An online evaluation method and system for metering performance of off-board charger for electric vehicles

Abstract

The present invention relates to an online evaluation method and system for metering performance of an off-board charger for electric vehicles. The method comprises: S1, collecting power data of the charger and normalizing and multi-granularity segmentation; S2, performing mask operation on the segmented multi-granularity slice data; S3, inputting the power data into an online evaluation model, performing channel-by-channel convolution on each granularity slice data through each depth-separable convolution module, extracting features in the slice data, and then grouping multiple slice data according to the relationship between variables and non-variables, performing point-by-point convolution on the grouped features, and performing comparative learning; thereby training the online evaluation model through power data; S4, using the obtained online evaluation model to perform online evaluation on the charger data to be tested, thereby obtaining an evaluation result of the metering performance of the off-board charger for electric vehicles. The method and system can efficiently, accurately and conveniently perform online detection of the metering performance of the off-board charger for electric vehicles.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric vehicle charging equipment, and in particular relates to an online evaluation method and system for metering performance of an off-board charger for an electric vehicle. Background Art

[0002] At present, the electric vehicle industry is developing rapidly. As part of the charging facilities, off-board chargers play an important role in promoting the popularization of electric vehicles. In order to ensure the fairness of electricity trade settlement, it is necessary to calibrate the metering performance of off-board chargers.

[0003] Traditional non-on-board charger metrology verification usually uses standard verification devices as measurement standards and conducts on-site verification according to corresponding verification procedures. However, due to the large number of charging piles and complex working environments, this method requires a lot of manpower and is inefficient and costly. Summary of the invention

[0004] The object of the present invention is to provide a method and system for online evaluation of the metering performance of an off-board charger for an electric vehicle. The method and system can efficiently, accurately and conveniently perform online detection of the metering performance of an off-board charger for an electric vehicle.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: an online evaluation method for the metering performance of an off-board charger for an electric vehicle, comprising the following steps:

[0006] S1. Collect the power data of the charger over a period of time, and normalize and divide the power data into multiple granularities;

[0007] S2, performing a mask operation on the multi-granularity slice data after segmentation;

[0008] S3, inputting the power data after the mask operation into the online evaluation model, the online evaluation model includes multiple depth-separable convolution modules, firstly, performing channel-by-channel convolution on each granularity slice data through each depth-separable convolution module, extracting features in the slice data, then grouping the multiple slice data according to the relationship between variables and non-variables, performing point-by-point convolution on the grouped features, and performing comparative learning on slice data of different granularities; thereby training the online evaluation model through the power data to obtain the final online evaluation model;

[0009] S4. Use the obtained online evaluation model to perform online evaluation on the charger data to be tested, so as to obtain an evaluation result of the metering performance of the off-board charger of the electric vehicle.

[0010] Furthermore, in step S1, the collected power data includes the power error value of the charger, the output voltage of the charger, the battery state of charge value, the charging time and the battery temperature.

[0011] Furthermore, the specific implementation method of step S1 is:

[0012] Collect the energy error values ​​of N groups of chargers over a period of time E = { e 1, e 2, ..., e m}、Output voltage of charger V = { v 1, v 2, ..., v m}、Battery state of charge value C = { c 1, c 2, ..., c m}、Charging time D = { d 1, d 2, ..., d m} and battery temperature T = { t 1, t 2, ..., t m};

[0013] Normalize the collected power data and map all data to between 0 and 1 to eliminate the dimension effect. The method is as follows:

[0014]

[0015] in, x express E , V , C , D , T The elements in x min , x max They are E , V , C , D , T The minimum and maximum values ​​of the elements in;

[0016] Then, the normalized power data are segmented into multiple granularities to obtain , , , , ; The first granularity data segment set is , , , , ,in E i 1 , V i 1 , C i 1 , D i 1 , T i 1 The same length, i = 1, 2, ..., k ; Each power data has l The data fragments in different granularity data fragment sets have different lengths, and the lengths increase in sequence, representing that the data granularity is from fine to coarse.

[0017] Furthermore, the specific implementation method of step S2 is:

[0018] Randomly add data to a collection of data fragments of different granularities s The number of masks, s is an adjustable parameter; the position where the mask is added in the first granularity data segment set is represented as , the position where the mask is added in the second granularity data segment set is expressed as , in order to l The position where the mask is added in the set of granular data segments is represented as , , that is, the locations where masks are added in sets of data fragments of different granularities are different;

[0019] The mask operation is specifically: replacing the element value at a random position selected from a set of data fragments of different granularities with 0 to mask the original element value.

