Iron ore sorting big data analysis and decision-making system

Abstract

The present invention provides an iron ore sorting big data analysis decision system, which is divided into a storage unit, a computing unit and a decision unit. The storage unit stores historical ore data, the computing unit calculates the relative content of iron and copper based on the ore data, and the decision unit constructs a multi-level decision tree to judge the order of selected parameters by information gain and give appropriate values, and apply such values ​​to future iron ore sorting. The present invention is based on big data calculation and combines relevant algorithms of artificial intelligence to realize the intelligent and automatic selection of values ​​in iron ore sorting.

Description

Technical Field

[0001] The invention relates to an iron ore sorting big data analysis decision-making system. Background Art

[0002] In recent years, with the rapid development of my country's energy industry, the consumption of iron ore has increased significantly. It should be noted that iron ore, as an intermediate product, cannot be directly used in industrial manufacturing. It must be sorted and refined before it can be effectively used in industrial production.

[0003] The current iron ore sorting process is still relatively extensive. The sorting of iron ore involves the application of multiple parameters. If the parameters are not selected correctly, it is likely to seriously affect the final grade of the iron ore after sorting, resulting in a large amount of ore waste.

[0004] In practice, the selection of such sorting parameters often still needs to rely on the experience of technical workers. So far, no scientific method has been proposed for intelligent selection, which leads to reduced work efficiency and seriously affects the final grade of iron ore after sorting, causing great waste and thus pushing up the corresponding raw material costs. Summary of the invention

[0005] The present invention provides an iron ore sorting big data analysis and decision-making system, which effectively solves the above-mentioned technical problems existing in the prior art.

[0006] Specifically, the present invention provides an iron ore sorting big data analysis and decision-making system. In the iron ore sorting process, firstly, the iron concentrate is screened out from the raw ore by magnetic separation with a specific magnetic field strength and the primary residual ore is left. Then, the copper concentrate is screened out from the primary residual ore by flotation with a specific pulp mass concentration and the secondary residual ore is left. Then, the tailings are screened out from the secondary residual ore by a specific number of sweeping selections. The system includes a storage unit, a computing unit, and a decision-making unit. The storage unit stores ore data of N iron ore sorting processes in history, wherein N≥5, Each ore data includes parameter data, i.e., magnetic field strength used in magnetic separation, pulp mass concentration used in flotation, and number of sweeping times used in sweeping, and also includes result data, i.e., iron content and copper content in iron concentrate, iron content and copper content in copper concentrate, and iron content and copper content in tailings. The basic iron content and basic copper content of the original ore are set in the calculation unit, and the relative iron and copper content of each ore data is calculated based on the basic iron content and basic copper content, so as to obtain the relative iron and copper content D1, D2, ... D of the ore data of N times. N , select the maximum value D MAX =Max{D1, D2, ... D N} and the minimum value D MIN =Min{D1, D2, ... D N}, forming the content interval [DMIN , D MAX , the decision-making unit first divides the content interval into three equal parts to form three sub-content intervals, uses the N ore data as the total data set, classifies the three sub-content intervals to obtain the empirical entropy H(D). Subsequently, a decision tree is constructed by classifying the total data set with any one of the three parameters, namely magnetic field strength, pulp mass concentration, and scavenging times, denoted as parameter T. In this decision tree, the root node is the classification of parameter T, and the leaf nodes are the classifications of the three sub-content intervals, obtaining the empirical conditional entropy H(D|T). Thus, the information gain ζ of parameter T is obtained as ζ = H(D|T) - H(D), that is, the information gain ζ1 regarding the magnetic field strength, the information gain ζ2 regarding the pulp mass concentration, and the information gain ζ3 regarding the scavenging times are calculated. The information gain ζ* with the largest value among ζ1, ζ2, and ζ3 is selected, and based on the information gain ζ*, a first selected parameter T1 corresponding to it is determined. In the decision tree corresponding to the first selected parameter T1, the leaf nodes [D MAX -(D MAX -D MIN ) / 3, D MAX are backtracked and selected. All the values of the first selected parameter T1 under this condition are obtained. Suppose the number of values of T1 is N1, then these N1 values of T1 correspond to N1 ore data, where N1 < N. Therefore, the average value of these N1 values of T1 is obtained as the value of the first selected parameter N1'. Using these N1 ore data as the secondary total data set, a secondary decision tree is constructed for the other two parameters except the selected parameter T1 based on this secondary total data set, and the information gains ζ1' and ζ2' of the other two parameters are calculated. If ζ1' > ζ2', then a secondary selected parameter T2 corresponding to ζ1' is determined, and in the decision tree, the leaf nodes [D MAX -(D MAX -D MIN ) / 3, D MAX are backtracked and selected. All the values of the secondary selected parameter T2 under this condition are obtained. Suppose the number of these values is N2, where N2 < N1. Therefore, the average value of these N2 values of T2 is obtained as the value of the secondary selected parameter N2'. Using the N2 ore data as the tertiary total data set, a tertiary decision tree is constructed for the remaining one parameter T3 except the selected parameters T1 and T2 based on this tertiary total data set. In the tertiary decision tree, the leaf nodes [D MAX -(D MAX -D MIN ) / 3, D MAXFor all the values of the third selected parameter T3 below, there are N3 such values, where N3 < N2. Thus, the average value of these N3 T3 parameter values is obtained as the value of the third selected parameter N3'. During the proposed (N + 1)-th iron ore separation process, the decision-making module assigns the three parameters of the magnetic field intensity used in magnetic separation, the pulp mass concentration used in flotation, and the number of scavenging operations used in scavenging to their respective corresponding values N1', N2', and N3'.

