A method and system for predicting the grade of copper matte in an Isa furnace based on data registration and data cleaning.

Abstract

This invention discloses a method and system for predicting the grade of ISA furnace matte based on data registration and data cleaning, belonging to the field of intelligent control technology for ISA furnace smelting processes. The prediction method of this invention analyzes the smelting mechanism of the ISA furnace, selects process variables affecting the grade of matte, uses a data registration method to register second-level and hour-level data, uses a data cleaning method to filter out abnormal samples, uses a maximum correlation minimum redundancy algorithm to select the optimal process variables, and uses a long short-term memory network to predict the grade of matte and plot the predicted grade curve. The prediction method of this invention can consider the influence of second-level variables on the grade of matte, improving prediction accuracy and accurately guiding production. The prediction system of this invention, through the collaborative work of its modules, can accurately predict the grade of matte, thereby accurately adjusting the operating parameters of the smelting process and improving product quality.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control technology for the Isa furnace smelting process, specifically relating to a method and system for predicting the grade of Isa furnace matte based on data registration and data cleaning. Background Technology

[0002] Isa furnace smelting is one of the main pyrometallurgical copper smelting methods in China. During Isa furnace copper smelting, the matte grade is a crucial indicator of the smelting quality and a key basis for adjusting operating parameters. Currently, to address the issue of the inability to measure matte grade in real time during Isa furnace smelting, existing methods include statistical models, machine learning algorithms, and mathematical modeling. These methods typically utilize information such as ore chemical composition data and historical production data to predict matte grade through analysis and modeling. Common prediction models include multiple linear regression, support vector machines, and artificial neural networks.

[0003] However, due to the complexity of the smelting reaction mechanism, the coexistence of solid, liquid, and gas phases within the furnace, and the severe coupling between parameters, real-time detection of matte grade is impossible. Laboratory methods are required to measure the matte grade, which hinders timely guidance for production operations. Furthermore, the inconsistent time scales of different process variables during smelting mean that traditional methods only consider hourly process variables with the same time scale as the target variable, neglecting second-level process variables. This leads to poor model prediction accuracy and misjudgments. Additionally, actual industrial sites experience equipment failures, environmental changes, and other factors that introduce outliers into the collected data, affecting prediction accuracy and significantly impacting overall industrial efficiency. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for predicting the grade of matte in an Isa furnace based on data registration and cleaning. The specific operation includes the following steps: establishing a data registration model to fully consider all factors affecting matte grade during the smelting process, achieving registration between second-level and hourly data; establishing a data cleaning model to filter out abnormal samples, improving the overall quality of the dataset; using a variable filtering algorithm to remove redundant or low-correlation process variables related to matte grade, improving the computational efficiency of the prediction model; and finally, inputting the preprocessed data into a Long Short-Term Memory (LSTM) network to achieve accurate prediction of matte grade, thereby providing timely guidance for production operations. This invention can achieve time-scale registration between data, maximizing the consideration of process variables affecting matte grade, and can clean the collected data to improve data quality, ultimately achieving accurate prediction of matte grade during the Isa furnace smelting process.

[0005] To achieve the above objectives, the technical solution adopted in this invention is a method for predicting the grade of Isa furnace matte based on data registration and data cleaning. The specific process includes the following steps:

[0006] S1. Analyze the smelting mechanism of the Isa furnace, select process variables that affect the grade of matte in the industrial field, and collect relevant data.

[0007] S2. Based on the sampling frequency when collecting relevant data, divide the process variables selected in S1 into second-level variables X according to the sampling frequency. si (i = 1…n) and hourly variable X hj (j = 1…m); where the second-level variable X si (i = 1…n) includes: nozzle oxygen supply pipe pressure, nozzle oxygen supply pipe oxygen flow rate, nozzle air supply pipe pressure, nozzle air supply pipe flow rate, nozzle back pressure, nozzle top pressure, and Isa furnace molten pool temperature; hourly variable X hj (j=1…m) includes: diesel fuel addition, coal addition, slag-forming agent silicon material addition, flux silicon material Si content, flux silicon material Fe content, flux silicon material H2O content, copper concentrate addition, copper concentrate Cu content, copper concentrate Ca content, copper concentrate Fe content, copper concentrate Zn content, copper concentrate H2O content, copper concentrate S content, copper concentrate Si content, and copper concentrate As content.

