Copper converter blowing endpoint dynamic prediction system and method based on multi-source feature fusion

Abstract

The present application relates to copper smelting technical field, disclose a copper converter blowing end dynamic prediction system and method based on multi-source feature fusion, the method comprises the following steps: collecting operation, detection, process and auxiliary monitoring multi-source data in blowing process, the data is preprocessed by removing outliers, missing data filling and deviation correction, based on the end point index screening fusion feature construction set, the end point working condition expansion label sample is reproduced after the copper making period, the working condition is classified according to the raw material batch and the matching feature set is matched, and the dynamic prediction result, the key influence parameter and its correlation degree are output. The present application realizes the accurate dynamic prediction of blowing end by means of multi-source data integration optimization, feature screening fusion, working condition adaptation and other means, synchronously determines the parameter correlation logic, and provides reliable technical support for copper converter blowing end judgment.

Description

Technical Field

[0001] This invention relates to the field of copper smelting technology, and more specifically, to a dynamic prediction system and method for the endpoint of copper converter blowing based on multi-source feature fusion. Background Technology

[0002] With the large-scale and refined development of the copper smelting industry, copper converter blowing, as the core process connecting matte smelting and blister copper production, directly affects product quality, production efficiency, and production safety due to the accuracy of its endpoint prediction. Currently, the operating conditions at the blowing site are complex, and multi-source data is easily interfered with by high temperatures and dust. Furthermore, operating conditions such as raw material batches, furnace age, and tuyeres condition are highly variable. Existing prediction technologies still have shortcomings in data processing, feature mining, and operating condition adaptation, making it difficult to achieve stable and reliable dynamic endpoint prediction.

[0003] Traditional methods for predicting the endpoint of copper converter blowing have many shortcomings. Data collection lacks systematic integration, and multi-source data often exhibits anomalies, missing data, and measurement biases. The preprocessing process is not standardized. Feature selection is not targeted enough and fails to effectively mine parameters that are strongly correlated with the endpoint indicators. The logic of feature fusion is lacking. The sample size near the endpoint is scarce, the working condition classification and feature set are poorly matched, and the key influencing parameters and their correlation with the prediction results cannot be clearly identified.

[0004] Therefore, it is necessary to design a dynamic prediction method for the end point of copper converter blowing based on multi-source feature fusion to solve the problems of insufficient prediction reliability caused by non-standard data preprocessing, unreasonable feature selection and fusion, insufficient samples, poor adaptability to operating conditions, and unclear correlation of key parameters in the existing technology. Summary of the Invention

[0005] In view of this, the present invention proposes a dynamic prediction system and method for the end point of copper converter blowing based on multi-source feature fusion, aiming to solve at least one of the problems in the background art.

[0006] In one aspect, this invention proposes a dynamic prediction method for the endpoint of copper converter blowing based on multi-source feature fusion, comprising: Collect multi-source data related to the blowing state during the copper converter blowing process. The multi-source data includes at least operating parameters, detection parameters, process parameters, and auxiliary monitoring parameters. The multi-source data is preprocessed, including anomaly screening and removal, missing data imputation, and measurement bias correction. Based on the refining endpoint index, feature filtering is performed on the preprocessed multi-source data, and the filtered feature parameters are fused to construct a feature set for endpoint prediction. Based on historical smelting data, the final working conditions of the later stage of copper production were reproduced. The data samples of the adjacent intervals of the final point were expanded, and the working conditions of the expanded samples were marked. The blowing conditions are classified according to raw material batch, furnace age stage and tuyere operating status, and the feature set is matched and adapted with the corresponding condition type. Based on the adapted feature set, the dynamic prediction result of the copper converter blowing endpoint is output, and the key influencing parameters of the prediction result and their correlation degree are given simultaneously.

[0007] Furthermore, the collection of multi-source data related to the blowing state during the copper converter blowing process includes: The corresponding data acquisition device is used to synchronously collect the operating parameters, detection parameters, process parameters, and auxiliary monitoring parameters; The operating parameters include blast flow rate, blast pressure, oxygen enrichment ratio, and adjustment timing data for each parameter; the detection parameters include furnace surface temperature, furnace gas outlet temperature, flue gas composition content, and detection node data for each parameter; the process parameters include total amount of matte injected, flux addition batches, and input time node data for each material; the auxiliary monitoring parameters include furnace vibration amplitude, melt level height, and continuous monitoring data for each parameter. All collected multi-source data are archived sequentially according to the blowing timeline to form a raw data set with time sequence identifiers.

[0008] Furthermore, the preprocessing of the multi-source data includes: The data in the original dataset is screened point by point to identify abnormal data that exceeds the fluctuation range of preset parameters, and the abnormal data is removed from the original dataset. For missing data segments in the dataset, valid data under the same blowing stage and working conditions are extracted, and time-series interpolation is used to fill in the missing data. Historical qualified data within the same raw material batch and furnace age range are selected as the correction benchmark. The deviation difference between the current data and the benchmark data is calculated, and the deviation is corrected based on the deviation difference. All data that have been screened, removed, filled in missing data, and corrected for biases are integrated to form a preprocessed dataset.

[0009] Furthermore, the step of screening the data in the original dataset point by point to identify abnormal data that exceeds the fluctuation range of preset parameters, and removing the abnormal data from the original dataset, includes: For different types of parameters, corresponding normal fluctuation ranges are set, and the normal fluctuation ranges are determined based on the statistical range of historical qualified data under the same working conditions. The data in the original dataset are checked point by point to determine whether each data point is within the normal fluctuation range of the corresponding parameter; For data points that exceed the normal fluctuation range, if the data points before and after them all exceed the normal fluctuation range of the corresponding parameter, then the data point is confirmed as abnormal data and removed from the original dataset. Record the location, value, and reason for removing abnormal data to form an abnormal data processing log.

[0010] Furthermore, the step of correcting the deviation of the current data based on the deviation difference includes: When the deviation between the current parameter data and the benchmark value is greater than the preset deviation threshold, the correction amount is calculated based on the deviation difference, and the correction operation is performed. Specifically, the deviation difference is compared with a preset adjustment coefficient mapping table, and the adjustment coefficient of the deviation difference is determined based on the comparison result; The product of the adjustment coefficient and the deviation difference is used as the correction amount; When the deviation between the current parameter data and the benchmark value is less than or equal to the preset deviation threshold, no correction operation is performed on the current parameter data, and the original data is directly retained.

