Model training method, material data processing method, device, equipment and medium

Abstract

The present disclosure provides a training method of a deep learning model, relates to the technical field of artificial intelligence, in particular to the technical field of deep learning and the technical field of industrial big data. The specific implementation scheme is: generating enhanced sample data according to preset cooling knowledge information and initial sample data, wherein the initial sample data includes an initial value of at least one candidate cooling parameter of a sample material, the enhanced sample data includes an enhanced value of the at least one candidate cooling parameter, and the preset cooling knowledge information is used to indicate the relationship between the candidate cooling parameter and the target cooling temperature of the sample material; inputting the enhanced sample data into the deep learning model to obtain an enhanced sample output value of the target cooling temperature; obtaining an enhanced sample loss according to the enhanced sample output value of the target cooling temperature and a label value of the target cooling temperature; and training the deep learning model according to the enhanced sample loss. The present disclosure also provides a material data processing method and device, an electronic device and a storage medium.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and more particularly to the fields of deep learning technology and industrial big data technology. More specifically, this disclosure provides a method for training a deep learning model, a method for processing material data, an apparatus, an electronic device, and a storage medium. Background Technology

[0002] With the development of artificial intelligence technology, the application scenarios of deep learning models are constantly increasing. In the production process of metal materials, cooling can be performed to reduce the minimum product temperature. Accurate temperature control during cooling can improve the material's performance and quality. By using deep learning models to process relevant parameters during the cooling process, the temperature of the material after cooling can be determined, which helps to accurately control the temperature of the metal material during cooling. Summary of the Invention

[0003] This disclosure provides a method for training a deep learning model, a method for processing material data, an apparatus, a device, and a storage medium.

[0004] According to one aspect of this disclosure, a method for training a deep learning model is provided. The method includes: generating augmented sample data based on preset cooling knowledge information and initial sample data, wherein the initial sample data includes initial values ​​of at least one candidate cooling parameter of the sample material, and the augmented sample data includes augmented values ​​of at least one candidate cooling parameter; the preset cooling knowledge information is used to indicate the relationship between the candidate cooling parameter and the target cooling temperature of the sample material; inputting the augmented sample data into a deep learning model to obtain an augmented sample output value of the target cooling temperature; obtaining an augmented sample loss based on the augmented sample output value of the target cooling temperature and the label value of the target cooling temperature; and training the deep learning model based on the augmented sample loss.

[0005] According to one aspect of this disclosure, a material data processing method is provided, the method comprising: inputting target material data of a target material into a deep learning model to obtain a target output cooling temperature, wherein the target material data includes detection values ​​of at least one candidate cooling parameter of the target material, and wherein the deep learning model is trained using the method provided in this disclosure.

[0006] According to another aspect of this disclosure, a training apparatus for a deep learning module is provided. The apparatus includes: a generation module for generating enhanced sample data based on preset cooling knowledge information and initial sample data, wherein the initial sample data includes initial values ​​of at least one candidate cooling parameter of the sample material, and the enhanced sample data includes enhanced values ​​of at least one candidate cooling parameter; the preset cooling knowledge information is used to indicate the relationship between the candidate cooling parameter and the target cooling temperature of the sample material; a first obtaining module for inputting the enhanced sample data into a deep learning model to obtain an enhanced sample output value of the target cooling temperature; a second obtaining module for obtaining an enhanced sample loss based on the enhanced sample output value of the target cooling temperature and the label value of the target cooling temperature; and a training module for training the deep learning model based on the enhanced sample loss.

[0007] According to another aspect of this disclosure, a material data processing apparatus is provided, the apparatus comprising: a third obtaining module for inputting target material data of a target material into a deep learning model to obtain a target output cooling temperature, wherein the target material data includes detection values ​​of at least one candidate cooling parameter of the target material, and wherein the deep learning model is trained using the apparatus provided in this disclosure.

[0008] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a method provided according to this disclosure.

[0009] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions for causing a computer to perform the methods provided according to this disclosure.

[0010] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method provided according to this disclosure.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0013] Figure 1 This is a flowchart of a training method for a deep learning model according to an embodiment of the present disclosure;

[0014] Figure 2 This is a schematic diagram of a training method for a deep learning model according to an embodiment of the present disclosure;

[0015] Figure 3 This is a schematic diagram of a deep learning model according to an embodiment of the present disclosure;

[0016] Figure 4 This is a schematic diagram of a training method for a deep learning model according to an embodiment of the present disclosure;

[0017] Figure 5 This is a flowchart of a material data processing method according to an embodiment of the present disclosure;

[0018] Figure 6 This is a block diagram of a training apparatus for a deep learning model according to an embodiment of the present disclosure;

[0019] Figure 7 This is a block diagram of a material data processing apparatus according to an embodiment of the present disclosure; and

[0020] Figure 8 This is a block diagram of an electronic device according to an embodiment of the present disclosure, which can apply deep learning model training methods and / or material data processing methods. Detailed Implementation

[0021] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0022] Taking steel plates as an example, during water cooling, process engineers can adjust relevant parameters to ensure the final cooling temperature is the expected temperature. For instance, after determining the initial cooling temperature and thickness of the steel plate, engineers can adjust parameters such as the cooling water volume and roller speed based on experience and standards to achieve the preset final cooling temperature. However, adjusting parameters based on experience may result in the final cooling temperature not reaching the expected level, leading to low accuracy and excessive reliance on skilled engineers.

[0023] In some embodiments, processing data of the water-cooling process over a certain period can be acquired as samples to train a deep learning model, resulting in a data-driven material data processing model. This model can determine the final temperature of the steel plate corresponding to different parameters. With sufficient data, the pure data-driven material processing model exhibits high accuracy and strong generalization ability. However, with limited data or data inherently biased, the output of the pure data-driven model may contradict physical or chemical knowledge.

[0024] Figure 1 This is a flowchart of a training method for a deep learning model according to an embodiment of the present disclosure.

[0025] like Figure 1 As shown, the method 100 may include operations S110 to S140.

[0026] In operation S110, enhanced sample data is generated based on preset cooling knowledge information and initial sample data.

[0027] In embodiments of this disclosure, the initial sample data includes initial values ​​for at least one candidate cooling parameter of the sample material. For example, the sample material may include various metals or plastics. Alternatively, the sample material may be a material that needs to be cooled during production or processing. In one example, the sample material may be a steel plate.

[0028] In this embodiment of the disclosure, the candidate cooling parameter can be a parameter related to preset cooling knowledge information. For example, the initial value of the candidate cooling parameter can come from the actual production or processing of the sample material.

