Prediction method, prediction device, and prediction program

Abstract

To provide a prediction method with improved prediction accuracy of the recovery probability of mineral particles by ore selection.SOLUTION: The prediction method is executed by a computer and comprises: a step of creating a prediction model by performing machine learning using feature data of mineral particles as training data; and a step of inputting the feature data into the prediction model, and predicting the recovery probability of the mineral particles by ore selection on the basis of a value output by the prediction model, wherein the feature data of the mineral particles includes individual particle feature data indicating the characteristics of the mineral particles and particle group feature data indicating the characteristics of a particle group composed of the plurality of mineral particles.SELECTED DRAWING: Figure 7

Description

Technical Field

[0001] The present invention relates to a prediction method, a prediction device, and a prediction program for predicting the result of mineral processing.

Background Art

[0002] Conventionally, a mineral processing technique is known that separates mined ore into useful minerals and gangue minerals by methods such as flotation, gravity separation, or magnetic separation. Since there are few test ore samples that can be collected at the exploration stage, it is required to predict the recovery rate, grade, etc. of useful minerals with a limited number of tests. Non-Patent Document 1 discloses a method of predicting the recovery rate, grade, etc. of useful minerals using a prediction model created by machine learning with characteristic data for each particle in minerals identified by a mineral particle analyzer (MLA: Mineral Liberation Analyzer) as training data. Further, Non-Patent Document 2 also discloses a method of predicting the recovery rate, grade, etc. of useful minerals using a machine-learned prediction model.

Prior Art Documents

Non-Patent Documents

[0003]

Non-Patent Document 1

Non-Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0004] The behavior of mineral particles is thought to be influenced by other mineral particles in the surrounding area. Therefore, when predicting the recovery probability of particles using a machine learning model that uses only the characteristic data of each particle as training data, the recovery rate and grade predicted often differ too greatly from the actual recovery rate and grade.

[0005] Therefore, the present invention has been made in view of these points, and aims to provide a prediction method, prediction device, and prediction program that improve the prediction accuracy of the probability of recovering mineral particles by ore dressing. [Means for solving the problem]

[0006] A prediction method according to a first aspect of the present invention is a computer-based prediction method comprising the steps of: creating a prediction model by machine learning using characteristic data of mineral particles as training data; inputting the characteristic data into the prediction model and predicting the probability of recovering mineral particles by ore dressing based on the values ​​output by the prediction model, wherein the characteristic data of mineral particles includes individual particle characteristic data that shows the characteristics of the mineral particles and particle group characteristic data that shows the characteristics of a group of particles composed of a plurality of the mineral particles.

[0007] The prediction method may include, in the step of creating the prediction model, the individual particle feature data and the particle group feature data used as training data, which may include individual particle feature data and particle group feature data of mineral particles from recovered concentrate and tailings.

[0008] The prediction method may include, in the step of predicting the recovery probability of the mineral particles, the individual particle characteristic data and the particle group characteristic data used, which include individual particle characteristic data and particle group characteristic data of mineral particles whose recovery probability is unknown.

[0009] The prediction method further comprises the steps of: acquiring first condition data indicating the conditions for carrying out ore bean dressing; acquiring second condition data indicating the conditions for which the training data was obtained; acquiring prior probability data indicating prior probabilities which are the ratio of concentrate particles to tailings particles or the ratio of tailings particles to the number of feed particles corresponding to each of a plurality of conditions for carrying out ore bean dressing; and identifying the optimal prior probability in the prior probability data that corresponds to the condition closest to the condition indicated by the first condition data among the plurality of conditions indicated by the second condition data. In the step of predicting the recovery probability, the recovery probability may be predicted by correcting the value output by the prediction model with the optimal prior probability.

[0010] The prediction method may further include the step of creating the prediction model by machine learning selected from random forest, neural network, and logistic regression analysis.

[0011] In the step of predicting the recovery probability, some features selected by dimensionality reduction from a plurality of features included in the acquired individual particle feature data, and some features selected by dimensionality reduction from a plurality of features included in the acquired particle group feature data, may be input to the prediction model.

[0012] A prediction program according to a second aspect of the present invention is a prediction program that causes a computer to perform the following steps: creating a prediction model by machine learning using characteristic data of mineral particles as training data; inputting the characteristic data into the prediction model and predicting the probability of recovering mineral particles by ore dressing based on the values ​​output by the prediction model, wherein the characteristic data of mineral particles includes individual particle characteristic data that shows the characteristics of the mineral particles and particle group characteristic data that shows the characteristics of a group of particles composed of a plurality of the mineral particles.

