Calibration result generation model training method, calibration result generation model generation method, calibration result generation system, calibration result generation equipment and medium

Abstract

The invention provides a calibration result generation model training method, a calibration result generation model generation method, a calibration result generation system, equipment and a medium. The training method comprises the following steps: acquiring a motor key design parameter sequence containing multiple labels; inputting the motor key design parameter sequence into a feature extraction network of a calibration result generation model, and extracting various basic features; inputting each basic feature into an expert network corresponding to each basic feature in a calibration result generation model, and correspondingly extracting a specific feature; inputting the basic features into a shared network of a calibration result generation model, and extracting universal features; inputting the general feature and the specific feature of each basic feature into a task tower network corresponding to each basic feature in a calibration result generation model, and generating a corresponding prediction calibration result; and updating the calibration result generation model according to the predicted calibration result and the difference degree of the corresponding label, and obtaining a trained calibration result generation model. The problems that the generation period of the calibration result of the motor is long, and the generation accuracy is low are solved.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and particularly relates to a training and generation method, system, device and medium for a calibration result generation model. Background Art

[0002] In the field of motor calibration, the current mainstream methods include the working condition sweep point method, the parameter identification method, and hybrid calibration methods such as sweep point + fitting. Limited by factors such as sample scatter, measurement deviation, and temperature time-variation, and with complicated subsequent data processing, the required cycle of various calibration methods is very long, the test cost is relatively high, and the calibration result may not be at the median value, resulting in an increase in the number of products that cannot meet the technical indicators. In addition, the motor calibration result is usually manually calibrated, which not only makes the generation cycle of the calibration result longer, but also affects the accuracy of the calibration result. Therefore, a training method, system, device and medium for a calibration result generation model are needed. Summary of the Invention

[0003] The present invention provides a training method for a calibration result generation model, so as to solve the problems in the prior art that the generation cycle of the calibration result of the motor is long and the generated accuracy is low.

[0004] The training method for a calibration result generation model provided by the present invention includes: obtaining a sequence of key design parameters of a motor with multiple labels; each label represents a motor calibration result; inputting the sequence of key design parameters of the motor into a feature extraction network of the calibration result generation model to extract multiple basic features in the sequence of key design parameters of the motor; each basic feature corresponds to a motor calibration result; respectively inputting each basic feature into an expert network corresponding to each basic feature in the calibration result generation model to extract specific features in each basic feature; jointly inputting each basic feature into a shared network of the calibration result generation model to extract general features in each basic feature; inputting the general features and the specific features of each basic feature into a task tower network corresponding to each basic feature in the calibration result generation model to generate a predicted calibration result corresponding to each basic feature; and updating the parameters of the calibration result generation model according to the difference degree between the predicted calibration result and the corresponding label to obtain a trained calibration result generation model.

[0005] In an embodiment of the present invention, inputting the motor key design parameter sequence into the feature extraction network of the calibration result generation model to extract various basic features in the motor key design parameter sequence, including: standardizing each motor key design parameter in the motor key design parameter sequence; performing correlation verification on each motor key design parameter in the standardized motor key design parameter sequence to obtain a verified motor key design parameter sequence; inputting the verified motor key design parameter sequence into the feature extraction network of the calibration result generation model to extract various basic features in the verified motor key design parameter sequence.

[0006] In an embodiment of the present invention, performing correlation verification on each motor key design parameter in the standardized motor key design parameter sequence to obtain a verified motor key design parameter sequence includes: calculating the correlation coefficient of each motor key design parameter in the standardized motor key design parameter sequence with respect to other motor key design parameters in the standardized motor key design parameter sequence based on the canonical correlation analysis method; when the correlation coefficient is greater than a preset coefficient threshold, deleting the two motor key design parameters corresponding to the correlation coefficient from the standardized motor key design parameter sequence to obtain a verified motor key design parameter sequence.

[0007] In an embodiment of the present invention, updating the parameters of the calibration result generation model according to the difference between the predicted calibration result and the corresponding label includes: calculating the difference between each predicted calibration result and the corresponding label according to a preset loss function; updating the parameters of the expert network and the task tower network corresponding to each label in the calibration result generation model according to the difference between each predicted calibration result and the corresponding label; performing dynamic weighted summation on the differences between each predicted calibration result and the corresponding label, and updating the parameters of the shared network and the feature extraction network of the calibration result generation model according to the weighted summation value.

