A method and system for real-time optimization control of grinding efficiency for a ball mill

Abstract

The application provides a mill efficiency real-time optimization control method and system for a ball mill, twelve real-time input data are acquired, the real-time input data are input into a pre-trained neural network model, a mill power prediction value for reflecting the mill efficiency output by the neural network model is accepted, and a deviation value between the mill power prediction value and a mill power expected value is determined; when the deviation value is in a first range, step S31 is entered; when the deviation value is in a second range, step S32 is entered; and when the deviation value is in a third range, step S33 is entered. The application can realize real-time optimization control of the mill efficiency, can realize high-precision control of the mill efficiency through regulation and control of target adjustment parameters, can ensure that the mill is in an expected mill efficiency, and can solve abnormal working conditions such as "underload" or "expansion" of the mill.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ball mill, in particular to a grinding efficiency real-time optimization control method and system for ball mill. BACKGROUND

[0002] Grinding is a key link in the production process of mineral processing, and is a particle size preparation operation before separation of crushed ore. The mill is the core equipment for grinding production process, and is one of the largest power consumption equipment in the whole plant. Its production cost and efficiency affect the economic and technical indicators of the mineral processing enterprise. How to improve the grinding efficiency and reduce the grinding cost is a research topic for many production enterprises.

[0003] Grinding operation is completed in a continuously rotating mill cylinder. The cylinder is filled with grinding media (such as steel balls, rods, and special-shaped ball rods). The grinding media are driven to produce complex impact, grinding, and shearing action during the rotation of the cylinder. The continuously fed ore is gradually ground under the action of the grinding media.

[0004] Ball mill is a common equipment for grinding operation. For example, a wet overflow type ball mill is a hollow cylinder with hollow bearings at both ends. Ore is fed from one end and discharged from the other end. The grinding medium of the ball mill is steel balls. When the mill rotates, the steel balls are brought to a certain height and then fall. The steel balls near the cylinder wall move in a parabolic motion with the cylinder, and the steel balls in the center of the cylinder roll. The falling steel balls impact and crush the ore, and the rolling steel balls grind and crush the ore. In the case of non-centrifugal motion, the greater the number and arc of the steel balls moving in a parabolic motion, the stronger the crushing ability of the steel balls on the ore, and the higher the grinding efficiency.

[0005] Please refer to the accompanying drawings Figure 1 The current common grinding system using ball mill includes ball mill, mill medium supplement belt, pump pool, underflow pump, and hydrocyclone. The mill medium supplement belt is communicated with the inlet of the ball mill. The discharge port of the ball mill is communicated with the pump pool through a pipeline. The outlet at the bottom of the pump pool is communicated with the underflow pump through a pipeline. The pump is pumped to the hydrocyclone through the underflow pump. The sand outlet of the hydrocyclone is communicated with the inlet of the ball mill, thereby forming a closed grinding system.

[0006] In the prior art, in order to obtain a higher grinding efficiency, the inlet parameters of the ball mill need to be adjusted artificially, so as to obtain a higher grinding efficiency. However, due to the large number of devices related to the production process of the ball mill, the models are different, and are affected by the production environment. When the desired grinding efficiency is obtained, the inlet parameters of the ball mill, such as the mill feed quantity, need to be adjusted multiple times, which is time-consuming and laborious.

[0007] Therefore, it is necessary to provide a grinding efficiency real-time optimization control method and system for ball mill to solve or at least alleviate the above defects. SUMMARY

[0008] The main purpose of the present application is to provide a grinding efficiency real-time optimization control method and system for a ball mill, to solve the problem that in the prior art, due to the large number of devices related to the production process of the ball mill, the models are different, and are affected by the production environment, when obtaining the desired grinding efficiency, the inlet parameters of the ball mill, such as the mill feed amount, need to be adjusted multiple times, which is time-consuming and laborious.

[0009] To achieve the above-mentioned purpose, the present application provides a grinding efficiency real-time optimization control method for a ball mill, comprising the steps of:

[0010] S1, obtaining real-time input data; wherein the real-time input data includes: cyclone inlet pipe slurry flow real-time value, cyclone inlet pipe slurry concentration real-time value, cyclone internal medium pressure real-time value, cyclone overflow fine ore grade real-time value, cyclone overflow fine ore concentration real-time value, cyclone overflow fine ore particle size real-time value, average filling amount of grinding medium per unit time, mill audio real-time value, total mass of material in the mill real-time value, mill internal material filling rate real-time value, grinding concentration real-time value, and mill speed real-time value;

[0011] S2, inputting the real-time input data into a pre-trained neural network model, and accepting the mill power prediction value output by the neural network model for reflecting the grinding efficiency; wherein the neural network model contains the mapping relationship between the real-time input data and the mill power prediction value;

[0012] S3, determining the deviation value between the mill power prediction value and the mill power expected value; when the deviation value is within a first range, entering step S31; when the deviation value is within a second range, entering step S32; when the deviation value is within a third range, entering step S33; wherein,

[0013] S31, increasing the target adjustment parameter by a preset increase, then inputting the adjusted target adjustment parameter into the neural network model again, and accepting the mill power prediction value output by the neural network model; then returning to step S3; wherein the target adjustment parameter includes one or more of cyclone inlet pipe slurry flow, average filling amount of grinding medium per unit time, mill internal material filling rate, and mill speed;

[0014] S32, reducing the target adjustment parameter by a preset decrease, then inputting the adjusted target adjustment parameter into the neural network model again, and accepting the mill power prediction value output by the neural network model; then returning to step S3;

[0015] S33, maintaining the current ball mill to continue production within the tolerance range of the target adjustment parameter.

