Production expert system-based iron ore blending optimization method

Abstract

The present invention relates to a production expert system-based iron ore blending optimization method. The method comprises: firstly, performing data enhancement on historical ore blending and beneficiation data by means of an improved generative adversarial network; then, establishing a case library, and by means of a case retrieval method, establishing a relational model between blended ore properties and ore concentrate grades on the basis of the historical ore blending and beneficiation data and expert experience; then, on the basis of material balance, establishing a blended ore property prediction mechanism model to realize blended ore property estimation after ore blending; and finally, by taking an optimal ore beneficiation concentrate grade and a maximum difficult-to-beneficiate ore use amount as optimization objectives, determining constraints on the basis of actual working conditions, obtaining a set of ore blending proportion non-inferior solutions by means of a multi-objective grey wolf algorithm, and deciding an optimal solution by means of a TOPSIS algorithm. The present invention has the following advantages: while ensuring the quality of ore beneficiation products, reasonable ore blending proportions are calculated, the use amount of difficult-to-beneficiate ores is increased, the sustainable mining capacity of mines is improved while ensuring the mining and beneficiation benefits, and the invention has universal applicability.

Description

Iron ore optimization blending method based on production expert system TECHNICAL FIELD

[0001] The present application belongs to the technical field of ore blending between mining and beneficiation, and particularly relates to an iron ore optimization blending method based on a production expert system. BACKGROUND

[0002] The stope of an iron mine contains various types of ore bodies, including both easily selected ores such as magnetic iron and difficult-to-select ores such as iron carbonate. If only the easily selected ores are mined, it is beneficial to improve the concentrate grade of subsequent beneficiation, but it is not conducive to the sustainable development of the mine. On the other hand, if too much difficult-to-select ore is used, it will have an adverse effect on the subsequent beneficiation and the yield of the finished product. Therefore, optimization of the blending scheme before beneficiation operation is beneficial to the sustainable development of the mine operation, and can make full use of the beneficiation capacity of the plant, thereby improving the product quality and overall efficiency.

[0003] Currently, the ore blending work in actual production of the mine stope mainly only considers the total iron grade index of the mixed ore, that is, the ratio of various materials obtained by using the trial-and-error method according to the material balance, without considering the use of difficult-to-select ores and the influence of the amount of difficult-to-select ores on the subsequent beneficiation index under the premise of ensuring the total iron grade index of the ore for beneficiation. The single consideration factor is the disadvantage of the trial-and-error method. Chinese patent CN111738580A is an optimization method for ore blending of Anshan-type iron ore. The invention establishes a blending objective function with the deviation absolute value of the average grade of each ore point in the stope and the given ore grade of the beneficiation plant and the beneficiation index of each ore point as the independent variables, and uses a genetic algorithm to solve the blending objective function. The optimized blending scheme is obtained by outputting the ore quality to the beneficiation plant, which reduces the fluctuation of the beneficiation production index caused by the fluctuation of the ore quality, and realizes the scientific and reasonable ore blending of the stope. Although the invention patent considers the beneficiation index of each ore point, the properties of the ores at each ore point are also different, and the beneficiation index will also change. The main disadvantage is that it only considers the total iron grade index of the mixed ore, and the consideration factors are less and single. Adding the amount of difficult-to-select ores to the ore blending design scheme can better utilize limited mineral resources and improve the product quality of the beneficiation plant and the overall efficiency of the mine.

[0004] SUMMARY

[0005] The present application provides an iron ore optimization blending method based on a production expert system. The present application establishes a soft measurement model of the properties of the mixed ore and the corresponding concentrate grade by an expert system, and reasonably configures difficult-to-select ores in the stope under the premise of ensuring the concentrate grade of beneficiation, which is beneficial to the sustainable development of the mine. In addition, the present application performs multi-objective optimization on the ore feeding amount of each ore point in the stope, focuses on the insufficient data problem of each stope, uses the production method to expand the data, realizes the operation guidance of the ore blending process, enhances the applicability, and has promotional value.

