A method and device for predicting raw ball ratio of a ball making machine

Abstract

The present application provides a method and device for predicting the proportion of raw balls of a pelletizing machine. The method comprises: quantizing the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine, the amount of water added, and the amount of feed obtained in the same prediction period to the same interval according to their respective shrinkage ratios, and obtaining the characteristic vector of the key influencing factors of pelletizing based on the quantified data and the types and proportions of the components of the mixture, the proportion of the binder, and the original moisture content of the mixture, and then inputting the characteristic vector into a pre-trained proportion prediction model, and obtaining the proportion prediction value of raw balls of different specifications based on the mapping relationship contained in the proportion prediction model. In this way, on the one hand, the method provided in the embodiment of the present application does not require manual screening, saves time and effort, and has high detection efficiency. On the other hand, the method can predict the proportion of raw balls in real time, realize real-time control of the pelletizing machine, and better guide actual production operations.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and device for predicting the raw ball ratio of a ball making machine. Background Art

[0002] In the steel industry, pelletizing is a commonly used technology for iron ore refining. Among them, the pelletizing process is an important process in the iron ore pelletizing production line. The stability and improvement of the quality of raw balls mainly depends on the pelletizing process. As the core equipment in the pelletizing process, the pelletizing machine mainly includes two types: disc pelletizing machine and cylinder pelletizing machine. Cylinder pelletizing machine is generally used in large-scale and high-output production lines. However, due to the current dominance of small and medium-sized pelletizing production lines, disc pelletizing machines are more common.

[0003] When the pelletizing machine is working, the materials move along their own different tracks in the pelletizing machine to form raw balls of different diameters, which are discharged from the pelletizing machine and fall into the subsequent raw ball receiving device. The pelletizing particle size of the pelletizing machine is a key parameter of the pelletizing process. The higher the proportion of qualified raw balls, the higher the output of the pelletizing machine. In the prior art, the method for evaluating the raw ball proportion of the pelletizing machine is to obtain a batch of raw balls from the ball outlet area of ​​the pelletizing machine, and then use several sieves of different specifications to screen out raw balls of different specifications, and then calculate the proportion of raw balls of different specifications. If the proportion of raw balls meets the range of process requirements, it is evaluated that the pelletizing particle size of the pelletizing machine meets the requirements. In the prior art, a more advanced method is to use machine vision technology to collect the raw ball images of the ball outlet area in real time, use image recognition and processing technology to calculate the particle size of each raw ball, and then calculate the proportion of raw balls of different specifications. If the raw ball proportion meets the preset threshold range, it is considered that the raw balls manufactured by the pelletizing machine meet the production requirements and the relevant setting parameters meet the pelletizing requirements.

[0004] However, the first method requires manual screening, which is time-consuming and labor-intensive, has low detection efficiency, and is a sampling offline detection method that cannot meet the real-time requirements of production. Although the second method is advanced and has good real-time performance, this method can only detect the proportion of raw balls after the ball making machine completes the ball making work. The parameter control can only be carried out through closed-loop control based on feedback, which is not real-time enough, resulting in the detection results of visual technology still cannot be used for real-time control of the ball making machine. Summary of the invention

[0005] The present application provides a method and device for predicting the raw ball ratio of a pelletizing machine, which can be used to solve the problem that the manual screening method of detecting raw ball quality in the prior art is time-consuming and labor-intensive, and cannot meet the real-time requirements of production; the method of machine vision technology can only detect the raw ball ratio after the pelletizing machine completes the pelletizing work, and the parameter control can only be carried out through feedback-based closed-loop control, which is not real-time enough, resulting in the detection results of the visual technology still cannot be used for real-time control of the pelletizing machine.

[0006] In a first aspect, an embodiment of the present application provides a method for predicting the raw ball ratio of a ball making machine, the method comprising:

[0007] Obtain the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine and the water supply provided by the water supply system in the current prediction period;

[0008] Obtaining current raw material information for making green balls by a ball making machine within a current forecast period, wherein the current raw material information includes feed amount, types and proportions of each component in the mixture, proportion of binder in the mixture, and original moisture content of the mixture;

[0009] Quantify the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine, the water supply amount and the material supply amount to the same interval according to their respective shrinkage ratios;

[0010] According to the quantified rotation speed of the pelletizing machine, the quantified inclination angle of the pelletizing disk of the pelletizing machine, the quantified water supply, the quantified feed amount, the type and proportion of each component in the mixture, the proportion of the binder in the mixture and the original moisture content of the mixture, a characteristic vector affecting the pelletizing is obtained;

[0011] The characteristic vector affecting ball making is input into a proportion prediction model to obtain the proportion prediction values ​​of raw balls of different specifications. The proportion prediction model includes a mapping relationship between the characteristic vector affecting ball making and the proportion prediction values ​​of raw balls of different specifications.

[0012] In combination with the first aspect, in an implementation manner of the first aspect, the step of generating a proportion prediction model includes:

[0013] Obtain sample data within N historical prediction periods; the sample data within each historical prediction period includes the historical rotation speed of the pelletizing machine, the historical inclination angle of the pelletizing disk of the pelletizing machine, the historical amount of water added, the historical raw material information of the pelletizing machine for manufacturing sample raw balls, and the actual measured values ​​of the proportion of sample raw balls of different specifications; the historical raw material information includes the historical feed amount, the types and proportions of each component in the historical mixture, the proportion of the binder in the historical mixture, and the original moisture content of the mixture in the historical mixture; the actual measured values ​​of the proportion of sample raw balls of different specifications are obtained by using machine vision to collect images of sample raw balls within each historical prediction period, analyze and calculate after processing;

[0014] Quantify the historical rotation speed of the pelletizing machine, the historical inclination angle of the pelletizing disk of the pelletizing machine, the historical water supply amount and the historical feed amount into the same interval according to their respective contraction ratios;

[0015] According to the quantified historical rotation speed of the pelletizing machine, the quantified historical pelletizing disk inclination angle of the pelletizing machine, the quantified historical water supply, the quantified feed amount, the types and proportions of each component of the historical mixture, the proportion of the binder in the historical mixture and the original moisture content of the mixture in the historical mixture, the characteristic vectors of N samples affecting the pelletizing are obtained;

[0016] The characteristic vectors of the N samples influencing ball making are used as the input of the prediction model, and the actual values ​​of the proportions of sample balls of different specifications in N historical prediction periods are used as the output of the prediction model, and the proportion prediction model is trained by using the time back propagation method;

[0017] The weight parameters, bias parameters and learning factors of the share prediction model are continuously updated through iterative training;

[0018] If the difference between the predicted value of the proportion of sample raw balls of different specifications by the proportion prediction model and the actual value of the proportion of sample raw balls of different specifications reaches the preset tolerance range, or the proportion prediction model reaches the set maximum number of iterations through iterative calculation, the training ends and the last updated weight parameter, bias parameter and learning factor are saved.

[0019] In combination with the first aspect, in one possible implementation of the first aspect, the proportion prediction model is established based on a long short-term memory neural network prediction model LSTM.

