Training method and device for automatic control model of crushing production line, equipment and medium
The regression network model dynamically optimizes crushing production line parameters, addressing PLC inefficiencies by adapting to real-time data for improved efficiency and quality.
Patent Information
- Application Number
- CN202411168841.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-23
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-08-23
AI Technical Summary
Existing Programmable Logic Controllers (PLCs) in crushing production lines require manual parameter setting, lack flexibility, and fail to adapt to changes in production conditions, leading to inefficiency and unstable product quality due to fixed procedures and insufficient use of real-time data.
A training method using a regression network model to predict optimal operating parameters based on historical data, preprocessing, and iterative optimization, enabling dynamic adjustment of equipment settings.
Enhances production efficiency and quality by allowing the automatic control model to adapt to real-time changes, reducing resource waste and ensuring stable operation.
Smart Images

Figure CN119065313A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of automatic control technology, and in particular relates to a training method and device, equipment and storage medium for an automatic control model of a crushing production line. Background Art
[0002] In modern crushing production lines, Programmable Logic Controllers (PLCs) are widely used to control various operational processes from material input to final product output. These control systems can process input signals from various sensors and adjust the operation of mechanical equipment according to preset logic, such as adjusting speed, temperature and pressure.
[0003] However, PLC requires manual setting of machine parameters before production begins. This reliance on manual input is not only time-consuming, but also easily affects production efficiency and product quality due to operator inexperience or improper parameter setting. In addition, PLC programs are usually fixed and lack flexibility, which makes it difficult to effectively respond to changes in production conditions or material characteristics, resulting in inefficiency and waste of resources.
[0004] Although PLC can process real-time data, it usually does not have the ability to learn and optimize the process from real-time data. Therefore, PLC cannot automatically improve or adjust its control strategy from production practice, which limits its ability in production process optimization.
[0005] In modern crushing production lines, although PLC provides basic automated control, its response to changes lacks flexibility and requires manual preset parameters, resulting in low efficiency and unstable production quality. In addition, PLC's fixed procedures and insufficient use of real-time data limit its ability to automatically optimize and adapt to changes in production practices. Summary of the invention
[0006] The purpose of the present invention is to provide a training method and device, equipment, and storage medium for an automatic control model of a crushing production line, so that the trained automatic control model can optimize and adjust the operating parameters of the crushing production line equipment, improve the ability of automatic optimization and adaptation to changes in production practice, and thus improve production efficiency and production quality.
[0007] The first aspect of the present invention discloses a training method for an automatic control model of a crushing production line, comprising:
[0008] Acquire historical operating parameters of a grinder in a crushing production line, and sample characteristic data at multiple sample moments from the historical operating parameters;
[0009] According to the characteristic data at each sample moment, the label data corresponding to each sample moment is obtained, wherein the label data is the real optimal parameter of the target device in the crushing production line, and the target device is any one or more devices in the crushing production line except the grinder;
[0010] Preprocessing the feature data at each sample moment to obtain sample data corresponding to each sample moment; dividing the sample data into a training set and a validation set;
[0011] Input each sample data in the training set into the constructed regression network model to obtain the predicted optimal parameters of the target device corresponding to each sample time output by the regression network model;
[0012] Calculate the loss value between the predicted optimal parameters of the target device corresponding to each sample time and the corresponding label data;
[0013] According to the loss value, the weight parameters of the regression network model are iteratively optimized until the performance index of the regression network model on the validation set reaches a performance threshold, and the training is terminated to obtain an automatic control model; wherein, the automatic control model is used to predict the target operating parameters of the target equipment of the crushing production line according to the real-time operating parameters of the grinder of the crushing production line, and control the target equipment to operate with the target operating parameters.
[0014] In some embodiments, preprocessing the feature data at each sample moment to obtain sample data corresponding to each sample moment includes:
[0015] Calculate the median of the characteristic data at each sample time, and fill in the missing values in the characteristic data at each sample time according to the median;
[0016] Based on the interquartile range method, outliers in the filled feature data are removed;
[0017] The feature data after removing outliers is normalized to obtain sample data corresponding to each sample time.
[0018] In some embodiments, before inputting each sample data in the training set into the constructed regression network model to obtain the predicted optimal parameters of the target device corresponding to each sample time output by the regression network model, the method further includes:
[0019] Load the ResNet50 pre-trained model based on pytorch, and set the number of nodes in the output layer according to the number of parameters of the target equipment of the crushing production line that needs to be predicted;
[0020] Remove the activation function of the last layer in the ResNet50 pre-trained model and readjust it to adapt to the requirements of the regression task to obtain the regression network model.