[0020] Furthermore, the specific implementation method of step S3 is:

[0021] The first granular power data , , , , Input to the first granularity depth-separable convolution module, and use a 1×M The convolution kernel performs channel-by-channel convolution:

[0022]

[0023] in, , , , , Indicates that the convolution kernel size is 1× M Output features of the convolution operation; W 1×M represents the size of the convolution kernel, b is bias; Conv is the convolution operation, φ is the ReLu activation function, and its expression is as follows:

[0024]

[0025] Then the channel-by-channel convolution output of the second granularity depth-wise separable convolution module is , , , , , in order to l The channel-by-channel convolution output of the granular depthwise separable convolution module is , , , , ; For the first granularity depth separable convolution module to the l The channel-by-channel convolution output of the granular depth-separable convolution module uses contrastive learning. Any univariate slice feature in the channel-by-channel convolution output of any granular depth-separable convolution module is selected as an anchor point, and the same univariate residual slice feature in the channel-by-channel convolution output of the same granular depth-separable convolution module is used as a positive sample. The same univariate slice feature in the channel-by-channel convolution output of different granular depth-separable convolution modules is used as a negative sample. The specific function form is as follows:

[0026]

[0027] in, S represents the specific category of the variable, sim() is the cosine similarity function, is the anchor point, is a positive sample, is a negative sample, r Indicates the granularity, x , y , z represents the ordinal number of the univariate slice in the channel-by-channel convolution output, , τ is the temperature parameter, M is the number of positive samples;

[0028] right , , , , According to the non-variable relationship, they are divided into P group, get feature group , , , , ; According to the relationship between variables, it is divided into Q group, get features , , h = 1, 2, ..., Q ; for features , , , , , j = 1,2, ..., P and Features Use the size of 1×1× N 1.1×1× N The convolution kernel of 2 performs point-by-point convolution to extract the periodic features within the same variable and the coupling features between different variables:

[0029]

[0030] in, F j out1_E , F j out1_V , F j out1_C , F j out1_D , F j out1_T , F h out1 is the output feature of point-by-point convolution; b 1. b 2 is bias;

[0031] Once again, the characteristics F j out1_E , F j out1_V ,F j out1_C , F j out1_D , F j out1_T According to the non-variable relationship, they are divided into P 1 group gets features F u out1_E , F u out1_V , F u out1_C , F u out1_D , F u out1_T , u = 1, 2, ..., P 1, and perform convolution with a kernel size of 1×1× M 1 point-by-point convolution to obtain features F u out2_E , F u out2_V , F u out2_C , F u out2_D , F u out2_T ; At the same time, once again F h out1 Get features by grouping according to variable relationships F v out1 , v = 1, 2, ..., Q 1, and perform convolution with a kernel size of 1×1× M 2 point-by-point convolution to obtain features F v out2 ;

[0032] Similarly, the output of the second granularity depthwise separable convolutional module is F u out2_E , F u out2_V , F uout2_C , F u out2_D , F u out2_T , F v out2 , in order to l The output of the granular depthwise separable convolution module is F u outl_E , F u outl_V , F u outl_C , F u outl _D , F u outl_T , F v outl ; Input the output of each granular depth-separable convolution module into different MLP networks to obtain the output ; Finally, the feature F out Input it into the MLP network again to get the final output result Y out ; The loss function of the output result is:

[0033]

[0034] in, N is the number of samples, Y n out is the true value of the charger power error, Estimating output value for the power error value of the online evaluation model;

[0035] The total loss function of the online evaluation model is:

[0036] .

[0037] The present invention also provides an online evaluation system for the metering performance of an off-board charger for an electric vehicle, comprising a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the above-mentioned method steps can be implemented.

[0038] Compared with the prior art, the present invention has the following beneficial effects: the present invention provides an online evaluation method and system for the metering performance of an electric vehicle non-on-board charger, the method and system collect the power data of the charger, perform multi-granularity segmentation and implement masking operations, and then input the deep separable convolution module to extract the intra-variable, inter-variable and periodic features, and input the extracted features into the MLP network to obtain the electric energy value estimation result, thereby realizing the online evaluation and detection of the metering performance of the electric vehicle non-on-board charger, with high detection accuracy and detection efficiency, convenient detection, no need to consume a lot of manpower and material resources, and low detection cost; the method uses multiple deep separable convolution modules to extract the internal features of a single variable data, the periodic features of the data and the coupling features between variables, which can reduce the number of parameters and the operation cost; at the same time, the idea of ​​comparative learning is introduced for slices of different granularities, so that the slice features with similar periodic characteristics are close to each other, while the slice features with different periodic characteristics are far away from each other, further strengthening the model's learning of the periodic characteristics in the features, and improving the accuracy of detection. Therefore, the present invention has strong practicality and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of the method implementation of the embodiment of the present invention. DETAILED DESCRIPTION

[0040] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0041] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0042] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0043] like Figure 1 As shown, this embodiment provides an online evaluation method for metering performance of an off-board charger for an electric vehicle, comprising the following steps:

[0044] S1. Collect the power data of the charger over a period of time, including the power error value of the charger, the output voltage of the charger, the battery state of charge value, the charging time and the battery temperature; then normalize and multi-granularize the power data.