[0011] Furthermore, based on the refining endpoint index, feature filtering is performed on the preprocessed multi-source data, and the filtered feature parameters are fused to construct a feature set for endpoint prediction, including: The core indicators for the final stage of the blowing process are determined, including crude copper grade, melt sulfur content, and melt temperature. Calculate the correlation between the values ​​of each parameter in the preprocessed dataset and the core indicator, and select parameters with a correlation higher than a preset threshold as valid feature parameters; The effective characteristic parameters are divided into operation characteristic group, detection characteristic group, process characteristic group and auxiliary monitoring characteristic group according to parameter type, and a corresponding mapping relationship between each type of parameter and the blowing endpoint index is established. The parameter data of each feature group are associated and spliced ​​according to the blowing time sequence, the blowing stage corresponding to each feature parameter is marked, and a feature set for endpoint prediction is formed. Redundancy is checked and redundant parameters are removed from the parameter data in the feature set.

[0012] Furthermore, the process of reproducing the final operating conditions of the later stage of copper production based on historical blowing data involves expanding the data samples in the adjacent intervals of the final stage and labeling the operating conditions of the expanded samples, including: Retrieve the final operating condition parameters of the later stage of copper production in the historical blowing process. The operating condition parameters include a combination of historical operating parameters, historical process parameters and historical detection parameters. The current reproduction conditions are set according to the historical endpoint operating conditions parameters to simulate the endpoint reaction environment in the later stage of copper production, and the operating conditions reproduction operation is performed; the reproduction conditions include material input, blowing operation standards and detection frequency; During the reproduction process, data of each parameter are collected simultaneously to form a single set of samples in the vicinity of the endpoint. The number of samples in the vicinity of the endpoint is expanded by repeatedly changing the raw material batch and furnace age status to reproduce the working conditions. For each expanded sample group, offline sampling and testing are performed simultaneously to obtain the measured values ​​of the endpoint indicators corresponding to the sample. The working condition of the sample is labeled according to the measured values ​​to form a labeled sample set.

[0013] Furthermore, the classification of blowing conditions according to raw material batch, furnace age stage, and tuyere operating status, and the matching and adaptation processing of the feature set with the corresponding condition type, includes: Based on the composition ratio, Fe / S ratio and mining source information of the copper matte raw material, different batches of raw materials are classified into different operating conditions. Based on the amount of furnace lining wear and the duration of furnace lining use, different furnace age stages are classified into different operating condition types. Based on the results of airflow and cleanliness tests at the air ventilator, different air ventilator operating conditions are classified. Extract the core parameters from the feature set and compare them with the feature parameter templates for each working condition type; When the matching degree between the core parameters of the feature set and the template of a certain working condition reaches the preset matching degree threshold, the feature set is associated with the working condition and the adaptation process is performed. When the matching degree between the core parameters of the feature set and all working condition type templates does not reach the preset matching degree threshold, the core parameters of the feature set are re-extracted, the accuracy of parameter extraction is checked, and the comparison operation is performed again until the matching is completed or it is confirmed that there is no matching working condition type and recorded. For the matched feature set, the time sequence correspondence and numerical correlation logic of the parameters are adjusted to make the feature set match the blowing reaction law of the corresponding working condition type, thus completing the adaptation process.

[0014] Furthermore, the dynamic prediction result of the copper converter blowing endpoint is output based on the feature set after adaptation processing, and key parameters that have a significant impact on the prediction result and their correlation degree are given simultaneously, including: The adapted feature set data is extracted at preset time intervals, combined with the labeled sample set, to generate dynamic prediction results of the copper converter blowing endpoint, and output in real time according to the blowing progress. In the statistical prediction process, the influence weight of each feature parameter on the prediction result is determined, and parameters with weights higher than the preset weight standard value are selected as key influencing parameters. Record the numerical correspondence between each key influencing parameter and the prediction results, clarify the correlation logic between each key parameter, and simultaneously output a list of key parameters and an explanation of the degree of correlation.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention collects and preprocesses multi-source relevant data during the copper converter blowing process, constructs a predictive feature set based on the blowing endpoint index by screening and integrating features, reproduces the endpoint conditions in the later stage of copper production to expand the labeled sample, classifies the conditions according to raw material batch, furnace age stage and tuyere operating status and matches them with the appropriate feature set, outputs dynamic prediction results and simultaneously provides key influencing parameters and their correlation degree, can systematically optimize data utilization, accurately mine effective features, adapt to diverse blowing conditions, enrich the foundation of endpoint prediction samples, clarify the core influencing factors of the prediction results, provide comprehensive and reliable technical support for the judgment of the copper converter blowing endpoint, and ensure the orderly progress of the blowing process.

[0016] On the other hand, this invention also proposes a dynamic prediction system for the endpoint of copper converter blowing based on multi-source feature fusion, which is applied to the dynamic prediction method for the endpoint of copper converter blowing based on multi-source feature fusion, including: Data acquisition module, data processing module, feature fusion module, working condition reproduction module, feature matching module, and prediction module; The data acquisition module is used to collect multi-source data related to the blowing state during the copper converter blowing process; The data processing module is used to preprocess the multi-source data; The feature fusion module is used to perform feature filtering on the preprocessed multi-source data based on the refining endpoint index, and to fuse the filtered feature parameters to construct a feature set for endpoint prediction. The working condition reproduction module is used to reproduce the final working condition of the later stage of copper production based on historical blowing data, expand the data sample of the adjacent interval of the final point, and mark the working condition status of the expanded sample. The feature matching module is used to classify the blowing conditions according to the raw material batch, furnace age stage and tuyere operating status, and to match and adapt the feature set with the corresponding operating condition type. The prediction module is used to output dynamic prediction results of the copper converter blowing endpoint based on the adapted feature set, and simultaneously provide the key influencing parameters of the prediction results and their correlation degree.

[0017] It is understandable that the above-mentioned dynamic prediction system and method for the end point of copper converter blowing based on multi-source feature fusion have the same beneficial effects, and will not be elaborated here. Attached Figure Description

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 The flowchart illustrates the dynamic prediction method for the end point of copper converter blowing based on multi-source feature fusion, as provided in this embodiment of the invention.

[0019] Figure 2 This is a functional block diagram of a dynamic prediction system for the end point of copper converter blowing based on multi-source feature fusion, provided in an embodiment of the present invention. Detailed Implementation

[0020] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] See Figure 1 As shown, in some embodiments of this application, this embodiment provides a method for dynamic prediction of the endpoint of copper converter blowing based on multi-source feature fusion, including the following steps: S100: Collect multi-source data related to the blowing state during the copper converter blowing process. The multi-source data includes at least operating parameters, detection parameters, process parameters, and auxiliary monitoring parameters.