[0006] The object of the present application is achieved by the following technical solutions:

[0007] The iron ore optimization blending method based on the production expert system, characterized in that it comprises the following steps:

[0008] Step 1, collect the historical data of the properties of the mixed ore for beneficiation, including the total iron grade, the content of iron carbonate and the content of ferrous oxide of the mixed ore for beneficiation, and collect the historical data of the concentrate grade after beneficiation of the mixed ore for beneficiation, and establish a database.

[0009] Step 2, check and remove abnormal value samples in the database that do not meet the process requirements, and screen to obtain the optimized data of the historical data of the properties of the mixed ore for beneficiation and the concentrate grade after beneficiation;

[0010] The checking and removing of abnormal value samples in the database that do not meet the process requirements are based on formula (1):

[0011] Wherein, P is a data group; n is the number of data, P i is the i-th data value in P, and the difference between P i and the average of P is greater than or equal to 3 times the standard deviation, and this group of data is removed.

[0012] Step 3, according to the optimized data, a case library is constructed, the optimized data of the historical data of the properties of the mixed ore for beneficiation and the concentrate grade after beneficiation screened in step 2 is expanded by using a generative adversarial network, and a soft measurement model of the properties of the mixed ore for beneficiation and the concentrate grade after beneficiation is established under the premise that the beneficiation process is operated according to the process requirements by using the method of case retrieval;

[0013] The step 3 comprises the following specific steps:

[0014] Step 3.1, according to the optimized data screened in step 2, a case library is constructed, the structure of the cases in the case library is composed of case description and case solution, the case description is composed of the total iron grade f1, the content of magnetic iron f2 and the content of iron carbonate f3, and the case solution is the concentrate grade j1. Therefore, the case based on the case reasoning technology is shown in formula (2):

[0015] C k ={F k ,J K} (2)

[0016] Wherein, C k is the k-th case in the case library, k=1, 2, …, m, m is the number of cases in the case library; F k ={fk,1 ,f k,2 ,f k,3} represents the feature description of the k-th case; J K ={j k,1} represents the solution for the k-th case;

[0017] Step 3.2: Expand the case library using a generative adversarial network, where the generator input is Gaussian mixture noise and the discriminator input is real cases and generated cases;

[0018] The improved loss function is shown in equation (3):

[0019] Where G is the generator and D is the discriminator. To generate samples, x is the real sample, σ i Let λ1 be the i-th Gaussian parameter, N be the number of Gaussian mixed noises, λ1 be the first hyperparameter, and λ2 be the second hyperparameter.

[0020] Step 3.3: Obtain real-time data on the properties of mixed ores as new cases. Perform case retrieval by comparing the new case description with historical cases recorded in the case library and new cases generated by the generative adversarial network to obtain a set of similarity values, as shown in formula (4):

[0021] Where, ω i For the case feature attribute weights, SIM(f) i ,f k.i () is a case description of the current working condition, f i The case description f corresponding to the kth case in the case library k.i The similarity is defined as shown in formula (5):

[0022] Cases from the case library corresponding to the similarity score will be used as reference cases for the current working condition;

[0023] Step 3.4: Sort the obtained similarity values ​​from largest to smallest, and use cross-validation to select the cases in the case library corresponding to the top n similarity values ​​as reference cases for the current working condition;

[0024] Step 3.5: Based on the reference cases retrieved from the case library, reuse the cases to obtain new case solutions, which will be used as the current feed amount for the mixed ore.