[0013] A prediction device according to a third aspect of the present invention is a prediction device comprising: a feature data acquisition unit that acquires feature data of mineral particles; and a prediction unit that inputs the feature data to a prediction model created by machine learning using the feature data as training data, and predicts the probability of recovering mineral particles by ore dressing based on the values ​​output by the prediction model, wherein the feature data of the mineral particles includes individual particle feature data that shows the characteristics of the mineral particles and particle group feature data that shows the characteristics of a group of particles composed of a plurality of the mineral particles. [Effects of the Invention]

[0014] According to the present invention, the accuracy of predicting the probability of recovering mineral particles through ore dressing is improved. [Brief explanation of the drawing]

[0015] [Figure 1] This figure illustrates the general outline of the prediction system according to this embodiment. [Figure 2] This is a diagram illustrating the overview of the feature data. [Figure 3] This figure shows an example of training data. [Figure 4] This figure shows an overview of the prediction data used when predicting the probability of recovery using the prediction model M. [Figure 5] This diagram shows the configuration of the prediction device 1. [Figure 6] This is a flowchart showing the processing flow of the learning phase in prediction device 1. [Figure 7] This is a flowchart showing the processing flow of the prediction phase in prediction device 1. [Figure 8] This table shows a list of predictive models used in the comparative experiment. [Figure 9] This figure shows the copper recovery rate and grade obtained by flotation of particles ground under the condition of a grinding time of 15 minutes. [Figure 10] This figure shows the predicted grades of multiple types of minerals obtained by flotation of particles ground under the condition of a grinding time of 15 minutes. [Figure 11] This is a diagram showing the copper recovery rate and grade obtained by subjecting particles ground for 35 minutes to flotation. [Figure 12] This is a diagram showing the results of predicting the grades of multiple types of minerals obtained by subjecting particles ground for 35 minutes to flotation. [Mode for Carrying Out the Invention]

[0016] [Overview of Prediction System] Figure 1 is a diagram for explaining the overview of the prediction system according to the present embodiment. The prediction system is a system for predicting at least any one of the recovery probability of mineral particles from mineral processing, and the recovery rate and grade of mineral particles obtained by mineral processing.

[0017] Mineral processing refers to separating valuable minerals (concentrates) and gangue minerals (tailings) in ore, and includes methods such as flotation, gravity separation, or magnetic separation. A valuable mineral is a mineral corresponding to the element targeted for recovery. Although the present invention can be applied to any arbitrary mineral processing method, the present specification will mainly describe the case where the mineral processing method is flotation as an example.

[0018] Flotation is one of the most frequently used ore separation methods in the ore processing industry, and is a method of separating valuable mineral particles and gangue mineral particles based on the difference in hydrophobicity of ore surfaces. For example, as shown in Figure 1, the feed ore before separation is fed into the grinder 100 for grinding, and the ground feed ore particles are fed into the flotation machine 200. Specifically, finely ground ore particles are fed together with a foaming agent into the flotation machine 200 containing a solvent such as water, and hydrophobic valuable mineral particles are made to adhere to bubbles generated by stirring, and the valuable mineral particles are collected by being floated. The floated minerals (mineral particles) obtained as a result of flotation are referred to as concentrates, and the precipitated minerals (mineral particles) are referred to as tailings.

[0019] Incidentally, the recovery rate of concentrate from mined ore and the grade, which is expressed as the mass percentage of the target element contained in the recovered concentrate, differ depending on the mine. In order to increase the economic profit of mining, it is necessary to improve the recovery rate and grade. Flotation has operational variables such as grinding time, flotation time, and pH, and by determining the operational variables that lead to the optimal grade and recovery rate, the recovery rate and grade can be improved.

[0020] Traditionally, these instrumental variables have been determined by repeatedly conducting flotation tests and modifying them based on empirical rules. When instrumental variables are determined based on empirical rules, the process of setting instrumental variable values, confirming the resulting test results (i.e., the grade and recovery rate of the target element), and resetting the instrumental variable values ​​is repeated, resulting in significant costs in terms of both time and money.

[0021] Therefore, in recent years, research has been conducted to reduce the number of trials required to determine instrumental variables by using machine learning-based predictive models to determine them. Machine learning is a computational method that uses past information such as experience and data to improve accuracy in a given prediction task, and in recent years, predictive models have been developed using machine learning to predict the results of flotation trials.

[0022] Conventional prediction models, when given feature data representing the characteristics of a particle, output the recovery probability of that particle. When the ore beneficiation method is flotation, the recovery probability is the probability that the mineral particle will float (i.e., the probability of becoming concentrate). Conventional prediction models were based on the assumption that interactions between particles do not affect the recovery probability.

[0023] However, it is considered that interparticle interactions in flotation have a significant impact on the beneficiation results. Therefore, in the prediction method according to this embodiment, the accuracy of predicting the recovery probability is improved by performing processing that takes into account the effect of interparticle interactions in flotation. Specifically, in the prediction method according to this embodiment, when creating the prediction model, in addition to the characteristic data of individual mineral particles, characteristic data indicating the characteristic quantities of the particle group containing those mineral particles is also used as training data. Furthermore, when predicting the recovery probability using this prediction model, in addition to the characteristic data of individual mineral particles, characteristic data indicating the characteristic quantities of the particle group containing those mineral particles is input into the prediction model. The prediction model is created by machine learning selected from, for example, random forest, neural network, and logistic regression analysis.