[0022] S200: Preprocessing of multi-source data, including outlier screening and removal, missing data imputation, and measurement bias correction.

[0023] S300: Based on the refining endpoint index, feature filtering is performed on the preprocessed multi-source data, and the filtered feature parameters are fused to construct a feature set for endpoint prediction.

[0024] S400: Based on historical smelting data, reproduce the final working conditions of the later stage of copper production, expand the data samples in the adjacent interval of the final point, and label the working conditions of the expanded samples.

[0025] S500: Classify the blowing conditions according to raw material batch, furnace age stage and tuyere operating status, and match and adapt the feature set with the corresponding condition type.

[0026] S600: Based on the adapted feature set, output the dynamic prediction result of the copper converter blowing endpoint, and simultaneously give the key influencing parameters of the prediction result and their correlation degree.

[0027] It is understandable that the comprehensive acquisition of multi-source data in step S100 of this embodiment ensures the completeness of the description of the blowing process state. The operating parameters directly reflect the current operating intensity, the detection parameters reflect the real-time chemical reaction state in the furnace, the process parameters provide a basic background for prediction, and the auxiliary monitoring parameters enrich the perception dimensions of the complex blowing environment. S200 effectively eliminates "noise" data caused by sensor failure, transient interference, etc. through abnormal data screening and elimination; missing data filling ensures the continuity of the data sequence and avoids the impact of information gaps on subsequent analysis; and measurement deviation correction ensures the consistency and reliability of the data from various sources in terms of dimensions and accuracy by comparing with standard values ​​or establishing a correction model, laying a solid data foundation for subsequent feature extraction and model training. In S300, feature screening based on the blowing endpoint index ensures that the final feature set closely revolves around the prediction target, avoids interference from irrelevant features, and improves prediction efficiency and accuracy. Feature fusion organically combines different types and dimensions of screened features to form a comprehensive feature space that can fully characterize the blowing state. S400 reproduces the endpoint conditions and expands the sample by using historical data, especially strengthening the critical interval near the endpoint. This effectively solves the problem of insufficient model training data caused by the scarcity of endpoint samples and the complex and variable operating conditions in actual production. The labeling of operating conditions provides the model with clear learning objectives. S500 classifies the blowing conditions according to raw material batch, furnace age, and tuyere operating status and matches them with feature sets, which is one of the core links to achieve dynamic prediction. Different raw material batches have different compositions and properties, different furnace ages affect the heat transfer efficiency and reaction space of the furnace, and the tuyere operating status directly affects the contact efficiency between oxygen and melt and the stirring intensity. These factors significantly affect the blowing process and endpoint judgment. Through classification and adaptation, the prediction model can call or adjust the corresponding parameters and strategies for different specific operating conditions, achieving accurate prediction "tailored to the situation". Ultimately, while outputting dynamic prediction results, the S600 simultaneously provides key influencing parameters and their correlation, which not only provides operators with an intuitive basis for endpoint judgment, but also helps process engineers to deeply understand the influence mechanism of various factors on the blowing endpoint. This provides data support and decision-making reference for optimizing operations and improving processes, demonstrating the guiding value of this method in practical production applications.

[0028] Specifically, multi-source data related to the blowing state during the copper converter blowing process are collected, including: The corresponding data acquisition device is used to synchronously collect operating parameters, detection parameters, process parameters and auxiliary monitoring parameters; The operating parameters include blast flow rate, blast pressure, oxygen enrichment ratio, and adjustment timing data for each parameter; the detection parameters include furnace surface temperature, furnace gas outlet temperature, flue gas composition content, and detection node data for each parameter; the process parameters include total amount of matte injected, flux addition batches, and input time node data for each material; and the auxiliary monitoring parameters include furnace vibration amplitude, melt level height, and continuous monitoring data for each parameter. All collected multi-source data are archived sequentially according to the blowing timeline to form a raw data set with time sequence identifiers.

[0029] In this embodiment, the flue gas composition content includes the volume fraction of flue gas components such as CO, CO2, and O2.

[0030] Understandably, by synchronously collecting the adjustment sequence of operating parameters, the node data of detection parameters, the input time points of process parameters, and the continuous data of auxiliary monitoring parameters, a multi-dimensional data matrix covering the entire copper converter blowing process can be constructed. This provides a complete and coherent data foundation for subsequent feature fusion and endpoint prediction.

[0031] Specifically, preprocessing of multi-source data includes: The data in the original dataset is screened point by point to identify abnormal data that exceeds the fluctuation range of preset parameters, and the abnormal data is removed from the original dataset. For missing data segments in the dataset, valid data under the same blowing stage and working conditions are extracted, and time-series interpolation is used to fill in the missing data. Historical qualified data within the same raw material batch and furnace age range are selected as the calibration benchmark. The deviation difference between the current data and the benchmark data is calculated, and the deviation is corrected based on the deviation difference. All data that have been screened, removed, filled in missing data, and corrected for biases are integrated to form a preprocessed dataset.

[0032] In this embodiment, the preset parameter fluctuation range is set comprehensively based on the process allowable threshold and historical normal operation data distribution characteristics of different parameters. For example, the fluctuation range of the blower flow rate can be set to ±15% of the normal average, and the fluctuation range of the furnace surface temperature can be set according to the heat conduction characteristics of different furnace age stages to ensure the accuracy of abnormal data identification. For missing data segments, if the missing duration is short, such as no more than 5 minutes, linear interpolation is used to fill in the missing data; if the missing duration is long, such as more than 5 minutes but no more than 30 minutes, trend extrapolation interpolation is performed in combination with the changing trend of historical data in the same period; for a very small number of severe missing data exceeding 30 minutes, it is necessary to combine manual operation records and the correlation of adjacent parameters for comprehensive judgment before filling in the missing data to restore the true change pattern of the data to the greatest extent. During the deviation correction process, for the detection parameters such as flue gas composition content, it is also necessary to periodically calibrate using standard gases and incorporate the calibration results into the dynamic update mechanism of the calibration benchmark to ensure the accuracy of data measurement during long-term operation.

