A method and system for accurate prediction of grinding efficiency for a semi-autogenous mill

Abstract

The application provides a grinding efficiency precise prediction method and system for a semi-autogenous mill, ten real-time input data are acquired, the real-time input data are input into a pre-trained neural network model, a mill power preliminary prediction value output by the neural network model is accepted, a deviation value of the mill power preliminary prediction value and a mill power measured value is determined, whether the deviation value is within a preset range is judged, when the deviation value is within the preset range, the mill power preliminary prediction value is output as a mill power final prediction value for reflecting the grinding efficiency, so as to realize the precise prediction of the grinding efficiency. The application adopts the deep learning mode, and proposes the mode of predicting the mill power to predict the grinding efficiency, so that the grinding efficiency can be timely and accurately predicted under the change of the working condition, and the problems of the existing mode that the discovery is not timely, the manpower is consumed, the judgment result is fuzzy or the analysis is lagged are solved.

Description

Technical Field

[0001] This invention relates to the field of semi-autogenous grinding mill technology, and in particular to a method and system for accurately predicting the grinding efficiency of a semi-autogenous grinding mill. Background Technology

[0002] Grinding is a crucial step in the mineral processing production process. It is the particle size preparation operation before the ore is crushed and sorted. Among them, the mill is the core equipment used in the grinding production process and is one of the equipment with the highest power consumption in the entire plant. Its production cost and efficiency affect various economic and technical indicators of mineral processing enterprises. How to improve grinding efficiency and reduce grinding cost is a research topic for many production enterprises.

[0003] Grinding is carried out inside a continuously rotating mill cylinder, which contains grinding media (such as steel balls, rods, and irregularly shaped rods). As the cylinder rotates, the grinding media are driven to generate complex impact, grinding, and shearing effects, and the ore continuously fed into the cylinder is gradually ground up under the action of the grinding media.

[0004] Semi-autogenous grinding mills are common equipment in ore grinding operations. When the ore particle size is large, only a portion of grinding media steel balls needs to be added. Please refer to the appendix. Figure 1 Currently, common grinding systems using semi-autogenous mills include a semi-autogenous mill, a mill media replenishment belt, a mill fresh feed belt, a screening machine (e.g., a linear screen), and a pump sump. The mill media replenishment belt and the mill fresh feed belt are both connected to the inlet of the semi-autogenous mill. The mill discharge outlet of the semi-autogenous mill is connected to the inlet of the screening machine. The outlet of the screening machine is connected to the pump sump through a pipeline. The return sand outlet of the screening machine is connected to the inlet of the semi-autogenous mill, thus forming a closed-loop grinding system.

[0005] In existing technologies, the grinding efficiency of semi-autogenous mills is generally judged based on manual experience. During manual inspection, operators frequently estimate the particle size of the slurry by observing its color or solid-liquid concentration. When the tested slurry particle size is too coarse or too fine, operators typically compare the test values ​​with process parameters and judge the mill load and grinding efficiency based on manual experience by observing the mill outlet discharge and grinding noise. However, due to frequent changes in ore properties (grindability grade), operators often cannot accurately and promptly determine the current condition of the mill, leading to slurry particle size exceeding its upper and lower limits, or even abnormal operating conditions such as mill underload or bloating, resulting in poor grinding efficiency. Furthermore, this manual method suffers from problems such as untimely detection, high manpower consumption, and unclear or delayed judgment results.

[0006] Therefore, it is necessary to propose a method and system for accurate prediction of grinding efficiency in semi-autogenous mills to solve or at least alleviate the above-mentioned defects. Summary of the Invention

[0007] The main objective of this invention is to provide a method and system for accurately predicting the grinding efficiency of a semi-autogenous mill, in order to solve the problems of untimely detection and delayed analysis in the prior art where the grinding efficiency of a semi-autogenous mill is judged by manual experience.

[0008] In a first aspect, to achieve the above objective, the present invention provides a method for accurately predicting the grinding efficiency of a semi-autogenous mill, comprising the following steps:

[0009] S1, acquire real-time input data; wherein, the real-time input data includes the real-time value of mill feed rate, real-time value of screening machine return rate, real-time value of water addition, average filling amount of grinding media per unit time, real-time value of mill audio, real-time value of total mass of material in mill, real-time value of material filling rate in mill, real-time value of grinding concentration, real-time value of the proportion of feed particle size greater than 10mm, and real-time value of mill speed.

