Mineral processing test method based on multi-data fusion algorithm

Abstract

The invention discloses a beneficiation test method based on a multi-data fusion algorithm, which comprises the following steps: collecting original ore data, and determining target parameters of a test; the collected data is preprocessed; predicting core test parameters based on the collected original ore data of the multiple attributes and the target parameters by using a multi-data fusion algorithm; on the basis of the selected core test parameters, alternative abrasion parameters and mineral separation schemes are specified, and then a test scheme is obtained; implementing a test scheme and collecting feedback information; and dynamically adjusting the test based on the feedback information. According to the method, intelligent analysis and decision making are achieved based on a multi-data fusion algorithm, a large amount of data processing can be completed in a short time, the optimal ore dressing scheme is rapidly obtained, accordingly, the ore dressing efficiency is improved, meanwhile, dependence on manpower can be reduced, and resources are saved.

Description

Technical Field

[0001] The invention belongs to the field of mining engineering and data science, and specifically relates to a mineral processing test method. Background Art

[0002] Ore dressing is a crucial link in mining engineering and has important economic and technical significance. With the gradual depletion of global resources and the continuous increase in mining costs, how to effectively and economically process ore and improve ore recovery rate and resource utilization has become a major challenge facing the industry.

[0003] Traditional mineral processing methods rely on manual experience, and are usually adjusted manually based on ore characteristics and previous test results. However, this method often shows problems such as blind decision-making, low test efficiency and waste of resources in practice. Specifically, it manifests as follows: 1. The selection of mineral processing technology and parameter setting often rely on the experience of mineral processing engineers, which may cause different engineers to make different choices under the same circumstances, resulting in inconsistent test results. 2. Mineral processing tests usually require multiple rounds, with a long test cycle, and the data analysis and integration after each round of tests are also very cumbersome, which increases the time cost. 3. Due to the lack of effective preliminary evaluation, a large number of mineral processing tests will cause unnecessary waste of resources and materials, affecting the economic benefits of the enterprise. Summary of the invention

[0004] The present invention proposes a mineral processing test method based on a multi-data fusion algorithm, the purpose of which is to solve the problems of high dependence on manual labor, long time and waste of resources in traditional methods.

[0005] The technical solution of the present invention is as follows:

[0006] A mineral processing test method based on a multi-data fusion algorithm comprises the following steps:

[0007] Step 1: Collect raw ore data and determine the target parameters of the test;

[0008] Step 2: pre-process the collected data;

[0009] Step 3: Use a multi-data fusion algorithm to predict core test parameters based on the collected raw ore data of multiple attributes and target parameters;

[0010] Step 4: Based on the selected core test parameters, specify the alternative grinding parameters and mineral processing scheme, and then obtain the test scheme;

[0011] Step 5: Implement the test plan and collect feedback information;

[0012] Step 6: Dynamically adjust the experiment based on the feedback information.

[0013] As a further improvement of the mineral processing test method based on the multi-data fusion algorithm: in step 1, the original ore data includes the grade, mineral composition, particle size distribution, mineral embedding, and physical and chemical properties of the ore; the target parameters include the set grade, fineness, recovery rate and output rate of the concentrate.

[0014] As a further improvement of the mineral processing test method based on the multi-data fusion algorithm: the preprocessing includes data screening based on threshold value, data cleaning based on statistical analysis method and data normalization processing.

[0015] As a further improvement of the mineral processing test method based on the multi-data fusion algorithm: the z-score method is used to identify and eliminate outliers during data cleaning;

[0016] During normalization, Min-Max normalization or Z-score normalization is used.

[0017] As a further improvement of the mineral processing test method based on the multi-data fusion algorithm: the multi-data fusion algorithm is based on the weighted average method, the principal component analysis-random forest method or the neural network.

[0018] As a further improvement of the mineral processing test method based on the multi-data fusion algorithm, the method of obtaining the core test parameters based on the weighted average method is: first, the weight of each attribute in the preprocessed data is calculated, and then the estimated value is calculated using the weight, and then based on the estimated value and the target parameter, the closest test case is found from the historical successful test database, and the test parameters in the test case are used as the core test parameters.

[0019] As a further improvement of the mineral processing test method based on the multi-data fusion algorithm, the weight calculation method is as follows: multiple batches of data of each attribute are taken to form a column of data, and the variance of each column is calculated to obtain the variance of each attribute: σ1, σ2...σ n , n is the number of attributes, and then calculate the weight corresponding to each attribute:

[0020]

[0021] W i is the weight of the ith attribute.

