Lightweight prediction method for dynamic normal stress of silo based on derivative distillation
By adopting a lightweight prediction method based on derivative distillation in the silo dynamic normal stress prediction model, the problems of insufficient consideration and high complexity of existing model factors are solved, and more efficient and accurate dynamic stress prediction is achieved, meeting the real-time prediction needs.
Patent Information
- Application Number
- CN202411906291.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-23
- Publication Date
- 2025-04-11
AI Technical Summary
The existing silo dynamic normal stress prediction model has limited considerations, and it is difficult to accurately predict complex dynamic stress changes during unloading. The model is highly complex and difficult to meet the real-time prediction needs.
Using a lightweight prediction method based on derivative distillation, a dynamic normal stress prediction model is constructed by acquiring and preprocessing the silo sample information and basic information, a dynamic normal stress prediction model is constructed, and a Spearman correlation coefficient is used to analyze feature correlations, redundant features are eliminated, and a derivative matrix is constructed to improve the model expression ability, and a lightweight network is trained through the distillation technology of deep learning models.
It significantly improves the prediction performance of the model, reduces the complexity of the model, improves generalization ability and prediction accuracy, and can be effectively used in engineering applications with high real-time performance.
Smart Images
Figure CN119808572A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present application relate to the field of building silo structure design, and specifically to a lightweight prediction method for dynamic normal stress of a silo based on derivative distillation. Background Art
[0002] As a structure for storing and transporting granular materials, silos have been widely used in many fields. However, due to the characteristics of granular materials in silos, various flow states (such as overall flow, mixed flow and funnel flow) will occur during the unloading process. The dynamic normal stress generated is an important cause of structural instability and failure. Since the material properties and geometric structural parameters of the silo are directly related to its dynamic normal stress. Most of the existing studies consider fewer influencing factors, while there are many factors that affect the dynamic normal stress of silos and there are very complex nonlinear relationships between them. Therefore, it is particularly important to establish an accurate prediction model that considers multiple influencing factors.
[0003] As machine learning and deep learning are widely used in civil engineering, more and more models are used to predict parameters. However, when dealing with complex nonlinear relationships, the model pays less attention to the efficient use of output information and information extraction. In addition, in order to improve the prediction accuracy of the model, the complexity of the model will increase, resulting in a longer prediction time. Therefore, it is very important to maintain the prediction accuracy of the model while achieving lightweight.
[0004] However, it is found in practice that when the above method is used for silo structure design, the following technical problems often occur:
[0005] First, the existing silo dynamic normal stress prediction models usually only consider limited factors. However, in actual operation, different flow states during the unloading process (such as bulk flow, mixed flow and funnel flow) will produce complex dynamic stress changes, and these factors have not been fully considered;
[0006] Second, the dynamic normal stress inside the silo is the result of the combined effect of multiple factors, which have complex nonlinear relationships with each other. Existing models often fail to accurately predict the changes in dynamic stress when dealing with these complex relationships.
[0007] Third, existing models do not pay enough attention to the efficient use of output information and feature extraction when dealing with complex nonlinear relationships;
[0008] Fourth, in actual engineering applications, a prediction model with higher real-time performance is required, which makes it difficult for highly complex models to meet engineering needs. Summary of the invention
[0009] The content of this application is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this application is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.
[0010] Some embodiments of the present application propose a lightweight prediction method, a computer device and a computer-readable storage medium for the dynamic normal stress of a silo based on derivative distillation to solve one or more of the technical problems mentioned in the above background technology section.
[0011] In a first aspect, some embodiments of the present application provide a lightweight prediction method for silo dynamic normal stress based on derivative distillation, the method comprising: obtaining a silo sample information set and silo basic information; preprocessing each silo sample information in the above silo sample information set to generate silo sample enhancement information, and obtaining a silo sample enhancement information set; preprocessing the above silo basic information to obtain silo basic enhancement information; constructing a silo dynamic normal stress prediction model based on the above silo sample enhancement information set; inputting the above silo basic enhancement information into the above silo dynamic normal stress prediction model to obtain a silo dynamic normal stress prediction value; in response to determining that the above silo dynamic normal stress prediction value is greater than a preset alarm threshold, sending the above silo dynamic normal stress prediction value to an alarm terminal for alarm processing.