[0033] Specifically, the data in the original dataset is screened point by point to identify abnormal data that exceeds the fluctuation range of preset parameters, and the abnormal data is removed from the original dataset, including: For different types of parameters, corresponding normal fluctuation ranges are set, and the normal fluctuation ranges are determined based on the statistical range of historical qualified data under the same working conditions. The data in the original dataset are checked point by point to determine whether each data point is within the normal fluctuation range of the corresponding parameter; For data points that exceed the normal fluctuation range, if the data points before and after them all exceed the normal fluctuation range of the corresponding parameter, then the data point is confirmed as abnormal data and removed from the original dataset. Record the location, value, and reason for removing abnormal data to form an abnormal data processing log.

[0034] It is worth noting that for data points that exceed the fluctuation range, if all adjacent data points before and after the data point exceed the normal fluctuation range of the corresponding parameter, it indicates that the data point is not an instantaneous normal fluctuation and is confirmed as abnormal data. The data point will be completely removed from the original dataset. If the data point only exceeds the fluctuation range itself and all adjacent data points before and after it are within the normal fluctuation range, the data point is determined to be an instantaneous normal fluctuation and is not included in the category of abnormal data. The data point will be retained in the original dataset and proceed to the subsequent processing flow.

[0035] It is understandable that this embodiment achieves accurate identification and removal of abnormal data by setting a dedicated normal fluctuation range according to parameter type and checking point by point in combination with adjacent data verification. This effectively avoids the problems of poor adaptability of a single fluctuation range and easy accidental deletion of a single data point. It preserves the effective information in the original data to the greatest extent and eliminates data noise caused by environmental interference, equipment failure and other factors. At the same time, by constructing an abnormal data processing log, the data processing process is made traceable, which facilitates subsequent verification of the rationality of data processing, optimization of parameter fluctuation range settings and investigation of potential equipment operation problems.

[0036] Specifically, deviation correction is performed on the current data based on the deviation difference, including: When the deviation between the current parameter data and the benchmark value is greater than the preset deviation threshold, the correction amount is calculated based on the deviation difference, and the correction operation is performed. Among them, the deviation difference is compared with the preset adjustment coefficient mapping table, and the adjustment coefficient of the deviation difference is determined based on the comparison result; The product of the adjustment factor and the deviation difference is used as the correction amount; When the deviation between the current parameter data and the benchmark value is less than or equal to the preset deviation threshold, no correction operation is performed on the current parameter data, and the original data is directly retained.

[0037] It is worth noting that the current parameter data is the data after anomaly removal and missing data filling.

[0038] It is worth noting that the preset adjustment coefficient mapping table records the correspondence between the deviation difference of different intervals and the corresponding adjustment coefficient. The adjustment coefficient is set according to the magnitude of the deviation difference, the parameter type, and the weight of the impact on the blowing endpoint index.

[0039] It is worth noting that the impact of historical deviation data on the prediction results of the blowing endpoint is statistically analyzed to determine the maximum deviation value that will not significantly affect the prediction accuracy. This value is then set as the preset deviation threshold. Different types of parameters (operational parameters, detection parameters, etc.) can correspond to different preset deviation thresholds.

[0040] It is worth noting that when performing a correction operation on the current parameter data based on the correction amount, the correction amount is subtracted from the current data. If the deviation difference is positive, the corrected data will fall back to the reference value; if the deviation difference is negative, the corrected data will rise to the reference value. After obtaining the corrected parameter data, it is included in the preprocessed data set.

[0041] It is worth noting that for all data that has undergone deviation correction, the original value, deviation difference, matching adjustment coefficient, calculated correction amount, and corrected value are recorded; for data that has not undergone correction, the original value, deviation difference, and reason for not correcting (deviation difference ≤ preset deviation threshold) are recorded; all recorded information is organized and archived according to the blowing time sequence to form a deviation correction processing log, which facilitates subsequent data traceability, adjustment coefficient optimization, and process rationality verification.

[0042] Understandably, this embodiment distinguishes deviation types by using threshold judgment, achieving targeted and reasonable deviation correction, avoiding blind correction of minor deviations and omission of correction of significant deviations, effectively reducing invalid operations and correction errors; by matching corresponding adjustment coefficients through a preset adjustment coefficient mapping table and calculating the correction amount in combination with the deviation difference, it achieves accurate correction of data with different degrees of deviation, ensuring that the corrected data conforms to the normal operating condition value rules, and eliminating systematic deviations caused by factors such as sensor aging and environmental interference; by recording correction process information, a traceable processing log is formed, providing important reference for subsequent process verification, parameter optimization, and equipment maintenance.

[0043] Specifically, based on the refining endpoint index, feature filtering is performed on the preprocessed multi-source data, and the filtered feature parameters are fused to construct a feature set for endpoint prediction, including: Determine the core indicators for the final stage of the blowing process. These core indicators include crude copper grade, melt sulfur content, and melt temperature. Calculate the correlation between the values ​​of each parameter and the core indicators in the preprocessed dataset, and select parameters with a correlation higher than a preset threshold as effective feature parameters; The effective characteristic parameters are divided into operation characteristic group, detection characteristic group, process characteristic group and auxiliary monitoring characteristic group according to parameter type, and a corresponding mapping relationship between each type of parameter and the blowing endpoint index is established. The parameter data of each feature group are associated and spliced ​​according to the blowing time sequence, the blowing stage corresponding to each feature parameter is marked, and a feature set for endpoint prediction is formed. Redundancy is checked and redundant parameters are removed from the parameter data in the feature set.

[0044] It is worth noting that, in conjunction with the process objectives and quality requirements of copper converter blowing, the core indicators of the blowing endpoint are clearly defined, and the value range and testing standards of each core indicator are sorted out, so as to provide a benchmark for subsequent numerical correlation calculations.

[0045] It is worth noting that the operation parameters, detection parameters, process parameters, and auxiliary monitoring parameters in the preprocessed dataset are extracted, and the numerical correlation between each parameter and the core indicators of the blowing endpoint is calculated one by one. The numerical correspondence between a single parameter and the changes in crude copper grade, melt sulfur content, and melt temperature is quantified. A preset threshold is retrieved, and the numerical correlation of each parameter is compared with the preset threshold. Parameters with a correlation higher than the preset threshold are selected as valid feature parameters and included in the subsequent processing. Parameters with a correlation lower than or equal to the preset threshold are judged as redundant and invalid parameters and are removed, and are not included in the subsequent feature fusion process.