[0010] S2, the real-time input data is input into a pre-trained neural network model, and the preliminary predicted value of the mill power output by the neural network model is received; wherein, the neural network model contains a mapping relationship between the real-time input data and the preliminary predicted value of the mill power;

[0011] S3, determine the deviation between the preliminary predicted value of the mill power and the measured value of the mill power, and determine whether the deviation is within a preset range;

[0012] S4, when the deviation value is within the preset range, the preliminary predicted value of the mill power is output as the final predicted value of the mill power to reflect the grinding efficiency; when the deviation value is not within the preset range, the corresponding real-time input data is re-inputted into the neural network model for training.

[0013] Preferably, the neural network model in step S2 is obtained through the following steps:

[0014] S21, acquire historical sample data for training and testing over N historical periods; the historical sample data includes historical sample input data and historical sample output data; wherein, the historical sample input data includes historical values ​​of mill feed rate, screening machine return rate, water addition, average filling amount of grinding media per unit time, mill audio frequency, total mass of material in mill, material filling rate in mill, grinding concentration, percentage of feed particles larger than 10mm, and mill rotation speed; the historical sample output data includes historical mill power.

[0015] S22, preprocess the historical sample data; wherein, the preprocessing includes smoothing and normalization.

[0016] S23, use N1 historical sample input data as input values ​​of the preliminary neural network model and N1 corresponding historical sample output data as output values ​​of the preliminary neural network model to train the preliminary neural network model;

[0017] S24, input N2 historical sample input data into the preliminary neural network model for testing, and output N2 network prediction values;

[0018] S25, calculate the mean square error between the N2 network prediction values ​​and the N2 historical sample output data corresponding to the N2 historical sample input data, and determine whether the mean square error is within the set range.

[0019] S26, when the mean square error is within the set range, the preliminary neural network model is used as the neural network model.

[0020] Preferably, the neural network model is an ANN neural network model, specifically:

[0021]

[0022] Where Y0 is the target output, X it WH represents the current input quantity. ij Here, WO represents the weighted neuron connecting the i-th input and the j-th hidden, where m is the number of input neurons. j f is the connection weight between the j-th hidden neuron and the output neuron. h f is the activation function for the hidden neuron. o b is the activation function for the output neuron. j For the j-th bias hidden neuron, b o H is the bias of the output neuron, and HN is the number of hidden neurons in the output neuron.

[0023] Preferably, the training of the preliminary neural network model in step S23 specifically includes the following steps:

[0024] S231, Initialize or update the network parameters of the model, the parameters including: WH ij , m, WO j f h f o b j b o HN;

[0025] S232, the input layer takes historical sample input data as input, and the hidden layer takes the weights between the variables as input;

[0026] S233, determine whether the total number of input variables is the same as the number of data points. If they are the same, calculate the mean square error function (MSE). If they are different, output a warning message and re-enter the historical sample input data.

[0027] S234, determine whether MSE is less than the preset value. If MSE is greater than or equal to the preset value, calculate the parameter error terms in the input layer and hidden layer, as well as the bias terms of the activation functions of the hidden neurons and the output neurons, through the target automatic correction algorithm.

[0028] S235, calculate the model network parameters to be adjusted based on the parameter error term and the bias term, and return the model network parameters to be adjusted to step S231 to update the network parameters;

[0029] S236. If the MSE is less than the preset value, the model training ends.

[0030] Preferably, the normalization process in step S22 specifically normalizes the historical values ​​of the mill feed rate, the screen return rate, the water addition, the average filling rate of the grinding media per unit time, the mill audio frequency, the total mass of materials in the mill, the grinding concentration, and the mill speed.

[0031] Preferably, in step S22, during the normalization process, a 0-1 normalization model is used to normalize the historical sample input data. The 0-1 normalization model is as follows:

[0032]

[0033] Where x represents the historical sample input data, max represents the maximum value of the historical sample input data, and min represents the minimum value of the historical sample input data.

[0034] Preferably, in the smoothing process of step S22, a mean smoothing model is used for smoothing, and the mean smoothing model is:

[0035]

[0036] Where n is the nth data point of the historical sample input data, and y is the moving smoother setting value.