[0045] Specifically, collect the energy error values ​​of N groups of chargers over a period of time E = { e 1, e 2, ..., e m}、Output voltage of charger V = { v 1, v 2, ..., v m}、Battery state of charge value C = { c 1, c 2, ..., c m}、Charging time D ={ d 1, d 2, ..., d m} and battery temperature T = { t 1, t 2, ..., t m}.

[0046] Normalize the collected power data and map all data to between 0 and 1 to eliminate the dimension effect. The method is as follows:

[0047]

[0048] in, x express E , V , C , D , T The elements in x min , x max They are E , V , C , D , T The minimum and maximum values ​​of the elements in .

[0049] Then, the normalized power data are segmented into multiple granularities to obtain , , , , ; The first granularity data segment set is , , , , ,in E i 1 , V i 1 , C i 1 , D i 1 , T i 1 The same length, i = 1, 2, ..., k ; Each power data has l The data fragments in different granularity data fragment sets have different lengths, and the lengths increase in sequence, representing that the data granularity is from fine to coarse.

[0050] S2. Perform a mask operation on the segmented multi-granularity slice data.

[0051] Specifically, randomly add data in a set of data segments of different granularities s The number of masks, s is an adjustable parameter; the position where the mask is added in the first granularity data segment set is represented as , the position where the mask is added in the second granularity data segment set is expressed as , in order to l The position where the mask is added in the set of granular data segments is represented as , , that is, the locations where masks are added in sets of data fragments of different granularities are different;

[0052] The masking operation is to replace the element values ​​at random positions selected from the data fragment sets of different granularities with 0 to mask the original element values. Since there is redundancy in the elements in the data fragment sets of different granularities, adding masks can increase the diversity of samples and improve the performance and generalization ability of the model.

[0053] S3. Inputting the power data after the mask operation into the online evaluation model, the online evaluation model includes multiple depth-separable convolution modules. First, each granularity slice data is convolved channel by channel through each depth-separable convolution module to extract features in the slice data. Then, the multiple slice data are grouped according to the relationship between variables and non-variables, and the grouped features are convolved point by point. Then, comparative learning is performed on slice data of different granularities; thereby, the online evaluation model is trained through the power data to obtain the final online evaluation model.

[0054] Specifically, the first granularity power data , , , , Input to the first granularity depth-separable convolution module, and use a 1× M The convolution kernel performs channel-by-channel convolution:

[0055]

[0056] in, , , , , Indicates that the convolution kernel size is 1× M Output features of the convolution operation; W 1×M represents the size of the convolution kernel, b is bias; Conv is the convolution operation, φ is the ReLu activation function, and its expression is as follows:

[0057]

[0058] Then the channel-by-channel convolution output of the second granularity depth-wise separable convolution module is , , , , , in order to l The channel-by-channel convolution output of the granular depthwise separable convolution module is , , , , ; For the first granularity depth separable convolution module to the l The channel-by-channel convolution output of the granular depth-separable convolution module uses contrastive learning. Any univariate slice feature in the channel-by-channel convolution output of any granular depth-separable convolution module is selected as an anchor point, and the same univariate residual slice feature in the channel-by-channel convolution output of the same granular depth-separable convolution module is used as a positive sample. The same univariate slice feature in the channel-by-channel convolution output of different granular depth-separable convolution modules is used as a negative sample. The specific function form is as follows:

[0059]

[0060] in, S represents the specific category of the variable, sim() is the cosine similarity function, is the anchor point, is a positive sample, is a negative sample, r Indicates the granularity, x , y , z represents the ordinal number of the univariate slice in the channel-by-channel convolution output, , τ is the temperature parameter, M is the number of positive samples; since slice features of different granularities have different time scales, and different time scales have different periodic characteristics, contrastive learning can reduce the distance between the anchor point and the positive sample and expand the distance between the anchor point and the negative sample, so that slice features with the same time scale are close to each other, and slice features with different time scales are far away from each other, that is, the periodic characteristics in slice features of different granularities can be distinguished from each other, and the learning of periodic characteristics in features by the online evaluation model can be strengthened.