[0024] [Summary of Feature Data] Figure 2 is a diagram illustrating the overview of the feature data. Figure 2(a) is an image taken with an electron microscope of a cross-section of the mineral for which the feature data is to be created. Each of the various shaped regions shown in the image in Figure 2(a) corresponds to a mineral particle (e.g., particle P1, particle P2). The multiple particles contained within the region enclosed by the quadrilateral on the outer edge of Figure 2(a) constitute a particle group PG.

[0025] Figure 2(b) shows an overview of the training data used to create the predictive model. The training data shown in Figure 2(b) corresponds to the particle group PG shown in Figure 2(a), and consists of the characteristic data of each particle from particle P1 to particle Pm (where m is an integer) included in the particle group PG.

[0026] The training data shown in Figure 2(b) includes individual particle feature data representing the characteristics of each individual particle P, corresponding to multiple particles P contained in the particle group PG; particle group feature data representing the characteristics of the particle group PG; and labels indicating whether particle P is concentrate or tailings. Since concentrate and tailings contain many particle groups PG, multiple training data sets corresponding to multiple particle groups PG are used as training data. As a result, many combinations of individual particle feature data, particle group feature data, and labels are used as training data.

[0027] As feature data, for example, MLA data showing multiple features of particles identified by a mineral particle analyzer is used, but feature data may be created by means other than MLA data. When creating training data, first, random sampling is performed from multiple MLA data corresponding to multiple particles so that the ratio of concentrate to tailings particles is 1:1. Next, bootstrap is performed multiple times on the randomly sampled multiple particles to create multiple datasets, and the individual particle feature data for each of the multiple particles contained in each of the multiple datasets obtained by bootstrap is identified. Next, by calculating the feature quantities of the particle group to which each of the multiple particles belongs, particle group feature data corresponding to each of the multiple particles is created. Then, by adding labels indicating whether each of the multiple particles is concentrate or tailings to the set of individual feature data and particle group feature data corresponding to each of the multiple particles, training data is created.

[0028] Figure 3 shows an example of training data. In the training data shown in Figure 3, the combination of particle group number and particle number is associated with features related to individual particles, features related to particle groups, and labels. Features related to individual particles include features related to particle shape and features related to particle composition. Features related to particle shape for individual particles include, for example, particle diameter and particle cross-sectional area. Features related to particle composition include, for example, the content ratio of minerals such as pyrite and chalcopyrite. For individual particle features, features corresponding to, for example, 20 types of features are used, and for particle composition features, more features than those related to individual particles, for example, features corresponding to 90 types of features are used. Note that the number of features is arbitrary, and some features may be selected from multiple available features to reduce the number of each feature.

[0029] The feature quantities related to a group of particles are values ​​that indicate collective characteristics highly relevant to interparticle interactions, such as the average diameter of multiple particles in the group, or the density of multiple particles in the group. The label is a value that indicates whether it is concentrate or tailings.

[0030] Figure 4 shows an overview of the prediction data used when predicting the recovery probability using the prediction model M. The prediction data is data in which individual particle feature data and particle group feature data are associated, but it differs from the training data shown in Figure 2(b) in that it does not include labels. The prediction data is created, for example, based on the MLA data of the ore supply. The prediction data shown in Figure 4 consists of feature data corresponding to particles included in the training data and feature data corresponding to different particles, so in Figure 4, the particle names are R1 to Rn (where n is an integer). When such prediction data is input to the prediction model, the prediction model outputs the recovery probability.

[0031] Incidentally, generally there is more tailing than concentrate, and the ratio of concentrate to concentrate distributed from the feed varies depending on the implementation conditions, which are determined by the conditions of ore dressing (e.g., flotation conditions), the specifications of the equipment used for ore dressing (e.g., flotation machine 200), and the distribution of features of the feed. Therefore, when creating training data by performing bootstrap sampling of MLA data so that the particle number ratio of concentrate to tailing is 1:1, as described above, an error may occur between the recovery probability output by the predictive model and the actual recovery probability under the implementation conditions.

[0032] Therefore, in the prediction method according to this embodiment, prior probabilities indicating the ratio of concentrate to tailings corresponding to the conditions under which ore beneficiation is performed are identified, and the recovery probability output by the prediction model is corrected using the prior probabilities corresponding to the conditions under which ore beneficiation is performed. This makes it possible to predict the recovery probability with high accuracy for each set of conditions. Details of the correction process using prior probabilities will be described later.

[0033] [Configuration of Prediction Device 1] Figure 5 shows the configuration of the prediction device 1. The prediction device 1 includes an external IF unit 11, a storage unit 12, and a control unit 13. The control unit 13 is, for example, a CPU (Central Processing Unit), and by executing a program stored in the storage unit 12, it functions as a condition data acquisition unit 131, a prior probability data acquisition unit 132, a specification unit 133, a feature data acquisition unit 134, a learning unit 135, a prediction unit 136, and a calculation unit 137. The control unit 13 also functions as a prediction model M by executing a program stored in the storage unit 12. The prediction model M may be located outside the control unit 13 (for example, on another computer).