[0037] Preferably, the mean square error in step S25 is calculated using a mean square error model, which is:

[0038]

[0039] Where MSE is the mean square error, Y a For the target output, Yo For network output.

[0040] In a second aspect, the present invention also provides a grinding efficiency accurate prediction system for a semi-autogenous mill, for executing the above-mentioned grinding efficiency accurate prediction method for a semi-autogenous mill, including an acquisition module, a preliminary prediction output value generation module and a final prediction value generation module.

[0041] The acquisition module is used to acquire the real-time input data;

[0042] The preliminary prediction output value generation module is used to input the real-time input data into a pre-trained neural network model and receive the preliminary prediction value of the mill power output by the neural network model.

[0043] The final prediction value generation module is used to determine the deviation between the preliminary prediction value of the mill power and the measured value of the mill power, and to determine whether the deviation value is within a preset range; when the deviation value is within the preset range, the preliminary prediction value of the mill power is output as the final prediction value of the mill power used to reflect the grinding efficiency; when the deviation value is not within the preset range, the corresponding real-time input data is re-inputted into the neural network model for training.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] This invention provides a method and system for accurately predicting the grinding efficiency of a semi-autogenous mill. By acquiring ten real-time input data points, these data are fed into a pre-trained neural network model to obtain a preliminary prediction of the mill power. Error analysis is then performed between the preliminary prediction and the measured mill power to achieve accurate prediction of grinding efficiency. This invention employs deep learning and proposes using mill power prediction to forecast grinding efficiency, thereby enabling timely and accurate prediction of grinding efficiency under changing operating conditions. It also solves the problems of untimely detection, high manpower consumption, ambiguous judgment results, or delayed analysis in existing methods. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0047] Figure 1 This is a diagram showing the connection relationships of equipment in the existing grinding process.

[0048] Figure 2 This is a flowchart illustrating one embodiment of the present invention;

[0049] Figure 3 This is a flowchart illustrating the process of obtaining the neural network model in step S2 of one embodiment of the present invention;

[0050] Figure 4 This is a flowchart illustrating the steps involved in training the preliminary neural network model in step S23 of one embodiment of the present invention.

[0051] Figure 5 This is a schematic diagram of an automatic target correction algorithm in one embodiment of the present invention;

[0052] Figure 6 This is a graph showing the relationship between grinding efficiency and active power of the mill in one embodiment of the present invention.

[0053] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0054] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0056] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0057] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0058] Firstly, please refer to the appendix. Figures 2-5 An embodiment of the present invention provides a method for accurately predicting the grinding efficiency of a semi-autogenous mill, comprising the following steps:

[0059] S1, acquire real-time input data; wherein, the real-time input data includes the real-time values ​​of mill feed rate, screening machine return rate, water addition rate, average filling rate of grinding media per unit time, mill audio frequency, total mass of material in the mill, material filling rate in the mill, grinding concentration, percentage of feed particles larger than 10mm, and mill rotation speed; it should be noted that the real-time input data are influencing factors that can directly affect mill power, and thus affect grinding efficiency. In other embodiments, those skilled in the art can also add or reduce other influencing factors affecting mill power according to actual needs, specifically as needed.

[0060] Preferably, the real-time input data can be obtained in the following ways: (A) Real-time value of mill feed rate, which can be detected in real time by a belt scale installed on the mill's new feed conveyor belt; (B) Real-time value of screen return rate, which can be detected in real time by a belt scale installed on the screen conveyor belt; (C) Real-time value of water addition, which can be detected in real time by an electromagnetic flow meter installed on the water addition pipeline; (D) Average filling amount of grinding media per unit time, which can be detected by weighing each time the media is added; (E) Real-time value of mill audio, which can be detected by an electromagnetic flow meter installed on the mill's feed conveyor belt. (F) The total mass of material inside the mill is measured in real time by a weighing sensor; (G) The material filling rate inside the mill can be obtained by a soft measurement method, i.e., the volume of inlet material / the effective space volume of the mill; (H) The grinding concentration is obtained by a slurry concentration analyzer; (I) The real-time value of the proportion of feed particles larger than 10mm can be obtained by a visual particle size analyzer installed above the feed belt; (J) The real-time value of the mill speed is measured in real time by a speed measuring instrument installed at the end of the motor and then converted.