[0020] In combination with the first aspect, in an implementable manner of the first aspect, the measured values ​​of the proportions of the sample green balls of different specifications are obtained specifically by the following steps:

[0021] An image acquisition device is used to obtain image information of the ball-discharging area of ​​the ball-making machine;

[0022] Performing image preprocessing on the image information of the ball output area to separate the image information of the sample raw ball and the background image information;

[0023] According to the image information of the sample green ball, a central bright spot of the sample green ball is obtained;

[0024] Determine the outline of the sample green ball according to the image information of the sample green ball and the background image information;

[0025] Obtaining the particle size of the sample green ball according to the central bright spot of the sample green ball and the outline of the sample green ball;

[0026] Determining the specifications of the sample green balls according to the particle size of the sample green balls and the corresponding relationship between the preset particle size range and the green ball specifications;

[0027] Count the total number of sample balls within the historical forecast period, as well as the number of sample balls of different specifications;

[0028] According to the total number of sample green balls in the historical prediction period and the number of sample green balls of different specifications, the actual measured values ​​of the proportions of sample green balls of different specifications are determined.

[0029] In combination with the first aspect, in an implementation manner of the first aspect, the current raw material information is obtained specifically by the following steps:

[0030] The current raw material information for producing green balls by the pelletizer during the current forecast period is obtained from the feeding system located upstream of the pelletizer.

[0031] In combination with the first aspect, in one possible implementation of the first aspect, the contraction ratio corresponding to the rotation speed of the pelletizing machine is the maximum rotation speed of the pelletizing machine;

[0032] The contraction ratio corresponding to the water supply volume is the maximum water supply volume of the water supply pipe in the water supply system;

[0033] The shrinkage ratio corresponding to the feeding amount is the maximum feeding amount of the feeding belt;

[0034] The contraction ratio corresponding to the inclination angle of the pelletizing disk of the pelletizing machine is the maximum inclination angle of the pelletizing machine.

[0035] In combination with the first aspect, in an implementation manner of the first aspect, the method further includes:

[0036] If the difference between the actual measured value of the proportion of raw balls and the predicted value of the model is greater than a preset threshold, the characteristic vector affecting ball making is used as the input of the prediction model, and the actual measured value of the proportion of raw balls and the actual value of the proportion of N sample raw balls are used as the output of the prediction model. The proportion prediction model is updated online to obtain an updated proportion prediction model.

[0037] In combination with the first aspect, in an implementable method of the first aspect, the predicted values ​​of the proportion of raw balls of different specifications include the predicted values ​​of the proportion of qualified large balls, the predicted values ​​of the proportion of unqualified large balls, the predicted values ​​of the proportion of qualified medium balls, the predicted values ​​of the proportion of qualified small balls and the predicted values ​​of the proportion of unqualified small balls.

[0038] In a second aspect, an embodiment of the present application provides a device for predicting the raw ball ratio of a ball making machine, the device comprising:

[0039] An acquisition unit is used to acquire the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine and the water supply provided by the water supply system in the current prediction period; and to acquire the current raw material information of the pelletizing machine for making raw balls in the current prediction period, wherein the current raw material information includes the feed amount, the type and proportion of each component in the mixture, the proportion of the binder in the mixture and the original moisture content of the mixture;

[0040] The processing unit is used to quantify the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine, the water supply amount and the feed amount to the same interval according to their respective shrinkage ratios; and obtain a characteristic vector affecting pelletizing according to the quantified rotation speed of the pelletizing machine, the quantified inclination angle of the pelletizing disk of the pelletizing machine, the quantified water supply amount, the quantified feed amount, the types and proportions of each component in the mixture, the proportion of the binder in the mixture and the original moisture content of the mixture;

[0041] The prediction unit is used to input the characteristic vector affecting ball making into a proportion prediction model to obtain the proportion prediction value of raw balls of different specifications. The proportion prediction model includes a mapping relationship between the characteristic vector affecting ball making and the proportion prediction value of raw balls of different specifications.

[0042] In conjunction with the second aspect, in an implementation manner of the second aspect, the step of generating a share prediction model includes:

[0043] Obtain sample data within N historical prediction periods; the sample data within each historical prediction period includes the historical rotation speed of the pelletizing machine, the historical inclination angle of the pelletizing disk of the pelletizing machine, the historical amount of water added, the historical raw material information of the pelletizing machine for manufacturing sample raw balls, and the actual measured values ​​of the proportion of sample raw balls of different specifications; the historical raw material information includes the historical feed amount, the types and proportions of each component in the historical mixture, the proportion of the binder in the historical mixture, and the original moisture content of the mixture in the historical mixture; the actual measured values ​​of the proportion of sample raw balls of different specifications are obtained by using machine vision to collect images of sample raw balls within each historical prediction period, analyze and calculate after processing;

[0044] Quantify the historical rotation speed of the pelletizing machine, the historical inclination angle of the pelletizing disk of the pelletizing machine, the historical water supply amount and the historical feed amount into the same interval according to their respective contraction ratios;

[0045] According to the quantified historical rotation speed of the pelletizing machine, the quantified historical pelletizing disk inclination angle of the pelletizing machine, the quantified historical water supply, the quantified feed amount, the types and proportions of each component of the historical mixture, the proportion of the binder in the historical mixture and the original moisture content of the mixture in the historical mixture, the characteristic vectors of N samples affecting the pelletizing are obtained;

[0046] The characteristic vectors of the N samples influencing ball making are used as the input of the prediction model, and the actual values ​​of the proportions of sample balls of different specifications in N historical prediction periods are used as the output of the prediction model, and the proportion prediction model is trained by using the time back propagation method;

[0047] The weight parameters, bias parameters and learning factors of the share prediction model are continuously updated through iterative training;

[0048] If the difference between the predicted value of the proportion of sample raw balls of different specifications by the proportion prediction model and the actual value of the proportion of sample raw balls of different specifications reaches the preset tolerance range, or the proportion prediction model reaches the set maximum number of iterations during iterative operation, the training ends and the last updated weight parameter, bias parameter and learning factor are saved.

[0049] In combination with the second aspect, in one possible implementation of the second aspect, the proportion prediction model is established based on a long short-term memory neural network prediction model LSTM.

[0050] In conjunction with the second aspect, in an implementable manner of the second aspect, the measured values ​​of the proportions of the sample green balls of different specifications are obtained specifically by the following steps:

[0051] An image acquisition device is used to obtain image information of the ball-discharging area of ​​the ball-making machine;

[0052] Performing image preprocessing on the image information of the ball output area to separate the image information of the sample raw ball and the background image information;

[0053] According to the image information of the sample green ball, a central bright spot of the sample green ball is obtained;

[0054] Determine the outline of the sample green ball according to the image information of the sample green ball and the background image information;

[0055] Obtaining the particle size of the sample green ball according to the central bright spot of the sample green ball and the outline of the sample green ball;

[0056] Determining the specifications of the sample green balls according to the particle size of the sample green balls and the corresponding relationship between the preset particle size range and the green ball specifications;

[0057] Count the total number of sample balls within the historical forecast period, as well as the number of sample balls of different specifications;

[0058] According to the total number of sample green balls in the historical prediction period and the number of sample green balls of different specifications, the actual measured values ​​of the proportions of sample green balls of different specifications are determined.

[0059] In conjunction with the second aspect, in an implementation manner of the second aspect, the current raw material information is obtained specifically by the following steps:

[0060] The current raw material information for producing green balls by the pelletizer during the current forecast period is obtained from the feeding system located upstream of the pelletizer.

[0061] In combination with the second aspect, in an implementation of the second aspect, the contraction ratio corresponding to the rotation speed of the pelletizing machine is the maximum rotation speed of the pelletizing machine;

[0062] The contraction ratio corresponding to the water supply volume is the maximum water supply volume of the water supply pipe in the water supply system;

[0063] The shrinkage ratio corresponding to the feeding amount is the maximum feeding amount of the feeding belt;

[0064] The contraction ratio corresponding to the inclination angle of the pelletizing disk of the pelletizing machine is the maximum inclination angle of the pelletizing machine.