[0016] Preferably, before the step of "then input the adjusted target adjustment parameter into the neural network model again" in the step S31, further comprising the step of:

[0017] S300, judging whether the adjusted target adjustment parameter is within the preset range, if yes, entering the step of "then input the adjusted target adjustment parameter into the neural network model again"; if no, outputting a preset alarm instruction;

[0018] And / or,

[0019] Before the step of "then input the adjusted target adjustment parameter into the neural network model again" in the step S32, further comprising the step of:

[0020] S301, judging whether the adjusted target adjustment parameter is within the preset range, if yes, entering the step of "then input the adjusted target adjustment parameter into the neural network model again"; if no, outputting a preset alarm instruction.

[0021] Preferably, before the step of "then returning to step S3" in the step S3, further comprising the step of:

[0022] S302, prioritizing the target adjustment parameters.

[0023] Preferably, the neural network model in the step S2 is obtained by the following steps:

[0024] S21, obtaining historical sample data for training and testing in N historical periods; the historical sample data includes historical sample input data and historical sample output data; wherein the historical sample input data includes cyclone inlet pipe slurry flow historical value, cyclone inlet pipe slurry concentration historical value, cyclone internal medium pressure historical value, cyclone overflow fine ore grade historical value, cyclone overflow fine ore concentration historical value, cyclone overflow fine ore particle size historical value, average filling amount of grinding medium per unit time historical value, mill audio historical value, total mass of material in the mill historical value, mill internal material filling rate historical value, grinding concentration historical value, and mill speed historical value; the historical sample output data includes mill power historical value;

[0025] S22, pre-processing the historical sample data; wherein the pre-processing includes smoothing processing and normalization processing;

[0026] S23, inputting the N1 historical sample input data as input values of the preliminary neural network model, inputting N1 corresponding historical sample output data as output values of the preliminary neural network model, and training the preliminary neural network model;

[0027] S24, inputting N2 historical sample input data into the preliminary neural network model for testing, and outputting N2 network prediction values;

[0028] S25, calculating the mean square error between the N2 network prediction values and N2 historical sample output data corresponding to the N2 historical sample input data, and determining whether the mean square error is within a set range;

[0029] S26, when the mean square error is within the set range, taking the preliminary neural network model as the neural network model.

[0030] Preferably, the neural network model is an ANN neural network model, specifically:

[0031]

[0032] wherein Y0 is a target output, X it is a current input, WH ij is a weight neuron of a link between an i-th input and a j-th hidden, m is a number of input neurons, WO j is a connection weight between a j-th hidden neuron and an output neuron, f h is a hidden neuron activation function, f o is an output neuron activation function, b j is a bias of a j-th hidden neuron, b o is a bias of an output neuron, and HN is a number of hidden neurons in the output neuron.

[0033] Preferably, the step S23 of training the preliminary neural network model specifically comprises the following steps:

[0034] S231, initializing or updating network parameters of the model, the parameters comprising: WH ij , m, WO j , f h , f o , b j , b o , and HN;

[0035] S232, inputting historical sample input data to an input layer, and inputting weights between variables to a hidden layer;

[0036] S233, judging whether the total number of input variables is same as the number of data points, if yes, calculating mean square error function MSE, if not, outputting a prompt warning information and re-inputting the historical sample input data;

[0037] S234, judging whether MSE is less than a preset value, if yes, calculating parameter error terms in input layer and hidden layer and bias terms of hidden neuron activation function and output neuron activation function through a target automatic correction algorithm;

[0038] S235, calculating the model network parameters to be adjusted according to the parameter error terms and the bias terms and returning the model network parameters to be adjusted to step S231 to update the network parameters;

[0039] S236, if MSE is less than the preset value, the model training is ended.

[0040] Preferably, the normalization processing step in step S22 specifically needs to normalize the historical values of the cyclone inlet pipe slurry flow, the cyclone internal medium pressure, the cyclone overflow fine ore particle size, the average filling amount of grinding media per unit time, the mill audio, the total mass of the material in the mill and the mill speed.

[0041] Preferably, in the normalization step in step S22, a 0-1 standardization model is used to normalize the historical sample input data, and the 0-1 standardization model is:

[0042]

[0043] wherein x is the historical sample input data, max and min are the maximum and minimum values of the historical sample input data.