[0061] S2, the real-time input data is input into a pre-trained neural network model, and the preliminary predicted value of the mill power output by the neural network model is received; wherein, the neural network model contains a mapping relationship between the real-time input data and the preliminary predicted value of the mill power. Those skilled in the art should note 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 according to actual needs. Those skilled in the art will understand that since the neural network model is pre-learned using a large amount of sample data, and the accuracy of the trained neural network model is higher, and since the neural network model contains a mapping relationship between the real-time input data and the preliminary predicted value of the mill power, inputting the real-time input data into the pre-trained neural network model can obtain the preliminary predicted value of the mill power, and based on this value, the grinding efficiency under the current operating conditions can be indirectly predicted.

[0062] S3, determine the deviation between the preliminary predicted value of the mill power and the measured value of the mill power, and determine whether the deviation is within a preset range. It is worth noting that in this embodiment, the measured value of the mill power is obtained manually or through fuzzy calculation, or it can be detected by a power transmitter (or, the power value can also be a current value, as the two are linearly related, and a current transformer is used for current detection). Calculating the deviation between the two values ​​reflects the degree of deviation between the preliminary predicted value of the mill power and the measured value of the mill power. The preset range can be flexibly adjusted according to the required accuracy.

[0063] S4, when the deviation value is within the preset range, the preliminary predicted value of the mill power is output as the final predicted value of the mill power used to reflect the grinding efficiency; when the deviation value is not within the preset range, the corresponding real-time input data is re-inputted into the neural network model for training. In other words, when the error between the preliminary predicted value of the mill power and the measured value of the mill power is small, it indicates that the pre-trained neural network model has high accuracy and can obtain a predicted value close to the true value. At this time, the preliminary predicted value of the mill power is output as the final predicted value of the mill power used to reflect the grinding efficiency, thereby accurately predicting the mill power under the current operating conditions, and thus accurately predicting the grinding efficiency under the current operating conditions.

[0064] It is worth noting that, according to research findings, under certain conditions, there is a clear correlation between grinding efficiency and the active power of the mill, and the curves are basically consistent. (See attached diagram.) Figure 6In particular, the region where grinding efficiency is optimal perfectly matches the region where the mill's active power is at its maximum. This curve effect is technically known to those skilled in the art. Therefore, when grinding efficiency cannot be measured, using mill power to reflect grinding efficiency is one of the important methods. It should be noted that the mill power in this application refers to the mill's active power.

[0065] The method for accurately predicting grinding efficiency in a semi-autogenous mill provided in this application acquires ten real-time input data points, feeds these data into a pre-trained neural network model to obtain a preliminary prediction of mill power, and then performs error analysis on the preliminary prediction and the measured mill power to achieve accurate prediction of grinding efficiency. This invention employs deep learning and proposes a method to predict grinding efficiency by predicting mill power, thereby enabling timely and accurate prediction of grinding efficiency under changing operating conditions. It also solves the problems of untimely detection, high manpower consumption, ambiguous judgment results, or delayed analysis in existing methods.

[0066] Furthermore, the neural network model in step S2 is obtained through the following steps:

[0067] S21, acquire historical sample data for training and testing over N historical periods; the historical sample data includes historical sample input data and historical sample output data; wherein, the historical sample input data includes historical values ​​of mill feed rate, screening machine return rate, water addition, average filling amount of grinding media per unit time, mill audio frequency, total mass of material in mill, material filling rate in mill, grinding concentration, percentage of feed particles larger than 10mm, and mill rotation speed; the historical sample output data includes historical mill power.

[0068] Those skilled in the art should understand that, in order to obtain a more accurate mill power prediction value, the total amount of historical sample data in this application is set to a range of 10,000 to 100,000. Furthermore, the preset values ​​(thresholds) for each input data can be preset differently depending on the scope of different processes. The input data obtained above is labeled, and redundant data not within the threshold is removed. For example, the preset range for mill audio in this invention is [4, 20], unit: mA. If the mill audio in a set of input data is lower than 4 mA or higher than 20 mA, then this set of input data is removed. The removal method for other data in the input data is similar and will not be elaborated further.

[0069] S22, preprocessing the historical sample data; wherein, the preprocessing includes smoothing and normalization; as a specific example: the normalization step in step S22 specifically normalizes the historical values ​​of mill feed rate, screening machine return rate, water addition rate, average filling rate of grinding media per unit time, mill audio frequency, total mass of material in the mill, grinding concentration, and mill rotation speed. Through normalization, the data can be limited to a certain range, thereby reducing the adverse effects caused by outlier sample data.

