Motor efficiency operating condition classification method, device, equipment and medium

Abstract

The application provides a motor efficiency operating condition classification method, device, equipment and medium. The motor efficiency MAP graph to be classified is obtained, the motor efficiency MAP graph to be classified is input into an attention residual model, the classification result of the motor efficiency MAP graph to be classified output by the attention residual model is obtained, and the classification result of the motor efficiency MAP graph to be classified is used to determine the classification operating condition corresponding to the motor efficiency MAP graph to be classified. In the scheme of the application, the operating condition information can be extracted through the motor efficiency MAP graph. The attention residual model is fitted and verified based on a residual neural network and a loss function according to a standard operating condition MAP graph set in advance. The scheme reduces the time and cost in motor efficiency operating condition analysis, and can improve the accuracy of motor efficiency operating condition classification.

Description

Technical Field

[0001] This application relates to the field of electric motors, and more particularly to a method, apparatus, equipment, and medium for classifying electric motor efficiency operating conditions. Background Technology

[0002] With continuous economic development and the maturation of new energy vehicle technology, people's demand for electric vehicles is increasing. To improve the performance and stability of electric vehicle drive motor systems, continuous analysis of the motor's efficient operating conditions is necessary.

[0003] Currently, the analysis of the high-efficiency operating area of ​​a motor mainly relies on manually collecting data and calculating the proportion of high-efficiency zones. However, manual calculations are prone to errors and are time-consuming and labor-intensive. Another method is to calculate the high-efficiency operating area of ​​the motor based on color indexing for pixels with efficiencies greater than 80% in the motor efficiency map. However, this method lacks preliminary judgment logic for motor efficiency conditions and has low accuracy in identifying operating conditions in the motor efficiency map. Therefore, the current problem to be solved is to improve the accuracy of motor efficiency condition classification. Summary of the Invention

[0004] This application provides a method, apparatus, equipment, and medium for classifying motor efficiency operating conditions, in order to improve the accuracy of motor efficiency operating condition classification.

[0005] On one hand, this application provides a method for classifying motor efficiency operating conditions, including: acquiring a motor efficiency MAP map to be classified; the motor efficiency MAP map includes pixel data characterizing the relationship between motor speed and motor torque; inputting the motor efficiency MAP map to be classified into an attention residual model to obtain the classification result of the motor efficiency MAP map to be classified output by the attention residual model, the classification result including the similarity probability of the motor efficiency MAP map to be classified under various standard operating conditions; wherein, the attention residual model pre-fits and validates the standard operating condition MAP map set based on a residual neural network and a loss function; the standard operating condition MAP map set includes standard operating condition labels and multiple motor efficiency MAP maps corresponding to the standard operating condition labels; and determining the classification operating condition corresponding to the motor efficiency MAP map to be classified based on the classification result of the motor efficiency MAP map to be classified.

[0006] In one possible implementation, the method further includes: establishing an initial model based on a residual neural network, fitting the current model to a standard working condition MAP atlas, and validating the current model based on a loss function; if the validation fails, adjusting the parameters of the current model and returning to execute the steps of fitting the current model to a standard working condition MAP atlas and validating the current model based on a loss function, until the validation passes, thus obtaining the attention residual model.

[0007] In one possible implementation, the current model is fitted according to a standard operating condition MAP set, and the current model is validated based on a loss function. This includes: fitting the current model sequentially to each motor efficiency MAP in the standard operating condition MAP set, and determining whether the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP; if the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, then determining whether the number of fitting iterations of the motor efficiency MAP has reached a preset threshold; if the preset threshold is reached, the fitting ends and the validation passes; if the preset threshold is not reached, the current model continues to be fitted according to the motor efficiency MAP, and the process returns to determining whether the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, until the validation passes; if the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is not less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, then the validation fails.

[0008] In one possible implementation, the attention residual model includes: a first module for converting the pixel data of the input motor efficiency MAP into first feature data; a second module for performing convolutional layer feature extraction, max pooling layer feature extraction, and activation function layer feature activation on the first feature data to obtain second feature data; and a third module for performing residual layer feature extraction multiple times on the second feature data to obtain the output of the attention residual model.

[0009] In one possible implementation, residual layer feature extraction includes: extracting features from the input features using convolutional layers with kernels of a×a, b×b, and a×a, and then performing average pooling layer feature extraction to obtain residual layer average pooling features; where a and b are both positive integers, and a is less than b; extracting features from the residual layer average pooling features using a first-angle attention layer to obtain first-angle attention features; wherein, the first-angle attention layer feature extraction includes extracting the residual layer average pooling features using a global average pooling layer, followed by convolutional layer extraction with kernel of b×b, second-angle attention removal, convolutional layer extraction with kernel of b×b, and nonlinear function transformation, batch normalization transformation, and activation. The first angle attention feature is obtained by activating the live function layer feature; the second angle attention feature is obtained by extracting the residual layer average pooling feature through the second angle attention layer feature; the second angle attention feature extraction includes extracting the residual layer average pooling feature through a global average pooling layer, extracting it through a convolutional layer with a kernel of b×b, removing the first angle attention, extracting it through a convolutional layer with a kernel of b×b, and performing nonlinear function transformation, batch normalization transformation, and activation function layer feature activation to obtain the second angle attention feature; the first angle attention feature and the second angle attention feature are added together, and the added result is connected to the current input feature through a residual skip connection to obtain the output result of the current residual layer feature extraction.

[0010] On the other hand, this application provides a motor efficiency operating condition classification device, including: an acquisition module for acquiring a motor efficiency MAP map to be classified; the motor efficiency MAP map includes pixel data characterizing the relationship between motor speed and torque; a simulation module for inputting the motor efficiency MAP map to be classified into an attention residual model to obtain the classification result of the motor efficiency MAP map to be classified output by the attention residual model, the classification result including the similarity probability of the motor efficiency MAP map to be classified under various standard operating conditions; wherein, the attention residual model pre-fits and validates the standard operating condition MAP map set based on a residual neural network and a loss function; the standard operating condition MAP map set includes standard operating condition labels and multiple motor efficiency MAP maps corresponding to the standard operating condition labels; and a classification module for determining the classification operating condition corresponding to the motor efficiency MAP map to be classified based on the classification result of the motor efficiency MAP map to be classified.