[0012] In a second aspect, the present application also provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the method described in any implementation of the first aspect is implemented.
[0013] In a third aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored, wherein when the computer program is executed by a processor, the method described in any implementation of the first aspect is implemented.
[0014] The above-mentioned embodiments of the present application have the following beneficial effects: The present invention systematically collects and organizes the simulation and experimental data related to the silo for the first time, and establishes a comprehensive and structured silo database. This database not only contains rich feature information, but also lays a solid foundation for subsequent statistical analysis and model training. By adopting the Spearman correlation coefficient, the correlation analysis of various features of the factors affecting the dynamic normal stress of the silo is performed. The correlation coefficient between each feature is calculated, and the strongly correlated features are eliminated. The database after removing redundant features significantly improves the prediction performance of the model. This not only reduces the complexity of the model, but also improves the generalization ability and prediction accuracy of the model. Through the derivative module, the present invention uses the features output by the teacher model and the student model to construct a derivative matrix. The matrix can quantify and fuse the relationship between different features, thereby improving the expression ability of the model. The derivative module realizes efficient use of features by extracting and fusing the diagonal and off-diagonal elements of the derivative matrix, and further improves the prediction performance of the model. The outputs of different levels of the deep learning model are used, and the correlation between these outputs is captured through the derivative matrix, thereby improving the expression ability and generalization ability of the model. A derivative distillation model was built to train a complex network model and compare its output with the true label to train a smaller lightweight network. By building a combination of student, teacher channels and derivative modules, a lightweight design of the model was achieved, reducing the number of parameters and storage space, while improving prediction performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.
[0016] Figure 1 is a flow chart of some embodiments of a lightweight prediction method for dynamic normal stress of a silo based on derivative distillation according to the present application;
[0017] Figure 2 is a flowchart of an experiment for constructing a derivative distillation model according to some embodiments of the lightweight prediction method of the dynamic normal stress of a silo based on derivative distillation of the present application;
[0018] Figure 3 is a structural schematic diagram of a derivative distillation model according to some embodiments of a lightweight prediction method for dynamic normal stress of a silo based on derivative distillation of the present application;
[0019] Figure 4 It is a structural schematic diagram of a first prediction model and a first channel model of a lightweight prediction method of a silo dynamic normal stress based on derivative distillation according to the present application;
[0020] Figure 5 is a structural schematic diagram of a second prediction model and a second channel model of a lightweight prediction method for dynamic normal stress of a silo based on derivative distillation according to the present application;
[0021] Figure 6 It is a schematic diagram of the structure of a computer device suitable for implementing some embodiments of the present application. DETAILED DESCRIPTION
[0022] Embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes and are not intended to limit the scope of protection of the present application.
[0023] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other.
[0024] It should be noted that the concepts such as "first" and "second" mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.
[0025] It should be noted that the modifications of "one" and "plurality" mentioned in the present application are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".
[0026] The names of the messages or information exchanged between multiple devices in the embodiments of the present application are only used for illustrative purposes and are not used to limit the scope of these messages or information.
[0027] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments.
[0028] Figure 1 The flowchart 100 of some embodiments of the lightweight prediction method of the dynamic normal stress of the silo based on the derivative distillation according to the present application is shown. The lightweight prediction method of the dynamic normal stress of the silo based on the derivative distillation comprises the following steps:
[0029] Step 101, obtaining a silo sample information set and silo basic information.