[0065] In conjunction with the second aspect, in an implementation manner of the second aspect, the processing unit is further used to:

[0066] If the difference between the actual measured value of the proportion of raw balls and the predicted value of the model is greater than a preset threshold, the characteristic vector affecting ball making is used as the input of the prediction model, and the actual measured value of the proportion of raw balls and the actual value of the proportion of N sample raw balls are used as the output of the prediction model. The proportion prediction model is updated online to obtain an updated proportion prediction model.

[0067] In combination with the second aspect, in an implementation method of the second aspect, the predicted values ​​of the proportion of raw balls of different specifications include the predicted values ​​of the proportion of qualified large balls, the predicted values ​​of the proportion of unqualified large balls, the predicted values ​​of the proportion of qualified medium balls, the predicted values ​​of the proportion of qualified small balls and the predicted values ​​of the proportion of unqualified small balls.

[0068] In the embodiment of the present application, the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk, the amount of water added, and the amount of feed obtained in the same prediction period are quantized to the same interval according to their respective shrinkage ratios, and the characteristic vector affecting the pelletizing is obtained based on the quantified data and the types and proportions of each component in the mixture, the proportion of the binder, and the original moisture content of the mixture. Then, the characteristic vector affecting the pelletizing is input into the pre-trained proportion prediction model, and the proportion prediction value of raw balls of different specifications is obtained according to the mapping relationship contained in the proportion prediction model. In this way, the proportion prediction value of raw balls of different specifications can be predicted in advance according to the rotation speed, inclination angle of the pelletizing disk, water supply and raw material information of the pelletizing machine obtained in real time before the pelletizing machine completes the pelletizing work. On the one hand, the method provided in the embodiment of the present application does not require manual screening, saves time and effort, and has high detection efficiency. On the other hand, the method can detect the proportion of raw balls in real time, realize real-time control of the pelletizing machine, and better guide actual production operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a structural schematic diagram of a ball making process in the prior art;

[0070] Figure 2 A schematic diagram of the structure of a raw ball quality prediction system for a ball making machine provided in an embodiment of the present application;

[0071] Figure 3A schematic flow chart corresponding to a method for predicting the proportion of raw balls in a ball making machine provided in an embodiment of the present application;

[0072] Figure 4 A flowchart corresponding to a method for generating a share prediction model according to an embodiment of the present application;

[0073] Figure 5 A schematic flow chart corresponding to a method for determining a measured value of a ratio of green balls provided in an embodiment of the present application;

[0074] Figure 6 It is a structural diagram based on the LSTM neural network model;

[0075] Figure 7 A flowchart corresponding to another method for generating a share prediction model provided in an embodiment of the present application;

[0076] Figure 8 A schematic structural diagram of a device for predicting the raw ball ratio of a ball making machine provided in an embodiment of the present application. DETAILED DESCRIPTION

[0077] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0078] Please refer to Figure 1 , Figure 1 It is a structural schematic diagram of a ball-making process in the prior art.

[0079] like Figure 1 As shown, the pelletizing process includes a disc pelletizing machine and a feeding system. The disc pelletizing machine includes a base 11, a central shaft 12, a motor 13, a disc 14, a scraper frame 15, a scraper 16 and a pelletizing belt 17; the feeding system includes a silo 21, a water spray pipe 22 and a feeding belt 23.

[0080] In the above structure, the central axis 12 can adjust the inclination angle of the disc 14, and the adjustment of the speed of the motor 13 can change the speed of the disc 14; the scraper frame 15 is supported on the disc 14, and the scraper 16 is used to promote the movement of the mixture in the disc 14 and prevent adhesion; the mixture in the silo 21 is transported to the disc 14 through the feeding belt 23, and the feeding amount of the mixture can be dynamically adjusted; the water spraying device 22 drops water on the position where the material falls into the disc 14, or drops water on the area where the raw balls grow in the disc 14, and the water volume can be adjusted; after the raw balls come out of the disc 14, they fall into the raw ball belt 17, and the position of the raw ball belt 17 can be regarded as the ball outlet area of ​​the disc ball making machine.

[0081] It should be noted that Figure 1 The disc pelletizing machine shown can be replaced by a cylinder pelletizing machine, and the present application is not limited to this.

[0082] exist Figure 1 On this basis, an embodiment of the present application provides a system for predicting the quality of raw balls of a pelletizing machine. Figure 2 The schematic diagram of the structure of a ball-making machine raw ball quality prediction system provided in the embodiment of the present application. Figure 2 As shown, the system mainly includes:

[0083] The image acquisition device 31 is used to acquire images of the ball output conditions in the ball output area of ​​the ball making machine.

[0084] Among them, in the selection of image acquisition devices, in order to accurately measure the particle size of the ball and suppress motion smear, a 5-megapixel industrial camera and a high-resolution lens are selected, such as the MV-CA013-20GM industrial camera and the MVL_HF2528M-6MP industrial lens, with a measurement accuracy of ±0.1mm.

[0085] The mixed material feeding amount acquisition device 24 is used to obtain the flow rate of the mixed material fed into the pelletizing machine during the current prediction period. Other mixed material information, such as the type and proportion of each component of the mixed material (because the proportion of the binder and the original moisture content of the mixed material have a greater impact on the pelletizing, they are extracted separately from this type of data), the proportion of the binder, and the original moisture content of the mixed material can be obtained from the material ratio data system.

[0086] The water volume acquisition device 25 is used to obtain the water volume provided by the water supply system in the current prediction period.

[0087] The data processing device 41 is respectively connected to the image acquisition device 31, the mixture acquisition device 24 and the water acquisition device 25. On the one hand, it is used to perform image analysis on the images collected by the image acquisition device 31 to obtain the measured values ​​of the proportion of sample raw balls of different specifications; and to train the proportion prediction model according to the characteristic vector of the sample affecting ball making and the measured values ​​of the proportion of sample raw balls of different specifications.

[0088] The data processing device 41, on the other hand, is used to obtain the rotation speed of the pelletizing machine and the inclination angle of the pelletizing disk of the pelletizing machine in the current prediction period, obtain the feed amount and water amount of the pelletizing machine for making raw balls in the current prediction period from the mixture feed amount acquisition device 24 and the water amount acquisition device 25, and obtain the type and proportion of each component in the mixture, the proportion of the binder in the mixture and the original moisture content of the mixture from the material proportioning data system; quantize the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine, the water supply and the feed amount to the same interval according to their respective shrinkage ratios; obtain the characteristic vector affecting the pelletizing according to the quantized rotation speed of the pelletizing machine, the quantized inclination angle of the pelletizing disk of the pelletizing machine, the quantified water supply, the quantified feed amount, the type and proportion of each component in the mixture, the proportion of the binder in the mixture and the original moisture content of the mixture; input the characteristic vector affecting the pelletizing into the proportion prediction model to obtain the proportion prediction value of raw balls of different specifications, and the proportion prediction model includes the mapping relationship between the characteristic vector affecting the pelletizing and the proportion prediction value of raw balls of different specifications.

[0089] It should be noted that the above embodiment only takes the disc pelletizing machine as an example. Of course, the raw ball quality prediction system provided in the embodiment of the present application is also applicable to the cylinder pelletizing machine.

[0090] based on Figure 2 The system architecture shown, Figure 3 The following is a flow chart showing a method for predicting the proportion of raw balls in a ball making machine provided in an embodiment of the present application. Figure 3 As shown, the specific steps include:

[0091] Step 301, obtaining the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine and the water supply amount provided by the water supply system in the pelletizing machine within the current prediction period.