[0008] S3. Constructing a data registration model: First, use a sliding window to select the data corresponding to each second-level variable within one smelting cycle to obtain a second-level data matrix M. Second, calculate the mean of each second-level variable within each smelting cycle. The mean values ​​calculated from different second-level data are concatenated with the hour-level data to complete data registration. The specific calculation formula is as follows:

[0009]

[0010] In the formula, q is the number of samples of second-level data collected within one smelting cycle, and x ip This represents the value of the i-th second-level variable at second p.

[0011] S4. Input the registered dataset into a Long Short-Term Memory (LSTM) network to obtain the predicted value of the matte grade for each smelting cycle, and calculate the error e between the actual and predicted values ​​of the matte grade for each smelting cycle. i The error sequence X = (e1, e2, ..., e) is obtained. n The error e i The specific expression is as follows:

[0012] e i =|yi -y i ′|(i=1...n)

[0013] In the formula y i y represents the true value of the grade of copper matte. i ′ represents the predicted value of the copper matte grade.

[0014] The error sequence X is fitted using a Gaussian distribution to obtain an error curve that conforms to a normal distribution; the allowable error range [a, b] is determined using the 3σ principle, i.e.

[0015] a=μ-3σ

[0016] b=μ+3σ

[0017] In the formula, μ represents the mean of the error sequence X, and σ represents the standard deviation of the error sequence X.

[0018] S5. Determine whether the prediction error of each sample is within the allowable error range [a,b]. If it is within the allowable error range [a,b], then consider this sample as a normal sample; if it is not within the allowable error range [a,b], then consider this sample as an abnormal sample and delete it. Repeat this process for the entire dataset to obtain the cleaned data.

[0019] S6. Use the Maximum Relevance Minimum Redundancy (MRMR) algorithm to screen variables, calculate the correlation coefficient between each variable and the correlation coefficient between each variable and the grade of copper matte, and remove variables with correlation coefficients greater than 0.8 between them and variables with correlation coefficients less than 0.3 with the grade of copper matte to obtain the optimal process variables.

[0020] S7. Input the optimal process variables after data registration, data cleaning, and variable screening into the Long Short-Term Memory (LSTM) network, output the accurate predicted value of the copper matte grade, and visualize the prediction results.

[0021] An Isa furnace copper matte grade prediction system based on data registration and data cleaning: The prediction system includes a data acquisition module, a data preprocessing module, a prediction module, and a communication module; the data acquisition module collects second-level data through sensors deployed on-site and hourly data obtained through offline testing and factory logs; the data preprocessing module registers the second-level data with the hourly data and cleans the registered data; the prediction module filters variables from the registered and cleaned data, selects key process variables to input into the prediction model, and finally outputs the predicted value of copper matte grade; the communication module is used for data transmission between the data acquisition module, the data preprocessing module, and the prediction module to ensure real-time data transmission between each module.

[0022] The data acquisition module includes a second-level data acquisition unit and an hour-level data acquisition unit. The second-level data acquisition unit is used to collect online data on the pressure of the oxygen supply pipe of the smelting equipment, the oxygen flow rate of the oxygen supply pipe of the smelting equipment, the pressure of the air supply pipe of the smelting equipment, the air flow rate of the air supply pipe of the smelting equipment, the back pressure of the smelting equipment, the top pressure of the smelting equipment, and the temperature of the molten pool of the Isa furnace in seconds. The hour-level data acquisition unit collects offline data on the following parameters through offline testing and log data reading: diesel fuel addition, coal addition, slag-forming agent silicon material addition, flux silicon material Si content, flux silicon material Fe content, flux silicon material H2O content, copper concentrate addition, copper concentrate Cu content, copper concentrate Ca content, copper concentrate Fe content, copper concentrate Zn content, copper concentrate H2O content, copper concentrate S content, copper concentrate Si content, and copper concentrate As content.

[0023] The data preprocessing module includes a data registration unit and a data cleaning unit. The data registration unit receives data from the data acquisition module, then concatenates the second-level data matrix with the hour-level data using a data registration model and normalizes the registered data. The data cleaning unit first calculates the error between the model's predicted value and the actual value, then obtains a reasonable error range through Gaussian distribution fitting, and finally deletes samples whose prediction error is outside the reasonable range. This module is used for the registration and cleaning of acquired data. Based on the aforementioned data acquisition module, second-level and hour-level data within different periods are acquired, and data preprocessing is achieved through data registration and data cleaning models.