[0061] right , , , , According to the non-variable relationship, they are divided into P group, get feature group , , , , ; According to the relationship between variables, it is divided into Q group, get features , , h = 1, 2, ..., Q ; for features , , , , , j = 1,2, ..., P and Features Use the size of 1×1× N 1.1×1× N The convolution kernel of 2 performs point-by-point convolution to extract the periodic features within the same variable and the coupling features between different variables:

[0062]

[0063] in, F j out1_E , F j out1_V , F j out1_C , F jout1_D , F j out1_T , F h out1 is the output feature of point-by-point convolution; b 1. b 2 is the bias.

[0064] Once again, the characteristics F j out1_E , F j out1_V , F j out1_C , F j out1_D , F j out1_T According to the non-variable relationship, they are divided into P 1 group gets features F u out1_E , F u out1_V , F u out1_C , F u out1_D , F u out1_T , u = 1, 2, ..., P 1, and perform convolution with a kernel size of 1×1× M 1 point-by-point convolution to obtain features F u out2_E , F u out2_V , F u out2_C , F u out2_D , F u out2_T ; At the same time, once again F h out1 Get features by grouping according to variable relationships F v out1 , v = 1, 2, ...,Q 1, and perform convolution with a kernel size of 1×1× M 2 point-by-point convolution to obtain features F v out2 .

[0065] Similarly, the output of the second granularity depthwise separable convolutional module is F u out2_E , F u out2_V , F u out2_C , F u out2_D , F u out2_T , F v out2 , in order to l The output of the granular depthwise separable convolution module is F u outl_E , F u outl_V , F u outl_C , F u outl _D , F u outl_T , F v outl ; Input the output of each granular depth-separable convolution module into different MLP networks to obtain the output ; Finally, the feature F out Input it into the MLP network again to get the final output result Y out ; The loss function of the output result is:

[0066]

[0067] in, N is the number of samples, Y n out is the true value of the charger power error, The output value is estimated for the electric energy error value of the online evaluation model.

[0068] The total loss function of the online evaluation model is:

[0069] .

[0070] S4. Use the obtained online evaluation model to perform online evaluation on the charger data to be tested, so as to obtain an evaluation result of the metering performance of the off-board charger of the electric vehicle.

[0071] This embodiment also provides an online evaluation system for the metering performance of an electric vehicle off-board charger, including a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the above-mentioned method steps can be implemented.

[0072] The method provided in this embodiment was used for experimental verification. The power data of 10 charging piles that have been offline verified were input into the online evaluation model constructed by this method. The evaluation results are shown in Table 1.

[0073] Table 1

[0074]

[0075] It can be seen from Table 1 that the measurement error of the method of the present invention is close to the offline measurement error, which proves the accuracy and effectiveness of the method of the present invention and has broad practical application prospects.