[0029] In this embodiment of the disclosure, preset cooling knowledge information can indicate the relationship between candidate cooling parameters and the target cooling temperature of the sample material. For example, the preset cooling knowledge information can be related to the material cooling process. The cooling knowledge information can be at least one of physical knowledge information and chemical knowledge information. For example, taking the preset cooling knowledge information as preset physical knowledge information and the sample material as a steel plate as an example, a preset cooling knowledge information can indicate the following relationship: as the water temperature of the water cooling equipment decreases, the temperature drop when the cooling of the steel plate is terminated increases. The parameter associated with this preset cooling knowledge information can be the water temperature of the water cooling equipment. In one example, the water cooling equipment can be an accelerated cooling control (ACC) device.

[0030] In this embodiment of the disclosure, the enhanced sample data includes an enhanced value for at least one candidate cooling parameter. For example, taking the water temperature of a water-cooled device as an example, the initial value of the water temperature can be 14 degrees Celsius. This initial value can be adjusted to obtain an enhanced value to generate enhanced sample data. An enhanced value for the water temperature of the water-cooled device can be 13 degrees Celsius.

[0031] It is understandable that augmented sample data is obtained by augmenting the initial sample data.

[0032] In operation S120, the enhanced sample data is input into the deep learning model to obtain the enhanced sample output value of the target cooling temperature.

[0033] In embodiments of this disclosure, the deep learning model can be various models. For example, the deep learning model can be implemented as a multilayer perceptron (MLP). As another example, the deep learning model can also be implemented as various neural networks (NNs).

[0034] In operation S130, the enhanced sample loss is obtained based on the enhanced sample output value of the target cooling temperature and the label value of the target cooling temperature.

[0035] In this embodiment of the disclosure, the label value of the target cooling temperature can be derived from the actual production or processing of the sample material. For example, if other parameter values ​​are fixed and the initial water temperature of the water-cooling equipment is 14 degrees Celsius, the actual detected temperature at which the cooling of the steel plate ends can be used as the label value. Alternatively, the difference between the actual detected temperature at which the cooling of the steel plate ends and the actual detected temperature at which the cooling of the steel plate begins can be used as the actual temperature drop value after the cooling of the steel plate ends. This actual temperature drop value can be used as the label value.

[0036] In the embodiments of this disclosure, the augmented sample output value and label value can be processed using any loss function to obtain the augmented sample loss. For example, the L1 loss function can be used to process the augmented sample output value and label value.

[0037] In operation S140, a deep learning model is trained based on the augmented sample loss.

[0038] In this embodiment of the disclosure, the parameters of the deep learning model can be adjusted to make the augmented sample loss converge, thereby training the deep learning model. For example, if the augmented sample loss is less than or equal to a preset augmented sample loss threshold, it can be determined that the augmented sample loss has converged.

[0039] Through the embodiments of this disclosure, data augmentation is performed on the initial sample data based on preset cooling knowledge information, introducing physical or chemical knowledge into the input data of the deep learning model. This makes the output results of the deep learning model more consistent with physical or chemical laws. Furthermore, the deep learning model is trained using augmented sample loss, adding physical or chemical knowledge constraints to the model. This further ensures that the output of the deep learning model aligns with actual production practices, facilitating accurate temperature control during the cooling process of metal materials, and ultimately effectively improving the performance and quality of metal material products.

[0040] It is understood that the overall process of the model training method provided in this disclosure has been described above. The sample materials of this disclosure will be described in detail below with reference to relevant embodiments.

[0041] In some embodiments, the sample material may be at least one of a sample metal material and a sample plastic material.

[0042] In this embodiment of the disclosure, the sample material may be a steel plate.

[0043] For example, the initial sample data for the sample material may include initial values ​​for at least one candidate cooling parameter. The at least one candidate cooling parameter may include at least one of the following: water temperature of the water-cooling equipment, roller speed, water intake of a single spray unit, water output of a single spray unit, and the number of spray units.

[0044] For example, the initial sample data of the sample material may also include initial values ​​for at least one initial cooling parameter, which may include at least one of the following: the target thickness of the steel plate, the temperature of the steel plate at the start of cooling, the preset target temperature at which cooling of the steel plate ends, and the cooling rate. In one example, the preset target temperature at which cooling of the steel plate ends may differ from the actual detected temperature at which cooling of the steel plate ends. It is understood that candidate cooling parameters and initial cooling parameters may be parameters related to the material cooling process.

[0045] For example, the initial sample data can be implemented as (x1, x2, x3, x4, x5, x6, x7, x8, x9). x1 can be the initial value of the target thickness of the steel plate. x2 can be the initial value of the temperature of the steel plate when cooling begins. x3 can be the initial value of the preset target temperature when cooling of the steel plate ends. x4 can be the initial value of the cooling rate. x5 can be the initial value of the water temperature of the water-cooling equipment. x6 can be the initial value of the roller conveyor speed. x7 can be the initial value of the water supply to a single spray unit. x8 can be the initial value of the water discharge to a single spray unit. x9 can be the initial value of the number of spray units.

[0046] It is understood that the sample materials provided in this disclosure have been described above. The following will provide a detailed description of the preset cooling knowledge information of this disclosure in conjunction with relevant embodiments.

[0047] In some embodiments, the preset cooling knowledge information may indicate the following relationship: as the value of the candidate cooling parameter decreases, the decrease in the target cooling temperature increases or decreases.

[0048] In this embodiment of the disclosure, the preset cooling knowledge information may include first preset cooling knowledge information, which may indicate the following relationship: as the value of the candidate cooling parameter decreases, the decrease in the target cooling temperature increases. For example, taking the water temperature of the water-cooling equipment as an example, the first preset cooling knowledge information related to this candidate cooling parameter may indicate the following relationship: as the water temperature of the water-cooling equipment decreases, the temperature decrease after the steel plate stops cooling increases. It can be understood that the temperature at which the steel plate stops cooling can be the target cooling temperature. Subtracting the temperature at which the steel plate stops cooling from the temperature at which the steel plate starts cooling yields the temperature decrease after the steel plate stops cooling.

[0049] In this embodiment of the disclosure, the preset cooling knowledge information may include second preset cooling knowledge information, which may indicate the following relationship: as the value of the candidate cooling parameter decreases, the decrease in the target cooling temperature decreases. For example, taking the water flow rate of a single set of spray devices as an example, the second preset cooling knowledge information related to the candidate cooling parameter may indicate the following relationship: as the water flow rate of a single set of spray devices decreases, the temperature decrease after the steel plate stops cooling decreases.

[0050] It is understood that the above description uses the example of preset cooling knowledge information, including first and second preset cooling knowledge information, for detailed explanation. However, this disclosure is not limited to this; preset cooling knowledge information can also indicate other relationships between candidate cooling parameters and target cooling temperatures. For example, third preset cooling knowledge information can indicate the following relationship: as the value of a candidate cooling parameter decreases, the decrease in target cooling temperature first increases and then decreases.

[0051] It is understood that the above explanation uses the example of preset cooling knowledge information being preset physical information, but this disclosure is not limited to this; preset cooling knowledge information can also be preset chemical information. Candidate cooling parameters related to preset chemical information can be the iron content in the steel plate.