[0025] The aforementioned case reuse involves calculating the weighted average of the solutions for each of the n reference cases, where the weighting coefficient is the similarity between each reference case and the new case. This completes the case reuse, yielding a new case solution, which serves as the optimal ore feed rate calculation result for the current working condition. The new case solution is shown in formula (6):

[0026] wherein J is a new case solution, J k is a case solution of a reference case;

[0027] Step 4, determining the number of each ore mining point participating in ore blending and the corresponding ore properties of each ore mining point, including: the total iron grade, the iron carbonate content, the ferrous oxide content, the minimum usage amount, and the maximum usage amount of the ore, and establishing a beneficiation mixed ore property prediction mechanism model according to the ore usage amount and the ore properties of each ore mining point;

[0028] The specific process of establishing the beneficiation mixed ore property prediction mechanism model according to the ore mining amount and the ore properties of each ore mining point is as follows:

[0029] Suppose that k ore mining points participate in ore blending, and the usage amount of each ore mining point is x i , i = 1, …, k, the total iron grade of each ore mining point is Fe i , the ferrous oxide content is FeO i , the iron carbonate content is MCFe i , the total iron grade of the mixed ore is Fe, the ferrous oxide content is FeO, and the iron carbonate content is MCFe. According to the material balance, the mixed ore property prediction mechanism model is as shown in formula group (7):

[0030] Step 5, taking the ore usage amount of each ore mining point as a decision variable, using a multi-objective optimization algorithm to set the highest concentrate grade as the first optimization target and the largest iron carbonate content as the second optimization target, and setting the constraint conditions combined with the actual production process to calculate the most ideal ore usage amount of each ore mining point.

[0031] The step 5 includes the following specific steps:

[0032] Step 5.1, taking the ore usage amount of each ore mining point as a decision variable, calculating the current beneficiation mixed ore properties and the corresponding concentrate grade according to the soft measurement model of step 3 and the beneficiation mixed ore property prediction mechanism model of step 4, determining the ore usage amount x i of each ore mining point as a decision variable, and setting the highest concentrate grade MFe and the largest iron carbonate content CFe as the first optimization target and the second optimization target respectively, as shown in formula group (8):

[0033] Setting the supply capacity of each ore mining point, the upper and lower limits of the total iron grade of the ore, the upper and lower limits of the concentrate grade, and the upper and lower limits of the iron carbonate content as constraint conditions, as shown in formula group (9):

[0034] wherein Qimin and Q imax respectively are the minimum percentage and the maximum percentage of the ore usage of each mining point, Fe min and Fe max respectively are the minimum value and the maximum value of the mixed ore grade, MFe min and MFe max respectively are the minimum value and the maximum value of the concentrate grade, CFe min and CFe max respectively are the minimum value and the maximum value of the iron carbonate content.

[0035] Step 5.2, the multi-objective grey wolf algorithm is used to calculate the optimal solution set of each decision variable, wherein the eigenvalue function is selected according to field experience;

[0036] Step 5.3, the TOPSIS decision method is used to decide the optimal solution from the optimal solution set as the most ideal ore usage ratio of each mining point.

[0037] Compared with the prior art, the advantages of the present application are:

[0038] 1) Compared with the traditional artificial trial and error method which only considers the mixed ore total iron grade index, the present application reduces the probability of unsatisfactory ore blending results caused by the business ability of operators and other problems;

[0039] 2) Compared with Chinese patent CN111738580A which only considers the mixed ore total iron grade index, the present application uses a generative adversarial network for data augmentation, solves the problem of insufficient data in each mine, and enhances the applicability, universality and precision of the application;

[0040] 3) The present application uses more iron carbonate difficult-to-select ore under the premise of ensuring the concentrate grade after beneficiation, realizes scientific and reasonable ore blending of the mining site, and provides mineral resources guarantee for the sustainable development of the mine. BRIEF DESCRIPTION OF DRAWINGS

[0041] Fig. 1 is a technical flow structure block diagram of the present application.

[0042] Fig. 2(a) is the distribution of the first generation of wolf population initialization algorithm.

[0043] Fig. 2(b) is the position of the wolf population after 50 iterations. DETAILED DESCRIPTION

[0044] The present application will be further described below in conjunction with the drawings and examples.

[0045] The specific embodiments of the present application will be further described in detail below in conjunction with the drawings and examples. The following examples are used to illustrate the present application, but are not used to limit the scope of the present application.