[0092] Step 302, obtaining the current raw material information of the pelletizing machine for making raw pellets within the current prediction period.

[0093] Step 303, quantize the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine, the water supply amount and the material supply amount to the same interval according to their respective contraction ratios.

[0094] Step 304, obtains a characteristic vector affecting ball making according to the quantified rotation speed of the ball making machine, the quantified inclination angle of the ball making disk of the ball making machine, the quantified water supply, the quantified feed amount, the types and proportions of each component in the mixture, the proportion of the binder in the mixture and the original moisture content of the mixture.

[0095] Step 305, input the characteristic vector affecting ball making into the proportion prediction model to obtain the proportion prediction value of raw balls of different specifications.

[0096] In the embodiment of the present application, the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk, the amount of water added, and the amount of feed obtained in the same prediction period are quantized to the same interval according to their respective shrinkage ratios, and the characteristic vector affecting the pelletizing is obtained based on the quantized data and the types and proportions of each component of the mixture, the proportion of the binder, and the original moisture content of the mixture. Then, the characteristic vector affecting the pelletizing is input into a pre-trained proportion prediction model, and the proportion prediction value of raw balls of different specifications is obtained according to the mapping relationship contained in the proportion prediction model. In this way, the proportion prediction value of raw balls of different specifications can be predicted in advance according to the rotation speed, inclination angle of the pelletizing disk, water supply and raw material information of the pelletizing machine obtained in real time before the pelletizing machine completes the pelletizing work. On the one hand, the method provided in the embodiment of the present application does not require manual screening, saves time and effort, and has high detection efficiency. On the other hand, the method can detect the proportion of raw balls in real time, realize real-time control of the pelletizing machine, and better guide actual production operations.

[0097] Specifically, in step 301, during the detection of the rotation speed of the pelletizing machine, if the detected rotation speed is in a stable state, the rotation speed in the stable state is saved and used as subsequent detection data; if the detected rotation speed changes, the saved rotation speed is updated in real time.

[0098] In the embodiment of the present application, the speed regulation of the ball making machine is usually carried out by frequency conversion regulation, and the speed detection can be directly measured from the feedback of the frequency converter. The inclination adjustment of the ball making disk of the ball making machine is usually provided with a hydraulic inclination adjustment device, and the inclination detection is directly measured by an inclination sensor.

[0099] In step 302, the current raw material information includes the feed amount, the types and proportions of each component in the mixture, the proportion of the binder in the mixture, and the original moisture content of the mixture.

[0100] The current raw material information can be obtained specifically from the feeding system located upstream of the pelletizing machine.

[0101] There are many ways to obtain the types and proportions of the components of the mixture, the proportion of the binder and the original moisture content of the mixture. One possible way is to obtain the types and proportions of the components of the mixture, the proportion of the binder and the original moisture content of the mixture once at a predetermined time interval, and use the newly obtained mixture parameters to update the previously obtained mixture parameters.

[0102] Another possible acquisition method is to obtain various parameters of the mixture in real time and determine the change rate of each parameter of the mixture respectively. If the change rate is small and within the preset error range, the previously obtained parameters of the mixture are used. If the change rate is large and exceeds the preset error range, the previously obtained parameters of the mixture are updated with the newly obtained parameters of the mixture. It should be noted that the various parameters of the mixture mentioned here are the types and proportions of the various components of the mixture, the proportion of the binder and the original moisture content of the mixture.

[0103] In step 303 and step 304, due to the different data types and data types of the types and proportions of the components of the mixture, the proportion of the binder, the original moisture content of the mixture, the feed amount, the water supply, the rotation speed of the pelletizer in the pelletizer, and the inclination angle of the pelletizing disk of the pelletizer at the same time, it is impossible to directly calculate these data amounts. In the embodiment of the present application, considering that the proportions of the components of the mixture, the proportions of the binder, and the original moisture content of the mixture are percentages, that is, the values ​​are in the interval (0,1), therefore, the rotation speed of the pelletizer, the inclination angle of the pelletizing disk of the pelletizer, the water supply and the feed amount can also be quantified to the interval (0,1), so that the types and proportions of the components of the mixture, the proportion of the binder, the original moisture content of the mixture, the feed amount, the water supply, the rotation speed of the pelletizer in the pelletizer, and the inclination angle of the pelletizing disk of the pelletizer can be calculated.

[0104] Specifically, when the rotation speed of the pelletizing machine (or the inclination angle of the pelletizing disk of the pelletizing machine) is quantified, the corresponding shrinkage ratio can be the maximum rotation speed of the pelletizing machine (or the maximum inclination angle of the pelletizing disk), which can be referred to formula (1):

[0105]

[0106] Among them, Norm(n) represents the quantized speed of the pelletizing machine, n represents the speed of the pelletizing machine, represents the maximum speed of the pelletizing machine; or, Norm(n) represents the quantized inclination angle of the pelletizing disk of the pelletizing machine, n represents the inclination angle of the pelletizing disk of the pelletizing machine, Indicates the maximum inclination angle of the pelletizing machine.

[0107] When quantifying the water supply, the corresponding contraction ratio can be the maximum water supply of the water supply pipe in the water supply system, which can be referred to formula (2):

[0108]

[0109] Where Norm(n) represents the quantified water supply, s represents the water supply, and s max Indicates the maximum water supply volume of the water supply pipe in the water supply system.

[0110] When quantifying the feed amount of the mixed material, the corresponding shrinkage ratio can be the maximum feed amount of the feed belt, which can be referred to formula (3):

[0111]

[0112] Where Norm(m) represents the quantified mixture feed amount, m represents the actual feed amount, and m max Indicates the maximum feeding amount of the feeding belt.

[0113] After quantification, the characteristic vectors that affect ball making are as follows:

[0114] X(k)=(N,x1(k),x2(k),x3(k),x4(k),x5(k),x6(k),x7(k))=Norm(n,s,m,w,r,p,q)

[0115] Among them, X(k) represents the characteristic vector used to affect ball making, x1(k), x2(k), x3(k), x4(k), x5(k), x6(k), and x7(k) are respectively the quantified rotation speed of the ball making machine, the quantified inclination angle of the ball making disk of the ball making machine, the quantified water supply, the quantified mixture supply, the distribution ratio of each group in the mixture, the proportion of the binder in the mixture and the original moisture content of the mixture; among them, x5(k) contains the type and proportion information of each component. In practical applications, each component can be numbered with a number, and the component number and the component proportion are corresponded in x5(k).

[0116] For example, if the mixture includes three components numbered 1, 2, and 3, then the eigenvector affecting ball formation can be as follows:

[0117]

[0118] Among them, the first 3 indicates that the mixture has three other components besides the binder, x1(k) represents the quantified rotation speed of the pelletizer, x2(k) represents the quantified inclination angle of the pelletizer, x3(k) represents the quantified water supply, and x4(k) represents the quantified feed amount of the mixture. Indicates the proportion of component numbered 1 in the mixture, Indicates the proportion of component numbered 2 in the mixture, represents the proportion of component numbered 3 in the mixture, x6(k) represents the proportion of binder in the mixture, and x7(k) represents the original moisture content of the mixture.

[0119] In step 305, the proportion prediction model includes a mapping relationship between the characteristic vectors affecting ball making and the proportion prediction values ​​of raw balls of different specifications.