[0030] In some embodiments, the execution subject of the lightweight prediction method of the silo dynamic normal stress based on derivative distillation can obtain the silo sample information set and the silo basic information from the storage terminal by wired connection or wireless connection. Among them, the above-mentioned storage terminal can be a terminal for storing the silo sample information set and the silo basic information. The silo sample information in the above-mentioned silo sample information set can include: a silo feature information set and a silo dynamic normal stress value. The silo feature information in the above-mentioned silo feature information set can be but not limited to at least one of the following: silo height, silo diameter, height-to-diameter ratio, funnel angle, discharge port size, storage external friction angle coefficient, storage internal friction angle coefficient, storage density, stress position height from the silo top and the ratio of the stress position to the silo height. The above-mentioned silo dynamic normal stress value can be the silo normal stress value actually measured. The above-mentioned silo basic information can be the basic information of the silo that needs to be predicted for dynamic normal stress. The above-mentioned silo basic information can include: a silo basic feature information set. The basic characteristic information of the silo in the above-mentioned basic characteristic information set of the silo may be but is not limited to at least one of the following: silo height, silo diameter, height-to-diameter ratio, funnel angle, discharge port size, storage material external friction angle coefficient, storage material internal friction angle coefficient, storage material density, height of the stress position from the silo top, and the ratio of the stress position to the silo height.
[0031] It should be noted that the above-mentioned wireless connection methods may include but are not limited to 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future.
[0032] Step 102 : pre-process each silo sample information in the silo sample information set to generate silo sample enhancement information, thereby obtaining a silo sample enhancement information set.
[0033] In some embodiments, the execution entity may pre-process each silo sample information in the silo sample information set to generate silo sample enhancement information to obtain a silo sample enhancement information set.
[0034] In some optional implementations of some embodiments, the execution subject preprocesses each silo sample information in the silo sample information set to generate silo sample enhancement information, which may include the following steps:
[0035] In the first step, each silo feature information in the silo feature information set included in the silo sample information is normalized to generate a silo feature normalized value, thereby obtaining a silo feature normalized value set. In particular, each silo feature information in the silo feature information set included in the silo sample information is normalized to generate a silo feature normalized value, thereby obtaining a silo feature normalized value set, using a preset normalization algorithm.
[0036] As an example, the above-mentioned preset normalization algorithm may be a maximum-minimum normalization algorithm.
[0037] The second step is to determine the characteristic correlation coefficient value corresponding to every two silo characteristic normalized values in the above silo characteristic normalized value set to generate a characteristic correlation coefficient value group, and obtain a characteristic correlation coefficient value group set. Among them, the characteristic correlation coefficient value corresponding to every two silo characteristic normalized values in the above silo characteristic normalized value set can be determined by a preset similarity formula to generate a characteristic correlation coefficient value group, and obtain a characteristic correlation coefficient value group set.
[0038] As an example, the above-mentioned preset similarity formula may be a Spearman correlation coefficient formula.
[0039] In the third step, for each silo feature normalized value in the above silo feature normalized value set, perform the following screening sub-steps:
[0040] In the first sub-step, the mean of each feature correlation coefficient value in the feature correlation coefficient value group corresponding to the above silo feature normalized value is determined as the correlation coefficient mean value.
[0041] The second sub-step is, in response to determining that the mean value of the correlation coefficient is within a preset interval, determining the silo feature information corresponding to the silo feature normalization value as the silo enhanced feature information.
[0042] As an example, the preset interval may be (0.3, 0.8).
[0043] The fourth step is to determine the enhanced feature information of each silo as the silo sample enhanced information.
[0044] Step 103: pre-process the basic information of the silo to obtain basic enhanced information of the silo.
[0045] In some embodiments, the execution subject may pre-process the basic silo information to obtain basic enhanced silo information. The specific implementation method of generating the basic enhanced silo information and the technical effects thereof may refer to step 102 in the above embodiment, which will not be described in detail here.
[0046] Step 104: construct a silo dynamic normal stress prediction model based on the silo sample enhanced information set.
[0047] In some embodiments, the execution entity may construct a silo dynamic normal stress prediction model based on the silo sample enhanced information set.