[0024] The prediction module includes a feature selection unit and a model prediction unit. The feature selection unit uses the Maximum Relevance Minimum Redundancy (MRMR) algorithm to select key process variables affecting the grade of copper matte, reducing the computational load of the model and improving computational efficiency. The model prediction unit inputs the selected optimal feature set into a Long Short-Term Memory (LSTM) network and outputs the predicted result of the copper matte grade. This module is used for accurate prediction of the copper matte grade. The registered and cleaned data is obtained through the data preprocessing module mentioned above, and then the key influencing factors of the copper matte grade are selected through the feature selection algorithm and input into the prediction model to obtain an accurate prediction result of the copper matte grade.

[0025] The communication service module includes: a data demodulation unit, a data exchange unit, and a communication management unit. The data demodulation unit receives analog signals acquired from sensors and converts them into digital signals, providing data to the subsequent data preprocessing and prediction modules. The data exchange unit transmits the demodulated data, offline test data, and log data to the data preprocessing unit, and simultaneously receives data processed by the data preprocessing module and transmits it to the prediction module. The communication management unit establishes an end-to-end communication framework based on the real-time data collected by the data acquisition module, specifically including a data acquisition module client, a data preprocessing module client, and a prediction module client. This module facilitates communication and data exchange between modules within the system, enabling collaboration and data sharing between different modules, specifically for data transmission, status synchronization, and fault recovery, thereby achieving high efficiency, speed, and stability of the entire system.

[0026] Compared with the prior art, the present invention has at least the following beneficial effects:

[0027] The present invention discloses an Isa furnace matte grade prediction method based on data registration and data cleaning. It considers the influence of Isa furnace operating parameters on matte grade during the smelting process and uses the sliding window method to establish data registration models at different time scales, thus fully taking into account the influence of Isa furnace operating parameters on matte grade.

[0028] The present invention discloses an Isa furnace copper matte grade prediction method based on data registration and data cleaning. The method calculates the prediction error and obtains a reasonable error range through Gaussian distribution, thereby deleting abnormal samples to clean the collected data. This avoids the impact of abnormal samples on the accurate prediction of copper matte grade, thereby improving production efficiency and enterprise benefits.

[0029] The Isa furnace matte grade prediction system based on data registration and data cleaning described in this invention can obtain accurate matte grade prediction values ​​during the production process through the cooperation of various modules, correctly guide the adjustment of production operation parameters, reduce energy consumption and production costs, and has broad application prospects. Attached Figure Description

[0030] Figure 1 This is a detailed process of an Isa furnace copper matte grade prediction method based on data registration and data cleaning;

[0031] Figure 2 This invention is a registration model for second-level and hour-level data.

[0032] Figure 3 It is an Isa furnace copper matte grade prediction system based on data registration and data cleaning;

[0033] Figure 4 This is a graph showing the predicted grade of copper matte according to the present invention. Detailed Implementation

[0034] To better illustrate the purpose, technical solution, and advantages of the present invention, the present invention will be further described below in conjunction with specific embodiments.

[0035] Example 1

[0036] An experiment was conducted to predict the grade of copper matte using the ISA furnace smelting process in an industrial setting as an example. The smelting process is equipped with one ISA furnace, one oxygen production workshop, one blower, one main coal bunker, one concentrate bunker, and one diesel chamber. The temperature during smelting in the ISA furnace is between 1200-1300℃. The ISA furnace is a key piece of equipment for increasing the copper matte content in the molten metal.

[0037] First, an Isa furnace matte grade prediction system based on data registration and data cleaning is established, as shown in the attached figure. Figure 3 As shown, the system includes a data acquisition module, a data preprocessing module, a prediction module, and a communication module. The data acquisition module collects second-level data through sensors deployed on-site and hourly data through offline testing and factory logs. The data preprocessing module registers the second-level and hourly data and cleans the registered data. The prediction module filters variables from the registered and cleaned data, selects key process variables to input into the prediction model, and finally outputs the predicted value of the copper matte grade. The communication module is used for data transmission between the data acquisition module, the data preprocessing module, and the prediction module to ensure real-time data transmission between each module.