[0008] In an embodiment of the present invention, the calculation method for performing dynamic weighted summation on the differences between each predicted calibration result and the corresponding label includes: where L is the weighted summation value, K is the number of labels, L i is the difference of the i-th label, and w i is the dynamic weight assigned to the difference of the i-th label based on the dynamic weight averaging method.

[0009] In an embodiment of the present invention, the task tower network is constructed based on the attention mechanism.

[0010] In an embodiment of the present invention, there is also provided a method for generating a calibration result, including: obtaining a sequence of key motor parameters; inputting the sequence of key motor parameters into a calibration result generation model trained by the training method of the calibration result generation model described in any one of the above, to obtain multiple motor calibration results.

[0011] In another aspect of the present invention, there is also provided a training system for a calibration result generation model. The system includes: a data acquisition module, configured to acquire a sequence of key motor design parameters with multiple labels; each label represents a motor calibration result; a basic feature extraction module, configured to input the sequence of key motor design parameters into a feature extraction network of the calibration result generation model to extract multiple basic features in the sequence of key motor design parameters; each basic feature corresponds to a motor calibration result; a specific feature extraction module, configured to input each basic feature into an expert network corresponding to each basic feature in the calibration result generation model to extract specific features in each basic feature; a general feature extraction module, configured to input all the basic features into a shared network of the calibration result generation model to extract general features in all the basic features; a prediction result acquisition module, configured to input the general features and the specific features of each basic feature into a task tower network corresponding to each basic feature in the calibration result generation model to generate a predicted calibration result corresponding to each basic feature; a parameter update module, configured to update the parameters of the calibration result generation model according to the difference degree between the predicted calibration result and the corresponding label, to obtain a trained calibration result generation model.

[0012] In an embodiment of the present invention, there is also provided an electronic device, which includes: one or more processors; a storage device, configured to store one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device implements the training method of the calibration result generation model described in any one of the above.

[0013] In an embodiment of the present invention, there is also provided a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor of a computer, the computer executes the training method of the calibration result generation model described in any one of the above.

[0014] A training method, system, device, and medium for a calibration result generation model proposed by the present invention input a sequence of key motor design parameters into the feature extraction network of the calibration result generation model to extract various basic features that can be used by subsequent networks. Input each basic feature into the expert network corresponding to the calibration result generation model to provide specific features unique to the current motor calibration result task. Input all basic features into the shared network of the calibration result generation model to extract the common features among these bases as general features. Input the general features and the corresponding specific features into the task tower network corresponding to the specific feature of the calibration result generation model to obtain the predicted calibration result corresponding to the task tower network. This multi-task learning model used in the present invention can accurately predict various different motor calibration results, and the time to generate various motor calibration results is much less than the time of manual calibration, effectively solving the problems of long calibration result generation cycle and low accuracy in the prior art for motors. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic flowchart of the training method for the calibration result generation model provided by an embodiment of the present invention;

[0016] Figure 2 It is shown as a structural block diagram of the training system for the calibration result generation model provided by an embodiment of the present invention;

[0017] Figure 3 It is shown as a schematic structural diagram of an electronic device for the training method of the calibration result generation model of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The following uses specific specific examples to illustrate the embodiments of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0019] It should be noted that the drawings provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner. Therefore, only the components related to the present invention are shown in the drawings, rather than being drawn according to the number, shape, and size of the components in actual implementation. The type, quantity, and ratio of each component in actual implementation can be arbitrarily changed, and the component layout type may also be more complex.

[0020] In the following description, numerous details are explored to provide a more thorough explanation of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the embodiments of the present invention.