[0048] As an example, the above-mentioned silo sample enhanced information set is used to construct a silo dynamic normal stress prediction model, which can be referred to Figure 2 FIG. 1 is a flowchart of an experiment for constructing a derivative distillation model according to some embodiments of the lightweight prediction method for dynamic normal stress of a silo based on derivative distillation of the present application. Figure 2 As shown, MAE stands for mean absolute error. 2 Here, the derived distillation model constructed is the above-mentioned silo dynamic normal stress prediction model. The above-mentioned derived distillation model can be referred to Figure 3 FIG. 1 is a schematic diagram of the structure of a derivative distillation model according to some embodiments of the lightweight prediction method of the silo dynamic normal stress based on derivative distillation of the present application. Figure 3 As shown, TM1 represents the first silo feature information. TM2 represents the first prediction information. TD1 represents the first feature encoding information. TD2 represents the first feature decoding information. SM1 represents the second silo feature information. SM2 represents the second prediction information. SD represents the second channel feature information. N generates a matrix. M1 represents a matrix generated according to the first feature encoding information of the second channel feature information. M2 represents a matrix generated according to the first feature encoding information. W represents a weight coefficient. D1 represents a 1-dimensional tensor containing all main diagonal elements returned after extracting the matrix diagonal elements. D2 represents a 1-dimensional tensor that is returned after extracting the non-diagonal elements of the matrix.
[0049] In some optional implementations of some embodiments, the execution subject constructs a silo dynamic normal stress prediction model based on the silo sample enhanced information set, which may include the following steps:
[0050] The first step is to select the target silo sample enhancement information from the above silo sample enhancement information set and perform the following training sub-steps:
[0051] In the first sub-step, the target silo sample enhancement information is input into the first prediction model to obtain the first silo feature information and the first prediction information. Among them, the silo sample enhancement information can be randomly selected from the above silo sample enhancement information set as the target silo sample enhancement information. The above first prediction information may include but is not limited to: a first prediction value sequence. The first prediction value in the above first prediction value sequence may be the silo dynamic normal stress prediction value output by the above first prediction model.
[0052] The second sub-step is to input the first silo feature information into the first channel model to obtain first feature encoding information and first feature decoding information.
[0053] As an example, the specific structures of the first prediction model and the first channel model can be referred to Figure 4 The structure diagram of the first prediction model and the first channel model of the lightweight prediction method of the silo dynamic normal stress based on derivative distillation according to the present application is shown. Figure 4 As shown, Teacher network represents the above-mentioned first prediction model. LSTM represents Long Short-Term Memory layer. BN represents BatchNormalization layer. ELU represents Exponential Linear Unit layer. GRU represents Gate Recurrent Unit layer. FC represents Fully Connected layer. TM1 represents the first silo feature information. TM2 represents the first prediction information. Label represents the silo dynamic normal stress value included in the target silo sample enhancement information. Teacher Loss represents the loss function used to adjust the first prediction model. Teacher channel represents the first channel model. LeakyReLu represents the (rectified linear unit) activation function. Conv1d represents a one-dimensional convolutional neural network. Stride represents the step size. TD1 represents the first feature encoding information. TD2 represents the first feature decoding information. Channel Loss represents the loss function used to adjust the first channel model.
[0054] The third sub-step is to input the target silo sample enhancement information into the second prediction model to obtain the second silo feature information and the second prediction information. The second prediction information may include but is not limited to: a second prediction value sequence. The second prediction value in the second prediction value sequence may be the silo dynamic normal stress prediction value output by the second prediction model.
[0055] The fourth sub-step is to input the second silo characteristic information into the second channel model to obtain the second channel characteristic information.
[0056] As an example, the specific structures of the second prediction model and the second channel model can be referred to Figure 5 The structure schematic diagram of the second prediction model and the second channel model of the lightweight prediction method of the silo dynamic normal stress based on derivative distillation according to the present application is shown. Figure 5 As shown, Student network represents the second prediction model. SM1 represents the second silo feature information. SM2 represents the second prediction information. SD represents the second channel feature information.