[0044] Preferably, in the smoothing step in step S22, a mean smoothing model is used for smoothing, and the mean smoothing model is:

[0045]

[0046] wherein n is the nth data of the historical sample input data, and y is the moving smoother setting value.

[0047] The application further provides a grinding efficiency real-time optimization control system, comprising a ball mill, a mill medium supplementing belt, a pump pool, an underflow pump and a hydrocyclone; wherein the mill medium supplementing belt is communicated with a feeding port of the ball mill, a discharge port of the ball mill is communicated with the pump pool through a pipeline, an outlet of the pump pool is communicated with an inlet of the underflow pump through a pipeline, an outlet of the underflow pump is communicated with an inlet of the hydrocyclone through a pipeline, and a sand outlet of the hydrocyclone is communicated with the feeding port of the ball mill through a pipeline; further comprising a control system, which is used for executing the grinding efficiency real-time optimization control method for the ball mill as described above; the control system comprises an acquisition module, a mill power prediction value generation module and a regulation and control module; wherein,

[0048] The acquisition module is used for acquiring the real-time input data.

[0049] The mill power prediction value generation module is used for inputting the real-time input data into a pre-trained neural network model and accepting a mill power prediction value for reflecting the grinding efficiency output by the neural network model.

[0050] The regulation and control module is used for determining a deviation value between the mill power prediction value and a mill power expected value; when the deviation value is within a first range, step S31 is entered; when the deviation value is within a second range, step S32 is entered; and when the deviation value is within a third range, step S33 is entered.

[0051] Compared with the prior art, the application has the following beneficial effects:

[0052] The application provides a grinding efficiency real-time optimization control method and system for a ball mill, which acquires twelve real-time input data, inputs the real-time input data into a pre-trained neural network model, accepts a mill power prediction value for reflecting the grinding efficiency output by the neural network model, determines a deviation value between the mill power prediction value and a mill power expected value, when the deviation value is within a first range, step S31 is entered, when the deviation value is within a second range, step S32 is entered, and when the deviation value is within a third range, step S33 is entered. The application can realize real-time optimization control of the grinding efficiency, can realize high-precision control of the grinding efficiency through regulation and control of target adjustment parameters, can ensure that the ball mill is in the expected grinding efficiency, and can solve the abnormal working conditions such as "underload" or "bulging" of the ball mill. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings described below only illustrate some of the embodiments of the present application, and all other drawings obtained by those of ordinary skill in the art without creative effort based on these drawings also belong to the protection scope of the present application.

[0054] Figure 1 A connection relationship diagram of the process equipment in the prior art grinding process;

[0055] Figure 2 A flowchart of an embodiment of the present application;

[0056] Figure 3 A flowchart of an embodiment of the present application;

[0057] Figure 4 A flowchart of an embodiment of the present application;

[0058] Figure 5 A corresponding relationship diagram of the grinding efficiency and the active power of the mill in an embodiment of the present application.

[0059] The purposes, functional features and advantages of the present application will be further illustrated with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0060] It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0061] The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort belong to the protection scope of the present application.

[0062] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present application are only used to explain the relative positional relationship, movement condition, etc. between the components in a certain specific posture (as shown in the drawings), and if the specific posture changes, the directional indications also change accordingly.

[0063] In addition, the description related to "first", "second" and the like in the present application is only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions of various embodiments can be combined with each other, but it must be based on the realization of the ordinary skilled in the art, when the combination of technical solutions appears contradictory or cannot be realized, it should be considered that the combination of technical solutions does not exist, nor within the protection scope required by the present application.

[0064] In the first aspect, please refer to the accompanying drawings Figures 2-4 In an embodiment of the present application, a method for real-time optimization control of grinding efficiency of a ball mill is provided, comprising the steps of:

[0065] S1, acquiring real-time input data; wherein the real-time input data includes: cyclone inlet pipe slurry flow real-time value, cyclone inlet pipe slurry concentration real-time value, cyclone internal medium pressure real-time value, cyclone overflow fine ore grade real-time value, cyclone overflow fine ore concentration real-time value, cyclone overflow fine ore particle size real-time value, average filling amount of grinding medium per unit time, mill audio real-time value, total mass of material in the mill real-time value, mill internal material filling rate real-time value, mill grinding concentration real-time value and mill speed real-time value; It should be noted that the real-time input data is an influencing factor that can directly affect the power of the mill, and then affect the grinding efficiency. In other embodiments, those skilled in the art can also increase or decrease other influencing factors that can affect the power of the mill according to actual needs, which can be determined according to actual needs.