[0022] As a further improvement of the mineral processing test method based on the multi-data fusion algorithm, the estimated value is calculated as follows: Where X i is the column data average of the i-th attribute.

[0023] As a further improvement of the mineral processing test method based on multi-data fusion algorithm, the method of finding the closest test case from the historical successful test database is: the obtained estimated value and the target parameter constitute a feature vector, and then use the similarity algorithm to find the test case in the historical successful test database whose feature vector is closest to the aforementioned feature vector, and use it as the closest test case.

[0024] As a further improvement of the mineral processing test method based on the multi-data fusion algorithm, the dynamic adjustment in step 6 includes short-term adjustment and long-term adjustment;

[0025] Short-term adjustment: Analyze the test results regularly. If there is any deviation from the expected target, quickly adjust the test conditions or process parameters through the feedback mechanism.

[0026] Long-term adjustment: Compare the test results with historical test data to determine the optimal test parameters, and then optimize the multi-data fusion algorithm in step 3 based on the comparison results to enhance the accuracy and efficiency of future tests.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1. The present invention realizes intelligent analysis and decision-making based on multi-data fusion algorithm, can complete a large amount of data processing in a short time, and quickly obtain the best mineral processing plan, thereby improving mineral processing efficiency.

[0029] 2. The present invention can reduce the reliance on manual experience, lighten the workload of engineers, save human resources, and reduce the cost waste caused by human errors.

[0030] 3. The present invention is based on comprehensive data analysis, and the optimized mineral processing scheme can effectively improve the recovery rate of ore, thereby increasing the economic benefits of the mine.

[0031] 4. The present invention promotes the sustainable development of the mining industry through more efficient resource utilization and waste reduction, which meets the basic requirements of current social environmental protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Schematic diagram of the process of this method;

[0033] Figure 2 Schematic diagram of weighted fusion algorithm. DETAILED DESCRIPTION

[0034] The technical solution of the present invention is described in detail below with reference to the accompanying drawings:

[0035] A mineral processing test method based on a multi-data fusion algorithm utilizes a large amount of mineral processing test data accumulated in the laboratory, and through comparing the properties of the ore with the mineral processing setting goals, performs automated intelligent evaluation and decision-making to obtain a suitable mineral processing test plan for the ore, thereby greatly reducing the blindness of manually designed experiments, improving the accuracy and efficiency of mineral processing experiments, significantly improving the test quality, shortening the test cycle, and reducing labor and consumables costs.

[0036] like Figure 1 , the method comprises the following steps:

[0037] Step 1: Collect raw ore data and determine the target parameters of the test.

[0038] Specifically, collect the original ore data related to the test, including the ore grade, mineral composition, particle size distribution, mineral embedding, physical and chemical properties, etc. The data sources of the original ore can include on-site exploration, laboratory analysis and historical mineral processing records. The target parameters include the set grade, fineness, recovery rate and output rate of the concentrate. The data source of the concentrate can be set according to the requirements of the subsequent production process.

[0039] Step 2: Preprocess the collected data.

[0040] The preprocessing includes data screening based on threshold values, data cleaning based on statistical analysis methods, and data normalization.

[0041] Data screening: For target minerals, the set thresholds (such as missing values, values ​​greater than or less than a reasonable range, etc.) are applied to screen the data and eliminate obvious invalid data.

[0042] Data cleaning: Use statistical analysis methods (such as the z-score method) to identify and remove outliers to achieve data cleaning.

[0043] Data normalization: For data with different attributes, Min-Max normalization or Z-score normalization is used to normalize the data to the same dimension to ensure that different data contribute consistently to the analysis results.

[0044] Step 3: Use a multi-data fusion algorithm to predict core test parameters based on the collected raw ore data of multiple attributes and target parameters.

[0045] The multi-data fusion algorithm is based on weighted average method, principal component analysis (PCA)-random forest method or neural network, and the appropriate algorithm can be selected according to the target requirements.

[0046] Specifically, the method of obtaining the core test parameters based on the weighted average method is as follows: Figure 2,Firstly calculate the weight of each attribute in the preprocessed data, and then use the weight to calculate the estimated value, and then based on the estimated value and the target parameter, find the closest test case from the historical successful test database, and use the test parameters in the test case as the core test parameters.