[0057] The fifth sub-step is to determine the first predicted loss value, the first channel loss value, the second predicted loss value and the second channel loss value based on the first silo characteristic information, the first prediction information, the first feature encoding information, the first feature decoding information, the second prediction information, the second channel characteristic information and the silo dynamic normal stress value included in the target silo sample enhancement information.
[0058] In a sixth sub-step, in response to determining that the first predicted loss value, the first channel loss value, the second predicted loss value and the second channel loss value are all less than the target threshold, the second prediction model is determined as a silo dynamic normal stress prediction model.
[0059] As a strength, the above target threshold can be 0.01.
[0060] Optionally, the above execution entity may further perform the following steps:
[0061] In the first step, in response to determining that the first prediction loss value is greater than or equal to the target threshold, the relevant parameters in the first prediction model are adjusted, the adjusted first prediction model is determined as the first prediction model, and the target silo sample enhancement information is selected from each silo sample enhancement information that has not been selected in the silo sample enhancement information set, so as to perform the above training step again. The relevant parameters in the first prediction model can be adjusted by a preset adjustment algorithm.
[0062] As an example, the preset adjustment algorithm mentioned above may be but is not limited to one of the following: a back propagation algorithm or a stochastic gradient algorithm.
[0063] In the second step, in response to determining that the first channel loss value is greater than or equal to the target threshold, the relevant parameters in the first channel model are adjusted, the adjusted first channel model is determined as the first channel model, and the target silo sample enhancement information is selected from the silo sample enhancement information set that has not been selected, so as to perform the training step again. The relevant parameters in the first channel model can be adjusted by the preset adjustment algorithm.
[0064] In the third step, in response to determining that the second prediction loss value is greater than or equal to the target threshold, the relevant parameters in the second prediction model are adjusted, the adjusted second prediction model is determined as the second prediction model, and the target silo sample enhancement information is selected from the silo sample enhancement information set that has not been selected, so as to perform the training step again. The relevant parameters in the second prediction model can be adjusted by the preset adjustment algorithm.
[0065] In step 4, in response to determining that the second channel loss value is greater than or equal to the target threshold, the relevant parameters in the second channel model are adjusted, the adjusted second channel model is determined as the second channel model, and the target silo sample enhancement information is selected from the silo sample enhancement information set that has not been selected, so as to perform the training step again. The relevant parameters in the second channel model can be adjusted by the preset adjustment algorithm.
[0066] In some optional implementations of some embodiments, the execution subject determines the first predicted loss value, the first channel loss value, the second predicted loss value, and the second channel loss value based on the first silo feature information, the first prediction information, the first feature encoding information, the first feature decoding information, the second prediction information, the second channel feature information, and the silo dynamic normal stress value included in the target silo sample enhancement information, which may include the following steps:
[0067] The first step is to determine the first predicted loss value corresponding to the first predicted information and the dynamic normal stress value of the silo based on the first preset loss function. The first preset loss function can be the following formula:
[0068]
[0069] Among them, L ST represents the first preset loss function. n represents the number of first prediction values in the first prediction value sequence included in the first prediction information. i represents the sequence number. y represents the dynamic normal stress value of the silo.
[0070] The second step is to determine the first channel loss value corresponding to the first silo feature information and the first feature decoding information based on the second preset loss function. The second preset loss function can be the following formula:
[0071]
[0072] Among them, L SC represents the first channel loss value. δ represents a preset value. For example, the preset value may be 0.5.
[0073] Step 3. Construct a feature derivative matrix based on the first feature encoding information and the second channel feature information. Based on the first feature encoding information and the second channel feature information, the feature derivative matrix can be constructed by the following formula:
[0074]
[0075] Where C represents the feature derivative matrix. b represents the number of silo sample enhancement information in the silo sample enhancement information set. n represents the first characteristic matrix. (T n )T Indicates T n The transposed matrix of . represents the conjugate matrix of the first feature encoding information. std represents the standard deviation function. std(TD1) represents the standard deviation value corresponding to the first feature encoding information. S n represents the second characteristic matrix. Represents the conjugate matrix of the second channel feature information. std(SD) represents the standard deviation value corresponding to the second channel feature information.