[0034] The external IF unit 11 has at least one of the following: a user interface such as a keyboard and mouse for the user of the prediction device 1 to input data, a display for displaying data, and a communication interface for sending and receiving data with an external device. The prediction device 1 acquires data from the user or an external device via the external IF unit 11, and outputs the results of calculating the concentration particle recovery probability, concentrate recovery rate, and grade.

[0035] The memory unit 12 has storage media such as ROM (Read Only Memory), RAM (Random Access Memory), and SSD (Solid State Drive). The memory unit 12 stores the program executed by the control unit 13. The memory unit 12 also stores various data used by the control unit 13 to calculate the recovery probability, recovery rate, and quality. The memory unit 12 may also store the weights of the neural network included in the prediction model M.

[0036] The condition data acquisition unit 131 acquires condition data indicating the conditions for performing ore beautification. The condition data includes, for example, operational variables for performing ore beautification, the model name of the flotation machine 200, or the distribution of features of the ore supply. The condition data acquisition unit 131 acquires condition data, for example, by displaying a screen for inputting condition data on a display. The condition data acquisition unit 131 may also acquire condition data entered by a user in an external device from an external device. The condition data acquisition unit 131 acquires, for example, first condition data indicating the conditions for performing ore beautification, and second condition data indicating the conditions for which training data has been obtained. The condition data acquisition unit 131 inputs the acquired condition data into the identification unit 133.

[0037] The condition data acquisition unit 131 may acquire data indicating the target element to be beneficiated. The data indicating the target element is, for example, text data indicating the element such as Cu or Fe. The condition data acquisition unit 131 inputs the data indicating the target element to the feature data acquisition unit 134. The data indicating the target element may be stored in the storage unit 12 in advance.

[0038] The prior probability data acquisition unit 132 acquires prior probability data that represents the prior probability of the ratio of concentrate or tailings corresponding to each of the multiple conditions under which ore dressing is performed. In other words, the prior probability data acquisition unit 132 acquires multiple prior probability data associated with each of the conditions.

[0039] Similar to the condition data acquisition unit 131, the prior probability data acquisition unit 132 may acquire prior probability data by displaying a screen for inputting prior probability data on a display, or it may acquire prior probability data entered by a user on an external device from an external device. The prior probability data acquisition unit 132 inputs the acquired prior probability data into the identification unit 133. The prior probability data acquisition unit 132 may also input the acquired prior probability data into the identification unit 133 by storing the acquired prior probability data in the storage unit 12.

[0040] The identification unit 133 identifies the optimal prior probability corresponding to the prior probability data that is closest to the condition indicated by the first condition data, based on the prior probability data input from the prior probability data acquisition unit 132. For example, the identification unit 133 selects the prior probability data that is closest to the condition under which ore beneficiation is performed, based on the distance between a vector consisting of multiple elements representing the conditions corresponding to each of the multiple prior probability data and a vector consisting of multiple elements representing the conditions under which ore beneficiation is performed. In other words, the identification unit 133 defines the prior probability indicated by the prior probability data corresponding to the vector consisting of multiple elements representing the conditions indicated by the first condition data as the optimal prior probability. The identification unit 133 notifies the prediction unit 136 of the identified optimal prior probability.

[0041] Furthermore, the multiple vectors consisting of multiple elements representing the conditions corresponding to each of the multiple prior probability data are also called the implemented ore beneficiation condition data and correspond to the second condition data. The vectors consisting of multiple elements representing the conditions for performing ore beneficiation are also called the unimplemented ore beneficiation condition data and correspond to the first condition data.

[0042] [Learning Phase] Next, we will explain the process in the learning phase for creating the predictive model M. The feature data acquisition unit 134 acquires individual feature data for learning that shows the characteristics of mineral particles, and group feature data for learning that shows the characteristics of a group of particles.

[0043] Specifically, the feature data acquisition unit 134 acquires first individual learning particle feature data that shows the characteristics of the learning particles contained in the concentrate. The feature data acquisition unit 134 acquires second individual learning particle feature data that shows the characteristics of the learning particles contained in the tailings. The feature data acquisition unit 134 acquires first learning particle group feature data that shows the characteristics of a group of learning particles contained in the concentrate, which consists of multiple learning particles including the learning particles. The feature data acquisition unit 134 acquires second learning particle group feature data that shows the characteristics of a group of learning particles contained in the tailings, which consists of multiple learning particles including the learning particles.

[0044] The feature data acquisition unit 134 creates first training data by labeling a set of first individual particle feature data and first group of particle feature data for learning as concentrate, and inputs the first training data into the learning unit 135. The feature data acquisition unit 134 also inputs second training data into the learning unit 135 by labeling a set of second individual particle feature data and second group of particle feature data for learning as tailings.