[0021] The present invention provides a method for training a calibration result generation model. By inputting a sequence of key motor design parameters into the feature extraction network of the calibration result generation model, various basic features that can be used by subsequent networks are extracted. Each basic feature is input into the corresponding expert network of the calibration result generation model to provide specific features unique to the current motor calibration result task. All the basic features are input into the shared network of the calibration result generation model, and the common features among these basic features are extracted as general features. The general features and the corresponding specific features are input into the task tower network corresponding to the specific feature of the calibration result generation model, and the predicted calibration result corresponding to the task tower network is obtained. This multi-task learning model used in the present invention can accurately predict various different motor calibration results, greatly shorten the calibration period, sharply reduce the calibration test cost, make the calibration result more robust, and greatly improve the yield rate. It can be flexibly extended and then applied to the product development for other purposes. It effectively solves the problems in the prior art that the calibration result generation period of the motor is long and the generated accuracy is low.

[0022] Please refer to Figure 1 , the method for training the calibration result generation model includes the following steps:

[0023] S1. Obtain a sequence of key motor design parameters with multiple labels; each label represents a motor calibration result;

[0024] S2. Input the sequence of key motor design parameters into the feature extraction network of the calibration result generation model, and extract various basic features in the sequence of key motor design parameters; each basic feature corresponds to a motor calibration result;

[0025] S3. Input each basic feature into the corresponding expert network in the calibration result generation model for the respective basic feature, and extract the specific features in each basic feature;

[0026] S4. Input all the basic features into the shared network of the calibration result generation model, and extract the general features in each basic feature;

[0027] S5. Input the general features and the specific features of each basic feature into the task tower network corresponding to each basic feature in the calibration result generation model to generate the predicted calibration result corresponding to each basic feature;

[0028] S6. Update the parameters of the calibration result generation model according to the difference between the predicted calibration result and the corresponding label, and obtain the trained calibration result generation model.

[0029] The following specifically describes the specific implementation process of each step:

[0030] S1. Obtain a sequence of motor key design parameters with multiple labels; each label represents a motor calibration result.

[0031] The motor key design parameters in this application refer to the design parameters that can affect the performance of the motor. Among them, these design parameters that affect the motor performance include but are not limited to material selection parameters, size parameters, and electrical parameters, etc. Exemplarily, the specific parameters include but are not limited to the motor magnet grade, magnet size and arrangement parameters, motor internal resistance, air gap, number of pole pairs, number of slots, number of winding layers, etc. In this application, by inputting the motor key design parameters into the calibration result generation model, the calibration result of the motor is obtained to realize the automatic calibration of the motor. Motor calibration refers to a series of tests and measurements on the parameters of the motor to confirm whether the performance indicators of the motor meet specific standards or intended uses. These tests include but are not limited to power output, torque, efficiency, current and voltage levels, and other important performance parameters. Each label represents a specific performance result of the motor under a certain specific test or measurement condition, as a calibration result of the motor. Through this multi-label classification of the motor design parameters, the performance of the motor under different working conditions or design changes can be understood more carefully.

[0032] In an embodiment of the present invention, input the sequence of motor key design parameters into the feature extraction network of the calibration result generation model, and extract various basic features in the sequence of motor key design parameters, including:

[0033] Standardize each motor key design parameter in the sequence of motor key design parameters;

[0034] Perform correlation verification on each motor key design parameter in the standardized sequence of motor key design parameters to obtain a verified sequence of motor key design parameters;

[0035] Input the verified sequence of motor key design parameters into the feature extraction network of the calibration result generation model, and extract various basic features in the verified sequence of motor key design parameters.

[0036] Considering that different dimensions may be involved in the key design parameters of the motor, it is necessary to standardize them to unify the dimensions. Further, in the training stage of the model, it is also necessary to unify the dimensions of all labels to reduce the errors caused by inconsistent units. To reduce the complexity of subsequent data mining, it is also necessary to perform a correlation check on the sequence of key motor design parameters after standardization and eliminate the key motor design parameters with high correlation. After the check, the verified key motor design parameters are input into the model for iterative training. It should be noted that this application is described by taking the Progressive Layered Extraction (PLE) model as an example, but other machine learning, deep learning models, or multi-task learning models based on hard parameter sharing can also be used, which are not limited herein.