[0046] It is worth noting that the operation characteristic group includes effective parameters related to the control of the blowing operation, such as blast flow rate and oxygen enrichment ratio; the detection characteristic group includes effective parameters related to the detection of the blowing status, such as furnace temperature and flue gas SO2 concentration; the process characteristic group includes effective parameters related to material input, such as matte injection amount and flux addition batch; and the auxiliary monitoring characteristic group includes effective parameters related to auxiliary monitoring of the furnace and melt status, such as furnace vibration amplitude and melt level. After grouping, a mapping relationship between each characteristic group and the core indicators of the blowing endpoint is established, clarifying that each individual parameter or combination of parameters in each group corresponds to one or more indicators affecting crude copper grade, melt sulfur content, and melt temperature, forming a clear parameter-indicator correlation logic.

[0047] It is worth noting that, following a complete timeline from the start of feeding to the end of blowing, the effective characteristic parameter data of the four characteristic groups are correlated and spliced ​​to ensure that the parameter data of different characteristic groups correspond one-to-one with the time nodes; the blowing stage corresponding to each effective characteristic parameter is marked, such as the feeding stage, slag formation stage, copper formation stage, and the near-end stage, clarifying the timing of the parameter's role in the entire blowing process; all effective characteristic parameters after splicing and marking are integrated to form an initial characteristic set; the parameter data in the initial characteristic set are checked for redundancy, and redundant parameters with duplicate associations, such as two parameters having completely consistent influence logic on the same end-point indicator and having no independent contribution, are identified and eliminated; the characteristic set after eliminating redundancy is used as the final characteristic set for end-point prediction, and the parameter composition, grouping information, temporal correlation relationship, and stage marking information of the characteristic set are recorded, archived, and synchronously input into subsequent working condition reproduction and sample expansion steps.

[0048] Understandably, this embodiment provides a precise benchmark for feature selection by clearly defining the core indicators of the blowing endpoint, ensuring that the selected effective feature parameters directly address the core needs of endpoint prediction. Through numerical correlation calculation and threshold comparison, it achieves accurate selection of effective feature parameters, eliminating redundant parameters with low correlation and reducing interference from invalid information in subsequent prediction processes. By dividing feature groups by type and establishing parameter-indicator mapping relationships, it clarifies the functional dimensions and correlation logic of each parameter, avoiding prediction logic distortion caused by chaotic parameter stacking. By splicing and fusing according to the blowing time sequence and labeling stage information, the feature set can completely and orderly represent the dynamic state of the entire blowing process, adapting to the temporal change characteristics of the blowing process. Redundancy checks further simplify the feature set structure, reducing the computational complexity of subsequent prediction processes and improving prediction efficiency.

[0049] Specifically, based on historical smelting data, the final operating conditions of the later stage of copper production are reproduced. The data samples in the adjacent intervals of the final stage are expanded, and the operating conditions of the expanded samples are labeled, including: Retrieve the final operating condition parameters of the later stage of copper production in the historical blowing process. The operating condition parameters include a combination of historical operating parameters, historical process parameters and historical detection parameters. Set the current reproduction conditions according to the historical endpoint operating parameters, simulate the endpoint reaction environment in the later stage of copper production, and execute the operating condition reproduction operation; the reproduction conditions include material input, blowing operation standards and detection frequency; During the reproduction process, data of each parameter are collected simultaneously to form a single set of samples in the vicinity of the endpoint. The number of samples in the vicinity of the endpoint is expanded by repeatedly changing the raw material batch and furnace age status to reproduce the working conditions. For each expanded sample group, offline sampling and testing are performed simultaneously to obtain the measured values ​​of the endpoint indicators corresponding to the sample. The working conditions of the sample are labeled according to the measured values ​​to form a labeled sample set.

[0050] It is worth noting that, from the historical blowing database, historical qualified blowing data that are consistent with the current blowing process type and equipment model are selected, with a focus on extracting the final operating parameters in the later stages of the copper production period. The operating parameters are combined parameters, specifically including historical operating parameters, such as the blast flow rate, oxygen enrichment ratio and adjustment sequence before the endpoint; historical process parameters, such as the matte injection balance and flux addition batches before the endpoint; and historical detection parameters, such as the furnace temperature and flue gas SO2 concentration change data before the endpoint.

[0051] It is worth noting that the focus was on collecting parameter data near the endpoint of the reproducibility test. This data was then organized according to the blowing sequence to form a single set of samples for the interval near the endpoint. Through repeated reproducibility tests and data collection, the number of samples for the interval near the endpoint was gradually increased, resulting in multiple sets of sample data covering different raw material batches and different furnace ages.

[0052] It is worth noting that when labeling the operating conditions of corresponding samples based on measured values, if the measured values ​​meet the qualification standards of the blowing endpoint, such as the crude copper grade, melt sulfur content, and melt temperature all being within the preset qualification range, then the sample is labeled as "endpoint qualified state"; if the measured values ​​do not meet the qualification standards, then the sample is labeled as "endpoint unqualified state," and the specific indicators and values ​​that were not qualified are also noted. All labeled samples are classified and organized according to raw material batch and furnace age to form a complete set of labeled samples, which are then archived and simultaneously input into subsequent operating condition classification and adaptation steps.

[0053] Understandably, this embodiment retrieves historical blowing data to reproduce the final conditions of the late copper-making period, effectively solving the problems of scarce samples and high collection difficulty in the vicinity of the final point in actual production, and providing a sufficient sample basis for final point prediction. By setting reproduction conditions that match historical conditions, the authenticity of the reproduced conditions is ensured, and the collected samples can accurately characterize the parameter change patterns before the final point, improving the reliability of the samples. By changing the raw material batch and furnace age multiple times to expand the sample size, the diversity of the samples is enriched, enabling the sample set to cover different production scenarios, avoiding the problem of poor prediction adaptability caused by a single sample, and ensuring the adaptability of the subsequent prediction model to various blowing conditions. Through offline sampling and testing and precise status labeling, the correspondence between samples and final point states is clarified, providing a clear basis for the prediction model to learn the correlation logic of "parameter change - final point state", helping to improve the accuracy of dynamic prediction of the final point.