[0011] In one possible implementation, the device further includes a training module, which is used to: establish an initial model based on the residual neural network, fit the current model according to the standard working condition MAP set, and validate the current model based on the loss function; if the validation fails, the current model is tuned and the steps of fitting the current model according to the standard working condition MAP set and validating the current model based on the loss function are returned to be executed until the validation passes, thus obtaining the attention residual model.

[0012] In one possible implementation, the training module is used to: sequentially fit the current model to each motor efficiency MAP in the standard operating condition MAP set, and determine whether the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP; if the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, then determine whether the number of fitting iterations of the motor efficiency MAP has reached a preset threshold; if the preset threshold is reached, then the fitting ends and the verification is passed; if the preset threshold is not reached, then the current model continues to be fitted to the motor efficiency MAP, and the process returns to determine whether the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, until the verification is passed; if the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is not less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, then the verification fails.

[0013] On the other hand, this application provides an electronic device, including: a processor, and a memory communicatively connected to the processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement the above method.

[0014] On the other hand, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the above-described method.

[0015] The motor efficiency operating condition classification method, apparatus, equipment, and medium provided in this application involve acquiring a motor efficiency MAP (Motor Efficiency Map) to be classified, inputting the MAP into an attention residual model, obtaining the classification result of the MAP output by the attention residual model, and determining the corresponding operating condition based on the classification result. In this application's scheme, operating condition information can be extracted from the motor efficiency MAP. The attention residual model pre-fits and validates the MAP based on a standard operating condition MAP set using a residual neural network and a loss function. This scheme reduces the time and cost in motor efficiency operating condition analysis and improves the accuracy of motor efficiency operating condition classification. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] Figure 1 The flowchart of the motor efficiency condition classification method provided in Embodiment 1 is illustrated in the figure below.

[0018] Figure 2 The diagram above illustrates a flowchart of another motor efficiency condition classification method provided in Embodiment 1.

[0019] Figure 3 The diagram above illustrates a flowchart of another motor efficiency condition classification method provided in Embodiment 1.

[0020] Figure 4 The diagram above illustrates a flowchart of another motor efficiency condition classification method provided in Embodiment 1.

[0021] Figure 5 The diagram above illustrates a flowchart of another motor efficiency condition classification method provided in Embodiment 1.

[0022] Figure 6 The diagram above illustrates the structure of the motor efficiency condition classification device provided in Embodiment 2.

[0023] Figure 7 The diagram below illustrates the structure of the electronic device provided in Embodiment 3.

[0024] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0025] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0026] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning. The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to be omnipresent but not exclusive. For example, a product or device that comprises a series of components is not necessarily limited to those components that are explicitly listed, but may include other components that are not explicitly listed or that are inherent to such products or devices. The term "module" as used in this application refers to any known or subsequently developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and / or software code capable of performing the functions associated with that element.

[0027] With the continuous development of new energy vehicle technology and the increasing demand for energy conservation and environmental protection, electric vehicles are becoming increasingly popular in the automotive market. Motor efficiency, the ratio of the motor's output shaft power to its input electrical power, is a crucial indicator for evaluating motors. During the performance testing phase of electric vehicles, continuous analysis of the motor's efficient operating conditions is necessary to improve the performance and stability of the electric vehicle drive motor system.

[0028] Currently, the analysis of high-efficiency operating conditions of drive motors mainly relies on manually collecting data and calculating the proportion of the high-efficiency zone. However, the accuracy of manual calculation depends on the number of operating points used in the experiment, leading to significant errors and high time and labor costs. Another approach is to calculate the high-efficiency operating area of ​​the motor based on color indexing for pixels with efficiencies greater than 80% in the motor efficiency map, obtaining the high-efficiency operating area of ​​the motor after curve fitting of irregular regions. However, this approach lacks preliminary judgment logic for motor efficiency conditions, resulting in low accuracy in motor calibration analysis and low recognition accuracy of operating conditions in the motor efficiency map. Therefore, the current problem to be solved is to improve the accuracy of motor efficiency condition classification.

[0029] The technical content provided in this application aims to solve the aforementioned technical problems in related technologies. In the motor efficiency operating condition classification method, apparatus, equipment, and medium of this application, a motor efficiency MAP map to be classified is obtained, and then input into an attention residual model to obtain the classification result of the motor efficiency MAP map output by the attention residual model. Based on the classification result of the motor efficiency MAP map, the corresponding operating condition can be determined. In the scheme of this application, operating condition information can be extracted through the motor efficiency MAP map. The attention residual model pre-fits and validates the model based on a residual neural network and loss function using a standard operating condition MAP map set. This scheme reduces the time and cost in motor efficiency operating condition analysis and improves the accuracy of motor efficiency operating condition classification.

[0030] The technical solutions of this application will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. In the description of this application, unless otherwise expressly specified and limited, the terms should be broadly understood within the art. The embodiments of this application will now be described with reference to the accompanying drawings.

[0031] Example 1

[0032] Figure 1 The example illustrates a flowchart of a motor efficiency condition classification method. The executing entity in this example can be a motor efficiency condition classification device, such as... Figure 1 As shown, the method includes:

[0033] Step 101: Obtain the motor efficiency MAP map to be classified; the motor efficiency MAP map includes pixel data representing the relationship between motor speed and motor torque;

[0034] Step 102: Input the motor efficiency MAP map to be classified into the attention residual model to obtain the classification result of the motor efficiency MAP map to be classified output by the attention residual model. The classification result includes the similarity probability of the motor efficiency MAP map to be classified under each standard operating condition. The attention residual model is pre-fitted and validated based on the standard operating condition MAP map set using a residual neural network and loss function. The standard operating condition MAP map set includes standard operating condition labels and multiple motor efficiency MAP maps corresponding to the standard operating condition labels.