[0052] The following will describe in detail some implementation methods for generating enhanced sample data, based on the first preset cooling knowledge information and related embodiments.

[0053] In some embodiments, when the preset cooling knowledge information is a first preset cooling knowledge information, the enhanced sample data includes at least one of the first enhanced sample data and the second enhanced sample data.

[0054] In some embodiments of operation S110 described above, generating enhanced sample data based on preset cooling knowledge information and initial sample data may include: in response to determining that the preset cooling knowledge information is a first preset cooling knowledge information, generating first enhanced sample data based on the difference information between the preset value and the initial value of the candidate cooling parameter.

[0055] In this embodiment, a first enhanced value for the candidate cooling parameter can be obtained by subtracting a preset value from the initial value. Replacing the initial value of the candidate cooling parameter in the initial sample data with the first enhanced value yields the first enhanced sample data. For example, taking the water temperature of a water-cooled device as the candidate cooling parameter, if the initial value x5 of the water temperature is 14 degrees Celsius and the preset value gap5 is 1 degree Celsius, the first enhanced value of the water temperature can be 13 degrees Celsius. Alternatively, the first enhanced sample data can be implemented as (x1, x2, x3, x4, x5 - gap5, x6, x7, x8, x9).

[0056] In some embodiments of operation S110 described above, generating enhanced sample data based on preset cooling knowledge information and initial sample data may include: in response to determining that the preset cooling knowledge information is first preset cooling knowledge information, generating second enhanced sample data based on the fusion result between the preset value and the initial value of the candidate cooling parameter.

[0057] In this embodiment, adding the initial value of the candidate cooling parameter to a preset value yields a second enhanced value for the candidate cooling parameter. Replacing the initial value of the candidate cooling parameter in the initial sample data with this second enhanced value yields second enhanced sample data. For example, taking the water temperature of a water-cooled device as the candidate cooling parameter, if the initial value x5 of the water temperature is 14 degrees Celsius and the preset value gap5 is 1 degree Celsius, the second enhanced value of the water temperature could be 15 degrees Celsius. Alternatively, the second enhanced sample data could be implemented as (x1, x2, x3, x4, x5 + gap, x6, x7, x8, x9).

[0058] It is understood that the above describes some implementation methods for generating enhanced sample data. The following will describe some implementation methods for obtaining enhanced sample output values ​​of the target cooling temperature in conjunction with relevant embodiments.

[0059] In some embodiments of operation S120 described above, inputting enhanced sample data into a deep learning model to obtain an enhanced sample output value of the target cooling temperature may include: inputting first enhanced sample data into a deep learning model to obtain a first enhanced sample output value of the target cooling temperature.

[0060] For example, by inputting the aforementioned first enhanced sample data (x1, x2, x3, x4, x5-gap5, x6, x7, x8, x9) into a deep learning model, the first enhanced sample output value of the target cooling temperature can be obtained. 15 .

[0061] In some embodiments of operation S120 described above, inputting enhanced sample data into a deep learning model to obtain an enhanced sample output value of the target cooling temperature may include: inputting second enhanced sample data into a deep learning model to obtain a second enhanced sample output value of the target cooling temperature.

[0062] For example, by inputting the aforementioned second-enhanced sample data (x1, x2, x3, x4, x5 + gap5, x6, x7, x8, x9) into a deep learning model, the second-enhanced sample output value of the target cooling temperature can be obtained. 35 .

[0063] As can be understood, the above describes some implementation methods for obtaining enhanced sample output values. The following will describe some implementation methods for obtaining enhanced sample loss in conjunction with relevant embodiments.

[0064] In some embodiments of operation S130 described above, obtaining the enhanced sample loss based on the enhanced sample output value of the target cooling temperature and the label value of the target cooling temperature may include obtaining the first enhanced sample loss based on the label value of the target cooling temperature and the first enhanced sample output value of the target cooling temperature.

[0065] In this embodiment of the disclosure, obtaining the first enhanced sample loss based on the label value of the target cooling temperature and the first enhanced sample output value of the target cooling temperature may include: subtracting the first enhanced sample output value of the target cooling temperature from the label value of the target cooling temperature to obtain a first difference value; determining a first weight based on first preset cooling knowledge information; and processing the first difference value using the first weight to obtain the first enhanced sample loss. For example, the first enhanced sample output value of the target cooling temperature may be subtracted from the label value y of the target cooling temperature. 15 The first difference value is obtained. Furthermore, based on the first preset cooling knowledge information, a first weight w can be determined. 15 For example, the loss L15 for the first augmented sample can be determined using the following formula:

[0066] L15 = w 15 (yo 15 (Formula 1)

[0067] yo 15 This can be used as the first difference value. In one example, during the generation of the first enhanced sample data, first preset cooling knowledge information is introduced. As mentioned above, the first enhanced value of the water temperature of the water-cooled device, x5 - gap5, can be less than the initial value of the water temperature of the water-cooled device, x5. Therefore, based on the first preset cooling knowledge information, the output value of the first enhanced sample can be greater than the label value, and the first difference value can be a value less than 0. To reduce the first difference value, the first weight can be determined to be a value greater than 0.

[0068] In some embodiments of operation S130 described above, obtaining the enhanced sample loss based on the enhanced sample output value of the target cooling temperature and the label value of the target cooling temperature may include obtaining a second enhanced sample loss based on the label value of the target cooling temperature and the second enhanced sample output value of the target cooling temperature.

[0069] In this embodiment of the disclosure, obtaining the second enhanced sample loss based on the label value of the target cooling temperature and the second enhanced sample output value of the target cooling temperature may include: subtracting the label value of the target cooling temperature from the second enhanced sample output value of the target cooling temperature to obtain a second difference value; determining a second weight based on first preset cooling knowledge information; and processing the second difference value using the second weight to obtain the second enhanced sample loss. For example, the second enhanced sample output value of the target cooling temperature can be used... 35 Subtracting the label value y of the target cooling temperature yields the second difference value. Furthermore, based on the first preset cooling knowledge information, a second weight w can be determined. 25 For example, the loss L35 for the second augmented sample can be determined using the following formula:

[0070] L35 = w 25 (o 35 -y) (Formula 2)

[0071] o 35-y can be used as the second difference value. In one example, a first preset cooling knowledge information is introduced during the generation of the second augmented sample data. As mentioned above, the second augmented value x5 + gap5 of the water temperature of the water-cooled device can be greater than the initial value x5 of the water temperature of the water-cooled device. Therefore, based on the first preset cooling knowledge information, the output value of the second augmented sample can be less than the label value, and the second difference value can be a value less than 0. In order to reduce the second difference value, the second weight can be determined to be a value greater than 0. It can be understood that in the embodiments of this disclosure, after obtaining at least one of the first augmented sample loss and the second augmented sample loss, the parameters of the deep learning model can be adjusted to train the deep learning model. Through the embodiments of this disclosure, a first preset cooling knowledge information is introduced during the generation of augmented sample data. Therefore, when determining the augmented sample loss, the positive and negative values ​​of the weights can be quickly determined, which helps to reduce the amount of computation, improve the performance of the model, and reduce resource consumption.

