processing, select the mean and standard deviation σ in the sample; by expanding the actual interval output value of the model, the sample variance 2σ, which is the sum of squares of the difference between the sample data and the expected value; train and verify sample classification, mainly The content includes dividing all experimental data into training samples and simulation verification samples to build a BP natural network model; and comparing the corresponding model errors by calibrating the training set and prediction set.

(2) The BP neural network model is an error back propagation neural network and needs to be constructed in the following three aspects: parameter input layer, hidden layer and prediction result output layer in smelting production; the model includes neurons, weights, thresholds, layers and Activation function; This embodiment establishes a BP neural network model based on the real-time state parameters of the desulfurization and sulfur production process and the recovery and utilization conversion rate of pollutant sulfur.

(3) First, the parameters of the input layer, hidden layer and output layer are established. The five state parameters in the database are used as input variables of the BP natural network model, namely pressure, temperature in the tower, liquid level, gas-liquid ratio and sulfur gas. traffic access; the output layer is the conversion rate and desulfurization efficiency of sulfuric acid; based on the calculation of the hidden layer of the BP neural network model, the BP neural network model is a network structure with a dimension of "5-25-15-4". The hidden layer between the smelting parameters of the secondary desulfurization process and the collaborative optimization of sulfuric acid is set to two layers, and the nodes are set according to the actual process.

(4) The conversion function for the above two hidden layers can be defined as

That is the input variable, such as/> is the reactor pressure during the smelting process/> is the temperature inside the tower,/> is the liquid level in the tower,/> is the gas-liquid ratio from the input pipe to the tower,/> The flow rate of sulfur gas;/> is the output quantity,/> then represents the output sulfuric acid conversion rate,/> is the production speed of sulfuric acid.

Repeat the recursive process of BPNLP's two-stage optimization prediction iteration until the Hth hidden layer, so the function values of the five-dimensional input x and the two-dimensional output y can be obtained :

(5) However, when actually applied to the copper smelting industry, since the desulfurization process operation is a typical complex process with multiple inputs and multiple outputs, large delays, severe nonlinearity, and strong parameter coupling, there are widespread problems in using the BP neural network model alone. The problem of insufficient generalization ability and inaccurate simulation results; therefore, it is necessary to perform secondary smoothing on the model output; the generalization accuracy of the model before non-smoothing is defined as the proportion of singular values (negative numbers) in the predicted values of all samples; making the generalization The accuracy can be set to an acceptable range.

(6) The output function of the BP neural network model is set to the maximum conversion rate of sulfuric acid and the maximum desulfurization efficiency in the smelting flue gas. In order to conduct better simulation predictions of the BP neural network model and solve the problem of minimizing the output function of the desulfurization of the smelting flue gas Question; the BPNLP neural network model used for pollution reduction optimization is expressed as the following mathematical function:

J represents the index function, which is used to represent the function index of the optimal desulfurization conversion rate and rate to be obtained. The function Min represents the predicted output obtained from the many results estimated by the two-stage BPNLP. mapping, where the input variable x (each input parameter) is the desired index; where c is a linear transformation vector, which is a 2×1 vector of sulfuric acid conversion rate and desulfurization efficiency; Represents the estimation result of the BP model prediction, and represents the linear transformation from the output estimation result of the BPNLP model with input variables to the expected index.

To minimize the emission problem, the nonlinear programming is as follows:

Among them, b, L, U are constant column vectors, f is a nonlinear function, x is a decision variable vector, which includes pressure, tower temperature, liquid level, gas-liquid ratio and sulfur dioxide flow rate. The incoming 5× 1 vector; A is a constant matrix, and the linear truncation function Ax represents the data after passing through the normalized domain and eliminating anomalies.

(7) Through the above fitting function N(x), x can generate multiple local minima, which can produce multiple solutions and further affect the global optimization; the random scatter method is used to select the initial value to adapt to the initial value optimization etc. Step size change; confirm that the minimum value of the model simulation is selected from the optimization results obtained from the optimal value in the desulfurization and acid production process.