[0038] The data acquisition module includes a second-level data acquisition unit and an hour-level data acquisition unit. The second-level data acquisition unit is used to collect online data on the pressure of the oxygen supply pipe of the smelting equipment, the oxygen flow rate of the oxygen supply pipe of the smelting equipment, the pressure of the air supply pipe of the smelting equipment, the air flow rate of the air supply pipe of the smelting equipment, the back pressure of the smelting equipment, the top pressure of the smelting equipment, and the temperature of the molten pool of the Isa furnace in seconds. The hour-level data acquisition unit collects offline data on the following parameters through offline testing and log data reading: diesel fuel addition, coal addition, slag-forming agent silicon material addition, flux silicon material Si content, flux silicon material Fe content, flux silicon material H2O content, copper concentrate addition, copper concentrate Cu content, copper concentrate Ca content, copper concentrate Fe content, copper concentrate Zn content, copper concentrate H2O content, copper concentrate S content, copper concentrate Si content, and copper concentrate As content.

[0039] The data preprocessing module includes a data registration unit and a data cleaning unit. The data registration unit receives data from the data acquisition module, then concatenates the second-level data matrix with the hourly data using a data registration model and normalizes the registered data. The data cleaning unit first calculates the error between the model's predicted values ​​and the actual values, then obtains a reasonable error range through Gaussian distribution fitting, and finally deletes samples whose prediction errors fall outside the reasonable range. This module is used for data registration and cleaning. Based on the aforementioned data acquisition module, second-level and hourly data within different periods are collected, and data preprocessing is achieved through data registration and data cleaning models.

[0040] The prediction module includes a feature selection unit and a model prediction unit. The feature selection unit uses the Maximum Relevance Minimum Redundancy (MRMR) algorithm to filter out redundant features or features with low correlation to copper matte grade, reducing the computational load and improving computational efficiency. The model prediction unit inputs the selected optimal feature set into a Long Short-Term Memory (LSTM) network and outputs the predicted copper matte grade. This module is used for accurate prediction of copper matte grade. It obtains registered and cleaned data through the aforementioned data preprocessing module, then uses the feature selection algorithm to select key influencing factors of copper matte grade, inputting them into the prediction model to obtain accurate copper matte grade prediction results.

[0041] The communication service module includes: a data demodulation unit, a data exchange unit, and a communication management unit. The data demodulation unit receives analog signals acquired from sensors and converts them into digital signals, providing data to the subsequent data preprocessing and prediction modules. The data exchange unit transmits the demodulated data, offline test data, and log data to the data preprocessing unit, and simultaneously receives data processed by the data preprocessing module and transmits it to the prediction module. The communication management unit establishes an end-to-end communication framework based on real-time data collected by the data acquisition module, specifically including a data acquisition module client, a data preprocessing module client, and a prediction module client. This module facilitates communication and data exchange between modules within the system, enabling collaboration and data sharing between different modules, specifically for data transmission, status synchronization, and fault recovery, thereby achieving high efficiency, speed, and stability for the entire system.

[0042] Next, an Isa furnace matte grade prediction system based on data registration and data cleaning is connected to copper smelting equipment such as the Isa furnace, oxygen production workshop, table blower, main coal bunker, concentrate bunker, and diesel chamber, and calculations are performed.

[0043] Example 2

[0044] The method for predicting the grade of Isa furnace matte based on data registration and data cleaning is as follows: Figure 1 As shown, it includes the following steps:

[0045] S1. Based on the smelting mechanism of the Isa furnace and experimental data, process variables that affect the grade of matte in the industrial field are selected.

[0046] S2, and divide the selected process variables into second-level variables X according to the sampling frequency. si (i = 1…n) and hourly variable X hj (j = 1…m); where the second-level variable X si (i = 1…n) includes: nozzle oxygen supply pipe pressure, nozzle oxygen supply pipe oxygen flow rate, nozzle air supply pipe pressure, nozzle air supply pipe flow rate, nozzle back pressure, nozzle top pressure, and Isa furnace molten pool temperature; hourly variable X hj (j=1…m) includes: diesel fuel addition, coal addition, slag-forming agent silicon material addition, flux silicon material Si content, flux silicon material Fe content, flux silicon material H2O content, copper concentrate addition, copper concentrate Cu content, copper concentrate Ca content, copper concentrate Fe content, copper concentrate Zn content, copper concentrate H2O content, copper concentrate S content, copper concentrate Si content, and copper concentrate As content.

[0047] S3, see also Figure 2 The data registration model extracts a second-level data matrix M for each smelting cycle using a sliding window, where the size of the sliding window is set to the length of one smelting cycle. The mean of each second-level variable is then calculated sequentially within each smelting cycle. The mean value calculated from the second-level data is then concatenated with the hour-level data to complete data registration. The specific calculation formula is as follows:

[0048]

[0049] In the formula, q is the number of samples of second-level data collected within one smelting cycle, and x ip This represents the value of the i-th second-level variable at second p.

