Method for optimizing copper flotation by recommending and adjusting slag composition based on machine learning

Abstract

The invention relates to a method for optimizing copper flotation by recommending and adjusting slag composition based on machine learning. The method comprises the following steps: firstly, collecting key chemical composition ratios S / Cu, Fe / SiO2 and CaO / SiO2 of the copper slag, and constructing a data set in combination with corresponding tailings copper grade measured values; and then, constructing a prediction model by adopting an XGBoost algorithm, and performing feature inversion in combination with an SHAP analysis method so as to reversely determine and recommend or iteratively recommend a key chemical composition recommendation interval which enables the copper grade of the predicted tailings to be lower than a target value. And adjusting the flotation raw materials or the previous process according to the recommended interval, so that the chemical composition falls into the interval, and then carrying out flotation. According to the method, conversion from passive raw material adaptation to active raw material design and adjustment is achieved, the optimal condition can be created for flotation from the source, the method has the advantages of being accurate in model, high in interpretability, good in process collaboration and the like, the copper slag phase can be remarkably optimized, the tailing copper grade is reduced, and the recovery rate is increased.

Description

Technical Field

[0001] This invention relates to the field of non-ferrous metal beneficiation technology, and in particular to a method for optimizing copper flotation based on machine learning-based recommendation and adjustment of slag composition. Background Technology

[0002] Various slags generated during copper smelting (such as converter slag, lean electric arc furnace slag, and flash furnace slag) are important secondary resources, typically containing 0.5% to 5% copper, and possess extremely high recovery value. Flotation is a key process for recovering copper from these slags. However, the chemical composition of these slags is complex and variable. The ratios of their key chemical components, especially S / Cu, Fe / SiO2, and CaO / SiO2, not only directly affect the crystallization size and dissociation characteristics of copper minerals, but also fundamentally determine the copper-bearing phase after the slag cools, thus affecting its floatability. Among these, copper sulfide, with its strong hydrophobic surface, readily binds to xanthate collectors, making it the easiest phase to recover by flotation. Metallic copper has slightly lower floatability and typically requires activator treatment to form a surface sulfide film before it can be effectively collected. Copper oxide, on the other hand, has a strong hydrophilic surface and also requires surface conversion with sodium sulfide before it can be floated. The most challenging phase is copper in sparingly soluble minerals such as fir olivine, which exists in a strongly bound state and is almost impossible to recover using conventional flotation methods. Therefore, the distribution differences of copper among different phases are the fundamental reason for the large fluctuations in flotation tailings grade and unstable recovery rates.

[0003] Traditional optimization approaches for copper slag with a fixed composition primarily involve passively improving separation performance by adjusting process parameters such as flotation reagent regime, pulp potential, and pH. However, this method struggles to overcome the limitations imposed by the inherently poor phase composition of the copper slag. For instance, when copper is abundant in the difficult-to-float fracturing olivine or glassy phase, not only are conventional collectors ineffective, but even the use of highly activated and specialized reagents results in extremely limited recovery. Furthermore, this can lead to a dramatic increase in reagent consumption and tailings impurities, ultimately impacting subsequent resource utilization or environmental emission requirements.

[0004] In the complex system of actual smelting and beneficiation, key slag parameters (such as S / Cu, CaO / SiO2, and Fe / SiO2) do not act in isolation, but are interconnected and synergistically affect the process mineralogical properties of copper slag, ultimately influencing flotation indices. However, most existing studies focus on the independent effects of single or a few variables, lacking a systematic quantitative description and global optimization scheme for how the synergistic effects of multiple parameters influence the final flotation indices. Current technologies lack effective means to reverse-engineer and quantitatively derive the optimal chemical composition range to be controlled at the smelting source, starting from the target of low tailings grade. Therefore, it is impossible to proactively and directionally optimize the crystallization behavior of copper slag, and it is difficult to systematically promote the enrichment of copper in the easily flotable copper sulfide phase and reduce its occurrence in sparingly soluble minerals, fundamentally limiting further improvements in flotation indices.

[0005] Data-driven methods, exemplified by machine learning, offer a novel paradigm for modeling and optimizing multivariable, nonlinear processes. Their successful application in mineral processing, particularly in process parameter optimization and index prediction, demonstrates their ability to effectively uncover the inherent patterns hidden within complex data. The introduction of interpretability analysis tools (such as SHAP) further enhances the transparency and comprehensibility of model conclusions, providing new perspectives for mechanistic exploration.

[0006] Therefore, developing an intelligent method that can integrate multivariate industrial data, analyze complex nonlinear relationships, and reversely recommend the optimal raw material composition range to guide precise control of upstream smelting is of great industrial significance for steadily improving copper recovery rate, reducing tailings grade, and achieving efficient resource utilization. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing copper flotation processes, which passively adapt to raw material fluctuations and struggle to control the copper occurrence morphology at the source. This invention provides a method and process for optimizing copper flotation based on machine learning-based recommendations and adjustments to slag composition. The core innovation of this method lies in the deep integration of data-driven modeling and interpretability analysis. It innovatively selects and demonstrates the regulatory role of the key parameter S / Cu on the sulfur potential and copper phase orientation of copper slag. Simultaneously, it clarifies the specific influence and optimization range of the conventional alkalinity parameter CaO / SiO2 on the copper occurrence state during the slow cooling of copper slag. By recommending the optimal raw material chemical composition range based on the model, it guides precise control in the upstream smelting process, thereby optimizing the copper phase distribution in the slag, increasing the proportion of easily floatable sulfide copper phases, and suppressing the formation of copper phases in insoluble minerals. Ultimately, through a matching flotation process, it achieves a stable reduction in tailings copper grade and an increase in recovery rate.