[0072] It is understood that the above describes some implementation methods for obtaining enhanced sample loss. The following will describe the deep learning model of this disclosure in conjunction with relevant embodiments.

[0073] Figure 2 This is a schematic diagram of a deep learning model according to an embodiment of the present disclosure.

[0074] like Figure 2 As shown, the deep learning model 200 can be, for example, a multilayer perceptron. The deep learning model 200 may include an input layer 210, a hidden layer 221, a hidden layer 222, and an output layer 230.

[0075] For example, input layer 210 can receive initial sample data and use it as input to hidden layer 221. Hidden layer 221 processes the initial sample data to obtain initial sample data features. Hidden layer 222 processes the initial sample data features to obtain initial sample output values. Output layer 230 can output the initial sample output values.

[0076] For example, input layer 210 can also receive augmented sample data and use it as input to hidden layer 221. Hidden layer 221 processes the augmented sample data to obtain augmented sample data features. Hidden layer 222 processes the augmented sample data features to obtain augmented sample output values. Output layer 230 can output the augmented sample output values.

[0077] Understandable. Figure 2The deep learning model shown includes two hidden layers. However, this disclosure is not limited to this, and the deep learning model of this disclosure may include a greater number of hidden layers. It is also understood that the deep learning model may be composed of various other neural networks, such as convolutional neural networks (CNNs).

[0078] As can be understood, the deep learning model disclosed herein has been described in detail above. The following will describe in detail some methods for training the deep learning model in conjunction with relevant embodiments.

[0079] In some embodiments, in some implementations of the above-described operation S140, training the deep learning model based on the augmented sample loss may include: inputting initial sample data into the deep learning model to obtain initial sample output values ​​of the target cooling temperature; obtaining an initial sample loss based on the initial sample output values ​​of the target cooling temperature and the label values ​​of the target cooling temperature; and training the deep learning model based on the initial sample loss and the augmented sample loss. The following will combine... Figure 3 Provide detailed explanation Figure 3 This is a schematic diagram of a training method for a deep learning model according to an embodiment of the present disclosure.

[0080] like Figure 3 As shown, based on the initial sample data 301 and the first preset cooling knowledge information, the first enhanced sample data 302 and the second enhanced sample data 303 can be generated.

[0081] Inputting the first enhanced sample data 302 into the deep learning model 300 yields the first enhanced sample output value 3021 for the target cooling temperature. Inputting the initial sample data 301 into the deep learning model 300 yields the initial sample output value 3011 for the target cooling temperature. Inputting the second enhanced sample data 303 into the deep learning model 300 yields the second enhanced sample output value 3031 for the target cooling temperature.

[0082] The first enhanced sample loss can be obtained based on the target cooling temperature label value and the first enhanced sample output value 3021. The initial sample loss can be obtained based on the target cooling temperature label value and the initial sample output value 3011. The second enhanced sample loss can be obtained based on the target cooling temperature label value and the second enhanced sample output value 3031.

[0083] In this embodiment of the disclosure, the total loss can be determined based on the initial sample loss and the augmented sample loss. For example, the first total loss 311 can be determined based on the first augmented sample loss, the initial augmented sample loss, and the second augmented sample loss. For example, the first total loss L can be determined using the following formula. x5 :

[0084] L x5 =(o2-y) 2 +w 15 (yo 15 )+w 25 (o 35 -y) (Formula 3)

[0085] y can be the label value for the target cooling temperature. o2 can be the initial sample output value. 15 This can be the output value for the first augmented sample. 35 This can be used to output the value of the second augmented sample. 15 It can be the first weight. 25 It can be a second weight. In one example, w 15 =w 25 =1.

[0086] In this embodiment of the disclosure, the parameters of the deep learning model are adjusted to bring the total loss to converge, thereby training the deep learning model. Based on the first total loss 311, the parameters of the deep learning model 300 can be adjusted to bring the first total loss 311 to converge.

[0087] It is understood that the method of this disclosure has been described in detail above in conjunction with the first preset cooling knowledge information, but this disclosure is not limited thereto. The method of this disclosure will be described in detail below in conjunction with the second preset cooling knowledge information. As mentioned above, the second preset cooling knowledge information can indicate the following relationship: as the value of the candidate cooling parameter increases, the decrease in the target cooling temperature increases.

[0088] In some embodiments, when the preset cooling knowledge information is a second preset cooling knowledge information, the enhanced sample data includes at least one of a third enhanced sample data and a fourth enhanced sample data.

[0089] In some embodiments, in other implementations of the above-described operation S110, generating enhanced sample data based on preset cooling knowledge information and initial sample data may include: in response to determining that the preset cooling knowledge information is a second preset cooling knowledge information, generating third enhanced sample data based on the fusion result between the preset value and the initial value of the candidate cooling parameter.

[0090] In this embodiment, adding the initial value of the candidate cooling parameter to a preset value yields a third enhanced value for the candidate cooling parameter. Replacing the initial value of the candidate cooling parameter in the initial sample data with this third enhanced value yields third enhanced sample data. For example, taking the water flow rate of a single spray device as the candidate cooling parameter, the preset value gap7 corresponding to the water flow rate x7 of the single spray device can be, for example, 1 cubic meter per hour. Another example is that the third enhanced sample data can be implemented as (x1, x2, x3, x4, x5, x6, x7 + gap7, x8, x9).

[0091] In some embodiments of operation S110 described above, generating enhanced sample data based on preset cooling knowledge information and initial sample data may include: in response to determining that the preset cooling knowledge information is a second preset cooling knowledge information, generating fourth enhanced sample data based on the difference information between the preset value and the initial value of the candidate cooling parameter.

[0092] In this embodiment, a fourth enhanced value for the candidate cooling parameter can be obtained by subtracting a preset value from the initial value of the candidate cooling parameter. Replacing the initial value of the candidate cooling parameter in the initial sample data with the fourth enhanced value yields fourth enhanced sample data. For example, taking the candidate cooling parameter as the water flow rate x7 of a single spray device, the fourth enhanced sample data can be implemented as (x1, x2, x3, x4, x5, x6, x7 - gap7, x8, x9).

[0093] It is understood that the above describes some implementation methods for generating enhanced sample data. The following will describe some implementation methods for obtaining enhanced sample output values ​​of the target cooling temperature in conjunction with relevant embodiments.

[0094] In some embodiments, in other implementations of the above-described operation S120, inputting enhanced sample data into a deep learning model to obtain an enhanced sample output value of the target cooling temperature may include: inputting third enhanced sample data into a deep learning model to obtain a third enhanced sample output value of the target cooling temperature.

[0095] For example, by inputting the aforementioned third-enhanced sample data (x1, x2, x3, x4, x5, x6, x7 + gap7, x8, x9) into a deep learning model, the third-enhanced sample output value of the target cooling temperature can be obtained. 17 .