[0066] Among them, preferably, the real-time input data can be obtained by the following methods respectively: (A) the real-time value of the flow of the slurry in the inlet pipe of the cyclone, which is detected by the flow meter installed on the conveying pipe; (B) the real-time value of the concentration of the slurry in the inlet pipe of the cyclone, which is detected by the concentration meter installed on the conveying pipe; (C) the real-time value of the medium pressure in the cyclone, which is detected by the pressure transmitter; (D) the real-time value of the fine ore grade of the overflow in the cyclone, which is detected by the grade meter installed on the overflow pipe; (E) the real-time value of the concentration of the fine ore overflow in the cyclone, which is detected by the concentration meter installed on the overflow pipe; (F) the real-time value of the particle size of the fine ore overflow in the cyclone, which is detected by the particle size meter installed on the overflow pipe; (G) the average filling amount of the grinding medium per unit time, which is detected by weighing when the medium is added each time; (H) the real-time value of the audio frequency of the mill, which is detected by the electric ear installed on the mill; (I) the total mass of the material in the mill, which is detected by the weighing sensor for the overall mill equipment; (J) the filling rate of the material in the mill, which is obtained by the soft measurement method, i.e. the volume of the inlet material / the effective space volume of the mill; (K) the grinding concentration, which is detected by the slurry concentration analyzer in real time; (L) the real-time value of the mill speed, which is detected by the speed meter installed at the end of the motor.

[0067] Among them, preferably, the target adjustment parameters can be adjusted by the following methods respectively: (a) the real-time value of the slurry flow in the inlet pipe of the cyclone, which can be adjusted by the regulating valve on the pipe; (b) the average filling amount of the grinding medium per unit time, which can be adjusted by controlling the input amount per unit time; (c) the filling rate of the material in the mill, which can be changed by adjusting the total amount of the inlet material; (d) the real-time value of the mill speed, which can be adjusted by the frequency conversion speed regulation or hydraulic speed regulation mechanism according to the actual configuration of the motor.

[0068] S2, input the real-time input data into the pre-trained neural network model, and accept the mill power prediction value output by the neural network model for reflecting the grinding efficiency; wherein the neural network model contains the mapping relationship between the real-time input data and the mill power prediction value;

[0069] It should be noted by those skilled in the art that the specific type of the neural network model can be set according to actual needs, for example, it can be an ANN-based neural network model or other types of neural network models, which can be selected by those skilled in the art according to actual needs. It can be understood by those skilled in the art that since the neural network model is obtained by learning a large amount of sample data in advance, and the accuracy of the trained neural network model is high, and the neural network model contains the mapping relationship between the real-time input data and the preliminary prediction value of the mill power, the real-time input data is input into the pre-trained neural network model, and the preliminary prediction value of the mill power is obtained, and the grinding efficiency under the current working condition can be indirectly predicted according to the value.

[0070] S3, determine the deviation value between the mill power prediction value and the mill power expected value; when the deviation value is in a first range, enter step S31; when the deviation value is in a second range, enter step S32; when the deviation value is in a third range, enter step S33; wherein,

[0071] S31, increase the target adjustment parameter by a preset increase amount, then input the adjusted target adjustment parameter into the neural network model again, and accept the mill power prediction value output by the neural network model; then return to step S3; wherein, the target adjustment parameter includes one or more of cyclone inlet pipe slurry flow, average filling amount of grinding medium per unit time, material filling rate in the mill, and mill speed;

[0072] S32, decrease the target adjustment parameter by a preset decrease amount, then input the adjusted target adjustment parameter into the neural network model again, and accept the mill power prediction value output by the neural network model; then return to step S3;

[0073] S33, maintain the current ball mill to continue production within the tolerance range of the target adjustment parameter.

[0074] It is worth noting that the present application determines the deviation value between the mill power prediction value and the mill power expected value, wherein preferably, when the mill power prediction value is greater than the mill power expected value preset value, step S31 is entered to increase the target adjustment parameter by a preset increase amount, wherein the target adjustment parameter includes one or more of the mill feed amount, the mill water amount, the average filling amount of the grinding medium per unit time, the material filling rate in the mill, and the mill speed; for example, the water amount is increased by 5%, and then the adjusted target adjustment parameter is input into the neural network again, and the mill power prediction value output by the neural network model is accepted, at this time the neural network can predict the mill power prediction value corresponding to the adjusted target adjustment parameter, and then returns to step S3 to continue to judge the deviation value between the mill power prediction value and the mill power expected value.

[0075] Similarly, when the deviation value is in the second range, that is, the mill power prediction value is less than the mill power expected value preset value, step S32 is entered to decrease the target adjustment parameter by a preset decrease amount, for example, the water amount is decreased by 5%, and then the adjusted target adjustment parameter is input into the neural network again, and the mill power prediction value output by the neural network model is accepted, at this time the neural network can predict the mill power prediction value corresponding to the adjusted target adjustment parameter, and then returns to step S3.

[0076] In addition, when the deviation value is in the third range, that is, the mill power prediction value and the mill power expected value are very close, it means that the mill power reaches the mill power expected value under the current state / adjustment parameter.

[0077] The present application adjusts the target adjustment parameter according to the preset adjustment amount in a cyclic judgment manner, and the target adjustment parameter can be selected by those skilled in the art according to actual needs, so that the mill power is always within the preset range of the mill power expected value, realizing real-time optimization control of the grinding efficiency, saving time and effort.