[0035] Step 103: Based on the classification results of the motor efficiency MAP diagram to be classified, determine the classification operating conditions corresponding to the motor efficiency MAP diagram to be classified.

[0036] In practical applications, the main body executing this method can be a motor efficiency condition classification device, which can be implemented in various ways. For example, it can be implemented through a computer program, such as application software; or it can be implemented as a medium storing relevant computer programs, such as a USB flash drive or cloud drive; or it can be implemented through a physical device that integrates or installs relevant computer programs, such as a chip.

[0037] In this example, a motor efficiency MAP (Motor Efficiency Map) is obtained for classification. This MAP includes pixel data representing the relationship between motor speed and torque. Specifically, the operating data of the motor to be classified can be obtained through calibration data. Based on the obtained operating data and the principle of contour plotting, a motor efficiency MAP is drawn using computer graphics software. In practical applications, the motor torque and efficiency corresponding to different speed gradients can be obtained according to a pre-set motor speed gradient, serving as the operating data for the motor to be classified. After drawing the motor efficiency MAP based on the obtained operating data, image smoothing processing can be performed on the MAP. The motor efficiency MAP, which includes pixel data representing the relationship between motor speed and torque, can be a two-dimensional color / grayscale MAP or a three-dimensional color / grayscale MAP.

[0038] After obtaining the motor efficiency MAP map to be classified, it is input into the attention residual model to obtain the classification result of the motor efficiency MAP map output by the attention residual model. The classification result includes the similarity probability of the motor efficiency MAP map under various standard operating conditions. In practical applications, the motor efficiency MAP map to be classified can be initially classified based on the engineer's experience, and an attention residual model can be built based on multiple standard operating conditions corresponding to the initial classification. In this example, the attention residual model is pre-fitted and validated based on the standard operating condition MAP map set using a residual neural network and a loss function. In practical applications, the residual neural network can be different types of residual neural networks, such as ResNetV2, Wider-ResNet, and Dilated-ResNet; the loss function can be a distance-based loss function such as the mean squared error loss function, or a probability distribution-based loss function such as the cross-entropy loss function. The standard operating condition MAP atlas includes standard operating condition labels and multiple motor efficiency MAPs corresponding to those labels. To improve the performance of the attention residual model, a preset MAP threshold can be set, ensuring that the number of motor efficiency MAPs corresponding to each standard operating condition label exceeds this threshold. In practical applications, the preset threshold can be 100 or 1000. After obtaining the classification results of the motor efficiency MAPs to be classified, the corresponding operating condition is determined based on these results. Specifically, the standard operating condition with the highest similarity probability among the motor efficiency MAPs to be classified can be used as the classification operating condition, or the standard operating condition with the highest probability after weighted averaging of multiple classification results can be used as the classification operating condition. Optionally, after classification, the motor efficiency MAPs to be classified and the classification operating conditions can be included as part of the standard operating condition MAP atlas to iterate the attention residual model and improve the accuracy of motor efficiency operating condition classification.

[0039] The motor efficiency operating condition classification method provided in this example obtains a motor efficiency MAP (Motor Efficiency Map) to be classified, inputs it into an attention residual model, and obtains the classification result of the motor efficiency MAP output by the attention residual model. Based on the classification result of the motor efficiency MAP, the corresponding operating condition can be determined. In this application's scheme, operating condition information can be extracted from the motor efficiency MAP. The attention residual model pre-fits and validates the standard operating condition MAP set based on a residual neural network and a loss function. This scheme reduces the time and cost in motor efficiency operating condition analysis and improves the accuracy of motor efficiency operating condition classification.

[0040] To ensure the realism and accuracy of the attention residual model, it is necessary to fit the model to a historically labeled atlas, and then determine the attention residual model after fitting. As an example, Figure 2 An exemplary flowchart of a motor efficiency operating condition classification method is shown. Based on any example, the motor efficiency operating condition classification method further includes:

[0041] Step 201: Establish an initial model based on the residual neural network, fit the current model according to the standard working condition MAP atlas, and validate the current model based on the loss function;

[0042] Step 202: If the verification fails, the parameters of the current model are tuned and the process is returned to perform the steps of fitting the current model according to the standard working condition MAP atlas and verifying the current model based on the loss function, until the verification passes and the attention residual model is obtained.

[0043] In this example, the standard operating condition MAP atlas can be stored on a server or a storage medium locally connected to the motor efficiency operating condition classification device. The standard operating condition MAP atlas includes standard operating condition labels and multiple motor efficiency MAP maps corresponding to those labels. For example, the standard operating condition labels can be: large proportion of motor high efficiency with no jumps, large proportion of motor high efficiency with jumps, large motor high efficiency with an upper limit threshold, and small proportion of motor high efficiency. Correspondingly, each standard operating condition label has multiple motor efficiency MAP maps. When fitting the attention residual model, the motor efficiency MAP map and its corresponding standard operating condition label are input into the attention residual model, and the current model is validated based on a loss function. The loss function represents the error value between the output of the attention residual model and the input standard operating condition label. In practical applications, the magnitude of the error value can be used to determine whether the validation passes. If the validation fails, the current model is tuned, and the process of fitting the current model according to the standard operating condition MAP atlas and validating the current model based on the loss function is repeated until the validation passes, resulting in the attention residual model. The parameters tuned include weights in the attention residual model, such as weights for convolutional layers, fully connected layers, and batch processing. In this example, the attention residual model is obtained upon successful validation. This model includes the model structure and multiple weights determined after fitting. The method provided in this example fits the model to a standard operating condition MAP dataset containing standard operating condition labels and multiple motor efficiency MAP maps corresponding to those labels, and tunes the model's parameters according to the loss function. This improves the accuracy and reliability of the attention residual model's classification.