[0047] The way to calculate the weight is: take multiple batches of data for each attribute to form a column of data, calculate the variance of each column, and then get the variance of each attribute: σ1, σ2...σ n , n is the number of attributes. Then calculate the weight corresponding to each attribute:

[0048]

[0049] This weight minimizes the root mean square error of the fused estimate, thereby improving the accuracy and reliability of the estimate.

[0050] The estimate is calculated as: Where X i is the column data average of the i-th attribute.

[0051] The method of finding the closest test case from the historical successful test database is: the obtained estimated value and the target parameter constitute a feature vector, and then use a similarity algorithm (such as Euclidean distance, cosine similarity, etc.) to find the test case in the historical successful test database whose feature vector is closest to the aforementioned feature vector, and use it as the closest test case.

[0052] It should be noted that when constructing a historical successful test database, for each test case, the weights and estimates need to be calculated in the above manner, and the target parameters need to be clarified to form a feature vector corresponding to the test case.

[0053] Optionally, in this step, principal component analysis (PCA) may be used to extract main features from the preprocessed data, and then the main features may be input into a random forest model to obtain core test parameters.

[0054] Optionally, in this step, a support vector machine (SVM) algorithm may be used to build a model, and then the preprocessed data is input into the support vector machine model to obtain core test parameters.

[0055] Optionally, in this step, linear regression analysis can be used to determine the main factors affecting the test, and a neural network model can be used to predict the effects that can be achieved by the test parameters, from which core test parameters can be selected.

[0056] Step 4: Based on the selected core test parameters, specify the alternative grinding parameters and mineral processing schemes, and then obtain the test scheme.

[0057] Specifically, according to the screened data and information, a detailed mineral processing test plan is formulated, including specific steps, required equipment, material requirements, environmental settings, etc. The evaluation criteria of the test, such as recovery rate, output rate, grade, etc., are determined, and quantitative indicators are established to ensure the feasibility and effectiveness of the plan.

[0058] Furthermore, the grinding includes any one or more combinations of autogenous grinding, ball grinding, gravel grinding, high pressure roller grinding, damp grinding, rod grinding, jaw crusher, cone crusher, impact crusher, impact crusher and hammer crusher, depending on the particle size distribution of the ore and the fineness of the target ore.

[0059] Furthermore, the mineral processing includes flotation, magnetic separation, gravity separation, electric separation, air separation and their combined processes, and the total number of roughing, cleaning and scavenging will be determined according to parameters such as the original particle size distribution of the ore.

[0060] Step 5: Implement the test plan and collect feedback information.

[0061] During the implementation of the experiment, various parameters and processing results are monitored in real time, and feedback information is collected.

[0062] Step 6: Dynamically adjust the experiment based on the feedback information.

[0063] Dynamic adjustments include short-term adjustments and long-term adjustments.

[0064] Short-term adjustments: Analyze test results regularly. If deviations from expected targets are found, quickly adjust test conditions or process parameters through feedback mechanisms.

[0065] Long-term adjustment: Compare the test results with historical test data to determine the optimal test parameters, and then optimize the multi-data fusion algorithm in step 3 based on the comparison results (such as updating the database, updating the knowledge base, optimizing model parameters, retraining the model, etc.) to enhance the accuracy and efficiency of future tests.

[0066] Through experiments, the optimal parameters for mineral processing under current conditions can be selected, and mineral processing can be carried out based on the optimal parameters.

[0067] It should be noted that this method is only applicable to single mineral sorting. For multi-mineral ores, different databases should be selected for different target minerals for evaluation and decision-making during sorting.

[0068] The parameters of the original ore may be parameters of any stage of the mineral processing process, that is, the method may be used to evaluate the progress of the mineral processing stage.

[0069] Example 1: Optimization of flotation process of lead-zinc ore.

[0070] Step 1: Data collection.

[0071] Data source: Ore sample information of a lead-zinc mine was collected, including the chemical composition of the ore (content of lead, zinc, sulfur, iron and other elements), mineral phases (such as galena, sphalerite, etc.), particle size distribution (-10μm, 10-30μm, 30-50μm), and dry density, humidity and other characteristics of the ore.

[0072] Historical data: collects data from past flotation tests, including the amount of different reagents used, flotation time, stirring rate, recovery rate and other results.

[0073] Step 2: Data preprocessing.

[0074] Clean data: remove samples with missing values ​​and delete data records that do not meet the conditions.

