CN114822723A - Method and device for predicting reaction endpoint of refining anode copper - Google Patents

Method and device for predicting reaction endpoint of refining anode copper Download PDF

Info

Publication number
CN114822723A
CN114822723A CN202210246653.2A CN202210246653A CN114822723A CN 114822723 A CN114822723 A CN 114822723A CN 202210246653 A CN202210246653 A CN 202210246653A CN 114822723 A CN114822723 A CN 114822723A
Authority
CN
China
Prior art keywords
prediction model
preset
test
model
reactant
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210246653.2A
Other languages
Chinese (zh)
Inventor
霍芊羽
吴金财
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China ENFI Engineering Corp
Original Assignee
China ENFI Engineering Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by China ENFI Engineering Corp filed Critical China ENFI Engineering Corp
Priority to CN202210246653.2A priority Critical patent/CN114822723A/en
Publication of CN114822723A publication Critical patent/CN114822723A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/70—Machine learning, data mining or chemometrics
    • C—CHEMISTRY; METALLURGY
    • C22—METALLURGY; FERROUS OR NON-FERROUS ALLOYS; TREATMENT OF ALLOYS OR NON-FERROUS METALS
    • C22B—PRODUCTION AND REFINING OF METALS; PRETREATMENT OF RAW MATERIALS
    • C22B15/00—Obtaining copper
    • C22B15/0026—Pyrometallurgy
    • C22B15/006—Pyrometallurgy working up of molten copper, e.g. refining
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/10—Analysis or design of chemical reactions, syntheses or processes

Abstract

The invention provides a method and a device for predicting a reaction endpoint of refining anode copper, which comprises the following steps: performing machine learning training on a preset basic model through a training data set of a preset characteristic parameter combination to obtain a primary prediction model of the consumption cumulant of the reactant; carrying out accuracy test on the primary prediction model; performing model stability test on the optimal prediction model, and selecting a prediction model of the consumption cumulant of the reactant; carrying out double-model nesting processing on the prediction model of the reactant consumption cumulant and a preset residual reaction time prediction model to obtain a reaction degree prediction model; and predicting the reaction end point of the anode copper through a reaction degree prediction model to obtain the residual reaction time of the anode copper from the reaction end point at any moment. The method can solve the problem that the reaction degree at any moment in the oxidation and reduction processes can not be judged when the oxidation-reduction end point of the refined anode copper is judged at present; the sampling and judgment of workers are required for many times.