[0021] In some embodiments, iteratively optimizing the weight parameters of the regression network model according to the loss value includes:
[0022] The Adam optimizer is used to automatically adjust the learning rate and calculate the gradient of the loss value relative to each weight parameter of the regression network model.
[0023] The Adam optimizer is used to update the weight parameters of the regression network model according to the learning rate and gradient.
[0024] In some embodiments, calculating the loss value between the predicted optimal parameter of the target device corresponding to each sample time and the corresponding label data includes:
[0025] The mean square error is used as the loss function to calculate the loss value between the predicted optimal parameters of the target device corresponding to each sample time and the corresponding label data. The loss value is expressed as:
[0026]
[0027] Among them, L represents the loss value, MSE represents the mean square error, and y i is the label data corresponding to the i-th sample moment, is the optimal parameter predicted by the regression network model for the i-th sample data, and n represents the total number of sample data in the training set.
[0028] The second aspect of the present invention discloses a training device for a crushing production line automatic control model, comprising:
[0029] A sampling unit, used to obtain historical operating parameters of a grinder of a crushing production line, and to obtain characteristic data at multiple sample moments from the historical operating parameters;
[0030] A label acquisition unit, used to acquire label data corresponding to each sample moment according to the feature data at each sample moment, wherein the label data is the real optimal parameter of the target device in the crushing production line, and the target device is any one or more devices in the crushing production line except the grinder;
[0031] A sample processing unit, used to pre-process the characteristic data at each sample moment to obtain sample data corresponding to each sample moment;
[0032] A division unit, used for dividing the sample data into a training set and a validation set;
[0033] A prediction unit, used to input each sample data in the training set into the constructed regression network model, and obtain the predicted optimal parameters of the target device corresponding to each sample time output by the regression network model;
[0034] A calculation unit, used to calculate the loss value between the predicted optimal parameter of the target device corresponding to each sample time and the corresponding label data;
[0035] An iterative unit is used to iteratively optimize the weight parameters of the regression network model according to the loss value until the performance index of the regression network model on the validation set reaches a performance threshold, and terminate the training to obtain an automatic control model; wherein the automatic control model is used to predict the target operating parameters of the target equipment of the crushing production line according to the real-time operating parameters of the grinder of the crushing production line, and control the target equipment to operate with the target operating parameters.
[0036] In some embodiments, the sample processing unit comprises:
[0037] The filling subunit is used to calculate the median of the characteristic data at each sample moment, and fill the missing values in the characteristic data at each sample moment according to the median;
[0038] Removing subunits is used to remove outliers in the filled feature data based on the interquartile range method;
[0039] The normalization subunit is used to normalize the feature data after removing the outliers to obtain the sample data corresponding to each sample moment.
[0040] In some embodiments, it also includes:
[0041] A loading unit, used to load a ResNet50 pre-trained model based on pytorch before the prediction unit inputs each sample data in the training set into the constructed regression network model to obtain the predicted optimal parameters of the target device corresponding to each sample moment output by the regression network model;
[0042] A construction unit is used to set the number of nodes in the output layer according to the number of parameters of the target equipment of the crushing production line that needs to be predicted; at the same time, the activation function of the last layer in the ResNet50 pre-trained model is removed and readjusted to adapt to the requirements of the regression task to obtain the regression network model.
[0043] The third aspect of the present invention discloses an electronic device, comprising a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the training method of the crushing production line automatic control model disclosed in the first aspect.
[0044] A fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the training method for the crushing production line automatic control model disclosed in the first aspect.
[0045] The beneficial effects of the present invention are that, by acquiring the historical operating parameters of the grinder of the crushing production line and sampling to obtain feature data at multiple sample moments, and acquiring the label data corresponding to each sample moment, the training set and the validation set are divided; the training set is input into the constructed regression network model to obtain the predicted optimal parameters of the target equipment corresponding to each sample moment output by the regression network model; the loss value between the predicted optimal parameters and the corresponding label data is calculated, and the weight parameters of the regression network model are iteratively optimized until the performance index of the regression network model on the validation set reaches the performance threshold, and the training is terminated to obtain the automatic control model, which is used to predict the target operating parameters of the target equipment of the crushing production line according to the real-time operating parameters of the grinder of the crushing production line, and control the operation of the target equipment. Therefore, the trained automatic control model can dynamically optimize and adjust the operating parameters of the crushing production line equipment based on the grinder parameters, improve the ability of automatic optimization and adaptation to changes in production practice, thereby improving production efficiency and production quality, and ensuring the efficiency and stability of the entire production process. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] The accompanying drawings herein show specific examples of the technical solutions described in the present invention, and together with the specific implementation methods, constitute a part of the specification, and are used to explain the technical solutions, principles and effects of the present invention.