[0007] Preferably, the basic iron content is set as A0, and the basic copper content is set as B0. In each ore data, the mass ratio of iron concentrate to raw ore is k1, the iron content in the iron concentrate is A1 and the copper content is B1, the mass ratio of copper concentrate to raw ore is k2, the iron content in the copper concentrate is A2 and the copper content is B2, the mass ratio of tailings to raw ore is k3, and the iron content in the tailings is A3 and the copper content is B3. Thus, the iron-copper relative content D in this ore data is calculated as follows:

[0008]

[0009] Thus, the iron-copper relative contents D1, D2,..., D of N ore data are obtained N , and then the maximum value D MAX = Max{D1, D2,..., D N} and the minimum value D MIN = Min{D1, D2,..., D N} are selected from these N iron-copper relative content values.

[0010] Preferably, the decision-making unit uses the N ore data as the total data set and classifies them into the above three sub-content intervals. Thus, the empirical entropy is obtained, and its calculation formula is:

[0011]

[0012] where K represents the numerical value of the number of sub-content intervals, that is, in the above summation formula, K takes values from 1 to 3, and N k is the number of samples corresponding to each sub-content interval.

[0013] Preferably, in the classification of the parameter T as the root node, there are N parameter values T1, T2,..., T of the parameter T corresponding to the ore data of N iron ore separations N . Then, the maximum value T MAX = Max{T1, T2,..., T N} and the minimum value T MIN = Min{T1, T2,..., T N} are obtained from these N parameter values. The parameter interval [T MIN , T MAXbe divided into three equal parts, that is, three sub-parameter intervals [T MIN , T MIN +(T MAX -T MIN ) / 3], [T MIN +(T MAX -T MIN ) / 3, T MAX -(T MAX -T MIN ) / 3], [T MAX -(T MAX -T MIN ) / 3, T MAX are formed. The classification of the root node for the parameter T is based on the above three sub-parameter intervals to classify the total data set.

[0014] Preferably, the calculation formula of the empirical conditional entropy is as follows:

[0015]

[0016] Among them, H(D|T) represents the empirical conditional entropy of the total data set classified with the sub-parameter interval as the root node and the sub-content interval as the leaf node, H(Di) represents the empirical entropy under each root node, Ti represents the number of each classification under a specific sub-parameter interval classification, and Dik represents the number of samples under each leaf node classification.