[0046] EXAMPLE

[0047] As shown in Figure 1, the iron ore optimization blending method based on the production expert system of the present application comprises the following steps:

[0048] Step 1, collect the historical data of the mixed ore properties, including the total iron grade, the iron carbonate content, the ferrous oxide content, and the historical data of the concentrate grade after beneficiation of the mixed ore;

[0049] In this embodiment, part of the historical data of the total iron grade, the iron carbonate content, the ferrous oxide content and the corresponding concentrate grade after beneficiation is shown in Table 1:

[0050] Table 1: Part of the historical data of the mixed ore properties and the corresponding beneficiation results

[0051] Step 2, check and remove the abnormal value samples in the collected data;

[0052] The collected mixed ore property data and beneficiation result data may have data abnormalities due to improper operation in the production process, information entry errors, etc., which will seriously affect the accuracy of subsequent optimization blending. The present application detects and removes abnormal value samples in the database based on formula (1):

[0053] Wherein, P is the item data group; n is the number of data, P i is the i-th data value in P, and the difference between P i and the average of P is greater than or equal to 3 times the standard deviation.

[0054] The item data group P can be any one of the data group of the total iron grade, the iron carbonate content, the ferrous oxide content, the minimum usage amount and the maximum usage amount of the ore.

[0055] Step 3: According to the preferred data, a case library is constructed, the data collected in steps 1 and 2 are expanded by using a generative adversarial network, and a soft measurement model of the properties of the mixed ore and the concentrate grade after beneficiation is established under the premise that the beneficiation process is operated according to the beneficiation process requirements by using the case retrieval method:

[0056] Step 3.1, the historical data of the properties of the mixed ore for beneficiation and the grade of the concentrate after beneficiation collected and screened in steps 1 and 2 are used to construct a case library. The total iron grade, the ferrous oxide content and the iron carbonate content of the ore are the case description, and the grade of the selected concentrate is the case solution;

[0057] Based on the historical data, production site investigation and theoretical analysis of the beneficiation process, the ore property data can directly reflect the grindability and selectability of the ore. When the concentrate index is constant, the property of the ore has a direct impact on the concentrate grade, and the proportion is large.

[0058] The structure of the case in the case base consists of case description and case solution. The case description is composed of the total iron grade f1, magnetic iron content f2 and iron carbonate content f3, and the case solution is the concentrate grade j1. Therefore, the case based on case-based reasoning technology is shown in formula (2):

[0059] C k ={F k ,J K} (2)

[0060] Where C k is the kth case in the case base, k = 1, 2, …, m, and m is the number of cases in the case base; F k ={f k,1 ,f k,2 ,f k,3} is the feature description of the kth case; J K ={j k,1} is the case solution of the kth case.

[0061] Step 3.2: The case base is expanded by using generative adversarial network to realize case expansion, where the generator input is Gaussian mixed noise to improve sample diversity and avoid case repetition, and the discriminator input is real case and generated case to improve the quality of the generated samples;

[0062] The number of samples directly affects the accuracy of the expert system, and the more cases in the case base can effectively avoid the mesh distance between cases, thereby improving the effectiveness of the scheme given by the expert system. Although the traditional interpolation method can increase the number of samples, it is difficult to generate effective samples that meet the internal rules of the samples. However, generative adversarial network can learn the true distribution of sample data through the binary zero-sum game between generator and discriminator. Due to the small number of training samples in some stope, it is difficult for the original data and generated data to have data overlap in high-dimensional space. In order to avoid the generator falling into gradient disappearance, Gaussian mixed noise is used instead of the original Gaussian or average noise of the generative adversarial network generator, which increases the diversity of the prior distribution of model input data by improving the complexity of random noise, thereby increasing the diversity of generated samples without increasing model parameters.

[0063] The improved loss function is shown in formula (3):

[0064] Where G is the generator, D is the discriminator, To generate samples, x is the real sample, σ i Let λ be the i-th Gaussian parameter, N be the number of Gaussian mixed noises, λ1 be the first hyperparameter, and λ2 be the second hyperparameter. Training employs a momentum-based optimizer (Adam), where the discriminator updates the generator every 5 training iterations to maintain the balance between the discriminator and generator. The initial training sample consisted of 147 groups, which were expanded to 1000 groups.