[0076] The fourth step is to perform feature extraction on the feature derivative matrix to obtain the feature derivative loss value. The feature derivative loss value can be obtained by performing feature extraction on the feature derivative matrix using the following formula:
[0077]
[0078] Among them, lossC represents the feature-derived loss value. α represents the first coefficient. β represents the second coefficient. w1 represents a weight coefficient in the second prediction model. mean represents the mean function. offdiagonal represents extracting the matrix diagonal elements and returning a 1-dimensional tensor containing all the main diagonal elements. offdiagonal represents extracting the matrix off-diagonal elements and returning a 1-dimensional tensor.
[0079] Step 5: Based on the third preset loss function, determine the second predicted loss value corresponding to the second prediction information and the silo dynamic normal stress value. The third preset loss function can be the following formula:
[0080]
[0081] Wherein, lossS represents the second prediction loss value. m represents the number of second prediction values in the second prediction value sequence included in the second prediction information. j represents the sequence number. j Indicates the dynamic normal stress value of the silo. j Represents the j-th second prediction value in the second prediction value sequence.
[0082] In the sixth step, the feature-derived loss value and the second predicted loss value are weighted and summed to obtain the second channel loss value. The weighted sum of the feature-derived loss value and the second predicted loss value to obtain the second channel loss value may be: first, the product of the feature-derived loss value and the preset weight coefficient is determined as the weighted loss value. Then, the sum of the weighted loss value and the second predicted loss value is determined as the second channel loss value.
[0083] As an example, the preset weight coefficient may be, but is not limited to, at least one of the following: 0.1 or 0.5.
[0084] Step 105: input the basic reinforcement information of the silo into the silo dynamic normal stress prediction model to obtain the predicted value of the silo dynamic normal stress.
[0085] In some embodiments, the execution entity may input the basic silo enhancement information into the silo dynamic normal stress prediction model to obtain a silo dynamic normal stress prediction value. The inputting the basic silo enhancement information into the silo dynamic normal stress prediction model to obtain a silo dynamic normal stress prediction value may be: first, inputting the basic silo enhancement information into the silo dynamic normal stress prediction model to obtain initial prediction information. The initial prediction information may include: an initial prediction value sequence. Then, the first initial prediction value in the initial prediction value sequence is determined as the silo dynamic normal stress prediction value.
[0086] Step 106, in response to determining that the predicted value of the silo dynamic normal stress is greater than a preset alarm threshold, the predicted value of the silo dynamic normal stress is sent to an alarm terminal for alarm processing.
[0087] In some embodiments, the execution subject may send the predicted value of the dynamic normal stress of the silo to an alarm terminal for alarm processing in response to determining that the predicted value of the dynamic normal stress of the silo is greater than a preset alarm threshold. The alarm terminal may be a terminal for displaying warning text or emitting a prompt sound in response to receiving the predicted value of the dynamic normal stress of the silo.
[0088] As an example, the preset alarm threshold may be 1000.
[0089] The above-mentioned embodiments of the present application have the following beneficial effects: The present invention systematically collects and organizes the simulation and experimental data related to the silo for the first time, and establishes a comprehensive and structured silo database. This database not only contains rich feature information, but also lays a solid foundation for subsequent statistical analysis and model training. By adopting the Spearman correlation coefficient, the correlation analysis of various features of the factors affecting the dynamic normal stress of the silo is performed. The correlation coefficient between each feature is calculated, and the strongly correlated features are eliminated. The database after removing redundant features significantly improves the prediction performance of the model. This not only reduces the complexity of the model, but also improves the generalization ability and prediction accuracy of the model. Through the derivative module, the present invention uses the features output by the teacher model and the student model to construct a derivative matrix. The matrix can quantify and fuse the relationship between different features, thereby improving the expression ability of the model. The derivative module realizes efficient use of features by extracting and fusing the diagonal and off-diagonal elements of the derivative matrix, and further improves the prediction performance of the model. The outputs of different levels of the deep learning model are used, and the correlation between these outputs is captured through the derivative matrix, thereby improving the expression ability and generalization ability of the model. A derivative distillation model was built to train a complex network model and compare its output with the true label to train a smaller lightweight network. By building a combination of student, teacher channels and derivative modules, a lightweight design of the model was achieved, reducing the number of parameters and storage space, while improving prediction performance.