[0047] Unless otherwise specified or defined, the same reference numerals in different drawings represent the same or similar technical features, and the same or similar technical features may also be represented by different reference numerals.
[0048] Figure 1 It is a flow chart of a training method for a crushing production line automatic control model disclosed in an embodiment of the present invention;
[0049] Figure 2 It is a structural schematic diagram of a training device for a crushing production line automatic control model disclosed in an embodiment of the present invention;
[0050] Figure 3 It is a structural schematic diagram of an electronic device disclosed in an embodiment of the present invention.
[0051] Description of reference numerals:
[0052] 201, sampling unit; 202, label acquisition unit; 203, sample processing unit; 204, division unit; 205, prediction unit; 206, calculation unit; 207, iteration unit; 301, memory; 302, processor. DETAILED DESCRIPTION
[0053] Unless otherwise specified or defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art. In the case of combining the technical solution of the present invention with realistic scenarios, all technical and scientific terms used herein may also have meanings corresponding to the purpose of implementing the technical solution of the present invention. The "first, second..." used herein is only used to distinguish the names and does not represent a specific quantity or order. The term "and / or" used herein includes any and all combinations of one or more related listed items.
[0054] It should be noted that when a component is considered to be "fixed to" another component, it can be directly fixed to the other component or there can be a central component; when an component is considered to be "connected to" another component, it can be directly connected to the other component or there can be a central component at the same time; when an component is considered to be "installed on" another component, it can be directly installed on the other component or there can be a central component at the same time. When an component is considered to be "set on" another component, it can be directly set on the other component or there can be a central component at the same time.
[0055] Unless otherwise specified or defined, the "said" and "the" used in this document refer to the technical features or technical contents mentioned or described before the corresponding position, and the technical features or technical contents may be the same as or similar to the technical features or technical contents mentioned therein. In addition, the terms "including" and "having" and any variations thereof used in this document are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.
[0056] The embodiment of the present invention discloses a training method for a crushing production line automatic control model, which can be implemented by computer programming. The execution subject of the method can be an electronic device such as a computer, a laptop, a tablet computer, or a training device for a crushing production line automatic control model embedded in an electronic device, and the present invention is not limited to this. In order to facilitate the understanding of the present invention, the specific embodiments of the present invention will be described in more detail below with reference to the drawings of the specification.
[0057] like Figure 1 As shown, the training method of the crushing production line automatic control model includes the following steps 110 to 160:
[0058] 110. Obtain historical operating parameters of the grinding machine of the crushing production line, and sample characteristic data at multiple sample moments from the historical operating parameters.
[0059] By installing sensors on the grinder in the crushing production line to monitor and record the operating parameters of the grinder in real time, including speed, pressure, temperature, etc., the operating parameters within a specified historical period (such as the past month or the past week) are obtained from the operating parameters of the grinder as historical operating parameters, and then characteristic data at multiple sample times are sampled from the historical operating parameters for subsequent analysis.
[0060] 120. According to the characteristic data at each sample moment, obtain the label data corresponding to each sample moment, where the label data is the real optimal parameter of the target device in the crushing production line, and the target device is any one or more devices in the crushing production line except the grinder.
[0061] Crushing production lines are widely used in a variety of industrial fields, including but not limited to ore processing, crushing of chemical raw materials, preparation of building materials and production of ceramic materials. In addition, similar production lines are also widely used in the pharmaceutical field, especially in the crushing and delivery of drugs.
[0062] In the crushing production line, the material is crushed for the first time by the crusher, and then crushed for the second time by the grinder, and then spray-dried by the cyclone separator and bag dust collector. The dried material enters the sending tank and is transmitted to the buffer warehouse for storage, and then enters the air flow screen for coarse and fine separation through the rotary valve control. The separated coarse material flows back to the crusher for further processing, and the fine material is conveyed to the first screw mixer through the first vacuum conveyor, mixed with other auxiliary materials or pure water and enters the powder silo, and then conveyed to the powder tank through the first screw conveyor, and then to the high-speed mixer through the second vacuum conveyor. At the same time, multiple auxiliary materials (such as liquid A, B, C, D) are mixed under the temperature control of the mold temperature controller and sent to the high-speed mixer together to mix with the fine material, and then the mixture is cooled by the chiller. The cooled mixture enters the second screw mixer, mixed again and sent to the powder tank, and then conveyed to the sending tank through the second screw conveyor. Finally, the mixture is sent out using the third vacuum conveyor to complete the entire production process.