[0065] Step 3.3: Obtain real-time data on the properties of mixed ores as new cases. Perform case retrieval by comparing the new case description with historical cases recorded in the case library and new cases generated by the generative adversarial network to obtain a set of similarity values; as shown in formula (4):

[0066] Where, ω i For the case feature attribute weights, SIM(f) i ,f k.i () is a case description of the current working condition, f i The case description f corresponding to the kth case in the case library k.i The similarity is defined as shown in formula (5):

[0067] Cases from the similarity-corresponding case library are used as reference cases for the current working conditions;

[0068] Step 3.4: Sort the obtained similarity values ​​from largest to smallest, and use cross-validation to select the cases in the case library corresponding to the top n similarity values ​​as reference cases for the current working condition;

[0069] The choice of the value of n has a significant impact on the algorithm's results. A smaller n value means that only historical cases close to the new case will affect the solution for the new case, but this can easily lead to overfitting. If the n value is larger, the advantage is that it can reduce the estimation error of the learning process, but the disadvantage is that the approximation error of the learning process increases, and historical cases far removed from the new case will also affect the prediction, causing the prediction to be incorrect. In practical applications, a smaller n value is generally chosen, and cross-validation is usually used to select the optimal n value. In this embodiment, n is set to 5.

[0070] Step 3.5: Based on the reference cases retrieved from the case library, reuse the cases to obtain new case solutions, which will be used as the ore input quantity corresponding to the current mixed ore.

[0071] According to the reference cases retrieved from the case library, a weighted average of each case solution of the n reference cases is calculated, wherein the weighting coefficient is the similarity of each reference case to the new case, thereby completing case reuse and obtaining a new case solution as the calculation result of the optimal selected ore quantity of the current working condition; the new case solution is shown in formula (6):

[0072] wherein J is the new case solution, J k is the case solution of the reference case.

[0073] Step 4: Determine the number of ore extraction points participating in ore blending and the properties of the ore corresponding to each ore bin, including: the total iron grade, the iron carbonate content, the ferrous oxide content, the minimum usage amount, and the maximum usage amount of the ore, and based on the material balance principle, a mixed ore property prediction mechanism model is established according to the ore usage amount of each ore extraction point and the ore properties;

[0074] Set k ore extraction points participating in ore blending, and the usage amount of each ore extraction point is x i , i = 1, …, k, the total iron grade of each ore bin is Fe i , the ferrous oxide content is FeO i , the iron carbonate content is MCFe i , the total iron grade of the mixed ore is Fe, the ferrous oxide content is FeO, and the iron carbonate content is MCFe. According to the material balance, the mixed ore property prediction mechanism model is shown in formula group (7):

[0075] The obtained real-time data of the ore properties in the embodiment are shown in Table 2:

[0076] Table 2 Input ore properties

[0077] Step 5: Take the ore usage amount of each ore extraction point as the decision variable, use a multi-objective optimization algorithm to set the highest concentrate grade as the first optimization target and the largest iron carbonate content as the second optimization target, and set the constraint conditions in combination with the actual production process to calculate the most ideal ore usage amount of each ore extraction point;

[0078] The step 5 includes the following specific steps:

[0079] Step 5.1, taking the ore usage amount of each ore extraction point as the decision variable, calculating the current mixed ore property and the corresponding concentrate grade according to the soft measurement model of step 3 and the mixed ore property prediction mechanism model of step 4, determining the ore usage amount x i of each ore extraction point as the decision variable, and setting the highest concentrate grade MFe and the largest iron carbonate content CFe as the first optimization target and the second optimization target respectively, as shown in formula group (8):

[0080] The supply capacity of each mining point, the upper and lower limits of the ore total iron grade, the upper and lower limits of the concentrate grade and the upper and lower limits of the iron carbonate content are set as constraint conditions, as shown in formula set (9):

[0081] Wherein, Q imin and Q imax are the minimum percentage and the maximum percentage of the ore usage amount of each mining point, Fe min and Fe max are the minimum value and the maximum value of the mixed ore grade, MFe min and MFe max are the minimum value and the maximum value of the concentrate grade, CFe min and CFe max are the minimum value and the maximum value of the iron carbonate content.