[0037] In an embodiment of the present invention, the correlation check of each key motor design parameter in the sequence of key motor design parameters after standardization to obtain the sequence of verified key motor design parameters includes:

[0038] Based on the canonical correlation analysis method, calculate the correlation coefficient of each key motor design parameter in the sequence of key motor design parameters after standardization with respect to other key motor design parameters in the sequence of key motor design parameters after standardization;

[0039] When the correlation coefficient is greater than the preset coefficient threshold, delete the two key motor design parameters corresponding to the correlation coefficient from the sequence of key motor design parameters after standardization to obtain the sequence of verified key motor design parameters.

[0040] Considering that there are a large number of design parameters that can affect the performance of the motor, after such a large number of parameters are input into the calibration result generation model, it will increase the computational workload of the model. To reduce the complexity of subsequent data mining, in this application, the correlation of the key design parameters of the motor is verified, and appropriate parameters are selected. Specifically, one of the design parameters that can affect the performance of the motor is used as the target value, and the canonical correlation analysis (CCA) method is adopted to calculate the correlation coefficient between the target value and the remaining design parameters that can affect the performance of the motor one by one. The obtained correlation coefficients are respectively compared with a preset coefficient threshold. Since the larger the correlation coefficient, the higher the degree of association between the two, to reduce the complexity of subsequent data mining, it is necessary to remove the two key design parameters of the motor corresponding to the correlation coefficient greater than the preset coefficient threshold from the sequence of standardized key design parameters of the motor, and update the sequence of key design parameters of the motor. After the update is completed, another design parameter that can affect the performance of the motor is selected as the target value, and the above process is continued until all the key design parameters of the motor are verified, so as to obtain the updated sequence of key design parameters of the motor, which is used as the verified sequence of key design parameters of the motor. It should be noted that to save the computational workload, in this application, the staff can also select the target value based on the motor principle and design experience, and improve the calculation rate by specifically selecting the target value. In this application, in the updated sequence of key design parameters of the motor, each key design parameter of the motor is classified and stored as a structured array according to the parameter category and function. Among them, the verified key design parameters of the motor include but are not limited to stator size, stator winding size, rotor size, silicon steel sheet size, magnet grade and size, number of poles and slots, air gap, maximum current, skew pole parameters, and winding material parameters.

[0041] S2. Input the sequence of the key design parameters of the motor into the feature extraction network of the calibration result generation model, and extract various basic features in the sequence of the key design parameters of the motor; each basic feature corresponds to a motor calibration result.

[0042] The basic feature refers to the initial feature extracted from the sequence of the key design parameters of the motor. After the sequence of the key design parameters of the motor is input into the feature extraction network, as the bottom layer of the model, the feature extraction network does not care about what task the feature will be used for, but tries to capture the key characteristics of the input data as much as possible. By learning the complex relationships between the key design parameters of the motor, it aims to capture various basic features in the sequence of the key design parameters of the motor. In addition, the feature extraction network can also help filter out the noise in the data, thus greatly improving the prediction accuracy of the calibration result generation model. The feature extraction network may include a normalization layer and one or more fully connected layers, and the intermediate output value of the feature extraction network is adjusted through the normalization layer to make it more stable.

[0043] S3. Input each basic feature into the expert network corresponding to the basic feature in the calibration result generation model, and extract specific features in each basic feature accordingly.

[0044] In the calibration result generation model, since each basic feature corresponds to an expert network, after inputting each basic feature into the corresponding expert network, each expert network will further extract features that are more advanced or more detailed than the corresponding basic feature from the corresponding basic feature as specific features. Among them, each expert network includes, but is not limited to, a fully connected layer, an activation layer, a regularization layer, etc. Each layer performs specific operations on the input data to learn the deep features of the data. Specifically, after inputting the basic feature into the expert network, the expert network further maps these features to a new space related to the corresponding calibration result prediction task through a series of transformations such as weighted sum and non-linear activation, and in the process of model training, by continuously adjusting the weights, obtain the features that are most useful for the calibration result prediction task, and use these features as specific features. Exemplarily, if the motor calibration result to be predicted is the motor operating efficiency, the corresponding expert network will focus on extracting several features related to power consumption, thermal efficiency, and load characteristics, etc. as the specific features of the output. Further, feature fusion can be performed between some specific features with strong correlation. This feature fusion can be achieved by setting a gating mechanism or a fusion layer to integrate the specific features output by different expert networks, so as to provide a more comprehensive perspective for the final decision-making. It can be understood that the specific network structure of each expert network is related to the complexity of the task, the dimension of the input basic feature, and the motor calibration result to be predicted. Those skilled in the art can adaptively set the specific structure based on the actual situation and are not limited here.