[0096] In some embodiments, in other implementations of the above-described operation S120, inputting enhanced sample data into a deep learning model to obtain an enhanced sample output value of the target cooling temperature may include: inputting fourth enhanced sample data into a deep learning model to obtain a fourth enhanced sample output value of the target cooling temperature.

[0097] For example, by inputting the aforementioned fourth augmented sample data (x1, x2, x3, x4, x5, x6, x7-gap7, x8, x9) into a deep learning model, the fourth augmented sample output value of the target cooling temperature can be obtained. 37 .

[0098] As can be understood, the above describes some implementation methods for obtaining enhanced sample output values. The following will describe some implementation methods for obtaining enhanced sample loss in conjunction with relevant embodiments.

[0099] In some embodiments, in other implementations of the above-described operation S130, obtaining the enhanced sample loss based on the enhanced sample output value of the target cooling temperature and the label value of the target cooling temperature may include obtaining a third enhanced sample loss based on the label value of the target cooling temperature and the third enhanced sample output value of the target cooling temperature.

[0100] In this embodiment of the disclosure, obtaining the third enhanced sample loss based on the label value of the target cooling temperature and the third enhanced sample output value of the target cooling temperature may include: subtracting the third enhanced sample output value of the target cooling temperature from the label value of the target cooling temperature to obtain a third difference value; determining a third weight based on second preset cooling knowledge information; and processing the third difference value using the third weight to obtain the third enhanced sample loss. For example, the third enhanced sample output value of the target cooling temperature may be subtracted from the label value y of the target cooling temperature. 17 The third difference value is obtained. Furthermore, based on the second preset cooling knowledge information, a third weight w can be determined. 17 For example, the loss L17 for the third augmented sample can be determined using the following formula:

[0101] L17 = w 17 (yo 17 (Formula 4)

[0102] yo 17This can be used as the third difference value. In one example, second preset cooling knowledge information is introduced during the generation of the third enhanced sample data. As mentioned above, the third enhanced value x7 + gap7 of the water supply of a single spray device can be greater than the initial value x7 of the water supply of a single spray device. Therefore, based on the second preset cooling knowledge information, the third enhanced sample output value can be greater than the label value, and the third difference value can be a value less than 0. To reduce the third difference value, the third weight can be determined to be a value greater than 0.

[0103] In some embodiments, in other implementations of the above-described operation S130, obtaining the enhanced sample loss based on the enhanced sample output value of the target cooling temperature and the label value of the target cooling temperature may include: obtaining the fourth enhanced sample loss based on the label value of the target cooling temperature and the fourth enhanced sample output value of the target cooling temperature.

[0104] In this embodiment of the disclosure, obtaining the fourth enhanced sample loss based on the label value of the target cooling temperature and the fourth enhanced sample output value of the target cooling temperature may include: subtracting the label value of the target cooling temperature from the fourth enhanced sample output value of the target cooling temperature to obtain a fourth difference value; determining a fourth weight based on second preset cooling knowledge information; and processing the fourth difference value using the fourth weight to obtain the fourth enhanced sample loss. For example, the fourth enhanced sample output value of the target cooling temperature can be used... 37 Subtracting the label value y of the target cooling temperature yields the fourth difference value. Furthermore, based on the second preset cooling knowledge information, a fourth weight w can be determined. 27 For example, the loss L27 for the fourth augmented sample can be determined using the following formula:

[0105] L27=w 27 (o 37 -y) (Formula 5)

[0106] o 37 -y can be used as the fourth difference value. In one example, a second preset cooling knowledge information is introduced during the generation of the fourth enhanced sample data. As mentioned above, the fourth enhanced value x7-gap7 of the water supply of a single set of spray devices can be less than the initial value x7 of the water supply of a single set of spray devices. Therefore, based on the second preset cooling knowledge information, the output value of the fourth enhanced sample can be less than the label value, and the fourth difference value can be a value less than 0. In order to reduce the fourth difference value, the fourth weight can be determined to be a value greater than 0. Through the embodiments of this disclosure, a second preset cooling knowledge information is introduced during the generation of enhanced sample data. Therefore, when determining the enhanced sample loss, the positive and negative signs of the weights can be quickly determined, which helps to reduce the amount of computation, improve the performance of the model, and reduce resource consumption.

[0107] It is understood that the above description has outlined some other implementation methods for obtaining enhanced sample loss. As mentioned above, the preset cooling knowledge information may include first preset cooling knowledge information and second preset cooling knowledge information. Based on this, some implementation methods for training deep learning models will be described below in conjunction with relevant embodiments.

[0108] Figure 4 This is a schematic diagram of a training method for a deep learning model according to an embodiment of the present disclosure.

[0109] like Figure 4 As shown, based on the initial sample data 401 and the first preset cooling knowledge information, the first enhanced sample data 402 and the second enhanced sample data 403 can be generated. Based on the initial sample data 401 and the second preset cooling knowledge information, the third enhanced sample data 404 and the fourth enhanced sample data 405 can be generated.

[0110] Inputting the first enhanced sample data 402 into the deep learning model 400 yields the first enhanced sample output value 4021 for the target cooling temperature. Inputting the initial sample data 401 into the deep learning model 400 yields the initial sample output value 4011 for the target cooling temperature. Inputting the second enhanced sample data 403 into the deep learning model 400 yields the second enhanced sample output value 4031 for the target cooling temperature.

[0111] The first enhanced sample loss can be obtained based on the label value of the target cooling temperature and the first enhanced sample output value 4021. The initial sample loss can be obtained based on the label value of the target cooling temperature and the initial sample output value 4011. The second enhanced sample loss can be obtained based on the label value of the target cooling temperature and the second enhanced sample output value 4031. Based on the first enhanced sample loss, the initial enhanced sample loss, and the second enhanced sample loss, the first total loss 411 can be determined.

[0112] The third enhanced sample loss can be obtained based on the target cooling temperature label value and the third enhanced sample output value 4041. The initial sample loss can be obtained based on the target cooling temperature label value and the initial sample output value 4011. The fourth enhanced sample loss can be obtained based on the target cooling temperature label value and the fourth enhanced sample output value 4051. The second total loss 412 can be determined based on the third enhanced sample loss, the initial enhanced sample loss, and the fourth enhanced sample loss. For example, the second total loss L can be determined using the following formula. x7 :

[0113] L x7 =(o2-y) 2 +w 17 (yo 17 )+w 27 (o37 -y) (Formula Six)

[0114] y can be the label value for the target cooling temperature. o2 can be the initial sample output value. 17 This can be the output value for the first augmented sample. 37 This can be used to output the value of the second augmented sample. 17 It can be the first weight. 27 It can be a second weight. In one example, w 17 =w 27 =1.

[0115] Next, based on the first total loss 411 and the second total loss 412, the parameters of the deep learning model are adjusted so that the first total loss 411 and the second total loss 412 converge.