[0063] The working parameters of the grinder directly affect the particle size, uniformity and fluidity of the material. If the parameters of the grinder change, the parameters of the subsequent equipment must be adjusted accordingly to adapt to the changes in the material. If adaptive adjustments are not made, it may lead to reduced material processing efficiency, unstable product quality, equipment blockage or even damage, and ultimately affect the operating efficiency of the entire production line and the product qualification rate.
[0064] Therefore, the target equipment specifically refers to other equipment located after the grinder in the production process of the crushing production line, such as cyclone separator, bag dust collector, air flow screen, first vacuum conveyor, first ribbon mixer, first screw conveyor, second vacuum conveyor, mold temperature controller, high-speed mixer, second ribbon mixer, second screw conveyor, third vacuum conveyor, etc. The actual optimal parameters of the target equipment are empirical values obtained through theoretical and experimental research, which represent the operating parameters of other equipment in the production line when the production efficiency and production quality reach the optimal performance under different grinder parameter conditions.
[0065] For example, feature data x={v g ,v g_avg ,l g ,s g_in ,s g_out ,h g_in ,h g_out ,t g}, where vg is the grinding machine speed, v g_avg is the average speed of the grinder in the first hour, l g is the grinding machine load, s g_in is the feed size of the grinding machine, s g_out is the particle size of the grinding machine, h g_in is the grinding machine feed humidity, h g_out is the humidity of the material discharged from the grinding mill, t g is the grinder temperature.
[0066] Label data y = {v cy ,v bdc ,v si ,v vac1 ,v r1 ,v sc1 ,v vac2 ,t mtc ,v h ,v r2 ,v sc2 ,v vac3}, where v cy is the cyclone separator speed, v bdc is the speed of the bag dust collector, v si is the air flow screen speed, v vac1 is the speed of the first vacuum conveyor, v r1 is the speed of the first ribbon mixer, v sc1 is the speed of the first screw conveyor, v vac2 is the speed of the second vacuum conveyor, t mtc is the temperature of the mold temperature controller, v h is the speed of high-speed mixer, v r2 is the speed of the second ribbon mixer, v sc2is the speed of the second screw conveyor, v vac3 is the third vacuum conveyor speed.
[0067] 130. Preprocess the feature data at each sample moment to obtain sample data corresponding to each sample moment, and divide the sample data into a training set and a validation set.
[0068] The feature data at each sample moment is preprocessed, including filling missing values, removing outliers, normalization, etc., and then divided into training set and validation set according to 8:2.
[0069] The implementation of the preprocessing may specifically include the following steps S11 to S13 which are not shown in the figure:
[0070] S11. Calculate the median of the feature data at each sample time, and fill in the missing values in the feature data at each sample time according to the median.
[0071] S12. Based on the interquartile range method, outliers in the filled feature data are removed.
[0072] First, the interquartile range (IQR) in the feature data is calculated, as shown in the following formula (1):
[0073] IQR=Q3-Q1 (1)
[0074] Among them, Q1 represents the feature value at the 25% position after the feature data of the same feature are sorted from small to large, and Q3 represents the feature value at the 75% position.
[0075] Then, the characteristic values in the characteristic data that are lower than Q1-1.5*IQR or higher than Q3+1.5*IQR are judged as outliers and screened out.
[0076] S13. Normalize the feature data after removing the outliers to obtain sample data corresponding to each sample time.
[0077] Each eigenvalue in the feature data is scaled to between 0 and 1, as shown in the following formula (2):
[0078] X′ i =(X i -min(X)) / (max(X)-min(X)) (2)
[0079] Among them, X i is the i-th eigenvalue in the feature data, X′ i is the normalized value of the i-th eigenvalue, min(X) is the minimum eigenvalue in the feature data, and max(X) is the maximum eigenvalue in the feature data.
[0080] Further optionally, after the training set is divided, data enhancement can be performed on multiple sample data corresponding to multiple sample moments in the training set. By performing data enhancement processing on the sample data of the training set, the generalization ability and robustness of the model can be improved, aiming to simulate various changes that may be encountered in the actual operating environment. Among them, data enhancement processing includes adding noise, scaling, random occlusion, etc.