[0044] To improve the applicability and training efficiency of attention residual models, validation criteria need to be established. As an example, Figure 3 An exemplary flowchart of a motor efficiency operating condition classification method is shown. Based on any example, step 201 fits the current model according to the standard operating condition MAP atlas and validates the current model based on the loss function. Specifically, it may include:

[0045] Step 301: Fit the current model to each motor efficiency MAP in the standard operating condition MAP set in turn, and determine whether the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP.

[0046] Step 302: If the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, then determine whether the number of fittings of the motor efficiency MAP has reached a preset threshold; if it has reached the preset threshold, then end the fitting and pass the verification; if it has not reached the preset threshold, then continue to fit the current model according to the motor efficiency MAP, and return to determine whether the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, until the verification passes.

[0047] Step 303: If the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is not less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, then the verification fails.

[0048] In this example, the mapping value of the loss function represents the error between the output of the motor efficiency MAP in the model and the standard operating condition label of the motor efficiency MAP in the standard operating condition MAP set. By limiting the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP to be less than the mapping value of the loss function corresponding to the previous fitting, the accuracy and efficiency of model fitting can be improved. Correspondingly, by setting a preset threshold for the number of fitting iterations of the motor efficiency MAP, the feature utilization of the motor efficiency MAP can be improved while avoiding overfitting. The method in this example, by setting validation conditions during the model fitting process, can improve the efficiency and accuracy of training and fitting the attention residual model.

[0049] As yet another example, attention residual models include:

[0050] The first module is used to convert the pixel data of the input motor efficiency MAP into first feature data;

[0051] The second module is used to extract features from the first feature data through convolutional layer features, max pooling layer features, and activation function layer features to obtain the second feature data.

[0052] The third module is used to extract residual layer features from the second feature data multiple times to obtain the output of the attention residual model.

[0053] In the first module, the pixel data of the input motor efficiency MAP is converted into first feature data, which can perform numerical conversion on the image information. Taking a two-dimensional color motor efficiency MAP as an example, the RGB color image can be converted into three images through the red, green and blue channels, and the corresponding values ​​are assigned according to the color saturation under each channel image to obtain the first feature data composed of three matrix data.

[0054] In the second module, the first feature data is extracted using convolutional layer features, max pooling layer features, and activation function layer features to obtain the second feature data. Specifically, the convolutional layer features extraction of the first feature data involves performing multiple convolutional layer features with an n×n kernel and a stride of m; where n is a positive integer and m is a positive integer greater than 1. For example, n can be 3 and m can be 2. Convolutional layer features extraction of the first feature data allows for the extraction of image features while preserving information from adjacent image patches in the two-dimensional image. Multiple convolutions also improve the learning ability of the attention residual model. Max pooling layer features remove redundant information and reduce computational cost. Activation function layer features perform non-linear transformations on the feature data, activating the feature information. In practical applications, Tanh, ReLU, or Leaky ReLU can be used as activation functions for the activation function layer.

[0055] In the third module, the second feature data undergoes multiple residual layer feature extractions to obtain the output of the attention residual model. In practical applications, 16 residual layer feature extractions can be performed on the second feature data to obtain the output of the attention residual model. Multiple residual layer feature extractions enable deep learning on the second layer feature data. The residual skip connections in the residual layers avoid the gradient vanishing problem and further improve the deep learning performance of the attention residual model.

[0056] To improve the fitting ability of the attention residual model, attention layers from different angles can be introduced to extract features from the input features in different dimensions. As an example, Figure 4 An exemplary flowchart of a motor efficiency condition classification method is shown. Based on any example, residual layer feature extraction may specifically include:

[0057] Step 401: Extract features from the input features using convolutional layers with kernels of a×a, b×b, and a×a, and extract features from average pooling layers to obtain residual average pooling features; where a and b are both positive integers, and a is less than b.

[0058] Step 402: Extract the residual layer average pooling features using the first angle attention layer to obtain the first angle attention features; wherein, the first angle attention layer feature extraction includes extracting the residual layer average pooling features using the global average pooling layer, extracting the features using a convolutional layer with a kernel of b×b, removing the second angle attention, extracting the features using a convolutional layer with a kernel of b×b, and performing nonlinear function transformation, batch normalization transformation, and activation function layer feature activation to obtain the first angle attention features;

[0059] Step 403: Extract second-angle attention features from the residual layer average pooling features to obtain second-angle attention features; wherein, the second-angle attention feature extraction includes extracting the residual layer average pooling features from the global average pooling layer, extracting from the convolutional layer with a kernel of b×b, removing the first-angle attention, extracting from the convolutional layer with a kernel of b×b, and performing nonlinear function transformation, batch normalization transformation, and activation function layer feature activation to obtain second-angle attention features;

[0060] Step 404: Add the first angle attention features and the second angle attention features together, and perform a residual skip connection between the added result and the current input features to obtain the output result of the current residual layer feature extraction.

[0061] In this example, the input features are extracted using convolutional layers with kernels of a×a, b×b, and a×a. This yields features with varying degrees of convolution and improves the non-linearity of the convolutional layer feature extraction. In practice, 'a' can be 1 and 'b' can be 3. Global average pooling layers in the first or second angle attention layer feature extraction are beneficial for subsequent model outputs. Extraction using a b×b convolutional layer avoids the use of identical kernels in upper and lower convolutional layers. The second angle extraction yields features of the first angle dimension, and vice versa. Further extraction using a b×b convolutional layer facilitates subsequent non-linear function transformations, which can be achieved using the sigmoid function in practice. Batch normalization facilitates subsequent feature addition, accelerates model convergence, mitigates gradient vanishing, and improves non-linear expressiveness. An activation function layer activates the batch-normalized features. In practical applications, the first angle can be the length dimension of the motor efficiency MAP, and the second angle can be the width dimension. After obtaining the attention features from the first and second angles, the first and second angle attention features are added together, and the added result is joined with the first feature data using a residual skip connection to obtain the output result of the current residual layer feature extraction. Specifically, feature addition can merge feature data from different angles; joining the added result with the first feature data using a residual skip connection can weight the residual layer feature extraction, thereby improving the model's fitting ability.