(8) In addition, the output function is smoothed to make the predicted value meaningful. Assume that the output sample takes a certain dimension value of the output value, and then the comparison of the model needs to be further expanded to simulate the output value, so that the output function smoothly takes the domain values in , and filter out unreasonable values (such as negative values).

(9) Finally complete the solution of the optimal parameters of the data twin model, and the results obtained, such as a collaborative optimization method based on digital twins for exhaust gas recovery and utilization in the copper smelting process, perform data visualization in step S6 and send it to unity3D to complete the human-machine completion The continuation of the interaction.

As a further preferred embodiment of the present invention:

S1: Retrieve the previous database, and use the data example of the desulfurization and acid production process in the copper smelting plant as real-time production data to demonstrate:

The physical parameters and state process parameters of this time are expressed as follows: the temperature in the reaction tower is set between 1000℃-1200℃, the pressure is about 2MPa, the height of the reaction tower is 70 meters, the diameter is 10 meters, and the sulfuric acid is introduced The flow rate is 500 cubic meters per hour and the flow rate is 5 meters per second.

S2: Connect the given equipment to the online prediction device for the remaining life of rolling bearings based on digital twins, including: database module, data analysis module, visual graphics module, communication service module, and equipment intelligent management module;

S3: Assume that the data collected in S1 is the real-time data collected in actual production, input the above-mentioned real-time data into the digital twin simulation model, and build a digital model with the above-mentioned data as the running state, with the purpose of achieving The data is dynamically interacted with the actual production equipment and linked with the virtual and real.

S4: Based on the digital twin model built in S3, the above-mentioned state parameters, physical parameters, and process parameter-related data are simulated through the algorithm library to simulate production conditions, and the main model in the algorithm library - the BPNLP neural network model is continuously repeated. Under simulation iteration, find and predict the optimal operating output parameters (the most suitable Desulfurization efficiency and sulfuric acid conversion rate); based on the above input, the optimal output simulated in this experiment is: desulfurization rate 94.23%, sulfuric acid conversion rate 98.9%, gas concentration 8%. The results of desulfurization rate and sulfuric acid conversion rate during the entire iteration process are shown in Figure 4(a) and Figure 4(b).

S5: According to the communication service module, connect the above-mentioned digital twin database and simulation data to the communication network. The communication module specifically includes: data packaging unit, communication architecture unit, data decoding unit, transmission scheduling unit, transmit them to the visualization platform, and perform Data visualization line chart is displayed and the optimal parameters are displayed.

S6: In actual production, operators can issue control instructions based on the comparison between the displayed optimal parameters and the current real-time operating parameters and the data weight. Under this simulation, the desulfurization rate of the optimal output parameters at this time is 94.23%, and The sulfuric acid conversion rate is 98.9%. The corresponding production indicators are found in the historical database. Based on the liquid level in the furnace, it is most appropriate to control the liquid level at about 40 meters. Then a control instruction is issued to target the liquid level at 40 meters.

S7: The above control instructions are connected to the feedback control system built by simulink. The control system completes the process parameter regulation based on the continuous feedback comparison between the input set value and the current value. Taking liquid level as an example, the entire liquid level control process is shown in Figure 4 ( c) shown.

S8: Connect the valves, regulators, and controllers under actual production conditions through the above communication service module to complete the guidance of actual production; and continuously repeat real-time data collection, digital twin simulation, visual chart update, and virtual chart updating according to actual production needs. Real control ensures long-term maximization of desulfurization efficiency and sulfuric acid production speed.

Example

A collaborative optimization system based on digital twins for tail gas recycling in the copper smelting process, as shown in Figure 2, includes a database module, a data analysis module, a visualization module, an equipment management control module, and a communication service module.