[0045] In order to create multiple predictive models M corresponding to multiple conditions under which ore beneficiation may be carried out, the feature data acquisition unit 134 may acquire training particle feature data and training particle group feature data in association with the conditions. That is, in the step of acquiring first training individual particle feature data, the feature data acquisition unit 134 may acquire multiple first training individual particle feature data corresponding to each of the multiple conditions under which ore beneficiation is carried out. In the step of acquiring second training individual particle feature data, the feature data acquisition unit 134 may acquire multiple second training individual particle feature data corresponding to each of the multiple conditions under which ore beneficiation is carried out.

[0046] In the step of acquiring first training particle group feature data, the feature data acquisition unit 134 may acquire multiple first training particle group feature data corresponding to each of the multiple conditions for performing ore beneficiation. In the step of acquiring second training particle group feature data, the feature data acquisition unit 134 may acquire multiple second training particle group feature data corresponding to each of the multiple conditions for performing ore beneficiation. The feature data acquisition unit 134 creates first training data and second training data for each condition and inputs the created first training data and second training data to the learning unit 135.

[0047] The learning unit 135 creates a predictive model M by machine learning using first training data, which consists of a set of first individual particle feature data and first group of particle feature data labeled to indicate that it is concentrate, and second training data, which consists of a set of second individual particle feature data and second group of particle feature data labeled to indicate that it is tailings. The learning unit 135 creates a predictive model M by inputting either the first or second training data into the predictive model M and updating the weights of the neural network included in the predictive model M.

[0048] The learning unit 135 may create a predictive model corresponding to each of multiple conditions by performing machine learning using first training data, which consists of a set of first individual particle feature data and first group of particle feature data labeled to indicate concentrate, and second training data, which consists of a set of second individual particle feature data and second group of particle feature data labeled to indicate tailings, for each condition under which ore beneficiation is performed. In this case, the learning unit 135 stores in the storage unit 12, for example, the weights of the neural networks included in multiple predictive models M, which differ for each condition, in association with the conditions under which ore beneficiation is performed.

[0049] [Prediction Phase] The prediction phase is executed after the learning phase. In the prediction phase, the prediction unit 136 predicts the probability of recovering mineral particles contained in the ore (supply) to be processed, and the calculation unit 137 outputs information indicating the recovery rate and grade of mineral particles by processing. The probability of recovering the mineral particles to be processed is assumed to be unknown.

[0050] In the prediction phase, the feature data acquisition unit 134 acquires individual particle feature data that indicates the characteristics of the mineral particles to be beneficiated. The feature data acquisition unit 134 also acquires particle group feature data that indicates the characteristics of a group of particles composed of multiple particles contained in the feed. The feature data acquisition unit 134 notifies the prediction unit 136 of the acquired individual particle feature data and particle group feature data.

[0051] The prediction unit 136 predicts the recovery probability for each of the multiple particles contained in the ore supply. It predicts the recovery probability of each mineral particle based on the values ​​output by the prediction model M, which is created by machine learning using a set of training individual particle feature data and training particle group feature data labeled to indicate whether it is concentrate or tailings, as training data, and inputting the individual particle feature data and particle group feature data into the prediction model M for each type of particle.

[0052] The prediction unit 136 may, for example, input some of the features selected by dimensionality reduction from multiple features included in the individual particle feature data acquired by the feature data acquisition unit 134, and some of the features selected by dimensionality reduction from multiple features included in the particle group feature data acquired by the feature data acquisition unit 134, into the prediction model M.

[0053] As mentioned above, if the individual particle feature data and group particle feature data used for training, where the ratio of concentrate to tailings is 1:1, are used when creating the prediction model M, the prediction model M will output a recovery probability that is higher than the recovery probability under the actual conditions of ore beneficiation.

[0054] Therefore, in the step of predicting the recovery probability, the prediction unit 136 predicts the recovery probability of each mineral particle by correcting the value output by the prediction model M with the optimal prior probability calculated by the identification unit 133. Specifically, the prediction unit 136 corrects the recovery probability of each target particle by multiplying the recovery probability of one particle output by the prediction model M by a value obtained by dividing the optimal prior probability by the ratio of the number of particles labeled as concentrate in the training data, and then normalizing the result. The prediction unit 136 inputs the corrected recovery probability of each mineral particle to the calculation unit 137.

[0055] The prediction unit 136 may also predict the recovery probability using a prediction model M corresponding to the conditions indicated by the first condition data. For example, the prediction unit 136 can select the prediction model M that has the greatest similarity between the conditions indicated by the second condition data, which corresponds to the individual particle feature data and group particle feature data used for learning by each of the multiple prediction models M, and the conditions indicated by the first condition data, and input the individual particle feature data and group particle feature data into the selected prediction model M, thereby increasing the recovery probability output by the prediction model M.