Description

Method and device for predicting reaction endpoint of refining anode copper
Technical Field
The invention relates to the technical field of non-ferrous metal smelting, in particular to a method and a device for predicting a reaction endpoint of refining anode copper.
Background
The pyrometallurgical copper smelting is a main method for producing copper, accounts for 80-90% of the total copper yield, and 4 steps of matte smelting, matte blowing, pyrometallurgical refining and electrolytic refining are needed to produce high-grade copper products. After copper concentrate containing 18-30% of copper enters a flash smelting furnace, part of iron is removed after oxidation and slagging, and matte containing 30-70% of copper is produced. And (4) allowing the matte to enter a converting furnace for further oxidation and slagging to produce crude copper with the grade of about 96-98%. And (3) the crude copper enters an anode furnace, sulfur elements and impurity elements are further removed through pre-oxidation and oxidation, the oxygen elements are reduced and removed after the crude copper reaches the oxidation end point, the anode copper with the grade of 99.2-99.7% is finally obtained, and the anode copper plate is obtained after the crude copper enters a disc casting system. The anode copper plate is used as a final product of pyrometallurgical smelting, is conveyed to an electrolysis workshop to be used as a production raw material, and is electrolyzed to produce the cathode copper plate with the grade of more than 99.99 percent.
The refining of the anode copper as the last link of the pyrometallurgy plays an important link for connecting the electrolysis process, so the grade of the anode copper is the key to control the quality of the copper product. In the traditional production process, in order to ensure the grade of the anode copper and reduce the cost as much as possible, the oxidation degree and the reduction degree of the reaction in the anode furnace need to be continuously monitored, so that the oxidation end point and the reduction end point of the reaction in the furnace are accurately judged, and the grade, the sulfur content, the oxygen content and the impurity rate of the anode copper meet the requirements. At present, because the reaction degree in the refining anode furnace is related to a plurality of factors such as raw copper feeding amount, raw copper components, various gas consumption, furnace temperature, negative pressure and the like, the redox degree is difficult to define through a certain production index, the redox degree in the current refining anode furnace is mainly judged by a method of manually sampling, observing a sample and a section after knocking in order to monitor the reaction degree in the anode furnace, and the manual experience judgment is specifically carried out according to the details such as the form, the color, the sulfur bubbles, the sulfur filaments, the crystal form and the like of the section of the sample. Generally, the method only carries out full-element component analysis and detection on the anode copper at the final reduction end point, and the detection result is used as the quality of a subsequent monitoring product and is not used as a judgment basis for production operation.
At present, conditional smelters are actively developing modern plants that are intelligent and unmanned, and therefore it is necessary to develop a method for intelligently predicting the redox end point. On the one hand, the automation degree of the refining anode furnace can be improved, so that the refining anode furnace can be better transformed into intelligent type, on the other hand, the traditional experience data can be recorded, the production can be better guided, and the related manpower and material resources cost can be saved
The prior art scheme is as follows: the bronze group corporation provides a method for measuring SO in an anode furnace 2 A CO content device and a smelting method (201910023191.6), wherein the method comprises the steps of extracting flue gas in an anode furnace through a gas sampling device, and carrying out SO content analysis through a gas analyzer 2 And the content of CO is detected by the content of CO and SO 2 The oxidation end point and the reduction end point are judged according to the content of the (A). The Jiangxi Liwode science and technology limited company also provides an intelligent endpoint judgment system (202010170459.1) for copper smelting anode furnace oxidation and reduction, which judges an oxidation endpoint and a reduction endpoint through an endpoint judgment model by acquiring signals on a DCS and a flue gas concentration value detected by a flue gas analyzer.
In summary, the prior art mainly utilizes SO 2 And a CO on-line flue gas analysis method, which combines DCS system data to establish an end point judgment model and judges the oxidation reduction end point of the refined anode copper.
The problem with the above technique is the following disadvantages: the prediction accuracy of the on-line analysis of the flue gas can be influenced by the change of the negative pressure and the air flow in the furnace; the method mainly aims at the end points of oxidation and reduction, but gives no specific operation guidance suggestion on the reaction degree in the oxidation and reduction processes; the analysis principle is mainly based on SO 2 The model coefficient is a constant value in a certain range, so the performance of the model is not improved along with the increase of industrial production data.
Disclosure of Invention
In view of the above problems, the present invention provides a method and an apparatus for predicting the reaction endpoint of a copper anode in refining to solve the problem of using SO 2 The CO on-line flue gas analysis method is characterized in that when the redox end point of the refined anode copper is judged, the reaction degree at any moment in the oxidation and reduction processes cannot be judged; the sampling and judgment of workers are required for many times.
The invention provides a method for predicting a reaction endpoint of refining anode copper, which comprises the following steps:
taking the reactant consumption cumulant as a first target quantity, and performing machine learning training on a preset basic model through a training data set of a preset characteristic parameter combination to obtain a primary prediction model of the reactant consumption cumulant corresponding to the preset characteristic parameter combination; wherein the preset characteristic parameter combination consists of parameters related to the cumulative amount of consumption of the reactants;
carrying out accuracy test on the primary prediction model, and taking the primary prediction model with the accuracy score higher than a preset accuracy score threshold value as a preferred prediction model of the reactant consumption cumulant;
performing model stability test on the optimal prediction model, and selecting the optimal prediction model with the best stability as the prediction model of the reactant consumption cumulant;
carrying out double-model nesting processing on the prediction model of the reactant consumption cumulant and a preset residual reaction time prediction model, and enabling the output of the prediction model of the reactant consumption cumulant to be used as the input of the preset residual reaction time prediction model to obtain a reaction degree prediction model;
and predicting the reaction end point of the anode copper through the reaction degree prediction model to obtain the residual reaction time of the anode copper from the reaction end point at any moment.
In addition, preferably, before the performing machine learning training on a preset basic model by using the cumulative reactant consumption as a first target quantity through a training data set of a preset feature parameter combination to obtain a primary prediction model of the cumulative reactant consumption corresponding to the preset feature parameter combination, the method further includes:
collecting data of characteristic parameters related to the consumption cumulant of the reactant, and establishing a characteristic parameter data set;
carrying out data normalization processing on the data in the characteristic parameter data set to enable units or dimensions of the data to be uniform, and obtaining a processed characteristic parameter data set;
performing correlation calculation processing on the data of the characteristic parameters in the processing characteristic parameter data set, and combining the data of the characteristic parameters with similar properties to obtain a dimension reduction characteristic parameter data set;
performing parameter combination processing on the dimensionality reduction characteristic parameter data set according to a preset characteristic parameter combination rule to obtain a characteristic parameter combination data set related to the consumption cumulant of the reactant; wherein the feature parameter combination dataset comprises: a characteristic parameter combination, parameter data corresponding to the characteristic parameter combination, and real data of a reactant consumption cumulative amount corresponding to the parameter data;
and dividing the characteristic parameter combination data set into a training data set and a testing data set.