[0062] The following formula exemplifies the residual layer feature extraction process described above:

[0063] F R =avpool(F a*a (F b*b (F a*a (x))))

[0064] Where x is the input feature for this operation, F a*a F represents feature extraction from a convolutional layer with an a×a kernel. b*b F represents feature extraction from a convolutional layer with a kernel size of b×b, avpool represents feature extraction from an average pooling layer, and F... R This represents the average pooling characteristic of the residual layer.

[0065]

[0066]

[0067] Among them, F RFor residual layer average pooling features, AvPool represents global average pooling layer extraction, F b*b This represents feature extraction from a convolutional layer with a kernel size of b×b. This represents the removal of second-angle attention. δ represents first-angle attention removal, δ represents nonlinear function transformation, and F BM Representative batch normalization transformation, F AF F represents the feature activation of the activation function layer. D1 F represents the first-person attention characteristic. D2 Second-angle attention characteristics.

[0068] F D =F D1 +F D2

[0069] F out =F D +x

[0070] Among them, F D1 F represents the first-person attention characteristic. D2 Second-angle attention characteristics, F D F represents the sum of the first-angle attention features and the second-angle attention features, where x is the input feature for this operation. out This is the output of the residual layer feature extraction. The method in this example improves the fitting ability of the attention residual model by introducing attention layers at different angles to extract features from the input features in different dimensions.

[0071] Combining any of the above examples, Figure 5 The diagram illustrates a flowchart of a method for classifying motor efficiency operating conditions. Figure 5 As shown, the motor efficiency MAP input to the attention residual model is transformed into first feature data by the first module of the attention residual model; the first feature data is transformed into second feature data by the convolutional layer, max pooling layer, and activation function layer of the second module of the attention residual model; the second feature data is then extracted by two residual layer features by the third module of the attention residual model, and the output result of the attention residual model is output. Specifically, in each residual layer feature extraction, the input features are processed by a convolutional layer and an average pooling layer to obtain residual layer average pooling features; these residual layer average pooling features are then passed through a first angle attention layer and a second angle attention layer to obtain corresponding first angle attention features and second angle attention features; the first angle attention features and the second angle attention features are summed, and the result of the feature sum is joined with the input features using a residual jump connection to obtain the output of the current residual layer feature extraction.

[0072] This embodiment uses a neural network model for working condition identification, which reduces the time and cost of analysis and improves computational efficiency compared to manual working condition identification schemes. The model built based on residual neural networks can extract attention feature information from multiple angles, avoiding the loss of feature information from different angles and improving the accuracy of working condition classification. By constructing multiple residual layer feature extractions, the neural network model has a deeper network layer, stronger learning ability, and fewer parameters.

[0073] In the motor efficiency operating condition classification method provided in this embodiment, the motor efficiency MAP map to be classified is obtained, and then input into the attention residual model to obtain the classification result of the motor efficiency MAP map output by the attention residual model. Based on the classification result of the motor efficiency MAP map, the corresponding operating condition can be determined. In the scheme of this application, operating condition information can be extracted through the motor efficiency MAP map. The attention residual model is pre-fitted and validated based on a standard operating condition MAP map set using a residual neural network and a loss function. This scheme reduces the time and cost in motor efficiency operating condition analysis and can improve the accuracy of motor efficiency operating condition classification.

[0074] Example 2

[0075] Figure 6 The diagram above illustrates the structure of the motor efficiency condition classification device provided in Embodiment 2 of this application. Figure 6 As shown, the device includes:

[0076] The acquisition module 21 is used to acquire the motor efficiency MAP map to be classified; the motor efficiency MAP map includes pixel data representing the relationship between motor speed and torque;

[0077] The simulation module 22 is used to input the motor efficiency MAP map to be classified into the attention residual model to obtain the classification result of the motor efficiency MAP map to be classified output by the attention residual model. The classification result includes the similarity probability of the motor efficiency MAP map to be classified under each standard operating condition. The attention residual model is pre-fitted and validated based on the standard operating condition MAP map set using a residual neural network and a loss function. The standard operating condition MAP map set includes standard operating condition labels and multiple motor efficiency MAP maps corresponding to the standard operating condition labels.

[0078] The classification module 23 is used to determine the classification condition corresponding to the motor efficiency MAP diagram to be classified based on the classification result of the motor efficiency MAP diagram to be classified.

[0079] In practical applications, there are several ways to implement this motor efficiency condition classification device. For example, it can be implemented through a computer program, such as application software; or it can be implemented as a medium storing the relevant computer program, such as a USB flash drive or cloud drive; or it can be implemented through a physical device that integrates or installs the relevant computer program, such as a chip.

[0080] In this example, a motor efficiency MAP (Motor Efficiency Map) is obtained for classification. This MAP includes pixel data representing the relationship between motor speed and torque. Specifically, the operating data of the motor to be classified can be obtained through calibration data. Based on the obtained operating data and the principle of contour plotting, a motor efficiency MAP is drawn using computer graphics software. In practical applications, the motor torque and efficiency corresponding to different speed gradients can be obtained according to a pre-set motor speed gradient, serving as the operating data for the motor to be classified. After drawing the motor efficiency MAP based on the obtained operating data, image smoothing processing can be performed on the MAP. The motor efficiency MAP, which includes pixel data representing the relationship between motor speed and torque, can be a two-dimensional color / grayscale MAP or a three-dimensional color / grayscale MAP.