[0090] The present application also provides a computer device 600. Figure 6 As shown, the computer device 600 includes: a bus 601, a processor 602, a memory 603 and a communication interface 604. The processor 602, the memory 603 and the communication interface 604 communicate with each other through the bus 601. The computer device 600 can be a server or a terminal device. It should be understood that the present application does not limit the number of processors and memories in the computer device 600.
[0091] The bus 601 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 The bus 601 is represented by only one line, but it does not mean that there is only one bus or one type of bus. The bus 601 may include a path for transmitting information between various components of the computer device 600 (for example, the memory 603, the processor 602, and the communication interface 604).
[0092] The processor 602 may include any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).
[0093] The memory 603 may include a volatile memory, such as a random access memory (RAM). The memory 603 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).
[0094] The memory 603 stores executable program codes, and the processor 602 executes the executable program codes to respectively implement the functions of the aforementioned acquisition module, sampling module, determination module, and mixing module, thereby implementing the aforementioned lightweight prediction method for the dynamic normal stress of the silo based on derivative distillation. That is, the memory 603 stores instructions for executing the aforementioned lightweight prediction method for the dynamic normal stress of the silo based on derivative distillation.
[0095] The communication interface 604 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computer device 600 and other devices or communication networks.
[0096] An embodiment of the present application also provides a chip, which includes a processor and a data interface. The processor reads instructions stored in a memory through the data interface to execute the above-mentioned lightweight prediction method for the dynamic normal stress of a silo based on derivative distillation.
[0097] The embodiment of the present application also provides a computer-readable storage medium. The above-mentioned computer-readable storage medium can be any available medium that can be stored by a computing device or a data storage device such as a data center containing one or more available media. The above-mentioned available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid-state hard disk), etc. The computer-readable storage medium includes instructions, and the above-mentioned instructions instruct the computing device to execute the above-mentioned lightweight prediction method of the dynamic normal stress of the silo based on derivative distillation.
[0098] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0099] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the protection scope of the technical solutions of the embodiments of the present application.
Claims
1. A lightweight prediction method for dynamic normal stress of a silo based on derivative distillation, comprising: Get silo sample information set and silo basic information; Preprocessing each silo sample information in the silo sample information set to generate silo sample enhancement information, thereby obtaining a silo sample enhancement information set; Preprocessing the basic information of the silo to obtain basic enhanced information of the silo; Based on the silo sample enhanced information set, a silo dynamic normal stress prediction model is constructed; Inputting the basic reinforcement information of the silo into the dynamic normal stress prediction model of the silo to obtain a predicted value of the dynamic normal stress of the silo; In response to determining that the predicted value of the silo dynamic normal stress is greater than a preset alarm threshold, the predicted value of the silo dynamic normal stress is sent to an alarm terminal for alarm processing.
2. The lightweight prediction method for dynamic normal stress of silo based on derivative distillation according to claim 1, wherein: The silo sample information in the silo sample information set includes: a silo feature information set; and the preprocessing of each silo sample information in the silo sample information set to generate silo sample enhancement information includes: Normalizing each silo feature information in the silo feature information set included in the silo sample information to generate a silo feature normalized value, thereby obtaining a silo feature normalized value set; Determine the characteristic correlation coefficient value corresponding to every two silo characteristic normalized values in the silo characteristic normalized value set to generate a characteristic correlation coefficient value group, and obtain a characteristic correlation coefficient value group set; For each silo feature normalized value in the silo feature normalized value set, the following screening steps are performed: Determine the mean of each characteristic correlation coefficient value in the characteristic correlation coefficient value group corresponding to the silo characteristic normalized value as the correlation coefficient mean; In response to determining that the correlation coefficient mean is within a preset interval, determining the silo feature information corresponding to the silo feature normalization value as silo enhanced feature information; The determined individual silo enhancement feature information is determined as silo sample enhancement information.