[0008] This invention discloses a method for optimizing copper flotation based on machine learning-based recommendation and adjustment of slag composition, comprising the following steps:

[0009] S1. Data Collection: The key chemical composition ratios of copper slag, S / Cu, Fe / SiO2, and CaO / SiO2, and the corresponding measured values ​​of copper grade in tailings were collected to form a modeling dataset. Among these, the use of the S / Cu ratio as a key modeling parameter for evaluating and controlling the properties of copper slag is proposed for the first time. It directly determines the sulfur potential of the slag and is a core control factor guiding copper enrichment towards easily flotation phases. CaO / SiO2, as a commonly used metallurgical parameter, has not previously had a clearly defined quantitative influence mechanism and recommended range on the copper occurrence state during the slow cooling process of copper slag. This invention is the first to incorporate it into a synergistic analysis system with flotation indicators.

[0010] S2. Model Construction: The XGBoost algorithm is used to train the dataset to construct a regression prediction model for the chemical composition ratios and tailings copper grade. The regression prediction model predicts the tailings copper grade by integrating N regression trees, and its final prediction model expression can be expressed as: ; in, For the first Predicted copper grade of tailings for a sample For the first The input feature vector of each sample, The number of ensemble trees in the model. For the k-th regression tree, the first... The predicted score for each sample. Let the function space be the sum of all possible regression trees. It refers to a specific sample among the samples from the 1st to the nth.

[0011] The training process of the model is expressed as follows: ; in, Let t be the current iteration round number (the t-th tree). After the t-th iteration, the model evaluates the samples... The predicted value of copper grade, This is the prediction result accumulated from the first t-1 rounds. For the newly generated tree in round t, fit the residuals of the previous round's predictions to correct the prediction bias for copper grade. It refers to a specific sample among the samples from the 1st to the nth.

[0012] S3. Interval Inversion and Recommendation: Based on the regression prediction model, the SHAP (SHapley Additive ex Planations) analysis method is applied to perform feature importance analysis and inversion, and to determine and recommend the recommended intervals of key chemical compositions S / Cu, Fe / SiO2, and CaO / SiO2 that would cause the model to predict that the copper grade of the tailings is lower than the preset target value.

[0013] S4. Composition Adjustment and Flotation: Based on the recommended range, adjust the raw material ratio or previous smelting process parameters of the copper slag to make the chemical composition of the copper slag fall within the recommended range, so as to optimize the flotation phase; use the adjusted copper slag as flotation feed for subsequent flotation production.

[0014] The slag described in this invention is preferably copper smelting slag.

[0015] Furthermore, the dataset in step S1 is derived from actual production data from the factory, with a sample size of no less than 500 groups and no less than 3 input variables. In the iterative optimization, other process parameters that can be included include, but are not limited to, one or more of the following: Al2O3 content and other chemical compositions in the slag, cooling rate of the slow cooling process, isothermal temperature and time, or properties of the cooling medium.

[0016] Furthermore, the specific steps of model construction in step S2 include:

[0017] a. Data preprocessing: First, the dataset is standardized to eliminate the influence of units, and then divided into training and test sets in a ratio of 8:2 to obtain standardized input features; the training set is used to build the prediction model, and the test set is used to verify the generalization ability of the model.

[0018] b. Model training and optimization: Using the standardized chemical composition ratio as the input feature and the corresponding tailings copper grade as the target, configure the XGBoost model and optimize the model hyperparameters using genetic algorithms or grid search methods. The model training is completed through the training set.

[0019] c. Model performance evaluation: Root mean square error (RMSE) and coefficient of determination (R²) are used. 2 The model's prediction accuracy was evaluated, and the SHAP analysis method was used to assess the contribution and importance of each input feature to the prediction results.

[0020] Furthermore, in step b, the grid search method refers to finding the optimal parameter grid in a pre-coarsely determined grid through 5-fold cross-validation.

[0021] Furthermore, in step b, the hyperparameters optimized using the genetic algorithm include at least the learning rate, the maximum depth of the tree, and the subsample ratio.

[0022] Furthermore, in step c, the formula for calculating the root mean square error (RMSE) is: ; The formula for calculating the coefficient of determination R² is as follows: ; in, For the first The actual copper grade value of each sample. For the first The predicted copper grade value for each sample, where n is the number of samples. This represents the average actual copper grade of the sample. It refers to a specific sample among the samples from the 1st to the nth.

[0023] Furthermore, in step c, SHAP analysis is used to quantitatively assess the importance of each input feature value (S / Cu, Fe / SiO2, CaO / SiO2) to the prediction results of copper grade in tailings.

[0024] Furthermore, in step c, the performance of the control model is: R 2 The value is between 0.9 and 1, and the RMSE is between 0 and 0.001.

[0025] After completing S3, model verification and iterative optimization are performed. Specifically, the copper slag adjusted according to the recommended range is tested. If the actual copper grade of the tailings does not reach the preset target, the process returns to step S1, the dataset is expanded, and other process parameters that affect slag crystallization and phase formation are further included as new input features on the basis of the key chemical composition ratio. The model is then retrained and optimized until the recommended range of the model can make the actual copper grade of the tailings stably meet the target. And / or, after completing S4, perform model verification and iterative optimization. Specifically, conduct industrial or scale-up tests on the copper slag adjusted according to the recommended range. If the actual copper grade of the tailings does not reach the preset target, return to step S1, expand the dataset, and further incorporate other process parameters that affect slag crystallization and phase formation as new input features based on the key chemical composition ratios. Retrain and optimize the model until the recommended range of the model can stably meet the actual copper grade of the tailings.

[0026] This design ensures that the developed processes are perfectly suited for industrial applications.

[0027] Furthermore, the recommended ranges for the key chemical compositions retrieved and recommended in step S3 are: S / Cu 0.20~0.34, Fe / SiO2 1.20~1.35, and CaO / SiO2 0.07~0.14. Where S / Cu, Fe / SiO2, and CaO / SiO2 are all mass ratios.