Further, it is preferable that the cumulative amount of consumption of the reactant is based on a reaction process of the anode copper, and includes a cumulative amount of consumption of the gas in the oxidation stage and a cumulative amount of consumption of the gas in the reduction stage.
In addition, the preferable scheme is that the machine learning training is one or a mixture of any several of a support vector machine, a neural network, a random forest, regression analysis and deep learning.
In addition, it is preferable that the performing the accuracy test on the primary prediction model, and using the primary prediction model with the accuracy score higher than a preset accuracy score threshold as the preferred prediction model of the cumulative amount of consumed reactant includes:
inputting the parameter data corresponding to the characteristic parameter combinations in the test data set into the primary prediction model to obtain predicted values of the reactant consumption cumulant, and acquiring real values of the reactant consumption cumulant corresponding to the parameter data from the test data set;
taking a square value of a correlation coefficient of the predicted value of the cumulative amount of consumed reactant and the true value of the cumulative amount of consumed reactant as an accuracy score of the primary prediction model;
and taking the primary prediction model with the certainty score higher than a preset accuracy score threshold value as a preferred prediction model of the cumulative amount of the consumed reactants.
In addition, preferably, the performing a model stability test on the preferred prediction model, and selecting the preferred prediction model with the best stability as the prediction model of the reactant consumption cumulative quantity includes:
acquiring a characteristic parameter combination corresponding to the optimal prediction model from the test data set, randomly selecting a group of parameter data from the parameter data corresponding to the characteristic parameter combination as test parameter data, and taking a true value of the reactant consumption cumulant corresponding to the test parameter data as a test true value;
inputting the test parameter data into the optimal prediction model according to a preset test frequency to obtain a test value set of the reactant consumption cumulant corresponding to the preset test frequency;
and selecting a preferred prediction model with the best stability as a prediction model of the accumulated consumption of the reactants according to the test value set and the test true value.
In addition, it is preferable that the selecting, according to the test value set and the test actual value, a preferred prediction model with the best stability as the prediction model of the reactant consumption cumulative quantity includes:
calculating the average value of the test values in the test value set as a test average value;
calculating the variance between the predicted value of each optimal prediction model and the test true value by adopting a variance calculation mode according to the test average value and the test true value;
and selecting a preferred prediction model with the minimum variance as a prediction model of the cumulative consumption of the reactants.
In addition, it is preferable that the training method of the prediction model of the preset residual reaction time includes:
collecting production data at a preset moment as a training data set of the residual reaction time; wherein the production data comprises an actual value of the cumulative amount of consumed reactants at a preset moment, an actual gas flow of the reactants and an actual reaction end point moment;
taking the residual reaction time at the preset moment as a target quantity, and performing machine learning iterative training on a basic model of the preset residual reaction time through a training data set of the residual reaction time to obtain a primary prediction model of the preset residual reaction time;
performing model accuracy test on the primary prediction model, and selecting the primary prediction model with the accuracy score higher than a preset time prediction accuracy score threshold value and preset residual reaction time as a preferred prediction model of the preset residual reaction time;
and performing stability test on the optimal prediction model of the preset residual reaction time, and selecting the optimal prediction model of the preset residual reaction time with the highest stability as the prediction model of the preset residual reaction time.
In addition, it is preferable that the production data is based on the reaction process of the anode copper, including production data of an oxidation stage and production data of a reduction stage; wherein the content of the first and second substances,
the production data of the oxidation stage includes: presetting an actual value of the accumulated amount of gas consumption for oxidation, an actual flow rate of gas for oxidation and an actual oxidation reaction end time;
the production data of the reduction stage comprises an actual value of the cumulative amount of gas consumption for reduction at a preset moment, the actual flow rate of gas for reduction and the actual reduction reaction end moment.
The invention provides a device for predicting reaction endpoint of refining anode copper, which comprises:
the model training module is used for performing machine learning training on a preset basic model through a training data set of a preset characteristic parameter combination by taking the reactant consumption cumulant as a first target quantity to obtain a primary prediction model of the reactant consumption cumulant corresponding to the preset characteristic parameter combination; wherein the preset characteristic parameter combination consists of parameters related to the cumulative amount of consumption of the reactants;
the accuracy testing module is used for testing the accuracy of the primary prediction model, and taking the primary prediction model with the accuracy score higher than a preset accuracy score threshold value as a preferred prediction model of the reactant consumption cumulant;
the stability testing module is used for carrying out model stability testing on the optimal prediction model, and selecting the optimal prediction model with the best stability as the prediction model of the reactant consumption cumulant;
the double-model nesting module is used for carrying out double-model nesting processing on the prediction model of the reactant consumption cumulant and a preset residual reaction time prediction model, so that the output of the prediction model of the reactant consumption cumulant is used as the input of the preset residual reaction time prediction model to obtain a reaction degree prediction model;
and the reaction end point prediction module is used for predicting the reaction end point of the anode copper through the reaction degree prediction model to obtain the residual reaction time of the anode copper from the reaction end point at any moment.
According to the technical scheme, the method and the device for predicting the reaction endpoint of the refining anode copper can comprehensively consider factors influencing the oxidation and reduction reactions of the refining anode copper through the training data set with the preset characteristic parameter combination, and analyze and use the factors in a characteristic parameter combination mode to reduce the dimensionality of data processing; the oxidation-reduction degree of the refined anode copper is predicted by adopting a dual-model nested mode, the cumulant of substances used for oxidation in the oxidation stage and the cumulant of substances used for reduction in the reduction stage are predicted, then the relative oxidation-reduction degree is given at any time in the oxidation and reduction stages, the time required by the residual oxidation and reduction is predicted, the visual guiding significance is provided for the actual production, and the sampling and judging times of workers are reduced; the established model adopts a machine learning algorithm, and along with the increase of industrial production data of the refined anode copper, the prediction precision of the model can be continuously improved, so that the utilization rate of industrial data is improved.
To the accomplishment of the foregoing and related ends, one or more aspects of the invention comprise the features hereinafter fully described. The following description and the annexed drawings set forth in detail certain illustrative aspects of the invention. These aspects are indicative, however, of but a few of the various ways in which the principles of the invention may be employed. Further, the present invention is intended to include all such aspects and their equivalents.
Drawings
Other objects and results of the present invention will become more apparent and more readily appreciated as the same becomes better understood by reference to the following description taken in conjunction with the accompanying drawings. In the drawings:
FIG. 1 is a schematic flow chart of a method for predicting the reaction endpoint of copper at a refining anode according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart of a method for predicting the reaction endpoint of the copper of the refining anode according to an embodiment of the present invention.