[0017] Preferably, a secondary decision tree is constructed for the other two parameters except the selected parameter T1 based on the secondary total data set. The secondary decision tree uses the sub-parameter interval as the root node and the sub-content interval as the leaf node for the other two parameters under the secondary total data set. Thus, the information gains ζ1' and ζ2' of the other two parameters are calculated. If ζ1'>ζ2', then based on ζ1', a corresponding secondary selected parameter T2 is determined, and the leaf node [D MAX -(D MAX -D MIN ) / 3, D MAX is backtracked and selected in the decision tree. The values of all secondary selected parameters T2 under this leaf node are set. There are N2 such values, where N2 < N1. Thus, the average value of these N2 T2 parameter values is obtained as the value of the secondary selected parameter N2'.

[0018] In summary, the present invention provides an iron ore sorting big data analysis decision system, which is divided into a storage unit, a computing unit and a decision unit. The storage unit stores historical ore data, the computing unit calculates the relative content of iron and copper based on the ore data, and the decision unit constructs a multi-level decision tree to judge the order of selected parameters by information gain and give appropriate values, and apply such values ​​to future iron ore sorting. The present invention is based on big data calculation and combines relevant algorithms of artificial intelligence to realize the intelligent and automatic selection of values ​​in iron ore sorting. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be discussed below. Obviously, the technical solutions described in conjunction with the drawings are only some embodiments of the present invention. For ordinary technicians in this field, other embodiments and their drawings can be obtained based on the embodiments shown in these drawings without paying creative work.

[0020] Figure 1 The operating flow chart of the iron ore sorting big data analysis and decision-making system according to the present invention is shown. DETAILED DESCRIPTION

[0021] The technical solutions of various embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments described in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] In summary, the present invention provides an iron ore sorting big data analysis and decision-making system, which uses the comprehensive grade of various result products to classify the historical data of iron ore sorting, thereby obtaining the classification results, and adjusts the process parameters of various result products based on the classification results, thereby gradually approaching the most suitable parameters to achieve the best sorting effect.

[0023] Figure 1 The operating flow chart of the iron ore sorting big data analysis and decision-making system according to the present invention is shown.

[0024] In the iron ore sorting big data analysis and decision-making system provided by the present invention, the raw ore is first separated into iron concentrate by magnetic separation. The so-called "magnetic separation", as the name implies, is to use a magnetic field to screen out the ferromagnetic substances in the raw ore. The screened substances are iron concentrates, which are rich in iron and contain a small amount of copper. For example, in iron concentrates, the iron content can reach 60% to 70%, and the copper content may reach 0.05% to 0.5%.

[0025] The key parameter in the magnetic separation process is the magnetic field strength, which can generally reach 100A / m to 200A / m. However, it should be noted that to ensure the high quality of iron ore concentrate, the magnetic field strength is not the greater the better. Due to the complexity of the ore composition, excessive magnetic field strength will interfere with the ore during the magnetic separation process, thereby bringing in unnecessary impurities after magnetic separation.

[0026] After magnetic separation, the remaining first residue ore is removed from the iron concentrate and then flotation is carried out. The first residue ore is introduced into water containing bubbles and stirred, so as to take away the designated ore powder and achieve the purpose of ore dressing.

[0027] The key to flotation is to introduce flotation adjusters to control the pulp mass concentration. Depending on the site, the pulp mass concentration can be adjusted between 20% and 40%.

[0028] Copper concentrate can be screened from the first residue by flotation. It should be noted that since the original ore is iron ore, although copper concentrate is named with "copper" as the first letter, its iron content is still the majority, but the copper content in copper concentrate is greatly increased compared with iron concentrate. For example, in copper concentrate, the iron content can reach 30% to 40%, while the copper content can reach 10% to 20%.