[0028] Furthermore, the key chemical composition recommendation ranges inverted and recommended in step S3 work synergistically through the following mechanism to achieve the regulation of the copper slag phase.

[0029] This invention, through modeling, analysis, and deduction, controls the S / Cu ratio (mass ratio) within the range of 0.20 to 0.34. This range can effectively increase the sulfur potential of the slag, thereby thermodynamically promoting the directional migration and enrichment of copper into the copper sulfide phase during the slag cooling process. This is a key innovative control point for optimizing the occurrence state of copper in the slag.

[0030] This invention, through modeling, analysis, and deduction, controls the CaO / SiO2 ratio (mass ratio) within the range of 0.07 to 0.14. This control range is defined for the first time using a data-driven model. This invention optimizes and selects this ratio range, which reduces viscosity by using free oxygen provided by CaO to break the silicon-oxygen network and depolymerize the melt structure. 2+ Ion exchange substitution of Fe in fir olivine 2 + Meanwhile, Ca 2+ It can also partially replace copper ions that may enter the crystal lattice, destroying the lattice and thus inhibiting the occurrence of copper in the insoluble silicate phase. On the other hand, the moderate alkalinity effectively inhibits the large-scale formation of high-melting-point spinel phase, thereby optimizing the overall viscosity and phase interface properties of the slag and promoting the aggregation and grain growth of copper minerals.

[0031] This invention, through modeling, analysis, and deduction, controls the Fe / SiO2 ratio (mass ratio) within the range of 1.20 to 1.35. This recommended range falls within a reasonable slag type range, aiming to regulate the viscosity and mineral composition of the slag, maintaining low viscosity and good fluidity, thereby reducing the aggregation resistance of copper matte droplets in the slag phase, promoting the aggregation and grain growth of copper minerals in the copper slag, and improving the embedding and dispersion of copper minerals.

[0032] Through the above-mentioned synergistic regulation, the copper occurrence state in the copper slag is ultimately satisfied: the copper content in copper sulfide is greater than 60%, and the copper content in sparingly soluble minerals is less than 10%, thus providing an ideal raw material phase basis for subsequent efficient flotation.

[0033] In step S4 of this invention, adjusting the raw material ratio or previous smelting process parameters of the copper slag includes adjusting at least one of the following: Al2O3 content in the slag, cooling rate of the slow cooling process, constant temperature and time, and cooling medium.

[0034] By adjusting step S4, this invention optimizes the mineral crystallization behavior and phase distribution of copper slag, so that the copper occurrence state in the copper slag meets the following requirements: the copper content in copper sulfide is greater than 60%, and the copper content in sparingly soluble minerals is less than 10%.

[0035] In this invention, the flotation feed is copper slag whose chemical composition falls within the recommended range after being adjusted by the method described above; the process includes: The feed ore is ground until the proportion of particles with a particle size of less than 0.045 mm reaches 78-90%. The grinding product is subjected to a flotation process consisting of primary roughing, secondary roughing, primary scavenging, secondary scavenging, and tertiary scavenging. Throughout the flotation process, the redox potential of the pulp is controlled within the range of 50 mV to -180 mV, the pH value is controlled between 8.0 and 10.0, and the pulp mass concentration during flotation is maintained between 33% and 45%.

[0036] The reagents added to each flotation stage in this invention and their unit addition amounts are as follows, wherein the unit addition amount is calculated based on the feed per ton of ore for that stage:

[0037] Activator: Selected from one or more of sodium sulfide, ammonia, triethanolamine, and hydroxylamine hydrochloride. 250~590 g / t is added during the first roughing stage, 0~150 g / t (preferably 50~150 g / t) is added during the second roughing stage, and 0~100 g / t (preferably 90~100 g / t), 0~85 g / t (preferably 70~85 g / t), and 0~65 g / t (preferably 40~60 g / t) are added during the first, second, and third scavenging stages, respectively. Collector: Selected from one or more of butyl xanthate, butylamine black powder, isopentyl potassium xanthate, pentyl sodium xanthate, sec-octyl xanthate, and Z200. The amount added is 40~60 g / t in the first roughing, second roughing, second scavenging, and third scavenging, and 60~90 g / t in the first scavenging. Foaming agent: selected from one or more of pine oil, No. 2 oil, methyl isobutyl methanol, and ether alcohols, added at 12~24 g / t, preferably 12~15 g / t during the first roughing stage, and at 24~48 g / t, preferably 24~30 g / t, during the second roughing stage, the first scavenging stage, the second scavenging stage, and the third scavenging stage, respectively. Dispersant: Selected from one or more of water glass, sodium hexametaphosphate, sodium carbonate, and sodium tripolyphosphate, added in the roughing stage, with a total addition of 0~600 g / t.

[0038] The flotation feed of the present invention is at least one of converter slag, lean electric furnace slag, or flash furnace slag.

[0039] Furthermore, in the method described above for optimizing copper flotation by recommending and adjusting slag composition based on machine learning, the flotation feed is one or more mixed slags selected from converter slag, depleted electric arc furnace slag, or flash furnace slag.

[0040] Compared with the prior art, the present invention has the following significant advantages:

[0041] 1. Existing technologies mainly target slag with fixed compositions, passively adapting to the flotation process by adjusting parameters such as reagents and potential. This approach struggles to overcome the bottleneck of poor raw material phase composition. This invention creatively moves the optimization process upstream to the smelting stage, using a data-driven model to recommend the optimal chemical composition range. This guides upstream batching and process adjustments. For the first time, S / Cu is introduced as a key control parameter into the copper slag phase optimization system, actively guiding the crystallization behavior of copper towards the easily floatable sulfide copper phase, fundamentally optimizing the phase composition of the flotation feed.