[0116] As can be understood, as mentioned above, candidate cooling parameters may also include roller speed, water flow rate of a single spray unit, and number of spray units.

[0117] For example, preset cooling knowledge information related to roller speed can indicate the following relationship: as the roller speed decreases, the temperature drop after the steel plate stops cooling increases. This preset cooling knowledge information can serve as a first preset cooling knowledge information. The method of training a deep learning model based on roller speed is the same as or similar to the method of training a deep learning model based on the water temperature of a water-cooling device, and will not be described in detail here.

[0118] For example, preset cooling knowledge information related to the water flow rate of a single spray unit can indicate the following relationship: as the water flow rate of a single spray unit decreases, the temperature drop of the steel plate after cooling is terminated decreases. As another example, preset cooling knowledge information related to the number of spray units can indicate the following relationship: as the number of spray units decreases, the temperature drop of the steel plate after cooling is terminated decreases. These two preset cooling knowledge pieces can be used as second preset cooling knowledge information. The method of training a deep learning model based on the water flow rate of a single spray unit or the number of spray units is the same as or similar to the method of training a deep learning model based on the water flow rate of a single spray unit, and will not be elaborated here.

[0119] Through the embodiments of this disclosure, compared with a purely data-driven deep learning model, a deep learning model that incorporates pre-defined physical knowledge exhibits stronger generalization ability, helps avoid overfitting caused by small datasets, enhances the reliability of the model output, and can more accurately predict the temperature at which the steel plate stops cooling. For example, based on the same test dataset, the mean square error (MSE) of a purely data-driven deep learning model can be 396.3, while the MSE of a deep learning model that incorporates pre-defined physical knowledge can be 367.7. It is evident that the deep learning model incorporating pre-defined physical knowledge demonstrates stronger generalization ability.

[0120] Figure 5 This is a flowchart of a material data processing method according to another embodiment of the present disclosure.

[0121] like Figure 5 As shown, the method 500 may include operation S510.

[0122] When operating the S510, the target material data is input into the deep learning model to obtain the target output parameters.

[0123] In embodiments of this disclosure, the target material may include at least one of a target metallic material and a target plastic material. For example, the target material may be a steel plate.

[0124] In this embodiment of the disclosure, the target material data includes the detected values ​​of at least one candidate cooling parameter of the target material. For example, when the target material is a steel plate, the target material data may include: the target thickness of the steel plate, the temperature of the steel plate when cooling begins, the preset target temperature when cooling of the steel plate ends, the cooling rate, the water temperature of the water cooling equipment, the roller conveyor speed, the water intake of a single spray device, the water output of a single spray device, and the number of spray devices.

[0125] In this embodiment of the disclosure, the target output parameter can be the temperature at which the target material ceases cooling. It is understood that the temperature at which the steel plate ceases cooling, output by the material data processing model, differs from the preset target temperature at which the steel plate ceases cooling.

[0126] In embodiments of this disclosure, the deep learning model is trained using the methods provided herein. For example, the deep learning model may be trained using method 100.

[0127] Figure 6 This is a block diagram of a training apparatus for a deep learning model according to an embodiment of the present disclosure.

[0128] like Figure 6 As shown, the device 600 may include a generation module 610, a first acquisition module 620, a second acquisition module 630, and a training module 640.

[0129] The generation module 610 is used to generate enhanced sample data based on preset cooling knowledge information and initial sample data. For example, the initial sample data includes the initial value of at least one candidate cooling parameter of the sample material, the enhanced sample data includes the enhanced value of at least one candidate cooling parameter, and the preset cooling knowledge information is used to indicate the relationship between the candidate cooling parameter and the target cooling temperature of the sample material.

[0130] The first acquisition module 620 is used to input the enhanced sample data into the deep learning model to obtain the enhanced sample output value of the target cooling temperature.

[0131] The second acquisition module 630 is used to obtain the enhancement sample loss based on the enhanced sample output value of the target cooling temperature and the label value of the target cooling temperature.

[0132] Training module 640 is used to train deep learning models based on augmented sample loss.

[0133] In some embodiments, the sample material includes at least one of sample metal material and sample plastic material.

[0134] In some embodiments, the preset cooling knowledge information includes first preset cooling knowledge information, which indicates the following relationship: as the value of the candidate cooling parameter decreases, the decrease in the target cooling temperature increases. When the preset cooling knowledge information is the first preset cooling knowledge information, the enhanced sample data includes at least one of first enhanced sample data and second enhanced sample data, where the first enhanced sample data includes a first enhanced value of the candidate cooling parameter, and the second enhanced sample data includes a second enhanced value of the candidate cooling parameter.

[0135] In some embodiments, the generation module includes at least one of the following sub-modules: a first generation sub-module, configured to generate first enhanced sample data based on the difference information between a preset value and the initial value of a candidate cooling parameter in response to determining that the preset cooling knowledge information is the first preset cooling knowledge information; and a second generation sub-module, configured to generate second enhanced sample data based on the fusion result between the preset value and the initial value of the candidate cooling parameter in response to determining that the preset cooling knowledge information is the first preset cooling knowledge information.

[0136] In some embodiments, the first obtaining module includes at least one of the following sub-modules: a first obtaining sub-module, configured to input first enhanced sample data into a deep learning model to obtain a first enhanced sample output value of the target cooling temperature; and a second obtaining sub-module, configured to input second enhanced sample data into a deep learning model to obtain a second enhanced sample output value of the target cooling temperature.

[0137] In some embodiments, the enhanced sample loss includes at least one of a first enhanced sample loss and a second enhanced sample loss, and the second obtaining module includes at least one of the following sub-modules: a third obtaining sub-module, configured to obtain the first enhanced sample loss based on the label value of the target cooling temperature and the first enhanced sample output value of the target cooling temperature; and a fourth obtaining sub-module, configured to obtain the second enhanced sample loss based on the label value of the target cooling temperature and the second enhanced sample output value of the target cooling temperature.

[0138] In some embodiments, the third obtaining submodule includes: a first obtaining unit, configured to subtract a first enhanced sample output value of the target cooling temperature from the label value of the target cooling temperature to obtain a first difference value; a first determining unit, configured to determine a first weight based on first preset cooling knowledge information; and a first processing unit, configured to process the first difference value using the first weight to obtain a first enhanced sample loss.

[0139] In some embodiments, the fourth obtaining submodule includes: a second obtaining unit, configured to subtract the label value of the target cooling temperature from the second enhanced sample output value of the target cooling temperature to obtain a second difference value; a second determining unit, configured to determine a second weight based on first preset cooling knowledge information; and a second processing unit, configured to process the second difference value using the second weight to obtain a second enhanced sample loss.