[0070] Furthermore, in step S22, during the normalization process, a 0-1 normalization model is used to normalize the historical sample input data. The 0-1 normalization model is as follows:

[0071]

[0072] Where x represents the historical sample input data, max represents the maximum value of the historical sample input data, and min represents the minimum value of the historical sample input data. For example, in this invention, the mill feed rate W... F The preset range is [400, 1000], then W F The minimum value min is 400 and the maximum value max is 1000. If the current period's W... F If the value is 460, then normalize it:

[0073]

[0074] In addition, it is worth noting that the material filling rate inside the mill is [0, 1], so no normalization is required.

[0075] Furthermore, to make the input data more accurate, it is smoothed. Smoothing methods mainly include mean smoothing, median smoothing, Gaussian smoothing, and bilateral smoothing. In this model, mean smoothing is primarily used for input data smoothing, and its principle expression is shown below:

[0076]

[0077] Where n is the nth data point of the historical sample input data, and y is the moving smoother setting value.

[0078] The preset value in this invention is 5, which can be set according to the specific amount of data obtained within the allowable range of the granulation process. If there are 5 sets of historical water addition data: F(1), F(2), F(3)...F(5) and the real-time water addition data FT in the current period, the calculation steps for smoothing them are as follows:

[0079] F(1)=F(2) (1)

[0080] F(2)=F(3) (2)

[0081] F(3)=F(4) (3)

[0082] F(4)=F(5) (4)

[0083] F(5)=F T (5)

[0084]

[0085] Here, the value of F' is the result of smoothing the water addition data within the current cycle. The smoothing method for other data in the input data is similar and will not be described in detail here. Finally, N sample data are randomly selected from the preprocessed input data for model training and model testing. In a preferred embodiment, the sampling ratio preset by the present invention is [50%, 80%].

[0086] S23, use N1 historical sample input data as input values ​​of the preliminary neural network model and N1 corresponding historical sample output data as output values ​​of the preliminary neural network model to train the preliminary neural network model;

[0087] S24, input N2 historical sample input data into the preliminary neural network model for testing, and output N2 network prediction values;

[0088] S25, calculate the mean square error between the N2 network predicted values ​​and the 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; it is worth noting that by calculating the mean square error between the N2 network predicted values ​​and the 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.

[0089] S26, when the mean square error is within the set range, the preliminary neural network model is used as the neural network model.

[0090] In this embodiment of the invention, a high-precision neural network model can be obtained by training and testing the neural network model with a sufficient amount of sample data and by calculating the mean square error of the testing process, thus laying the foundation for subsequent high-precision prediction.

[0091] In a preferred embodiment of the present invention, the neural network model is an ANN neural network model, specifically:

[0092]

[0093] Where Y0 is the target output, X it WH represents the current input quantity. ij Here, WO represents the weighted neuron connecting the i-th input and the j-th hidden, where m is the number of input neurons. j f is the connection weight between the j-th hidden neuron and the output neuron. h f is the activation function for the hidden neuron. o b is the activation function for the output neuron. j For the j-th bias hidden neuron, b o H is the bias of the output neuron, and HN is the number of hidden neurons in the output neuron.

[0094] Furthermore, the training of the preliminary neural network model in step S23 specifically includes the following steps:

[0095] S231, Initialize or update the network parameters of the model, the parameters including: WH ij , m, WO j f h f o b j b o HN;

[0096] S232, the input layer takes historical sample input data as input, and the hidden layer takes the weights between the variables as input;

[0097] S233, determine whether the total number of input variables is the same as the number of data points. If they are the same, calculate the mean square error function (MSE). If they are different, output a warning message and re-enter the historical sample input data.

[0098] S234, determine whether MSE is less than the preset value. If MSE is greater than or equal to the preset value, calculate the parameter error terms in the input layer and hidden layer, as well as the bias terms of the activation functions of the hidden neurons and the output neurons, through the target automatic correction algorithm.

[0099] S235, calculate the model network parameters to be adjusted based on the parameter error term and the bias term, and return the model network parameters to be adjusted to step S231 to update the network parameters;

[0100] S236. If the MSE is less than the preset value, the model training ends.