[0076] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0077] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0078] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0079] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0080] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. An online evaluation method for the metering performance of an off-board charger for an electric vehicle, characterized in that: The following steps are involved: S1. Collect the power data of the charger over a period of time, and normalize and divide the power data into multiple granularities; S2, performing a mask operation on the multi-granularity slice data after segmentation; S3, inputting the power data after the mask operation into an online evaluation model, wherein the online evaluation model includes a plurality of depthwise separable convolution modules, firstly performing channel-by-channel convolution on the multi-granularity slice data through the plurality of depthwise separable convolution modules, wherein each depthwise separable convolution module performs channel-by-channel convolution on only one granularity slice data, extracting features in the slice data, then grouping the plurality of slice data, performing point-by-point convolution on the grouped features, and performing comparative learning on slice data of different granularities; Thus, the online evaluation model is trained through the power data to obtain the final online evaluation model; S4. Using the obtained online evaluation model to perform online evaluation on the charger data to be tested, thereby obtaining an evaluation result of the metering performance of the off-board charger for the electric vehicle; In step S1, the collected power data includes the power error value of the charger, the output voltage of the charger, the battery state of charge value, the charging time and the battery temperature; The specific implementation method of step S1 is: Collect the energy error values ​​of N groups of chargers over a period of time E = { e 1, e 2, ..., e m }、Output voltage of charger V = { v 1, v 2, ..., v m }、Battery state of charge value C = { c 1, c 2, ..., c m }、Charging time D = { d 1, d 2,..., d m } and battery temperature T = { t 1, t 2, ..., t m }; Normalize the collected power data and map all data to between 0 and 1 to eliminate the dimension effect. The method is as follows: ； in, x express E , V , C , D , T The elements in x min , x max They are E , V , C , D , T The minimum and maximum values ​​of the elements in; Then, the normalized power data are segmented into multiple granularities to obtain , , , , ; The first granularity data segment set is , , , , ,in E i 1 , V i 1 , C i 1 , D i 1 , T i 1 The same length, i = 1, 2, ..., k ; Each power data has l A set of data fragments of different granularities. The lengths of data fragments in different sets of data fragments are different. The lengths increase in sequence, representing that the data granularity is from fine to coarse. The specific implementation method of step S2 is: Randomly add data to a collection of data fragments of different granularities s The number of masks, s is an adjustable parameter; the position where the mask is added in the first granularity data segment set is represented as , the position where the mask is added in the second granularity data segment set is expressed as , in order to l The position where the mask is added in the set of granular data segments is represented as , , that is, the locations where masks are added in sets of data fragments of different granularities are different; The mask operation is specifically as follows: replacing the element values ​​at random positions selected from the sets of data fragments of different granularities with 0 to mask the original element values; The specific implementation method of step S3 is: The first granular power data , , , , Input to the first granularity depth-separable convolution module, and use a 1× M The convolution kernel performs channel-by-channel convolution: ； in, , , , , Indicates that the convolution kernel size is 1× M Output features of the convolution operation; W 1×M represents the size of the convolution kernel, b is bias; Conv is the convolution operation, φ is the ReLu activation function, and its expression is as follows: ； Then the channel-by-channel convolution output of the second granularity depth-wise separable convolution module is , , , , , in order to l The channel-by-channel convolution output of the granular depthwise separable convolution module is , , , , ; For the first granularity depth-separable convolution module to the l The channel-by-channel convolution output of the granular depth-separable convolution module uses contrastive learning. Any univariate slice feature in the channel-by-channel convolution output of any granular depth-separable convolution module is selected as an anchor point, and the same univariate residual slice feature in the channel-by-channel convolution output of the same granular depth-separable convolution module is used as a positive sample. The same univariate slice feature in the channel-by-channel convolution output of different granular depth-separable convolution modules is used as a negative sample. The specific function form is as follows: ； in, S represents the specific category of the variable, sim() is the cosine similarity function, is the anchor point, is a positive sample, is a negative sample, r Indicates the granularity, x , y , z represents the ordinal number of the univariate slice in the channel-by-channel convolution output, , τ is the temperature parameter, MP is the number of positive samples; right , , , , According to the non-variable relationship, they are divided into P group, get feature group , , , , ; According to the relationship between variables, it is divided into Q group, get features , , h = 1, 2, ..., Q ; for features , , , , , j = 1,2, ..., P and Features Use the size of 1×1× N 1.1×1× N The convolution kernel of 2 performs point-by-point convolution to extract the periodic features within the same variable and the coupling features between different variables: ； in, F j out1_E , F j out1_V , F j out1_C , F j out1_D , F j out1_T , F h out1 is the output feature of point-by-point convolution; b 1. b 2 is bias; Once again, the characteristics F j out1_E , F j out1_V , F j out1_C , F j out1_D , F j out1_T According to the non-variable relationship, they are divided into P 1 group gets features F u out1_E , F u out1_V , F u out1_C , F u out1_D , F u out1_T , u = 1, 2, ..., P 1, and the features F u out1_E , F u out1_V , F u out1_C , F u out1_D , F u out1_T The convolution kernel size is 1×1× M 1 point-by-point convolution to obtain features F u out2_E , F u out2_V , F u out2_C , F u out2_D , F u out2_T ; At the same time, once again F h out1 Get features by grouping according to variable relationships F v out1 , v = 1, 2, ..., Q 1, and F v out1 The convolution kernel size is 1×1× M 2 point-by-point convolution to obtain features F v out2 ; Similarly, the output of the second granularity depthwise separable convolutional module is F u out2_E , F u out2_V , F u out2_C , F u out2_D , F u out2 _T , F v out2 , in order to l The output of the granular depthwise separable convolution module is F u outl_E , F u outl_V , F u outl_C , F u outl_D , F u outl_T , F v outl ; Input the output of each granular depth-separable convolution module into different MLP networks to obtain the output ; Finally, the feature F out Input it into the MLP network again to get the final output result Y out ; The loss function of the output result is: ； in, N is the number of samples, Y n out is the true value of the charger power error, Estimating output value for the power error value of the online evaluation model; The total loss function of the online evaluation model is: 。 2. An online evaluation system for the metering performance of an off-board charger for electric vehicles, characterized in that: The invention comprises a memory, a processor and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method according to claim 1 can be implemented.

Images