[0042] 2. To address the complex nonlinear relationship between slag chemical composition and flotation indicators, this invention employs the XGBoost algorithm to construct a high-precision prediction model and utilizes SHAP analysis to rank feature importance and interpret contribution. This method not only achieves accurate prediction of tailings grade but also transparently reveals the synergistic influence mechanism of multiple key parameters such as S / Cu, Fe / SiO2, and CaO / SiO2. For the first time, it clarifies the influence of CaO / SiO2 on the copper occurrence state during the slow cooling process of copper slag using a data-driven approach, thereby scientifically and quantitatively deriving the globally optimal composition control range and overcoming the limitations of traditional research relying on single-factor experiments and empirical judgment.

[0043] 3. This invention does not consider individual chemical parameters in isolation, but rather systematically regulates the sulfur potential, basicity, viscosity, and mineral composition of the slag by recommending a synergistically optimized combination of ranges. This synergistic effect effectively promotes the enrichment of copper into the copper sulfide phase, while significantly inhibiting its occurrence in sparingly soluble minerals such as fritillary olivine, creating ideal process mineralogical conditions for subsequent efficient flotation, which cannot be achieved by simply adjusting flotation parameters.

[0044] 4. This invention provides a complete process solution from intelligent recommendation and front-end adjustment to back-end flotation. Based on obtaining copper slag with optimized phase composition, precisely matched flotation process parameters are designed. Example results show that this system can stably reduce the copper grade of flotation tailings to below 0.20% while maintaining a high copper recovery rate, with overall performance significantly better than traditional methods.

[0045] 5. The method of this invention is based on modeling from actual production data, with clearly defined and specific recommended ranges. It can directly guide the smelting control and flotation production of various copper slags, such as converter slag, lean electric arc furnace slag, and flash furnace slag. This method achieves its goal through intelligent optimization of upstream batching or smelting process parameters, without requiring large-scale modifications or investments to downstream flotation equipment. Therefore, it is easy to implement and promote in existing industrial systems, and has significant value for achieving efficient and stable recovery of secondary resources from copper smelting.

[0046] 6. This invention further establishes a closed-loop optimization process of "modeling-recommendation-verification-iteration": When the copper slag produced after adjustment according to the recommended range fails to stably meet the standards in actual flotation, the system can expand the dataset, incorporate new features such as Al2O3 content and slow cooling process parameters, retrain the model, and iteratively optimize the recommended range until the copper grade of the tailings stably meets the preset target. Based on the finally determined recommended range (i.e., the optimized range), the raw materials or preceding processes of the copper slag entering the flotation are adjusted so that their chemical composition falls within this range, thereby optimizing the copper occurrence state from the source and achieving the goal of copper sulfide phase ratio greater than 60% and copper content in sparingly soluble minerals less than 10%. Attached Figure Description

[0047] Figure 1 The flowchart illustrates a method for optimizing copper flotation based on machine learning to recommend and adjust the phase composition of slag, as provided in an embodiment of the present invention.

[0048] Figure 2 This is a graph showing the fitting effect between the XGBoost model prediction and the actual copper grade of the tailings in the example.

[0049] Figure 3 A flotation process flow diagram provided for an embodiment of the present invention. Detailed Implementation

[0050] The present invention will be further described below with reference to embodiments, but the embodiments do not limit the present invention in any way. Simple modifications or substitutions made to the methods, steps or conditions of the present invention without departing from the spirit and substance of the present invention are all within the scope of the present invention; unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0051] Example 1

[0052] The method for optimizing copper flotation based on machine learning to recommend and adjust slag phase composition provided by this invention is as follows: Figure 1As shown. The core of this method lies in optimizing the phase distribution of the slag by actively adjusting its key chemical composition (S / Cu, Fe / SiO2, CaO / SiO2), aiming to significantly improve the enrichment of copper in the easily floatable copper sulfide phase and reduce its occurrence in the difficult-to-recover phases. To solve the problem of large fluctuations in chemical composition and phase distribution and difficulty in controlling the tailings grade in copper slag flotation, based on the systematic analysis of production data and flotation mechanism, this invention applies the aforementioned intelligent optimization method, achieving a reduction in tailings copper grade to below 0.20%, an increase in concentrate copper grade to above 18.00%, and a simultaneous and significant increase in copper recovery rate to above 90.0%.

[0053] The following combination Figure 1 The implementation process of the method described in this embodiment will be explained in detail below:

[0054] 1) Data Collection: Production data from the past year was extracted from the central control system of a copper smelter, and 530 valid records were selected. Each record includes: the S / Cu (mass ratio), Fe / SiO2 (mass ratio), and CaO / SiO2 (mass ratio) analysis values ​​of the raw copper slag, as well as the copper grade of the tailings after flotation of that batch of copper slag. This constitutes the original dataset. This invention specifically uses the S / Cu ratio as one of the key input features for modeling, which is the first time it has been proposed to systematically evaluate and control the process mineralogical characteristics of copper slag; at the same time, the conventional metallurgical parameter CaO / SiO2 is included in this analysis system, aiming to quantitatively reveal its specific impact on the copper occurrence state during the slow cooling process of copper slag for the first time.

[0055] 2) Model Construction: Z-score standardization was applied to the three features S / Cu, Fe / SiO2, and CaO / SiO2 to eliminate the influence of dimensions. The standardized data were then randomly divided into training and testing sets in an 8:2 ratio. Subsequently, the XGBoost algorithm was used to construct a regression prediction model: using the standardized chemical composition as input and the copper grade of the tailings as output, a genetic algorithm was used to optimize hyperparameters such as learning rate, maximum tree depth, and subsample ratio. The mean negative squared error of 5-fold cross-validation was used as the fitness function. The optimal parameter combination was obtained through iterative optimization, and the final model was trained.