[0029] After the primary residue is removed from the copper concentrate by flotation, the secondary residue is scavenged. After multiple scavenging, the final tailings can be obtained. Obviously, after magnetic separation and flotation, the quality of the tailings obtained by scavenging is obviously not comparable to that of iron concentrate and copper concentrate. The iron content in the tailings is generally only 10% to 15%, and the copper content generally does not exceed 0.5%.

[0030] The quality of tailings obtained by scavenging is often closely related to the number of scavenging times.

[0031] In the iron ore sorting big data analysis and decision-making system provided by the present invention, a storage unit is first set up, which stores ore data of N iron ore sortings in history (generally speaking, the integer N≥5 is required). The ore data of each time includes parameter data, that is, the magnetic field strength used in magnetic separation, the slurry mass concentration used in flotation, and the number of sweeps used in sweeping, and also includes result data, that is, the iron content and copper content in iron concentrate, the iron content and copper content in copper concentrate, and the iron content and copper content in tailings.

[0032] Next, a calculation unit is also provided in the system, in which the basic iron content and basic copper content of the original ore are set, and based on this, the relative iron and copper content of each ore data is calculated.

[0033] The following will describe the calculation process of the relative iron and copper content of each ore data.

[0034] Assume the basic iron content is A0, the basic copper content is B0, in each ore data, the mass proportion of iron concentrate to the original ore is k1, the iron content in the iron concentrate is A1 and the copper content is B1, the mass proportion of copper concentrate to the original ore is k2, the iron content in the copper concentrate is A2 and the copper content is B2, the mass proportion of tailings to the original ore is k3, the iron content in the tailings is A3 and the copper content is B3, and the relative content of iron and copper D in the ore data is calculated:

[0035]

[0036] Thus, the relative content of iron and copper in N ore data D1, D2, ... D N , and then select the maximum value D among the N values ​​of the relative content of iron and copper MAX =Max{D1, D2, ... D N} and the minimum value D MIN =Min{D1, D2, ... D N}.

[0037] The above calculation of the relative content of iron and copper effectively integrates the iron content and copper content of the three intermediate products (iron concentrate, copper concentrate, and tailings), that is, six different attributes, into one indicator (i.e., the relative content of iron and copper), thus laying the foundation for subsequent analysis and decision-making.

[0038] Next, the system also sets up a decision-making unit, which converts the content interval [D MIN ,D MAX ] is divided into three equal parts, that is, forming three sub-content intervals [D MIN ,D MIN +(D MAX -D MIN ) / 3]、[D MIN

[0039] +(D MAX -D MIN ) / 3,D MAX -(D MAX -D MIN ) / 3]、[D MAX -(D MAX -D MIN ) / 3,D MAX ].

[0040] Furthermore, the decision-making unit uses N times of ore data as the total data set and classifies them according to the above three sub-content intervals to obtain the empirical entropy, which is calculated as follows:

[0041]

[0042] Wherein, K represents the numerical value of the number of sub-content intervals, that is, in the above summation formula, K ranges from 1 to 3, N k is the number of samples corresponding to each sub-content interval.

[0043] Subsequently, the data set is classified according to any one of the three parameters, namely, the magnetic field intensity used in magnetic separation, the pulp mass concentration used in flotation, and the number of sweeping times used in scavenging (hereinafter written as T for convenience) to construct a decision tree, in which the root node is the classification of parameter T and the leaf nodes are the classifications of the three sub-content intervals.

[0044] In the classification of parameter T as the root node, there are N parameter values ​​T1, T2, ..., T of parameter T corresponding to the ore data of N iron ore sorting. N , so find the maximum value T among these N parameter values MAX =Max{T1, T2, ..., T N} and the minimum value T MIN =Min{T1, T2, ..., T N}. Set the parameter interval [T MIN ,T MAX ] is divided into three equal parts, that is, forming three sub-parameter intervals [T MIN ,T MIN

[0045] +(T MAX -T MIN ) / 3]、[T MIN +(T MAX -T MIN ) / 3,T MAX -(T MAX -T MIN ) / 3]、[T MAX -(T MAX -T MIN ) / 3, T MAX ]. Therefore, the classification of parameter T by the above root node can classify the total data set based on the above three sub-parameter intervals.