[0078] It is worth noting that according to research and development, under certain conditions, the grinding efficiency has a clear corresponding relationship with the mill active power, and the curve effect is basically consistent, which can be referred to in the accompanying Figure 5 , especially the region where the grinding efficiency is best coincides with the region where the mill active power is maximum, and the curve effect is known to those skilled in the art, so in the case where the grinding efficiency cannot be detected, the mill power is one of the important means to reflect the grinding efficiency. It should be noted that the mill power of the present application is the mill active power.

[0079] In the technical solution of the present application, twelve real-time input data are acquired, the real-time input data are input into a pre-trained neural network model, a mill power prediction value reflecting the grinding efficiency output by the neural network model is accepted, a deviation value between the mill power prediction value and a mill power expected value is determined, when the deviation value is in a first range, step S31 is entered, when the deviation value is in a second range, step S32 is entered, and when the deviation value is in a third range, step S33 is entered. The present application can realize real-time optimization control of the grinding efficiency, can realize high-precision control of the grinding efficiency through target adjustment parameter adjustment, and can ensure that the mill is in the expected grinding efficiency, thereby solving the abnormal working conditions such as mill "underload" or "bulging".

[0080] As a preferred embodiment, before the step "then the adjusted target adjustment parameter is input into the neural network model again" in step S31, the step of:

[0081] S300, determining whether the adjusted target adjustment parameter is in a preset range, if yes, entering the step of "then the adjusted target adjustment parameter is input into the neural network model again", and if no, outputting a preset alarm instruction.

[0082] And / or,

[0083] Before the step "then the adjusted target adjustment parameter is input into the neural network model again" in step S32, the step of:

[0084] S301, determining whether the adjusted target adjustment parameter is in a preset range, if yes, entering the step of "then the adjusted target adjustment parameter is input into the neural network model again", and if no, outputting a preset alarm instruction.

[0085] Those skilled in the art should understand that the target adjustment parameter has a better adjustable range, and the effect is not obvious or other problems will be caused when the target adjustment parameter is higher or lower than the adjustable range. Based on this, the present embodiment determines whether the adjusted target adjustment parameter is in a preset range, if yes, the subsequent step is entered normally, and if no, it is indicated that the adjusted target adjustment parameter is not in the preset range, and a preset alarm instruction is output at this time, for example, an alarm instruction is sent to a background or an operation and maintenance personnel.

[0086] In other embodiments, other adjustment methods can also be requested, for example, the traditional method mentioned in the background art can be switched to at this time, so that the final mill power is in the expected value range.

[0087] As a preferred embodiment, before the step "then returning to step S3" in step S3, the step of:

[0088] S302, the target adjustment parameters are prioritized. It is worth noting that according to the above description, the target adjustment parameters include one or more of the cyclone inlet pipe pulp flow, the average filling amount of the grinding medium per unit time, the material filling rate in the mill and the mill speed, and the influence factor of each adjustment parameter on the mill power is different, so in the embodiment of the application, the target adjustment parameters are prioritized, for example, the cyclone inlet pipe pulp flow can be adjusted first, followed by the average filling amount of the grinding medium per unit time, and so on. Those skilled in the art can set it according to actual needs. Thus, the mill power can be quickly and reasonably adjusted to the desired value.

[0089] Further, the neural network model in step S2 is obtained by the following steps:

[0090] S21, obtaining historical sample data for training and testing in N historical periods; the historical sample data includes historical sample input data and historical sample output data; wherein the historical sample input data includes cyclone inlet pipe pulp flow historical value, cyclone inlet pipe pulp concentration historical value, cyclone medium pressure historical value, cyclone overflow fine ore grade historical value, cyclone overflow fine ore concentration historical value, cyclone overflow fine ore particle size historical value, average filling amount of grinding medium per unit time historical value, mill audio historical value, total mass of material in the mill historical value, material filling rate in the mill historical value, grinding concentration historical value and mill speed historical value; the historical sample output data includes mill power historical value;

[0091] Those skilled in the art should understand that in order to obtain a higher precision mill power prediction value, the total amount of historical sample data in the present application is set to 10000-100000. In addition, the pre-set values (thresholds) of each input data can be pre-set different data according to different process categories. The above obtained input data is labeled and redundant data outside the threshold is removed. For example, the pre-set range of the mill audio in the present application is [4, 20], if the mill audio in a group of input data is lower than 4mA or higher than 20mA, this group of input data is removed, and the removal of other data in the input data is the same, and is not repeated here.

[0092] S22, pre-processing the historical sample data; wherein the pre-processing includes smoothing processing and normalization processing; as a specific example: the normalization processing step in the step S22 specifically needs to normalize the cyclone inlet pipe slurry flow historical value, cyclone internal medium pressure historical value, cyclone overflow fine ore particle size historical value, grinding medium average filling amount per unit time historical value, mill audio historical value, mill internal material total mass historical value, grinding concentration historical value and mill speed historical value. Through normalization processing, the data can be limited within a certain range, thereby reducing the adverse effects caused by singular sample data.