[0101] The target automatic correction algorithm comprises three layers: an input layer, a hidden layer, and an output layer. The invention pre-defines 10 input nodes in the input layer: x1 is the mill feed rate; x2 is the screen return rate; x3 is the water addition rate; x4 is the average filling amount of grinding media per unit time; x5 is the mill audio signal; x6 is the total mass of material in the mill; x7 is the material filling rate in the mill; x8 is the grinding concentration; x9 is the percentage of feed particles larger than 10mm; x 10 The input layer specifies the mill speed and an offset node (marked with a circle of +1). Input nodes can be added or removed as needed. After receiving sample input data, the input layer will convert x1...x... 10 With weight matrix WH respectively ij Combined, with WH ij The input to the hidden layer is in the form *x + b (bias b = +1), and then passes through the activation function f of the hidden neurons. h (x n The processing of ) yields the output results a1……a 10 Then, combined with the corresponding weights and biases, it is used as the input to the output layer. After being processed by the activation function of the output neuron, the final result is output in the output layer, which is: mill power.

[0102] Furthermore, the mean square error in step S25 is calculated using a mean square error model, which is as follows:

[0103]

[0104] Where MSE is the mean square error, Y a For the target output, Y o For network output.

[0105] In a second aspect, the present invention also provides a system for accurately predicting grinding efficiency, for performing the above-mentioned method for accurately predicting grinding efficiency for a semi-autogenous mill, including an acquisition module, a preliminary prediction output value generation module, and a final prediction value generation module.

[0106] The acquisition module is used to acquire the real-time input data;

[0107] The preliminary prediction output value generation module is used to input the real-time input data into a pre-trained neural network model and receive the preliminary prediction value of the mill power output by the neural network model.

[0108] The final prediction value generation module is used to determine the deviation between the preliminary prediction value of the mill power and the measured value of the mill power, and to determine whether the deviation value is within a preset range; when the deviation value is within the preset range, the preliminary prediction value of the mill power is output as the final prediction value of the mill power used to reflect the grinding efficiency; when the deviation value is not within the preset range, the corresponding real-time input data is re-inputted into the neural network model for training.

[0109] As can be seen from the above technical solution, the present invention provides a grinding efficiency accurate prediction system for a semi-autogenous mill, which is used to execute the grinding efficiency accurate prediction method for a semi-autogenous mill as described above. It can achieve accurate prediction of grinding efficiency, and can predict grinding efficiency in a timely and accurate manner under changing operating conditions. It also solves the problems of untimely detection, high manpower consumption, ambiguous judgment results, or delayed analysis in existing methods.

[0110] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for accurately predicting the grinding efficiency of a semi-autogenous mill, characterized in that, Including the following steps: S1, acquire real-time input data; wherein, the real-time input data includes the real-time value of mill feed rate, real-time value of screening machine return rate, real-time value of water addition, average filling amount of grinding media per unit time, real-time value of mill audio, real-time value of total mass of material in mill, real-time value of material filling rate in mill, real-time value of grinding concentration, real-time value of the proportion of feed particle size greater than 10mm, and real-time value of mill speed. S2, the real-time input data is input into a pre-trained neural network model, and the preliminary predicted value of the mill power output by the neural network model is received; wherein, the neural network model contains a mapping relationship between the real-time input data and the preliminary predicted value of the mill power; S3, determine the deviation between the preliminary predicted value of the mill power and the measured value of the mill power, and determine whether the deviation is within a preset range; S4, when the deviation value is within the preset range, the preliminary predicted value of the mill power is output as the final predicted value of the mill power to reflect the grinding efficiency; when the deviation value is not within the preset range, the corresponding real-time input data is re-inputted into the neural network model for training. The material filling rate inside the mill is obtained using a soft measurement method, which is the inlet material volume / effective mill space volume. The region where the grinding efficiency is optimal coincides perfectly with the region where the mill's active power is maximum. When grinding efficiency cannot be measured, the mill power is used to reflect the grinding efficiency. The neural network model is an ANN neural network model, specifically: Where Y0 is the target output, X it WH represents the current input quantity. ij Here, WO represents the weighted neuron connecting the i-th input and the j-th hidden, where m is the number of input neurons. j f is the connection weight between the j-th hidden neuron and the output neuron. h f is the activation function for the hidden neuron. o b is the activation function for the output neuron. j Let b0 be the bias hidden neuron of the j-th neuron, b0 be the bias of the output neuron, and HN be the number of hidden neurons in the output neuron.