[0081] After obtaining the motor efficiency MAP map to be classified, it is input into the attention residual model to obtain the classification result of the motor efficiency MAP map output by the attention residual model. The classification result includes the similarity probability of the motor efficiency MAP map under various standard operating conditions. In practical applications, the motor efficiency MAP map to be classified can be initially classified based on the engineer's experience, and an attention residual model can be built based on multiple standard operating conditions corresponding to the initial classification. In this example, the attention residual model is pre-fitted and validated based on the standard operating condition MAP map set using a residual neural network and a loss function. In practical applications, the residual neural network can be different types of residual neural networks, such as ResNetV2, Wider-ResNet, and Dilated-ResNet; the loss function can be a distance-based loss function such as the mean squared error loss function, or a probability distribution-based loss function such as the cross-entropy loss function. The standard operating condition MAP atlas includes standard operating condition labels and multiple motor efficiency MAPs corresponding to those labels. To improve the performance of the attention residual model, a preset MAP threshold can be set, ensuring that the number of motor efficiency MAPs corresponding to each standard operating condition label exceeds this threshold. In practical applications, the preset threshold can be 100 or 1000. After obtaining the classification results of the motor efficiency MAPs to be classified, the corresponding operating condition is determined based on these results. Specifically, the standard operating condition with the highest similarity probability among the motor efficiency MAPs to be classified can be used as the classification operating condition, or the standard operating condition with the highest probability after weighted averaging of multiple classification results can be used as the classification operating condition. Optionally, after classification, the motor efficiency MAPs to be classified and the classification operating conditions can be included as part of the standard operating condition MAP atlas to iterate the attention residual model and improve the accuracy of motor efficiency operating condition classification.

[0082] The motor efficiency operating condition classification device provided in this example acquires a motor efficiency MAP (Motor Efficiency Map) to be classified, inputs the MAP into an attention residual model, obtains the classification result of the MAP output by the attention residual model, and determines the corresponding operating condition based on the classification result. In this application's solution, operating condition information can be extracted from the motor efficiency MAP. The attention residual model pre-fits and validates the MAP based on a standard operating condition MAP set using a residual neural network and a loss function. This solution reduces the time and cost in motor efficiency operating condition analysis and improves the accuracy of motor efficiency operating condition classification.

[0083] In one example, the device also includes a training module 24, which is used for:

[0084] An initial model is established based on a residual neural network. The current model is then fitted using a standard working condition MAP dataset, and the model is validated based on a loss function.

[0085] If the validation fails, the parameters of the current model are tuned and the process is repeated to fit the current model to the standard working condition MAP atlas and validate the current model based on the loss function until the validation passes, thus obtaining the attention residual model.

[0086] In this example, the standard operating condition MAP atlas can be stored on a server or a storage medium locally connected to the motor efficiency operating condition classification device. The standard operating condition MAP atlas includes standard operating condition labels and multiple motor efficiency MAP maps corresponding to those labels. For example, the standard operating condition labels can be: large proportion of motor high efficiency with no jumps, large proportion of motor high efficiency with jumps, large motor high efficiency with an upper limit threshold, and small proportion of motor high efficiency. Correspondingly, each standard operating condition label has multiple motor efficiency MAP maps. When fitting the attention residual model, the motor efficiency MAP map and its corresponding standard operating condition label are input into the attention residual model, and the current model is validated based on a loss function. The loss function represents the error value between the output of the attention residual model and the input standard operating condition label. In practical applications, the magnitude of the error value can be used to determine whether the validation passes. If the validation fails, the current model is tuned, and the process of fitting the current model according to the standard operating condition MAP atlas and validating the current model based on the loss function is repeated until the validation passes, resulting in the attention residual model. The parameters tuned include weights in the attention residual model, such as weights for convolutional layers, fully connected layers, and batch processing. In this example, the attention residual model is obtained upon successful validation. This model includes the model structure and multiple weights determined after fitting. The device provided in this example fits the model to a standard operating condition MAP dataset containing standard operating condition labels and multiple motor efficiency MAP maps corresponding to those labels, and tunes the model's parameters according to the loss function, thereby improving the accuracy and reliability of the attention residual model's classification.

[0087] In one example, training module 24 is specifically used for:

[0088] The current model is fitted sequentially to each motor efficiency MAP in the standard operating condition MAP set, and it is determined whether the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP.

[0089] If the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, then it is determined whether the number of fitting attempts of the motor efficiency MAP has reached a preset threshold. If it has reached the preset threshold, the fitting ends and the verification is passed. If it has not reached the preset threshold, the current model continues to be fitted based on the motor efficiency MAP, and the process returns to determine whether the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, until the verification is passed.

[0090] If the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is not less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, then the verification fails.

[0091] In this example, the mapping value of the loss function represents the error between the output of the motor efficiency MAP in the model and the standard operating condition label of the motor efficiency MAP in the standard operating condition MAP set. By limiting the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP to be less than the mapping value of the loss function corresponding to the previous fitting, the accuracy and efficiency of model fitting can be improved. Correspondingly, by setting a preset threshold for the number of fitting iterations of the motor efficiency MAP, the feature utilization of the motor efficiency MAP can be improved while avoiding overfitting. The device in this example, by setting validation conditions during the model fitting process, can improve the efficiency and accuracy of training and fitting the attention residual model.

[0092] In one example, the attention residual model obtained by training module 24 includes: a first module for converting the pixel data of the input motor efficiency MAP into first feature data; a second module for performing convolutional layer feature extraction, max pooling layer feature extraction, and activation function layer feature activation on the first feature data to obtain second feature data; and a third module for performing multiple residual layer feature extractions on the second feature data to obtain the output result of the attention residual model.

[0093] In the first module, the pixel data of the input motor efficiency MAP is converted into first feature data, which can perform numerical conversion on the image information. Taking a two-dimensional color motor efficiency MAP as an example, the RGB color image can be converted into three images through the red, green and blue channels, and the corresponding values ​​are assigned according to the color saturation under each channel image to obtain the first feature data composed of three matrix data.