[0093] Further, in the step of normalization in the step S22, a 0-1 standardization model is used to normalize the historical sample input data, and the 0-1 standardization model is as follows:

[0094]

[0095] Wherein x is the historical sample input data, max is the maximum value of the historical sample input data, and min is the minimum value of the historical sample input data.

[0096] In addition, it is worth noting that the mill internal material filling rate belongs to [0, 1], so it does not need to be normalized.

[0097] Further, at the same time, in order to make the input data more accurate, the input data is smoothed. The smoothing processing mainly includes mean smoothing, median smoothing, Gaussian smoothing and bilateral smoothing methods. In the present model, the mean smoothing method is mainly used for smoothing the input data, and the principle expression is as follows:

[0098]

[0099] Wherein n is the nth data of the historical sample input data, and y is the moving smoother setting value.

[0100] The present application is preset to 5, which can be set according to the specific amount of data obtained within the allowed range of the granulation process. If there are 5 groups of water addition historical data: F(1), F(2), F(3)……F(5) and the real-time water addition data FT in the current period, the calculation steps of smoothing processing are as follows:

[0101]

[0102] Wherein, the value of F' is the result value of the water adding amount data in the current period after smoothing processing. The smoothing processing method of other data in the input data is the same, and will not be described one by one. Finally, N sample data are randomly extracted from the preprocessed input data for model training and model testing, and the extraction ratio preset by the application is [50%, 80%].

[0103] S23, input N1 historical sample input data as input values of the preliminary neural network model, input N1 corresponding historical sample output data as output values of the preliminary neural network model, and train the preliminary neural network model;

[0104] S24, input N2 historical sample input data to the preliminary neural network model for testing, and output N2 network prediction values;

[0105] S25, calculate the mean square error between the N2 network prediction values and N2 historical sample output data corresponding to the N2 historical sample input data, and determine whether the mean square error is within a set range; wherein, it is worth noting that by calculating the mean square error between the N2 network prediction values and N2 historical sample output data corresponding to the N2 historical sample input data, the prediction accuracy of the neural network model can be further improved.

[0106] S26, when the mean square error is within the set range, the preliminary neural network model is used as the neural network model.

[0107] In the embodiment of the application, the neural network model is trained and tested by a sufficient amount of sample data, and a high-precision neural network model can be obtained by calculating the mean square error of the test process, thereby laying a foundation for subsequent high-precision prediction.

[0108] As a preferred embodiment of the application, the neural network model is an ANN neural network model, specifically:

[0109]

[0110] Wherein, Y0 is the target output, X it is the current input, WH ij is the weight neuron of the link between the i-th input and the j-th hidden, m is the number of input neurons, WO j is the connection weight between the j-th hidden neuron and the output neuron, f h is the hidden neuron activation function, f o is the output neuron activation function, b j is the bias of the j-th hidden neuron, b o is the bias of the output neuron, and HN is the number of hidden neurons in the output neuron.

[0111] Further, the step S23 of training the preliminary neural network model specifically comprises the following steps:

[0112] S231, initializing or updating network parameters of the model, the parameters comprising: WH ij , m, WO j , f h , f o , b j , b o , HN;

[0113] S232, inputting historical sample input data into an input layer and inputting weights between variables into a hidden layer;

[0114] S233, judging whether the total number of input variables is the same as the number of data points, if yes, calculating a mean square error (MSE) function, if not, outputting a prompt warning information and inputting historical sample input data again;

[0115] S234, judging whether the MSE is less than a preset value, if no, calculating parameter error terms in the input layer and the hidden layer and bias terms of a hidden neuron activation function and an output neuron activation function through a target automatic correction algorithm;

[0116] S235, calculating the model network parameters to be adjusted according to the parameter error terms and the bias terms and returning the model network parameters to be adjusted to the step S231 to update the network parameters;

[0117] S236, if the MSE is less than the preset value, ending the model training.

[0118] The target automatic correction algorithm comprises three layers: an input layer, a hidden layer and an output layer. The preset input layer of the application has 12 input nodes: x1 is the flow rate of the slurry at the inlet pipe of the cyclone; x2 is the concentration of the slurry at the inlet pipe of the cyclone; x3 is the medium pressure in the cyclone; x4 is the fine ore grade of the overflow of the cyclone; x5 is the concentration of the fine ore of the overflow of the cyclone; x6 is the particle size of the fine ore of the overflow of the cyclone; x7 is the average filling amount of the grinding medium per unit time; x8 is the audio frequency of the mill; x9 is the total mass of the material in the mill; x 10 is the filling rate of the material in the mill; x 11 is the grinding concentration; x 12 is the rotational speed of the mill; and one bias node (a circle marked with +1), the input nodes can be increased or decreased according to requirements. After the input layer receives the sample input data, x1……x 12 are combined with a weight matrix WH ij , respectively, to obtain WH ij x+b (bias b = +1) form input hidden layer, through hidden neuron activation function f h (x n ) processing, get output results a ­1 ……a 12 , and then combined with the corresponding weight, bias, as the output layer input through the output neuron activation function processing, the output layer output final results, its output is: mill power.