[0140] In some embodiments, the preset cooling knowledge information includes second preset cooling knowledge information, which indicates the following relationship: as the value of the candidate cooling parameter increases, the decrease in the target cooling temperature increases. When the preset cooling knowledge information is the second preset cooling knowledge information, the enhanced sample data includes at least one of third enhanced sample data and fourth enhanced sample data, where the third enhanced sample data includes a third enhanced value of the candidate cooling parameter, and the fourth enhanced sample data includes a fourth enhanced value of the candidate cooling parameter.

[0141] In some embodiments, the generation module includes at least one of the following sub-modules: a third generation sub-module, configured to generate third enhanced sample data based on the fusion result between the preset value and the initial value of the candidate cooling parameter in response to determining that the preset cooling knowledge information is the second preset cooling knowledge information; and a fourth generation sub-module, configured to generate fourth enhanced sample data based on the difference information between the preset value and the initial value of the candidate cooling parameter in response to determining that the preset cooling knowledge information is the second preset cooling knowledge information.

[0142] In some embodiments, the first obtaining module includes at least one of the following sub-modules: a fifth obtaining sub-module, configured to input the third enhanced sample data into the deep learning model to obtain the third enhanced sample output value of the target cooling temperature; and a sixth obtaining sub-module, configured to input the fourth enhanced sample data into the deep learning model to obtain the fourth enhanced sample output value of the target cooling temperature.

[0143] In some embodiments, the enhanced sample loss includes at least one of a third enhanced sample loss and a fourth enhanced sample loss. The second obtaining module includes at least one of the following sub-modules: a seventh obtaining sub-module, configured to obtain a third enhanced sample loss based on the label value of the target cooling temperature and the third enhanced sample output value of the target cooling temperature; and an eighth obtaining sub-module, configured to obtain a fourth enhanced sample loss based on the label value of the target cooling temperature and the fourth enhanced sample output value of the target cooling temperature.

[0144] In some embodiments, the seventh submodule includes: a third obtaining unit, configured to subtract the third enhanced sample output value of the target cooling temperature from the label value of the target cooling temperature to obtain a third difference value; a third determining unit, configured to determine a third weight based on second preset cooling knowledge information; and a third processing unit, configured to process the third difference value using the third weight to obtain a third enhanced sample loss.

[0145] In some embodiments, the eighth obtaining submodule includes: a fourth obtaining unit, configured to subtract the label value of the target cooling temperature from the fourth enhancement value of the target cooling temperature to obtain a fourth difference value; a fourth determining unit, configured to determine a fourth weight based on second preset cooling knowledge information; and a fourth processing unit, configured to process the fourth difference value using the fourth weight to obtain a fourth enhanced sample loss.

[0146] In some embodiments, the training module includes: a ninth obtaining submodule, used to input initial sample data into the deep learning model to obtain initial sample output values ​​of the target cooling temperature; a tenth obtaining submodule, used to obtain initial sample loss based on the initial sample output values ​​of the target cooling temperature and the label values ​​of the target cooling temperature; and a training submodule, used to train the deep learning model based on the initial sample loss and the augmented sample loss.

[0147] In some embodiments, the training submodule includes: a fifth determining unit, configured to determine the total loss based on the initial sample loss and the augmented sample loss; and an adjusting unit, configured to adjust the parameters of the deep learning model so that the total loss converges.

[0148] In some embodiments, the sample material includes steel plate, and at least one candidate cooling parameter includes at least one of the following: water temperature of the water cooling equipment, roller speed, water intake of a single spray device, water output of a single spray device, and number of spray devices, and the target cooling temperature includes the temperature at which the steel plate is finally cooled.

[0149] In some embodiments, the initial sample data further includes an initial value for at least one initial cooling parameter, which includes at least one of the following: the target thickness of the steel plate, the temperature of the steel plate when cooling begins, the preset target temperature when cooling of the steel plate ends, and the cooling rate.

[0150] Figure 7 This is a block diagram of a material data processing apparatus according to another embodiment of the present disclosure.

[0151] like Figure 7 As shown, the device 700 may include a third acquisition module 710.

[0152] The third acquisition module 710 is used to input the target material data of the target material into the deep learning model to obtain the target output parameters.

[0153] In the embodiments of this disclosure, the deep learning model is trained using the apparatus provided in this disclosure.

[0154] In this embodiment of the disclosure, the target material data includes the detected value of at least one candidate cooling parameter of the target material.

[0155] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0156] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0157] Figure 8 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0158] like Figure 8As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0159] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0160] The computing unit 801 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as deep learning model training methods and / or material data processing methods. For example, in some embodiments, the deep learning model training methods and / or material data processing methods can be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the deep learning model training methods and / or material data processing methods described above can be performed. Alternatively, in other embodiments, the computing unit 801 may be configured by any other suitable means (e.g., by means of firmware) to perform deep learning model training methods and / or material data processing methods.

[0161] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0162] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0163] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0164] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) monitor or an LCD (liquid crystal display)) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0165] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0166] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

[0167] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0168] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for training a deep learning model, comprising: Based on preset cooling knowledge information and initial sample data, enhanced sample data is generated. The initial sample data includes initial values ​​of at least one candidate cooling parameter of the sample material, and the enhanced sample data includes enhanced values ​​of at least one candidate cooling parameter. The preset cooling knowledge information is used to indicate the physical relationship between the candidate cooling parameter and the target cooling temperature of the sample material. The preset cooling knowledge information includes first preset cooling knowledge information, which indicates the following relationship: as the value of the candidate cooling parameter decreases, the decrease in the target cooling temperature increases. The enhanced sample data is input into the deep learning model to obtain the enhanced sample output value of the target cooling temperature; The enhanced sample loss is obtained based on the enhanced sample output value of the target cooling temperature and the label value of the target cooling temperature; The initial sample data is input into the deep learning model to obtain the initial sample output value of the target cooling temperature; Based on the initial sample output value of the target cooling temperature and the label value of the target cooling temperature, the initial sample loss is obtained; and The deep learning model is trained based on the initial sample loss and the augmented sample loss.

2. The method according to claim 1, wherein, The sample material includes at least one of sample metal material and sample plastic material.

3. The method according to claim 1, wherein, When the preset cooling knowledge information is the first preset cooling knowledge information, the enhanced sample data includes at least one of first enhanced sample data and second enhanced sample data. The first enhanced sample data includes a first enhanced value of the candidate cooling parameter, and the second enhanced sample data includes a second enhanced value of the candidate cooling parameter. The process of generating enhanced sample data based on preset cooling knowledge information and initial sample data includes at least one of the following operations: In response to determining that the preset cooling knowledge information is the first preset cooling knowledge information, the first enhanced sample data is generated based on the difference information between the preset value and the initial value of the candidate cooling parameter; In response to determining that the preset cooling knowledge information is the first preset cooling knowledge information, the second enhanced sample data is generated based on the fusion result between the preset value and the initial value of the candidate cooling parameter.