[0094] In the second module, the first feature data is extracted using convolutional layer features, max pooling layer features, and activation function layer features to obtain the second feature data. Specifically, the convolutional layer features extraction of the first feature data involves performing multiple convolutional layer features with an n×n kernel and a stride of m; where n is a positive integer and m is a positive integer greater than 1. For example, n can be 3 and m can be 2. Convolutional layer features extraction of the first feature data allows for the extraction of image features while preserving information from adjacent image patches in the two-dimensional image. Multiple convolutions also improve the learning ability of the attention residual model. Max pooling layer features remove redundant information and reduce computational cost. Activation function layer features perform non-linear transformations on the feature data, activating the feature information. In practical applications, Tanh, ReLU, or Leaky ReLU can be used as activation functions for the activation function layer.

[0095] In the third module, the second feature data undergoes multiple residual layer feature extractions to obtain the output of the attention residual model. In practical applications, 16 residual layer feature extractions can be performed on the second feature data to obtain the output of the attention residual model. Multiple residual layer feature extractions enable deep learning on the second layer feature data. The residual skip connections in the residual layers avoid the gradient vanishing problem and further improve the deep learning performance of the attention residual model.

[0096] In one example, training module 24 is specifically used for: in residual layer feature extraction,

[0097] The input features are extracted using convolutional layers with kernels of a×a, b×b, and a×a, and then extracted using average pooling layers to obtain residual average pooling features; where a and b are both positive integers, and a is less than b.

[0098] The residual layer average pooling features are subjected to first-angle attention layer feature extraction to obtain first-angle attention features. The first-angle attention layer feature extraction includes extracting the residual layer average pooling features through a global average pooling layer, extracting them through a convolutional layer with a kernel of b×b, removing second-angle attention, extracting them through a convolutional layer with a kernel of b×b, and performing nonlinear function transformation, batch normalization transformation, and activation function layer feature activation to obtain first-angle attention features.

[0099] The residual layer average pooling features are subjected to second-angle attention layer feature extraction to obtain second-angle attention features. The second-angle attention layer feature extraction includes extracting the residual layer average pooling features through a global average pooling layer, extracting them through a convolutional layer with a kernel of b×b, removing the first-angle attention, extracting them through a convolutional layer with a kernel of b×b, and performing nonlinear function transformation, batch normalization transformation, and activation function layer feature activation to obtain second-angle attention features.

[0100] The first-angle attention features and the second-angle attention features are added together, and the added result is connected to the current input features via a residual skip connection to obtain the output result of the current residual layer feature extraction.

[0101] In this example, the input features are extracted using convolutional layers with kernels of a×a, b×b, and a×a. This yields features with varying degrees of convolution and improves the non-linearity of the convolutional layer feature extraction. In practice, 'a' can be 1 and 'b' can be 3. Global average pooling layers in the first or second angle attention layer feature extraction are beneficial for subsequent model outputs. Extraction using a b×b convolutional layer avoids the use of identical kernels in upper and lower convolutional layers. The second angle extraction yields features of the first angle dimension, and vice versa. Further extraction using a b×b convolutional layer facilitates subsequent non-linear function transformations, which can be achieved using the sigmoid function in practice. Batch normalization facilitates subsequent feature addition, accelerates model convergence, mitigates gradient vanishing, and improves non-linear expressiveness. An activation function layer activates the batch-normalized features. In practical applications, the first angle can be the length dimension of the motor efficiency MAP, and the second angle can be the width dimension. After obtaining the attention features from the first and second angles, the first and second angle attention features are added together, and the added result is joined with the first feature data using a residual skip connection to obtain the output result of the current residual layer feature extraction. Specifically, feature addition can merge feature data from different angles; joining the added result with the first feature data using a residual skip connection can weight the residual layer feature extraction, thereby improving the model's fitting ability.

[0102] The following formula exemplifies the residual layer feature extraction process described above:

[0103] F R =avpool(F a*a (F b*b (F a*a (x))))

[0104] Where x is the input feature for this operation, F a*a F represents feature extraction from a convolutional layer with an a×a kernel. b*b F represents feature extraction from a convolutional layer with a kernel size of b×b, avpool represents feature extraction from an average pooling layer, and F... R This represents the average pooling characteristic of the residual layer.

[0105]

[0106]

[0107] Among them, F R For residual layer average pooling features, AvPool represents global average pooling layer extraction, F b*b This represents feature extraction from a convolutional layer with a kernel size of b×b. This represents the removal of second-angle attention. δ represents first-angle attention removal, δ represents nonlinear function transformation, and F BM Representative batch normalization transformation, F AF F represents the feature activation of the activation function layer. D1 F represents the first-person attention characteristic. D2 Second-angle attention characteristics.

[0108] F D =F D1 +F D2

[0109] F out =F D +x

[0110] Among them, F D1 F represents the first-person attention characteristic. D2 Second-angle attention characteristics, F D F represents the sum of the first-angle attention features and the second-angle attention features, where x is the input feature for this operation. out This is the output result of the residual layer feature extraction. The device in this example improves the fitting ability of the attention residual model by introducing attention layers at different angles to extract features of different dimensions from the input features.

[0111] The motor efficiency condition classification device provided in this embodiment acquires a motor efficiency MAP (Motor Efficiency Map) to be classified, inputs the MAP into an attention residual model, obtains the classification result of the MAP output by the attention residual model, and determines the corresponding operating condition based on the classification result. In this application's solution, operating condition information can be extracted from the motor efficiency MAP. The attention residual model pre-fits and validates the MAP based on a standard operating condition MAP set using a residual neural network and a loss function. This solution reduces the time and cost in motor efficiency condition analysis and improves the accuracy of motor efficiency condition classification.

[0112] Example 3

[0113] Figure 7 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of this application. The electronic device includes:

[0114] The electronic device includes a processor 291 and a memory 292; it may also include a communication interface 293 and a bus 294. The processor 291, memory 292, and communication interface 293 can communicate with each other via the bus 294. The communication interface 293 can be used for information transmission. The processor 291 can invoke logical instructions stored in the memory 292 to execute the methods described in the example above.