[0056] 3) Model Evaluation and Interpretation: Training set RMSE=0.0087, R 2 =0.9779, test set RMSE=0.0033, R 2 =0.9340, indicating that the model has high prediction accuracy and good generalization ability. Figure 2To further reveal the influence mechanism of key chemical components, the SHAP method was used for model interpretability analysis. The results showed that the importance of the features was ranked as Fe / SiO2 > S / Cu > CaO / SiO2. SHAP dependency analysis further indicated that when Fe / SiO2 was in the range of 1.20–1.35 (mass ratio), S / Cu was in the range of 0.20–0.34 (mass ratio), and CaO / SiO2 was in the range of 0.07–0.14 (mass ratio), the SHAP value showed a significant negative contribution, meaning that this chemical composition combination had a positive promoting effect on reducing the copper grade of tailings.

[0057] 4) Recommended Range Inversion, Industrial Validation, and Closed-Loop Iteration: Based on the combined model predictions and SHAP analysis results, the recommended ranges for key chemical compositions required to stably control the copper grade of tailings below 0.20% were inverted and recommended: S / Cu: 0.20~0.34 (mass ratio), Fe / SiO2: 1.20~1.35 (mass ratio), CaO / SiO2: 0.07~0.14 (mass ratio). Based on test set samples, within these recommended ranges, the compliance rate of the copper grade in tailings exceeded 96%, validating the effectiveness of the model optimization strategy.

[0058] To construct a closed-loop optimization process, the recommended range was applied to a one-month industrial pilot production. Results showed that most batches of tailings met the grade target, but approximately 5% of batches exhibited fluctuations, slightly exceeding the 0.20% target. Therefore, iterative model optimization was initiated, returning to the data collection step. The original dataset was supplemented with the slow-cooling process parameters of the copper slag from these fluctuating batches (such as the slow-cooling endpoint temperature and cooling rate) as new input features. The expanded data was then used to retrain the XGBoost model—using the new model test set R. 2 The SHAP analysis confirmed that the newly added parameters made an interpretable contribution, with the RMSE value increased to 0.945 and the RMSE value decreased to 0.0028. Based on the new model, a more refined recommended range was obtained. Subsequent production verification showed that the copper grade compliance rate of tailings was steadily increased to over 98.1% after adopting this corrected range, thus successfully achieving the closed-loop optimization goal.

[0059] 5) Composition Adjustment: The recommended range of key chemical compositions mentioned above was used as the production control target to guide the upstream smelting and batching process. By adjusting the S / Cu ratio of the mixed copper concentrate and precisely controlling the ratio of quartz sand (adjusting Fe / SiO2) to limestone (adjusting CaO / SiO2), the composition of copper slag was actively optimized. The actual composition of the test batch of copper slag obtained after adjustment was: S / Cu = 0.21 (mass ratio), Fe / SiO2 = 1.30 (mass ratio), CaO / SiO2 = 0.11 (mass ratio), which completely fell within the recommended range.

[0060] 6) Phase optimization: Process mineralogical analysis of the adjusted copper slag revealed a significant optimization of the copper occurrence state. The copper slag contained 66.67% copper sulfide, 12.59% copper oxide, 11.11% metallic copper, and 9.63% copper in the sparingly soluble minerals. This result achieved the expected technical effect (copper sulfide >60%, copper in sparingly soluble minerals <10%), indicating that chemical composition regulation effectively promoted the enrichment of copper in easily flotable phases.

[0061] 7) Flotation Production: The feed ore is ground to a particle size of -0.045mm, with 85% of the particles being fine. The post-grinding pulp is processed using a "two-roughing, three-scavenging" flotation process, sequentially passing through Roughing I (first roughing), Roughing II (second roughing), Scavenging I (first scavenging), Scavenging II (second scavenging), and Scavenging III (third scavenging). Throughout the flotation process, key physicochemical parameters of the pulp are precisely controlled. By adding sodium sulfide and synergistically adjusting the aeration rate, the redox potential (Eh) of the pulp is stabilized within the range of -100 mV ± 30 mV. Simultaneously, the pH value of the pulp is stabilized at an alkaline environment of 9.0 ± 0.5. The pulp mass concentration at each flotation point is controlled at 35% to ensure optimal hydrodynamic conditions during the flotation process. This process employs a staged, refined reagent addition system. Sodium sulfide was added as the activator at 250 g / t before roughing stage I, 100 g / t before roughing stage II, and 100, 80, and 60 g / t before scavenging stages I, II, and III, respectively. Butyl xanthate was used as the collector, with 40 g / t added before roughing stages I and II, and 60 g / t before scavenging stage I. No. 2 oil was added as the frother at 12 g / t before roughing stage I, and 24 g / t before roughing stage II and all subsequent scavenging stages. The flotation results are shown in Table 1.

[0062] Table 1 Flotation test data of Example 1 ; Applying the method and process described in this embodiment to treat the optimized copper slag, the final copper grade of the flotation tailings was reduced to 0.17%, the copper grade of concentrate I was 19.51%, and the copper recovery rate reached 90.6%. This fully verifies the effectiveness and superiority of the entire process solution of this invention, from intelligent recommendation and control at the source to optimization of the flotation process.

[0063] Example 2

[0064] This embodiment applies the method described in Example 1 to adjust the slag composition and perform copper flotation. Copper slag with the adjusted composition of S / Cu = 0.21 (mass ratio), Fe / SiO2 = 1.30 (mass ratio), and CaO / SiO2 = 0.11 (mass ratio) was used as the flotation sample. The feed was milled until the proportion of particles with a fineness of -0.045 mm reached 88%, and the post-milling process was also treated using a "two-coarse, three-scavenger" flotation procedure. By adding sodium sulfide and synergistically adjusting the aeration rate, the oxidation-reduction potential (Eh) of the pulp was stabilized at -100 mV ± 20 mV. Simultaneously, the pH value of the pulp was stabilized at 9.0 ± 0.5. The pulp mass concentration at each flotation point was controlled at 35%. Sodium sulfide, the activator, was added at 250 g / t before roughing stage I, 100 g / t before roughing stage II, and 100, 80, and 60 g / t before scavenging stages I, II, and III, respectively. The collector was a combination of butyl xanthate and butylamine black powder in a mass ratio of 2:1. 60 g / t was added before roughing stages I and II, and scavenging stages II and III, and 90 g / t was added before scavenging stage I. Foaming agent No. 2 oil was added at 12 g / t before roughing stage I, and 24 g / t before roughing stage II and all subsequent scavenging stages. The flotation results are shown in Table 2.