The database module described in this embodiment is used to collect real-time data during the operation of desulfurization and sulfuric acid production in the smelting process, and obtain data information of the tower in the entire process flow; including a geometric parameter data unit, an operating status parameter unit, and real-time process parameters. Unit, neural network model unit.

The geometric parameter data unit described in this embodiment contains actual production field equipment data: including the relevant dimensions, height, production parameters, etc. of the production furnace body, and long-term operation data is stored and retrieved;

The operating status parameter unit described in this embodiment contains operating data in production smelting, including temperature, pressure, liquid level and other relevant information in the production furnace, describing the status and performance of the system or equipment during operation; and providing information about the entire system or equipment. Information on equipment health, work efficiency, resource utilization, etc., and store and retrieve long-term operating data.

The real-time process parameter unit described in this embodiment includes updating the pipeline input and production output in real-time into the reaction, which is used to describe and evaluate the current status and performance of the system, equipment or process, and provide information about the health status and work of the entire system or equipment. Efficiency, resource utilization and other aspects of information, and long-term operation data storage and retrieval.

The neural network model unit described in this embodiment records each operation log record of the neural network model under long-term operation, and records each relevant parameter weight, iteration number, and output-input ratio.

The data analysis module described in this embodiment obtains status data from the above-mentioned database module, preprocesses and analyzes the data to obtain optimal parameter prediction values through the digital twin model, and achieves the purpose of collaborative optimization of resource recovery in the desulfurization and acid production process; including Algorithm library unit, data preprocessing unit, feature extraction unit, and data mining unit.

The algorithm library unit described in this embodiment is used to perform secondary smoothing processing on the output of the BP model, using its strong nonlinear mapping ability and flexible network structure to improve prediction reliability. Large sample data tends to fit better, and combined with the global optimal search method, the optimal parameters of the data twin model are solved.

The data preprocessing unit described in this embodiment preprocesses the missing data and large deviation noise that often occur in real-time data collection in advance by changing the optimal parameters in historical data to avoid repeating a large amount of the same optimization on equipment data.

The feature extraction unit described in this embodiment uses the BP neural network model and uses the five state parameters in the database as the input variables of the BP natural network model. The output layer is the conversion rate and desulfurization efficiency of sulfuric acid, as well as the settings of the hidden layer. Perform predictive training and verify sample classification.

The data mining unit described in this embodiment discovers hidden patterns, association rules and trends in long-term production data. Before feature extraction, it performs advance data analysis to mine the overall characteristics of the obtained production input information to identify Identify trends, unusual patterns, and correlations in your data.

The visualization module described in this embodiment is mainly used to graphically display the data of the above two modules; it includes a fusion model unit, a result transmission unit, a data graphical unit, and a graphic dynamic display unit.

The fusion model unit model fusion described in this embodiment combines the prediction results of multiple data analysis modules to obtain more accurate and reliable predictions; by combining the prediction results obtained in the model with different data and different dates, it is possible to improve Overall system performance for higher accuracy and robustness.

The result transmission unit described in this embodiment regularly retrieves the optimal parameters obtained by the data analysis module, updates the obtained optimal parameters in real time, and transmits these output results to the back-end data analysis system in a suitable form, so that Support decision-making, provide insights, drive subsequent visualization processes and conduct further analysis.

The data graphical unit described in this embodiment performs a visual graphical display of the optimal parameter data obtained by the result transmission unit, and uses bar charts, line charts, scatter charts, pie charts, etc. to display output parameters, process parameters, etc. Convey relationships, trends, and patterns among data.

The graphic dynamic display unit described in this embodiment mainly drives the entire module to continuously retrieve real-time data from the data analysis module and update the graphics for data visualization, so that the graphics can be updated with real-time or frequent changes in the data, so as to present the latest information in a timely manner. and trends.