In the drawings, the same reference numerals indicate similar or corresponding features or functions.
Detailed Description
In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of one or more embodiments. It may be evident, however, that such embodiment(s) may be practiced without these specific details.
Currently utilized SO proposed for the foregoing 2 The CO on-line flue gas analysis method is characterized in that when the redox end point of the refined anode copper is judged, the reaction degree at any moment in the oxidation and reduction processes cannot be judged; the method and the device for predicting the reaction endpoint of the refined anode copper are provided.
Specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
In order to illustrate the method for predicting the reaction endpoint of the refining anode copper, provided by the invention, fig. 1 shows a flow of the method for predicting the reaction endpoint of the refining anode copper according to an embodiment of the invention; fig. 2 shows a specific flow of the method for predicting the reaction endpoint of the copper of the refining anode according to the embodiment of the invention.
As shown in figure 1 and figure 2 together, the method for predicting the reaction endpoint of the refining anode copper provided by the invention comprises the following steps:
s1, taking the reactant consumption cumulant as a first target quantity, and performing machine learning training on a preset basic model through a training data set of a preset characteristic parameter combination to obtain a primary prediction model of the reactant consumption cumulant corresponding to the preset characteristic parameter combination; wherein the preset characteristic parameter combination is composed of parameters related to the cumulative amount of consumption of the reactants.
Specifically, in order to establish a model between characteristic parameters such as the raw copper input amount, the raw copper component, the temperature and the flue gas amount and the oxidation reduction degree of the refining anode copper, the gas substance accumulation amount used for oxidation in the pre-oxidation stage and the gas substance accumulation amount used for reduction in the reduction stage are respectively selected as a first target amount of the model according to actual production experience and a metallurgical principle.
The target quantity selected by the primary predictive model of cumulative reactant consumption is the cumulative quantity of the reactant consumption for the oxidation and reduction stages, i.e., the cumulative quantity of the reactant consumption during the oxidation and reduction stages, wherein the substance includes, but is not limited to, oxygen, compressed air, natural gas. For example, the reactant for oxidation is oxygen or oxygen-enriched air, and the reactant for reduction is natural gas, coal gas, etc. The first target amount of substance is therefore a reactive substance for oxidation, reduction in actual use.
The method comprises the steps of performing machine learning training on a preset basic model by using a training data set of preset characteristic parameter combinations by adopting a machine learning algorithm, and obtaining a primary prediction model of the reactant consumption cumulant after iterative training of model parameters.
As a preferred aspect of the present invention, before performing machine learning training on a preset basic model through a training data set of a preset feature parameter combination with the reactant consumption cumulant as a first target quantity to obtain a primary prediction model of the reactant consumption cumulant corresponding to the preset feature parameter combination, the method further includes:
collecting data of characteristic parameters related to the consumption cumulant of the reactant, and establishing a characteristic parameter data set;
carrying out data normalization processing on the data in the characteristic parameter data set to unify units or dimensions of the data to obtain a processed characteristic parameter data set;
carrying out correlation calculation processing on the data of the characteristic parameters in the characteristic parameter data set, and combining the data of the characteristic parameters with similar properties to obtain a dimension-reduced characteristic parameter data set;
according to a preset characteristic parameter combination rule, performing parameter combination processing on the dimension reduction characteristic parameter data set to obtain a characteristic parameter combination data set related to the consumption cumulant of the reactant; wherein the feature parameter combination dataset comprises: the characteristic parameter combination, the parameter data corresponding to the characteristic parameter combination and the real data of the reactant consumption cumulant corresponding to the parameter data;
the feature parameter combination dataset is divided into a training dataset and a testing dataset.
Specifically, the data related to the oxidation and reduction degrees of the fire refining anode copper, that is, the data of the characteristic parameters related to the reactant consumption cumulant, may be factory DCS system data, data related to raw materials input into the refining anode furnace, various energy data required for anode copper production, operation data of related equipment in anode copper production, data related to products produced in the anode copper refining process, data related to heat in the anode copper production process, time parameters of each period in the anode copper refining process, production data at a certain moment in production, data related to qualified anode copper products, and data related to anode plate waste. And the sample number, the production date, the shift, the furnace number and the operator data corresponding to the characteristic data.
For example: the mass, composition, temperature of the blister copper; the consumption and flow of oxygen, natural gas, compressed air and nitrogen; equipment operation angle and system negative pressure; slag content, slag composition, flue gas content, flue gas composition; the temperature in the furnace, the temperature of the furnace shell, the real-time temperature of molten copper, the temperature of flue gas and the casting temperature; feeding, pre-oxidizing, reducing, casting, slagging-off and sampling; the residual oxidation time, the residual reduction time, the accumulated amount of oxidizing substances, the accumulated amount of reducing substances, the weight, the quantity and the components of the anode plate at a certain moment; weight, number, cause of scrap board. The process of refining the anode copper comprises the steps of starting feeding the raw copper material into the anode furnace and finishing casting the anode copper plate by the anode furnace.
And carrying out data normalization processing on the data in the characteristic parameter data set, and avoiding adverse effects of data units or dimensions on the training and prediction results of the subsequent model. And then, carrying out correlation calculation processing on the data of the characteristic parameters in the characteristic parameter data set, and combining the data of the characteristic parameters with similar properties, so as to avoid increasing the calculation difficulty due to overlarge data dimension in the model training calculation process. And then, carrying out parameter combination processing on the dimensionality reduction characteristic parameter data set according to a preset characteristic parameter combination rule to obtain a characteristic parameter combination data set related to the consumption cumulant of the reactant, wherein the preset characteristic parameter combination rule can be set according to actual experience and can also be randomly generated. The number of the parameters in the characteristic parameter combination data set can be one or more; such as the mass of the blister copper and the flow rate of the compressed air as a combination of characteristic quantities; the total amount of compressed air, the total amount of natural gas and the S content in the blister copper are another characteristic quantity combination. Through different combination modes, a large number of characteristic quantity combinations can be generated and used for selecting a subsequent model.
As a preferable mode of the present invention, the cumulative amount of consumed reactant is based on the reaction course of the anode copper, including the cumulative amount of consumed gas in the oxidation stage and the cumulative amount of consumed gas in the reduction stage.
Specifically, the refining of the anode copper includes two processes, i.e., an oxidation stage and a reduction stage, and in order to cover the two stages, the reactant consumption cumulative amount includes a gas consumption cumulative amount in the oxidation stage and a gas consumption cumulative amount in the reduction stage.
As a preferred scheme of the invention, the machine learning training is one or a mixture of any several of a support vector machine, a neural network, a random forest, regression analysis and deep learning.
Specifically, the machine learning training may be selected from one or a mixture of any of a support vector machine, a neural network, a random forest, regression analysis, and deep learning, as long as a primary prediction model of the reactant consumption cumulant can be obtained.