[0075] Standardization processing: All numerical data (such as the content of each element) are standardized to eliminate the influence of different dimensions, and the Z-score standardization method is used to process the data set.

[0076] Step 3: Application of multi-data fusion algorithm.

[0077] Feature extraction: The principal component analysis (PCA) method was used to extract key features and found factors that were strongly correlated with the recovery of lead and zinc, such as the ratio of sphalerite and galena in the ore.

[0078] Modeling: The random forest algorithm was used to establish a recovery prediction model. The input features were ore composition, reagent usage, and flotation parameters. The output was the recovery prediction of lead and zinc.

[0079] Step 4: Experimental design.

[0080] Design optimization plan: Based on the model results, design a new test plan to determine the best dosage of reagents (such as collectors, depressants), optimize the flotation time (15 minutes), and adjust the stirring speed (set to 800 rpm).

[0081] Step 5: Implementation and feedback.

[0082] Conduct tests: Implement optimized flotation tests and record the recovery rate data for each test. The test results show that the recovery rate of lead increased from 65% to 78%, and the recovery rate of zinc increased from 60% to 75%.

[0083] Record feedback: Compare real-time monitoring data with the prediction model, analyze the differences, and confirm the optimization effect of the drug and treatment time.

[0084] Step 6: Dynamic adjustment.

[0085] Subsequent adjustments: During the test, the reagent ratio was adjusted in time for the special ore samples, and the preliminary test results were verified for the second time, ultimately achieving an overall recovery rate of lead-zinc ore of more than 80%.

[0086] Example 2: Optimization of flotation parameters of copper ore.

[0087] Step 1: Data collection.

[0088] Data source: The characteristics of ore samples from a copper mine were collected. The data included ore grade (copper content), mineral composition (such as the ratio of chalcocite and chalcopyrite), particle size, adsorbent usage, operating temperature, pH value, etc.

[0089] Historical records: Organize the results of historical flotation tests, especially the type of flotation reagents, dosage and test year.

[0090] Step 2: Data preprocessing.

[0091] Cleaning and processing: Screen the collected data, delete incomplete or obviously erroneous data, and normalize the data.

[0092] Statistical analysis: Analyze the distribution of data, with special attention to the grade of copper ore and other key factors affecting recovery.

[0093] Step 3: Application of multi-data fusion algorithm

[0094] Model building: A support vector machine (SVM) algorithm was used to build the model and input historical data to predict copper recovery under different flotation parameters.

[0095] Verification and tuning: Optimize the model through cross-validation method to ensure the accuracy and reliability of the model.

[0096] Step 4: Experimental design.

[0097] Optimization design: Arrange an action plan based on the model prediction results, determine the dosage of reagents (such as increasing the dosage of key reagents by 10%), adjust the flotation time (standard time 30 minutes) and the stirring rate (adjusted to 700rpm).

[0098] Step 5: Implementation and feedback.

[0099] Test implementation: Flotation tests were carried out according to the new scheme, and test data were recorded. It was found that the copper recovery rate was increased to 82%.

[0100] Real-time monitoring: Monitor changes in froth quality during the flotation process to ensure that the implemented parameters remain within the optimal range.

[0101] Step 6: Dynamic adjustment.

[0102] Optimization feedback: timely adjustment of reagent formula, found that the flotation effect decreased slightly after a certain flotation cycle, quickly adjusted the reagent ratio, and after another test, the copper recovery rate increased to 85%.

[0103] Example 3: Optimization of tungsten ore dressing process.

[0104] Step 1: Data collection.

[0105] Data basis: At a tungsten mine, sample characteristics were collected, including tungsten content, distribution of various metals in the ore (such as tungsten bismuth ore, quartz, etc.), particle size distribution, and sample processing methods.

[0106] Test history: Statistics of relevant data of historical mineral processing tests, including the output of tungsten concentrate, processing conditions, etc.

[0107] Step 2: Data preprocessing.

[0108] Data cleaning: Eliminate noise and erroneous records in the data to ensure data quality.

[0109] Normalization: Normalize the values ​​of different features to ensure comparability of data analysis.

[0110] Step 3: Application of multi-data fusion algorithm

[0111] Characteristic analysis: Linear regression analysis was used to determine the main factors affecting tungsten recovery, including grinding particle size and flotation reagent dosage.