The equipment intelligent management module described in this embodiment is used to combine the above module results, including based on the above database module, to achieve equipment life cycle information acquisition, complete equipment real-time intelligent optimization operation and maintenance, and based on the optimal operating parameters, in the output of the control system Under the control, the actual production status parameters are controlled in real time and efficiently; including evaluation index unit, control module unit, intelligent decision-making unit and control instruction unit.

The evaluation index unit described in this embodiment mainly compares the obtained optimal operating parameters with the real-time operating parameters, and determines which parameters need to be weighted according to the evaluation index. High weights are required for regulation and measurement under different evaluation indexes. Model accuracy, performance, and reliability help workers understand model performance and make decisions.

The control module unit described in this embodiment performs a liquid level-flow cascade control system for gas and liquid in and out, a liquid level-pressure cascade control system in the tower, and a dual input and The dual-output hierarchical decoupling control, under the adjustment of feedback, completes the control from digital twin simulation to actual production.

The intelligent decision-making unit described in this embodiment uses the existing data and information of the data analysis module as a basis, and identifies patterns, trends and laws through analysis, mining and integration of optimal parameters and actual on-site production data, thereby providing effective Recommendations for targeted and reliable production decisions.

The control instruction unit described in this embodiment refers to the key component responsible for generating, verifying and sending control instructions after intelligent decision-making, monitoring the system status and receiving feedback after the intelligent decision-making unit; the operator controls the actual production by sending commands to the system. The valves, regulators, and controllers in the actual production control parameters issue control instructions related to valve opening and controller opening adjustment to guide the collaborative optimization of the ultimate goal of copper smelting tail gas desulfurization and sulfuric acid resource recovery.

The communication service module mainly realizes the correlation between the above four modules, transfers the data in the database module to the data analysis module in real time, and sends the results obtained by the data analysis module to the visualization module, and finally obtains the results based on the data visualization. Applied to the equipment intelligent management module, it completes the information retrieval and association between various modules; including production data text, on-site images and other types of messages; and provides real-time communication functions to complete the real-time update function of images in the visualization module, and ensure The correctness, stability and security of communication ensure that the intelligent operation and maintenance module of the equipment issues correct instructions and completes the control; and records the operation logs of each module to record the platform for subsequent analysis and troubleshooting.

Claims (5)