And S2, carrying out accuracy test on the primary prediction model, and taking the primary prediction model with the accuracy score higher than a preset accuracy score threshold value as a preferred prediction model of the reactant consumption cumulant.
Specifically, due to the fact that parameter data corresponding to different characteristic parameter combinations are used for training, a plurality of primary prediction models are obtained, and models with good prediction effects are selected, so that accuracy testing is conducted on the primary prediction models, and the primary prediction models with accuracy scores higher than a preset accuracy score threshold value are used as the optimal prediction models of the reactant consumption cumulant.
As a preferred embodiment of the present invention, the accuracy test of the primary prediction model, and the step of using the primary prediction model with an accuracy score higher than a preset accuracy score threshold as the preferred prediction model of the cumulative amount of consumed reactant includes:
inputting parameter data corresponding to the characteristic parameter combinations in the test data set into the primary prediction model to obtain predicted values of the reactant consumption cumulant, and acquiring real values of the reactant consumption cumulant corresponding to the parameter data from the test data set;
taking the square value of the correlation coefficient of the predicted value of the cumulative amount of the consumed reactant and the true value of the cumulative amount of the consumed reactant as the accuracy score of the primary prediction model;
and taking the primary prediction model with the certainty score higher than the preset accuracy score threshold value as a preferred prediction model of the cumulative amount of the consumed reactant.
Specifically, the accuracy test is performed on the primary prediction model by using parameter data corresponding to the characteristic parameter combinations in the test data set, a square value of a correlation coefficient between a predicted value of the reactant consumption cumulant and a true value of the reactant consumption cumulant is used as an accuracy score of the primary prediction model, when the predicted values are multiple, the accuracy score of the average machine model can be calculated, the primary prediction model with the accuracy score higher than a preset accuracy score threshold value is used as a preferred prediction model of the reactant consumption cumulant, and for example, the primary prediction model with the score higher than 90 is selected as the preferred prediction model.
And S3, performing model stability test on the optimal prediction model, and selecting the optimal prediction model with the best stability as a prediction model of the consumption accumulation of the reactants.
Specifically, in order to avoid the contingency of prediction of the preferred prediction model, the preferred prediction model needs to be subjected to a stability test through the same set of parameter data, and according to the volatility, the preferred prediction model with the best stability is selected as the prediction model of the reactant consumption accumulation amount.
As a preferred embodiment of the present invention, the model stability test is performed on the preferred prediction model, and the selecting the preferred prediction model with the best stability as the prediction model of the cumulative amount of consumed reactant includes:
acquiring a characteristic parameter combination corresponding to the optimal prediction model from the test data set, randomly selecting a group of parameter data from the parameter data corresponding to the characteristic parameter combination as test parameter data, and taking the true value of the reactant consumption cumulant corresponding to the test parameter data as a test true value;
inputting the test parameter data into an optimal prediction model according to the preset test times to obtain a test value set of the reactant consumption cumulant corresponding to the preset test times;
and selecting the optimal prediction model with the best stability as a prediction model of the cumulative consumption of the reactants according to the test value set and the test true value.
As a preferred scheme of the invention, the step of selecting the optimal prediction model with the best stability as the prediction model of the cumulative consumption amount of the reactants according to the test value set and the test actual value comprises the following steps:
calculating the average value of the test values in the test value set as a test average value;
calculating the variance between the predicted value of each optimal prediction model and the test true value by adopting a variance calculation mode according to the test average value and the test true value;
and selecting a preferred prediction model with the minimum variance as a prediction model of the cumulative amount of the consumed reactant.
Specifically, the stability of different optimal prediction models is compared by a method of taking an average value and a variance through multiple experiments. The smaller the variance, the better the stability.
And S4, performing double-model nesting processing on the prediction model of the reactant consumption cumulant and the preset residual reaction time prediction model, and enabling the output of the prediction model of the reactant consumption cumulant to be used as the input of the preset residual reaction time prediction model to obtain a reaction degree prediction model.
Specifically, the target amounts of the prediction model of the cumulative amounts of consumed reactants are cumulative amounts of consumed reactants for oxidation and reduction in the oxidation and reduction processes, respectively, rather than the degree of oxidation and reduction. Therefore, the output of the prediction model of the cumulative amount of consumed reactant is taken as the input of the preset residual reaction time prediction model, wherein the degree of oxidation reaction can be regarded as the ratio between the cumulative amount of the reaction gas used for oxidation at a certain time and the total amount of the reaction gas used for oxidation in the whole oxidation stage, and the degree of reduction can be regarded as the ratio between the cumulative amount of the reaction gas used for reduction at a certain time and the total amount of the reaction gas used for reduction in the whole reduction stage, and in order to further facilitate the actual worker operation, the degree of reaction progress is embodied by using the predicted residual reaction time (including the residual oxidation reaction time and the residual reduction reaction time) at a certain time.
As a preferred aspect of the present invention, a method for training a prediction model of a preset residual reaction time includes:
collecting production data at a preset moment as a training data set of the residual reaction time; the production data comprises an actual value of the reactant consumption cumulant at a preset moment, the gas flow of an actual reactant and an actual reaction end point moment;
taking the residual reaction time at the preset moment as a target quantity, and performing machine learning iterative training on a basic model of the preset residual reaction time through a training data set of the residual reaction time to obtain a primary prediction model of the preset residual reaction time;
performing model accuracy test on the primary prediction model, and selecting the primary prediction model with the accuracy score higher than a preset time prediction accuracy score threshold value and preset residual reaction time as a preferred prediction model of the preset residual reaction time;
and performing stability test on the prediction model of the preset residual reaction time, and selecting the optimal prediction model of the preset residual reaction time with the highest stability as the prediction model of the preset residual reaction time.
Specifically, the preset time can be any time of the refined anode copper in the oxidation stage and the reduction stage. Production data at this time include, but are not limited to: when the anode furnace is in an oxidation stage, the consumption cumulant of compressed air or other gases for oxidation reaction at the moment in actual production, the gas flow of the actually used oxidation reaction and the actual oxidation end point moment; when the anode furnace is in a reduction stage, the accumulated natural gas consumption at the moment in actual production or other accumulated gas consumption for reduction reaction, the actually used natural gas flow and the actual reduction end moment. The gas flow at this moment can be regarded as the gas flow at this stage, because the flow of the introduced gas is generally adjusted in a stepwise manner in the actual production, when the production plan is stable, the gas flow is generally not changed greatly.
As a preferred embodiment of the present invention, the production data is based on the reaction process of the anode copper, including the production data of the oxidation stage and the production data of the reduction stage; wherein the content of the first and second substances,