3. The lightweight prediction method for dynamic normal stress of silo based on derivative distillation according to claim 1, wherein: The silo sample information in the silo sample information set further includes: a silo dynamic normal stress value; and constructing a silo dynamic normal stress prediction model based on the silo sample enhanced information set includes: Select target silo sample enhancement information from the silo sample enhancement information set and perform the following training steps: Inputting the target silo sample enhancement information into the first prediction model to obtain first silo feature information and first prediction information; Inputting the first silo feature information into the first channel model to obtain first feature encoding information and first feature decoding information; Inputting the target silo sample enhancement information into the second prediction model to obtain second silo feature information and second prediction information; Inputting the second silo characteristic information into the second channel model to obtain the second channel characteristic information; Determine a first predicted loss value, a first channel loss value, a second predicted loss value, and a second channel loss value based on the first silo feature information, the first prediction information, the first feature encoding information, the first feature decoding information, the second prediction information, the second channel feature information, and the silo dynamic normal stress value included in the target silo sample enhancement information; In response to determining that the first predicted loss value, the first channel loss value, the second predicted loss value, and the second channel loss value are all less than the target threshold, the second prediction model is determined to be a silo dynamic normal stress prediction model.
4. The lightweight prediction method for dynamic normal stress of silo based on derivative distillation according to claim 3, wherein: The method further comprises: In response to determining that the first prediction loss value is greater than or equal to the target threshold, adjusting relevant parameters in the first prediction model, determining the adjusted first prediction model as the first prediction model, and selecting target silo sample enhancement information from each silo sample enhancement information that has not been selected in the silo sample enhancement information set, so as to perform the training step again; In response to determining that the first channel loss value is greater than or equal to the target threshold, adjusting relevant parameters in the first channel model, determining the adjusted first channel model as the first channel model, and selecting target silo sample enhancement information from each silo sample enhancement information that has not been selected in the silo sample enhancement information set, so as to perform the training step again; In response to determining that the second prediction loss value is greater than or equal to the target threshold, adjusting relevant parameters in the second prediction model, determining the adjusted second prediction model as the second prediction model, and selecting target silo sample enhancement information from each silo sample enhancement information that has not been selected in the silo sample enhancement information set, so as to perform the training step again; In response to determining that the second channel loss value is greater than or equal to the target threshold, relevant parameters in the second channel model are adjusted, the adjusted second channel model is determined as the second channel model, and target silo sample enhancement information is selected from each silo sample enhancement information that has not been selected in the silo sample enhancement information set, so as to perform the training step again.
5. The lightweight prediction method for dynamic normal stress of silo based on derivative distillation according to claim 3, wherein: The method of determining the first predicted loss value, the first channel loss value, the second predicted loss value, and the second channel loss value based on the first silo feature information, the first prediction information, the first feature encoding information, the first feature decoding information, the second prediction information, the second channel feature information, and the silo dynamic normal stress value included in the target silo sample enhancement information comprises: Based on the first preset loss function, determining a first predicted loss value corresponding to the first predicted information and the dynamic normal stress value of the silo; Based on a second preset loss function, determining a first channel loss value corresponding to the first silo feature information and the first feature decoding information; Based on the first feature encoding information and the second channel feature information, construct a feature derivative matrix; Perform feature extraction processing on the feature derivative matrix to obtain the feature derivative loss value; Based on a third preset loss function, determining a second predicted loss value corresponding to the second predicted information and the dynamic normal stress value of the silo; The feature-derived loss value and the second prediction loss value are weighted and summed to obtain the second channel loss value.
6. A computer device, wherein: The computer device comprises a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 5 are implemented.
7. A computer-readable storage medium, wherein: The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.