[0120] The mapping relationship includes the predicted values ​​of the proportion of raw balls at different time lengths. According to the actual engineering needs, the entire ball making process of the ball making machine can be divided into multiple steps. For example, if a step length is 10 seconds, the proportion prediction model can obtain the following multiple step length proportion prediction values ​​according to the mapping relationship:

[0121] y(k|k), y(k|k+1), y(k|k+2),…, y(k|k+j)

[0122] Among them, j is the prediction step size and k is the specific time of prediction.

[0123] In the embodiment of the present application, the predicted values ​​of the proportion of raw balls of different specifications include the predicted values ​​of the proportion of qualified large balls, the predicted values ​​of the proportion of unqualified large balls, the predicted values ​​of the proportion of qualified medium balls, the predicted values ​​of the proportion of qualified small balls and the predicted values ​​of the proportion of unqualified small balls.

[0124] According to the diameter of the raw balls, the raw balls can be divided into five specifications: qualified large balls, unqualified large balls, qualified medium balls, qualified small balls and unqualified small balls. As shown in Table 1, it is an example of the proportion range of raw balls of different specifications that meet the requirements of the ball making process.

[0125] Among them, raw balls with a diameter greater than or equal to 5mm and less than 8mm are unqualified small balls; raw balls with a diameter greater than or equal to 8mm and less than 11mm are qualified small balls; raw balls with a diameter greater than or equal to 11mm and less than 14mm are qualified medium balls; raw balls with a diameter greater than or equal to 14mm and less than 16mm are qualified large balls; raw balls with a diameter greater than or equal to 16mm are unqualified large balls.

[0126] The setting range for the proportion of unqualified small balls is less than or equal to 7%; the setting range for the proportion of qualified small balls is less than or equal to 30%; the setting range for the proportion of qualified medium balls is less than or equal to 55%; the setting range for the proportion of qualified large balls is less than or equal to 20%; the setting range for the proportion of unqualified large balls is less than or equal to 15%.

[0127] Table 1: An example of different sizes of green balls

[0128] Ball Specifications Unqualified ball Qualified small ball Qualified ball Qualified big ball Unqualified big ball Ball diameter (mm) 5≤d＜8 8≤d＜11 11≤d＜14 14≤d＜16 ≥16 Proportion setting range (%) <![CDATA[0<η1≤7]]> <![CDATA[0<η2≤30]]> <![CDATA[0<η3≤55]]> <![CDATA[0<η4≤20]]> <![CDATA[0<η5≤15]]>

[0129] When the proportion of raw balls of various specifications meets the set range in Table 1 above, and the total sum of the proportions is 100%, it can be considered that the quality of the raw balls is better. On the contrary, it can be considered that the quality of the raw balls is unqualified.

[0130] In the embodiment of the present application, the proportion prediction model can be established based on the long short-term memory neural network prediction model LSTM. Figure 4FIG. 1 is a flow chart of a method for generating a share prediction model provided in an embodiment of the present application, which specifically includes the following steps:

[0131] Step 401, obtaining sample data within N historical prediction periods.

[0132] Among them, the sample data in each historical prediction period include the historical rotation speed of the pelletizing machine, the historical pelletizing disk inclination angle of the pelletizing machine, the historical water supply, the historical raw material information of the pelletizing machine for manufacturing sample raw balls, and the actual measured values ​​of the proportion of sample raw balls of different specifications; the historical raw material information includes the historical feed amount, the types and proportions of each component in the historical mixture, the proportion of binders in the historical mixture and the original moisture content of the mixture in the historical mixture; the actual measured values ​​of the proportions of sample raw balls of different specifications are obtained by collecting images and analyzing and calculating the sample raw balls in each historical prediction period using the particle size analysis method.

[0133] Specifically, there are many ways to obtain the actual measured value of the proportion of sample raw balls of different specifications. For example, the actual measured value of the proportion of sample raw balls of different specifications can be obtained through manual screening methods. For example, the actual measured value of the proportion of sample raw balls of different specifications can be obtained through machine vision methods.

[0134] The following is a detailed introduction to the process of obtaining the measured values ​​of the proportion of sample raw balls of different specifications through machine vision methods. Please refer to Figure 5 , which exemplarily shows a flow chart corresponding to a method for determining a measured value of a ratio of green balls provided in an embodiment of the present application, which specifically includes the following steps:

[0135] Step 501, using an image acquisition device to obtain image information of the ball output area of ​​a ball making machine.

[0136] Step 502, performing image preprocessing on the image information of the ball output area to separate the image information of the sample raw ball and the background image information.

[0137] Step 503: Obtain the central bright spot of the sample green ball according to the image information of the sample green ball.

[0138] Step 504, determining the outline of the sample green ball according to the image information of the sample green ball and the background image information.

[0139] Step 505: Obtain the particle size of the sample green ball according to the central bright spot of the sample green ball and the outline of the sample green ball.

[0140] Step 506, determining the specifications of the sample green balls according to the particle size of the sample green balls and the corresponding relationship between the preset particle size range and the green ball specifications.

[0141] Step 507, counting the total number of sample balls within the historical prediction period, and the number of sample balls of different specifications.

[0142] Step 508, determining the measured values ​​of the proportions of sample raw balls of different specifications based on the total number of sample raw balls and the number of sample raw balls of different specifications in the historical prediction period.

[0143] Step 402, quantize the historical rotation speed of the pelletizing machine, the historical inclination angle of the pelletizing disk of the pelletizing machine, the historical water supply and the historical mixed material supply to the same interval according to their respective contraction ratios.

[0144] It should be noted that the specific quantification method can refer to the description of step 303 above, which will not be repeated here.

[0145] Step 403, obtain the characteristic vectors of N samples affecting ball making according to the quantified historical rotation speed of the ball making machine, the quantified historical ball making disk inclination angle of the ball making machine, the quantified historical water supply, the quantified historical mixture supply, the types and proportions of each component in the historical mixture, the proportion of the binder in the historical mixture and the original moisture content of the mixture in the historical mixture.

[0146] Step 404, taking the feature vectors of N samples influencing ball making as the input of the prediction model, and taking the actual values ​​of the proportion of sample balls of different specifications in N historical prediction periods as the output of the prediction model, and using the time back propagation method to train the proportion prediction model.

[0147] The proportion prediction training module uses the input and output of the training samples to train the LSTM neural network model using the time back propagation method. The time back propagation method is a learning algorithm suitable for multi-layer neural networks. It guides the multi-layer neural network's response (output) to the input to reach a predetermined target range through repeated cycles of excitation propagation and weight update.

[0148] Step 405, continuously updating the weight parameters, bias parameters and learning factors of the share prediction model through iterative training.

[0149] Step 406: If the difference between the predicted value of the proportion of sample raw balls of different specifications by the proportion prediction model and the actual value of the proportion of sample raw balls of different specifications reaches the preset tolerance range, or the proportion prediction model reaches the set maximum number of iterations through iterative operation, the training ends and the last updated weight parameter, bias parameter and learning factor are saved. A dynamic mapping relationship / dynamic prediction model between the feature vector affecting ball making and the predicted value of the proportion of raw balls of different specifications based on the LSTM neural network model is obtained.

[0150] In this process, the forward signal flow at time k (i.e., the output of LSTM at time k) is expressed as formula (4):

[0151]

[0152] In formula (4), Y k-1 is the output at time k-1, X k is the input vector at time k, σ represents the Sigmoid function, W f and b f is the weight vector and bias term of the forget gate, W i and b i is the weight vector and bias term of the input gate, W c and b c is the weight vector and bias term of the unit state, W o and b o is the weight vector and bias term of the output gate, c k For the immediate state, c k-1 The status at the previous moment.