[0119] Further, the mean square error in the step S25 is calculated by a mean square error model, the mean square error model is:

[0120]

[0121] Wherein, MSE is mean square error, Y a is target output, Y o is network output.

[0122] Secondly, the application also provides a grinding efficiency real-time optimization control system, comprising a ball mill, a mill medium supplement belt, a pump pool, an underflow pump and a hydrocyclone;The mill medium supplement belt is communicated with the feed inlet of the ball mill, the discharge outlet of the ball mill is communicated with the pump pool through a pipeline, the outlet of the pump pool is communicated with the inlet of the underflow pump through a pipeline, the outlet of the underflow pump is communicated with the inlet of the hydrocyclone through a pipeline, and the sand outlet of the hydrocyclone is communicated with the feed inlet of the ball mill through a pipeline;Further comprising a control system, the control system is used for executing the grinding efficiency real-time optimization control method for ball mill as described above;The control system comprises an acquisition module, a mill power prediction value generation module and a regulation and control module;Wherein,

[0123] The acquisition module is used for acquiring the real-time input data;

[0124] The mill power prediction value generation module is used for inputting the real-time input data into a pre-trained neural network model, and accepting the mill power prediction value output by the neural network model, which reflects the grinding efficiency;

[0125] The regulation and control module is used for determining the deviation value between the mill power prediction value and the mill power expected value;When the deviation value is in the first range, step S31 is entered;When the deviation value is in the second range, step S32 is entered;When the deviation value is in the third range, step S33 is entered.

[0126] In the technical scheme of the present application, twelve real-time input data are acquired, the real-time input data are input into a pre-trained neural network model, a deviation value between a mill power prediction value output by the neural network model and a mill power expected value is determined, when the deviation value is within a first range, step S31 is entered, when the deviation value is within a second range, step S32 is entered, and when the deviation value is within a third range, step S33 is entered. The present application can realize real-time optimization control of the grinding efficiency, can realize high-precision control of the grinding efficiency through regulation of the target adjustment parameter, and can ensure that the mill is in the expected grinding efficiency, thereby solving the abnormal working conditions such as mill "underload" or "bulging".

[0127] The above are only preferred embodiments of the present application, and do not limit the patent range of the present application, and any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, are also included in the patent protection range of the present application.

[0128] The above are only preferred embodiments of the present application, and do not limit the patent range of the present application, and any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, are also included in the patent protection range of the present application.

Claims

1. A method for real-time optimization control of grinding efficiency for a ball mill, characterized in that, The method comprises the steps of: S1, obtaining real-time input data; wherein the real-time input data comprises: real-time values of cyclone inlet pipe ore pulp flow, cyclone inlet pipe ore pulp concentration, medium pressure in the cyclone, overflow fine ore grade in the cyclone, overflow fine ore concentration in the cyclone, overflow fine ore particle size in the cyclone, average filling amount of grinding medium per unit time, real-time value of mill audio, real-time value of total mass of material in the mill, real-time value of material filling rate in the mill, real-time value of grinding concentration, and real-time value of mill speed; S2, inputting the real-time input data into a pre-trained neural network model, and accepting a mill power prediction value output by the neural network model for reflecting the grinding efficiency; wherein the neural network model comprises a mapping relationship between the real-time input data and the mill power prediction value; S3, determining a deviation value between the mill power prediction value and a mill power expected value; when the deviation value is within a first range, entering step S31; when the deviation value is within a second range, entering step S32; when the deviation value is within a third range, entering step S33; wherein, S31, increasing a target adjustment parameter by a preset increase amount, then inputting the adjusted target adjustment parameter into the neural network model again, and accepting a mill power prediction value output by the neural network model; then returning to step S3; wherein the target adjustment parameter comprises one or more of cyclone inlet pipe ore pulp flow, average filling amount of grinding medium per unit time, material filling rate in the mill, and mill speed; S32, decreasing the target adjustment parameter by a preset decrease amount, then inputting the adjusted target adjustment parameter into the neural network model again, and accepting a mill power prediction value output by the neural network model; then returning to step S3; S33, maintaining the current ball mill to continue production within the tolerance range of the target adjustment parameter; The neural network model in step S2 is obtained by the following steps: S21, obtaining historical sample data meeting the training and testing in N historical periods; the historical sample data comprises historical sample input data and historical sample output data; wherein the historical sample input data comprises historical values of cyclone inlet pipe ore pulp flow, cyclone inlet pipe ore pulp concentration, medium pressure in the cyclone, overflow fine ore grade in the cyclone, overflow fine ore concentration in the cyclone, overflow fine ore particle size in the cyclone, average filling amount of grinding medium per unit time, mill audio, total mass of material in the mill, material filling rate in the mill, grinding concentration, and mill speed; and the historical sample output data comprises mill power historical values; S22, pre-processing the historical sample data; wherein the pre-processing comprises smoothing processing and normalization processing; S23, input N1 historical sample input data as input value of the preliminary neural network model, input N1 corresponding historical sample output data as output value of the preliminary neural network model, and train the preliminary neural network model; S24, input N2 historical sample input data into the preliminary neural network model for testing, and output N2 network prediction values; S25, calculate mean square error between N2 network prediction values and N2 corresponding historical sample output data of the N2 historical sample input data, and determine whether the mean square error is within a set range; S26, when the mean square error is within the set range, take the preliminary neural network model as the neural network model; The neural network model is an ANN neural network model, and specifically is: where Y0is the target output, X it is the current input, WH ij is the weight of the link between the ith input and the jth hidden neuron, m is the number of input neurons, WO j is the weight of the connection between the jth hidden neuron and the output neuron, f h is the hidden neuron activation function, f o is the output neuron activation function, b j is the bias of the jth hidden neuron, b o is the bias of the output neuron, HNis the number of hidden neurons in the output neuron.