[0050] S4. Input the registered dataset into a Long Short-Term Memory (LSTM) network to obtain the predicted value of the matte grade for each smelting cycle, and calculate the error e between the actual and predicted values ​​of the matte grade for each smelting cycle. i The error sequence X = (e1, e2, ..., e) is obtained. n The error e i The specific expression is as follows:

[0051] e i =|y i -y i ′|(i=1...n)

[0052] In the formula yi y represents the true value of the grade of copper matte. i ′ represents the predicted value of the copper matte grade.

[0053] The error sequence X is fitted using a Gaussian distribution to obtain an error curve that conforms to a normal distribution. The reasonable range of error [a, b] is determined using the 3σ principle.

[0054] a=μ-3σ

[0055] b=μ+3σ

[0056] In the formula, μ represents the mean of the error sequence X, and σ represents the standard deviation of the error sequence X.

[0057] S5. Determine whether the prediction error of each sample is within the error range [a, b]. If it is within this range, consider this sample as a normal sample; otherwise, consider this sample as an abnormal sample and delete it. Repeat this process for the entire dataset to obtain the cleaned data.

[0058] S6. Use the Maximum Correlation Minimum Redundancy (MRMR) algorithm to calculate the correlation coefficient between each variable and the correlation coefficient between each variable and the grade of copper matte. Filter out variables with correlation coefficients greater than 0.8 between them and variables with correlation coefficients less than 0.3 with the grade of copper matte to obtain the optimal process variables.

[0059] S7. Input the optimal process variables into a Long Short-Term Memory (LSTM) network, output the accurate predicted value of the copper matte grade, and visualize the prediction results.

[0060] Finally, see Figure 4 Observing the prediction results of the copper matte grade of this invention, it can be seen that the predicted values ​​of copper matte grade obtained from the second-level and hour-level data collected during the Isa furnace smelting process are basically consistent with the actual values, verifying the reliability and accuracy of the method in terms of prediction precision. Furthermore, the calculated mean square error (MSE) is within 2%, indicating that the system can meet the actual production requirements of enterprises. Therefore, this system is of great significance in actual production, helping enterprises improve the accuracy of operating parameter adjustments and providing a strong guarantee for efficient production.

Claims

1. A method for predicting the grade of copper matte in an Isa furnace based on data registration and data cleaning, characterized in that: Includes the following steps: S1. Analyze the smelting mechanism of the Isa furnace, select process variables that affect the grade of copper matte in the industrial field, and collect relevant data; The process variables include: second-level variables and hour-level variables; among which, the second-level variables include: spray gun oxygen supply pipe pressure, spray gun oxygen supply pipe oxygen flow rate, spray gun air supply pipe pressure, spray gun air supply pipe flow rate, spray gun back pressure, spray gun top pressure, and Isa furnace molten pool temperature; the hour-level variables include: diesel fuel addition amount, coal addition amount, slag-forming agent silicon material addition amount, flux silicon material Si content, flux silicon material Fe content, flux silicon material H2O content, copper concentrate addition amount, copper concentrate Cu content, copper concentrate Ca content, copper concentrate Fe content, copper concentrate Zn content, copper concentrate H2O content, copper concentrate S content, copper concentrate Si content, and copper concentrate As content; S2. Based on the sampling frequency when collecting relevant data, the process variables selected in S1 are divided into second-level variables and hour-level variables. S3. Constructing a data registration model: First, use a sliding window to select the data corresponding to each second-level variable within one smelting cycle to obtain a second-level data matrix M. Second, calculate the mean of the data corresponding to each second-level variable within that cycle. Finally, the mean values ​​of the data at different second levels are concatenated with the hourly data to complete the registration of the second-level and hourly data within a period. S4. Input the registered dataset into the Long Short-Term Memory network to obtain the predicted value of the matte grade for each smelting cycle, and calculate the error between the actual and predicted values ​​of the matte grade for each smelting cycle. e i The error sequence for predicting the grade of copper matte, X = ( e 1 ,e 2 ,…,e n ); by fitting a Gaussian distribution and 3 σ The principle is to obtain the allowable error range. a,b ] Determine whether the prediction error for each sample is within the allowable error range; S5. If the error between the actual value and the predicted value of the matte grade is within the allowable error range... a,b If the sample is within the allowable error range, then the sample is considered a normal sample; if the error between the actual and predicted values ​​of the copper matte grade is within the allowable error range, then the sample is considered a normal sample. a,b If the sample is found to be outside the specified range, it is considered an anomalous sample and deleted. The entire dataset is then traversed to obtain the cleaned data. S6. Use the maximum correlation minimum redundancy algorithm to filter variables in the cleaned data, calculate the correlation coefficient between each variable and the correlation coefficient between each variable and the grade of copper matte, and filter out variables with correlation coefficients between each variable and the grade of copper matte that are greater than 0.8 and less than 0.