[0065] Table 2 Flotation test data of Example 2 ; After processing according to the procedure in this embodiment, the copper grade of the final flotation tailings was reduced to 0.18%, the copper grade of concentrate I was 24.77%, and the copper recovery rate reached 91.4%.

[0066] Example 3

[0067] This embodiment demonstrates another specific implementation case of applying the method described in Example 1 to adjust slag composition and perform copper flotation. Copper slag with the adjusted chemical composition of S / Cu = 0.29 (mass ratio), Fe / SiO2 = 1.25 (mass ratio), and CaO / SiO2 = 0.11 (mass ratio) was used as the flotation feed, and its composition fell entirely within the recommended range. Phase analysis showed that the phase composition of this copper slag was significantly optimized, with a copper distribution rate as high as 81.51% in copper sulfides, while the copper distribution rate in sparingly soluble minerals was only 4.11%, as shown in Table 3, fully meeting the phase optimization target.

[0068] Table 3. Analysis results of Cu phase composition of copper slag falling within the recommended range / % ; After grinding the ore sample to a particle size of -0.045 mm (82%), a "two-stage roughing and three-stage scavenging" flotation process was adopted. By adding sodium sulfide and synergistically adjusting the aeration rate, the redox potential (Eh) of the pulp was stabilized within the range of -100 mV ± 40 mV. Simultaneously, the pH value of the pulp was stabilized at 9.0 ± 0.5; and the pulp mass concentration at each flotation point was controlled at 40%. This process adopts a staged and refined reagent addition system: activator (sodium sulfide) is added at 250 g / t before roughing I, 100 g / t before roughing II, and 100, 80, and 60 g / t before scavenging I, II, and III, respectively; collector (n-butyl xanthate) is added at 40 g / t before roughing I and II and scavenging II and III, and 60 g / t before scavenging I; frother (2# oil) is added at 12 g / t before roughing I, and 24 g / t before roughing II and each subsequent scavenging. The flotation results are shown in Table 4.

[0069] Table 4 Flotation test data of Example 3 ; By applying the method and process described in this embodiment to treat the optimized copper slag, the final copper grade of the flotation tailings was reduced to 0.16%, the copper grade of concentrate I was 27.41%, and the copper recovery rate reached 90.1%.

[0070] Comparative Example 1

[0071] To verify the necessity of the recommended range of key chemical compositions described in this invention, this comparative example was set up. The same modeling and adjustment logic as Example 1 was used, except that the adjusted composition was S / Cu = 0.14 (mass ratio), Fe / SiO2 = 1.24 (mass ratio), and CaO / SiO2 = 0.16. This composition did not fall within the recommended range of key chemical compositions. Phase analysis showed that the phase composition of this copper slag was significantly deteriorated, with a very low proportion of insoluble copper (only 35.37%) and a high proportion of copper in the insoluble minerals (19.05%), as shown in Table 5.

[0072] Table 5. Analysis results of Cu phase composition of copper slag whose chemical composition does not fall within the recommended range / % ; Subsequently, the batch of copper slag was treated using the same flotation process parameters and reagent regimen as in Example 1. Specific conditions were: grinding fineness of -0.045 mm accounted for 85%, pulp potential Eh controlled within the range of -100 mV ± 30 mV, pH value controlled within 9.0 ± 0.5, and pulp concentration of 35%. Reagent regimen: Activator (sodium sulfide) was added at 250 g / t before roughing stage I, 100 g / t before roughing stage II, and 100, 80, and 60 g / t before scavenging stages I, II, and III, respectively; Collector (n-butyl xanthate) was added at 40 g / t before roughing stages I and II, and 60 g / t before scavenging stage I; Frothing agent (No. 2 oil) was added at 12 g / t before roughing stage I, and 24 g / t before roughing stage II and all subsequent scavenging stages. The results are shown in Table 6.

[0073] Table 6 Flotation test data of Comparative Example 1 ; Under these comparative conditions, the copper grade in the flotation tailings was 0.37%, the copper grade in concentrate I was only 10.24%, and the copper recovery rate was only 74.4%. This result clearly demonstrates that even using the exact same advanced flotation process, if the chemical composition of the copper slag fed into the flotation does not fall within the recommended range derived from machine learning and SHAP analysis in this invention, its phase composition (low proportion of copper sulfide and high proportion of sparingly soluble minerals) will directly lead to severe deterioration of flotation performance. Therefore, adjusting the feedstock to a specific chemical composition range is a necessary prerequisite for achieving the ideal of low tailings grade and high recovery rate.

[0074] Comparative Example 2

[0075] To verify the importance of grinding fineness in the flotation process parameters, this comparative example was set up. The same copper slag with optimized chemical composition as in Example 1 (S / Cu = 0.21 (mass ratio), Fe / SiO2 = 1.30 (mass ratio), CaO / SiO2 = 0.11 (mass ratio)) was used as feed. The difference was that the grinding fineness was only -0.045 mm, accounting for 60%, significantly lower than the recommended range of 78%~90%. Other flotation process conditions remained the same as in Example 1: the mill adopted a "two roughing and three scavenging" process, the pulp potential (Eh) was controlled at -100 mV ± 30 mV, the pH value was controlled at 9.0 ± 0.5, and the pulp concentration was 35%; the reagent system was as follows: activator (sodium sulfide) was added at 250 g / t before roughing I, 100 g / t before roughing II, and 100, 80, and 60 g / t before scavenging I, II, and III, respectively; collector (n-butyl xanthate) was added at 40 g / t before roughing I and II and scavenging II and III, and 60 g / t before scavenging I; frother (2# oil) was added at 12 g / t before roughing I, and at 24 g / t before roughing II and each subsequent scavenging. The results are shown in Table 7.