[0153] like Figure 6 As shown, it is a structural diagram based on the LSTM neural network model, wherein the structure of the middle layer mainly consists of a forget gate, an input gate and an output gate, each gate has a corresponding connection weight matrix and a bias term, and the multi-layer neural network is trained through training samples, and the weight parameters, bias parameters and learning factors are continuously updated to obtain the dynamic prediction model.

[0154] During the specific training process of the LSTM neural network model in the embodiment of the present application, the LSTM neural network model divides multiple groups of training samples into two parts, including 2 / 3 groups of training data and 1 / 3 groups of test data, and uses the input of the 2 / 3 groups of training samples and the output of the training samples as the training data of the LSTM neural network model, and continuously updates the weight parameters, bias parameters and learning factors; and uses the input of the 1 / 3 group of training samples and the output of the training samples as the quality test data of the LSTM neural network model to test the prediction quality of the neural network model.

[0155] Taking into account that in the actual process, equipment wear, migration of working conditions, changes in detection points, etc. may cause the proportion prediction model to be inapplicable, in order to improve the applicability and accuracy of the prediction model, an embodiment of the present application also provides a method for online updating of the proportion prediction model.

[0156] Specifically, it is to adjust the model parameters appropriately according to the deviations in the online application prediction process of the training model.

[0157] In the embodiment of the present application, the granularity index of the model is mainly used to determine whether the model needs to be corrected and how it needs to be corrected. Usually, the mean square error between the predicted value and the measured value can be used as the granularity index, and then according to the statistical distribution law of the granularity index, the statistical confidence limit is preset to determine whether an update needs to be triggered and the required update method.

[0158] If the process characteristics are judged to be gradual according to the index analysis results, the model recursion method is selected and the moving window recursion method is used to update the prediction model. The steps are as follows:

[0159] Assume that the sample set of the original prediction model is S = {[X 1 ,Y 1 ],...,[X t ,Y t ]}, t is the total number of samples. When a new measurement value [X m ,Y m ], then add it to the sample set and eliminate the oldest sample, then the new sample set is:

[0160] S={[X 2 ,Y 2 ],...,[X t ,Y t ],[X m ,Y m ]}

[0161] Then use the new samples to train the learning algorithm and obtain a new prediction model.

[0162] If the process characteristics are judged to be mutations based on the indicator analysis results, the just-in-time learning method is selected to select data samples similar to the current measurement status in the sample data within the historical prediction period and reconstruct the prediction model.

[0163] In the embodiment of the present application, the proportion prediction model can also be established based on the proportion prediction table of raw balls of different specifications. Figure 7 FIG. 1 is a flow chart of another method for generating a share prediction model provided in an embodiment of the present application, which specifically includes the following steps:

[0164] S701, obtaining sample data within N historical prediction periods.

[0165] Among them, the sample data in each historical prediction period include the historical speed of the pelletizing machine, the historical water supply, the historical raw material information of the pelletizing machine for manufacturing sample raw balls, and the actual measured values ​​of the proportion of sample raw balls of different specifications; the historical raw material information includes the historical feed amount, the types and proportions of each component in the historical mixture, the proportion of binders in the historical mixture and the original moisture content of the mixture in the historical mixture; the actual measured values ​​of the proportions of sample raw balls of different specifications are calculated after screening the sample raw balls in each historical prediction period using the particle size analysis method.

[0166] S702, statistically analyze N groups of independent mixture parameters, water supply, pelletizing machine state parameters and corresponding measured values ​​of the proportion of sample raw balls of different specifications, and establish a proportion prediction table.

[0167] Among them, the mixture parameters include the historical feed amount, the types and proportions of each component in the historical mixture, the proportion of binder in the historical mixture, and the original moisture content of the mixture in the historical mixture; the pelletizing machine state parameters include the rotation speed of the pelletizing machine and the inclination angle of the pelletizing machine's pelletizing disk.

[0168] Statistical analysis includes data preprocessing. First, the feed amount of the mixture, the water supply, the rotation speed of the pelletizer, the inclination angle of the pelletizer's pelletizing disk, and the shrinkage ratio are quantified to the same range. Then, the quantified data and multiple groups of data with the same mixture parameters are extracted, and the actual measured values ​​of the proportion of raw balls of samples of different specifications corresponding to different groups are compared.

[0169] If the actual measured values ​​of the proportion of sample raw balls of corresponding different specifications are also the same, then only one set of data is retained and the same data in other sets are removed.

[0170] If the actual measured values ​​of the proportion of raw balls of samples of different specifications are different, but the deviation is within the allowable range, the average value of the proportions corresponding to the multiple sets of data will be used as the corresponding actual measured value of the proportions, and only one set of data will be retained. For multiple sets of data with the same mixture parameters, water supply, and pelletizer status parameters, the actual measured values ​​of the proportions of raw balls of samples of different specifications are different, and the deviation exceeds the allowable range, and the sources of the multiple sets of data are marked and retained in the isolation area. The data source here refers to the pelletizer corresponding to the data collection and the collection time. The isolation area refers to a separate area in the proportion prediction table, which is used to record abnormal data as a reference data for equipment maintenance.

[0171] For multiple groups of independent mixture parameters, water supply, ball making machine state parameters and corresponding measured values ​​of the proportion of raw balls of different specifications after data preprocessing, they are sorted according to the set multiple index items, and each index item is sorted from small to large. Among them, at least 6 index items are included, such as the 6 index items set in the embodiment of the present application, and the index order, the first-level index item is the number of components of the mixture (excluding binders), the second-level index item is the proportion of each component in the mixture, wherein the component types of the mixture are pre-numbered, and the component index items are arranged according to the number size, the third-level index item is the quantified feed amount, the fourth-level index item is the proportion of the adhesive in the mixture, the fifth-level index item is the original moisture content of the mixture, the sixth-level index item is the quantified water supply, the seventh-level index item is the quantified speed of the ball making machine, and the eighth-level index item is the quantified inclination angle of the ball making machine. According to the above rules, the establishment of the proportion prediction table is completed, and the proportion prediction table contains the mapping relationship between the characteristic vector affecting ball making and the proportion prediction value of raw balls of different specifications, such as shown in Table 3, which is an example of some data of the proportion prediction table.

[0172] Table 3: An example of partial data in the proportion prediction table

[0173]

[0174] When using the proportion prediction table for prediction, the data in the prediction samples are searched one by one in the proportion prediction table according to the order of the index items in the proportion prediction table until the proportion prediction values ​​of raw balls of different specifications are determined.

[0175] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0176] Figure 8 The following is a schematic diagram showing the structure of a device for predicting the raw ball ratio of a ball making machine provided in an embodiment of the present application. Figure 8 As shown, the device has the function of realizing the method for predicting the raw ball ratio of the ball making machine, and the function can be realized by hardware or by hardware executing corresponding software. The device may include: an acquisition unit 801, a processing unit 802 and a prediction unit 803.

[0177] The acquisition unit 801 is used to acquire the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine and the water supply provided by the water supply system in the current prediction period; and to acquire the current raw material information of the pelletizing machine for making raw balls in the current prediction period, wherein the current raw material information includes the feed amount, the type and proportion of each component in the mixture, the proportion of the binder in the mixture and the original moisture content of the mixture;

[0178] The processing unit 802 is used to quantize the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine, the water supply amount and the feed amount to the same interval according to their respective shrinkage ratios; and obtain a characteristic vector affecting pelletizing according to the quantified rotation speed of the pelletizing machine, the quantified inclination angle of the pelletizing disk of the pelletizing machine, the quantified water supply amount, the quantified feed amount, the types and proportions of each component in the mixture, the proportion of the binder in the mixture and the original moisture content of the mixture;

[0179] The prediction unit 803 is used to input the characteristic vector affecting ball making into the proportion prediction model to obtain the proportion prediction values ​​of raw balls of different specifications. The proportion prediction model includes the mapping relationship between the characteristic vector affecting ball making and the proportion prediction values ​​of raw balls of different specifications.