3. S7. Input the optimal process variables after data registration, data cleaning, and variable screening into the long short-term memory network, and output the predicted value of the copper matte grade.

2. The method for predicting the grade of Isa furnace matte based on data registration and data cleaning according to claim 1, characterized in that: The allowable error range in S4 [ a,b In the symbol ], 'a' and 'b' represent the following: In the formula μ Let X represent the mean of the error sequence. σ Let X represent the standard deviation of the error sequence X, and let X = ( e 1 ,e 2 ,…,e n ).

3. A copper matte grade prediction system for Isa furnaces based on data registration and data cleaning, used to implement the method as described in claim 1, characterized in that: The system includes a data acquisition module, a data preprocessing module, a prediction module, and a communication module. The data acquisition module collects second-level data through sensors deployed on-site and hourly data through offline testing and factory logs. The data preprocessing module registers the second-level and hourly data and cleans the registered data. The prediction module filters variables from the registered and cleaned data, selects key process variables to input into the prediction model, and finally outputs the predicted value of the copper matte grade. The communication module is used for data transmission between the data acquisition module, the data preprocessing module, and the prediction module to ensure real-time data transmission between each module.

4. The Isa furnace copper matte grade prediction system based on data registration and data cleaning according to claim 3, characterized in that: The data acquisition module includes a second-level data acquisition unit and an hour-level data acquisition unit. The second-level data acquisition unit is used to collect online data on the pressure of the oxygen supply pipe of the smelting equipment, the oxygen flow rate of the oxygen supply pipe of the smelting equipment, the pressure of the air supply pipe of the smelting equipment, the air flow rate of the air supply pipe of the smelting equipment, the back pressure of the smelting equipment, the top pressure of the smelting equipment, and the temperature of the molten pool of the Isa furnace in seconds. The hour-level data acquisition unit collects offline data on the following parameters through offline testing and log data reading: diesel fuel addition, coal addition, slag-forming agent silicon material addition, flux silicon material Si content, flux silicon material Fe content, flux silicon material H2O content, copper concentrate addition, copper concentrate Cu content, copper concentrate Ca content, copper concentrate Fe content, copper concentrate Zn content, copper concentrate H2O content, copper concentrate S content, copper concentrate Si content, and copper concentrate As content.

5. The Isa furnace copper matte grade prediction system based on data registration and data cleaning according to claim 3, characterized in that: The data preprocessing module includes a data registration unit and a data cleaning unit. The data registration unit receives data from the data acquisition module, and then uses a data registration model to concatenate and register the mean value of the second-level data with the hour-level data, and normalizes the registered data. The data cleaning unit first calculates the error between the model's predicted value and the actual value, then obtains a reasonable error range by fitting a Gaussian distribution, and finally deletes samples whose prediction error is outside the reasonable range.

6. The Isa furnace copper matte grade prediction system based on data registration and data cleaning according to claim 3, characterized in that: The prediction module includes a feature selection unit and a model prediction unit. The feature selection unit uses the maximum correlation minimum redundancy algorithm to screen out key process variables that affect the grade of copper matte. The model prediction unit inputs the screened key process variables into a long short-term memory network to obtain the predicted value of the copper matte grade.

7. The Isa furnace copper matte grade prediction system based on data registration and data cleaning according to claim 3, characterized in that: The communication module includes a data demodulation unit, a data exchange unit, and a communication management unit. The data demodulation unit receives analog signals acquired from sensors and converts them into digital signals, thereby providing data to the subsequent data preprocessing module and prediction module. The data exchange unit transmits the demodulated data, offline test data, and log data to the data preprocessing unit, and simultaneously receives the data processed by the data preprocessing module and transmits it to the prediction module. The communication management unit is used to establish an end-to-end communication framework, specifically including a data acquisition module client, a data preprocessing module client, and a prediction module client.

Images