[0054] Specifically, the blowing conditions are classified according to raw material batch, furnace age stage, and tuyere operating status, and the feature set is matched and adapted to the corresponding condition type, including: Based on the composition ratio, Fe / S ratio and mining source information of the copper matte raw material, different batches of raw materials are classified into different operating conditions. Based on the amount of furnace lining wear and the duration of furnace lining use, different furnace age stages are classified into different operating condition types. Based on the results of airflow and cleanliness tests at the air ventilator, different air ventilator operating conditions are classified. Extract the core parameters from the feature set and compare them with the feature parameter templates for each working condition type; When the matching degree between the core parameters of the feature set and the template of a certain working condition reaches the preset matching degree threshold, the feature set is associated with the working condition and the adaptation process is performed. When the matching degree between the core parameters of the feature set and all working condition type templates does not reach the preset matching degree threshold, the core parameters of the feature set are re-extracted, the accuracy of parameter extraction is checked, and the comparison operation is performed again until the matching is completed or it is confirmed that there is no matching working condition type and recorded. For the matched feature set, the time sequence correspondence and numerical correlation logic of the parameters are adjusted to make the feature set match the blowing reaction law of the corresponding working condition type, thus completing the adaptation process.

[0055] It is worth noting that for each combination (or single type) of the above three types of working conditions, typical core parameters are extracted from the historical qualified blowing data under that working condition type, and organized into four categories: operation, detection, process, and auxiliary monitoring, to construct a feature parameter template exclusive to each working condition type. From the feature set constructed in the early stage, parameters that can characterize the core features of the current blowing state are extracted as core parameters. In the extraction process, parameters with high correlation with the blowing endpoint index are selected first to ensure the representativeness of the core parameters.

[0056] It is worth noting that, for the established feature sets and operating condition types, the timing correspondence of parameters in the feature sets is adjusted (e.g., adjusting the timing weights of parameters at different blowing stages to adapt to the parameter change rhythm under the operating condition) in conjunction with the blowing reaction patterns of that operating condition type (e.g., reaction intensity corresponding to raw material batch operating condition type, temperature stability corresponding to furnace age stage operating condition type, reaction uniformity corresponding to tuyere operating state operating condition type). At the same time, the numerical correlation logic of parameters within the feature sets is optimized (e.g., strengthening the correlation weight between a certain type of parameter and the endpoint index to adapt to the reaction characteristics under the operating condition). After the adjustment is completed, the fit between the feature sets and the blowing reaction patterns of the corresponding operating condition types is verified to ensure that the adjusted feature sets can accurately represent the blowing state under the operating condition, thus completing the adaptation process.

[0057] Understandably, this embodiment categorizes blowing conditions according to three core variables: raw material batch, furnace age stage, and tuyere operating status. This achieves precise breakdown of complex blowing conditions, covering key scenarios affecting the blowing reaction law and avoiding prediction bias caused by single-condition adaptation logic. By constructing dedicated feature parameter templates, a clear reference is provided for feature set matching, improving the accuracy of matching judgments. Through core parameter extraction and multi-template comparison, combined with differentiated matching operations (standard association, non-standard re-extraction and verification), misjudgment and omissions in matching are effectively avoided, ensuring the effectiveness of matching feature sets with operating condition types. By specifically adjusting the temporal correspondence and numerical association logic of feature sets, a high degree of fit between feature sets and corresponding operating condition reaction laws is achieved, eliminating adaptation bias and ensuring that feature sets can accurately represent the blowing state under specific operating conditions.

[0058] Specifically, based on the adapted feature set, the dynamic prediction results of the copper converter blowing endpoint are output, and the key parameters that have a significant impact on the prediction results and their correlation degree are given simultaneously, including: The adapted feature set data is extracted at preset time intervals, combined with the labeled sample set, to generate dynamic prediction results of the copper converter blowing endpoint, and output in real time according to the blowing progress. In the statistical prediction process, the influence weight of each feature parameter on the prediction result is determined, and parameters with weights higher than the preset weight standard value are selected as key influencing parameters. Record the numerical correspondence between each key influencing parameter and the prediction results, clarify the correlation logic between each key parameter, and simultaneously output a list of key parameters and an explanation of the degree of correlation.

[0059] It is worth noting that the preset weight standard value is set based on the statistical requirements of historical parameter influence weights and the accuracy of prediction, and is used to distinguish between parameters with significant influence and parameters with general influence.

[0060] It is worth noting that the preset time interval is set based on the rate of change of the blowing process parameters and the on-site control response requirements, ensuring that changes in operating conditions can be captured in a timely manner without increasing redundant calculations due to excessive frequency.

[0061] It is worth noting that, at this time interval, the adapted feature set data is periodically extracted. The extracted content includes the real-time values ​​of each feature parameter, the time-series change trend, and the corresponding blowing stage information. The extracted feature set data is compared and matched with the labeled sample set. The association logic of "parameter change - endpoint status" in the sample set is called. Combined with the current blowing progress, a dynamic prediction result of the copper converter blowing endpoint is generated. The prediction result specifically includes: the remaining time to reach the blowing endpoint, the predicted value of the endpoint crude copper grade, the predicted value of the endpoint melt sulfur content, and the predicted value of the endpoint melt temperature. The generated dynamic prediction result is output in real time according to the blowing progress and synchronously fed back to the on-site control terminal to ensure that operators can obtain the latest prediction information in a timely manner.

[0062] It is worth noting that during the dynamic generation of prediction results, the influence weight of each feature parameter on the accuracy of the prediction results is simultaneously calculated: a quantitative calculation method is used to analyze the influence of the numerical change of each feature parameter on the prediction results (remaining time, predicted value of core indicators), and the influence magnitude is converted into a standardized influence weight.

[0063] It is worth noting that for each of the selected key influencing parameters, the numerical correspondence between them and the prediction results was recorded one by one: the quantitative correlation between the range of change of each key influencing parameter and the magnitude of change in the prediction result was clarified, such as an increase of X units in the blower flow rate, a decrease of Y minutes in the remaining time at the endpoint, and an increase of Z percentage in the grade of crude copper at the endpoint, forming a numerical correspondence table; at the same time, the correlation logic between each key influencing parameter was analyzed: the synergistic effect between parameters was clarified, such as the comprehensive impact on the endpoint temperature when the oxygen enrichment ratio and blower flow rate were adjusted synchronously; the restrictive effect was clarified, such as the interaction law that when the value of a certain parameter is too high, it will weaken the positive impact of another parameter on the prediction result; the preliminary list of key influencing parameters, the influence weight of each parameter, the numerical correspondence table, and the correlation logic between parameters were systematically organized to form a standardized list of key parameters and a document explaining the degree of correlation; this document was output synchronously with the dynamic prediction results.

[0064] It is worth noting that each time a new dynamic prediction result is generated at a preset time interval, the impact weight statistics, numerical correspondence, and correlation logic of key influencing parameters are updated synchronously (to adapt to real-time changes in operating conditions). If there are significant adjustments to the operating conditions (such as changes in raw material batches or furnace age stages), the key influencing parameter screening and correlation logic sorting operations are re-executed to ensure that the output list of key parameters and correlation descriptions always fit the current operating conditions, providing accurate guidance for on-site control.