1. A collaborative optimization method for tail gas recycling in the copper smelting process based on digital twins, which is characterized by: including the following steps: Step1: Retrieve the state parameters, process parameters, and evaluation indicators of desulfurization and acid production in the smelting process from the previous database, and perform data preprocessing, data prediction, and data mining to build a previous database; Step2: Collect real-time data, including status parameters and process parameters, which together form a real-time database; the status parameters include temperature, pressure, liquid level, and gas-liquid ratio in the furnace; the process parameters include the input concentrations of sulfur dioxide and hydrogen peroxide and Flow rate, concentration and flow rate of output sulfuric acid; Step3: The previous database and the real-time database constitute the digital twin dynamic model database. The data interoperability connection between the digital twin dynamic model database and the server, the server and the front-end interface composed of unity3D is realized through the TCP communication protocol, and the relevant data in the digital twin dynamic database are The visual graphics are displayed on the front-end interface; Step 4: Introduce the digital twin dynamic model database obtained in Step 3 as the digital twin model simulation signal into the digital twin model of desulfurization and acid production. The model is simulated through data. The simulation model uses nonlinearity based on the two-stage BPNLP neural network. The optimization is combined with an approximate global optimization algorithm to continuously optimize the process parameters corresponding to the optimal parameters obtained through the digital twin model. The most suitable production optimal engineering parameters are obtained based on the maximizing parameters of the sulfur dioxide desulfurization rate and sulfuric acid conversion rate. ; Step 5: Perform data mining similar aggregation, noise reduction, and missing value processing on the process parameters predicted to maximize the sulfur dioxide desulfurization rate and sulfuric acid conversion rate with previous databases, and then compare parameters with the real-time database to determine the current real-time data Which parameters still need to be improved; finally, the optimal parameter improvement plan is obtained by optimizing the global algorithm under real-time data; Step6: Visualize the optimal parameter data improvement plan through the TCP protocol, and display the visual graphics of the data on the human-machine interface of the unity3D front-end; Step7: In the optimal parameter improvement plan, the operator issues control instructions based on the obtained parameter improvement plan and manual experience; Step8: The control instructions are connected to Simulink's desulfurization and acid production simulation control model through the TCP protocol to form a control system. The control system is remotely connected to the actual industrial server, compares the optimal parameter values with the real-time data values, and evaluates the actual furnace body based on the difference. Carry out parameter control and control valves and regulators in production to guide actual production; Step9: Continue to repeat the steps from Step2 to Step5, collect real-time data, obtain the optimal control parameters through simulating the neural network digital twin model, and then compare the two data, and then feed back the control furnace process parameters, and repeat continuously for a long time. Ensure desulfurization efficiency and sulfuric acid production speed are maximized; The simulation model described in Step 4 is built using nonlinear optimization based on the two-stage BPNLP neural network combined with an approximate global optimization algorithm, which specifically includes the following steps: S1: Process the data in the data model library and specify the input samples transmitted from the database; according to the production process of desulfurization and acid production in copper smelting, when the input samples obey the normal distribution, select the mean and sum of the samples Standard deviation σ; S2: The BP neural network model is an error back propagation neural network and needs to be constructed in the following three aspects: input layer, hidden layer and output layer; the model includes neurons, weights, thresholds, layers and activation functions; according to the process of desulfurization and acid production Establish a BP neural network model based on the real-time state parameters and the recycling conversion rate of pollutant sulfur; S3: First, the parameters of the input layer, hidden layer and output layer are established. The five state parameters in the database are used as input variables of the BP natural network model, namely pressure, temperature in the tower, liquid level, gas-liquid ratio and sulfur gas. Flow input, the output layer is the conversion rate and desulfurization efficiency of sulfuric acid. Based on the hidden layer calculation of the BP neural network model, the hidden layer between the smelting parameters of the secondary desulfurization process and the collaborative optimization of sulfuric acid is set to two layers, and based on Actual process setting node; S4: The conversion function for the above two hidden layers can be defined as x _i = (x ⁱ ₁ , x ⁱ ₂ ,..., x ⁱ _p ), i=1,..., N y _i =(y ⁱ ₁ , y ⁱ ₂ ,..., y ⁱ _q ), i=1,...,N where p and q are the dimensions of the input and output variables, N is the number of samples; i is the hidden layer of the neuron, i=1 represents the input layer, x _p is the input variable, and y _p represents the output variable parameter, And the number of samples for each type of input variable in the above x _p is defined as Calculations are performed for each sample as follows: Because i is the hidden layer of the neuron, its output value needs to be calculated for each sample details as follows: Assume that for the kth layer, there are s neurons; l (l=1, 2,...H) is the number of hidden layers; where represents the weight corresponding to each of the