[0180] Optionally, the step of generating a share prediction model includes:

[0181] Obtain sample data within N historical prediction periods; the sample data within each historical prediction period includes the historical rotation speed of the pelletizing machine, the historical inclination angle of the pelletizing disk of the pelletizing machine, the historical amount of water added, the historical raw material information of the pelletizing machine for manufacturing sample raw balls, and the actual measured values ​​of the proportion of sample raw balls of different specifications; the historical raw material information includes the historical feed amount, the types and proportions of each component in the historical mixture, the proportion of the binder in the historical mixture, and the original moisture content of the mixture in the historical mixture; the actual measured values ​​of the proportion of sample raw balls of different specifications are obtained by using machine vision to collect images of sample raw balls within each historical prediction period, analyze and calculate after processing;

[0182] Quantify the historical rotation speed of the pelletizing machine, the historical inclination angle of the pelletizing disk of the pelletizing machine, the historical water supply amount and the historical feed amount into the same interval according to their respective contraction ratios;

[0183] According to the quantified historical rotation speed of the pelletizing machine, the quantified historical pelletizing disk inclination angle of the pelletizing machine, the quantified historical water supply, the quantified feed amount, the types and proportions of each component of the historical mixture, the proportion of the binder in the historical mixture and the original moisture content of the mixture in the historical mixture, the characteristic vectors of N samples affecting the pelletizing are obtained;

[0184] The characteristic vectors of the N samples influencing ball making are used as the input of the prediction model, and the actual values ​​of the proportions of sample balls of different specifications in N historical prediction periods are used as the output of the prediction model, and the proportion prediction model is trained by using the time back propagation method;

[0185] The weight parameters, bias parameters and learning factors of the share prediction model are continuously updated through iterative training;

[0186] If the difference between the predicted value of the proportion of sample raw balls of different specifications by the proportion prediction model and the actual value of the proportion of sample raw balls of different specifications reaches the preset tolerance range, or the proportion prediction model reaches the set maximum number of iterations during iterative operation, the training ends and the last updated weight parameter, bias parameter and learning factor are saved.

[0187] Optionally, the proportion prediction model is established based on a long short-term memory neural network prediction model LSTM.

[0188] Optionally, the measured values ​​of the proportions of the sample green balls of different specifications are obtained by the following steps:

[0189] An image acquisition device is used to obtain image information of the ball-discharging area of ​​the ball-making machine;

[0190] Performing image preprocessing on the image information of the ball output area to separate the image information of the sample raw ball and the background image information;

[0191] According to the image information of the sample green ball, a central bright spot of the sample green ball is obtained;

[0192] Determine the outline of the sample green ball according to the image information of the sample green ball and the background image information;

[0193] Obtaining the particle size of the sample green ball according to the central bright spot of the sample green ball and the outline of the sample green ball;

[0194] Determining the specifications of the sample green balls according to the particle size of the sample green balls and the corresponding relationship between the preset particle size range and the green ball specifications;

[0195] Count the total number of sample balls within the historical forecast period, as well as the number of sample balls of different specifications;

[0196] According to the total number of sample green balls in the historical prediction period and the number of sample green balls of different specifications, the actual measured values ​​of the proportions of sample green balls of different specifications are determined.

[0197] Optionally, the current raw material information is obtained by the following steps:

[0198] The current raw material information for producing green balls by the pelletizer during the current forecast period is obtained from the feeding system located upstream of the pelletizer.

[0199] Optionally, the shrinkage ratio corresponding to the rotation speed of the pelletizing machine is the maximum rotation speed of the pelletizing machine;

[0200] The contraction ratio corresponding to the water supply volume is the maximum water supply volume of the water supply pipe in the water supply system;

[0201] The shrinkage ratio corresponding to the feeding amount is the maximum feeding amount of the feeding belt;

[0202] The contraction ratio corresponding to the inclination angle of the pelletizing disk of the pelletizing machine is the maximum inclination angle of the pelletizing machine.

[0203] Optionally, the processing unit 802 is further configured to:

[0204] If the difference between the actual measured value of the proportion of raw balls and the predicted value of the model is greater than a preset threshold, the characteristic vector affecting ball making is used as the input of the prediction model, and the actual measured value of the proportion of raw balls and the actual value of the proportion of N sample raw balls are used as the output of the prediction model. The proportion prediction model is updated online to obtain an updated proportion prediction model.

[0205] Optionally, the predicted values ​​of the proportions of raw balls of different specifications include the predicted values ​​of the proportions of qualified large balls, the predicted values ​​of the proportions of unqualified large balls, the predicted values ​​of the proportions of qualified medium balls, the predicted values ​​of the proportions of qualified small balls and the predicted values ​​of the proportions of unqualified small balls.

[0206] In the embodiment of the present application, the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine, the amount of water added, and the amount of feed obtained in the same prediction period are quantized to the same interval according to their respective shrinkage ratios, and the characteristic vector affecting the pelletizing is obtained according to the quantified data and the types and proportions of each component in the mixture, the proportion of the binder, and the original moisture content of the mixture. Then, the characteristic vector affecting the pelletizing is input into the pre-trained proportion prediction model, and the proportion prediction value of raw balls of different specifications is obtained according to the mapping relationship contained in the proportion prediction model. In this way, the proportion prediction value of raw balls of different specifications can be predicted in advance according to the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine, the amount of water added, and the raw material information obtained in real time before the pelletizing machine completes the pelletizing work. On the one hand, the method provided in the embodiment of the present application does not require manual screening, saves time and effort, and has high detection efficiency. On the other hand, the method can detect the proportion of raw balls in real time, realize real-time control of the pelletizing machine, and better guide actual production operations.

[0207] Similar parts between the embodiments provided in this application can be referenced to each other. The specific implementation methods provided above are only a few examples under the general concept of this application and do not constitute a limitation on the protection scope of this application. For those skilled in the art, any other implementation methods expanded based on the scheme of this application without creative work belong to the protection scope of this application.