[0081] First, adding noise refers to adding noise to the sample data of the training set to simulate the error and uncertainty in the actual measurement process. The mathematical expression is shown in the following formula (3):
[0082] x noisy =x+τ,τ~N(0,1)(3)
[0083] Among them, x is the sample data before adding noise, N(0,1) is the standard normal distribution with mean 0 and variance 1, τ is the noise extracted from the standard normal distribution, and x noisy is the sample data after adding noise.
[0084] Secondly, scaling refers to simulating changes in sensor sensitivity or different operating conditions by scaling sample data. The mathematical expression is shown in the following formula (4):
[0085] x scaled =αx (4)
[0086] Among them, α is the scaling factor, x is the sample data before scaling, and x scaled is the scaled sample data.
[0087] Third, random occlusion refers to randomly occluding a portion of sample data to simulate sensor failure or data loss. The mathematical expression is shown in the following formula (5):
[0088]
[0089] Where i is the index of the sample data, M is the randomly selected occluded training set index, x[i] is the value of the original sample data x at index i, and x masked[i] is the value of the sample data at index i after random occlusion.
[0090] 140. Input each sample data in the training set into the constructed regression network model to obtain the predicted optimal parameters of the target device corresponding to each sample time output by the regression network model.
[0091] Before executing step 140, the regression network model can also be adjusted and constructed based on the existing ResNet50 pre-trained model. Exemplarily, the ResNet50 pre-trained model is loaded based on pytorch, and the number of nodes in the output layer is set according to the number of parameters of the target equipment of the crushing production line that needs to be predicted; at the same time, the activation function of the last layer in the ResNet50 pre-trained model is removed, and it is readjusted to adapt to the requirements of the regression task to obtain the (ResNet50 Regression, referred to as ResNet50-R) regression network model. Among them, adapting to the requirements of the regression task means that the output layer directly predicts continuous values.
[0092] Specifically, the last output layer of the ResNet50 pre-trained model is represented by z = σ(F fc (x out ), where F fc represents the fully connected layer, x out represents the feature of the output of the last convolutional layer, and σ represents the activation function. To ensure that the final output of the network is a continuous value, the activation function is removed and readjusted to adapt to the requirements of the regression task, and the ResNet50-R regression network model is obtained. That is, the ResNet50-R regression network model is modified and adjusted on the basis of the ResNet50 pre-trained model. The automatic control model obtained after training can be used to predict the optimal operating parameters of the target equipment of the crushing production line. Among them, the output layer of the ResNet50-R regression network model can be expressed as shown in the following formula (6):
[0093]
[0094] Among them, W, b represent the weight and bias of the fully connected layer, It is represented as the final output result of the regression network model, that is, the predicted optimal parameters.
[0095] 150. Calculate the loss value between the predicted optimal parameters of the target device corresponding to each sample time and the corresponding label data.
[0096] During the training process, the mean absolute error, Huber loss or mean square error (MSE) can be used as a loss function to evaluate the loss value between the predicted optimal parameters of the target device output by the model and the corresponding label data to guide model training. For example, using MSE as a loss function to calculate the loss value between the predicted optimal parameters of the target device corresponding to each sample moment and the corresponding label data, the loss value L can be expressed as shown in the following formula (7):
[0097]
[0098] Among them, L represents the loss value, MSE represents the mean square error, and y i is the label data corresponding to the i-th sample moment, is the optimal prediction parameter output by the regression network model for the ith sample data, which also corresponds to the ith sample time. n represents the total number of sample data in the training set.
[0099] 160. According to the loss value, iteratively optimize the weight parameters of the regression network model until the performance index of the regression network model on the validation set reaches the performance threshold, and terminate the training to obtain the automatic control model.
[0100] Among them, the model can be optimized by using optimizers such as gradient descent, stochastic gradient descent or Adam. For example, the Adam optimizer is used to automatically adjust the learning rate, and the weight parameters of the regression network model are iteratively optimized according to the learning rate and loss value until the performance index of the regression network model on the validation set reaches the performance threshold, and the training is terminated to obtain an automatic control model.