[0045] S4. Input all the basic features into the shared network of the calibration result generation model to extract the common features in each basic feature.

[0046] The shared network, as a part of the calibration result generation model, is designed to process the input features common to all tasks and extract the common information that is useful for all tasks as common features. Specifically, the shared network consists of multiple layers, including but not limited to fully connected layers, convolutional layers, and recurrent layers, etc. The shared network processes the input basic features through these layers to capture the general patterns and rules in these basic features. In addition, the shared network converts the input features into a representation of common features through weighted summation of network layers and activation functions, etc. These common features capture the useful information across various tasks, and this useful information is manifested as the basic physical laws and electrical laws that can affect various performance indicators of the motor. By focusing on the most informative features, the overall performance and processing efficiency of the model are greatly improved.

[0047] Furthermore, the calibration result generation model has multiple depth feature extraction modules that accumulate layer by layer. Each depth feature extraction module includes a shared network and multiple expert networks, and each expert network corresponds to a motor calibration result. For each expert network: the specific features output by the expert network, the general features output by the shared network, and the input value of the input layer of the depth feature extraction module are feature-converged through a gating mechanism. The output of the previous depth feature extraction module is used as the input of the next depth feature extraction module. Through this multi-layer feature extraction, complex data representations from the bottom layer to the high layer can be captured, providing a more refined information processing method for the complex task of inputting a set of motor key design parameter sequences and outputting multiple different motor calibration results, and improving the overall performance of the calibration result generation model. Further, the gating network can automatically determine which features output by the shared network and the expert network are more important by learning the requirements of each motor calibration task. Through the gating mechanism, the amount of information flowing from the shared network and the expert network to the next depth feature extraction module or the task tower network can be dynamically controlled. By selectively using features in this way, the model can maintain good performance on each task. The gating mechanism described in the present invention has a network structure including but not limited to a long short-term memory (LSTM) network or a gated recurrent unit (GRU).

[0048] S5. Input the general features and the specific features of each basic feature into the task tower network corresponding to each basic feature in the calibration result generation model to generate the predicted calibration results corresponding to each basic feature.

[0049] The task tower network, as the final part of the calibration result generation model, is responsible for converting the features extracted by the shared network and the corresponding expert network into specific prediction results, which are used as the motor calibration results output. Specifically, each task tower network receives two types of features: general features from the shared network and specific features extracted by the corresponding expert network. By combining the general features and specific features, the corresponding motor calibration results can be accurately predicted. Each task tower network can not only be optimized for specific motor calibration results but also benefit from the shared network through the general features provided by the shared network that are helpful for all motor calibration results. Therefore, in this way, the calibration result generation model described in the present invention can not only effectively utilize the common information in the data but also provide customized solutions for each specific motor calibration result, so that for each motor calibration result, the generated prediction result is more accurate. Further, in an embodiment of the present invention, the task tower network is constructed based on an attention mechanism so as to be able to focus more on the part of the input data that is most relevant to the current task and improve the accuracy of the model. The motor calibration results refer to various data stored in the motor controller software after the motor and the motor controller are calibrated in cooperation to control the operation of the motor. The motor calibration results described in the present invention refer to the relevant variables indicating the motor characteristics in the calibration results, and these variables include but are not limited to the direct-axis and quadrature-axis current distribution table, the inductance table in the electric mode, the inductance table in the generating mode, the magnetic flux and correction parameters, the saturation correction table, the minimum direct-axis current limit table, and the motor torque external characteristic table.

[0050] S6. Update the parameters of the calibration result generation model according to the difference degree between the predicted calibration result and the corresponding label, and obtain the trained calibration result generation model.