Claims

1. A method for predicting the raw ball ratio of a ball making machine, characterized in that: The method comprises: Obtain the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine and the water supply provided by the water supply system in the current prediction period; Obtaining current raw material information for making green balls by a ball making machine within a current forecast period, wherein the current raw material information includes feed amount, types and proportions of each component in the mixture, proportion of binder in the mixture, and original moisture content of the mixture; Quantify the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine, the water supply amount and the material supply amount to the same interval according to their respective shrinkage ratios; According to the quantified rotation speed of the pelletizing machine, the quantified inclination angle of the pelletizing disk of the pelletizing machine, the quantified water supply, the quantified feed amount, the type and proportion of each component in the mixture, the proportion of the binder in the mixture and the original moisture content of the mixture, a characteristic vector affecting the pelletizing is obtained; Inputting the characteristic vector affecting ball making into a proportion prediction model to obtain the proportion prediction values ​​of raw balls of different specifications, wherein the proportion prediction model includes a mapping relationship between the characteristic vector affecting ball making and the proportion prediction values ​​of raw balls of different specifications; The step of generating a proportion prediction model includes: Obtain sample data within N historical prediction periods; the sample data within each historical prediction period includes the historical rotation speed of the pelletizing machine, the historical inclination angle of the pelletizing disk of the pelletizing machine, the historical amount of water added, the historical raw material information of the pelletizing machine for manufacturing sample raw balls, and the actual measured values ​​of the proportion of sample raw balls of different specifications; the historical raw material information includes the historical feed amount, the types and proportions of each component in the historical mixture, the proportion of the binder in the historical mixture, and the original moisture content of the mixture in the historical mixture; the actual measured values ​​of the proportion of sample raw balls of different specifications are obtained by using machine vision to collect images of sample raw balls within each historical prediction period, analyze and calculate after processing; Quantify the historical rotation speed of the pelletizing machine, the historical inclination angle of the pelletizing disk of the pelletizing machine, the historical water addition amount and the historical feed amount into the same interval according to their respective shrinkage ratios; According to the quantified historical rotation speed of the pelletizing machine, the quantified historical pelletizing disk inclination angle of the pelletizing machine, the quantified historical water addition amount, the quantified feeding amount, the types and proportions of each component of the historical mixture, the proportion of the binder in the historical mixture and the original moisture content of the mixture in the historical mixture, the characteristic vectors of N samples affecting the pelletizing are obtained; The characteristic vectors of the N samples influencing ball making are used as the input of the prediction model, and the actual values ​​of the proportions of sample balls of different specifications in N historical prediction periods are used as the output of the prediction model, and the proportion prediction model is trained by using the time back propagation method; The weight parameters, bias parameters and learning factors of the share prediction model are continuously updated through iterative training; If the difference between the predicted value of the proportion of sample raw balls of different specifications by the proportion prediction model and the actual value of the proportion of sample raw balls of different specifications reaches the preset tolerance range, or the proportion prediction model reaches the set maximum number of iterations through iterative operation, the training ends and the last updated weight parameter, bias parameter and learning factor are saved; The measured values ​​of the proportions of the sample green balls of different specifications are obtained by the following steps: An image acquisition device is used to obtain image information of the ball-discharging area of ​​the ball-making machine; Performing image preprocessing on the image information of the ball output area to separate the image information of the sample raw ball and the background image information; According to the image information of the sample green ball, a central bright spot of the sample green ball is obtained; Determine the outline of the sample green ball according to the image information of the sample green ball and the background image information; Obtaining the particle size of the sample green ball according to the central bright spot of the sample green ball and the outline of the sample green ball; Determining the specifications of the sample green balls according to the particle size of the sample green balls and the corresponding relationship between the preset particle size range and the green ball specifications; Count the total number of sample balls within the historical forecast period, as well as the number of sample balls of different specifications; According to the total number of sample green balls in the historical prediction period and the number of sample green balls of different specifications, the actual measured values ​​of the proportions of sample green balls of different specifications are determined.

2. The method according to claim 1, characterized in that The proportion prediction model is established based on the long short-term memory neural network prediction model LSTM.

3. The method according to claim 1, characterized in that The current raw material information is obtained by the following steps: The current raw material information for producing green balls by the pelletizer during the current forecast period is obtained from the feeding system located upstream of the pelletizer.

4. The method according to claim 1, characterized in that: The shrinkage ratio corresponding to the speed of the pelletizing machine is the maximum speed of the pelletizing machine; The contraction ratio corresponding to the water supply volume is the maximum water supply volume of the water supply pipe in the water supply system; The shrinkage ratio corresponding to the feeding amount is the maximum feeding amount of the feeding belt; The contraction ratio corresponding to the inclination angle of the pelletizing disk of the pelletizing machine is the maximum inclination angle of the pelletizing machine.

5. The method according to claim 1, characterized in that The method further comprises: If the difference between the actual measured value of the proportion of raw balls and the predicted value of the model is greater than a preset threshold, the characteristic vector affecting ball making is used as the input of the prediction model, and the actual measured value of the proportion of raw balls and the actual value of the proportion of N sample raw balls are used as the output of the prediction model. The proportion prediction model is updated online to obtain an updated proportion prediction model.

6. The method according to any one of claims 1 to 5, characterized in that The predicted values ​​of the proportion of raw balls of different specifications include the predicted value of the proportion of qualified large balls, the predicted value of the proportion of unqualified large balls, the predicted value of the proportion of qualified medium balls, the predicted value of the proportion of qualified small balls and the predicted value of the proportion of unqualified small balls.

7. A device for predicting the proportion of raw balls in a ball making machine, characterized in that: The device comprises: An acquisition unit is used to acquire the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine and the water supply provided by the water supply system in the current prediction period; and to acquire the current raw material information of the pelletizing machine for making raw balls in the current prediction period, wherein the current raw material information includes the feed amount, the type and proportion of each component in the mixture, the proportion of the binder in the mixture and the original moisture content of the mixture; The processing unit is used to quantify the rotation speed of the pelletizing machine, the inclination angle of the pelletizing disk of the pelletizing machine, the water supply amount and the feed amount into the same interval according to their respective shrinkage ratios; and obtain a characteristic vector affecting pelletizing according to the quantified rotation speed of the pelletizing machine, the quantified inclination angle of the pelletizing disk of the pelletizing machine, the quantified water supply amount, the quantified feed amount, the types and proportions of each component in the mixture, the proportion of the binder in the mixture and the original moisture content of the mixture; A prediction unit, used for inputting the feature vector affecting ball making into a proportion prediction model to obtain the proportion prediction values ​​of raw balls of different specifications, wherein the proportion prediction model includes a mapping relationship between the feature vector affecting ball making and the proportion prediction values ​​of raw balls of different specifications; The step of generating a proportion prediction model includes: Obtain sample data within N historical prediction periods; the sample data within each historical prediction period includes the historical rotation speed of the pelletizing machine, the historical inclination angle of the pelletizing disk of the pelletizing machine, the historical amount of water added, the historical raw material information of the pelletizing machine for manufacturing sample raw balls, and the actual measured values ​​of the proportion of sample raw balls of different specifications; the historical raw material information includes the historical feed amount, the types and proportions of each component in the historical mixture, the proportion of the binder in the historical mixture, and the original moisture content of the mixture in the historical mixture; the actual measured values ​​of the proportion of sample raw balls of different specifications are obtained by using machine vision to collect images of sample raw balls within each historical prediction period, analyze and calculate after processing; Quantify the historical rotation speed of the pelletizing machine, the historical inclination angle of the pelletizing disk of the pelletizing machine, the historical water addition amount and the historical feed amount into the same interval according to their respective shrinkage ratios; According to the quantified historical rotation speed of the pelletizing machine, the quantified historical pelletizing disk inclination angle of the pelletizing machine, the quantified historical water addition amount, the quantified feeding amount, the types and proportions of each component of the historical mixture, the proportion of the binder in the historical mixture and the original moisture content of the mixture in the historical mixture, the characteristic vectors of N samples affecting the pelletizing are obtained; The characteristic vectors of the N samples influencing ball making are used as the input of the prediction model, and the actual values ​​of the proportions of sample balls of different specifications in N historical prediction periods are used as the output of the prediction model, and the proportion prediction model is trained by using the time back propagation method; The weight parameters, bias parameters and learning factors of the share prediction model are continuously updated through iterative training; If the difference between the predicted value of the proportion of sample raw balls of different specifications by the proportion prediction model and the actual value of the proportion of sample raw balls of different specifications reaches the preset tolerance range, or the proportion prediction model reaches the set maximum number of iterations through iterative calculation, the training ends and the last updated weight parameter, bias parameter and learning factor are saved.

Images