[0082] In this embodiment, the constraint conditions given in combination with the corresponding working conditions of the field situation are shown in Table 3:

[0083] Table 3 Constraint conditions corresponding to field situation

[0084] Step 5.2, the multi-objective grey wolf optimization (MOGWO) is used to calculate the optimal solution set of each decision variable, wherein the eigenvalue function is selected according to field experience;

[0085] Specifically, the maximum number of iterations is 1000, the population size is 300, the convergence factor is 2, the attraction factor is between 0 and 1, the crowding distance is 0.5, and the initial population position is

-10, 10

[0086] The two optimization objectives are normalized to reduce the influence of the order of magnitude on the optimization weight. In this embodiment, the initial population size is 100, and the grey wolf position is updated, and the results of the first generation and iteration 50 times are shown in Figure 2.

[0087] Figure 2(a) is the distribution of the first generation of wolf population after the leader is selected and the wolf population position is updated, and Figure 2(b) is the wolf population position after iteration 50 times. The non-inferior solution set of the proportion of the ore amount used by each mining point obtained is shown in Table 4:

[0088] Table 4 Optimal non-inferior solution set

[0089] Step 5.3, the TOPSIS decision method is used to decide the optimal solution from the optimal solution set as the most ideal ore usage amount proportion of each mining point.

[0090] In this embodiment, the TOPSIS method is used to make decision for the non-inferior solution set, and the final operation guidance is shown in Table 5.

[0091] Table 5: Optimal result data table

[0092] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not to limit it; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope defined by the claims of the present application.

Claims

1. A method for iron ore blending optimization based on a production expert system, characterized in that: The method comprises the following steps: Step 1, collect the historical data of the properties of the mixed ore, including the total iron grade, the iron carbonate content and the ferrous oxide content of the mixed ore, and collect the corresponding concentrate grade data after beneficiation of the mixed ore, and establish a database; Step 2, check and remove abnormal value samples in the database that do not meet the process requirements, and screen to obtain optimized data of the historical data of the properties of the mixed ore and the historical data of the concentrate grade after beneficiation; Step 3, according to the optimized data, a case library is constructed, the optimized data is expanded by using a generative adversarial network, and a soft measurement model of the properties of the mixed ore and the concentrate grade is established by using a case retrieval method under the premise of operating according to the process requirements in the beneficiation process; Step 4, determine the number of each ore mining point participating in ore blending and the corresponding properties of the ore, including the total iron grade, the iron carbonate content, the ferrous oxide content, the minimum usage amount and the maximum usage amount, and establish a mixed ore property prediction mechanism model according to the ore usage amount and the ore properties of each ore mining point; Step 5, based on the soft measurement model of the properties of the mixed ore and the concentrate grade and the mixed ore property prediction mechanism model, taking the ore usage amount of each ore mining point as a decision variable, using a multi-objective optimization algorithm to set the highest concentrate grade as the first optimization target and the largest iron carbonate content as the second optimization target, and setting the constraint conditions according to the actual production process, the most ideal ore usage amount of each ore mining point is calculated.

2. The iron ore optimization blending method based on the production expert system according to claim 1, characterized in that: The step 2 comprises the following specific steps: The abnormal value samples not meeting the process requirements in the database are checked and removed, based on formula (1): P is a term data group; n is the number of data, P i is the i-th data value in P, P i When the difference between the average and P is greater than or equal to 3 times the standard deviation, the data is rejected; thus, the data is screened to obtain the input The optimized data of the properties of the mixed ore and the concentrate grade after beneficiation.