[0112] Machine Learning Modeling: A neural network model was built to predict tungsten recovery and compared with actual test results.

[0113] Step 4: Experimental design.

[0114] Specific test plan: Design a new mineral processing test based on the model results, including optimizing the grinding particle size (setting the target to 100 mesh), the amount of reagent used (such as increasing the tungsten flotation collector by 20%) and setting the flotation time (set to 12 minutes).

[0115] Step 5: Implementation and feedback.

[0116] Test execution: Conduct tests, record test results, and find that the recovery rate of tungsten concentrate is increased from 67% to 84%.

[0117] Feedback analysis: record test data in real time and evaluate the implementation effect by comparing the recovery rate with the model prediction value.

[0118] Step 6: Dynamic adjustment.

[0119] Optimization and adjustment: When a certain ore sample was found to be special, the reagent ratio was adjusted in real time, and the tungsten recovery rate was stabilized at 86% in subsequent tests.

[0120] in conclusion

[0121] The above three examples show in detail the specific application of this method on different ore types. Each example has clear steps of data collection, preprocessing, modeling, experimental design and dynamic adjustment, and practice has proved the effectiveness and practicality of this method.

Claims

1. A mineral processing test method based on a multi-data fusion algorithm, characterized in that: The steps include: Step 1: Collect raw ore data and determine the target parameters of the test; Step 2: pre-process the collected data; Step 3: Use a multi-data fusion algorithm to predict core test parameters based on the collected raw ore data of multiple attributes and target parameters; Step 4: Based on the selected core test parameters, specify the alternative grinding parameters and mineral processing scheme, and then obtain the test scheme; Step 5: Implement the test plan and collect feedback information; Step 6: Dynamically adjust the experiment based on the feedback information.

2. The ore dressing test method based on multi-data fusion algorithm according to claim 1, characterized in that: In step 1, the original ore data includes the grade, mineral composition, particle size distribution, mineral embedding, and physical and chemical properties of the ore; the target parameters include the set grade, fineness, recovery rate, and output rate of the concentrate.

3. The ore dressing test method based on multi-data fusion algorithm according to claim 1, characterized in that: The preprocessing includes data screening based on threshold values, data cleaning based on statistical analysis methods, and data normalization.

4. The ore dressing test method based on multi-data fusion algorithm according to claim 3 is characterized in that: The z-score method was used to identify and remove outliers during data cleaning; During normalization, Min-Max normalization or Z-score normalization is used.

5. The ore dressing test method based on multi-data fusion algorithm according to claim 1, characterized in that: The multi-data fusion algorithm is based on weighted average method, principal component analysis-random forest method or neural network.

6. The ore dressing test method based on multi-data fusion algorithm according to claim 5, characterized in that: The method of obtaining the core test parameters based on the weighted average method is as follows: first, the weight of each attribute in the preprocessed data is calculated, and then the estimated value is calculated using the weight. Then, based on the estimated value and the target parameter, the closest test case is found from the historical successful test database, and the test parameters in the test case are used as the core test parameters.

7. The ore dressing test method based on multi-data fusion algorithm according to claim 6, characterized in that: The way to calculate the weight is: take multiple batches of data for each attribute to form a column of data, calculate the variance of each column, and then get the variance of each attribute: σ1, σ2...σ n , n is the number of attributes, and then calculate the weight corresponding to each attribute: W i is the weight of the ith attribute.

8. The ore dressing test method based on multi-data fusion algorithm according to claim 7, characterized in that: The estimate is calculated as: Where X i is the column data average of the i-th attribute.

9. The ore dressing test method based on multi-data fusion algorithm according to claim 6, characterized in that: The method of finding the closest test case from the historical successful test database is: the obtained estimated value and the target parameter constitute a feature vector, and then use a similarity algorithm to find the test case in the historical successful test database whose feature vector is closest to the aforementioned feature vector, and use it as the closest test case.

10. The ore dressing test method based on multi-data fusion algorithm according to any one of claims 1 to 9, characterized in that: The dynamic adjustment in step 6 includes short-term adjustment and long-term adjustment; short-term adjustment: regularly analyze the test results, and if deviations from the expected goals are found, quickly adjust the test conditions or process parameters through the feedback mechanism; Long-term adjustment: Compare the test results with historical test data to determine the optimal test parameters, and then optimize the multi-data fusion algorithm in step 3 based on the comparison results to enhance the accuracy and efficiency of future tests.

Images