[0046] In the above decision tree, find the empirical conditional entropy of parameter T:

[0047] The calculation formula of the empirical conditional entropy is as follows:

[0048]

[0049] Among them, H(D|T) represents the empirical conditional entropy of the total data set classified with the sub-parameter interval as the root node and the sub-content interval as the leaf node, H(Di) represents the empirical entropy under each root node, Ti represents the number of each classification under the specific sub-parameter interval classification, and Dik represents the number of samples under each leaf node classification.

[0050] Further obtain the information gain ζ of parameter T as ζ = H(D|T) - H(D). Thus, the information gain ζ1 regarding the magnetic field strength used in magnetic separation, the information gain ζ2 regarding the pulp mass concentration used in flotation, and the information gain ζ3 regarding the number of scavenging times used in scavenging can be calculated.

[0051] Then compare the numerical magnitudes of the three information gains ζ1, ζ2, and ζ3, select the information gain ζ* with the largest numerical value among ζ1, ζ2, and ζ3, and determine a first selected parameter T1 corresponding to it based on the information gain ζ*, that is, one of the three parameters: the magnetic field strength used in magnetic separation, the pulp mass concentration used in flotation, and the number of scavenging times used in scavenging.

[0052] In the decision tree corresponding to the first selected parameter T1, backtrack to select all the numerical values of the first selected parameter T1 under the leaf node [D MAX -(D MAX -D MIN ) / 3, D MAX . Suppose there are N1 such numerical values. Then these N1 numerical values of T1 correspond to N1 ore data, where N1 < N. Thus, the average value of these N1 numerical values of T1 is obtained as the numerical value of the first selected parameter N1'.

[0053] Next, by analogy, using the N1 ore data as the secondary total data set, construct a secondary decision tree for the other two parameters except the selected parameter T1 based on this secondary total data set. This secondary decision tree uses the sub-parameter interval as the root node and the sub-content interval as the leaf node for the other two parameters under the secondary total data set. Thus, calculate the information gains ζ1' and ζ2' of the other two parameters. If ζ1' > ζ2', then determine a secondary selected parameter T2 corresponding to it based on ζ1', and backtrack to select the leaf node [D MAX -(D MAX -D MIN ) / 3, D MAX in the decision tree and select all the numerical values of the secondary selected parameter T2. Suppose there are N2 such numerical values, where N2 < N1. Thus, the average value of these N2 numerical values of T2 is obtained as the numerical value of the secondary selected parameter N2'.

[0054] Next, by analogy, using the N2 ore data as the total set of three - time data, a three - level decision tree is constructed for the remaining one parameter T3 except the selected parameters T1 and T2 based on this total set of three - time data. The three - level decision tree uses sub - parameter intervals as root nodes and sub - content intervals as leaf nodes for parameter T3 under the total set of three - time data, and selects the leaf nodes [D MAX -(D MAX -D MIN ) / 3, D MAX . All the values of the three - time selected parameter T3 under these leaf nodes are set. Suppose there are N3 such values, where N3 < N2. Then, the average value of these N3 values of parameter T3 is obtained as the value of the three - time selected parameter, which is N3’.

[0055] It should be noted that, different from the previous two selected parameters T1 and T2, there is no process of information - gain comparison for the selection of the value of parameter T3. Because at this time, only the value of one parameter needs to be selected, and one parameter means there is no need to compare or make a choice, and it can be determined by itself.

[0056] In the planned (N + 1) - th iron - ore separation process, the decision - making module assigns the values N1’, N2’, and N3’ corresponding to the magnetic - field intensity used in magnetic separation, the pulp mass concentration used in flotation, and the number of scavenging times used in scavenging to their respective parameters, thereby realizing the selection of parameters for future iron - ore separation.