[0051] Specifically, in an embodiment of the present invention, the updating of the parameters of the calibration result generation model according to the difference degree between the predicted calibration result and the corresponding label includes:

[0052] Calculate the difference degree between each predicted calibration result and the corresponding label according to a preset loss function;

[0053] Update the parameters of the expert network and the task tower network corresponding to each label in the calibration result generation model according to the difference degree between each predicted calibration result and the corresponding label;

[0054] Perform dynamic weighted summation on the difference degrees between each predicted calibration result and the corresponding label, and update the parameters of the shared network and the feature extraction network of the calibration result generation model according to the weighted summation value.

[0055] Each task tower network has a corresponding loss function. For one of the task tower networks: according to its corresponding loss function, calculate the difference degree between the predicted calibration result of the output and the corresponding label, and use the difference degree between the two as the value of the loss function. Through the gradient descent method, update the parameters of the corresponding expert network and task tower network in the reverse direction. Dynamically weighted sum the loss function values of each task tower network to obtain the total loss, and update the parameters of the shared network and feature extraction network according to the total loss. When the model iteration reaches the preset number of times, the training can be terminated, so as to obtain a trained calibration result generation model. Specifically, in an embodiment of the present invention, the calculation method for dynamically weighted summing the difference degrees between each predicted calibration result and the corresponding label is shown in formula (1):

[0056]

[0057] where L is the weighted sum value, K is the number of labels, L i is the difference degree of the i-th label, w i is the dynamic weight assigned to the difference degree of the i-th label based on the dynamic weight averaging method. Through this dynamic weight, dynamically weighted sum different loss functions, so that the model can adaptively adjust the focus according to the current performance of each motor calibration result task. If a certain motor calibration result task performs worse than other tasks, the model will give a greater weight to the loss function corresponding to this task, improving the priority, so as to prompt the model to pay more attention to such poorly performing tasks during training. As the training progress changes, the dynamic weighting provides the flexibility to adjust the weights of the loss functions, enabling the model to adapt to the requirements of various training stages. Therefore, the calibration result generation model proposed by the present invention can handle the complexity of multi-task learning more intelligently and efficiently, so that the model can ultimately predict various calibration results of the motor more accurately. Further, the calibration result generation model described in the present invention can be adjusted to performance output for motor performance simulation, thereby reversely supporting motor design. In addition, such a calibration result generation model can also be combined with other motor simulation models, upper transmission systems, and even vehicle simulation systems to construct different forms of digital twin systems to achieve various development work under different environments and for different purposes.

[0058] In an embodiment of the present invention, a calibration result generation method is further provided, including:

[0059] Obtain a sequence of key motor parameters;

[0060] Input the sequence of key motor parameters into the calibration result generation model trained by the training method of the calibration result generation model described in any one of the above, and obtain various motor calibration results.

[0061] For a newly designed motor, after obtaining the sequence of key motor parameters, input it into the trained calibration result generation model to automatically obtain various motor calibration results. Further, after obtaining the motor calibration results, the obtained motor calibration results can be placed on a test bench for physical verification and fine-tuned if necessary to make the motor calibration results more accurate.

[0062] Please refer to Figure 2 , the training system 100 of the calibration result generation model includes: a data acquisition module 110, a basic feature extraction module 120, a specific feature extraction module 130, a general feature extraction module 140, and a prediction result acquisition module 150. The above data acquisition module 110 is used to acquire the sequence of key motor design parameters with multiple labels; each label represents a motor calibration result. The basic feature extraction module 120 is used to input the sequence of key motor design parameters into the feature extraction network of the calibration result generation model to extract various basic features in the sequence of key motor design parameters; each basic feature corresponds to a motor calibration result. The specific feature extraction module 130 is used to input each basic feature into the expert network corresponding to each basic feature in the calibration result generation model to extract the specific features in each basic feature. The general feature extraction module 140 is used to input all the basic features into the shared network of the calibration result generation model to extract the general features in each basic feature. The prediction result acquisition module 150 is used to input the general features and the specific features of each basic feature into the task tower network corresponding to each basic feature in the calibration result generation model to generate the predicted calibration results corresponding to each basic feature. The parameter update module 160 is used to update the parameters of the calibration result generation model according to the difference between the predicted calibration results and the corresponding labels to obtain the trained calibration result generation model.