[0065] Understandably, this embodiment achieves precise adaptation between the prediction results and real-time changes in the blowing conditions by dynamically generating and outputting the endpoint prediction results at preset time intervals. This effectively avoids the deviation problem caused by fluctuations in operating conditions due to fixed prediction results, ensuring that the prediction information obtained by operators is always timely and reliable. By screening key influencing parameters, the core objects of on-site control are focused, avoiding operators blindly searching for control targets among numerous parameters and improving control efficiency. By sorting out the numerical correspondence between key parameters and prediction results and the correlation logic between parameters, clear quantitative control basis is provided for operators, clarifying the impact of parameter adjustments on the endpoint state. This helps operators to optimize operating parameters in a targeted manner (such as adjusting the blower flow rate, oxygen enrichment ratio, etc.), reducing problems such as substandard crude copper quality, increased metal loss, and increased energy consumption caused by blind control.

[0066] As can be seen, this invention collects and preprocesses multi-source relevant data during the copper converter smelting process, constructs a predictive feature set based on the fusion features selected from the smelting endpoint indicators, reproduces the endpoint conditions in the later stages of copper production to expand the labeled sample, classifies the conditions according to raw material batch, furnace age stage, and tuyere operating status, matches and adapts the feature set, outputs dynamic prediction results, and simultaneously provides key influencing parameters and their correlation degree. It can systematically optimize data utilization, accurately mine effective features, adapt to diverse smelting conditions, enrich the foundation of endpoint prediction samples, clarify the core influencing factors of the prediction results, provide comprehensive and reliable technical support for the determination of the endpoint of copper converter smelting, and ensure the orderly progress of the smelting process.

[0067] refer to Figure 2 In another embodiment of this application, a dynamic prediction system for the endpoint of copper converter blowing based on multi-source feature fusion is applied to a dynamic prediction method for the endpoint of copper converter blowing based on multi-source feature fusion, including: a data acquisition module, a data processing module, a feature fusion module, a working condition reproduction module, a feature matching module, and a prediction module.

[0068] Specifically, the data acquisition module, data processing module, feature fusion module, working condition reproduction module, feature matching module, and prediction module are connected via a communication bus.

[0069] Specifically, the data acquisition module is used to collect multi-source data related to the blowing state during the copper converter blowing process; The data processing module is used to preprocess multi-source data; The feature fusion module is used to filter features from preprocessed multi-source data based on the refining endpoint index, and then fuse the filtered feature parameters to construct a feature set for endpoint prediction. The working condition reproduction module is used to reproduce the final working condition of the later stage of copper production based on historical blowing data, expand the data sample of the adjacent interval of the final point, and mark the working condition status of the expanded sample. The feature matching module is used to classify the blowing conditions according to raw material batch, furnace age stage and tuyere operating status, and to match and adapt the feature set with the corresponding operating condition type. The prediction module is used to output dynamic prediction results of the copper converter blowing endpoint based on the adapted feature set, and simultaneously provide the key influencing parameters of the prediction results and their correlation degree.

[0070] It is understandable that the dynamic prediction system and method for the end point of copper converter blowing based on multi-source feature fusion have the same beneficial effects, which will not be elaborated here.

[0071] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0072] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0073] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0074] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for dynamic prediction of the end point of copper converter blowing based on multi-source feature fusion, characterized in that, include: Collect multi-source data related to the blowing state during the copper converter blowing process. The multi-source data includes at least operating parameters, detection parameters, process parameters, and auxiliary monitoring parameters. The multi-source data is preprocessed, including anomaly screening and removal, missing data imputation, and measurement bias correction. Based on the refining endpoint index, feature filtering is performed on the preprocessed multi-source data, and the filtered feature parameters are fused to construct a feature set for endpoint prediction. Based on historical smelting data, the final working conditions of the later stage of copper production were reproduced. The data samples of the adjacent intervals of the final point were expanded, and the working conditions of the expanded samples were marked. The blowing conditions are classified according to raw material batch, furnace age stage and tuyere operating status, and the feature set is matched and adapted with the corresponding condition type. Based on the adapted feature set, the dynamic prediction result of the copper converter blowing endpoint is output, and the key influencing parameters of the prediction result and their correlation degree are given simultaneously.

2. The method for dynamic prediction of the end point of copper converter blowing based on multi-source feature fusion according to claim 1, characterized in that, The collection of multi-source data related to the blowing state during the copper converter blowing process includes: The corresponding data acquisition device is used to synchronously collect the operating parameters, detection parameters, process parameters, and auxiliary monitoring parameters; The operating parameters include blast flow rate, blast pressure, oxygen enrichment ratio, and adjustment timing data for each parameter; the detection parameters include furnace surface temperature, furnace gas outlet temperature, flue gas composition content, and detection node data for each parameter; the process parameters include total amount of matte injected, flux addition batches, and input time node data for each material; the auxiliary monitoring parameters include furnace vibration amplitude, melt level height, and continuous monitoring data for each parameter. All collected multi-source data are archived sequentially according to the blowing timeline to form a raw data set with time sequence identifiers.

3. The method for dynamic prediction of the end point of copper converter blowing based on multi-source feature fusion according to claim 2, characterized in that, The preprocessing of the multi-source data includes: The data in the original dataset is screened point by point to identify abnormal data that exceeds the fluctuation range of preset parameters, and the abnormal data is removed from the original dataset. For missing data segments in the dataset, valid data under the same blowing stage and working conditions are extracted, and time-series interpolation is used to fill in the missing data. Historical qualified data within the same raw material batch and furnace age range are selected as the correction benchmark. The deviation difference between the current data and the benchmark data is calculated, and the deviation is corrected based on the deviation difference. All data that have been screened, removed, filled in missing data, and corrected for biases are integrated to form a preprocessed dataset.

4. The method for dynamic prediction of the end point of copper converter blowing based on multi-source feature fusion according to claim 3, characterized in that, The step of screening the data in the original dataset point by point, identifying abnormal data that exceeds the fluctuation range of preset parameters, and removing the abnormal data from the original dataset includes: For different types of parameters, corresponding normal fluctuation ranges are set, and the normal fluctuation ranges are determined based on the statistical range of historical qualified data under the same working conditions. The data in the original dataset are checked point by point to determine whether each data point is within the normal fluctuation range of the corresponding parameter; For data points that exceed the normal fluctuation range, if the data points before and after them all exceed the normal fluctuation range of the corresponding parameter, then the data point is confirmed as abnormal data and removed from the original dataset. Record the location, value, and reason for removing abnormal data to form an abnormal data processing log.