[0057] So far, the present invention has been basically described. Generally speaking, the present invention provides an iron - ore separation big - data analysis and decision - making system, which is divided into a storage unit, an operation unit, and a decision - making unit. The storage unit stores historical ore data. The operation unit calculates the relative content of iron and copper based on the ore data. The decision - making unit constructs a multi - level decision tree to judge the order of the selected parameters by information gain and gives appropriate values, and applies such values to future iron - ore separation. The present invention relies on big - data calculation, combines relevant algorithms of artificial intelligence, and realizes the intelligence and automation of value selection in iron - ore separation.

[0058] The above are only exemplary embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the scope of protection of the present invention.

Claims

1. An iron ore sorting big data analysis and decision-making system, in the iron ore sorting process, firstly, the iron concentrate is screened out from the raw ore by magnetic separation with a specific magnetic field strength and the primary residual ore is left, then the copper concentrate is screened out from the primary residual ore by flotation with a specific pulp mass concentration and the secondary residual ore is left, and then the tailings are screened out from the secondary residual ore by a specific number of scavenging, characterized in that: The system includes a storage unit, a computing unit, and a decision unit. The storage unit stores ore data from N iron ore sorting processes in history, wherein N≥5. Each ore data includes parameter data, i.e., magnetic field strength used in magnetic separation, pulp mass concentration used in flotation, and the number of sweeping times used in sweeping, and also includes result data, i.e., iron content and copper content in iron concentrate, iron content and copper content in copper concentrate, and iron content and copper content in tailings. The basic iron content and basic copper content of the raw ore are set in the calculation unit, and the relative iron and copper content of each ore data is calculated based on the basic iron content and basic copper content, so as to obtain the relative iron and copper content D1, D2, ... D of the N ore data. N , select the maximum value D MAX =Max{D1, D2, ... D N } and the minimum value D MIN =Min{D1, D2, ... D N }, forming the content interval [D MIN ,D MAX ], The decision unit first divides the content interval into three equal parts to form three sub-content intervals, and uses N times of ore data as the total data set to classify the three sub-content intervals to obtain the empirical entropy H(D). Subsequently, the total data set is classified using any one of the three parameters, namely magnetic field strength, pulp mass concentration, and scavenging times, denoted as parameter T, to construct a decision tree. In this decision tree, the root node is the classification of parameter T, and the leaf nodes are the classifications of three sub-content intervals. The empirical conditional entropy H(D|T) is obtained, and thus the information gain ζ of parameter T is obtained as ζ = H(D|T) - H(D). That is, the information gain ζ1 regarding the magnetic field strength, the information gain ζ2 regarding the pulp mass concentration, and the information gain ζ3 regarding the scavenging times are calculated. The information gain ζ* with the largest value among ζ1, ζ2, and ζ3 is selected, and based on the information gain ζ*, a first selected parameter T1 corresponding to it is determined. In the decision tree corresponding to the first selected parameter T1, the leaf node [D MAX -(D MAX -D MIN ) / 3, D MAX is backtracked and selected. All the values of the first selected parameter T1 under this leaf node are obtained. Suppose there are N1 values of T1. Then these N1 values of T1 correspond to N1 ore data, where N1 < N. Thus, the average value of these N1 values of T1 is obtained, and the value of the first selected parameter is N1’. Using the N1 ore data as the total secondary data set, a secondary decision tree is constructed for the other two parameters except the selected parameter T1 based on this total secondary data set, and the information gains ζ1’ and ζ2’ of the other two parameters are calculated. If ζ1’ > ζ2’, then a corresponding secondary selected parameter T2 is determined based on ζ1’, and the leaf nodes are backtracked and selected in the decision tree MAX -(D MAX -D MIN ) / 3, D MAX of all the numerical values of the secondary selected parameter T2. Suppose there are N2 such numerical values, where N2 < N1. Thus, the average value of these N2 numerical values of the T2 parameter is obtained as the numerical value of the secondary selected parameter N2’ Using N2 ore data as the total cubic data set, a three-level decision tree is constructed for the remaining one parameter T3 except for the selected parameters T1 and T2 based on this total cubic data set. In the three-level decision tree, the leaf nodes are selected [D MAX -(D MAX -D MIN ) / 3, D MAX All the values of the three selected parameters T3 below are set. There are N3 such values, where N3 < N2. Thus, the average value of these N3 T3 parameter values is obtained as the value of the three selected parameters N3’. In the proposed N+1th iron ore sorting process, the decision module assigns corresponding values ​​N1', N2', and N3' to the three parameters of magnetic field intensity used in magnetic separation, pulp mass concentration used in flotation, and number of sweeping times used in sweeping.