[0101] During the training process, the Adam optimizer is first used to automatically adjust the learning rate and calculate the gradient of the loss value relative to each weight parameter of the regression network model. Then, the Adam optimizer is used to update the weight parameters of the regression network model based on the learning rate and these gradients. The update rule can be shown as follows:
[0102]
[0103] Among them, ∈ is a small number to prevent the denominator from being 0, usually 10 -8 θ t represents the weight parameter of the model after the tth iteration, θ t+1 represents the weight parameter of the model after the t+1th iteration, η represents the learning rate, is the bias-corrected first-order moment estimate (i.e., the exponential moving average of the gradient), expressed as shown in the following formula (9):
[0104]
[0105] Among them, m t =β1m t-1 +(1-β1)g t , g t is the gradient after the tth iteration, β1 is the exponential decay rate of the first moment estimate;
[0106] represents the second-order moment estimate of the gradient (i.e., the exponential moving average of the square of the gradient), which is expressed as shown in the following formula (10):
[0107]
[0108] in, β2 is the exponential decay rate of the second moment estimate.
[0109] Next, a performance threshold thresh is set, and the back-propagation iterative optimization process is continued until the performance index of the regression network model on the validation set reaches the performance threshold or no longer improves significantly. It is determined that the loss function is minimized and the performance of the regression network model is maximized at this time, and the training process is terminated. The regression network model with the last weight parameter update is determined as the automatic control model obtained after the training is completed.
[0110] Among them, the accuracy rate acc can be used as a standard for judging performance, that is, the performance indicator of the regression network model. The calculation formula of acc can be expressed as shown in the following formula (11):
[0111]
[0112] Among them, λ represents the allowable error range between the true value (i.e., the true optimal parameter) and the predicted value (i.e., the predicted optimal parameter). After each training iteration, the accuracy acc is calculated using the sample data on the validation set and compared with the preset performance threshold thresh. If acc ≥ thresh, it is considered that the model has achieved the ideal performance on the validation set, and the training process is terminated.
[0113] The automatic control model obtained after training is used to predict the target operating parameters of the target equipment of the crushing production line according to the real-time operating parameters of the grinder of the crushing production line, and control the target equipment of the crushing production line to operate with the target operating parameters. Specifically, the real-time operating parameters of the grinder of the crushing production line to be predicted (for example, grinder speed, average speed of the grinder in the previous hour, grinder load, grinder feed particle size, grinder discharge particle size, grinder feed humidity, grinder discharge humidity, grinder temperature) can be input into the automatic control model, and the target operating parameters of the target equipment of the crushing production line (for example, cyclone separator speed, bag dust collector speed, air flow screen speed, first vacuum conveyor speed, first ribbon mixer speed, first screw conveyor speed, second vacuum conveyor speed, mold temperature controller temperature, high-speed mixer speed, second ribbon mixer speed, second screw conveyor speed, third vacuum conveyor speed) are obtained according to the model output, and the target equipment of the crushing production line is controlled to operate according to the target operating parameters.
[0114] Therefore, the trained automatic control model can dynamically optimize and adjust the operating parameters of the crushing production line equipment according to the real-time operating parameters of the grinder, improve the ability to automatically optimize and adapt to changes in production practice, thereby improving production efficiency and production quality, and ensuring the efficiency and stability of the entire production process.
[0115] In summary, by implementing the embodiments of the present invention, by integrating advanced sensor technology and deep learning models (ResNet50-R), real-time monitoring and automatic optimization of crushing production line equipment parameters are achieved, thereby improving production efficiency and product quality. Using MSE and Adam optimizers, network weights can be accurately adjusted and optimized to ensure the stability and high performance of the model under various production conditions. In addition, through intelligent data processing and performance monitoring, the present invention significantly reduces resource waste, supports sustainable production goals, and promotes the development of industrial automation and intelligent manufacturing.
[0116] like Figure 2 As shown, the embodiment of the present invention discloses a training device for a crushing production line automatic control model, including a sampling unit 201, a label acquisition unit 202, a sample processing unit 203, a division unit 204, a prediction unit 205, a calculation unit 206, and an iteration unit 207, wherein:
[0117] The sampling unit 201 is used to obtain the historical operating parameters of the grinding machine of the crushing production line, and obtain characteristic data at multiple sample moments from the historical operating parameters;
[0118] The label acquisition unit 202 is used to acquire label data corresponding to each sample moment according to the feature data at each sample moment, where the label data is the real optimal parameter of the target device in the crushing production line, and the target device is any one or more devices in the crushing production line except the grinder;
[0119] The sample processing unit 203 is used to pre-process the feature data at each sample moment to obtain sample data corresponding to each sample moment;
[0120] A division unit 204 is used to divide the sample data into a training set and a validation set;
[0121] The prediction unit 205 is used to input each sample data in the training set into the constructed regression network model to obtain the predicted optimal parameters of the target device corresponding to each sample time output by the regression network model;
[0122] A calculation unit 206, used to calculate the loss value between the predicted optimal parameter of the target device corresponding to each sample time and the corresponding label data;
[0123] The iteration unit 207 is used to iteratively optimize the weight parameters of the regression network model according to the loss value until the performance index of the regression network model on the validation set reaches the performance threshold, and terminate the training to obtain the automatic control model; wherein the automatic control model is used to predict the target operating parameters of the target equipment of the crushing production line according to the real-time operating parameters of the grinder of the crushing production line, and control the target equipment to operate with the target operating parameters.