the production data for the oxidation stage included: presetting an actual value of the accumulated amount of gas consumption for oxidation, an actual flow rate of gas for oxidation and an actual oxidation reaction end time;
the production data of the reduction stage includes actual values of the cumulative amounts of gas consumption for reduction at preset timings, the flow rates of the gases actually used for reduction, and the actual reduction reaction end timings.
And S5, predicting the reaction end point of the anode copper through the reaction degree prediction model to obtain the residual reaction time of the anode copper from the reaction end point at any time.
Specifically, when the residual reaction time of the refining anode from the reaction end point at a certain moment is predicted, the actual parameter data of the characteristic parameter combination corresponding to the prediction model of the reactant consumption cumulant at the moment is input into the prediction model of the reactant consumption cumulant, the reactant consumption cumulant at the moment is predicted through the prediction model of the reactant consumption cumulant, and then the predicted reactant consumption cumulant is input into the preset residual reaction time prediction model, so that the residual reaction time of the refining anode copper from the reaction end point at the moment is obtained. Wherein, the reaction end point is the reaction end point of two stages, namely the reaction end point of an oxidation stage and the reaction end point of a reduction stage.
The invention provides a refining anode copper reaction end point prediction device, which comprises:
the model training module is used for performing machine learning training on a preset basic model through a training data set of a preset characteristic parameter combination by taking the reactant consumption cumulant as a first target quantity to obtain a primary prediction model of the reactant consumption cumulant corresponding to the preset characteristic parameter combination; wherein the preset characteristic parameter combination consists of parameters related to the cumulative amount of consumption of the reactants;
the accuracy testing module is used for testing the accuracy of the primary prediction model, and taking the primary prediction model with the accuracy score higher than a preset accuracy score threshold value as an optimal prediction model of the reactant consumption cumulant;
the stability testing module is used for carrying out model stability testing on the optimal prediction model, and selecting the optimal prediction model with the best stability as a prediction model of the consumption cumulant of the reactant;
the double-model nesting module is used for carrying out double-model nesting processing on the prediction model of the reactant consumption cumulant and the preset residual reaction time prediction model, so that the output of the prediction model of the reactant consumption cumulant is used as the input of the preset residual reaction time prediction model to obtain a reaction degree prediction model;
and the reaction end point prediction module is used for predicting the reaction end point of the anode copper through the reaction degree prediction model to obtain the residual reaction time of the anode copper from the reaction end point at any moment.
The following examples are presented to further illustrate the present invention so that those skilled in the art may better understand the advantages and features of the present invention.
Firstly, collecting the smelting data of a fire method anode plate of a certain factory, and collecting the industrial DCS data, the operation data, the raw copper input component data and the anode plate output data of each time, wherein the data are shown in tables 1-2.
Table 1 shows the industrial DCS data and the operation data for different production runs.
Figure BDA0003544988590000131
TABLE 1
Table 2 shows the composition data of the raw copper input for different production runs.
Figure BDA0003544988590000132
TABLE 2
And (4) carrying out data cleaning normalization on the data, and removing the influence of units and dimensions on numerical values. And (4) comprehensively considering the correlation calculation result of the parameter data of the characteristic parameters and worker experience, and screening characteristic parameter combinations for training and testing a primary prediction model of the reactant consumption cumulant.
Taking the data in table 1 as an example, it can be seen that the plant oxidation gas is compressed air and the reduction gas is natural gas. Therefore, the first target amount is the consumption accumulation amount of the compressed air in the pre-oxidation stage and the oxidation stage respectively, and the predicted value is recorded as Voi; natural gas in reduction stageThe total usage, predicted value is noted as Vci. Correspondingly, the total amount of compressed air in the actual oxidation stage is recorded as Vo; the total amount of natural gas in the actual reduction stage is denoted as Vc. According to the square value (R) of the correlation coefficient of Voi and Vo and Vc 2 ) As a criterion for evaluating the model. And establishing a model for the parameter data of the input characteristic parameter combination and the first target quantity through a support vector regression algorithm in machine learning.
The model finds suitable algorithm parameters in continuous iterative training or grid optimization to obtain a higher-grade model. And (4) carrying out model stability test on the model with the score higher than 90 to obtain the mean value and the variance of the model score. The most stable model is selected from the stable models whose variance approaches 0 as a prediction model for predicting the cumulative amount of consumption of the oxidizing and reducing reactants.
And recording and storing the predicted value of the prediction model of the reactant consumption cumulant for calling a prediction model of the preset residual reaction time of a subsequent model.
Table 3 is a data set of redox levels at a given time.
Figure BDA0003544988590000141
TABLE 3
Table 3 is an example of a training data set required to preset the residual reaction time prediction model. Recording a certain moment when the refined anode copper in a factory is in an oxidation stage, and recording the accumulated amount of compressed air for oxidation at the moment as Vot; the compressed air flow at this moment is recorded as Fo; the time from this point To the end of oxidation is denoted as Δ To. Recording a certain moment when the refined anode copper is in a reduction stage, and recording the accumulated amount of natural gas for reduction at the moment as Vct; the natural gas flow at that time is recorded as Fc; the time from this point to the end of the reduction was denoted as Δ Tc.
The prediction results Voi, Vci, Vot, Vct, Fo, Fc of the prediction models of Vot, Vct, Fo, Fc and the cumulative amount of consumed reactant are used as input, and the model 2 is established by an algorithm by using the Δ To and the Δ Tc as target amounts. TrainedModel 2, providing the time needed for predicting the residual oxidation, and recording as delta Toi; the predicted time required for residual reduction is given and noted as Δ Tci. Squared values of correlation coefficients (R) between Δ Toi and Δ To and between Δ Tci and Δ Tc 2 ) And when the score of the model is higher than 90, the prediction result of the preset residual reaction time prediction model has guiding significance for actual production.
According to the method and the device for predicting the reaction endpoint of the refining anode copper, provided by the invention, the factors influencing the oxidation and reduction reactions of the refining anode copper can be comprehensively considered through the training data set with the preset characteristic parameter combination, and the factors are analyzed and used in a characteristic parameter combination mode, so that the dimensionality of data processing is reduced; the oxidation-reduction degree of the refined anode copper is predicted by adopting a dual-model nested mode, the cumulant of substances used for oxidation in the oxidation stage and the cumulant of substances used for reduction in the reduction stage are predicted, then the relative oxidation-reduction degree is given at any time in the oxidation and reduction stages, the time required by the residual oxidation and reduction is predicted, the visual guiding significance is provided for the actual production, and the sampling and judging times of workers are reduced; the established model adopts a machine learning algorithm, and along with the increase of industrial production data of the refined anode copper, the prediction precision of the model can be continuously improved, so that the utilization rate of industrial data is improved.
The method and apparatus for predicting the reaction endpoint of a copper refining anode according to the present invention are described above by way of example with reference to the accompanying drawings. However, it will be appreciated by those skilled in the art that various modifications may be made to the method and apparatus for predicting the reaction endpoint of the copper anode in refining according to the present invention without departing from the scope of the present invention. Therefore, the scope of the present invention should be determined by the contents of the appended claims.