[0056] The calculation unit 137 calculates the recovery rate of the element for which recovery is to be calculated and the grade of the element in the concentrate for which grade is to be calculated, based on the recovery probability of each particle notified by the prediction unit 136. Specifically, the calculation unit 137 calculates the recovery rate of the element for which recovery is to be calculated by ore dressing by multiplying the recovery probability of each particle predicted by the prediction unit 136 by the mass of each particle obtained by MLA analysis and the grade of the element in each particle for which recovery is to be calculated, adding the results, and then dividing by the result obtained by multiplying the mass of each particle obtained by MLA analysis and the grade of the element in each particle for which recovery is to be calculated, and then adding the results.

[0057] More specifically, the calculation unit 137 calculates the recovery rate R using the following formula, assuming that the number of target element particles for which the prediction unit 136 has predicted the recovery probability is m, and the corrected recovery probability for the j-th particle Xj is y(Xj). The calculation unit 137 uses the mass of Xj and the wt% grade of the target element in Xj. In the following explanation, the target element is the element for which the recovery rate is to be calculated.

number

[0058] Furthermore, in order to calculate the grade indicating the proportion of the target element in the concentrate obtained by ore dressing, the calculation unit 137 first multiplies the recovery probability of each of the multiple target particles predicted by the prediction unit 136 by the mass of each of the multiple target particles and then adds them together. Then, the calculation unit 137 calculates the grade indicating the proportion of the target element in the concentrate obtained by ore dressing based on the ratio of the summation value to the value obtained by multiplying the recovery probability of each of the multiple target particles by the mass of each of the multiple target particles and the mass percentage of the target particles contained in the feed.

[0059] Specifically, the calculation unit 137 is X j The mass of MX j When the mass percentage of the target element is denoted as wt, the grade Q is calculated using the following formula.

number

[0060] The calculation unit 137 outputs the calculated recovery rate and quality values ​​via the external IF unit 11. The calculation unit 137 may display the recovery rate and quality values ​​on a display or transmit them to an external computer.

[0061] [Processing flow in prediction device 1] Figure 6 is a flowchart showing the processing flow of the learning phase in the prediction device 1. The feature data acquisition unit 134 acquires first individual particle feature data for learning corresponding to the concentrate (S11). The feature data acquisition unit 134 acquires first group of particle feature data for learning corresponding to the concentrate (S12). The feature data acquisition unit 134 acquires second individual particle feature data for learning corresponding to the tailings (S13). The feature data acquisition unit 134 acquires second group of particle feature data for learning corresponding to the tailings (S14). The order of processing from S11 to S14 is arbitrary.

[0062] The feature data acquisition unit 134 creates first training data by labeling the first individual particle feature data and the first group of particle feature data for learning with a label indicating that they are concentrates, and creates second training data by labeling the second individual particle feature data and the second group of particle feature data for learning with a label indicating that they are tailings (S15). The learning unit 135 creates a prediction model M using the created first and second training data (S16).

[0063] Figure 7 is a flowchart showing the processing flow of the prediction phase in the prediction device 1. The condition data acquisition unit 131 acquires first condition data indicating the conditions for beneficiating minerals containing mineral particles of the target element for which the recovery rate and grade are to be predicted (S21). The identification unit 133 identifies the prior probability corresponding to the first condition data acquired by the condition data acquisition unit 131 (S22) and notifies the prediction unit 136 of the identified prior probability.

[0064] The feature data acquisition unit 134 acquires individual particle feature data for each mineral particle of the target element for which recovery rate and grade are to be predicted (S23). The feature data acquisition unit 134 also acquires particle group feature data corresponding to the particle (S24). The feature data acquisition unit 134 inputs the individual particle feature data and particle group feature data into the prediction model M (S25). The feature data acquisition unit 134 sequentially inputs the individual particle feature data and particle group feature data corresponding to each of the multiple particles into the prediction model M.

[0065] Next, the prediction unit 136 corrects the recovery probability corresponding to each of the multiple mineral particles output by the prediction model M based on the prior probability notified by the identification unit 133 (S26). The calculation unit 137 calculates the recovery rate and grade of the mineral particles corresponding to the target element based on the corrected recovery probability for each mineral particle (S27). Note that processes S23 to S25 may be performed before processes S21 to S22.

[0066] <Experimental Results> [Collection of ore samples] Twelve 1kg samples were collected from copper ore mined from a mine. All other conditions were kept the same, except for the grinding time, and four copper ore samples were assigned to each of three grinding time conditions (15 minutes, 25 minutes, and 35 minutes). Twelve resin-embedded samples (3 concentrates and 9 tailings) were prepared from the concentrate and tailings, and MLA analysis was performed. The MLA analysis mode was set to XBSE, and data of at least 50,000 particles was obtained for each sample.

[0067] [Creating a predictive model] To verify the effectiveness of the prediction method according to this embodiment, we created several types of prediction models corresponding to conventional prediction methods and a prediction model corresponding to the prediction method according to this embodiment. 80,000 particles (40,000 from concentrate and 40,000 from tailings) were randomly sampled from the MLA data of concentrate and tailings. The collected MLA data had 108 features, and to minimize correlation between features and shorten the training time, 10 features were selected using the Minimum Redundancy Maximum Algorithm (MRMR) algorithm. D50, D80, diameter standard deviation, average density, density standard deviation, average copper mass percentage, average iron mass percentage, and average sulfur mass percentage were used as training particle group feature data to create the prediction model corresponding to the prediction method according to this embodiment.

[0068] Figure 8 is a table showing a list of prediction models used in the comparative experiment. Model M15 is a prediction model created using individual particle feature data obtained by grinding minerals over 15 minutes as training data. Model M35 is a prediction model created using individual particle feature data obtained by grinding minerals over 35 minutes as training data. Model M-Mix is ​​a prediction model created using individual particle feature data of multiple particles obtained by grinding minerals over 15, 25, and 35 minutes as training data.

[0069] Model M-New is a predictive model M according to this embodiment, created using training data that includes individual particle feature data of multiple particles obtained by crushing minerals over time periods of 15 minutes, 25 minutes, and 35 minutes, along with group particle feature data.

[0070] [Comparison results] Figure 9 shows the recovery rate and grade of copper obtained by flotation of mineral particles crushed under the condition of a crushing time of 15 minutes. In Figure 9, the circles represent the actual measured recovery rate and grade, which are true values. The squares in Figure 9 represent the recovery rate and grade calculated using the prediction model M of M-New according to this embodiment. These recovery rates and grades are calculated by correcting the values ​​output by the prediction model M using prior probabilities (e.g., optimal prior probabilities). Since the crushing time of the particles to be predicted is 15 minutes, the recovery rate and grade when the prediction model named M35 is used deviate significantly from the true values. When the prediction models named M15 and M-Mix are used, the recovery rates and grades are closer to the true values ​​than the values ​​corresponding to model M35, but the values ​​corresponding to model M-New are closer to the true values. Thus, the effectiveness of predicting the recovery probability using particle group characteristic data has been confirmed.

[0071] Figure 10 shows the results of predicting the grades of multiple types of minerals obtained by flotation of particles ground under the condition of a grinding time of 15 minutes. The leftmost value corresponding to each mineral is the true value, and the rightmost value is the grade value predicted using the prediction model M of model name M-New. Referring to the results for chalcopyrite and pyrite, the grade values ​​predicted using the prediction model M of model name M-New are almost equal to the true values, which further confirms the effectiveness of predicting the recovery probability using particle group characteristic data.

[0072] Figure 11 shows the copper recovery rate and grade obtained by flotation of particles ground under the condition of a grinding time of 35 minutes. In Figure 11, the circles represent the actual measured recovery rate and grade, which are true values. In Figure 11, the squares represent the recovery rate and grade calculated using the prediction model M of M-New according to this embodiment. Since the grinding time of the particles to be predicted is 35 minutes, the recovery rate and grade when the prediction model named M15 is used deviate significantly from the true values. When the prediction models named M35 and M-Mix are used, the recovery rate and grade are closer to the true values ​​than the values ​​corresponding to model name M15, but the values ​​corresponding to model name M-New are closer to the true values. This example also confirms the effectiveness of predicting the recovery probability using particle group characteristic data.

[0073] Figure 12 shows the results of predicting the grades of multiple types of minerals obtained by flotation of particles ground under the condition of a grinding time of 35 minutes. The value on the far left corresponding to each mineral is the true value, and the value on the far right is the grade value predicted using the prediction model M of model name M-New. Referring to the results for chalcopyrite and pyrite, the grade values ​​predicted using the prediction model M of model name M-New are almost equal to the true values, which further confirms the effectiveness of predicting the recovery probability using particle group characteristic data.

[0074] [Effects of the prediction method according to this embodiment] As described above, in the prediction method according to this embodiment, the computer predicts the probability of particle recovery based on the values ​​output by the prediction model, which is created by machine learning using a set of training individual particle feature data and training particle group feature data labeled to indicate whether it is concentrate or tailings, as training data, and inputs individual particle feature data and particle group feature data into the prediction method. By predicting the probability of particle recovery using particle group feature data in this way, the accuracy of predicting the recovery rate and grade of mineral particles obtained by ore dressing is improved.

[0075] Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments, and various modifications and changes are possible within the scope of its gist. For example, all or part of the apparatus can be configured by functionally or physically distributing and integrating in any unit. Furthermore, new embodiments resulting from any combination of multiple embodiments are also included in the embodiments of the present invention. The effects of the new embodiments resulting from the combinations are combined with the effects of the original embodiments.

[0076] For example, in the above explanation, the prediction device 1 created a prediction model M and predicted the recovery rate and grade of the target mineral. However, if the prediction device 1 does not have a learning unit 135, a pre-created prediction model M may be used. Also, although the example given shows a 1:1 ratio between the first training data corresponding to concentrate and the second training data corresponding to tailings in the training data, this ratio is also arbitrary.

[0077] Furthermore, while the above explanation illustrates the use of MLA data to create individual particle feature data and particle group feature data, other data may be used. In particular, a particle size distribution analyzer may be used to create particle group feature data. [Explanation of Symbols]

[0078] 1. Prediction device 11 External IF section 12 Storage section 13 Control Unit 100 Crusher 131 Condition Data Acquisition Unit 132 Prior Probability Data Acquisition Unit 133 Specific part 134 Feature Data Acquisition Unit 135 Learning Department 136 Prediction Section 137 Calculation Section 200 flotation machine

Claims

1. The steps include creating a predictive model by machine learning using training feature data of training mineral particles as training data, and The steps include inputting characteristic data of mineral particles to be processed into the prediction model and predicting the probability of recovering the mineral particles based on the values ​​output by the prediction model, A computer-based prediction method having, The characteristic data of the learning mineral particles includes individual learning particle characteristic data that shows the characteristics of the learning mineral particles, and group learning particle characteristic data that shows the collective characteristics of a group of particles composed of a plurality of the learning mineral particles. The characteristic data of the mineral particles to be processed includes characteristic data of individual particles to be processed, which shows the characteristics of the mineral particles to be processed, and characteristic data of a group of particles to be processed, which shows the collective characteristics of a group of particles composed of multiple mineral particles to be processed. The learning particle group characteristic data and the ore-dressing target particle group characteristic data include at least one of the average diameter of multiple particles included in the particle group, D50, D80, diameter standard deviation, density of multiple particles included in the particle group, average density, density standard deviation, average copper mass percentage, average iron mass percentage, or average sulfur mass percentage. Prediction method.

2. In the step of creating the prediction model, the individual particle feature data and the group of particle feature data used as training data are This includes individual particle feature data and particle group feature data of mineral particles from recovered concentrate and tailings. The prediction method according to claim 1.

3. The individual particle characteristic data and the group of particle characteristic data used in the step of predicting the recovery probability of the mineral particles are Includes individual characteristic data and particle group characteristic data of mineral particles with unknown recovery probability. The prediction method according to claim 1 or 2.

4. A step of obtaining first condition data that indicates the conditions for carrying out ore dressing, The steps include obtaining second condition data that indicates multiple conditions for which the aforementioned training data was obtained, A step of obtaining prior probability data that represents the prior probability of the ratio of concentrate particles to tailings particles to the number of feed particles corresponding to each of several conditions for performing ore dressing, The steps include: identifying the optimal prior probability corresponding to the condition closest to the condition shown by the first condition data, based on the similarity between the multiple elements representing the condition shown by the first condition data and the multiple elements representing the condition shown by the second condition data, in the prior probability data; It further possesses, In the step of predicting the probability of recovery, the probability of recovery is predicted by correcting the value output by the prediction model with the optimal prior probability. The prediction method according to claim 1.

5. The method further comprises the step of creating the predictive model using machine learning selected from random forest, neural network, and logistic regression analysis. The prediction method according to claim 1.

6. The steps include creating a predictive model by machine learning using training feature data of training mineral particles as training data, and The steps include inputting characteristic data of mineral particles to be processed into the prediction model and predicting the probability of recovering the mineral particles based on the values ​​output by the prediction model, A prediction program to be executed by a computer, The characteristic data of the learning mineral particles includes individual learning particle characteristic data that shows the characteristics of the learning mineral particles, and group learning particle characteristic data that shows the collective characteristics of a group of particles composed of a plurality of the learning mineral particles. The characteristic data of the mineral particles to be processed includes characteristic data of individual particles to be processed, which shows the characteristics of the mineral particles to be processed, and characteristic data of a group of particles to be processed, which shows the collective characteristics of a group of particles composed of multiple mineral particles to be processed. The learning particle group characteristic data and the ore-dressing target particle group characteristic data include at least one of the average diameter of multiple particles included in the particle group, D50, D80, diameter standard deviation, density of multiple particles included in the particle group, average density, density standard deviation, average copper mass percentage, average iron mass percentage, or average sulfur mass percentage. Prediction program.

7. A feature data acquisition unit that acquires feature data of mineral particles, A prediction unit that inputs characteristic data of mineral particles to be processed into a prediction model created by machine learning using training characteristic data of training mineral particles as training data, and predicts the probability of recovering the mineral particles to be processed based on the values ​​output by the prediction model, A prediction device having, The characteristic data of the learning mineral particles includes individual learning particle characteristic data that shows the characteristics of the learning mineral particles, and group learning particle characteristic data that shows the collective characteristics of a group of particles composed of a plurality of the learning mineral particles. The characteristic data of the mineral particles to be processed includes characteristic data of individual particles to be processed, which shows the characteristics of the mineral particles to be processed, and characteristic data of a group of particles to be processed, which shows the collective characteristics of a group of particles composed of multiple mineral particles to be processed. The learning particle group characteristic data and the ore-dressing target particle group characteristic data include at least one of the average diameter of multiple particles included in the particle group, D50, D80, diameter standard deviation, density of multiple particles included in the particle group, average density, density standard deviation, average copper mass percentage, average iron mass percentage, or average sulfur mass percentage. Prediction device.

Images