5. The method for dynamic prediction of the end point of copper converter blowing based on multi-source feature fusion according to claim 3, characterized in that, The step of correcting the deviation of the current data based on the deviation difference includes: When the deviation between the current parameter data and the benchmark value is greater than the preset deviation threshold, the correction amount is calculated based on the deviation difference, and the correction operation is performed. Specifically, the deviation difference is compared with a preset adjustment coefficient mapping table, and the adjustment coefficient of the deviation difference is determined based on the comparison result; The product of the adjustment coefficient and the deviation difference is used as the correction amount; When the deviation between the current parameter data and the benchmark value is less than or equal to the preset deviation threshold, no correction operation is performed on the current parameter data, and the original data is directly retained.

6. The method for dynamic prediction of the end point of copper converter blowing based on multi-source feature fusion according to claim 5, characterized in that, The process involves feature filtering of the preprocessed multi-source data based on the refining endpoint index, followed by fusion of the filtered feature parameters to construct a feature set for endpoint prediction, including: The core indicators for the final stage of the blowing process are determined, including crude copper grade, melt sulfur content, and melt temperature. Calculate the correlation between the values ​​of each parameter in the preprocessed dataset and the core indicator, and select parameters with a correlation higher than a preset threshold as valid feature parameters; The effective characteristic parameters are divided into operation characteristic group, detection characteristic group, process characteristic group and auxiliary monitoring characteristic group according to parameter type, and a corresponding mapping relationship between each type of parameter and the blowing endpoint index is established. The parameter data of each feature group are associated and spliced ​​according to the blowing time sequence, the blowing stage corresponding to each feature parameter is marked, and a feature set for endpoint prediction is formed. Redundancy is checked and redundant parameters are removed from the parameter data in the feature set.

7. The method for dynamic prediction of the end point of copper converter blowing based on multi-source feature fusion according to claim 6, characterized in that, The process of reproducing the final operating conditions of the later stage of copper production based on historical blowing data involves expanding the data sample in the adjacent interval of the final point and labeling the operating conditions of the expanded sample, including: Retrieve the final operating condition parameters of the later stage of copper production in the historical blowing process. The operating condition parameters include a combination of historical operating parameters, historical process parameters and historical detection parameters. The current reproduction conditions are set according to the historical endpoint operating conditions parameters to simulate the endpoint reaction environment in the later stage of copper production, and the operating conditions reproduction operation is performed; the reproduction conditions include material input, blowing operation standards and detection frequency; During the reproduction process, data of each parameter are collected simultaneously to form a single set of samples in the vicinity of the endpoint. The number of samples in the vicinity of the endpoint is expanded by repeatedly changing the raw material batch and furnace age status to reproduce the working conditions. For each expanded sample group, offline sampling and testing are performed simultaneously to obtain the measured values ​​of the endpoint indicators corresponding to the sample. The working condition of the sample is labeled according to the measured values ​​to form a labeled sample set.

8. The method for dynamic prediction of the end point of copper converter blowing based on multi-source feature fusion according to claim 7, characterized in that, The process of classifying the blowing conditions according to raw material batch, furnace age stage, and tuyere operating status, and matching and adapting the feature set with the corresponding condition type, includes: Based on the composition ratio, Fe / S ratio and mining source information of the copper matte raw material, different batches of raw materials are classified into different operating conditions. Based on the amount of furnace lining wear and the duration of furnace lining use, different furnace age stages are classified into different operating condition types. Based on the results of airflow and cleanliness tests at the air ventilator, different air ventilator operating conditions are classified. Extract the core parameters from the feature set and compare them with the feature parameter templates for each working condition type; When the matching degree between the core parameters of the feature set and the template of a certain working condition reaches the preset matching degree threshold, the feature set is associated with the working condition and the adaptation process is performed. When the matching degree between the core parameters of the feature set and all working condition type templates does not reach the preset matching degree threshold, the core parameters of the feature set are re-extracted, the accuracy of parameter extraction is checked, and the comparison operation is performed again until the matching is completed or it is confirmed that there is no matching working condition type and recorded. For the matched feature set, the time sequence correspondence and numerical correlation logic of the parameters are adjusted to make the feature set match the blowing reaction law of the corresponding working condition type, thus completing the adaptation process.

9. The method for dynamic prediction of the end point of copper converter blowing based on multi-source feature fusion according to claim 8, characterized in that, The dynamic prediction result of the copper converter blowing endpoint is output based on the feature set after adaptation processing, and the key parameters that have a significant impact on the prediction result and their correlation degree are given simultaneously, including: The adapted feature set data is extracted at preset time intervals, combined with the labeled sample set, to generate dynamic prediction results of the copper converter blowing endpoint, and output in real time according to the blowing progress. In the statistical prediction process, the influence weight of each feature parameter on the prediction result is determined, and parameters with weights higher than the preset weight standard value are selected as key influencing parameters. Record the numerical correspondence between each key influencing parameter and the prediction results, clarify the correlation logic between each key parameter, and output a list of key parameters and an explanation of the degree of correlation.

10. A dynamic prediction system for the endpoint of copper converter blowing based on multi-source feature fusion, applied to the dynamic prediction method for the endpoint of copper converter blowing based on multi-source feature fusion as described in any one of claims 1-9, characterized in that, include: The system includes a data acquisition module, a data processing module, a feature fusion module, a working condition reproduction module, a feature matching module, and a prediction module. The data acquisition module is used to collect multi-source data related to the blowing state during the copper converter blowing process; The data processing module is used to preprocess the multi-source data; The feature fusion module is used to perform feature filtering on the preprocessed multi-source data based on the refining endpoint index, and to fuse the filtered feature parameters to construct a feature set for endpoint prediction. The working condition reproduction module is used to reproduce the final working condition of the later stage of copper production based on historical blowing data, expand the data sample of the adjacent interval of the final point, and mark the working condition status of the expanded sample. The feature matching module is used to classify the blowing conditions according to the raw material batch, furnace age stage and tuyere operating status, and to match and adapt the feature set with the corresponding operating condition type. The prediction module is used to output dynamic prediction results of the copper converter blowing endpoint based on the adapted feature set, and simultaneously provide the key influencing parameters of the prediction results and their correlation degree.

Images