[0124] As an optional implementation, the sample processing unit 203 includes the following subunits (not shown):
[0125] The filling subunit is used to calculate the median of the characteristic data at each sample moment, and fill the missing values in the characteristic data at each sample moment according to the median;
[0126] Removing subunits is used to remove outliers in the filled feature data based on the interquartile range method;
[0127] The normalization subunit is used to normalize the feature data after removing the outliers to obtain the sample data corresponding to each sample moment.
[0128] Optionally, the above-mentioned training device for the crushing production line automatic control model further includes the following units not shown in the figure:
[0129] A loading unit, used to load a ResNet50 pre-trained model based on pytorch before the prediction unit 205 inputs each sample data in the training set into the constructed regression network model to obtain the predicted optimal parameters of the target device corresponding to each sample moment output by the regression network model;
[0130] A construction unit is used to set the number of nodes in the output layer according to the number of parameters of the target equipment of the crushing production line that needs to be predicted; at the same time, the activation function of the last layer in the ResNet50 pre-trained model is removed and readjusted to adapt to the requirements of the regression task to obtain the regression network model.
[0131] Optionally, the above-mentioned iteration unit 207 is specifically used to automatically adjust the learning rate using the Adam optimizer, and calculate the gradient of the loss value relative to each weight parameter of the regression network model, and use the Adam optimizer to update the weight parameters of the regression network model according to the learning rate and the gradient until the performance indicator of the regression network model on the validation set reaches the performance threshold, and the training is terminated to obtain the automatic control model.
[0132] like Figure 3 As shown, an embodiment of the present invention discloses an electronic device, including a memory 301 storing executable program codes and a processor 302 coupled to the memory 301;
[0133] The processor 302 calls the executable program code stored in the memory 301 to execute the training method of the crushing production line automatic control model described in the above embodiments.
[0134] An embodiment of the present invention further discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the training method of the crushing production line automatic control model described in the above embodiments.
[0135] The purpose of the above embodiments is to exemplarily reproduce and deduce the technical solution of the present invention, and to fully describe the technical solution, purpose and effect of the present invention. Its purpose is to make the public understand the disclosed content of the present invention more thoroughly and comprehensively, and it does not limit the scope of protection of the present invention.
[0136] The above embodiments are not exhaustive enumerations of the present invention, and there may be multiple other implementations not listed. Any replacement and improvement made without violating the concept of the present invention shall fall within the protection scope of the present invention.
Claims
1. A training method for an automatic control model of a crushing production line, characterized in that: include: Acquire historical operating parameters of a grinder in a crushing production line, and sample characteristic data at multiple sample moments from the historical operating parameters; According to the characteristic data at each sample moment, the label data corresponding to each sample moment is obtained, wherein the label data is the real optimal parameter of the target device in the crushing production line, and the target device is any one or more devices in the crushing production line except the grinder; Preprocessing the feature data at each sample moment to obtain sample data corresponding to each sample moment; dividing the sample data into a training set and a validation set; Input each sample data in the training set into the constructed regression network model to obtain the predicted optimal parameters of the target device corresponding to each sample time output by the regression network model; Calculate the loss value between the predicted optimal parameters of the target device corresponding to each sample time and the corresponding label data; According to the loss value, the weight parameters of the regression network model are iteratively optimized until the performance index of the regression network model on the validation set reaches a performance threshold, and the training is terminated to obtain an automatic control model; wherein, the automatic control model is used to predict the target operating parameters of the target equipment of the crushing production line according to the real-time operating parameters of the grinder of the crushing production line, and control the target equipment to operate with the target operating parameters.
2. The training method for the crushing production line automatic control model according to claim 1, characterized in that: Preprocess the feature data at each sample moment to obtain sample data corresponding to each sample moment, including: Calculate the median of the characteristic data at each sample time, and fill in the missing values in the characteristic data at each sample time according to the median; Based on the interquartile range method, outliers in the filled feature data are removed; The feature data after removing outliers is normalized to obtain sample data corresponding to each sample time.