[0076] Table 7 Flotation test data of Comparative Example 2 ; Under these comparative conditions, the copper grade in the flotation tailings was 0.28%, the copper grade in concentrate I was 24.59%, and the copper recovery rate was only 84.9%. This result indicates that even with optimized raw material chemical composition, optimal performance cannot be achieved if the grinding fineness does not meet the requirements of the synergistic process. Insufficient grinding fineness leads to incomplete liberation of copper mineral monomers, and a large number of intergrowths cannot be effectively identified and collected by conventional collectors, resulting in their loss in the tailings and low-grade middlings, leading to decreased recovery rate and increased tailings grade. This further demonstrates the necessity and rigor of the grinding fineness range provided by this invention for achieving the goals of high copper slag recovery and low tailings grade.

[0077] Comparative Example 3

[0078] To verify the appropriate range of pulp concentration in the flotation process parameters, this comparative example was set up. The same copper slag with optimized chemical composition as in Example 1 (S / Cu = 0.21 (mass ratio), Fe / SiO2 = 1.30 (mass ratio), CaO / SiO2 = 0.11 (mass ratio)) was used as feed, the difference being that the flotation concentration reached 50%, significantly exceeding the recommended upper limit of 33%~45%. The flotation process and other key conditions remained consistent with those in Example 1: grinding fineness of -0.045 mm accounted for 85%, pulp potential Eh was controlled within the range of -100 mV ± 30 mV, and pH value was controlled within 9.0 ± 0.5; reagent system: activator (sodium sulfide) was added at 250 g / t before roughing I, 100 g / t before roughing II, and 100, 80, and 60 g / t before scavenging I, II, and III, respectively; collector (n-butyl xanthate) was added at 40 g / t before roughing I and II and scavenging II and III, and 60 g / t before scavenging I; frother (2# oil) was added at 12 g / t before roughing I, and at 24 g / t before roughing II and all subsequent scavenging stages. The results are shown in Table 8.

[0079] Table 8 Flotation test data of Comparative Example 3 ; Under these comparative conditions, the copper grade in the flotation tailings was 0.23%, the copper grade in concentrate I was 18.71%, and the copper recovery rate was only 86.8%. This result indicates that even with optimized feedstock chemical composition, exceeding the recommended range for the synergistic process in pulp concentration will still impair overall separation efficiency. Excessively high pulp concentration increases pulp viscosity, worsens hydrodynamic conditions, leads to poor bubble dispersion, reduces the probability of mineral-bubble collisions, and exacerbates mechanical entrainment of intergrowths. This comparative example further demonstrates that the pulp concentration range determined in this invention is one of the precise synergistic conditions necessary to achieve optimal flotation performance.

[0080] Comparative Example 4

[0081] To verify the suitable range of redox potential and pH value in the flotation process parameters, this comparative example was set up. The same copper slag with optimized chemical composition as in Example 1 (S / Cu = 0.21 (mass ratio), Fe / SiO2 = 1.30 (mass ratio), CaO / SiO2 = 0.11 (mass ratio)) was used as feed. The difference was that the potential fluctuated significantly, ranging from -48.0 to -242.0 mV, and the pH fluctuated from 8.3 to 10.3, both significantly exceeding the recommended stable range. The flotation process and other key conditions remained consistent with Example 1: grinding fineness of -0.045 mm accounted for 85%, and pulp concentration was 35%; reagent regimen: activator (sodium sulfide) was added at 250 g / t before rougher I, 100 g / t before rougher II, and 100, 80, and 60 g / t before scavengers I, II, and III, respectively; collector (n-butyl xanthate) was added at 40 g / t before rougher I and II and scavengers II and III, and at 60 g / t before scavenger I; frother (2# oil) was added at 12 g / t before rougher I, and at 24 g / t before rougher II and all subsequent scavengers. The results are shown in Table 9.

[0082] Table 9. Flotation test data for Comparative Example 4 ; Under these comparative conditions, the copper grade in the flotation tailings was 0.30%, the copper grade in concentrate I was 19.77%, and the copper recovery rate was only 78.7%. This result is due to the pulp potential being as low as -242.0 mV and fluctuating significantly, along with the pH value being too high at 10.3, exceeding the stable range. These factors combined to disrupt the electrochemical stability of the flotation system and the reagent action environment, leading to decreased collector adsorption selectivity and unstable mineral surface properties. Consequently, a large amount of copper minerals lost to the tailings due to decreased floatability, which is detrimental to flotation. This comparative example further demonstrates that the stable range of key parameters such as pulp potential and pH determined in this invention is one of the precise synergistic conditions necessary to achieve optimal flotation performance.