2. The real-time optimization control method for the milling efficiency of a ball mill according to claim 1, characterized in that, The step S31 further includes the following steps before "then input the adjusted target adjustment parameter into the neural network model again": S300, determine whether the adjusted target adjustment parameter is within a preset range, if yes, enter the step of "then input the adjusted target adjustment parameter into the neural network model again"; if no, output a preset alarm instruction; And / or, The step S32 further includes the following steps before "then input the adjusted target adjustment parameter into the neural network model again": S301, determine whether the adjusted target adjustment parameter is within a preset range, if yes, enter the step of "then input the adjusted target adjustment parameter into the neural network model again"; if no, output a preset alarm instruction.

3. The real-time optimization control method for the milling efficiency of a ball mill according to claim 1, characterized in that, The step S3 further includes the following steps before "then return to step S3": S302, perform priority sorting on the target adjustment parameter.

4. The real-time optimization control method for the milling efficiency of a ball mill according to claim 1, characterized by, The step S23 of training the preliminary neural network model specifically includes the following steps: S231, initialize or update network parameters of the model, the parameters comprising: WH ij , m, WO j , f h , f o , b j , b o , HN; S232, input historical sample input data to an input layer, and input weights between variables to a hidden layer; S233, determine whether the total number of input variables is the same as the number of data points, if yes, calculate a mean square error function MSE, if no, output a prompt warning information, and input historical sample input data again; S234, determine whether the MSE is less than a preset value, if the MSE is greater than or equal to the preset value, calculate parameter error items in the input layer and the hidden layer and bias items of a hidden neuron activation function and an output neuron activation function through a target automatic correction algorithm; S235, calculate a model network parameter to be adjusted according to the parameter error items and the bias items, and return the model network parameter to be adjusted to step S231 to update the network parameter; S236, if the MSE is less than the preset value, the model training is completed.

5. The real-time optimization control method for the milling efficiency of a ball mill according to claim 1, characterized by, The normalization processing step in the step S22 needs to perform normalization processing on the cyclone inlet pipeline ore slurry flow historical value, the cyclone internal medium pressure historical value, the cyclone overflow fine ore particle size historical value, the grinding medium average filling amount per unit time historical value, the mill audio historical value, the mill internal material total mass historical value and the mill rotation speed historical value.

6. The real-time optimization control method for the milling efficiency of a ball mill according to claim 1, characterized in that, In the step S22, the historical sample input data is normalized by using a 0-1 standardization model, which is: wherein x is the historical sample input data, max is the maximum value of the historical sample input data, and min is the minimum value of the historical sample input data.

7. The mill efficiency real time optimization control method for a ball mill according to claim 1, characterized by, In the step S22, the historical sample input data is smoothed by using a mean smoothing model, which is: wherein n is the nth data of the historical sample input data, and y is a moving smoother setting value.

8. A real-time grinding efficiency optimization control system, comprising a ball mill, a mill media replenishment belt, a pump sump, an underflow pump, and a hydrocyclone; wherein, The mill medium supplementing belt is communicated with the feeding port of the ball mill, the discharge port of the ball mill is communicated with the pump pool through a pipeline, the outlet of the pump pool is communicated with the inlet of the underflow pump through a pipeline, the outlet of the underflow pump is communicated with the inlet of the hydrocyclone through a pipeline, and the sand outlet of the hydrocyclone is communicated with the feeding port of the ball mill through a pipeline. The control system is used for executing the real-time optimization control method for the grinding efficiency of the ball mill according to any one of claims 1-7. The acquisition module is used for acquiring the real-time input data. The mill power prediction value generation module is used for inputting the real-time input data into a pre-trained neural network model and accepting the mill power prediction value output by the neural network model, which is used for reflecting the grinding efficiency. The control module is used for determining the deviation value between the mill power prediction value and the mill power expected value, entering step S31 when the deviation value is within a first range, entering step S32 when the deviation value is within a second range, and entering step S33 when the deviation value is within a third range.

Images