above input variables, θ is the offset value,/> That is the bias value of the first hidden layer of the kth layer, p is a fixed value indicating the number of samples;/> is the sample node of the input variable in the kth layer; Similarly, with As input to the model, the output can be obtained/> In order to make the result close to the target, the output The error from the actual output is transmitted backward from the output layer to the next layer of the network; Represents the bias value of the i-th hidden layer in each pass. Each pass, the threshold and weight of the neuron are adjusted; this recursive process is repeated until the k-th hidden layer, so the function of the output layer can be obtained value/> S5: Perform secondary smoothing on the model output; define the generalization accuracy of the model before non-smoothing as the proportion of singular values in the predicted values of all samples; so that the generalization accuracy can be set to an acceptable range; S6: Set the output function of the BP neural network model to the maximum conversion rate of sulfuric acid and the maximum desulfurization efficiency in the smelting flue gas. The BPNLP neural network model optimized for pollution emission reduction is expressed as the following mathematical function: J represents the index function, which is used to represent the function index of the optimal desulfurization conversion rate and rate to be obtained. The function Min represents the mapping of the output from the two-stage BPNLP estimation results, where the input variable x is the desired index; where c is the linear transformation The vector is a 2×1 vector of sulfuric acid conversion rate and desulfurization efficiency; represents the estimation result predicted by the BP model, c ^T represents the linear transformation from the output estimation result of the BPNLP model with input variables to the expected index; To minimize the emission problem, the nonlinear programming is as follows: Ax≤b; f(x)≤0; L≤x≤U Among them, b, L, U are constant column vectors, f is a nonlinear function, x is a decision variable vector, which includes pressure, tower temperature, liquid level, gas-liquid ratio and sulfur dioxide flow rate. The incoming 5× 1 vector; A is a constant matrix, and the linear truncation function Ax represents the data after passing through the normalized domain and eliminating anomalies; S7: By fitting the function N(x), x can generate multiple local minima, generate multiple solutions and further affect the global optimization; use the random scatter method to select the initial value to adapt to the equal step changes of the initial value optimization; Confirm that the minimum value of the model simulation is selected from the optimization results in the optimal value; for the entire neural network model, the area of the variable to be optimized is gridded using the "grid and sample" method, combined with local search skills, to evolve The solution process within the framework is used to find the global optimal solution; S8: Smooth the output function; assuming that the output sample takes a certain dimension value of the output value, then the comparison of the model needs to be further expanded to simulate the output value, so that the output function smoothly takes the value in the domain and filters out the unreasonable value; S9: Finally complete the solution of the optimal parameters of the data twin model, and perform data visualization on the results and send them to unity3D to complete the subsequent content of human-computer interaction. 2. A collaborative optimization system based on digital twins for exhaust gas recycling in the copper smelting process, which is characterized by: including a database module, a data analysis module, a visualization module, an equipment management control module, and a communication service module; The database module is used to collect real-time data during the operation of desulfurization and sulfuric acid production in the smelting process, and obtain the data information of the tower in the entire process flow; including a geometric parameter data unit, an operating status parameter unit, a real-time process parameter unit, and a neural network. network model unit; The data analysis module obtains status data from the above-mentioned database module, preprocesses and analyzes the data to obtain the optimal parameter prediction values through the digital twin model, and achieves the purpose of collaborative optimization of resource recovery in the desulfurization and sulfuric acid production process; including algorithm library units , data preprocessing unit, feature extraction unit, data mining unit; The algorithm library unit is used to perform secondary smoothing on the output of the BP model, using its nonlinear mapping capabilities and network structure to improve prediction reliability, and tends to be better for large sample data such as copper smelting desulfurization and acid production processes. Fitting, combined with the global optimal search method, to complete the solution of the optimal parameters of the data twin model; The data preprocessing unit preprocesses the missing data and large deviation noise that often occur in real-time data collection in advance by changing the optimal parameters in historical data to avoid repeating a large amount of the same optimization on equipment data; The feature extraction unit uses the BP neural network model and uses the five state parameters in the database as input variables of the BP natural network model. The output layer is the conversion rate and desulfurization efficiency of sulfuric acid, as well as the settings of the hidden layer to perform prediction training on the data. and verify sample classification; The data mining unit discovers hidden patterns, association rules and trends in long-term production data. Before feature extraction, it conducts advance data analysis and mines the overall characteristics of the obtained production input information, thereby identifying the features in the data. trends, unusual patterns and correlations; The visualization module is mainly used to graphically display the data of the