Claims (10)

1. A method for predicting a reaction endpoint of refining anode copper is characterized by comprising the following steps:
taking the reactant consumption cumulant as a first target quantity, and performing machine learning training on a preset basic model through a training data set of a preset characteristic parameter combination to obtain a primary prediction model of the reactant consumption cumulant corresponding to the preset characteristic parameter combination; wherein the preset characteristic parameter combination consists of parameters related to the cumulative amount of consumption of the reactants;
carrying out accuracy test on the primary prediction model, and taking the primary prediction model with the accuracy score higher than a preset accuracy score threshold value as a preferred prediction model of the reactant consumption cumulant;
performing model stability test on the optimal prediction model, and selecting the optimal prediction model with the best stability as the prediction model of the reactant consumption cumulant;
carrying out double-model nesting processing on the prediction model of the reactant consumption cumulant and a preset residual reaction time prediction model, and enabling the output of the prediction model of the reactant consumption cumulant to be used as the input of the preset residual reaction time prediction model to obtain a reaction degree prediction model;
and predicting the reaction end point of the anode copper through the reaction degree prediction model to obtain the residual reaction time of the anode copper from the reaction end point at any moment.
2. The method for predicting the reaction endpoint of the refining anode copper according to claim 1, wherein before the step of performing machine learning training on a preset basic model through a training data set of a preset characteristic parameter combination by using the reactant consumption cumulant as a first target quantity to obtain a primary prediction model of the reactant consumption cumulant corresponding to the preset characteristic parameter combination, the method further comprises:
collecting data of characteristic parameters related to the consumption cumulant of the reactant, and establishing a characteristic parameter data set;
carrying out data normalization processing on the data in the characteristic parameter data set to enable units or dimensions of the data to be uniform, and obtaining a processed characteristic parameter data set;
performing correlation calculation processing on the data of the characteristic parameters in the processing characteristic parameter data set, and combining the data of the characteristic parameters with similar properties to obtain a dimension reduction characteristic parameter data set;
performing parameter combination processing on the dimensionality reduction characteristic parameter data set according to a preset characteristic parameter combination rule to obtain a characteristic parameter combination data set related to the consumption cumulant of the reactant; wherein the feature parameter combination dataset comprises: a characteristic parameter combination, parameter data corresponding to the characteristic parameter combination, and real data of a reactant consumption cumulative amount corresponding to the parameter data;
and dividing the characteristic parameter combination data set into a training data set and a testing data set.
3. The method of predicting the reaction endpoint of refining anode copper according to claim 1, wherein the cumulative amount of consumed reactant is based on the reaction process of the anode copper, and includes a cumulative amount of consumed gas in an oxidation stage and a cumulative amount of consumed gas in a reduction stage.
4. The method for predicting the reaction endpoint of the refining anode copper according to claim 1, wherein the machine learning training is one or a mixture of any of a support vector machine, a neural network, a random forest, regression analysis and deep learning.
5. The method for predicting the reaction endpoint of the refining anode copper as recited in claim 2, wherein the step of performing an accuracy test on the primary prediction model, and the step of using the primary prediction model with an accuracy score higher than a preset accuracy score threshold as the preferred prediction model of the cumulative amount of consumed reactants comprises the steps of:
inputting the parameter data corresponding to the characteristic parameter combinations in the test data set into the primary prediction model to obtain predicted values of the reactant consumption cumulant, and acquiring real values of the reactant consumption cumulant corresponding to the parameter data from the test data set;
taking a square value of a correlation coefficient of a predicted value of the reactant consumption cumulative quantity and a true value of the reactant consumption cumulative quantity as an accuracy score of the primary prediction model;
and taking the primary prediction model with the certainty score higher than a preset accuracy score threshold value as a preferred prediction model of the cumulative amount of the consumed reactants.
6. The method for predicting the reaction endpoint of the refining anode copper according to claim 2, wherein the step of performing model stability test on the preferred prediction model, and the step of selecting the preferred prediction model with the best stability as the prediction model of the reactant consumption cumulant comprises the following steps:
acquiring a characteristic parameter combination corresponding to the optimal prediction model from the test data set, randomly selecting a group of parameter data from the parameter data corresponding to the characteristic parameter combination as test parameter data, and taking a true value of the reactant consumption cumulant corresponding to the test parameter data as a test true value;
inputting the test parameter data into the optimal prediction model according to a preset test frequency to obtain a test value set of the reactant consumption cumulant corresponding to the preset test frequency;
and selecting a preferred prediction model with the best stability as a prediction model of the accumulated consumption of the reactants according to the test value set and the test true value.
7. The method for predicting the reaction endpoint of the refining anode copper according to claim 6, wherein the step of selecting the optimal prediction model with the best stability as the prediction model of the cumulative consumption amount of the reactant according to the test value set and the test actual value comprises the following steps:
calculating the average value of the test values in the test value set as a test average value;
calculating the variance between the predicted value of each optimal prediction model and the test true value by adopting a variance calculation mode according to the test average value and the test true value;
and selecting a preferred prediction model with the minimum variance as a prediction model of the cumulative consumption of the reactants.
8. The method for predicting the reaction endpoint of the refining anode copper according to claim 1, wherein the method for training the preset residual reaction time prediction model comprises the following steps:
collecting production data at a preset moment as a training data set of the residual reaction time; wherein the production data comprises an actual value of the cumulative amount of consumed reactants at a preset moment, an actual gas flow of the reactants and an actual reaction end point moment;
taking the residual reaction time at the preset moment as a target quantity, and performing machine learning iterative training on a basic model of the preset residual reaction time through a training data set of the residual reaction time to obtain a primary prediction model of the preset residual reaction time;
performing model accuracy test on the primary prediction model, and selecting the primary prediction model with the accuracy score higher than a preset time prediction accuracy score threshold value and preset residual reaction time as a preferred prediction model of the preset residual reaction time;
and performing stability test on the optimal prediction model of the preset residual reaction time, and selecting the optimal prediction model of the preset residual reaction time with the highest stability as the prediction model of the preset residual reaction time.
9. The method of claim 8, wherein the production data is based on the reaction profile of the anode copper, including production data from an oxidation stage and production data from a reduction stage; wherein the content of the first and second substances,
the production data of the oxidation stage includes: presetting an actual value of the accumulated amount of gas consumption for oxidation, an actual flow rate of gas for oxidation and an actual oxidation reaction end time;
the production data of the reduction stage comprises an actual value of the cumulative amount of gas consumption for reduction at a preset moment, the actual flow rate of gas for reduction and the actual reduction reaction end moment.
10. A device for predicting the reaction endpoint of refining anode copper, which is characterized by comprising:
the model training module is used for performing machine learning training on a preset basic model through a training data set of a preset characteristic parameter combination by taking the reactant consumption cumulant as a first target quantity to obtain a primary prediction model of the reactant consumption cumulant corresponding to the preset characteristic parameter combination; wherein the preset characteristic parameter combination consists of parameters related to the cumulative amount of consumption of the reactants;
the accuracy testing module is used for testing the accuracy of the primary prediction model, and taking the primary prediction model with the accuracy score higher than a preset accuracy score threshold value as a preferred prediction model of the reactant consumption cumulant;
the stability testing module is used for carrying out model stability testing on the optimal prediction model, and selecting the optimal prediction model with the best stability as the prediction model of the reactant consumption cumulant;
the double-model nesting module is used for carrying out double-model nesting processing on the prediction model of the reactant consumption cumulant and a preset residual reaction time prediction model, so that the output of the prediction model of the reactant consumption cumulant is used as the input of the preset residual reaction time prediction model to obtain a reaction degree prediction model;
and the reaction end point prediction module is used for predicting the reaction end point of the anode copper through the reaction degree prediction model to obtain the residual reaction time of the anode copper from the reaction end point at any moment.
CN202210246653.2A 2022-03-14 2022-03-14 Method and device for predicting reaction endpoint of refining anode copper Pending CN114822723A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210246653.2A CN114822723A (en) 2022-03-14 2022-03-14 Method and device for predicting reaction endpoint of refining anode copper