Claims

1. A method for optimizing copper flotation based on machine learning-based recommendation and adjustment of slag composition, characterized in that: Includes the following steps; S1. Data Collection: Collect the key chemical composition ratios of copper slag, S / Cu, Fe / SiO2, and CaO / SiO2, as well as the corresponding measured values ​​of copper grade in the tailings, to form a dataset; S2. Model Construction: The XGBoost algorithm is used to train the dataset to construct a regression prediction model of the chemical composition ratios and tailings copper grade. The regression prediction model predicts the tailings copper grade by integrating N regression trees, and its final prediction model expression can be expressed as: ； in, For the first Predicted copper grade of tailings for a sample For the first The input feature vector of each sample, The number of ensemble trees in the model. For the k-th regression tree, the first... The predicted score for each sample. The function space consisting of all possible regression trees. It refers to a specific sample among the samples from the 1st to the nth sample; The training process of the model is expressed as follows: ； in, Let t be the current iteration round number (the t-th tree). After the t-th iteration, the model evaluates the samples... The predicted value of copper grade, This is the prediction result accumulated from the first t-1 rounds. For the newly generated tree in round t, fit the residuals of the previous round's predictions to correct the prediction bias for copper grade. It refers to a specific sample among the samples from the 1st to the nth sample; S3. Interval Inversion and Recommendation: Based on the regression prediction model, the SHAP analysis method is applied to perform feature importance analysis and inversion, and to determine and recommend the recommended intervals of key chemical compositions S / Cu, Fe / SiO2, and CaO / SiO2 that would cause the model to predict that the copper grade of the tailings is lower than the preset target value. S4. Composition Adjustment and Flotation: Based on the recommended range, adjust the raw material ratio or previous smelting process parameters of the copper slag to make the chemical composition of the copper slag fall within the recommended range, so as to optimize its flotation phase; use the adjusted copper slag as flotation feed for subsequent flotation production.

2. The method according to claim 1, characterized in that: The specific steps for model construction in step S2 include: a. Data preprocessing: Standardize the dataset and divide it into training and test sets in an 8:2 ratio; b. Model training and optimization: Using the standardized chemical composition ratio as the input feature and the corresponding tailings copper grade as the target, the XGBoost model is configured, and the model hyperparameters are optimized using a genetic algorithm. The model training is completed using the training set. c. Model performance evaluation: The root mean square error (RMSE) and coefficient of determination (R²) are used to evaluate the model's prediction accuracy, and the SHAP analysis method is used to evaluate the contribution and importance of each input feature to the prediction results.

3. The method according to claim 2, characterized in that: In step b, the hyperparameters optimized using the genetic algorithm include at least the learning rate, the maximum depth of the number, and the subsample ratio.

4. The method according to claim 1, characterized in that: After completing S3, model verification and iterative optimization are performed. Specifically, the copper slag adjusted according to the recommended range is tested. If the actual copper grade of the tailings does not reach the preset target, the process returns to step S1, the dataset is expanded, and other process parameters that affect slag crystallization and phase formation are further included as new input features on the basis of the key chemical composition ratio. The model is then retrained and optimized until the recommended range of the model can make the actual copper grade of the tailings stably meet the target. And / or, after completing S4, perform model verification and iterative optimization. Specifically, conduct industrial or scale-up tests on the copper slag adjusted according to the recommended range. If the actual copper grade of the tailings does not reach the preset target, return to step S1, expand the dataset, and further incorporate other process parameters that affect slag crystallization and phase formation as new input features based on the key chemical composition ratios. Retrain and optimize the model until the recommended range of the model can stably meet the actual copper grade of the tailings.

5. The method according to claim 1, characterized in that, The recommended ranges for the key chemical compositions inverted and recommended in step S3 are: S / Cu: 0.20~0.34, Fe / SiO2: 1.20~1.35, CaO / SiO2: 0.07~0.14, where S / Cu, Fe / SiO2, and CaO / SiO2 are all mass ratios.

6. The method according to claim 1, characterized in that: In step S4, the raw material ratio or previous smelting process parameters of the copper slag are adjusted, including adjusting the Al2O3 content in the slag, the cooling rate of the slow cooling process, the constant temperature and time, and at least one of the cooling media.

7. The method according to claim 1, characterized in that: By adjusting step S4, the copper occurrence state in the copper slag is made to meet the following conditions: the copper content in copper sulfide is greater than 60%, and the copper content in sparingly soluble minerals is less than 10%.

8. The method according to any one of claims 1 to 7, characterized in that: The flotation feed is copper slag whose chemical composition falls within the recommended range after being adjusted by the aforementioned method; the process includes: The feed ore is ground until the proportion of particles with a particle size of less than 0.045 mm reaches 78-90%. The grinding product is subjected to a flotation process consisting of primary roughing, secondary roughing, primary scavenging, secondary scavenging, and tertiary scavenging. Throughout the flotation process, the redox potential of the pulp is controlled within the range of 50 mV to -180 mV, the pH value is controlled between 8.0 and 10.0, and the pulp mass concentration during flotation is maintained between 33% and 45%.

9. The method according to claim 8, characterized in that: The reagents added to each flotation stage and their unit addition amounts are as follows, whereby the unit addition amount is calculated based on one ton of feed per stage: Activator: Selected from one or more of sodium sulfide, ammonia, triethanolamine, and hydroxylamine hydrochloride. Add 250~590 g / t during the first roughing, add 0~150 g / t during the second roughing, and add 0~100 g / t, 0~85 g / t, and 0~65 g / t during the first, second, and third scavenging, respectively. Collector: Selected from one or more of butyl xanthate, butylamine black powder, isopentyl potassium xanthate, pentyl sodium xanthate, sec-octyl xanthate, and Z200. The amount added is 40~60 g / t in the first roughing, second roughing, second scavenging, and third scavenging, and 60~90 g / t in the first scavenging. Foaming agent: selected from one or more of pine oil, No. 2 oil, methyl isobutyl methanol, and ether alcohols, 12~24 g / t is added during the first roughing, and 24~48 g / t is added during the second roughing, first scavenging, second scavenging and third scavenging respectively; Dispersant: Selected from one or more of water glass, sodium hexametaphosphate, sodium carbonate, and sodium tripolyphosphate, added in the roughing stage, with a total addition of 0~600 g / t.

10. The method according to claim 8, characterized in that: The flotation feed includes at least one of converter slag, lean electric furnace slag, or flash furnace slag.

Images