above two modules; it includes a fusion model unit, a result transmission unit, a data graphical unit, and a graphic dynamic display unit; The equipment intelligent management module is used to combine the above-mentioned module results, including the above-mentioned database module, to achieve equipment life cycle information acquisition, complete equipment real-time intelligent optimization operation and maintenance, and according to the optimal operating parameters and the output of the control system, Real-time and efficient control of actual production status parameters; including evaluation index unit, control module unit, intelligent decision-making unit and control instruction unit; The communication service module mainly realizes the correlation between the above four modules, transfers the data in the database module to the data analysis module in real time, and sends the results obtained by the data analysis module to the visualization module, and finally obtains the results based on the data visualization. Apply to the device intelligent management module to complete information retrieval and association between various modules. 3. The collaborative optimization system for exhaust gas recycling in the copper smelting process based on digital twins according to claim 2, characterized by: The geometric parameter data unit contains actual production field equipment data: including the relevant size, height, and production parameters of the production furnace body, and long-term operation data is stored and retrieved; The operating status parameter unit contains operating data during production and smelting, including information related to temperature, pressure, and liquid level in the production furnace, describing the status and performance of the system or equipment during operation; and providing information about the health status of the entire system or equipment, Information on work efficiency and resource utilization, and storage and retrieval of long-term operating data; The real-time process parameter unit contains updated pipeline inputs and production outputs that enter the reaction in real time, and is used to describe and evaluate the current status and performance of the system, equipment or process, and provide information about the health status, work efficiency, and resources of the entire system or equipment. Utilize situational information and store and retrieve long-term operating data; The neural network model unit records each operation log record of the neural network model under long-term operation, and records each relevant parameter weight, number of iterations, and output-input ratio. 4. The collaborative optimization system for tail gas recycling in the copper smelting process based on digital twins according to claim 2, characterized by: The fusion model unit model fusion is to combine the prediction results of multiple data analysis modules to obtain more accurate and reliable predictions; by combining the prediction results obtained in the model with different data and different dates, the efficiency of the entire system is improved. performance for higher accuracy and robustness; The result transmission unit regularly retrieves the optimal parameters obtained by the data analysis module, updates the obtained optimal parameters in real time, and transmits these output results to the back-end data analysis system in an appropriate form to support decision-making, Provide insights, drive subsequent visualization processes and conduct further analysis; The data graphical unit performs a visual graphical display of the result optimal parameter data obtained by the result transmission unit, and uses bar charts, line charts, scatter charts and pie charts to convey the relationship between the output parameters and process parameters. , trends, patterns; The graphic dynamic display unit mainly drives the entire module to continuously retrieve real-time data from the data analysis module and perform graphic updates for data visualization, so that the graphics can be updated with real-time or frequent changes in data, so as to present the latest information and trends in a timely manner. 5. The collaborative optimization system for exhaust gas recycling in the copper smelting process based on digital twins according to claim 2, characterized by: The evaluation index unit mainly compares the obtained optimal operating parameters with the real-time operating parameters, and determines which parameters need to be weighted according to the evaluation index. Under different evaluation indexes, high weights are required for regulation and measurement of the accuracy of the model. accuracy, performance and reliability to help staff understand model performance and make decisions; The control module unit is a liquid level-flow cascade control system for gas and liquid in and out, a liquid level-pressure cascade control system in the tower, and dual input and dual output classification of the temperature in the tower based on the changed values obtained from the comparison. Decoupled control, under the regulation of feedback, completes the control from digital twin simulation to actual production; The intelligent decision-making unit uses the existing data and information of the data analysis module as a basis to identify patterns, trends and laws through analysis, mining and integration of optimal parameters and actual on-site production data, thereby providing targeted and Reliable production decision-making recommendations; The control instruction unit refers to the key component responsible for generating, verifying and sending control instructions after intelligent decision-making, monitoring the system status and receiving feedback after the intelligent decision-making unit; completing the operation of the operator to control the actual production by sending commands to the system. The valves, regulators, and controllers that actually produce control parameters issue control instructions regarding valve opening and controller opening adjustments to guide the collaborative optimization of the final target copper smelting tail gas desulfurization and sulfuric acid resource recovery.