3. The iron ore optimization blending method based on the production expert system according to claim 1, characterized in that: The step 3 comprises the following specific steps: Step 3.1, according to the optimized data screened in step 2, a case library is constructed, the structure of the cases in the case library is composed of case description and case solution, the case description is composed of the total iron grade f1, the magnetic iron content f2 and the iron carbonate content f3, and the case solution is the concentrate grade j1. Therefore, the case based on the case reasoning technology is shown in formula (2): C k = {F k J K} (2) Wherein, C k is the kth case in the case base, k = 1, 2, …, m, m is the number of cases in the case base; F k = {f k,1 ,f k,2 ,f k,3} is the feature description of the kth case; J K = {j k,1} is the solution of the kth case. Step 3.2, the case library is expanded by using a generative adversarial network, wherein the generator input is a Gaussian mixed noise, and the discriminator input is a real case and a generated case; The improved loss function is shown in equation (3): wherein G is a generator and D is a discriminator, To generate the samples, x is the real sample, σ i is the i-th Gaussian parameter; Step 3.3, obtain real-time data of the mixed ore properties as a new case, perform case retrieval of the new case description with the historical cases recorded in the case library to obtain a set of similarity values, as shown in formula (4): where ω i is the weight of the case feature attribute, SIM(f i ,f k.i ) is the case description of the current working condition, f i is the similarity between the case description f k.i corresponding to the kth case in the case base and the case description of the current working condition, and the similarity is defined as shown in equation (5): The case in the case library corresponding to the similarity is taken as the reference case of the current working condition; Step 3.4, the obtained similarity values are arranged from large to small, and the cross-validation method is used to select the cases in the case library corresponding to the first n similarities as the reference cases of the current working condition; Step 3.5, according to the reference cases retrieved from the case library, case reuse is performed to obtain a new case solution as the beneficiation amount corresponding to the current mixed ore; The case reuse is performed by seeking a weighted average of each case solution of n reference cases, wherein the weighted coefficient is the similarity of each reference case to the new case, thereby completing the case reuse to obtain a new case solution as the optimal selected ore quantity calculation result of the current working condition, and the new case solution is shown in formula (6): where J is the new case solution, J k is the case solution of the reference case.

4. The iron ore optimization blending method based on the production expert system according to claim 1, characterized in that: The process of establishing the mixed ore property prediction mechanism model according to the ore mining amount and the ore properties of each ore mining point is as follows: k mining points are set to participate in blending, and the usage of each mining point is x i , i = 1, …, k, the total iron grade of each mining point is Fe i , the content of ferrous oxide is FeO i , the content of iron carbonate is MCFe i , the total iron grade of the mixed ore is Fe, the content of ferrous oxide is FeO, and the content of iron carbonate is MCFe. According to the material balance, the mixed ore property prediction mechanism model is shown in formula group (7):

5. The iron ore optimization blending method based on the production expert system according to claim 1, characterized in that: The step 5 comprises the following specific steps: Step 5.1, taking the ore quantity of each mining point as the decision variable, calculating the current mixed ore properties and the corresponding concentrate grade according to the soft measurement model of step 3 and the mixed ore property prediction mechanism model of step 4, and determining the ore quantity x of each mining point i Taking the ore quantity of each mining point as the decision variable, and setting the highest concentrate grade MFe and the largest iron carbonate content CFe as the first optimization target and the second optimization target respectively, as shown in formula group (8): The supply capacity of each ore mining point, the upper and lower limits of the total iron grade of the ore, the upper and lower limits of the concentrate grade, and the upper and lower limits of the iron carbonate content are set as constraint conditions, as shown in the formula group (9): wherein Q imin and Q imax are the minimum and maximum percentage of ore used from each mine site, respectively, Fe min and Fe max are the minimum and maximum grade of the mixed ore, respectively, MFe min and MFe max are the minimum and maximum grade of the concentrate, respectively, CFe min and CFe max are the minimum and maximum content of iron carbonate, respectively. Step 5.2, the multi-objective grey wolf algorithm is used to calculate the optimal solution set of each decision variable, wherein the characteristic value function is selected according to the field experience; Step 5.3, the TOPSIS decision method is used to decide the optimal solution from the optimal solution set as the most ideal ore usage amount of each ore mining point.

Images