2. The method for accurately predicting the grinding efficiency of a semi-autogenous mill according to claim 1, characterized in that, The neural network model in step S2 is obtained through the following steps: S21, acquire historical sample data for training and testing over N historical periods; the historical sample data includes historical sample input data and historical sample output data; wherein, the historical sample input data includes historical values ​​of mill feed rate, screening machine return rate, water addition, average filling amount of grinding media per unit time, mill audio frequency, total mass of material in mill, material filling rate in mill, grinding concentration, percentage of feed particles larger than 10mm, and mill rotation speed; the historical sample output data includes historical mill power. S22, preprocess the historical sample data; wherein, the preprocessing includes smoothing and normalization. S23, use N1 historical sample input data as input values ​​of the preliminary neural network model and N1 corresponding historical sample output data as output values ​​of the preliminary neural network model to 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 the mean square error between the N2 network prediction values ​​and the N2 historical sample output data corresponding to the N2 historical sample input data, and determine whether the mean square error is within the set range. S26, when the mean square error is within the set range, the preliminary neural network model is used as the neural network model.

3. The method for accurately predicting the grinding efficiency of a semi-autogenous mill according to claim 1, characterized in that, Step S23, "training the preliminary neural network model," specifically includes the following steps: S231, Initialize or update the network parameters of the model, the parameters including: WH ij , m, WO j f h f o b j b0, HN; S232, the input layer inputs the historical sample input data, and the hidden layer inputs the weights between the variables; S233, determine whether the total number of input variables is the same as the number of data points. If they are the same, calculate the mean square error function (MSE). If they are different, output a warning message and re-enter the historical sample input data. S234, determine whether MSE is less than the preset value. If MSE is greater than or equal to the preset value, calculate the parameter error terms in the input layer and hidden layer, as well as the bias terms of the activation functions of the hidden neurons and the output neurons, through the target automatic correction algorithm. S235, calculate the model network parameters to be adjusted based on the parameter error term and the bias term, and return the model network parameters to be adjusted to step S231 to update the network parameters; S236. If the MSE is less than the preset value, the model training ends.

4. The method for accurately predicting the grinding efficiency of a semi-autogenous mill according to claim 2, characterized in that, The normalization process in step S22 specifically normalizes the historical values ​​of the mill feed rate, the screen return rate, the water addition, the average filling rate of the grinding media per unit time, the mill audio frequency, the total mass of materials in the mill, the grinding concentration, and the mill speed.

5. The method for accurately predicting the grinding efficiency of a semi-autogenous mill according to claim 2, characterized in that, In step S22, during the normalization process, a 0-1 normalization model is used to normalize the historical sample input data. The 0-1 normalization model is as follows: Where x represents the historical sample input data, max represents the maximum value of the historical sample input data, and min represents the minimum value of the historical sample input data.

6. The method for accurately predicting the grinding efficiency of a semi-autogenous mill according to claim 2, characterized in that, In step S22, the smoothing process is performed using a mean smoothing model, which is as follows: Where n is the nth data point of the historical sample input data, and y is the moving smoother setting value.

7. The method for accurately predicting grinding efficiency of a semi-autogenous mill according to claim 2, characterized in that, The mean square error in step S25 is calculated using a mean square error model, which is as follows: Where MSE is the mean square error, Y a For the target output, Y o For network output.

8. A system for accurately predicting the grinding efficiency of a semi-autogenous mill, used to execute the method for accurately predicting the grinding efficiency of a semi-autogenous mill as described in any one of claims 1-7, characterized in that, It includes an acquisition module, a preliminary prediction output value generation module, and a final prediction value generation module; The acquisition module is used to acquire the real-time input data; The preliminary prediction output value generation module is used to input the real-time input data into a pre-trained neural network model and receive the preliminary prediction value of the mill power output by the neural network model. The final prediction value generation module is used to determine the deviation between the preliminary prediction value of the mill power and the measured value of the mill power, and to determine whether the deviation value is within a preset range; when the deviation value is within the preset range, the preliminary prediction value of the mill power is output as the final prediction value of the mill power used to reflect the grinding efficiency. If the deviation value is not within the preset range, the corresponding real-time input data is re-inputted into the neural network model for training.

Images