[0063] It should be noted that, in order to highlight the innovative part of the present invention, modules that are not closely related to solving the technical problems proposed by the present invention are not introduced in this embodiment, but this does not mean that there are no other modules in this embodiment.

[0064] Please refer to Figure 3 , the electronic device 1 may include a memory 12, a processor 13, and a bus, and may also include a computer program stored in the memory 12 and executable on the processor 13, such as the training program of the calibration result generation model.

[0065] Among them, the memory 12 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), magnetic memory, magnetic disk, optical disc, etc. The memory 12 can be an internal storage unit of the electronic device 1 in some embodiments, such as the mobile hard disk of the electronic device 1. The memory 12 can also be an external storage device of the electronic device 1 in other embodiments, such as a plug-in mobile hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the electronic device 1. Further, the memory 12 can also include both the internal storage unit and the external storage device of the electronic device 1. The memory 12 can be used not only to store application software installed on the electronic device 1 and various types of data, such as the code for training the calibration result generation model, etc., but also to temporarily store the data that has been output or will be output.

[0066] The processor 13 can be composed of integrated circuits in some embodiments. For example, it can be composed of a single packaged integrated circuit, or can be composed of multiple integrated circuits with the same or different functions packaged, including a combination of one or more Central Processing Units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips, etc. The processor 13 is the control core (Control Unit) of the electronic device 1, connecting various components of the entire electronic device 1 through various interfaces and lines, and by running or executing programs or modules stored in the memory 12 (such as the training program of the calibration result generation model, etc.), and calling the data stored in the memory 12, to execute various functions of the electronic device 1 and process data.

[0067] The processor 13 executes the operating system of the electronic device 1 and various installed application programs. The processor 13 executes the application programs to implement the steps in the above-mentioned calibration result generation model training method.

[0068] Exemplarily, the computer program can be divided into one or more modules, and the one or more modules are stored in the memory 12 and executed by the processor 13 to complete this application. The one or more modules can be a series of computer program instruction segments capable of completing specific functions, and this instruction segment is used to describe the execution process of the computer program in the electronic device 1. For example, the computer program can be divided into a data acquisition module 110, a basic feature extraction module 120, a specific feature extraction module 130, a general feature extraction module 140, and a prediction result acquisition module 150.

[0069] The integrated unit implemented in the form of software functional modules described above can be stored in a computer-readable storage medium, which can be non-volatile or volatile. The above software functional modules are stored in a storage medium and include several instructions for causing a computer device (which can be a personal computer, a computer device, or a network device, etc.) or a processor to execute some functions of the training method of the calibration result generation model according to the embodiments of the present application.

[0070] In summary, the present invention discloses a training and generation method, system, device, and medium for a calibration result generation model. By inputting the motor key design parameter sequence into the feature extraction network of the calibration result generation model, various basic features that can be used by subsequent networks are extracted. Each basic feature is input into the expert network corresponding to the calibration result generation model to provide specific features unique to the current motor calibration result task. All basic features are input into the shared network of the calibration result generation model, and the common features among these basic features are extracted as general features. The general features and the corresponding specific features are input into the task tower network corresponding to the calibration result generation model and the specific feature, and the predicted calibration result corresponding to the task tower network is obtained. The present invention follows the trend of advanced technologies and uses the relevant data of mature motors to automatically output the corresponding calibration results. The multi-task learning model used in the present invention can accurately predict various different motor calibration results, and the time for generating various motor calibration results is much less than that of manual calibration, and the cost is low. The calibration results are in the median, effectively solving the problems of long calibration result generation cycle and low accuracy in the prior art. Therefore, the present invention effectively overcomes various disadvantages in the prior art and has high industrial utilization value.

[0071] The above embodiments are only illustrative of the principles and effects of the present invention, and are not used to limit the present invention. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical idea disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. A training method for a calibration result generation model, characterized in that, the training method includes: obtaining a sequence of motor key design parameters with multiple labels; each label represents a motor calibration result; inputting the sequence of motor key design parameters into the feature extraction network of the calibration result generation model to extract multiple basic features in the sequence of motor key design parameters; each basic feature corresponds to a motor calibration result; inputting each basic feature into the expert network corresponding to each basic feature in the calibration result generation model to extract specific features in each basic feature; inputting all basic features into the shared network of the calibration result generation model to extract general features in all basic features; inputting the general features and the specific features of each basic feature into the task tower network corresponding to each basic feature in the calibration result generation model to generate predicted calibration results corresponding to each basic feature; updating the parameters of the calibration result generation model according to the difference degree between the predicted calibration result and the corresponding label, and obtaining a trained calibration result generation model.