3. The training method of the crushing production line automatic control model according to claim 1, characterized in that: Before inputting each sample data in the training set into the constructed regression network model to obtain the predicted optimal parameters of the target device corresponding to each sample time output by the regression network model, the following steps are also included: Load the ResNet50 pre-trained model based on pytorch, and set the number of nodes in the output layer according to the number of parameters of the target equipment of the crushing production line that needs to be predicted; Remove the activation function of the last layer in the ResNet50 pre-trained model and readjust it to adapt to the requirements of the regression task to obtain the regression network model.
4. The training method for the crushing production line automatic control model according to claim 1, characterized in that: According to the loss value, iteratively optimizing the weight parameters of the regression network model includes: The Adam optimizer is used to automatically adjust the learning rate and calculate the gradient of the loss value relative to each weight parameter of the regression network model. The Adam optimizer is used to update the weight parameters of the regression network model according to the learning rate and gradient.
5. The training method for the crushing production line automatic control model according to any one of claims 1 to 4, characterized in that: Calculate the loss value between the predicted optimal parameters of the target device corresponding to each sample time and the corresponding label data, including: The mean square error is used as the loss function to calculate the loss value between the predicted optimal parameters of the target device corresponding to each sample time and the corresponding label data. The loss value is expressed as: Among them, L represents the loss value, MSE represents the mean square error, and y i is the label data corresponding to the i-th sample moment, is the optimal parameter predicted by the regression network model for the i-th sample data, and n represents the total number of sample data in the training set.
6. A training device for an automatic control model of a crushing production line, characterized in that: include: A sampling unit, used to obtain historical operating parameters of a grinder of a crushing production line, and to obtain characteristic data at multiple sample moments from the historical operating parameters; A label acquisition unit, used to acquire label data corresponding to each sample moment according to the feature data at each sample moment, wherein the label data is the real optimal parameter of the target device in the crushing production line, and the target device is any one or more devices in the crushing production line except the grinder; A sample processing unit, used to pre-process the characteristic data at each sample moment to obtain sample data corresponding to each sample moment; A division unit, used for dividing the sample data into a training set and a validation set; A prediction unit, used to input each sample data in the training set into the constructed regression network model, and obtain the predicted optimal parameters of the target device corresponding to each sample time output by the regression network model; A calculation unit, used to calculate the loss value between the predicted optimal parameter of the target device corresponding to each sample time and the corresponding label data; An iterative unit is used to iteratively optimize the weight parameters of the regression network model according to the loss value until the performance index of the regression network model on the validation set reaches a performance threshold, and terminate the training to obtain an automatic control model; wherein the automatic control model is used to predict the target operating parameters of the target equipment of the crushing production line according to the real-time operating parameters of the grinder of the crushing production line, and control the target equipment to operate with the target operating parameters.
7. The training device for the crushing production line automatic control model according to claim 6, characterized in that: The sample processing unit comprises: The filling subunit is used to calculate the median of the characteristic data at each sample moment, and fill the missing values in the characteristic data at each sample moment according to the median; Removing subunits is used to remove outliers in the filled feature data based on the interquartile range method; The normalization subunit is used to normalize the feature data after removing the outliers to obtain the sample data corresponding to each sample moment.
8. The training device for the crushing production line automatic control model according to claim 6, characterized in that: Also includes: A loading unit, used to load a ResNet50 pre-trained model based on pytorch before the prediction unit inputs each sample data in the training set into the constructed regression network model to obtain the predicted optimal parameters of the target device corresponding to each sample moment output by the regression network model; A construction unit is used to set the number of nodes in the output layer according to the number of parameters of the target equipment of the crushing production line that needs to be predicted; at the same time, the activation function of the last layer in the ResNet50 pre-trained model is removed and readjusted to adapt to the requirements of the regression task to obtain the regression network model.
9. An electronic device, characterized in that: It comprises a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the training method of the crushing production line automatic control model according to any one of claims 1 to 5.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program enables a computer to execute the training method for the crushing line automatic control model according to any one of claims 1 to 5.
Citation Information
Patent Citations
Prediction method for metal stamping forming process parameters
CN117743793A
Integrated learning semi-supervised industrial data prediction method based on pseudo label optimization
CN118133238A
Intelligent ore grinding control method and device, medium and equipment
CN118417046A