4. The method according to claim 3, wherein, The step of inputting the enhanced sample data into the deep learning model to obtain the enhanced sample output value of the target cooling temperature includes at least one of the following operations: The first enhanced sample data is input into the deep learning model to obtain the first enhanced sample output value of the target cooling temperature; The second enhanced sample data is input into the deep learning model to obtain the second enhanced sample output value of the target cooling temperature.

5. The method according to claim 4, wherein, The augmented sample loss includes at least one of a first augmented sample loss and a second augmented sample loss. The process of obtaining the enhanced sample loss based on the enhanced sample output value of the target cooling temperature and the label value of the target cooling temperature includes at least one of the following operations: The first enhanced sample loss is obtained based on the label value of the target cooling temperature and the first enhanced sample output value of the target cooling temperature; The second enhanced sample loss is obtained based on the label value of the target cooling temperature and the second enhanced sample output value of the target cooling temperature.

6. The method according to claim 5, wherein, The step of obtaining the first enhanced sample loss based on the label value of the target cooling temperature and the first enhanced sample output value of the target cooling temperature includes: The first difference value is obtained by subtracting the first enhanced sample output value of the target cooling temperature from the label value of the target cooling temperature; Based on the first preset cooling knowledge information, a first weight is determined; and The first difference value is processed using the first weight to obtain the first enhanced sample loss.

7. The method according to claim 5, wherein, The step of obtaining the second enhanced sample loss based on the label value of the target cooling temperature and the second enhanced sample output value of the target cooling temperature includes: The second difference value is obtained by subtracting the label value of the target cooling temperature from the second enhanced sample output value of the target cooling temperature; Based on the first preset cooling knowledge information, a second weight is determined; and The second difference value is processed using the second weight to obtain the second enhanced sample loss.

8. The method according to claim 1, wherein, The preset cooling knowledge information includes second preset cooling knowledge information, which indicates the following relationship: as the value of the candidate cooling parameter increases, the decrease in the target cooling temperature increases. When the preset cooling knowledge information is the second preset cooling knowledge information, the enhanced sample data includes at least one of a third enhanced sample data and a fourth enhanced sample data. The third enhanced sample data includes a third enhanced value of the candidate cooling parameter, and the fourth enhanced sample data includes a fourth enhanced value of the candidate cooling parameter. The process of generating enhanced sample data based on preset cooling knowledge information and initial sample data includes at least one of the following operations: In response to determining that the preset cooling knowledge information is the second preset cooling knowledge information, the third enhanced sample data is generated based on the fusion result between the preset value and the initial value of the candidate cooling parameter; In response to determining that the preset cooling knowledge information is the second preset cooling knowledge information, the fourth enhanced sample data is generated based on the difference information between the preset value and the initial value of the candidate cooling parameter.

9. The method according to claim 8, wherein, Inputting augmented sample data into a deep learning model to obtain the augmented sample output value of the target cooling temperature includes at least one of the following operations: The third enhanced sample data is input into the deep learning model to obtain the third enhanced sample output value of the target cooling temperature; The fourth enhanced sample data is input into the deep learning model to obtain the fourth enhanced sample output value of the target cooling temperature.

10. The method according to claim 9, wherein, The augmented sample loss includes at least one of the third augmented sample loss and the fourth augmented sample loss. The process of obtaining the enhanced sample loss based on the enhanced sample output value of the target cooling temperature and the label value of the target cooling temperature includes at least one of the following operations: The third enhanced sample loss is obtained based on the label value of the target cooling temperature and the third enhanced sample output value of the target cooling temperature; The fourth enhanced sample loss is obtained based on the label value of the target cooling temperature and the fourth enhanced sample output value of the target cooling temperature.

11. The method according to claim 10, wherein, The step of obtaining the third enhanced sample loss based on the label value of the target cooling temperature and the third enhanced sample output value of the target cooling temperature includes: The third difference value is obtained by subtracting the third enhanced sample output value of the target cooling temperature from the label value of the target cooling temperature; Based on the second preset cooling knowledge information, a third weight is determined; and The third difference value is processed using the third weight to obtain the third enhanced sample loss.

12. The method according to claim 10, wherein, The step of obtaining the fourth enhanced sample loss based on the label value of the target cooling temperature and the fourth enhanced sample output value of the target cooling temperature includes: The fourth difference value is obtained by subtracting the label value of the target cooling temperature from the fourth enhancement value of the target cooling temperature; Based on the second preset cooling knowledge information, determine the fourth weight; and The fourth difference value is processed using the fourth weight to obtain the fourth enhanced sample loss.

13. The method according to claim 1, wherein, Training the deep learning model based on the initial sample loss and the augmented sample loss includes: The total loss is determined based on the initial sample loss and the augmented sample loss; and Adjust the parameters of the deep learning model so that the total loss converges.

14. The method according to claim 1, wherein, The sample material includes steel plates, and at least one of the candidate cooling parameters includes at least one of the following: water temperature of the water cooling equipment, roller speed, water intake of a single spray device, water output of a single spray device, and number of spray devices. The target cooling temperature includes the temperature of the steel plate at the point of final cooling.

15. The method according to claim 14, wherein, The initial sample data also includes an initial value for at least one initial cooling parameter, wherein the initial cooling parameter includes at least one of the following: the target thickness of the steel plate, the temperature of the steel plate when cooling begins, the preset target temperature of the steel plate when cooling ends, and the cooling rate.

16. A material data processing method, comprising: The target material data is input into a deep learning model to obtain the target output cooling temperature. The target material data includes the detected value of at least one candidate cooling parameter of the target material. The deep learning model is trained using the method described in any one of claims 1 to 15.

17. A training device for a deep learning model, comprising: A generation module is used to generate enhanced sample data based on preset cooling knowledge information and initial sample data. The initial sample data includes initial values ​​of at least one candidate cooling parameter of the sample material, and the enhanced sample data includes enhanced values ​​of at least one candidate cooling parameter. The preset cooling knowledge information indicates the physical relationship between the candidate cooling parameter and the target cooling temperature of the sample material. The preset cooling knowledge information includes first preset cooling knowledge information, which indicates the following relationship: as the value of the candidate cooling parameter decreases, the decrease in the target cooling temperature increases. The first obtaining module is used to input the enhanced sample data into the deep learning model to obtain the enhanced sample output value of the target cooling temperature; The second acquisition module is used to obtain the enhancement sample loss based on the enhanced sample output value of the target cooling temperature and the label value of the target cooling temperature; and A training module is used to train the deep learning model based on the augmented sample loss, including: inputting the initial sample data into the deep learning model to obtain an initial sample output value of the target cooling temperature; obtaining an initial sample loss based on the initial sample output value of the target cooling temperature and the label value of the target cooling temperature; and training the deep learning model based on the initial sample loss and the augmented sample loss.

18. A material data processing apparatus, comprising: The third module is used to input the target material data into the deep learning model to obtain the target output cooling temperature. The target material data includes the detected value of at least one candidate cooling parameter of the target material. The deep learning model is trained using the apparatus of claim 17.

19. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 16.

20. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 16.

21. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 16.

Images