[0115] Furthermore, the logic instructions in the aforementioned memory 292 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0116] The memory 292, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the methods in the embodiments of this application. The processor 291 executes functional applications and data processing by running the software programs, instructions, and modules stored in the memory 292, that is, it implements the methods in the above method examples.

[0117] The memory 292 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 292 may include high-speed random access memory and may also include non-volatile memory.

[0118] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method in any of the embodiments.

[0119] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0120] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for classifying motor efficiency operating conditions, characterized in that, include: Obtain the motor efficiency MAP to be classified; The motor efficiency MAP includes pixel data characterizing the relationship between motor speed and motor torque; The motor efficiency MAP map to be classified is input into an attention residual model to obtain the classification result of the motor efficiency MAP map to be classified output by the attention residual model. The classification result includes the similarity probability of the motor efficiency MAP map to be classified under each standard operating condition. The attention residual model is pre-fitted and validated based on a residual neural network and a loss function according to the standard operating condition MAP map set. The standard operating condition MAP map set includes standard operating condition labels and multiple motor efficiency MAP maps corresponding to the standard operating condition labels. Based on the classification results of the motor efficiency MAP diagram to be classified, the corresponding classification operating conditions of the motor efficiency MAP diagram to be classified are determined.

2. The method according to claim 1, characterized in that, The method further includes: An initial model is established based on a residual neural network. The current model is then fitted according to the standard working condition MAP dataset, and the current model is validated based on the loss function. If the verification fails, the parameters of the current model are tuned and the process returns to the steps of fitting the current model according to the standard working condition MAP atlas and verifying the current model based on the loss function, until the verification passes and the attention residual model is obtained.

3. The method according to claim 2, characterized in that, The process of fitting the current model to the standard operating condition MAP atlas and validating the current model based on the loss function includes: The current model is fitted sequentially according to each motor efficiency MAP in the standard operating condition MAP set, and it is determined whether the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP. If the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, then it is determined whether the number of fitting attempts of the motor efficiency MAP has reached a preset threshold. If the preset threshold is reached, the fitting ends and the verification is passed. If the preset threshold is not reached, the current model continues to be fitted based on the motor efficiency MAP, and the process of determining whether the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP is repeated until the verification is passed. If the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is not less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, then the verification fails.

4. The method according to any one of claims 1 to 3, characterized in that, The attention residual model includes: The first module is used to convert the pixel data of the input motor efficiency MAP into first feature data; The second module is used to extract features from the first feature data by performing convolutional layer feature extraction, max pooling layer feature extraction, and activation function layer feature activation to obtain the second feature data. The third module is used to perform multiple residual layer feature extractions on the second feature data to obtain the output result of the attention residual model.

5. The method according to claim 4, characterized in that, The residual layer feature extraction includes: The input features are extracted using convolutional layers with kernels of a×a, b×b, and a×a, and then extracted using average pooling layers to obtain residual average pooling features; where a and b are both positive integers, and a is less than b. The residual layer average pooling features are subjected to first angle attention layer feature extraction to obtain first angle attention features; wherein, the first angle attention layer feature extraction includes extracting the residual layer average pooling features by a global average pooling layer, extracting by a convolutional layer with a kernel of b×b, removing second angle attention, extracting by a convolutional layer with a kernel of b×b, and performing nonlinear function transformation, batch normalization transformation, and activation function layer feature activation to obtain first angle attention features; The residual layer average pooling features are subjected to second-angle attention layer feature extraction to obtain second-angle attention features; wherein, the second-angle attention layer feature extraction includes extracting the residual layer average pooling features through a global average pooling layer, extracting through a convolutional layer with a kernel of b×b, removing first-angle attention, extracting through a convolutional layer with a kernel of b×b, and performing nonlinear function transformation, batch normalization transformation, and activation function layer feature activation to obtain second-angle attention features; The first angle attention feature and the second angle attention feature are added together, and the added result is connected to the current input feature through a residual skip connection to obtain the output result of the current residual layer feature extraction.

6. A motor efficiency condition classification device, characterized in that, include: The acquisition module is used to acquire the motor efficiency MAP map to be classified; The motor efficiency MAP includes pixel data characterizing the relationship between motor speed and torque; The simulation module is used to input the motor efficiency MAP map to be classified into an attention residual model to obtain the classification result of the motor efficiency MAP map to be classified output by the attention residual model. The classification result includes the similarity probability of the motor efficiency MAP map to be classified under various standard operating conditions. The attention residual model is pre-fitted and validated based on a residual neural network and a loss function according to a standard operating condition MAP map set. The standard operating condition MAP map set includes standard operating condition labels and multiple motor efficiency MAP maps corresponding to the standard operating condition labels. The classification module is used to determine the classification condition corresponding to the motor efficiency MAP to be classified based on the classification result of the motor efficiency MAP to be classified.

7. The apparatus according to claim 6, characterized in that, The device further includes a training module, the training module being used for: An initial model is established based on a residual neural network. The current model is then fitted according to the standard working condition MAP dataset, and the current model is validated based on the loss function. If the verification fails, the parameters of the current model are tuned and the process returns to the steps of fitting the current model according to the standard working condition MAP atlas and verifying the current model based on the loss function, until the verification passes and the attention residual model is obtained.

8. The apparatus according to claim 7, characterized in that, The training module is used for: The current model is fitted sequentially according to each motor efficiency MAP in the standard operating condition MAP set, and it is determined whether the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP. If the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, then it is determined whether the number of fittings of the motor efficiency MAP has reached a preset threshold. If the preset threshold is reached, the fitting ends and the verification is passed; if the preset threshold is not reached, the current model continues to be fitted according to the motor efficiency MAP, and the process returns to determine whether the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, until the verification is passed. If the mapping value of the loss function corresponding to the current fitting of the motor efficiency MAP is not less than the mapping value of the loss function corresponding to the previous fitting of the motor efficiency MAP, then the verification fails.

9. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement the method as described in any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 5.

Images