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210246653.2A CN114822723A (en) 2022-03-14 2022-03-14 Method and device for predicting reaction endpoint of refining anode copper

Publications (1)

Publication Number Publication Date
CN114822723A true CN114822723A (en) 2022-07-29

Family

ID=82529133

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210246653.2A Pending CN114822723A (en) 2022-03-14 2022-03-14 Method and device for predicting reaction endpoint of refining anode copper

Country Status (1)

Country Link
CN (1) CN114822723A (en)

Similar Documents

Publication Publication Date Title
CN104630410B (en) A kind of pneumatic steelmaking quality real-time dynamic forecast method based on data parsing
US9996074B2 (en) System and predictive modeling method for smelting process control based on multi-source information with heterogeneous relatedness
CN113343576B (en) Prediction method of calcium yield in calcium treatment process based on deep neural network
CN110991916A (en) Casting blank quality judgment system and method
CN110989510A (en) Hot galvanizing product full-process quality control and grade automatic judgment system
CN111893237A (en) Method for predicting carbon content and temperature of molten pool of converter steelmaking in whole process in real time
US20150337404A1 (en) Method and device for predicting, controlling and/or regulating steelworks processes
JP2007052739A (en) Method and device for generating model, method and device for predicting state, and method and system for adjusting state
CN107630122B (en) A kind of RH dynamic decarburization optimization method based on flue gas analysis
CN114822723A (en) Method and device for predicting reaction endpoint of refining anode copper
Ruuska et al. Mass-balance based multivariate modelling of basic oxygen furnace used in steel industry
CN107563656B (en) Method for evaluating running state of gold hydrometallurgy cyaniding leaching process
CN113420426A (en) Method, device, medium and computer equipment for determining forward running condition of blast furnace
CN116911057A (en) Real-time prediction method for temperature of molten pool in converter steelmaking converting process
WO2014167982A1 (en) Correction device, correction method and steel refining method
Manolescu et al. Net carbon consumption in aluminum electrolysis: impact of anode properties and reduction cell-operation variables
Cui et al. Soft sensing of alumina concentration in aluminum electrolysis industry based on deep belief network
CN114185976A (en) Visual intelligent perception platform of blast furnace
Sorsa et al. Data-driven multivariate analysis of basic oxygen furnace used in steel industry
CN116976148B (en) Method and system for monitoring ion content change in copper electrolysis process
CN217483223U (en) Rotary kiln production control system and rotary kiln
CN115537494B (en) Method and system for monitoring water leakage of converter flue
CN116579670B (en) Economic benefit calculation and feasibility assessment method for recycling thermal refining slag
Schaaf et al. Real-time hybrid predictive modeling of the Teniente Converter
CN111291501B (en) Intelligent end point judging system for oxidation reduction of copper smelting anode furnace

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination