Grinding grading intelligent decision-making method and system based on machine learning

Abstract

The invention provides an ore grinding grading intelligent decision-making method and system based on machine learning, and the method comprises the steps: S1, collecting the manual control data of a sensor data field, and constructing a training data set and a test data set; s2, analyzing and decoupling a plurality of groups of manual control data; and training a decision tree model according to the training data set, and predicting parameters of the ore grinding classification equipment through the decision tree model to obtain a prediction result. According to the method, the multivariable coupling nonlinear dynamic optimization problem in the ore grinding and grading process can be effectively solved, and the operation efficiency, the product quality and the production efficiency of ore grinding and grading are comprehensively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of automatic control in ore dressing, and specifically relates to an intelligent decision-making method and system for grinding classification based on machine learning. Background Art

[0002] Grinding classification is an important production link in a concentrator. It has huge production energy consumption, complex process parameter regulation, and the traditional PID-based control method cannot adapt to all working conditions, requiring a large amount of manual regulation. Due to the superposition of factors such as skills, physiology, and physical strength, manual operation cannot accurately regulate operation parameters for a long time.

[0003] Therefore, the automatic control optimization of the grinding classification operation can significantly improve the operation quality, enhance the equipment efficiency, reduce the production cost, relieve the manual operation intensity, and achieve high-quality output and energy conservation and emission reduction.

[0004] There are many random factors in the grinding classification control operation, such as changes in the properties of raw ore, fluctuations in sensor data, complex mechanisms, large time lags, and large inertia, which make the automatic control of this operation difficult. The four important parameters of the grinding classification process flow include the feed rate of the ball mill, the water feed rate of the ball mill, the additional water feed rate of the pump sump, and the feed pressure of the cyclone. During the production process, the above four control variables are dynamically time-varying and strongly coupled with each other, and it is difficult to coordinate and control them, and optimal parameters need to be given in real time.

[0005] Patent document CN112317110A discloses a grinding particle size prediction system and method based on deep learning. This solution uses the perception technology of deep learning to establish a double-time recursive neural particle size prediction model. This prediction method may encounter problems such as gradient disappearance and gradient explosion, especially when dealing with long-sequence data, which may cause the model to be difficult to capture long-term dependence relationships. Moreover, the system can only predict the grinding particle size and grinding control optimization suggestions according to the real-time production situation, and cannot give the optimal parameters for grinding classification control in real time, affecting the efficiency of the grinding classification operation.

[0006] Patent document CN116422452A discloses a method for automatically controlling and optimizing the particle size range of grinding overflow, and this solution belongs to the technical field of grinding particle size control. In the production of the present invention, production data is obtained through sensors and sent to the central control computer; in the early stage, a deep forest model is trained through long-term data accumulation and stored in the central control computer; the overflow particle size value obtained by the sensor is judged. For the situation not within the specified range, the optimal parameters of each device suitable for the current production state are obtained through the grid method under the deep forest model, and the parameter adjustment results are fed back to the production system through the sending module, and finally the parameter adjustment is completed through the control module. This solution realizes the automatic adjustment and correction of parameters in the grinding process, greatly improving the stability of the overflow particle size and production efficiency in grinding. The problem solved by this solution is to adjust the parameters of each device to meet the overflow particle size of grinding. The ultimate goal is to make the overflow particle size at the outlet of the first-stage grinding reach the target value. However, this solution does not solve the problem of multi-variable coupling non-linear dynamic optimization in the grinding classification process. This problem needs to be solved urgently. Summary of the Invention

[0007] Aiming at the defects in the prior art, the purpose of the present invention is to provide a grinding classification intelligent decision-making method and system based on machine learning.

[0008] A grinding classification intelligent decision-making method based on machine learning provided by the present invention includes:

[0009] Step S1: Collect sensor data and manual control data to construct a training data set and a test data set;

[0010] Step S2: Analyze and decouple multiple groups of the manual control data; train a decision tree model according to the training data set, and then predict the parameters of the grinding classification equipment through the decision tree model to obtain a prediction result.

[0011] Preferably, in the step S1, the sensor data includes: powder ore bin level, feeder status and frequency, belt conveyor ore quantity measurement value, ball mill feed quantity, ball mill water supply quantity, ball mill grinding sound and current, pump sump level and makeup water quantity, slurry pump pump frequency and current, hydrocyclone feed pressure and valve status, hydrocyclone feed concentration and flow rate, and hydrocyclone overflow particle size and concentration;

[0012] The manual control data includes: ball mill feed quantity, ball mill water supply quantity, pump sump makeup water, and hydrocyclone feed pressure.

[0013] Preferably, in the step S2, multiple groups of the manual control data are analyzed by correlation analysis; the number of items in each group of the manual control data is 4.

[0014] Preferably, in the step S2, perform correlation analysis on the ore feeding amount, water feeding amount of the ball mill, additional water feeding amount of the pump sump, and the ore feeding pressure of the hydrocyclone for the manual control data;

[0015] The prediction step of the ore feeding amount of the ball mill includes:

[0016] Step A1: Screen out the target data according to the operation flag bit;

[0017] Step A2: Truncate the outliers through the three-sigma principle;

[0018] Step A3: Filter out the data of the debugging actions of the target data to obtain an effective training data set;

[0019] Step A4: According to the decision tree and the effective training data set, obtain the prediction result of the ore feeding amount of the ball mill;

[0020] The prediction steps of the water feeding amount of the ball mill and the additional water feeding amount of the pump sump include:

[0021] Step B1: Let represent the predicted value of the water feeding amount of the ball mill, represent the predicted value of the additional water feeding amount of the pump sump, and construct a prediction formula;

[0022] Step B2: According to the prediction formula, obtain the values of the water feeding amount of the ball mill and the additional water feeding amount of the pump sump;

[0023] The prediction formula is:

[0024]

[0025] wherein, k1 and k2 are the first coefficient and the second coefficient respectively; both the first coefficient and the second coefficient are obtained by statistically analyzing the multiple relationships of the ore feeding amount, water feeding amount of the ball mill, and additional water feeding amount of the pump sump in each action interval in history; represent the predicted value of the ore feeding amount obtained based on LightGBM;

[0026] The mathematical expression of k1 is:

[0027] k1 = ∑ a∈A w a / y a

[0028] wherein, w a represents the water feeding amount of the ball mill when the action is a; y n represents the ore feeding amount when the action is a;

[0029] The mathematical expression of k2 is:

[0030] k2 = ∑ a∈A pa / y a

[0031] Among them, a is an action; A is a set of actions; p a represents the additional water volume added to the pump sump when the action is a;

[0032] The predicted value of the hydrocyclone feed pressure takes the mean value of the hydrocyclone feed pressure, and the mathematical expression is:

[0033]

[0034] Among them, T represents the number of actions; represents the mean value of the hydrocyclone feed pressure; t represents the time, and the value range of t is [t - 1 - T, t - 1]; q i represents the hydrocyclone feed pressure; i represents an index different from t, indicating that starting from the current moment, T moments are pushed forward;

[0035] A grinding and classification intelligent decision-making system based on machine learning provided by the present invention includes:

[0036] Module M1: Collect sensor data and manual control data, and construct a training data set and a test data set;

[0037] Module M2: Analyze and decouple multiple groups of the manual control data; train a decision tree model according to the training data set, and then predict the parameters of the grinding and classification equipment through the decision tree model to obtain a prediction result.

[0038] Preferably, in the module M1, the sensor data includes: powder ore bin level, feeder status and frequency, belt conveyor ore quantity measurement value, ball mill feed quantity, ball mill water supply quantity, ball mill grinding sound and current, pump sump liquid level and additional water quantity, slurry pump frequency and current, hydrocyclone feed pressure and valve status, hydrocyclone feed concentration and flow rate, and hydrocyclone overflow particle size and concentration;

[0039] The manual control data includes: ball mill feed quantity, ball mill water supply quantity, additional water added to the pump sump, and hydrocyclone feed pressure.

[0040] Preferably, in the module M2, perform correlation analysis on multiple groups of the manual control data; the number of items in each group of the manual control data is 4.

[0041] Preferably, in the module M2, perform correlation analysis on the ball mill feed quantity, ball mill water supply quantity, additional water quantity added to the pump sump, and hydrocyclone feed pressure of the manual control data;

[0042] The prediction module for the ball mill feed quantity includes:

[0043] Module A1: Filter out the target data according to the running flag bit;

[0044] Module A2: Truncate the outliers through the three-sigma principle;

[0045] Module A3: Filter out the data of the debugging actions of the target data to obtain an effective training data set;

[0046] Module A4: Obtain the prediction result of the ore feeding amount of the ball mill according to the decision tree and the effective training data set;

[0047] The ball mill water supply amount and pump pool makeup water amount prediction module includes:

[0048] Module B1: Let represent the predicted value of the ball mill water supply amount, represent the predicted value of the pump pool makeup water amount, and construct a prediction formula;

[0049] Module B2: Obtain the values of the ball mill water supply amount and the pump pool makeup water amount according to the prediction formula;

[0050] The prediction formula is:

[0051]

[0052] where k1 and k2 are the first coefficient and the second coefficient respectively; both the first coefficient and the second coefficient are obtained by statistically analyzing the multiple relationships among the ore feeding amount, the ball mill water supply amount, and the pump pool makeup water amount in each action interval in history; represent the predicted value of the ore feeding amount obtained based on LightGBM;

[0053] The mathematical expression of k1 is:

[0054] k1 = ∑ a∈A w a / y a

[0055] where w a represents the ball mill water supply amount when the action is a; y a represents the ore feeding amount when the action is a;

[0056] The mathematical expression of k2 is:

[0057] k2 = ∑ a∈A p a / y a

[0058] where a is an action; A is the set of actions; p a represents the pump pool makeup water amount when the action is a;

[0059] The predicted value of the hydrocyclone feed pressure is taken as the mean value of the hydrocyclone feed pressure, and the mathematical expression is:

[0060]

[0061] where T represents the number of times of the action; represents the mean value of the hydrocyclone feed pressure; t represents the time, and the value range of t is [t - 1 - T, t - 1]; q i represents the hydrocyclone feed pressure; i represents an index different from t, indicating that from the current moment, T moments are pushed forward;

[0062] According to a computer-readable storage medium storing a computer program provided by the present invention, when the computer program is executed by a processor, the steps of the intelligent decision-making method for grinding and classification based on machine learning are implemented.

[0063] According to an electronic device provided by the present invention, including a memory, a processor, and a computer program stored on the memory and executable on the processor, when the computer program is executed by the processor, the steps of the intelligent decision-making method for grinding and classification based on machine learning are implemented.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] 1. Based on the collected historical data, the present invention uses the correlation analysis method to perform correlation analysis and decoupling on four manually controllable parameters, reducing the correlation dimension.

[0066] 2. The present invention can effectively solve the multi-variable coupling non-linear dynamic optimization problem in the grinding and classification process, comprehensively improving the operation efficiency, product quality and production efficiency of grinding and classification.

[0067] 3. This solution uses the decision tree algorithm based on machine learning and the correlation analysis method to realize the prediction of the feed amount of the ball mill, the water supply amount of the ball mill, the supplementary water addition to the pump sump, and the hydrocyclone feed pressure. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] By reading the following detailed description of the non-limiting embodiments with reference to the accompanying drawings, other features, objects, and advantages of the present invention will become more apparent:

[0069] Figure 1 It is a schematic diagram of the correlation analysis of the important features provided by the present invention;

[0070] Figure 2 It is a comparison diagram of the prediction effects of the important control parameters of grinding and classification provided by the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0071] The present invention will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that those of ordinary skill in the art can make several changes and improvements without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0072] Based on machine learning, this paper proposes a multi-variable decoupling intelligent decision-making method for grinding and classification; specifically, this method uses the fusion application of correlation analysis and decision analysis, and solves the problem of dynamic prediction of control parameters in the grinding and classification process by the idea of decoupling control parameters into single or small-dimensional control parameter combinations through correlation analysis and then predicting each control parameter group one by one. Its main innovation points include:

[0073] 1. This solution proposes a method for decoupling control parameters for grinding and classification;

[0074] 2. This solution uses the decision tree algorithm based on machine learning and the correlation analysis method to realize the prediction of the feed rate of the ball mill, the water supply of the ball mill, the supplementary water addition to the pump sump, and the feed pressure of the cyclone;

[0075] What the present invention solves is the grinding efficiency, and the purpose is to improve the grinding efficiency by adjusting various equipment parameters, which is different from its optimization goal.

[0076] A grinding and classification intelligent decision-making method based on machine learning according to the present invention includes:

[0077] Step 1: Collect various key sensor data and corresponding manual control parameters in the grinding and classification production system, and store them in the real-time database as training and test data sets.

[0078] Step 2: Based on the collected training data set, use the correlation analysis method to perform correlation analysis and decoupling on four manually controllable parameters to reduce the correlation dimension.

[0079] Step 3: Train and test the gradient boosting decision tree model with historical data to achieve R 2 goodness of fit; predict the feed rate parameter of the ball mill based on the gradient boosting decision tree, predict the water supply of the ball mill and the supplementary water addition to the pump sump according to the correlation relationship, and predict the feed pressure of the cyclone according to the small interval averaging method. Specifically, the gradient boosting decision tree refers to LightGBM, and the detailed steps are in 3.1 below.

[0080] In the step 1, it includes:

[0081] Step 1.1: In the DCS / PLC system of the grinding and classification process flow of ore dressing, collect sensor data and manual control parameters through the OPC communication interface and transfer them to the real-time database.

[0082] Among them, the sensor data is 152 items, including: the powder ore bin level, the feeder status and frequency, the measured value of the ore quantity on the belt conveyor, the ore feeding quantity of the ball mill, the water feeding quantity of the ball mill, the grinding sound and current of the ball mill, the liquid level of the pump sump and the additional water feeding quantity, the pump frequency and current of the slurry pump, the feeding pressure and valve status of the hydrocyclone, the feeding concentration and flow rate of the hydrocyclone, and the overflow particle size and concentration of the hydrocyclone.

[0083] The manual control parameters are 4 items, including: the ore feeding quantity of the ball mill, the water feeding quantity of the ball mill, the additional water feeding to the pump sump, and the feeding pressure of the hydrocyclone.

[0084] In the said step 2, it includes:

[0085] Step 2.1: Use the correlation analysis method to analyze the correlation of the four control parameters, Figure 1 A correlation coefficient matrix is given. It is found from it that the correlation between the ore feeding quantity of the ball mill and the water feeding quantity of the ball mill and the additional water feeding quantity of the pump sump is relatively high.

[0086] Therefore, first predict the ore feeding quantity of the ball mill. After the ore feeding quantity of the ball mill is known, then predict the former two according to the correlation coefficients among the water feeding quantity of the ball mill, the additional water feeding quantity of the pump sump, and the ore feeding quantity of the ball mill.

[0087] However, since the correlation between the ore feeding quantity of the ball mill and the feeding pressure of the hydrocyclone is relatively low, but through analysis, it is found that the change range of the feeding pressure of the hydrocyclone is very small, the range value is 70 kPa to 100 kPa, and the variance is 6.0.

[0088] Therefore, set it as the average value in the recent time zone.

[0089] Ball mill feed rate Ball mill water feed rate Make-up water volume in the pump sump Hydrocyclone feed pressure Ball mill feed rate 1.000000 O.555341 0.730171 -0.292373 Ball mill water feed rate 0.555341 1.000000 0.635431 -0.358401 Make-up water volume in the pump sump 0.730171 0.635431 1.000000 -0.355880 Hydrocyclone feed pressure -0.292373 -0.358401 -O.355880 1.000000

[0090] Table 1: Correlation analysis of important parameters

[0091] In the said step 3, it includes:

[0092] Step 3.1: Prediction of the ore feeding quantity of the ball mill:

[0093] Preprocess the historical data dumped into the real-time database. First, according to the equipment operation flag bit, screen out the valid data in the equipment operation state, that is, the target data; then, use the three-sigma principle to truncate the outliers; finally, filter out a large number of debugging actions according to the action duration. One action refers to a quadruple formed by binding the four control parameters. As long as any one changes, a new action is generated, and finally an effective training data set is formed.

[0094] Specifically, the debuggable actions refer to trial actions. During the operation, by trying to set a value first and finding it unreasonable, it will be adjusted immediately. Therefore, the duration of such actions will be relatively short.

[0095] Therefore, all the actions in the training set are sorted according to the duration, and then the 5% of the actions with the shortest duration are removed. In this embodiment, the actions with a duration less than 5 minutes are removed.

[0096] Relevant feature data related to the feed rate of the ball mill is found through decision tree importance analysis, such as Figure 2 shown.

[0097] The prediction of the feed rate of the ball mill is a regression problem, and the efficient machine learning algorithm LightGBM is widely used in the industrial field for regression prediction problems. Therefore, LightGBM is used to fit the non-linear relationship between the features and the feed rate of the ball mill.

[0098] D = {(x1, y1),..., (x i , y i ),..., (x N , y N )) represents the data set for the prediction of the grinding amount, where x i represents the feature vector, and y i represents the manually given grinding amount. The number of training samples in this training is n = 1,334,228, and the number of test samples is N; based on the number of training samples, N - n = 333,557. The loss function uses the following mean square error, that is, MSE, and the mathematical expression is:

[0099]

[0100] where, n represents the number of training samples; y i represents the manually given grinding amount; represents the predicted value of the feed rate;

[0101] The loss function defines how the model measures the difference between the predicted value and the actual value, and guides the model to minimize this difference through continuous optimization. Specifically, it is usually gradient descent. After training the above LightGBM model, the predicted value R 2 of the feed rate of the test set is 0.6595, achieving a good fitting effect, indicating a high prediction accuracy.

[0102] Step 3.2: Prediction of the water supply amount of the ball mill and the supplementary water amount of the pump sump:

[0103] On the basis of knowing the feed rate of the ball mill, the water supply amount of the ball mill and the supplementary water amount of the pump sump are solved through the correlation coefficient in Step 2.1. Let represent the predicted value of the water supply amount of the ball mill, Represents the predicted value of the water volume added to the pump pool, and the mathematical expression is:

[0104]

[0105] Among them, k1 and k2 are obtained by statistically analyzing the multiples of the ball mill feed volume, ball mill water feed volume, and pump pool water replenishment volume in each action interval in history. k1=∑ a∈A w a / y a , k2=∑ a∈A p a / y a , a is an action; A is a set of actions; * represents a multiplication sign;

[0106] In other words, k1 and k2 are the first coefficient and the second coefficient respectively; the first coefficient and the second coefficient are obtained by statistically analyzing the multiples of the ball mill feed, the ball mill water feed and the pump pool water supply in each action interval in history; Indicates the predicted value of feed volume based on LightGBM;

[0107] The mathematical expression of k1 is:

[0108] k1=∑ a∈A w a / y a

[0109] Among them, w a Indicates the water supply of the ball mill when the action is a; y a Indicates the ore feeding amount when the action is a;

[0110] The mathematical expression of k2 is:

[0111] k2=∑ a∈A p a / y a

[0112] Among them, a is an action; A is a set of actions; p a It indicates the amount of water added to the pump pool when the action is a;

[0113] Step 3.3: Prediction of cyclone feed pressure:

[0114] The cyclone feed pressure is calculated by the average cyclone feed pressure of the last T actions:

[0115]

[0116] Among them, T represents the number of actions; represents the average value of the hydrocyclone feed pressure; t represents the time, and the value range of t is [t - 1 - T, t - 1]; q i represents the hydrocyclone feed pressure; i represents an index different from t, indicating T moments pushed forward from the current moment;

[0117] After predicting the important parameters of the grinding and classification process flow, namely the feed rate of the ball mill, the water feed rate of the ball mill, the supplementary water addition to the pump sump, and the hydrocyclone pressure respectively by the above method, the trend comparison chart of the actual value and the predicted value is as Figure 2 shown, and it can be seen that the fitting effect is good.

[0118] The present invention also provides a grinding and classification intelligent decision-making system based on machine learning. The grinding and classification intelligent decision-making system based on machine learning can be implemented by executing the process steps of the grinding and classification intelligent decision-making method based on machine learning. That is, those skilled in the art can understand the grinding and classification intelligent decision-making method based on machine learning as the preferred implementation manner of the grinding and classification intelligent decision-making system based on machine learning.

[0119] According to a grinding and classification intelligent decision-making system based on machine learning provided by the present invention, it includes:

[0120] Module M1: Collect sensor data and manual control data, and construct a training data set and a test data set;

[0121] Module M2: Analyze and decouple multiple groups of the manual control data; train a decision tree model according to the training data set, and then predict the parameters of the grinding and classification equipment through the decision tree model to obtain a prediction result.

[0122] Those skilled in the art know that in addition to implementing the system and its various devices, modules, and units provided by the present invention in the form of pure computer-readable program code, the method steps can be logically programmed to enable the system and its various devices, modules, and units provided by the present invention to be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers, etc. to achieve the same function. Therefore, the system and its various devices, modules, and units provided by the present invention can be regarded as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be regarded as the structure within the hardware component; the devices, modules, and units for implementing various functions can also be regarded as both software modules for implementing the method and the structure within the hardware component.

[0123] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. Without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. An intelligent decision-making method for grinding classification based on machine learning, characterized in that, Including: Step S1: Collect sensor data and manual control data to construct a training data set and a test data set; Step S2: Analyze and decouple multiple groups of the manual control data; Train a decision tree model based on the training data set, and then predict the parameters of the grinding and classification equipment through the decision tree model to obtain a prediction result.

2. The intelligent decision-making method for grinding classification based on machine learning according to claim 1, wherein In the step S1, the sensor data includes: powder ore bin level, feeder status and frequency, belt conveyor ore quantity measurement value, ball mill feed quantity, ball mill water supply quantity, ball mill grinding sound and current, pump sump level and makeup water quantity, slurry pump frequency and current, hydrocyclone feed pressure and valve status, hydrocyclone feed concentration and flow rate, and hydrocyclone overflow particle size and concentration; The manual control data includes: ball mill feed quantity, ball mill water supply quantity, pump sump makeup water, and hydrocyclone feed pressure.

3. The intelligent decision-making method for grinding classification based on machine learning according to claim 1, characterized in that, In the step S2, perform correlation analysis on multiple groups of the manual control data; the number of items in each group of the manual control data is 4.

4. The intelligent decision-making method for grinding and classification based on machine learning according to claim 1, characterized in that, In the step S2, perform correlation analysis on the ball mill feed quantity, ball mill water supply quantity, pump sump makeup water quantity, and hydrocyclone feed pressure of the manual control data; The prediction steps for the ball mill feed quantity include: Step A1: Screen out target data according to the operation flag bit; Step A2: Truncate outliers through the three-sigma principle; Step A3: Filter out the data of the debugging actions of the target data to obtain an effective training data set; Step A4: Obtain the prediction result of the ball mill feed quantity according to the decision tree and the effective training data set; The prediction steps for the ball mill water supply quantity and the pump sump makeup water quantity include: Step B1: Set to represent the predicted value of the water supply volume of the ball mill, to represent the predicted value of the additional water supply volume of the pump sump, and construct a prediction formula; Step B2: Obtain the values of the ball mill water supply quantity and the pump sump makeup water quantity according to the prediction formula; The prediction formula is: wherein, k1 and k2 are the first coefficient and the second coefficient respectively; both the first coefficient and the second coefficient are obtained by statistically analyzing the multiple relationships among the ore feeding amount, the water feeding amount of the ball mill, and the additional water feeding amount of the pump sump in each historical action interval; denotes the predicted value of the ore feeding amount obtained based on LightGBM; The mathematical expression of k1 is: k 1 = a∈A w a / y a Among them, w a represents the water feeding amount of the ball mill when the action is a; y a represents the ore feeding amount when the action is a; The mathematical expression of k2 is: k 2 = a∈A p a / y a where a is an action; A is a set of actions; p a represents the amount of water replenished to the pump sump when the action is a; The predicted value of the hydrocyclone feed pressure takes the mean value of the hydrocyclone feed pressure, and the mathematical expression is: Among them, T represents the number of times of the action; represents the average value of the cyclone feed pressure; t represents the moment, and the value range of t is [t-1-, t-1]; q i represents the cyclone feed pressure; i represents an index different from t, indicating T moments pushed forward from the current moment.

5. An intelligent decision-making system for grinding and classification based on machine learning, characterized in that, Including: Module M1: Collect sensor data and manual control data to construct a training data set and a test data set; Module M2: Analyze and decouple multiple groups of the manual control data; Train a decision tree model based on the training data set, and then predict the parameters of the grinding and classification equipment through the decision tree model to obtain a prediction result.

6. The intelligent decision-making system for grinding and classification based on machine learning according to claim 5, characterized in that In the module M1, the sensor data includes: powder ore bin level, feeder status and frequency, belt conveyor ore quantity measurement value, ball mill feed quantity, ball mill water supply quantity, ball mill grinding sound and current, pump sump level and makeup water quantity, slurry pump frequency and current, hydrocyclone feed pressure and valve status, hydrocyclone feed concentration and flow rate, and hydrocyclone overflow particle size and concentration; The manual control data includes: ball mill feed quantity, ball mill water supply quantity, pump sump makeup water, and hydrocyclone feed pressure.

7. The intelligent decision-making system for grinding and classification based on machine learning according to claim 5, characterized in that, In the module M2, perform correlation analysis on multiple groups of the manual control data; the number of items in each group of the manual control data is 4.

8. The intelligent decision-making system for grinding and classification based on machine learning according to claim 5, characterized in that, In the module M2, perform correlation analysis on the ball mill feed quantity, ball mill water supply quantity, pump sump makeup water quantity, and hydrocyclone feed pressure of the manual control data; The prediction module for the ore feeding amount of the ball mill includes: Module A1: Filter out target data according to the operation flag bit; Module A2: Truncate outliers through the three-sigma principle; Module A3: Filter out the data of the debugging actions of the target data to obtain an effective training data set; Module A4: Obtain the prediction result of the ore feeding amount of the ball mill according to the decision tree and the effective training data set; The prediction module for the water feeding amount of the ball mill and the additional water feeding amount of the pump sump includes: Module B1: Set Represents the predicted value of the water supply amount of the ball mill, Represents the predicted value of the additional water supply amount of the pump sump, and constructs a prediction formula; Module B2: Obtain the values of the water feeding amount of the ball mill and the additional water feeding amount of the pump sump according to the prediction formula; The prediction formula is: wherein, k1 and k2 are the first coefficient and the second coefficient respectively; both the first coefficient and the second coefficient are obtained by statistically analyzing the multiple relationships among the ore feeding amount, the water feeding amount of the ball mill, and the additional water feeding amount of the pump sump in each historical action interval; indicating the predicted value of the ore feeding amount obtained based on LightGBM; The mathematical expression of k1 is: k1 = ∑ a∈A w a / y a Among them, w a represents the water supply of the ball mill when the action is a; y a represents the ore feeding amount when the action is a; The mathematical expression of k2 is: k2 = ∑ a∈A p a / y a where a is an action; A is a set of actions; p a represents the additional water volume in the pump sump when the action is a; The predicted value of the ore feeding pressure of the hydrocyclone takes the mean value of the ore feeding pressure of the hydrocyclone, and the mathematical expression is: Among them, T represents the number of times of the action; represents the average value of the hydrocyclone feed pressure; t represents the moment, and the value range of t is [t-1-, t-1]; q i represents the hydrocyclone feed pressure; i represents an index different from t, indicating T moments pushed forward from the current moment.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the machine learning-based intelligent decision-making method for grinding and classification described in any one of claims 1 to 4.

10. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the machine learning-based intelligent decision-making method for grinding and classification described in any one of claims 1 to 4.

Images