2. The training method for a calibration result generation model according to claim 1, characterized in that, the step of inputting the sequence of motor key design parameters into the feature extraction network of the calibration result generation model to extract multiple basic features in the sequence of motor key design parameters includes: performing standardization processing on each motor key design parameter in the sequence of motor key design parameters; performing correlation verification on each motor key design parameter in the sequence of standardized motor key design parameters to obtain a sequence of verified motor key design parameters; inputting the sequence of verified motor key design parameters into the feature extraction network of the calibration result generation model to extract multiple basic features in the sequence of verified motor key design parameters.

3. The training method for a calibration result generation model according to claim 2, characterized in that, the step of performing correlation verification on each motor key design parameter in the sequence of standardized motor key design parameters to obtain a sequence of verified motor key design parameters includes: based on the canonical correlation analysis method, calculating the correlation coefficient of each motor key design parameter in the sequence of standardized motor key design parameters relative to other motor key design parameters in the sequence of standardized motor key design parameters; when the correlation coefficient is greater than a preset coefficient threshold, deleting the two motor key design parameters corresponding to the correlation coefficient from the sequence of standardized motor key design parameters to obtain a sequence of verified motor key design parameters.

4. The training method for a calibration result generation model according to claim 1, characterized in that, the step of updating the parameters of the calibration result generation model according to the difference degree between the predicted calibration result and the corresponding label includes: calculating the difference degree between each predicted calibration result and the corresponding label according to a preset loss function; updating the parameters of the expert network and the task tower network corresponding to each label in the calibration result generation model according to the difference degree between each predicted calibration result and the corresponding label. Dynamically weighted sum the differences between each predicted calibration result and the corresponding label, and update the calibration result according to the weighted sum value to generate the parameters of the shared network and the feature extraction network of the model.

5. The training method for the calibration result generation model according to claim 4, wherein, The calculation method of dynamically weighted summation of the differences between the respective prediction calibration results and the corresponding labels includes: where L is the weighted summation value, K is the number of labels, L i is the difference degree of the i-th label, w i is the dynamic weight corresponding to the difference degree of the i-th label assigned based on the dynamic weight averaging method.

6. The training method for the calibration result generation model according to claim 1, wherein, the task tower network is constructed based on the attention mechanism.

7. A calibration result generation method, wherein, the method includes: Obtain the motor key parameter sequence; Input the motor key parameter sequence into the calibration result generation model trained by the training method of the calibration result generation model according to any one of claims 1-6 to obtain multiple motor calibration results.

8. A training system for a calibration result generation model, wherein, the system includes: A data acquisition module, configured to acquire a motor key design parameter sequence with multiple labels; each label represents a motor calibration result; A basic feature extraction module, configured to input the motor key design parameter sequence into the feature extraction network of the calibration result generation model to extract multiple basic features in the motor key design parameter sequence; each basic feature corresponds to a motor calibration result; A specific feature extraction module, configured to input each basic feature into the expert network corresponding to each basic feature in the calibration result generation model, and extract specific features in each basic feature correspondingly; A general feature extraction module, configured to input all basic features into the shared network of the calibration result generation model to extract general features in all basic features; A prediction result acquisition module, configured to input the general features and the specific features of each basic feature into the task tower network corresponding to each basic feature in the calibration result generation model to generate a predicted calibration result corresponding to each basic feature; A parameter update module, configured to update the parameters of the calibration result generation model according to the difference degree between the predicted calibration result and the corresponding label, and obtain the trained calibration result generation model.

9. An electronic device, wherein: the electronic device includes: One or more processors; A storage device, configured to store one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device implements the training method of the calibration result generation model according to any one of claims 1 to 7.

10. A computer-readable storage medium, wherein, a computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer executes the training method of the calibration result generation model according to any one of claims 1 to 7.

Images