2. The system according to claim 1, characterized in that Assume the basic iron content is A0, the basic copper content is B0, in each ore data, the mass proportion of iron concentrate to the original ore is k1, the iron content in the iron concentrate is A1 and the copper content is B1, the mass proportion of copper concentrate to the original ore is k2, the iron content in the copper concentrate is A2 and the copper content is B2, the mass proportion of tailings to the original ore is k3, the iron content in the tailings is A3 and the copper content is B3, and the relative content of iron and copper D in the ore data is calculated: Thus, the relative content of iron and copper in N ore data D1, D2, ... D N , and then select the maximum value D among the N values ​​of the relative content of iron and copper MAX =Max{D1, D2, ... D N } and the minimum value D MIN =Min{D1, D2, ... D N }.

3. The system according to claim 1, characterized in that The decision-making unit uses N times of ore data as the total data set and classifies them according to the above three sub-content intervals to obtain the empirical entropy, which is calculated as follows: Wherein, K represents the numerical value of the number of sub-content intervals, that is, in the above summation formula, K ranges from 1 to 3, N K is the number of samples corresponding to each sub-content interval.

4. The system according to claim 3, characterized in that In the classification of parameter T as the root node, there are N parameter values ​​T1, T2, ..., T of parameter T corresponding to the ore data of N iron ore sorting. N , so find the maximum value T among these N parameter values MAX =Max{T1, T2, ..., T N } and the minimum value T MIN =Min{T1, T2, ..., T N }, the parameter interval [T MIN ,T MAX ] is divided into three equal parts, that is, forming three sub-parameter intervals [T MIN ,T MIN +(T MAX -T MIN ) / 3]、[T MIN +(T MAX -T MIN ) / 3,T MAX -(T MAX -T MIN ) / 3]、[T MAX -(T MAX -T MIN ) / 3, T MAX ], the above root node classifies the parameter T, and classifies the total data set based on the above three sub-parameter intervals.

5. The system according to claim 4, characterized in that The calculation formula of the empirical conditional entropy is as follows: Among them, H(D|T) represents the empirical conditional entropy of the total data set when the sub-parameter interval is classified as the root node and the sub-content interval is classified as the leaf node, and H(D i ) represents the empirical entropy under each root node, T i represents the number of each category under a specific sub-parameter interval classification, D ik Indicates the number of samples classified in each leaf node.

6. The system according to claim 1, characterized in that Construct a secondary decision tree for two other parameters except the selected parameter T1 based on the secondary data set. The secondary decision tree takes sub-parameter intervals as root nodes and sub-content intervals as leaf nodes for the two other parameters under the secondary data set, thereby calculating the information gains ζ1’ and ζ2’ of the two other parameters. If ζ1’ > ζ2’, determine a corresponding secondary selected parameter T2 based on ζ1’, and backtrack in the decision tree to select the leaf node [D MAX -(D MAX -D MIN ) / 3, D MAX and obtain the values of all secondary selected parameters T2. Suppose there are N2 such values, where N2 < N1. Thus, the average value of these N2 T2 parameter values is obtained as the value of the secondary selected parameter N2’.

Images